![Brain tumour detection using deep neural network via MRI images
Shadmaan, Rajat Panwar, Prajwal Kanaujia, Kushal Gautam, Sur Singh Rawat
Vimal Gupta
Department of Computer Science and Engineering, JSS Academy of Technical Education, Noida
ABSTRACT: The brain’s own aberrant and unregulated cell division is what causes brain
tumors. The patient cannot heal if the growth increases by more than 50%. the identification
of Brain tumor diagnosis must be swift and precise. The capture of a brain MRI scan is the
initial step, after which digital imaging techniques are used to determine the precise position
and size of the tumor. Gray and white matter make up MRI pictures, and the tumor-
containing area is more intense. In order to enhance the given brain MRI scan, noise filters
are initially utilized to remove background noise. This study aims to give a comprehensive
review on detection of brain tumors.
1 INTRODUCTION
Cell proliferation that is not under control leads to tumours. Because they don’t penetrate
the tissues around them, these tumours could only grow in smaller areas. However, if these
tumours grow close to a vital location, they may pose a risk. Malignant cancers, on the other
hand, can evolve and spread in such a way that they ultimately create a fatal kind of cancer.
MRI is recommended over other medical imaging modalities because it yields the most
contrast images of brain tumours. In instance, it has been demonstrated in numerous studies
that the transfer learning technique improves classification performance on the target dataset
by applying the knowledge acquired from one task to another that is similar [1,2]. A deep
convolutional neural network (DCNN) model must often be trained using a huge dataset,
which includes a high level of computational complexity.
2 LITERATURE REVIEW
Using MR tests, numerous researchers created a number of procedures algorithms, and
tactics to identify brain tumours, strokes, and other forms of variations in the human brain.
Brain Tumor Identification and Segmentation [4–6] outline methods for the detection of
brain tumours including segmentation, histograms, thresholding and morphology.
Fuzzy C means (FCM) does a good job of precisely segmenting tumour tissue. Svm
[11,12] was used to recognise segmentation.Utilising learning algorithm, fundamental com-
ponent analysis, as well as the wavelet transform, a hybrid technique is shown in [13,14]
algorithms, wherein the accuracy of brain tumour detection is attained 98.6%.
Three multi-resolution images from [17,18] includes the various methods. Here, the sug-
gested study asserted thatit had achieved an accuracy of 96.05%.The following approaches
are presented in the Table 1 for detecting structural abnormalities, including tumours.
DOI: 10.1201/9781003393580-3 17
Artificial Intelligence, Blockchain, Computing and Security – Dagur et al. (Eds)
© 2024 The Author(s), ISBN: 978-1-032-49393-0](https://image.slidesharecdn.com/26938674-250516084923-2257e64f/85/Artificial-Intelligence-Blockchain-Computing-And-Security-Volume-1-1st-Edition-Arvind-Dagur-48-320.jpg)
![3 WORKING OF THE MODEL
An executive control when processing medical images is the convolutional neural network. A
model of CNN as shown in the above figure is a type of machine learning used for image
analysis that focuses on learning technique component knowledge. The image processing
stage of this study comprises a variety of operations.
Table 1. Descriptions of some brain tumor detection algorithms.
Dataset
Method for Extracting
Features Classification Method Accuracy
70 MR Images Hybrid Technique Discrete
Wavelet Transformation
Artificial Neural Network with
Feedforward Backpropagation
Algorithm
97%
T2 weighted 255 MRI
images
Transform(DWPT),
Shanon Entropy(SE) and
Tsallis Entropy(TE)
Generalized Eigenvalue Proximate
Support Vector Machine
(GEPSVM)
99.61%
1800 MRI Images CNN CNN 98.60%
239 MRI Images (SGLD) ANN 99%
T2 weighted brain MR
images. Dataset: 66 and
Dataset: 255
Wavelet Transform
Curvelet Transform and
Shearlet Transform.
Support Vector Machine and
Particle Swarm Optimization
97.38%
253 MRI Images Hyper Colum Attention
Module and residual
block.
CNN 96.05%
500 MR Images Fully Automatic
Heterogeneous
Segmentation(FAHS).
SVM 98.51%
253 MR Images Improved ResNet50 CNN 97.01%
253 MR Images Deep CNN Deep CNN 98%
250 MR Images DenseNet-169 Multiple Classifier Ensemble
Multiple Classifier Ensemble
Multiple Classifier Ensemble
92.37%
3000 MR Images ResNext-101 Ensemble of multiple Classifier 93.13%
Figure 1. Analysis of brain tumor segmentation using CNN with MRI image [19].
18](https://image.slidesharecdn.com/26938674-250516084923-2257e64f/85/Artificial-Intelligence-Blockchain-Computing-And-Security-Volume-1-1st-Edition-Arvind-Dagur-49-320.jpg)
![To begin the pre-processing process, the original gray level MR pictures are retrieved in a
variety of sizes. Step 2 determines the region of interest utilising the active contours-based
segmentation technique by establishing one of the largest contour. A contour is made up of
an accumulation of points that have been interpolated utilising different residual methods to
approximate the curve in a picture, such as linear, polynomials, or splines.
The third step in a thresholding method is to select the extreme spots. Thresholding is an
essential non contextual segmentation process that generates a binary area map with a single
threshold by trying to convert a greyscale or coloured image to an image pixels [16]. Models
are used to extract features using the MR brain. image library. Data from the major
ImageNet dataset is used to train the pre-trained CNN models [20]. InceptionResNetV2 [19],
‘VGG-16, and VGG-19, Xception, ResNet50, and InceptionV3, as well as DenseNet201, are
a few of CNN models that have been trained and used in the project.
4 CONCLUSION
Healthcare image processing and classification methods have gotten a lot of interest recently.
A dramatically higher accuracy can be accomplished by getting a superior dataset with high-
resolution images that were clearly taken from the MRI scanner. To further improve quality,
classifier boosting technologies can also be used, trying to make this tool a meet its objective
for any medical facility treat brain tumors.
To further reduce the noise, classifier boosting technologies could be used, making this
tool a meet its objective for any medical facility treated brain tumors. MRI is the imaging
technique that is most advantageous for detecting brain tumors. This study aims to give a
comprehensive review on detection of brain tumors.
REFERENCES
[1] Deepak C. Dhanwani, Mahip M. Bartere, “Survey on Various Techniques of Brain Tumour Detection
from MRI Images”, IJCER, Vol.04, issue.1, Issn 2250-3005, January 2014, pg. 24–26.
[2] Megha A joshi, Shah D. H., “Survey of Brain Tumor Detection Techniques Through MRI Images”,
AIJRFANS, ISSN:2328-3785, March–May 2015, pp.09
[3] Gupta, V. and Bibhu, V., 2022. Deep Residual Network Based Brain Tumor Segmentation and
Detection with MRI Using Improved Invasive Bat Algorithm. Multimedia Tools and Applications,
pp.1–23
[4] Manoj K Kowear and Sourabh Yadev, “Brain Tumor Detection and Segmentation Using Histogram
Thresholding”, International Journal of engineering and Advanced Technology, April 2012.
[5] Rajesh C. patil, Bhalchandra A.S., “Brain Tumor Extraction from MRI Images Using MAT Lab”,
IJECSCSE, ISSN: 2277- 9477, Volume 2, issue 1.
[6] Vinay Parmeshwarappa, Nandish S, “A Segmented Morphological Approach to Detect Tumor in
Brain Images”, IJARCSSE, ISSN: 2277 128X, volume 4, issue 1, January 2014
[7] Preetha R., Suresh G. R., “Performance Analysis of Fuzzy C Means Algorithm in Automated
Detection ofBrain Tumor”, IEEE CPS, WCCCT, 2014.
[8] Amer Al-Badarnech, Hassan Najadat, Ali M. Alraziqi, “A Classifier to Detect Tumor Disease in MRI
Brain Images”, IEEE Computer Society, ASONAM. 2012, 142
[9] Palvika, Shatakshi, Sharma, Y., Dagur, A., Chaturvedi, R. (2019). Automated Bug Reporting
System with Keyword-driven Framework. In Soft Computing and Signal Processing: Proceedings of
ICSCSP 2018, Volume 2(pp. 271–277). Springer Singapore.
[10] Kumar, A., Alam, B. (2019). Energy Harvesting Earliest Deadline First Scheduling Algorithm for
Increasing Lifetime of Real Time Systems. International Journal of Electrical and Computer
Engineering, 9(1), 539.
[11] Gudigar A., Raghavendra U., San T. R., Ciaccio E. J.,and Acharya U. R., “Application of
Multiresolution Analysis for Automated Detection of Brain Abnormality Using MR Images: A
Comparative Study,” Future Gener. Comput. Syst., vol. 90, pp. 359–367, Jan. 2019.
19](https://image.slidesharecdn.com/26938674-250516084923-2257e64f/85/Artificial-Intelligence-Blockchain-Computing-And-Security-Volume-1-1st-Edition-Arvind-Dagur-50-320.jpg)
![[12] Toğaçar M., Ergen B., and Cömert Z., “BrainMRNet: Brain Tumor Detection Using Magnetic
Resonance Images with a Novel Convolutional Neural Network Model,” Med.
[13] Jia Z. and Chen D., “Brain Tumor Identification and Classification of MRI Images Using Deep
Learning Techniques,” IEEE Access, early access, Aug. 13, 2020, doi: 10.1109/ACCESS.2020.3016319.
[14] Zhuang F., Qi Z., K. Duan, Xi D., Zhu Y., Zhu H., Xiong H., and He Q., “A Comprehensive Survey
on Transfer Learning,” Proc. IEEE, vol. 109, no. 1, pp. 43–76, Jul. 2020
[15] Szegedy C., Ioffe S., Vanhoucke V., and Alemi A. A., “Inception-v4, Inception-ResNet and the Impact
of Residual Connections on Learning,” in Proc. 31st AAAI Conf. Artif. Intell., San Francisco, CA,
USA, 2017, pp. 4278–4284.
[16] He K., Zhang X., Ren S., and Sun J., “Deep Residual Learning for Image Recognition,” in Proc. IEEE
Conf. Comput. Vis. Pattern Recognit. (CVPR), Las Vegas, NV, USA, Jun. 2016, pp. 770–778.
[17] Chollet F., “Xception: Deep Learning with Depthwise Separable Convolutions,” in Proc. IEEE Conf.
Comput. Vis. Pattern Recognit. (CVPR), Honolulu, HI, USA, Jul. 2017, pp. 1800–1807.
[18] Szegedy C., Vanhoucke V., Ioffe S., Shlens J., and Wojna Z., “Rethinking the Inception Architecture
for Computer Vision,” in Proc. IEEE Conf.nComput. Vis. Pattern Recognit. (CVPR), Las Vegas, NV,
USA, Jun. 2016, pp. 2818–2826.
[19] Huang G., Liu Z., van der Maaten L., and Weinberger K. Q., “Densely connected convolutional
networks,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), Honolulu, HI, USA, Jul.
2017, pp. 2261–2269.
20](https://image.slidesharecdn.com/26938674-250516084923-2257e64f/85/Artificial-Intelligence-Blockchain-Computing-And-Security-Volume-1-1st-Edition-Arvind-Dagur-51-320.jpg)
![Variable lighting and weather conditions, a crowded background, a different pose, human
photographic flaws, different perspectives, and occlusions are all factors. Due to all of these
challenges, an effective algorithm for classification with the highest level of accuracy is required.
Convolutional neural networks are a kind of novel simulated neural network and deep learning
algorithm designed for efficient image processing. In recent years, multilayer neural networks have
been successfully applied in decision-making, learning, pattern recognition, and classification.
2 LITERATURE SURVEY
This section contains papers on all-inclusive object recognition and focus mechanisms in image
manipulation, and identification and categorization of animal species. Several image classifica-
tion techniques were covered in this study. The most common methods for classifying images are
classifiers that are object-oriented, spectral, contextual, spectral-contextual, per-pixel, per-field,
as well as robust and weak categories. The most often utilised techniques are covered in this
section. This survey provides theoretical information about categorization techniques as well as
recommendations for the best ones. The author proposed a model called “Machine Learning,
Neural Networks, and Convolutional Neural Networks” in [1]. The study of digital models that
are planned to improve effortlessly via training on example data is known as machine learning.
In [2] the author proposed a model called neural systems which was constructed by the means of
artificial neurons. In [3] the author proposed a model titled “Challenges of Camera-trap Images
for Convolutional Neural Networks.” Like all machine learning models, CNN animal classifi-
cation for camera-trap photos has to be taken into account and also generalization issues. They
demonstrated that when classifying images from untrained camera sites, unfamiliarity with
cutting-edge neural network classifiers accuracy drops.
In [4] the author proposed a model titled “Deep Convolutional Neural Networks for
Automated Wildlife Monitoring: Animal Recognition and Identification.” In this instance, they
applied a convolutional neural network. It automatically detects important features without the
need for human intervention. [5] Will focus on observing animal behavior in the wild employing
facial detection and tracing. It will show a formula for detecting and tracing fauna faces in biota
videos. Vincent Miele describes how efforts are still being made to develop species identification
criteria that are based not only measures of the craniofacial region as well as external morphol-
ogy, particularly on the cranium and maxilla, in [6]. In this case study, three different mouse
species—the house mouse Mus musculus, the European wood mouse Apodemus sylvaticus, and
the Cairo spiny mouse Acomys cahirinus—were examined. In [7], researchers used transfer
learning to classify and forecast images in a google collabrative for the ImageNet collected
information. In this study, MobileNet, MobileNetV2, VGG16, VGG19, and ResNet50 are
employed as transfer learning models The Google Colab notepad served as aid for picture cate-
gorization and prediction. The main goal in [8] was to see if they could limit a sample of manually
annotated camera-trap photos, educate a deep learning machine to recognize animals, including a
particular species. In [9] features five elements of the detection were compared to wild, a fresh
dataset of actual animal sightings focusing on difficult detection scenarios. In [10], to develop a
sensor that examines video images captured by camera traps in real time. The classes to identify
are rhinos, people and a collection of six typical big animals on the savannah of Africa. The scope
of this project also includes the extraction of significant events. Peiyi Zeng uses Python and a
simple 2D CNN to classify similar animal images in [11]. Its main focus is on the dichotomous
division between snub-nosed monkeys and other monkey species. A Python crawler is used to
build the database, each class having 800 and 200 photos serving as test and training data
respectively. The training accuracy is 96.67% when no anomalies exist. The algorithm developed
using the Missouri Hidden Cameras database [12] in Nawaz Sheikh’s MobileNet architecture and
the model appears to have performed well having an F1 score of 0.68 as opposed to InceptionV3’s
and VGG-16’s respective scores of 0.70 and 0.62. According to [13], the overarching goal of this
research was to provide managers with guidance on the most of the project objectives, accurate
22](https://image.slidesharecdn.com/26938674-250516084923-2257e64f/85/Artificial-Intelligence-Blockchain-Computing-And-Security-Volume-1-1st-Edition-Arvind-Dagur-53-320.jpg)
![models for camera-trap image analysis are required. In [14] the goal of this study, however, is to
find the creature even before hunt, not while it is going on. We propose a new method employing
machine learning techniques to separate predators from non-predators by extracting animal traits.
[15]. In [16], the proposed model was written in Python and tested in Visual Studio on a dataset
containing 12,984 images from six different animal species belonging to six different animal
classes/kingdoms. For six animal classes, the model had an accuracy of 87.22 percent.
3 MATERIAL AND METHODOLOGY
3.1 Convolutonal Neural Network (CNN)
We used Convolutional Neural Network (CNN) approach as shown in the Figure 2. This is a widely
used technique in applications for computer vision. It is a type of deep neural network used to analyse
visual imagery. This architectural style predominates when identifying objects in a photograph or
video. Applications such as neural language processing, image or video recognition, and so on.
Five layers make up a convolution neural network: input, convolution, pooling, fully
connected, and output layers. Warning capabilities, massive data capacity and picture per-
fect. Slower operation and long training period.
3.2 Deep learning
Categorization of camera-traps automatically in Nilgai by utilizing wildlife conservation in-depth
learning A branch of machine learning called deep learning, attempts to take out knowledge from
large information sets by understanding the underlying levels of more significant depiction or
characteristics (Chollet 2018). A neural network is a deep learning model composed of several
layers which are taught on labelled Information and pattern extraction in a hierarchical fashion. In
order to generate predictions that will be used to inform subsequent layers. Using predicted and
actual values, the neural network computes a rating of failures, which is then transmitted returning
via the system to modify weighing value. Knowledge is acquired in a iterative manner on new
unlabelled data by adjusting weights such that it maximises the capacity to lower it’s error causing
modest score, weights from previous layering are kept and used.
3.3 K-Nearest Neighbor (KNN) classifier
Fuzzy K-Nearest Neighbour Classifier is used to classify cattle. No training period, simple
implementation, and new data may be added at any moment without altering the model
since there is no preparation stage. Does not work well with large datasets, high dimen-
sionality, or noisy or missing data.
3.4 Transfer learning
In colab notebook, image classification and prediction are accomplished by utilising transfer learn-
ing. Adaptive learning allows developers to prevent the requirement for an abundance of new data.
The One of the most important restrictions on transfer learning is the problem of negative transfer.
Figure 2. CNN Architecture layers.
23](https://image.slidesharecdn.com/26938674-250516084923-2257e64f/85/Artificial-Intelligence-Blockchain-Computing-And-Security-Volume-1-1st-Edition-Arvind-Dagur-54-320.jpg)
![This ebook is for the use of anyone anywhere in the United
States and most other parts of the world at no cost and with
almost no restrictions whatsoever. You may copy it, give it away
or re-use it under the terms of the Project Gutenberg License
included with this ebook or online at www.gutenberg.org. If you
are not located in the United States, you will have to check the
laws of the country where you are located before using this
eBook.
Title: The Life of Sir Isaac Newton
Author: David Brewster
Release date: October 18, 2016 [eBook #53311]
Most recently updated: October 23, 2024
Language: English
Credits: E-text prepared by Sonya Schermann, Charlie Howard,
and the Online Distributed Proofreading Team
(http://www.pgdp.net) from page images generously
made available by Internet Archive
(https://archive.org)
*** START OF THE PROJECT GUTENBERG EBOOK THE LIFE OF SIR
ISAAC NEWTON ***](https://image.slidesharecdn.com/26938674-250516084923-2257e64f/85/Artificial-Intelligence-Blockchain-Computing-And-Security-Volume-1-1st-Edition-Arvind-Dagur-60-320.jpg)
Artificial Intelligence Blockchain Computing And Security Volume 1 1st Edition Arvind Dagur
- 1. Artificial Intelligence Blockchain Computing And Security Volume 1 1st Edition Arvind Dagur download https://ebookbell.com/product/artificial-intelligence-blockchain- computing-and-security-volume-1-1st-edition-arvind-dagur-53877348 Explore and download more ebooks at ebookbell.com
- 2. Here are some recommended products that we believe you will be interested in. You can click the link to download. Artificial Intelligence Blockchain Computing And Security Volume 2 1st Edition Arvind Dagur https://ebookbell.com/product/artificial-intelligence-blockchain- computing-and-security-volume-2-1st-edition-arvind-dagur-53877310 Blockchain Technology In Healthcare Concepts Methodologies And Applications Applied Artificial Intelligence In Data Science Cloud Computing And Iot Frameworks Kamila https://ebookbell.com/product/blockchain-technology-in-healthcare- concepts-methodologies-and-applications-applied-artificial- intelligence-in-data-science-cloud-computing-and-iot-frameworks- kamila-54979184 Regulation Of Innovative Technologies Blockchain Artificial Intelligence And Quantum Computing Rosario Girasa https://ebookbell.com/product/regulation-of-innovative-technologies- blockchain-artificial-intelligence-and-quantum-computing-rosario- girasa-43638978 Digital Fluency Understanding The Basics Of Artificial Intelligence Blockchain Technology Quantum Computing And Their Applications For Digital Transformation 1st Ed Volker Lang https://ebookbell.com/product/digital-fluency-understanding-the- basics-of-artificial-intelligence-blockchain-technology-quantum- computing-and-their-applications-for-digital-transformation-1st-ed- volker-lang-23942238
- 3. Digital Fluency Understanding The Basics Of Artificial Intelligence Blockchain Technology Quantum Computing And Their Applications For Digital Transformation Volker Lang https://ebookbell.com/product/digital-fluency-understanding-the- basics-of-artificial-intelligence-blockchain-technology-quantum- computing-and-their-applications-for-digital-transformation-volker- lang-46247660 Artificial Intelligence Blockchain And Iot For Smart Healthcare Sharma https://ebookbell.com/product/artificial-intelligence-blockchain-and- iot-for-smart-healthcare-sharma-43872142 Artificial Intelligence Governance And The Blockchain Revolution 2024th Edition Qiqi Gao https://ebookbell.com/product/artificial-intelligence-governance-and- the-blockchain-revolution-2024th-edition-qiqi-gao-56247568 Artificial Intelligence Machine Learning And Blockchain In Quantum Satellite Drone And Network Thiruselvan Subramanian https://ebookbell.com/product/artificial-intelligence-machine- learning-and-blockchain-in-quantum-satellite-drone-and-network- thiruselvan-subramanian-44710168 Artificial Intelligence Machine Learning And Blockchain In Quantum Satellite Drone And Network 1st Edition Thiruselvan Subramanian https://ebookbell.com/product/artificial-intelligence-machine- learning-and-blockchain-in-quantum-satellite-drone-and-network-1st- edition-thiruselvan-subramanian-44746370
- 6. ARTIFICIAL INTELLIGENCE, BLOCKCHAIN, COMPUTING AND SECURITY, VOLUME 1 This book contains the conference proceedings of ICABCS 2023, a non-profit conference with the objective to provide a platform that allows academicians, researchers, scholars and students from various institutions, universities and industries in India and abroad to exchange their research and innovative ideas in the field of Artificial Intelligence, Blockchain, Computing and Security. It explores the recent advancements in the field of Artificial Intelligence, Blockchain, Communication and Security in this digital era for novice to profound knowledge about cutting edges in Artificial Intelligence, financial, secure transaction, monitoring, real time assistance and security for advanced stage learners/ researchers/ academicians. The key features of this book are: l Broad knowledge and research trends in AI and Blockchain with security and their role in smart living assistance l Depiction of system model and architecture for clear picture of AI in real life l Discussion on the role of AI and Blockchain in various real-life problems across sectors including banking, healthcare, navigation, communication, security l Explanation of the challenges and opportunities in AI and Blockchain based healthcare, education, banking, and related industries This book will be of great interest to researchers, academicians, undergraduate students, postgraduate students, research scholars, industry professionals, technologists and entrepreneurs.
- 8. PROCEEDINGS OF THE INTERNATIONAL CONFERENCE ON ARTIFICIAL INTELLIGENCE, BLOCKCHAIN, COMPUTING AND SECURITY (ICABCS 2023), GR. NOIDA, UP, INDIA, 24–25 FEBRUARY 2023 Artificial Intelligence, Blockchain, Computing and Security Volume 1 Edited by Arvind Dagur School of Computing Science and Engineering, Galgotias University, Gr. Noida Karan Singh School of Computer & Systems Sciences, JNU New Delhi Pawan Singh Mehra Department of Computer Science and Engineering, Delhi Technological University, New Delhi Dhirendra Kumar Shukla School of Computing Science and Engineering, Galgotias University, Gr. Noida
- 9. First published 2023 by CRC Press/Balkema 4 Park Square, Milton Park, Abingdon, Oxon, OX14 4RN and by CRC Press/Balkema 2385 NW Executive Center Drive, Suite 320, Boca Raton FL 33431 CRC Press/Balkema is an imprint of the Taylor & Francis Group, an informa business ’ 2024 selection and editorial matter, Arvind Dagur, Karan Singh, Pawan Singh Mehra & Dhirendra Kumar Shukla; individual chapters, the contributors The right of Arvind Dagur, Karan Singh, Pawan Singh Mehra & Dhirendra Kumar Shukla to be identified as the authors of the editorial material, and of the authors for their individual chapters, has been asserted in accordance with sections 77 and 78 of the Copyright, Designs and Patents Act 1988. All rights reserved. No part of this book may be reprinted or reproduced or utilised in any form or by any electronic, mechanical, or other means, now known or hereafter invented, including photocopying and recording, or in any information storage or retrieval system, without permission in writing from the publishers. Although all care is taken to ensure integrity and the quality of this publication and the information herein, no responsibility is assumed by the publishers nor the author for any damage to the property or persons as a result of operation or use of this publication and/or the information contained herein. British Library Cataloguing-in-Publication Data A catalogue record for this book is available from the British Library Library of Congress Cataloging-in-Publication Data A catalog record has been requested for this book SET ISBN: 978-1-032-66966-3 (hbk) ISBN: 978-1-032-68590-8 (pbk) Volume 1 ISBN: 978-1-032-49393-0 (hbk) ISBN: 978-1-032-49397-8 (pbk) ISBN: 978-1-003-39358-0 (ebk) DOI: 10.1201/9781003393580 Volume 2 ISBN: 978-1-032-67841-2 (hbk) ISBN: 978-1-032-68498-7 (pbk) ISBN: 978-1-032-68499-4 (ebk) DOI: 10.1201/9781032684994 Typeset in Times New Roman by MPS Limited, Chennai, India
- 10. Table of Contents Preface xvii Acknowledgements xix Committee Members xxi National Advisory Committee xxiii Organizing committee xxv Artificial Intelligence An ensemble learning approach for large scale birds species classification 3 Harsh Vardhan, Aryan Verma & Nagendra Pratap Singh Comprehensive analysis of human action recognition and object detection in aerial environments 9 Mrugendrasinh Rahevar, Amit Ganatra, Hiren Mewada & Krunal Maheriya Brain tumour detection using deep neural network via MRI images 17 Shadmaan, Rajat Panwar, Prajwal Kanaujia, Kushal Gautam, Sur Singh Rawat & Vimal Gupta A review on wildlife identification and classification 21 Kartikeyea Singh, Manvi Singhal, Nirbhay Singh, Sur Singh Rawat & Vimal Gupta Image caption for object identification using deep convolution neural network 26 Sarthak Katyal, Dhyanendra Jain & Prashant Singh Brain tumor detection using texture based LBP feature on MRI images using feature selection technique 30 Vishal Guleria, Aryan Verma, Rishabh Dhenkawat, Uttkarsh Chaurasia & Nagendra Pratap Singh Towards computationally efficient and real-time distracted driver detection using convolutional neutral networks 37 Ramya Thatikonda, Sambit Satpathy, Shabir Ali, Munesh Chandra Trivedi & Mohit Choudhry A systematic study of networking design for co-working space environment 47 Rohit Vashisht, Rahul Kumar Sharma & Gagan Thakral Application of neural network algorithms in early detection of breast cancer 53 D.K. Mukhamedieva & M.E. Shaazizova Stock market prediction using DQN with DQNReg loss function 58 Alex Sebastian, K.V. Habis & Samiksha Shukla Towards improving the efficiency of image classification using data augmentation and transfer learning techniques 64 A. Christy, S. Prayla Shyry & M.D. Anto Praveena v Artificial Intelligence, Blockchain, Computing and Security – Dagur et al. (Eds) © 2024 The Editor(s), ISBN: 978-1-032-49393-0
- 11. Predicting stock market price over the years by utilizing machine learning algorithms 72 Mrignainy Kansal, Pancham Singh, Sachin Kumar & Ritu Sibal A survey and classification of lung, breast, thyroid, and prostate cancer detection 79 Dhananjay Kumar Sharma, Manoj Kumar Pal, Ashutosh Kumar Singh & Vijay Kumar Dwivedi Modified attention based cryptocurrency price presage with convolutional Bi-LSTM 84 Vibha Srivastava, Ashutosh Kumar Singh & Vijay Kumar Dwivedi Classification of vegetation, soil and water bodies of Telangana region using spectral indices 93 Devulapalli Sudheer, S. Nagini, Naga Sreenija Meka, Yasaswini Kolli, Anudeep Eloori, Nithish Kumar Chowdam & Rushikesh Reddy Dorolla Ovarian cancer identification using transfer learning 100 Rishabh Dhenkawat, Samridhi Singh & Nagendra Pratap Singh A study on automatic mathematical word problem solvers 108 Madhavi Alli, Balaga Sateesh, Duggasani Yaswanth Reddy, Potlapelli Sai Koushik & Thelukuntla Sai Chandra Predicting learning styles in personalized E-learning platforms 116 A. Madhavi, A. Nagesh & A. Govardhan Exploratory Data Analysis (EDA) based on demographical features for students’ performance prediction 126 Neeraj Kumar Srivastava, Prafull Pandey & Vikas Mishra Vehicle detection using Artificial Intelligence for traffic surveillance 134 Soma Ajay, Sai Vardhan Reddy, Tharun, Santhosh Kumar Pandian & T. Shakila Predicting Encopresis & Enuresis treatment: Utilizing AI 144 Rolly Gupta & Dr. Lalit Kumar Sagar Object detection from images by convolutional neural networks for embedded systems using Cifar-10 images 150 Tushar Singh & Vinod Kumar Recognition of Indian sign language using hand gestures 155 Umang Rastogi, Anand Pandey & Vinesh Kumar Potato plant leaf diseases detection and identification using convolutional neural networks 160 Sriram Gurusamy, B. Natarajan, R. Bhuvaneswari & M. Arvindhan Review: Recent advancements on Artificial Intelligence 166 Meeta Singh, Poonam Chahal, Deepa Bura & Srishty Embeddings of knowledge graphs for link prediction: A systematic analysis 172 Neelam Jain & Krupa Mehta Predictive system on the car market trend using AI & ML 176 Ansh Shankar, Dhruv Varshney & Arvind Nath Sinha vi
- 12. Object detection system with voice output using Artificial Intelligence 181 K. Sivaraman, Pinnika Gopi, Katta Karthik & Kamsani Venkata Upendar Reddy Multi-objective optimization-based methodological framework for net zero energy building design in India 187 Pushpendra Kumar Chaturvedi, Nand Kumar & Ravita Lamba A comparative study of different BERT modifications 195 S. Agarwal & M. Jain Prediction of cardiovascular diseases using explainable AI 201 Anuradha S. Deokar & M.A. Pradhan Music generation using RNNs and LSTMs 207 H. Aditya, J. Dev, S. Das & A. Yadav Effectiveness of virtual education during Covid-19: An empirical study in Delhi NCR 216 Girish Kumar Bhasin & Manisha Gupta Block chain Land transaction and registration system using blockchain 233 Anubhavi Agrawal, Ayush Teotia, Dhrubb Gupta, Akash Srivastava, G. Mahesh & B.C. Girish Kumar E-policing and information management system using blockchain technology 238 G. Mahesh, B.C. Girish Kumar, Shivani Pathak, M. Surekha, K.G. Harsha & Mukesh Raj A survey on Automated Market-Makers (AMM) for non-fungible tokens 244 Rishav Uppal, Ojuswi Rastogi, Priyam Anand, Vimal Gupta, Sur Singh Rawat & Nitima Malsa Blockchain based prophecy of cardiovascular disease using modified XGBoost 250 Vibha Srivastava, Ashutosh Kumar Singh & Vijay Kumar Dwivedi A survey on crowdfunding using blockchain 259 Nikunj Garg, Siddharth Seth, Naincy Rastogi, Rajiv Kumar, Vimal Gupta, Sur Singh Rawat & Nitima Malsa Data provenance for medical drug supply chain using blockchain-based framework 264 Martin Parmar & Parth Shah Blockchain technology for agricultural data sharing and sustainable development of the ecosystem 272 Ashok Kumar Koshariya, Virendra Kumar, Vashi Ahmad, Bachina Harish Babu, B. Umarani & S. Ramesh Problems of developing a decentralized system based on blockchain technology 277 D.T. Muhamediyeva, A.N. Khudoyberdiev & J.R. Abdurazzokov Authenticating digital documents using block chain technology 283 E. Benitha Sowmiya, D. Isaiah Ramaswamy, S. Hemanth Sai, T. Vignesh & S. Madhav Sai vii
- 13. Communications Vehicles communication and safe distancing using IOT and ad-hoc network 289 Raj Kumar Sharma, Roushan, Rajneesh Dev Singh & Isha Nair PG Radar 294 Yash Grover, Aditya & Kadambari Agarwal A new framework for distributed clustering based data aggregation in WSN 298 Anuj Kumar Singh, Shashi Bhushan & Ashish Kumar Designing composite codes to mitigate side-lobe levels in MIMO radar using polyphase codes 305 Ankur Thakur & Bobbinpreet Kaur Design and implementation of industrial fire detection and control system using internet of things 310 Tanushree Bharti, Madan Lal Saini, Ashok Kumar & Rajat Tiwari Implementation of optimized protocol for secure routing in cloud based wireless sensor networks 316 Radha Raman Chandan, Sushil Kumar, Sushil Kumar Singh, Abdul Aleem & Basu Dev Shivahare A cross CNN-LSTM model for sarcasm identification in sentiment analysis 322 Sandeep Kumar, Anuj Kumar Singh, Shashi Bhushan & Vineet Kumar Singh General track Detection of hate speech in multi-modal social post 331 Abhishek Goswami, Ayushi Rawat, Shubham Tongaria & Sushant Jhingran IC-TRAIN – an advance and dynamically trained data structure 337 Ochin Sharma Big Bang theory improved shortest path, construction, evolution and status model based course like environment machine learning 343 Tejinder Kaur, Abhijeet Singh, Yuvraj Singh Behl, Sanjoy Kumar Debnath & Susama Bagchi A module lattice based construction of post quantum blockchain for secure transactions in Internet of Things 351 Dharminder Chaudhary, M.S.P. Durgarao, Pratik Gupta, Saurabh Rana & Soumyendra Singh Entertainment based website: A review and proposed solution for lightning fast webpages 358 Prashante, Arslan Firoz, Vishesh Khullar & Abdul Aleem Sentiment analysis based brand recommendation system: A review 364 Chaitanya Rastogi, Darshika Singh, Ashutosh Dwivedi, Anshika Chaudhary, Sahil Kumar Aggarwal & Ruchi Jain Models for integrating Artificial Intelligence approaches & the future the humans 369 Ojas Sharma & Tejinder Kaur viii
- 14. Deep learning driven automated malaria parasite detection in thin blood smears 375 Aryan Verma, Sejal Mansoori, Adithya Srivastava, Priyanka Rathee & Nagendra Pratap Singh The future of mobile computing in smart phones and its potentiality- A survey 381 Mohd Shahzad & Geetinder Saini Auto scaling in cloud computing environments with AWS 387 Nazish Baliyan & Sukhmeet Kaur Analysis of cryptanalysis methods applied to stream encryption algorithms 393 Rakhmatullayev Ilkhom Rakhmatullaevich & Ilkhom Boykuziyev Mardanokulovich Early recognition of Alzheimer’s disease using machine learning 402 Prajwal Nagaraj, Anjan K. Koundinya & G. Thippeswamy Detecting malign in leaves using deep learning algorithm model ResNet for smart framing 407 Anmol Kushwaha & E. Rajesh COVID-19 prediction using deep learning VGG16 model from X-ray images 412 Narenthira Kumar Appavu & C. Nelson Kennedy Babu EDGE computing as a mapping study 419 Md Sarazul Ali & Ramneet Kaur Reverse and inverse engineering using machine and deep learning: Futuristic opportunities and applications 425 Sanjeev Kumar, Pankaj Agarwal, Jay Shankar Prasad, D. Pandey & Saurabh Chandra A Comprehensive study of risk prediction techniques for cardiovascular disease 433 Huma Parveen, Syed Wajahat, Abbas Rizvi & Raja Sarath Kumar Boddu WeSafe: A safety app for all 441 Reshma Kanse, Supriya Ajagekar, Trupti Patil, Harish Motekar, Vinod Rathod, Rahul Papalkar & Shabir Ali Detection of toxic comments over the internet using deep learning methods 447 Akash Naskar, Rohan Harchandani & K.T. Thomas Performance testing of scheduling algorithms for finding the availability factor 455 Prathamesh Vijay Lahande & Parag Ravikant Kaveri Higher education recommendation system using data mining algorithm 460 S. Ponmaniraj, S. Naga Kishore, G. Shashi Kumar, C.H. Abhinay & B. Harish A brief evaluation of deep learning-based retinal disease approaches 466 Reetika Regotra, Tamana, Samridhi Singh & Shekhar Yadav Multifunctional pose estimator workout guider 474 D. Burad, B. Gurav, S. Desai, S. Banerjee & S. Agrawal ix
- 15. Volume control using hand gesture recognition 481 Ashish Kumar Mallick, Adil Islam & Abdul Aleem Connecting faces: Secure social interconnection 486 Aatif Jamshed, Ankit Bhardwaj, Avi Nigam, Ujjwal Gupta & Sachin Goel 2019-nCovid Safe – a deep learning application for crowd management 491 Prachi Pundhir, Aatif Jamshed, Puneet Kumar Aggarwal, Sukrati Pateriya, Vaani Tyagi & Vanshita Garg Stock market price forecasting 498 Raja Jadon, Shivam Yadav & Abdul Aleem Analysis of node security optimization in WSN 504 K. Sharma, S. Chhabra & S. Rani Online roadside vehicle assistance: A review 510 Rohan Dass Gujrati, Roshi Kumar, Rupali Chaubey, Shikha Singh & Sahil Kumar Aggarwal Security approaches in software defined networks using machine learning – a critical review 515 Zahirabbas J. Mulani & Suhasini Vijaykumar Arduino based fire detection alarm in rural areas 522 Omkar Bhattarai, Abhay Aditya Dubey, Shashank Singh & Avjeet Singh Recent advances and future technologies in IoT, blockchain and 5G Fog enabling technologies in healthcare: A review 531 Aditya Yadav, Onesimus Chandra Pradhan, Ruqaiya Khanam & Amrita Microstrip patch antenna with high gain and dual bands for secure 5G communication 539 V. Kalai Priya, D. Sugumar, K. Vijayalakshmi, V. Vanitha, Charanjeet Singh & A. Yasminebegum Blockchain-based access control and interoperability framework for electronic health records (ANCILE) 544 G. Senthilkumar, Aravindan Srinivasan, J. Venkatesh, Ramu Kuchipudi, K. Vinoth & A. Ramamoorthy Deep learning based approach for rice prediction from authenticated block chain mode 550 V.V. Satyanarayana Tallapragada, Sumit Chaudhary, J. Sherine Glory, G. Venkatesan, B. Uma Maheswari & E. Rajesh Kumar Digital media industry driven by 5G and blockchain technology 557 B. Md. Irfan, Ramakrishnan Raman, Hirald Dwaraka Praveena, G. Senthilkumar, Ashok Kumar & Ruhi Bakhare Deep learning approach for smart home security using 5G technology 563 M. Amanullah, Sumit Chaudhary, R. Yalini, M. Balaji, M. Vijaya Sudha & Joshuva Arockia Dhanraj x
- 16. IoT based deep learning approach for online fault diagnosis against cyber attacks 569 A. Yovan Felix, V. Sharmila, S. Nandhini Devi, S. Deena, Ajay Singh Yadav & K. Jeyalakshmi Intrusion detection system using soft computing techniques in 5G communication systems 574 D. Dhanya, Shankari, I. Kathir, Ramu Kuchipudi, I. Thamarai & E. Rajesh Kumar Reducing power consumption in 2 tier H-CRAN using switch active/sleep of small cell RRHs 580 Amit Kumar Tiwari, Pavan Kumar Mishra & Sudhakar Pandey A Blockchain-based AI approach towards smart home organization security 589 Sarfraz Fayaz Khan, S. Sharon Priya, Mukesh Soni, Ismail Keshta & Ihtiram Raza Khan Multi-party secure communication using blockchain over 5G 597 K. Archana, Z.H. Kareem, Liwa H. Al-Farhani, K. Bagyalakshmi, Ignatia K. Majella Jenvi & Ashok Kumar Parallel Byzantine fault tolerance method for blockchain 605 Kumar Pradyot Dubey, C.N. Gnanaprakasam, Ihtiram Raza Khan, Md Shibli Sadik, Liwa H. Al-Farhani & Samrat Ray Fuzzy random proof of work for consensus algorithm in blockchain 613 Akhilesh Kumar, Z.H. Kareen, Mustafa Mudhafar, Gioia Arnone, Mekhmonov Sultonali Umaralievich & Avijit Bhowmick Security model to identify block withholding attack in blockchain 621 Ismail Keshta, Faheem Ahmad Reegu, Adeel Ahmad, Archana Saxena, Radha Raman Chandan & V. Mahalakshmi Threshold public key-sharing technique in block chain 630 Sagar Dhanraj Pande, Gurpreet Singh, Djabeur Mohamed Seifeddine Zekrifa, Shilpa Prashant Kodgire, Sunil A. Patel & Viet-Thanh Le 5G geological data for seismic inversion data detection based on wide-angle reflection wave technology 640 M. Thiyagesan, B. Md. Irfan, Ramakrishnan Raman, N. Ponnarasi, P. Ramakrishnan & G.A. Senthil Performance evaluation and comparison of blockchain mechanisms in E-healthcare 645 Prikshat Kumar Angra, Aseem Khanna, Gopal Rana, Manvendra Singh, Pritpal Singh & Ashwani Kumar Blockchain-Aware secure lattice aggregate signature scheme 653 Motashim Rasool, Arun Khatri, Renato R. Maaliw, G. Manjula, M.S. Kishan Varma & Sohit Agarwal Developing secure framework using blockchain technology for E-healthcare 662 Pritpal Singh, K. Jithin Gangadharan, Ashwani Kumar, Priya Chanda, Prikshat Kumar & Aseem Khanna Blockchain-based trusted dispute resolution service architecture 670 Ravi Mohan Sharma, V. Rama Krishna, Tirtha Saikia, Ashish Suri, Richard Rivera & Dinesh Mavaluru xi
- 17. Deep learning based federated learning scheme for decentralized blockchain 679 Gowtham Ramkumar, S. Sivakumar, Mukesh Soni, Yasser Muhammed, Hayder Mahmood Salman & Arsalan Muhammad Soomar Blockchain-aware federated anomaly detection scheme for multivariate data 690 V. Selvakumar, Renato R. Maaliw, Ravi Mohan Sharma, Rajvardhan Oak, Pavitar Parkash Singh & Ashok Kumar Recent advancements and challenges in Artificial Intelligence, machine learning, cyber security and blockchain technologies Relative study on machine learning techniques for opinion analysis of social media contents 701 V. Malik & N. Tyagi Review of permission-based malware detection in Android 708 Nishant Rawat, Amrita & Avjeet Singh A two-way online speech therapy system 714 Monika Garg, Mohini Joshi & Anchal Choudhary Comparative analysis of electronic voting methods based on blockchain technology 719 Zarif Khudoykulov, Umida Tojiakbarova, Ikbola Xolimtayeva & Barno Shamsiyeva Insider threat detection of ransomware using AutoML 724 R. Bhuvaneswari, Enaganti Karun Kumar, Annadanam Padmasini & K.V. Priyanka Varma Multispectral image processing using ML based classification approaches in satellite images 734 V.V. Satyanarayana Tallapragada, G. Venkatesan, G. Manisha, N. Sivakumar, Ashok Kumar & J. Karthika Malicious data detection in IoT using deep learning approach 739 Srinivas Kolli, Aravindan Srinivasan, R. Manikandan, Shalini Prasad, Ashok Kumar & S. Ramesh Detecting cross-site scripting attacks using machine learning: A systematic review 743 D. Baniya, Amrita & A. Chaudhary A speech emotion recognition system using machine learning 749 Reshma Kanse, Supriya Ajagekar, Trupti Patil, Harish Motekar, Vinod Rathod, Rahul Papalkar & Shabir Ali A predictive approach of property price prediction using regression models 755 Saad Khan, Shikha Singh, Bramha Hazela & Garima Srivastava Biological immune system based risk mitigation monitoring system: An analogy 760 Nida Hasib, Syed Wajahat Abbas Rizvi & Vinodani Katiyar A review on malicious link detection techniques 768 Ashim Chaudhary, K.C. Krishna, Md Shadik & Dharm Raj xii
- 18. Application of state-of-the-art blockchain and AI research in healthcare, supply chain, e-governance etc. V2E: Blockchain based E-voting system 781 Bipin Kumar Rai, Mukul Kumar Sahu & Viraaj Akulwar Blockchain based supply chain management system 786 Bipin Kumar Rai, Dhananjay Singh & Nitin Sharma Sign language detection using computer vision 791 Shivani Sharma, Bipin Kumar Rai, Manak Rawal & Kaustubh Ranjan An overview of thalassemia: A review work 796 Ruqqaiya Begum, G. Suryanarayana, B.V. Saketha Rama & N. Swapna Cloud computing architecture and adoption for agile system and devOps Smart face recognition attendance system using AWS 807 Nidhi Sharma, Samarth Gaur & Preksha Pratap A review on identification of fake news by using machine learning 812 Naeema Ahmed & Mukesh Rawat Optimal resource allocation in cloud: Introduction to hybrid optimization algorithm 817 Shubham Singh, Pawan Singh & Sudeep Tanwar Fake news detection on social-media: A 360 degree survey view 825 Vivek Kumar, Satveer, Waseem Ahmad & Satyaveer Singh Cloud computing in education 830 S. Singh, A. Singh & A. Singh Cloud economics and its influence on business 835 F. Nadeem & A. Singh Live virtual machine migration towards energy optimization in cloud datacenters 840 Rohit Vashisht, Gagan Thakral & Rahul Kumar Sharma Flexible-responsive data replication methodology for optimal performance in cloud computing 846 Snehal Kolte & Madhavi Ajay Pradhan Trends in cloud computing and bigdata analytics An innovative technique for management of personal data using intelligence 857 A. Sardana, A. Moral, S. Gupta & V. Kiran Crypt Cloud+: Cloud storage access control: Expressive and secure 863 Rajesh Bojapelli, Manikanta, Srinivasa Rao, Chinna Rao & R. Jagadeeswari Application of MCDM methods in cloud computing: A literature review 868 A. Kumar, A.K. Singh & A. Garg xiii
- 19. A review: Map-reduce (Hadoop) based data clustering for big data 874 Mili Srivastava, Hitesh Kansal, Aditi Gautam & Shivani Modeling of progressive Alzheimer’s disease using machine learning algorithms 879 Mitu Ranjan & Sushil Kumar Electronics and scientific computing to solve real-world problems An overview of electric vehicle and enhancing its performances 889 Lipika Nanda, Suchibrata Dash, Babita Panda & Rudra Narayan Dash Prophet-based energy forecasting of large-scaled solar photovoltaic plant 894 Akash Tripathi, Brijesh Singh & Jitendra Kumar Seth Skin cancer detection by using squeeze and excitation method 900 Shaurya Pandey, Rishabh Dhenkawat, Shekhar Yadav & Nagendra Pratap Singh Augmentation of medical image dataset using GAN 908 Harsh Sheth, Samridhi Singh, Nagendra Pratap Singh & Priyanka Rathee A review on tunable UWB antenna with multi-band notching techniques 913 Amit Madhukar Patil & Om Prakash Sharma System for predicting soil moisture using Arduino-UNO 919 A. Kapahi, V. Kapahi, D. Gupta, H. Verma & M. Singh Tank water flow automation 924 Aniketh Santhan, Aaditya Kumar Tomar, Vikalp Arora & Dharm Raj Recognition techniques of medicinal plants: A review 929 Nidhi Tiwari, Bineet Kumar Gupta & Rajat Sharma Security and privacy in the cloud computing Secure data storage using erasure-data in cloud environment 939 K. Bala, Balakrishna Reddy Mule, Rishi Raj Kumar, Srinivasulu Gude & Ranga Uday Sudheer Gaddam Security aspects in E-voting system using cloud computing 945 Shreyas Agrawal & Mohammad Junedul Haque Enhanced-honey bee based load balancing algorithm for cloud environment 951 Saurabh Singhal, Shabir Ali, Dhirendra Kumar Shukla, Arvind Dagur, Rahul Papalkar, Vinod Rathod & Mohan Awasthy A detail study on feature extraction technique for content based image retrieval for secure cloud computing 957 J. Sheeba Selvapattu & Suchithra R. Nair Secure data storage based on efficient auditing scheme 964 R. Selvaganesh, K. Akash Sriram, K. Venkatesh & K. Sai Teja Decision trees to detect malware in cloud computing environment 969 Poovidha Ayyappa, Govindu Kiran Kumar Reddy, Katamreddy Siva Satish, Prakash Rachakonda & Pujari Manjunatha xiv
- 20. An optimized feature selection guided light-weight machine learning models for DDoS attacks detection in cloud computing 975 Rahul R. Papalkar, A.S. Alvi, Shabir Ali, Mohan Awasthy & Reshma Kanse Analysis of methods for multiple reviews based sentiment analysis 983 Syed Zeeshan Ali Abrar Alvi & Ajay B. Gadicha Review of unknown attack detection with deep learning techniques 989 Rahul Rajendra Papalkar & Abrar S. Alvi Author index 999 xv
- 22. Preface On the behalf of organising committee, I would like to extend my heartiest welcome to the first international conference on Artificial Intelligence, Blockchain, Computing and Security (ICABCS 2023). ICABCS 2023 is a non-profit conference and the objective is to provide a platform for academicians, researchers, scholars and students from various institutions, universities and industries in India and abroad, to exchange their research and innovative ideas in the field of Artificial Intelligence, Blockchain, Computing and Security. We invited all students, research scholars, academicians, engineers, scientists and industrialists working in the field of Artificial Intelligence, Blockchain, Computing and Security from all over the world. We warmly welcomed all the authors to submit their research in conference ICABCS 2023 to share their knowledge and experience among each other. This two-day international conference (ICABCS 2023) was organized at Galgotias University on 24th and 25th February 2023. The inauguration was done on 24th February 2023 at Swami Vivekananda Auditorium of Galgotias University. In the inauguration ceremony, Professor Shri Niwas Singh, Director, Atal Bihari Bajpai Indian Institute of Information Technology and Management, Gwalior attended as Chief Guest. Professor Rajeev Tripathi, former Director Motilal Nehru National Institute of Technology Allahabad and Professor D.K. Lobiyal, Jawaharlal Nehru University attended as Guests of Honour. In the inauguration ceremony of the program, the Vice-Chancellor of the University, Professor K. Mallikarjuna Babu, Advisor to Chancellor, Professor Renu Luthra, Dean SCSE, Professor Munish Sabharwal welcome the guests with welcome address. The Registrar, COE and Deans of all the Schools were present. Conference Chair Professor Arvind Dagur told that in this conference more than 1000 research papers were received from more than ten countries, on the basis of blind review of two reviewers, more than 272 research papers were accepted and invited for presentation in the conference. The Chief Guest, Honorable Guests and Experts delivered lectures on Artificial Intelligence, Block Chain and Computing Security and motivated the participants for quality research. The Pro Vice-Chancellor, Professor Avadhesh Kumar, delivered the vote of thanks to conclude the inauguration ceremony. During the two-day conference, more than 272 research papers were presented in 22 technical sessions. The closing ceremony was presided over by Prof. Awadhesh Kumar, Pro-VC of the University and Conference Chair Professor Arvind Dagur, on behalf of the Organizing Committee. Conference Chair, Professor Arvind Dagur thanked Chancellor Mr. Sunil Galgotia, CEO Mr. Dhruv Galgotia, Director Operation Ms. Aradhana Galgotia, Vice Chancellor, Pro Vice-Chancellor, Registrar, Dean SCSE, Dean Engineering and university family for their co-operation and support. Finally, once again I would like to thank to all participants for their contribution to the conference and all the organising committee members for their valuable support to organise the conference successfully. I highly believed that this conference was a captivating and fascinating platform for every participant. On the behalf of editors Dr. Arvind Dagur xvii Artificial Intelligence, Blockchain, Computing and Security – Dagur et al. (Eds) © 2024 The Editor(s), ISBN: 978-1-032-49393-0
- 24. Acknowledgements It gives me immense pleasure to note that Galgotias University, Greater Noida, India is organizing the International Conference on Artificial Intelligence, Blockchain, Computing and Security (ICABCS 2023) on 24th and 25th February 2023. On behalf of the organizing committee, I would like to convey my sincere thanks to our Chief Patron, Honorable Shri Sunil Galgotia, Chancellor, GU and Hon’ble Shri Dhruv Galgotia CEO, GU for providing all the necessary support and facilities required to make ICABCS-2023 a successful con- ference. I convey my thanks to Prof. (Dr) K. Mallikharjuna Babu, Vice Chancellor and Prof. (Dr) Renu Luthra advisor to the chancellor for their continuous support and encourage- ment, without which it was not possible to achieve. I want to convey my sincere thanks to them for providing technical sponsorship and for showing their confidence in Galgotias University to provide us the opportunity to organize ICABCS-2023 and personally thank to all the participants of ICABCS 2023. I heartily welcome all the distinguished keynote speakers, guest, session chairs and all the authors presenting papers. In the end, I would convey my thanks to all the reviewers, organizing committee members, faculty and student volunteers for putting their effort into making the conference ICABCS 2023 a grand success. Thank you, Prof.(Dr.) Arvind Dagur Organizing Chair ICABCS 2023, Galgotias University xix Artificial Intelligence, Blockchain, Computing and Security – Dagur et al. (Eds) © 2024 The Editor(s), ISBN: 978-1-032-49393-0
- 26. Committee Members Scientific committee Prof. Valentina Emilia Balas Aurel Vlaicu University of Arad, Romania Prof. Toshio Fukuda Nagoya University, Japan Dr. Vincenzo Piuri University of Milan, Italy Dr. Ahmad Elngar Beni-Suef University, Egypt Dr. Malik Alazzam Lone Star College – Victory Center. Houston,TX, United States Dr. Osamah Ibrahim Khalaf Profesor, Al-Nahrain University, College of Information Engineering, Baghdad, Iraq Dr. TheyaznHassnHadi King Faisal University, Saudi Arabia Md Atiqur Rahman Ahad Osaka University, Japan,University of Dhaka, Bangladesh Prof. (Dr) Sanjay Nadkarni Director of Innovation and Research, The Emirates Academy of Hospitality Management, Dubai, UAE Dr. Ghaida Muttashar Abdulsahib Department of Computer Engineering, University of Technology, Baghdad, Iraq Dr. R. John Martin Assistant Professor, School of Computer Science and Information Technology, Jazan University Dr. Mohit Vij Associate Professor, Liwa College of Technology, Abu Dhabi, United Arab Emirates Dr. Syed MD Faisal Ali khan Lecture & Head – DSU, CBA, Jazan University Dr. Dilbag Singh Research Professor, School of Electrical Engineering and Computer Science, Gwangju Institute of Science and Technology, South Korea Dr. S B Goyal Dean & Director, Faculty of Information Technology, City University, Malaysia Dr. Shakhzod Suvanov Faculty of Digital Technologies, Department of Mathematical Modeling, Samarkand State University, Samarkand Uzbekistan xxi Artificial Intelligence, Blockchain, Computing and Security – Dagur et al. (Eds) © 2024 The Editor(s), ISBN: 978-1-032-49393-0
- 27. Dr. Upasana G Singh University of KwaZulu-Natal, South Africa Dr. Ouissem Ben Fredj ISSAT, University of Kairouan, Tunisia Dr. Ahmad Elngar Beni-Suef University, Egypt Dr. Omar Cheikhrouhou CES Lab, ENIS, University of Sfax, Tunisia Dr. Gordon Hunter Associate Professor, Mekelle University, Kingston University, UK Dr. Lalit Garg Computer Information Systems, Faculty of Information & Communication Technology, University of Malta, Malta Dr. Sanjeevi Kumar Padmanaban Aarhus University, Denmark Prof (Dr.) Alex Khang Professor of Information Technology, AI Expert and Data Scientist, GRITEx VUST SEFIX EDXOPS, Vietnam and USA Dr. Jiangtao Xi 1st degree connection 1st, Professor, Head of School of Electrical, Computer and Telecommunications Engineering at University of Wollongong, Greater Sydney Dr. Rabiul Islam Senior Lecturer at University of Wollongong, Australia Prof. Lambros Lambrinos Cyprus University of Technology, Cyprus Dr. Xiao-Zhi Gao University of Eastern Finland, Finland Dr. Sandeep Singh Sanger University of Copenhagen Dr. Mohamed Elhoseny University of Sharjah, United Arab Emirates Dr. Vincenzo Piuri University of Milan, Italy xxii
- 28. National Advisory Committee Dr. S. N. Singh, IIT Kanpur Dr. Rajeev Tripathi, MNNIT, Allahabad Dr. R. S. Yadav, MNNIT Allahabad Dr. Satish Chand, JNU, New Delhi Dr. M. N. Doja, IIIT Sonepat Dr. Bashir Alam, JMI, New Delhi Dr. Shailesh Tiwari, KEC, Ghaziabad Dr. Mansaf Alam, JMI, New Delhi Dr. Ompal, DST, New Delhi Dr. Rajeev Kumar, DTU, New Delhi Dr. Parma Nand, Sharda University, India Dr. Pavan Kumar Mishra, NIT Raipur Dr. Nagendra Pratap Singh, NIT Hamirpur Dr. Santarpal Singh, Thapar University Dr. Samayveer Singh, NIT Jalandhar Dr. Ankur Chaudhary, Sharda University Dr. Ranvijay, NIT Allahabad Dr. Manu Vardhan, NIT Raipur Dr. Pramod Yadav, NIT Srinagar Dr. Vinit Kumar, GCET Gr. Noida Dr. Anoop Kumar Patel, NIT Kurukshetra Dr. Suyash Kumar, DU Delhi Dr. Hitendra Garg, GLA University Mathura Dr. Chanchal Kumar, JMI New Delhi Dr. Vivek Sharma, GLBITM, Gr. Noida Dr. Anand Prakesh Shukla, DTE, UP Dr. Biru Rajak, MNNIT Allahabad Dr. Gopal Singh Kushwaha, Bhopal Dr. Rajeev Pandey, SRMS Brailly Dr. D. Pandey, KIET Ghaziabad Dr. D.S. Kushwaha, MNNIT Allahabad Dr. Sarsij Tripathi, MNNIT Allahabad Dr. Shivendra Shivani, Thapar University, Punjab Dr. Divakar Yadav, NIT Hamirpur Dr. Pradeep Kumar, NIT Kurukshetra Dr. Anand Sharma, AIT Aligarh Dr. Udai Pratap Rao, SVNIT Surat Dr. Vikram Bali, JSSATE Noida Dr. Gaurav Dubey, Amity University, Noida xxiii Artificial Intelligence, Blockchain, Computing and Security – Dagur et al. (Eds) © 2024 The Editor(s), ISBN: 978-1-032-49393-0
- 30. Organizing committee Chief Patron Shri Suneel Galgotia, Chancellor, Galgotias University, Greater Noida, India Patrons Shri Dhruv Galgotia, CEO, Galgotias University, Greater Noida, India Prof.(Dr.) Mallikharjuna Babu Kayala, Vice-Chancellor, Galgotias University, Greater Noida, India Ms. Aradhna Galgotia, Director Operations, Galgotias University, Greater Noida, India General Chairs Prof. (Dr.) Avadhesh Kumar, Pro-VC, Galgotias University, Greater Noida, India Prof. (Dr.) Munish Sabharwal, Dean, SCSE, Galgotias University, Greater Noida, India Conference Chairs Prof. (Dr.) Arvind Dagur, Professor, Galgotias University, Greater Noida, India Dr. Karan Singh, Professor, JNU New Delhi, India Dr. Pawan Singh Mehra, DTU, New Delhi Conference Co-Chairs Prof. (Dr.) Dr. Amit Kumar Goel, HOD (CSE) and Professor, Galgotias University, Greater Noida, India Prof. (Dr.) Krishan Kant Agarwal, Professor, Galgotias University, Greater Noida, India Dr. Dhirendra Kumar Shukla, Associate Professor, Galgotias University, Greater Noida, India Organizing Chairs Dr. Abdul Aleem, Associate Professor, Galgotias University, Greater Noida, India Dr. Vikash Kumar Mishra, Assistant Professor, Galgotias University, Greater Noida, India Technical Program Chairs Dr. Shiv Kumar Verma, Professor, SCSE, Galgotias University Dr. SPS Chauhan, Professor, SCSE, Galgotias University Dr. Ganga Sharma, Professor, SCSE, Galgotias, University Dr. Anshu Kumar Dwivedi, Professor, BIT, Gorakhpur Finance Chair Dr. Aanjey Mani Tripathi, Associate Professor, Galgotias University Dr. Dhirendra Kumar Shukla, Associate Professor, Galgotias University xxv Artificial Intelligence, Blockchain, Computing and Security – Dagur et al. (Eds) © 2024 The Editor(s), ISBN: 978-1-032-49393-0
- 31. Conference Organizing Committee Dr. Gambhir Singh, Professor, Galgotias University Dr. Arvinda Kushwaha, Professor, ABESIT, Ghaziabad Dr. Sanjeev Kumar Prasad, Professor, SCSE, Galgotias University Dr. Sampath Kumar K, Professor, Galgotias University Dr. Vimal Kumar, Associate Professor, Galgotias University Dr. T. Ganesh Kumar, Associate Professor, Galgotias University Dr. Atul Kumar Singh, Assistant Professor, Galgotias University Dr. Anuj Kumar Singh, Assistant Professor, Galgotias University Media and Publicity Chairs Dr. Ajay Shanker Singh, Professor, Galgotias University Dr. Ajeet Kumar, Professor, Galgotias University Dr. Santosh Srivastava, Professor, Galgotias University Cultural Program Chairs Ms. Garima Pandey, Assistant Professor, Galgotias University Ms. Heena Khera, Assistant Professor, Galgotias University Ms. Ambika Gupta, Assistant Professor, Galgotias University Ms Kimmi Gupta, Assistant Professor, Galgotias University xxvi
- 34. An ensemble learning approach for large scale birds species classification Harsh Vardhan, Aryan Verma & Nagendra Pratap Singh Department of Computer Science and Engineering, National Institute of Technology, Hamirpur ABSTRACT: Birds are vertebrate animals that are adapted for flight due to the presence of hollow bone structures. The entire population of birds contributes 0.08 % to the total animal biomass. In the past two decades, there has been a continuous loss and degradation of natural habitats resulting in a threat to bird population survival. The United Nations calculates that 49% of the bird population is declining, and some 1500 species have already gone extinct in the last 100 years. Researchers are studying the behavior and morphological characteristics of different bird species to understand them so that necessary steps can be taken for their protection. It is evinced that manually classifying bird species is a very inefficient and time-consuming task. Through the use of Automatic Bird Species Classification, this time can be reduced from hours to minutes. This paper features an automatic bird species classification system utilizing an ensemble of deep neural networks. Our proposed method trains individual state-of-the-art architectures like VGG 19, DenseNet 201, and ViT to classify 400 bird species. Further, the performance of these models is evaluated using metrics like F1 scores, precision, and recall. Our developed ensemble is better generalized and adapted to the problem with excellent accuracy of 99.40%. Results have stated that our approach is notably much better than existing works on bird species classification. Keywords: Ensemble Learning, Birds Species Classification, Image Classification, Deep Learning, Pretrained Model: DenseNet-201, VGG-19, ViT 1 INTRODUCTION Birds are members of classes aves; their feathers distinguish them from other classes. According to evolution theory, birds evolved from dinosaurs (Brusatte et al. 2015) They are a crucial member of the ecosystem due to their vital role-playing in functioning as natural pollinators, maintaining ecological balance, and keeping the pest population under control (Sekercioglu et al. 2016) Moreover, birds act as essential indicators for studying the state of the environment due to their susceptibility towards habitat change and the fact that they are accessible for census. These features make them an ecologist’s favorite tool. According to a report (Lehikoinen et al. 2019) One in every seven birds is under threat of extinction. A recent study (Pimm et al. 2018) highlights that there are more than 10400 living species of birds at present on this entire planet. In the past decade, bird populations had been severely affected due to many factors, such as global warming, deforestation, and the spreading of the communication network. Taking into concern, much research on wildlife bird monitoring has taken up the pace, and lots of government and semi-government programs have been initiated to protect the bird population. For this task, advanced technologies such as AI and IoT are aiding researchers in protecting the bird population. Authors in (Huang & Basanta 2019) have recog- nized endemic bird species through the deep learning algorithm CNN with skip connections. Authors of (Tóth & Czeba 2016) have used a convolutional neural network-based approach to classify birds’ songs in a noisy environment. Researchers in (Gavali & Banu DOI: 10.1201/9781003393580-1 3 Artificial Intelligence, Blockchain, Computing and Security – Dagur et al. (Eds) © 2024 The Author(s), ISBN: 978-1-032-49393-0
- 35. 2020) have combined DCNN and GoogleNet to classify bird species. Advanced Technologies are helping in a task such as classifying birds, monitoring the migratory birds’ status, establishing the pattern, conserving endangered species etc. 1.1 Main contributions l The ensemble model is developed that can classify 400 bird species by training the pre- trained networks like DenseNet-201, VGG-19, and VIT. l The dataset is cleaned and precisely pre-processed to remove the excessive noisy images that cause huge loss in feature extraction. l Extensive performance evaluation on a wide range of parameters like accuracy, F1 score, precision, and recall is performed for drawing comparison between models. The rest of our paper draws out the following structure. Related work is highlighted in section 2, the proposed methodology is defined in section 3, section 4 explains materials and methods, section 5 covers results and discussion, and finally section 6 includes comparative analysis section, the conclusion and future work part is covered under section 7, conflict of interest statement is in section 8 along with funding status in 9 and finally references are covered in last section. 2 RELATED WORK Traditionally bird classification is performed by hand-picking the features after physically exam- ining the bird image. Specifying bird species through physical examination requires excellent experience, which only expert ornithologists possess. Further, this task is very time-consuming and prone to numerous errors. Today, the biggest challenge for researchers in studying birds is that multiple species appear similar in initial appearance, which causes a delay in further research examination. Even expert ornithologists have limited study and exposure to rare bird species throughout their career. Much research has been done to solve this problem, and many papers have been published employing different tools and technologies. A technique for automating bird classification with the help of CNN has been proposed in (Gavali & Banu 2020) by employing DCNN(Deep Convolutional Neural Network) on Google Net framework. This system works by converting bird images into a grayscale format through autograph technology. A transfer learning approach has been presented in (Kumar & Das 2018) in which training is done using a multistage process and an ensemble model was formed, which consists of Inception Nets and Inception Res- Nets from localization. An existing VGG 16 architecture was implemented in (Islam et al. 2019) to extract the features for initiating bird species classification. In this paper comparison was made between different classification approaches such as Random Forest, K-nearest neighbor, and SVM. (Huang & Basanta 2021) Developed a new Inception ResNet v2-based transfer learning method to detect and classify endemic bird species. Their technique involves swapping missed classified data between training and test sets and then implementing it to validate the model performance. A comparison of existing recurrent convolutional networks for large-scale bird classification on acoustics has been drawn in (Gupta et al. 2021) it examines hybrid modeling approach that includes CNN and RCNN. An approach using a practical classification of bird Figure 1. Sample Images from data set. 4
- 36. species by transfer learning was implemented in (Alswaitti et al. 2022) this paper assesses the performance of traditional machine learning and deep learning by forming comparison between different groups of classifiers. 3 PROPOSED METHODOLOGY Our methodology highlights the use of a pre-trained deep neural network for automatic feature extraction. Let us discuss their structure to understand more about them. 3.1 Deep neural network Deep neural networks (Samek et al. 2021) are developed by stacking up more than two neural networks. This neural network automatically extracts relevant features for the clas- sification task. Our proposed ensemble learning model consists of VGG-19, DenseNet 201, and ViT. These three models are entirely different in their architecture and working. 3.1.1 VGG-19 VGG-19 is a variation of VGG-16 that consists of 19 layers instead of the standard 16 layers. Out of which (16 layers are convolution, three are fully connected, 5 are MaxPooling layers, and the remaining single layer is the Softmax function layer. Derived from Alex Net (Alom et al. 2018) this model improvised the traditional convolution neural network by a relatively large extent. It takes an image of size 224* 224 as input along with 3 * 3 kernel size. We trained it on our dataset by taking pre-trained weights of ImageNet itself. (Deng et al. 2009) 3.1.2 DenseNet-201 DenseNet-201 (Huang et al. 2017) works by connecting every layer with the other one in the network. It is often characterized as a Densely Connected Convolutional Network (Zhu & Newsam 2017) It uses transition layers between the DenseNet blocks. The transition layers consist of a batch-norm layer, then a 1x1 convolution layer, followed by a 2x2 average pooling layer. DenseNet architecture aims to make the connection between input and output layers deeper but shorter because it governs more accessible training and better feature extraction. Our method utilizes DenseNet-201 because it is much more efficient and easier to train than its other versions. 3.1.3 VIT transformer Vision Transformer applies a transformer (Cho et al. 2014) based architecture over the image patches. It splits the image into fixed-sized patches, then connects it with a transformer-encoder stacked with multi-layer perceptron (MLP), layer norm (LN), and multi-headed self-attention layer. The resultant self-attention layer is implemented to spread out the information globally. To perform the image classification, it uses a standard approach consisting of an extra learnable ”classification token” with the initial sequence. 3.2 Ensemble learning The ensemble learning approach combines the performance of several other models to generate one optimal model. It explores a predictive model that is supposed to perform better than any constituting predictive model alone. Technically there are numerous ways to generate an ensemble model, but three main famous classes of ensemble learning are bag- ging, boosting, and stacking. Through the combination of models, many benefits are acquired, such as improved predictive accuracy, precision, recall, and better statistics. In our work, we have sketched a complex voting-based ensemble learning model that combines the prediction of each class label with the predicted class label having the most votes. The hard 5
- 37. voting is outlined in the mathematical equation (1). 1 xi W ¼ fx1; x2:::::::::::xng (1) (a) Lets specify no. of iterations by integer T (b) Now randomly drawing F percent of V by taking itself replica of T (c) Represent weak-learn with Vt and receive the hypothesis(classifier) ht Now simply evaluate the ensemble E E ¼ fh1; h2:::::::::::hT gon x (2) Let at; j ¼ fh1; h2:::::::::::hT gon x (3) 4 MATERIAL AND METHODS 4.1 Dataset This paper utilized the dataset taken from Kaggle 400 Species Image Classification for testing and training the effectiveness of the developed model. The dataset consists of more than 50000 different images of birds separated into 400 species. This dataset comes from a public source that is con- stantly being updated. The moment we used this dataset for our research, it consisted of 400 classes (this may or may not be the same afterwards). Each image is of 224 x 224 size format. The dataset is already pre-divided into the test, train, and validation sets. During the data cleaning step, the images with excessive noise are removed due to their influence on extracting relevant features. 4.2 Tools used We have employed NVIDIA GPU (Tesla P100 16GB HBM2), Python 3.9, and PyTorch 1.12.1 for training our model. However, the actual implementation of neural networks was done using PyTorch 1.12.1. The complete training and testing procedures are done entirely on the Google Colab platform. This platform features free GPU access for max 12 straight hours. The pre-trained weights of the model are imported from the torchvision models package. 5 RESULT AND DISCUSSION This section illustrates a detailed analysis and discussion of the obtained results of our experiments performed on Kaggle Dataset(Wah et al. 2011) using the optimal ensemble learning model of three different neural networks VGG 19, DenseNet201, and ViT. Based on Table 1, it is proffered that the ViT outperformed other models on both the training and test sets. It achieved an accuracy of 99.75 and 99.40 on the training and test sets, respectively. To examine the overall characteristics of the proposed method result, a detailed exploration of metrics like precision, recall, and accuracy are also expressed in the Table 2. All models are Table 1. Summary of performance of different models used for birds classification in this work. Model Train Accuracy Train Loss Test Accuracy Test Loss VGG 19 75.27 0.912 90.50 0.350 DenseNet 201 93.52 0.303 97.20 0.124 ViT 99.61 0.008 99.50 0.016 Ensemble Model 99.75 0.005 99.40 0.012 6
- 38. trained for ten epochs. Hyper-parameters like Adam Optimizer, Cross Entropy Loss function, and Learning rate 103 are kept constant during the entire training process. The average time taken for one epoch completion is 550s for VGG 19. Similarly, it is 480 for DenseNet 201, and for ViT, it is 615s. In the end, the model is saved in h5 format for further research examination. 6 COMPARATIVE ANALYSIS In this section, we have presented a comparison between the classification result of our approach with other approaches used for classifying bird species. Table 3 contrasts bird classification results comparison between our approach and existing one. In (Gavali Banu 2020) authors have applied an approach to convert the bird image into an autograph from the grayscale format, then examined each autograph to calculate the score of a particular bird species. Authors of (Y.-P Huang Basanta 2019) have implemented a skip connection-based CNN network to improve feature extraction accuracy. A novel method is proposed in the paper (Marini et al. 2013) that extracts colored features from unconstrained images. Table 3, simply corroborates that our method is significantly much more accurate than other developed classification models. 7 CONCLUSION AND FUTURE WORK This paper advances the potential of deep learning and ensemble learning for automatic bird species classification. This paper illustrated an optimal ensemble (Sagi Rokach 2018) model from individual trained networks, i.e., VGG-19-bn, Dense Net 201, and ViT. The experimental results derive the f1 score from being (90.50,97.20,99.50,99.40) for VGG 19, Dense Net 201, ViT Transformer, and Hard Voting Ensemble, respectively. This ensemble learning technique is a promising approach for automatic bird species classification. We plan to implement data augmentation to increase the training dataset size in the future. Also, some aspects of the deep neural network and its underlying filters are expected to be mod- ified to achieve much better performance, reducing the time and cost outlays. CONFLICT OF INTEREST STATEMENT The authors declare no conflict of interest. Table 2. Performance of the models on some evaluation metrics. Technique Precision Recall F1 Score VGG 19 0.95 0.85 0.90 DenseNet 201 0.94 0.90 0.97 ViT 0.99 0.98 0.99 Ensemble Model 0.99 0.99 0.99 Table 3. Comparison of accuracy achieved using different approaches for birds classification. Technique Classification Method No. of Classes Accuracy (Gavali Banu 2020) Deep CNN (Google Net) 200 88.33 (Huang Basanta 2019) Skip Connections CNN 27 99.00 (Marini et al. 2013) Colour + Segmentation 200 90.00 Proposed Method Ensemble Approach 400 99.40 7
- 39. FUNDING STATUS The author states that no funding was received for this work. REFERENCES Alom, Md Zahangir, Tarek M Taha, Christopher Yakopcic, Stefan Westberg, Paheding Sidike, Mst Shamima Nasrin, Brian C Van Esesn, Abdul A S Awwal, and Vijayan K Asari. 2018. “The History Began From Alexnet: A Comprehensive Survey on Deep Learning Approaches.” arXiv preprint arXiv:1803.01164. Alswaitti, Mohammed, Liao Zihao, Waleed Alomoush, Ayat Alrosan, and Khalid Alissa. 2022. “Effective Classification of Birds’ Species Based on Transfer Learning.” International Journal of Electrical Computer Engineering (2088-8708) 12 (4) Brusatte, Stephen L Jingmai K O’Connor, and Erich D Jarvis. 2015. “The Origin and Diversification of Birds.” Current Biology 25 (19) R888–R898. Cho, Kyunghyun, Bart Van Merrienboer, Dzmitry Bahdanau, and Yoshua Bengio. 2014. “On The Properties of Neural Machine Translation: Encoder-decoder Approaches.” arXiv preprint arXiv:1409.1259. Deng, Jia, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. 2009. “Imagenet: A Large-scale Hierarchical Image Database.” In 2009 IEEE Conference on Computer Vision and Pattern Recognition, 248–255. IEEE. Gavali, Pralhad, and J Saira Banu. 2020. “Bird Species Identification Using Deep Learning on GPU Platform.” In 2020 International Conference on Emerging Trends in Information Technology and Engineering (ic-ETITE) 1–6. IEEE Gupta, Gaurav, Meghana Kshirsagar, Ming Zhong, Shahrzad Gholami, and Juan Lavista Ferres. 2021. “Comparing Recurrent Convolutional Neural Networks for Large Scale Bird Species Classification.” Scientific Reports 11 (1) 1–12. Huang, Gao, Zhuang Liu, Laurens Van Der Maaten, and Kilian Q Weinberger. 2017. “Densely Connected Convolutional Networks.” In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 4700–4708. Huang, Yo Ping, and Haobijam Basanta. 2021. “Recognition of Endemic Bird Species Using Deep Learning Models.” IEEE Access 9:102975–102984. Huang, Yo-Ping, and Haobijam Basanta. 2019. “Bird Image Retrieval and Recognition Using a Deep Learning Platform.” IEEE Access 7:66980–66989. Islam, Shazzadul, Sabit Ibn Ali Khan, Md Minhazul Abedin, Khan Mohammad Habibullah, and Amit Kumar Das. 2019. “Bird Species Classification from An Image using VGG-16 network.” In Proceedings of the 2019 7th International Conference on Computer and Communications Management, 38–42. Kumar, A., Alam, B. (2016). Real-Time Fault Tolerance Task Scheduling Algorithm with Minimum Energy Consumption. In Proceedings of the Second International Conference on Computer and Communication Technologies: IC3T 2015, Volume 2 (pp. 441–448). Springer India. Kumar, Akash, and Sourya Dipta Das. 2018. “Bird Species Classification Using Transfer Learning with Multistage Training.” In Workshop on Computer Vision Applications, 28–38. Springer. Lehikoinen, Aleksi, Lluıs Brotons, John Calladine, Tommaso Campedelli, Virginia Escandell, Jiri Flousek, Christoph Grueneberg, Fredrik Haas, Sarah Harris, Sergi Herrando, et al. 2019. “Declining Population Trends of European Mountain Birds.” Global Change Biology 25 (2) 577–588. Marini, Andréia, Jacques Facon, and Alessandro L Koerich. 2013. “Bird Species Classification Based on Color Features.” In 2013 IEEE International Conference on Systems, Man, and Cybernetics, 4336–4341. IEEE Mehra, P. S., Mehra, Y. B., Dagur, A., Dwivedi, A. K., Doja, M. N., Jamshed, A. (2021). COVID-19 Suspected Person Detection and Identification Using Thermal Imaging-based Closed Circuit Television Camera and Tracking Using Drone in Internet of Things. International Journal of Computer Applications in Technology, 66(3–4), 340–349. Pimm, Stuart L Clinton N Jenkins, and Binbin V Li. 2018. “How to Protect Half of Earth to Ensure it Protects Sufficient Biodiversity.” Science Advances 4 (8) eaat2616. Sagi, Omer, and Lior Rokach. 2018. “Ensemble Learning: A Survey.” Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery 8 (4) e1249. Tóth, Bálint Pál, and Bálint Czeba. 2016. “Convolutional Neural Networks for Large-Scale Bird Song Classification in Noisy Environment.” In CLEF (Working Notes), 560–568. Wah, Catherine, Steve Branson, Peter Welinder, Pietro Perona, and Serge Belongie. 2011. “The Caltech-ucsd Birds-200-2011 Dataset.” Zhu, Yi, and Shawn Newsam. 2017. “Densenet for Dense Flow.” In 2017 IEEE International Conference on Image Processing (ICIP) 790–794. IEEE 8
- 40. Comprehensive analysis of human action recognition and object detection in aerial environments Mrugendrasinh Rahevar Chandubhai S. Patel Institute of Technology, Charotar University of Science and Technology, Changa, Anand, Gujarat, India Amit Ganatra Parul Univerity, Vadodara, India Hiren Mewada Prince Mohammad Bin Fahd University, Al Khobar, Kingdom of Saudi Arabia Krunal Maheriya Chandubhai S. Patel Institute of Technology, Charotar University of Science and Technology, Changa, Anand, Gujarat, India ABSTRACT: Drone-based aerial view analysis is the newly emerging technique helpful in topological, regional analysis and interpretation of objects and features. Due to its relevance to environment monitoring, Human action recognition (HAR) and Object detection (OD) from aerial view to search and rescue is the technical challenges. They are difficulties owing to diverse views, the tiny size of persons and objects, and involved constraints in processing. The deep learning models are proven accurate in image and video processing applications. However, the impact of Drone ego-motion on object identification, human activity detection, and crowded backgrounds may weaken the deep learning applicability in aerial view analysis. This assessment establishes current trends and progress in HAR and OD. Initially, the study on various datasets, including UCF-ARG, Okutama, BirdsEye View, and DOTA, is presented. Then, a summary and comparative analysis of various areal perspective algorithms are discussed. Finally, chal- lenges and new directions in aerial view-based HAR and OD are discussed in depth. 1 INTRODUCTION HAR and OD are computer vision applications where HAR identifies human behavior, for example, running, walking, fighting, sky-diving, etc. And OD is a technique to locate an object’s instance and provide its labeling like trucks, cars, animals, humans, etc. Identifying human actions and objects from the ground camera is easy as we can see objects and humans correctly. Still, it will be difficult in aerial view because of large variations in human body pose, differences in the appearance of interacted objects, occlusions, motions of cameras, and minimal size of the objects make it difficult to identify objects and recognizes human behavior. Also, there are fewer datasets in aerial view for both action recognition and OD. The spatiotemporal and motion aspects are the most significant to recognition actions since they influence the learning of spatial-temporal repre- sentations to comprehend the category of an action class. The spatiotemporal records the link between spatial information at distinct timestamps, whereas the motion captures features between surrounding frames. CNNs have demonstrated powerful effectiveness in collecting high level representational features in pictures unique to a given task. As a result of its versatility and high modeling capabilities, it has been widely used for picture classification (Chollet 2017; Jmour et al. 2018) tasks, allowing to learn spatial representations (Zuo et al. 2015) from visual data for tackling the challenge of human action detection. DOI: 10.1201/9781003393580-2 9 Artificial Intelligence, Blockchain, Computing and Security – Dagur et al. (Eds) © 2024 The Author(s), ISBN: 978-1-032-49393-0
- 41. In order to simulate human vision and cognition, OD focuses on techniques for recognizing multiple sorts of objects within a shared framework. Image classification has advanced significantly since launch ImageNet and offered AlexNet. By multiplying the layers in the network, demonstrated VGGNet, introduced GoogLeNet, ResNet Using a residual network in the design of its image categorization, which outperforms an average person by 3.57%. HAR has wider applications, such as driving assistance, sports analysis, and video surveillance. Many problems in jobs using computer vision, such as captioning images, and object tracking, rely on OD. Nowadays, drones are used widely for an assortment of intention, which may include search and rescue, sports analysis, agri- culture, and surveillance, because of their ability to capture wide areas and reach difficult arias. For example, we can use drones at the country’s border to identify any suspicious activity of terrorists and detect weapons. Such surveillance cannot be done through any human or ground camera, so using drones to identify objects and human behavior at the country border is easy. HAR and OD are important steps of search and rescue operations, their in-depth study and investigation supplement its development and better progress. This survey paper focuses on recent research to identify human actions and detected objects from an aerial view as well as which dataset is available for both studies, OD and recognition of human activities. In this paper, we focus on l Available Dataset: Discussion on an available dataset for both studies action recognition and OD in aerial view. l Recent research: Discussion of recent research of the past 4 to 5 years on HAR and OD in aerial view. l Application and challenges: What applications may be made for aerial view object iden- tification and HAR, and what problems will we have in identifying items and human activities from an aerial perspective. Figure 1. Number of researches done in both for OD and HAR aerial view perspective according to year. Figure 2. Methods for recognizing human actions and detecting objects in aerial views (best of our knowledge). 10
- 42. 2 RELATED WORK we first examine the datasets that are available for OD and HAR, and then we talk about recent OD and HAR research. Finally, challenges and applications are discussed. 2.1 Dataset for HAR in aerial view Some research on aerial view human action has been conducted, most of which utilizes the UCFARG dataset. This section will provide an overview of aerial view HAR datasets. UCG-ARG Dataset collection, which is a Multiview dataset of human activity, is made up of aerial, rooftop, and ground cameras from the University of Central Florida. The set if 10 actions, such as walking, throwing, digging, boxing, carrying, and clapping, were exe- cuted by 12 different performers forming the UCG-ARG team. The acts were recorded using three different cameras: one places on the ground, another at a height of 100 feet on a rooftop, and the third on the payload platform of a Kingfisher Aerostat helium balloon, which is 13 feet high. Each actor performed each action four times, except for the opening and closing of trunks, which were performed three times on three different parked cars. The footage was captured in high quality at a resolution of 1920 x 1080, with a frame rate of 60 frames per second.4 Okutoma Dataset (Barekatain et al. 2017) A video dataset is used to find concurrent human actions in aerial perspectives. There are 12 action classes in its 43-minute-long, fully annotated sequences, including human-to-human, human-to-object, and non-interaction. Several problems unique to OkutomaAction include dynamic action transitions, rapid camera movements, dramatic size and aspect ratio shifts, and characters with multiple labels. This dataset is more complicated than others as a result, and it will help advance the dis- cipline and enable practical applications. Game Action Dataset to gather game activity datasets, the games GTA-5 and FIFA are used (Sultani Shah 2021). Record the same action from several angles. Seven human behaviors—cycling, fighting, soccer kicking, running, walking, shooting, and skydiving— were recorded. The game’s kicking action is included, while GTA-5 is used for the other motions. For a total of 14000 footage over seven activities, the dataset consists of 200 movies (100 ground and 100 aerial) for each activity. The YouTube Aerial Dataset was compiled using drone footage clips sourced from YouTube. The dataset focuses on eight distinct activities, such as golf swinging, skate- boarding, horseback riding, kayaking, and surfing, among others. The aerial videos in the dataset were filmed at different altitudes and feature rapid and extensive camera movements. There are a total of 50 videos for each activity. The dataset is partitioned into three subsets: 60% of the videos are reserved for training, 10% for validation, and 30% for testing. Drone-Action dataset action recognition data for drones was collected using the Drone- Action dataset (Perera et al. 2019). The 13 actions include clapping, hitting with a bottle, hitting with a stick, jogging f/b, jogging f/b, kicking, punching, running side stabbing, walking side, and waving hands. 24 high definition video clips totaling 66,919 frames are Table 1. Most used dataset for aerial view HAR. Dataset No. of action classes Published year UCF-ARG 10 2008 Okutoma Action (Barekatain et al. 2017) 12 2017 Game Action Dataset 7 – YouTube Aerial 8 – Drone-Action (Perera et al. 2019) 13 2019 11
- 43. part of the collection. To capture as many human position details in a relatively good quality, the entire movie was taken at a low height and slowly. 2.2 Dataset of OD in aerial view This section provides current datasets that may be accessed and used for OD tasks in aerial views. Drones, Google Earth, and satellites are used to collect some datasets. BirdsEye View Dataset7 for Object Classification this dataset comprises 5000 photos, each thoroughly annotated according to the PASCAL VOC criteria. The dataset contains diverse situations for which they used different datasets such as UCF-ARG and PNNL Parking dataset and selected an appropriate (i.e., can be used for OD). Parking Lot, Action Test, Routine Life, Outdoor Living, Harbour, and Social Party are the scenes. Captures frames from over 70 films as well as photos from various scenes. DOTA More than 1,793,658 annotated object instances are included in this dataset, which is divided into 18 different categories8. These categories include airplanes, ships, tanks, baseball diamonds, tennis courts, basketball courts, ground track fields, harbors, bridges, large vehicles, helicopters, roundabout soccer fields, swimming pools, container cranes, airports, and helipads. Google Earth, the JL-1 and GF-2 satellites of the China Centre for Resources Satellite Data and Application, and other sources provided the images for this collection. Due to the vast amount of instances of an item, random orientations, many categories, a variety of aerial scenes, and a density distribution, DOTA is challenging. Nevertheless, DOTA’s features make it worthwhile for real-world applications. The UAVDT Benchmark dataset (Du et al. 2018) primarily concerns difficult challenging scenarios. It includes over 80,000 sample frames from 10 hours of raw video thoroughly labeled with bounding boxes. Introduce 14 characteristics for the core assignments in com- puter vision, namely identifying objects, tracking a single object, and tracking multiple objects, involve a range of factors such as weather conditions, altitude of flight, camera angle, classification of vehicles, and obstructions. Visdrone Dataset (Zhu et al. 2018) comprises of 10,209 pictures and 263 videos with annotated frames such as bounding boxes, item occlusion, truncation ratios, and classifica- tions, etc. 2.5 million annotations were found in 179,264 image/video frames. This dataset spans 14 distinct countries in China, from north to south. The dataset may be utilized for the following four tasks: single-object tracking, multi-object tracking, video detection, and image detection. 2.3 HAR in aerial view HAR from an aerial perspective is a tough issue; however, due to the increased usage of deep learning methods, numerous types of studies have been conducted in recent years. Mmekreki’s et al. (2021) research employed the pre-train YOLOv3 model, with re-searchers adjusting its configuration file to make the model compatible with identifying human actions from an aerial perspective. Different video frames are sent into the yolov3 model as input and a label text file. They achieve a high validation accuracy with the aid of this approach. Ketan Kotecha et al. Table 2. Most recent dataset of OD in aerial view. Dataset No of images/ Video clips Published year BirdsEye View (Qi et al. 2019) 5000 2019 DOTA (Xia et al. 2018) 1,793,658 2021 UAVDTBenchmark (Du et al. 2018) 80,000 2018 Visdrone (Zhu et al. 2018) 2.5 million images 2018 12
- 44. (2021) provides a solution that will deal with this issue. This approach first takes video as input, which is of this complexity level, and then takes an image frame as input. Faster motion feature modeling was utilized to identify persons. After identifying humans, accurate action recognition was used. SoftMax function was used for the final layer to classify human actions. Because this aerial view study is not widely explored, the dataset is insufficient. So, how will the model be trained in the absence of data? Sultani Shah (2021) proposes one model for this problem that uses GAN generated dataset from ground camera features. They begin by extracting the features of various datasets, including Game Aerial Videos, Aerial Videos dataset, and Ground Videos. After extracting features from ground videos, GAN Network will be utilized to build aerial features, and all the features will be fed into the Feed-Forward Network. The authors obtained an average validation accuracy using this approach. Authors (Mliki et al. 2020) separated the process of recognizing humans and human actions into two phases: offline and inference. In the offline phase model development is done for identifying humans/nonhumans and recognizing human actions from an aerial perspective. The datasets obtained using potential motion recognition are utilized to con- struct human/ non-human models. The inference phase is used to classify human actions into two categories: instant classification and entire classification. Instant classification classifies human activity frame by frame, whereas the whole classification produces an average of instant classification. With this approach, they achieve very promising accuracy detection and good accuracy for both instant classifications and entire classifications with this approach. EfficientDetD7 was used to identify humans with high accuracy, EfficientNetB7 to extract features, and LSTM to classify human actions. Get an average accuracy in recognizing activities under diverse conditions such as blurring, noise addition, lighting, and darkness. A fully autonomous UAV-based activity detection system based on aerial photography has been developed by Peng Razi (2020). It overcomes issues with aerial imaging tech- nologies like camera vibration and motion, small human size, and poor resolution. With this method, they were able to identify every video level with excellent accuracy. A Lightweight Action Recognition Method for Unmanned-Aerial-Vehicle Video (LARMUV) (Ding et al. 2020) was presented. The approach was built on a teacher-student network (TSN) and employed MobileNetv3’s backbone. Self-Attention was used to gather temporal information across many frames. 2.4 OD from aerial view We will have difficulty identifying anything from an aerial perspective due to the items’ diminutive size. Because of the restricted dataset, it will not be easy to recognize things from an aerial perspective, particularly the top view, and road view, an aerial view. As a result, we discovered issues challenging cases. To overcome this issue, Hong et al. (2019) presented a hard chip mining approach in patch-level augmentation for object recognition in an aerial view study. The first multiscale chip was designed to impart object-detecting knowledge. To create an object pool, they extract patches from the dataset in the second stage. To address the issue of class imbalance, these modifications will be included in the dataset. The final model is then trained after hard chips are formed from misclassified locations. However, after a calamity, like a flood or a tsunami, identifying items from an aerial perspective will be challenging. The main problem is recognizing and mapping things of interest in real time. (Pi et al. 2020) presented CNN detecting objects from an aerial view. These models were widely used for image classification tasks and could detect roofs, automobiles, and flooded areas. OD in aerial images is also difficult because pixel occupancy varies across varied object sizes, the non-uniform distribution of items in aerial photographs, variations in an object’s appearance due to different view angles and lighting conditions, and variations in the number of objects, and even when they are of the same type, across images (Chalavadi et al. 2022). Therefore, Chalavadi et al. proposed a hierarchical dilated convolutions operation 13
- 45. and developed a mSODANet network for multiscale object recognition in aerial images. They used parallel dilated convolutions to learn the context information of various sorts of objects at diverse sizes and fields of vision. As a result, it helped to get visual information more efficiently and improved the model’s accuracy. We discovered that identifying a car from an aerial view image was more challenging than a ground view image due to the tiny vehicle size and complicated background. Michael Ying Yang employed a Focal Loss convolutional neural network (DFL-CNN) in vehicle recog- nition in aerial images (Yang et al. 2019) to recognize a vehicle from an aerial perspective. There are skip connections used in CNN structures to improve feature learning. In addition, the focal loss function is used in the region proposal network and the final classifier to replace the usual-cross-entropy loss function (Yang et al. 2019). Traffic, urban planning, defense, and agriculture all depend heavily on object identifica- tion, and convolutional neural network-based research is excellently detecting pictures. Still, high density, tiny object size, and complicated backdrop fundamental models are not per- forming well (Long et al. 2019). To appropriately identify things, Hao Long presented a method called feature Fusion Deep Network in his research on object recognition in high- altitude images using feature fusion deep learning (Long et al. 2019). The problem of positive and negative anchor boxes is solved by the horizontal key point-based object detector in the paper, oriented OD using boundary box-aware vectors in aerial images (Yi et al. 2021). To capture the oriented bounding boxes, they first determine the object’s center, and then they employ the determined center BBAVectors (Yi et al. 2021). Fuyan Lin’s study (Lin et al. 2020) improved the YOLOv3 model. To enhance the identification of tiny objects, the YOLOv3 model was updated by changing the anchor values and building the 4x down sampling prediction layer (Lin et al. 2020). 3 CHALLENGES AND APPLICATIONS 3.1 Challenges in HAR and OD in aerial view The first challenge is the human’s small size. The size of a human appears to be the smallest in the image from an aerial view. Because of their small size, humans cannot be seen properly by UAVs (Unmanned Aerial Vehicles), and also, we cannot see humans properly with our own eyes. Because we can’t see humans properly, human parts like legs and hands aren’t correctly identified, which makes it difficult to find human actions from an aerial view. Another problem is UAV camera motion; movies obtained by a UAV cannot be stabilized, and video stabilization is required to detect human movement reliably. From an aerial view, Human activities like walking and running appear practically identical. It will be impossible to distinguish that activity. Aerial view HAR research has fewer datasets available. Most analyses are based on the UCF-ARG dataset. Because the dataset is nearly 14 years old, the background environment may impair model accuracy. Furthermore, models of deep learning needs hundreds of films for training in human air action, and gathering a huge amount of action data is challenging. One of the challenges in distinguishing human actions from an overhead perspective is the style changes, variation in view, human changes, and changes in clothes, tracking complexity of various objects and identifying anomalies and aberrant crowd behavior. It is difficult to identify many persons’ actions from a single image. Objects seem different as humans from various perspectives such as top view, road view, and aerial view. Helicopters and unmanned aerial vehicles (UAVs) were employed to detect disaster damages. It is more challenging due to the tiny size of items from an aerial view. Sometimes a photograph obtained from a high height (i.e., an aerial view) has reduced pixel density or is blurry, making item identification harder. There is also a limited quantity of datasets available for aerial view object identification. Background clutter, diverse types of things, such as more than one object in one image, make detecting more than one object from one image difficult, especially from an aerial view perspective. 14
- 46. 3.2 Applications for human actions and OD in aerial view One of the most promising applications for OD and action recognition is a surveillance system. We can cover more ground area from an overhead view since we can see things and people from a higher height. We can, for example, conduct surveillance in a retail mall or a fair to detect suspicious behavior. While it is hard to trace terrorists’ movements from the ground, aerial views allow for simple detection of suspicious activities near borders utilizing drones (UAVs). We may use aerial views to identify damage caused by natural disasters such as earthquakes and tsunamis. Strange occurrences like a lone individual loitering, many people interacting (like fights and personal assaults), people interacting with vehicles (like vehicle injury), and people interacting with facilities or locations (e.g., Object left behind and trespassing). Use of HAR is to use surveillance systems to detect people walking. Using a double helical signature approach, Identify human walking activity in surveillance footage. Using DHS characteristics including human size, viewing different angles, camera motion, and extreme occlusion, crowded scenes may simultaneously separate people in frequent motion and identify body parts. Occlusion makes it difficult to see and count people in thick crowds. Yilmaz et al. (2006) conducted a thorough analysis of tracking techniques and divided them into groups based on the object and motion representations they employed. In addition, gender may be categorized using security cameras. Using patch characteristics to represent various body parts, Cao et al. (Cao et al. 2008) developed a part-based gender reverberation system that could accurately identify from a single frontal or rear shot, the gender picture. One of the main uses of HAR and OD is the identification of pedestrians and the prevention of falls in the elderly. 4 CONCLUSION This article discusses current research in human action identification and OD, some of the most extensively used and promising datasets in OD and HAR, and some of the issues we face when identifying objects and human activity from an aerial perspective. We are only focused on a current study from the last 5 to 8 years, thus our article contains re-cent research most promising for an aerial perspective in recent years. There are some challenges such as changes in human appearance, different objects, and changes in camera view which need to be resolved and are addressed in this survey. We also found that a broad range of applications exists, such as in the field of surveillance, where utilizing people for monitoring costs more in terms of time and money than using a UAV. Therefore, we can create a model that can accomplish the same thing, and the studies are ideal for that. Most studies use pre- trained models, RNNs, and LSTM-based models, but we now have a new, SOTA model called a transformers model that can produce results that are more effective because it incorporates a self-attention model. Through this research, we have also found that research on attention and this self-attention mechanism is not as prevalent as it could be. REFERENCES Barekatain, M., Martí, M., Shih, H.-F., Murray, S., Nakayama, K., Matsuo, Y., Prendinger, H. (2017). Okutama-action: An Aerial View Video Dataset for Concurrent Human Action Detection. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops, 28–35. Cao, L., Dikmen, M., Fu, Y., Huang, T. S. (2008). Gender Recognition From Body. Proceedings of the 16th ACM International Conference on Multimedia, 725–728. Chalavadi, V., Jeripothula, P., Datla, R., Ch, S. B., C, K. M. (2022). mSODANet: A Network for Multi- scale Object Detection in Aerial Images using Hierarchical Dilated Convolutions. Pattern Recognition, 126, 108548. https://doi.org/10.1016/j.patcog.2022.108548 15
- 47. Chollet, F. (2017). Xception: Deep Learning with Depthwise Separable Convolutions. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 1251–1258. Ding, M., Li, N., Song, Z., Zhang, R., Zhang, X., Zhou, H. (2020). A Lightweight Action Recognition Method for Unmanned-Aerial-Vehicle Video. 2020 IEEE 3rd International Conference on Electronics and Communication Engineering (ICECE), 181–185. Du, D., Qi, Y., Yu, H., Yang, Y., Duan, K., Li, G., Zhang, W., Huang, Q., Tian, Q. (2018). The Unmanned Aerial Vehicle Benchmark: Object Detection and Tracking. Proceedings of the European Conference on Computer Vision (ECCV), 370–386. Hong, S., Kang, S., Cho, D. (2019). Patch-level Augmentation for Object Detection in Aerial images. Proceedings of the IEEE/CVF International Conference on Computer Vision Workshops, 0–0. Jmour, N., Zayen, S., Abdelkrim, A. (2018). Convolutional Neural Networks for Image Classification. 2018 International Conference on Advanced Systems and Electric Technologies (IC_ASET), 397–402. Kotecha, K., Garg, D., Mishra, B., Narang, P., Mishra, V. K. (2021). Background Invariant Faster Motion Modeling for Drone Action Recognition. Drones, 5(3), 87. Lin, F., Zheng, X., Wu, Q. (2020). Small Object Detection in Aerial View Based on Improved YoloV3 Neural Network. 2020 IEEE International Conference on Advances in Electrical Engineering and Computer Applications (AEECA), 522–525. Long, H., Chung, Y., Liu, Z., Bu, S. (2019). Object Detection in Aerial Images Using Feature Fusion Deep Networks. IEEE Access, 7, 30980–30990. Mliki, H., Bouhlel, F., Hammami, M. (2020). Human Activity Recognition from UAV-captured Video Sequences. Pattern Recognition, 100, 107140. Mmereki, W., Jamisola, R. S., Mpoeleng, D., Petso, T. (2021). YOLOv3-based Human Activity Recognition as Viewed from a Moving High-altitude Aerial Camera. 2021 7th International Conference on Automation, Robotics and Applications (ICARA), 241–246. Peng, H., Razi, A. (2020). Fully Autonomous UAV-based Action Recognition System using Aerial Imagery. International Symposium on Visual Computing, 276–290. Perera, A. G., Law, Y. W., Chahl, J. (2019). Drone-action: An Outdoor Recorded Drone Video Dataset for Action Recognition. Drones, 3(4), 82. Pi, Y., Nath, N. D., Behzadan, A. H. (2020). Convolutional Neural Networks for Object Detection in Aerial Imagery for Disaster Response and Recovery. Advanced Engineering Informatics, 43, 101009. Qi, Y., Wang, D., Xie, J., Lu, K., Wan, Y., Fu, S. (2019). Birdseyeview: Aerial View Dataset for Object Classification and Detection. 2019 IEEE Globecom Workshops (GC Wkshps), 1–6. Sultani, W., Shah, M. (2021). Human Action Recognition in Drone Videos Using a Few Aerial Training Examples. Computer Vision and Image Understanding, 206, 103186. Xia, G.-S., Bai, X., Ding, J., Zhu, Z., Belongie, S., Luo, J., Datcu, M., Pelillo, M., Zhang, L. (2018). DOTA: A Large-scale Dataset for Object Detection in Aerial Images. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 3974–3983. Kumar, A., Yadav, R. S., Ranvijay, A. J. (2011). Fault Tolerance in Real Time Distributed system. International Journal on Computer Science and Engineering, 3(2), 933–939. Tripathi, A. M., Singh, A. K., Kumar, A. (2012). Information and Communication Technology for rural Development. International Journal on Computer Science and Engineering, 4(5), 824. Kumar, A., Mehra, P. S., Gupta, G., Jamshed, A. (2012). Modified Block Playfair Cipher using Random Shift Key Generation. International Journal of Computer Applications, 58(5). Zhu, P., Wen, L., Du, D., Bian, X., Ling, H., Hu, Q., Nie, Q., Cheng, H., Liu, C., Liu, X. (2018). Visdrone- det2018: The Vision Meets Drone Object Detection in Image Challenge Results. Proceedings of the European Conference on Computer Vision (ECCV) Workshops, 0–0. Zuo, Z., Shuai, B., Wang, G., Liu, X., Wang, X., Wang, B., Chen, Y. (2015). Convolutional Recurrent Neural Networks: Learning Spatial Dependencies for Image Representation. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops, 18–26. 16
- 48. Brain tumour detection using deep neural network via MRI images Shadmaan, Rajat Panwar, Prajwal Kanaujia, Kushal Gautam, Sur Singh Rawat Vimal Gupta Department of Computer Science and Engineering, JSS Academy of Technical Education, Noida ABSTRACT: The brain’s own aberrant and unregulated cell division is what causes brain tumors. The patient cannot heal if the growth increases by more than 50%. the identification of Brain tumor diagnosis must be swift and precise. The capture of a brain MRI scan is the initial step, after which digital imaging techniques are used to determine the precise position and size of the tumor. Gray and white matter make up MRI pictures, and the tumor- containing area is more intense. In order to enhance the given brain MRI scan, noise filters are initially utilized to remove background noise. This study aims to give a comprehensive review on detection of brain tumors. 1 INTRODUCTION Cell proliferation that is not under control leads to tumours. Because they don’t penetrate the tissues around them, these tumours could only grow in smaller areas. However, if these tumours grow close to a vital location, they may pose a risk. Malignant cancers, on the other hand, can evolve and spread in such a way that they ultimately create a fatal kind of cancer. MRI is recommended over other medical imaging modalities because it yields the most contrast images of brain tumours. In instance, it has been demonstrated in numerous studies that the transfer learning technique improves classification performance on the target dataset by applying the knowledge acquired from one task to another that is similar [1,2]. A deep convolutional neural network (DCNN) model must often be trained using a huge dataset, which includes a high level of computational complexity. 2 LITERATURE REVIEW Using MR tests, numerous researchers created a number of procedures algorithms, and tactics to identify brain tumours, strokes, and other forms of variations in the human brain. Brain Tumor Identification and Segmentation [4–6] outline methods for the detection of brain tumours including segmentation, histograms, thresholding and morphology. Fuzzy C means (FCM) does a good job of precisely segmenting tumour tissue. Svm [11,12] was used to recognise segmentation.Utilising learning algorithm, fundamental com- ponent analysis, as well as the wavelet transform, a hybrid technique is shown in [13,14] algorithms, wherein the accuracy of brain tumour detection is attained 98.6%. Three multi-resolution images from [17,18] includes the various methods. Here, the sug- gested study asserted thatit had achieved an accuracy of 96.05%.The following approaches are presented in the Table 1 for detecting structural abnormalities, including tumours. DOI: 10.1201/9781003393580-3 17 Artificial Intelligence, Blockchain, Computing and Security – Dagur et al. (Eds) © 2024 The Author(s), ISBN: 978-1-032-49393-0
- 49. 3 WORKING OF THE MODEL An executive control when processing medical images is the convolutional neural network. A model of CNN as shown in the above figure is a type of machine learning used for image analysis that focuses on learning technique component knowledge. The image processing stage of this study comprises a variety of operations. Table 1. Descriptions of some brain tumor detection algorithms. Dataset Method for Extracting Features Classification Method Accuracy 70 MR Images Hybrid Technique Discrete Wavelet Transformation Artificial Neural Network with Feedforward Backpropagation Algorithm 97% T2 weighted 255 MRI images Transform(DWPT), Shanon Entropy(SE) and Tsallis Entropy(TE) Generalized Eigenvalue Proximate Support Vector Machine (GEPSVM) 99.61% 1800 MRI Images CNN CNN 98.60% 239 MRI Images (SGLD) ANN 99% T2 weighted brain MR images. Dataset: 66 and Dataset: 255 Wavelet Transform Curvelet Transform and Shearlet Transform. Support Vector Machine and Particle Swarm Optimization 97.38% 253 MRI Images Hyper Colum Attention Module and residual block. CNN 96.05% 500 MR Images Fully Automatic Heterogeneous Segmentation(FAHS). SVM 98.51% 253 MR Images Improved ResNet50 CNN 97.01% 253 MR Images Deep CNN Deep CNN 98% 250 MR Images DenseNet-169 Multiple Classifier Ensemble Multiple Classifier Ensemble Multiple Classifier Ensemble 92.37% 3000 MR Images ResNext-101 Ensemble of multiple Classifier 93.13% Figure 1. Analysis of brain tumor segmentation using CNN with MRI image [19]. 18
- 50. To begin the pre-processing process, the original gray level MR pictures are retrieved in a variety of sizes. Step 2 determines the region of interest utilising the active contours-based segmentation technique by establishing one of the largest contour. A contour is made up of an accumulation of points that have been interpolated utilising different residual methods to approximate the curve in a picture, such as linear, polynomials, or splines. The third step in a thresholding method is to select the extreme spots. Thresholding is an essential non contextual segmentation process that generates a binary area map with a single threshold by trying to convert a greyscale or coloured image to an image pixels [16]. Models are used to extract features using the MR brain. image library. Data from the major ImageNet dataset is used to train the pre-trained CNN models [20]. InceptionResNetV2 [19], ‘VGG-16, and VGG-19, Xception, ResNet50, and InceptionV3, as well as DenseNet201, are a few of CNN models that have been trained and used in the project. 4 CONCLUSION Healthcare image processing and classification methods have gotten a lot of interest recently. A dramatically higher accuracy can be accomplished by getting a superior dataset with high- resolution images that were clearly taken from the MRI scanner. To further improve quality, classifier boosting technologies can also be used, trying to make this tool a meet its objective for any medical facility treat brain tumors. To further reduce the noise, classifier boosting technologies could be used, making this tool a meet its objective for any medical facility treated brain tumors. MRI is the imaging technique that is most advantageous for detecting brain tumors. This study aims to give a comprehensive review on detection of brain tumors. REFERENCES [1] Deepak C. Dhanwani, Mahip M. Bartere, “Survey on Various Techniques of Brain Tumour Detection from MRI Images”, IJCER, Vol.04, issue.1, Issn 2250-3005, January 2014, pg. 24–26. [2] Megha A joshi, Shah D. H., “Survey of Brain Tumor Detection Techniques Through MRI Images”, AIJRFANS, ISSN:2328-3785, March–May 2015, pp.09 [3] Gupta, V. and Bibhu, V., 2022. Deep Residual Network Based Brain Tumor Segmentation and Detection with MRI Using Improved Invasive Bat Algorithm. Multimedia Tools and Applications, pp.1–23 [4] Manoj K Kowear and Sourabh Yadev, “Brain Tumor Detection and Segmentation Using Histogram Thresholding”, International Journal of engineering and Advanced Technology, April 2012. [5] Rajesh C. patil, Bhalchandra A.S., “Brain Tumor Extraction from MRI Images Using MAT Lab”, IJECSCSE, ISSN: 2277- 9477, Volume 2, issue 1. [6] Vinay Parmeshwarappa, Nandish S, “A Segmented Morphological Approach to Detect Tumor in Brain Images”, IJARCSSE, ISSN: 2277 128X, volume 4, issue 1, January 2014 [7] Preetha R., Suresh G. R., “Performance Analysis of Fuzzy C Means Algorithm in Automated Detection ofBrain Tumor”, IEEE CPS, WCCCT, 2014. [8] Amer Al-Badarnech, Hassan Najadat, Ali M. Alraziqi, “A Classifier to Detect Tumor Disease in MRI Brain Images”, IEEE Computer Society, ASONAM. 2012, 142 [9] Palvika, Shatakshi, Sharma, Y., Dagur, A., Chaturvedi, R. (2019). Automated Bug Reporting System with Keyword-driven Framework. In Soft Computing and Signal Processing: Proceedings of ICSCSP 2018, Volume 2(pp. 271–277). Springer Singapore. [10] Kumar, A., Alam, B. (2019). Energy Harvesting Earliest Deadline First Scheduling Algorithm for Increasing Lifetime of Real Time Systems. International Journal of Electrical and Computer Engineering, 9(1), 539. [11] Gudigar A., Raghavendra U., San T. R., Ciaccio E. J.,and Acharya U. R., “Application of Multiresolution Analysis for Automated Detection of Brain Abnormality Using MR Images: A Comparative Study,” Future Gener. Comput. Syst., vol. 90, pp. 359–367, Jan. 2019. 19
- 51. [12] Toğaçar M., Ergen B., and Cömert Z., “BrainMRNet: Brain Tumor Detection Using Magnetic Resonance Images with a Novel Convolutional Neural Network Model,” Med. [13] Jia Z. and Chen D., “Brain Tumor Identification and Classification of MRI Images Using Deep Learning Techniques,” IEEE Access, early access, Aug. 13, 2020, doi: 10.1109/ACCESS.2020.3016319. [14] Zhuang F., Qi Z., K. Duan, Xi D., Zhu Y., Zhu H., Xiong H., and He Q., “A Comprehensive Survey on Transfer Learning,” Proc. IEEE, vol. 109, no. 1, pp. 43–76, Jul. 2020 [15] Szegedy C., Ioffe S., Vanhoucke V., and Alemi A. A., “Inception-v4, Inception-ResNet and the Impact of Residual Connections on Learning,” in Proc. 31st AAAI Conf. Artif. Intell., San Francisco, CA, USA, 2017, pp. 4278–4284. [16] He K., Zhang X., Ren S., and Sun J., “Deep Residual Learning for Image Recognition,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), Las Vegas, NV, USA, Jun. 2016, pp. 770–778. [17] Chollet F., “Xception: Deep Learning with Depthwise Separable Convolutions,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), Honolulu, HI, USA, Jul. 2017, pp. 1800–1807. [18] Szegedy C., Vanhoucke V., Ioffe S., Shlens J., and Wojna Z., “Rethinking the Inception Architecture for Computer Vision,” in Proc. IEEE Conf.nComput. Vis. Pattern Recognit. (CVPR), Las Vegas, NV, USA, Jun. 2016, pp. 2818–2826. [19] Huang G., Liu Z., van der Maaten L., and Weinberger K. Q., “Densely connected convolutional networks,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), Honolulu, HI, USA, Jul. 2017, pp. 2261–2269. 20
- 52. A review on wildlife identification and classification Kartikeyea Singh, Manvi Singhal, Nirbhay Singh, Sur Singh Rawat Vimal Gupta Department of Computer Science and Engineering, JSS Academy of Technical Education, Noida, India ABSTRACT: Decisions on conservation and management must be supported by effective and trustworthy observation of wild creatures in their native environment. Automatic covert cameras, sometimes known as “camera traps,” are becoming a more and more common method for animal surveillance due to their efficiency and dependability in quietly, regularly, and in great quantities gathering data on wildlife. However, manually processing such a massive number of photos and movies taken with camera traps is very expensive, tedious, and time-consuming. This is a sig- nificant barrier for ecologists and scientists trying to observe wildlife in a natural setting. This study suggests that present developments in deep learning techniques can be used in computer vision. This work presents a comprehensive review using current breakthroughs in deep learning methods Keywords: Image classification, CNN, SVM. 1 INTRODUCTION Ecology’s primary goal is to learn about wild creatures in their natural habitats. By overusing natural resources, the rapid increase in population of people and the never-ending need to pursue economic growth are prompting quick, innovative, and significant updates to the systems of life on earth. Human activity has altered the population, habitat, and behaviour of wildlife on a growing area of land surface. Overusing natural resources to cause rapid, innovative, and significant changes to the Earth’s ecosystems. In response to these alterations, contemporary techniques for observing wild animals, such as tracking by satellite and GPS, wireless sensor network, radio, and movement cam trap, surveillance have been developed. Because of their unique qualities, greater commercial accessibility, and simplicity of setup and performance, an increasing number of people are using remote motion-activated cameras, also known as “camera traps,” to monitor wildlife. A standard model of a hidden camera, for example, is capable of taking pictures that are not just in high definition but also collecting image data such as the moon phase, the time, and the temperature, as well as information about the day and the night a shown in the Figure 1. The enormous image collections, however, and the limitations of low-quality photographs, have a significant impact on the speed and, at times, accuracy of human classification. Images taken in a field setting present a difficult classification problem because they appear in a field setting. Figure 1. Example of wildlife. DOI: 10.1201/9781003393580-4 21 Artificial Intelligence, Blockchain, Computing and Security – Dagur et al. (Eds) © 2024 The Author(s), ISBN: 978-1-032-49393-0
- 53. Variable lighting and weather conditions, a crowded background, a different pose, human photographic flaws, different perspectives, and occlusions are all factors. Due to all of these challenges, an effective algorithm for classification with the highest level of accuracy is required. Convolutional neural networks are a kind of novel simulated neural network and deep learning algorithm designed for efficient image processing. In recent years, multilayer neural networks have been successfully applied in decision-making, learning, pattern recognition, and classification. 2 LITERATURE SURVEY This section contains papers on all-inclusive object recognition and focus mechanisms in image manipulation, and identification and categorization of animal species. Several image classifica- tion techniques were covered in this study. The most common methods for classifying images are classifiers that are object-oriented, spectral, contextual, spectral-contextual, per-pixel, per-field, as well as robust and weak categories. The most often utilised techniques are covered in this section. This survey provides theoretical information about categorization techniques as well as recommendations for the best ones. The author proposed a model called “Machine Learning, Neural Networks, and Convolutional Neural Networks” in [1]. The study of digital models that are planned to improve effortlessly via training on example data is known as machine learning. In [2] the author proposed a model called neural systems which was constructed by the means of artificial neurons. In [3] the author proposed a model titled “Challenges of Camera-trap Images for Convolutional Neural Networks.” Like all machine learning models, CNN animal classifi- cation for camera-trap photos has to be taken into account and also generalization issues. They demonstrated that when classifying images from untrained camera sites, unfamiliarity with cutting-edge neural network classifiers accuracy drops. In [4] the author proposed a model titled “Deep Convolutional Neural Networks for Automated Wildlife Monitoring: Animal Recognition and Identification.” In this instance, they applied a convolutional neural network. It automatically detects important features without the need for human intervention. [5] Will focus on observing animal behavior in the wild employing facial detection and tracing. It will show a formula for detecting and tracing fauna faces in biota videos. Vincent Miele describes how efforts are still being made to develop species identification criteria that are based not only measures of the craniofacial region as well as external morphol- ogy, particularly on the cranium and maxilla, in [6]. In this case study, three different mouse species—the house mouse Mus musculus, the European wood mouse Apodemus sylvaticus, and the Cairo spiny mouse Acomys cahirinus—were examined. In [7], researchers used transfer learning to classify and forecast images in a google collabrative for the ImageNet collected information. In this study, MobileNet, MobileNetV2, VGG16, VGG19, and ResNet50 are employed as transfer learning models The Google Colab notepad served as aid for picture cate- gorization and prediction. The main goal in [8] was to see if they could limit a sample of manually annotated camera-trap photos, educate a deep learning machine to recognize animals, including a particular species. In [9] features five elements of the detection were compared to wild, a fresh dataset of actual animal sightings focusing on difficult detection scenarios. In [10], to develop a sensor that examines video images captured by camera traps in real time. The classes to identify are rhinos, people and a collection of six typical big animals on the savannah of Africa. The scope of this project also includes the extraction of significant events. Peiyi Zeng uses Python and a simple 2D CNN to classify similar animal images in [11]. Its main focus is on the dichotomous division between snub-nosed monkeys and other monkey species. A Python crawler is used to build the database, each class having 800 and 200 photos serving as test and training data respectively. The training accuracy is 96.67% when no anomalies exist. The algorithm developed using the Missouri Hidden Cameras database [12] in Nawaz Sheikh’s MobileNet architecture and the model appears to have performed well having an F1 score of 0.68 as opposed to InceptionV3’s and VGG-16’s respective scores of 0.70 and 0.62. According to [13], the overarching goal of this research was to provide managers with guidance on the most of the project objectives, accurate 22
- 54. models for camera-trap image analysis are required. In [14] the goal of this study, however, is to find the creature even before hunt, not while it is going on. We propose a new method employing machine learning techniques to separate predators from non-predators by extracting animal traits. [15]. In [16], the proposed model was written in Python and tested in Visual Studio on a dataset containing 12,984 images from six different animal species belonging to six different animal classes/kingdoms. For six animal classes, the model had an accuracy of 87.22 percent. 3 MATERIAL AND METHODOLOGY 3.1 Convolutonal Neural Network (CNN) We used Convolutional Neural Network (CNN) approach as shown in the Figure 2. This is a widely used technique in applications for computer vision. It is a type of deep neural network used to analyse visual imagery. This architectural style predominates when identifying objects in a photograph or video. Applications such as neural language processing, image or video recognition, and so on. Five layers make up a convolution neural network: input, convolution, pooling, fully connected, and output layers. Warning capabilities, massive data capacity and picture per- fect. Slower operation and long training period. 3.2 Deep learning Categorization of camera-traps automatically in Nilgai by utilizing wildlife conservation in-depth learning A branch of machine learning called deep learning, attempts to take out knowledge from large information sets by understanding the underlying levels of more significant depiction or characteristics (Chollet 2018). A neural network is a deep learning model composed of several layers which are taught on labelled Information and pattern extraction in a hierarchical fashion. In order to generate predictions that will be used to inform subsequent layers. Using predicted and actual values, the neural network computes a rating of failures, which is then transmitted returning via the system to modify weighing value. Knowledge is acquired in a iterative manner on new unlabelled data by adjusting weights such that it maximises the capacity to lower it’s error causing modest score, weights from previous layering are kept and used. 3.3 K-Nearest Neighbor (KNN) classifier Fuzzy K-Nearest Neighbour Classifier is used to classify cattle. No training period, simple implementation, and new data may be added at any moment without altering the model since there is no preparation stage. Does not work well with large datasets, high dimen- sionality, or noisy or missing data. 3.4 Transfer learning In colab notebook, image classification and prediction are accomplished by utilising transfer learn- ing. Adaptive learning allows developers to prevent the requirement for an abundance of new data. The One of the most important restrictions on transfer learning is the problem of negative transfer. Figure 2. CNN Architecture layers. 23
- 55. Random documents with unrelated content Scribd suggests to you:
- 59. The Project Gutenberg eBook of The Life of Sir Isaac Newton
- 60. This ebook is for the use of anyone anywhere in the United States and most other parts of the world at no cost and with almost no restrictions whatsoever. You may copy it, give it away or re-use it under the terms of the Project Gutenberg License included with this ebook or online at www.gutenberg.org. If you are not located in the United States, you will have to check the laws of the country where you are located before using this eBook. Title: The Life of Sir Isaac Newton Author: David Brewster Release date: October 18, 2016 [eBook #53311] Most recently updated: October 23, 2024 Language: English Credits: E-text prepared by Sonya Schermann, Charlie Howard, and the Online Distributed Proofreading Team (http://www.pgdp.net) from page images generously made available by Internet Archive (https://archive.org) *** START OF THE PROJECT GUTENBERG EBOOK THE LIFE OF SIR ISAAC NEWTON ***
- 61. The Project Gutenberg eBook, The Life of Sir Isaac Newton, by David Brewster Note:Images of the original pages are available through Internet Archive. See https://archive.org/details/56010330R.nlm.nih.gov
- 62. HARPER’S FAMILY LIBRARY. DESIGNED FOR ADULT PERSONS. “Books that you may carry to the fire, and hold readily in your hand, are the most useful after all. A man will often look at them, and be tempted to go on, when he would have been frightened at books of a larger size, and of a more erudite appearance.”—Dr. Johnson. The proprietors of the Family Library feel themselves stimulated to increased exertions by the distinguished favour with which it has already been received. The volumes now before the public may be confidently appealed to as proofs of zeal on the part of the publishers to present to their readers a series of productions, which, as they are connected, not with ephemeral, but with permanent subjects, may, years hence as well as now, be consulted for lively amusement as well as solid instruction. To render this Library still more worthy of patronage, the proprietors propose incorporating in it such works of interest and value as may appear in the various Libraries and Miscellanies now preparing in Europe, particularly “Constable’s Miscellany,” the “Edinburgh Cabinet” Library, c. All these productions, as they emanate from the press, will be submitted to literary gentlemen for inspection; and none will be reprinted but such as shall be found calculated to sustain the exalted character which this Library has already acquired.
- 63. Several well-known authors have been engaged to prepare for it original works of an American character, on History, Biography, Travels, c. c. Every distinct subject will in general be comprehended in one volume, or at most in three volumes, which may form either a portion of the series or a complete work by itself; and each volume will be embellished with appropriate engravings. The entire series will be the production of authors of eminence, who have acquired celebrity by their literary labours, and whose names, as they appear in succession, will afford the surest guarantee to the public for the satisfactory manner in which the subjects will be treated. Such is the plan by which it is intended to form an American Family Library, comprising all that is valuable in those branches of knowledge which most happily unite entertainment with instruction. The utmost care will be taken, not only to exclude whatever can have an injurious influence on the mind, but to embrace every thing calculated to strengthen the best and most salutary impressions. With these arrangements and facilities, the publishers flatter themselves that they shall be able to present to their fellow-citizens a work of unparalleled merit and cheapness, embracing subjects adapted to all classes of readers, and forming a body of literature deserving the praise of having instructed many, and amused all; and above every other species of eulogy, of being fit to be introduced, without reserve or exception, by the father of a family to the domestic circle. Meanwhile, the very low price at which it is charged renders more extensive patronage necessary for its support and prosecution. The immediate encouragement, therefore, of those who approve its plan and execution is respectfully solicited. The work may be obtained in complete sets, or in separate numbers, from the principal booksellers throughout the United States.
- 64. OPINIONS OF THE FAMILY LIBRARY. “The publishers have hitherto fully deserved their daily increasing reputation by the good taste and judgment which have influenced the selections of works for the Family Library.”—Albany Daily Advertiser. “The Family Library—A title which, from the valuable and entertaining matter the collection contains, as well as from the careful style of its execution, it well deserves. No family, indeed, in which there are children to be brought up, ought to be without this Library, as it furnishes the readiest resources for that education which ought to accompany or succeed that of the boarding-school or the academy, and is infinitely more conducive than either to the cultivation of the intellect.”—Monthly Review. “It is the duty of every person having a family to put this excellent Library into the hands of his children.”—N. Y. Mercantile Advertiser. “It is one of the recommendations of the Family Library, that it embraces a large circle of interesting matter, of important information and agreeable entertainment, in a concise manner and a cheap form. It is eminently calculated for a popular series— published at a price so low, that persons of the most moderate income may purchase it—combining a matter and a style that the most ordinary mind may comprehend it, at the same time that it is calculated to raise the moral and intellectual character of the people.”—Constellation. “We have repeatedly borne testimony to the utility of this work. It is one of the best that has ever been issued from the American press, and should be in the library of every family desirous of treasuring up useful knowledge.”—Boston Statesman. “We venture the assertion that there is no publication in the country more suitably adapted to the taste and requirements of the great mass of community, or better calculated to raise the
- 65. intellectual character of the middling classes of society, than the Family Library.”—Boston Masonic Mirror. “We have so often recommended this enterprising and useful publication (the Family Library), that we can here only add, that each successive number appears to confirm its merited popularity.”— N. Y. American. “The little volumes of this series truly comport with their title, and are in themselves a Family Library.”—N. Y. Commercial Advertiser. “We recommend the whole set of the Family Library as one of the cheapest means of affording pleasing instruction, and imparting a proper pride in books, with which we are acquainted.”—U. S. Gazette. “It will prove instructing and amusing to all classes. We are pleased to learn that the works comprising this Library have become, as they ought to be, quite popular among the heads of families.”— N. Y. Gazette. “The Family Library is, what its name implies, a collection of various original works of the best kind, containing reading useful and interesting to the family circle. It is neatly printed, and should be in every family that can afford it—the price being moderate.”—New- England Palladium. “We are pleased to see that the publishers have obtained sufficient encouragement to continue their valuable Family Library.”— Baltimore Republican. “The Family Library presents, in a compendious and convenient form, well-written histories of popular men, kingdoms, sciences, c. arranged and edited by able writers, and drawn entirely from the most correct and accredited authorities. It is, as it professes to be, a Family Library, from which, at little expense, a household may prepare themselves for a consideration of those elementary subjects of education and society, without a due acquaintance with which neither man nor woman has claim to be well bred, or to take their
- 66. proper place among those with whom they abide.”—Charleston Gazette.
- 67. BOY’S AND GIRL’S LIBRARY. PROSPECTUS. The Publishers of the “Boy’s and Girl’s Library” propose, under this title, to issue a series of cheap but attractive volumes, designed especially for the young. The undertaking originates not in the impression that there does not already exist in the treasures of the reading world a large provision for this class of the community. They are fully aware of the deep interest excited at the present day on the subject of the mental and moral training of the young, and of the amount of talent and labour bestowed upon the production of works aiming both at the solid culture and the innocent entertainment of the inquisitive minds of children. They would not therefore have their projected enterprise construed into an implication of the slightest disparagement of the merits of their predecessors in the same department. Indeed it is to the fact of the growing abundance rather than to the scarcity of useful productions of this description that the design of the present work is to be traced; as they are desirous of creating a channel through which the products of the many able pens enlisted in the service of the young may be advantageously conveyed to the public. The contemplated course of publications will more especially embrace such works as are adapted, not to the extremes of early childhood or of advanced youth, but to that intermediate space which lies between childhood and the opening of maturity, when the trifles of the nursery and the simple lessons of the school-room have ceased to exercise their beneficial influence, but before the taste for a higher order of mental pleasure has established a fixed
- 68. ascendency in their stead. In the selection of works intended for the rising generation in this plastic period of their existence, when the elements of future character are receiving their moulding impress, the Publishers pledge themselves that the utmost care and scrupulosity shall be exercised. They are fixed in their determination that nothing of a questionable tendency on the score of sentiment shall find admission into pages consecrated to the holy purpose of instructing the thoughts, regulating the passions, and settling the principles of the young. In fine, the Publishers of the “Boy’s and Girl’s Library” would assure the Public that an adequate patronage alone is wanting to induce and enable them to secure the services of the most gifted pens in our country in the proposed publication, and thus to render it altogether worthy of the age and the object which calls it forth, and of the countenance which they solicit for it.
- 69. SIR. G. KNELLER PINX. ENG.d BY GIMBER. Printed by R. Miller SIR ISAAC NEWTON. HARPER’S FAMILY LIBRARY
- 70. Harper’s Stereotype Edition. THE L I F E OF SIR ISAAC NEWTON. BY DAVID BREWSTER, LL.D. F.R.S. Ergo vivida vis animi pervicit, et extra Processit longe flammantia mœnia mundi; Atque omne immensum peragravit mente amimoque. Lucret. lib. i. 1. 73.
- 71. The Birthplace of Newton. NEW-YORK: PRINTED AND PUBLISHED BY J. J. HARPER; NO. 82 CLIFF-STREET, AND SOLD BY THE BOOKSELLERS GENERALLY THROUGHOUT THE UNITED STATES. 1833.
- 72. TO THE RIGHT HONOURABLE LORD BRAYBROOKE. The kindness with which your lordship intrusted to me some very valuable materials for the composition of this volume has induced me to embrace the present opportunity of publicly acknowledging it. But even if this personal obligation had been less powerful, those literary attainments and that enlightened benevolence which reflect upon rank its highest lustre would have justified me in seeking for it the patronage of a name which they have so justly honoured. DAVID BREWSTER. Allerly, June 1st, 1831.
- 73. PREFACE. As this is the only Life of Sir Isaac Newton on any considerable scale that has yet appeared, I have experienced great difficulty in preparing it for the public. The materials collected by preceding biographers were extremely scanty; the particulars of his early life, and even the historical details of his discoveries, have been less perfectly preserved than those of his illustrious predecessors; and it is not creditable to his disciples that they have allowed a whole century to elapse without any suitable record of the life and labours of a master who united every claim to their affection and gratitude. In drawing up this volume, I have obtained much assistance from the account of Sir Isaac Newton in the Biographia Britannica; from the letters to Oldenburg, and other papers in Bishop Horsley’s edition of his works; from Turnor’s Collections for the History of the Town and Soke of Grantham; from M. Biot’s excellent Life of Newton in the Biographie Universelle; and from Lord King’s Life and Correspondence of Locke. Although these works contain much important information respecting the Life of Newton, yet I have been so fortunate as to obtain many new materials of considerable value. To the kindness of Lord Braybrooke I have been indebted for the interesting correspondence of Newton, Mr. Pepys, and Mr. Millington, which is now published for the first time, and which throws much light upon an event in the life of our author that has recently acquired an unexpected and a painful importance. These letters, when combined with those which passed between Newton and Locke, and with a curious extract from the manuscript diary of Mr.
- 74. Abraham Pryme, kindly furnished to me by his collateral descendant Professor Pryme of Cambridge, fill up a blank in his history, and have enabled me to delineate in its true character that temporary indisposition which, from the view that has been taken of it by foreign philosophers, has been the occasion of such deep distress to the friends of science and religion. To Professor Whewell, of Cambridge, I owe very great obligations for much valuable information. Professor Rigaud, of Oxford, to whose kindness I have on many other occasions been indebted, supplied me with several important facts, and with extracts from the diary of Hearne in the Bodleian Library, and from the original correspondence between Newton and Flamstead, which the president of Corpus Christi College had for this purpose committed to his care; and Dr. J. C. Gregory, of Edinburgh, the descendant of the illustrious inventor of the reflecting telescope, allowed me to use his unpublished account of an autograph manuscript of Sir Isaac Newton, which was found among the papers of David Gregory, Savilian Professor of Astronomy at Oxford, and which throws some light on the history of the Principia. I have been indebted to many other friends for the communication of books and facts, but especially to Sir William Hamilton, Bart., whose liberality in promoting literary inquiry is not limited to the circle of his friends. D. B. Allerly, June 1st, 1831.
- 75. CONTENTS. Page CHAPTER I. The Pre-eminence of Sir Isaac Newton’s Reputation—The Interest attached to the Study of his Life and Writings—His Birth and Parentage—His early Education—Is sent to Grantham School—His early Attachment to Mechanical Pursuits—His Windmill—His Water-clock—His Self-moving Cart—His Sun- dials—His Preparation for the University 17 CHAPTER II. Newton enters Trinity College, Cambridge— Origin of his Propensity for Mathematics—He studies the Geometry of Descartes unassisted —Purchases a Prism—Revises Dr. Barrow’s Optical Lectures—Dr. Barrow’s Opinion respecting Colours—Takes his Degrees—Is appointed a Fellow of Trinity College— Succeeds Dr. Barrow in the Lucasian Chair of Mathematics 26
- 76. CHAPTER III. Newton occupied in grinding Hyperbolical Lenses—His first Experiments with the Prism made in 1666—He discovers the Composition of White Light, and the different Refrangibility of the Rays which compose it—Abandons his Attempts to improve Refracting Telescopes, and resolves to attempt the Construction of Reflecting ones—He quits Cambridge on account of the Plague—Constructs two Reflecting Telescopes in 1668, the first ever executed—One of them examined by the Royal Society, and shown to the King—He constructs a Telescope with Glass Specula— Recent History of the Reflecting Telescope— Mr. Airy’s Glass Specula—Hadley’s Reflecting Telescopes—Short’s—Herschel’s—Ramage’s— Lord Oxmantown’s 30 CHAPTER IV. He delivers a Course of Optical Lectures at Cambridge—Is elected Fellow of the Royal Society—He communicates to them his Discoveries on the different Refrangibility and Nature of Light—Popular Account of them— They involve him in various Controversies— His Dispute with Pardies—Linus—Lucas—Dr. Hooke and Mr. Huygens—The Influence of these Disputes on the mind of Newton 47
- 77. CHAPTER V. Mistake of Newton in supposing that the Improvement of Refracting Telescopes was hopeless—Mr. Hall invents the Achromatic Telescope—Principles of the Achromatic Telescope explained—It is reinvented by Dollond, and improved by future Artists—Dr. Blair’s Aplanatic Telescope—Mistakes in Newton’s Analysis of the Spectrum—Modern Discoveries respecting the Structure of the Spectrum 63 CHAPTER VI. Colours of thin Plates first studied by Boyle and Hooke—Newton determines the Law of their Production—His Theory of Fits of easy Reflection and Transmission—Colours of thick Plates 75 CHAPTER VII. Newton’s Theory of the Colours of Natural Bodies explained—Objections to it stated— New Classification of Colours—Outline of a new Theory proposed 82 CHAPTER VIII. Newton’s Discoveries respecting the Inflection or Diffraction of Light—Previous Discoveries 98
- 78. of Grimaldi and Dr. Hooke—Labours of succeeding Philosophers—Law of Interference of Dr. Young—Fresnel’s Discoveries—New Theory of Inflection on the Hypothesis of the Materiality of Light CHAPTER IX. Miscellaneous Optical Researches of Newton— His Experiments on Refraction—His Conjecture respecting the Inflammability of the Diamond—His Law of Double Refraction— His Observations on the Polarization of Light —Newton’s Theory of Light—His “Optics” 106 CHAPTER X. Astronomical Discoveries of Newton—Necessity of combined Exertion to the completion of great Discoveries—Sketch of the History or Astronomy previous to the time of Newton— Copernicus, 1473–1543—Tycho Brahe, 1546– 1601—Kepler, 1571–1631—Galileo, 1564– 1642 110 CHAPTER XI. The first Idea of Gravity occurs to Newton in 1666—His first Speculations upon it— Interrupted by his Optical Experiments—He resumes the Subject in consequence of a Discussion with Doctor Hooke—He discovers the true Law of Gravity and the Cause of the Planetary Motions—Dr. Halley urges him to 140
- 79. publish his Principia—His Principles of Natural Philosophy—Proceedings of the Royal Society on this Subject—The Principia appears in 1687—General Account of it, and of the Discoveries it contains—They meet with great Opposition, owing to the Prevalence of the Cartesian System—Account of the Reception and Progress of the Newtonian Philosophy in Foreign Countries—Account of its Progress and Establishment in England CHAPTER XII. Doctrine of Infinite Quantities—Labours of Pappus—Kepler—Cavaleri—Roberval—Fermat —Wallis—Newton discovers the Binomial Theorem and the Doctrine of Fluxions in 1606 —His Manuscript Work containing this Doctrine communicated to his Friends—His Treatise on Fluxions—His Mathematical Tracts —His Universal Arithmetic—His Methodus Differentialis—His Geometria Analytica—His Solution of the Problems proposed by Bernouilli and Leibnitz—Account of the celebrated Dispute respecting the Invention of Fluxions—Commercium Epistolicum— Report of the Royal Society—General View of the Controversy 168 CHAPTER XIII. James II. attacks the Privileges of the University of Cambridge—Newton chosen one of the Delegates to resist this Encroachment—He is 200
- 80. elected a Member of the Convention Parliament—Burning of his Manuscript—His supposed Derangement of Mind—View taken of this by foreign Philosophers—His Correspondence with Mr. Pepys and Mr. Locke at the time of his Illness—Mr. Millington’s Letter to Mr. Pepys on the subject of Newton’s Illness—Refutation of the Statement that he laboured under Mental Derangement CHAPTER XIV. No Mark of National Gratitude conferred upon Newton—Friendship between him and Charles Montague, afterward Earl of Halifax—Mr. Montague appointed Chancellor of the Exchequer in 1694—He resolves upon a Recoinage—Nominates Mr. Newton Warden of the Mint in 1695—Mr. Newton appointed Master of the Mint in 1699—Notice of the Earl of Halifax—Mr. Newton elected Associate of the Academy of Sciences in 1699—Member for Cambridge in 1701—and President of the Royal Society in 1703—Queen Anne confers upon him the Honour of Knighthood in 1705 —Second Edition of the Principia, edited by Cotes—His Conduct respecting Mr. Ditton’s Method of finding the Longitude 223 CHAPTER XV. Respect in which Newton was held at the Court of George I.—The Princess of Wales delighted with his Conversation—Leibnitz endeavours to 234
- 81. prejudice the Princess against Sir Isaac and Locke—Controversy occasioned by his Conduct—The Princess obtains a Manuscript Abstract of his System of Chronology—The Abbé Conti is, at her request, allowed to take a Copy of it on the promise of Secrecy—He prints it surreptitiously in French, accompanied with a Refutation by M. Freret— Sir Isaac’s Defence of his System—Father Souciet attacks it, and is answered by Dr. Halley—Sir Isaac’s larger Work on Chronology published after his Death—Opinions respecting it—Sir Isaac’s Paper on the Form of the most ancient Year CHAPTER XVI. Theological Studies of Sir Isaac—Their Importance to Christianity—Motives to which they have been ascribed—Opinions of Biot and La Place considered—His Theological Researches begun before his supposed Mental Illness—The Date of these Works fixed —Letters to Locke—Account of his Observations on Prophecy—His Lexicon Propheticum—His Four Letters to Dr. Bentley —Origin of Newton’s Theological Studies— Analogy between the Book of Nature and that of Revelation 242 CHAPTER XVII. The Minor Discoveries and Inventions of Newton—His Researches on Heat—On Fire 265
- 82. and Flame—On Elective Attraction—On the Structure of Bodies—His supposed Attachment to Alchymy—His Hypothesis respecting Ether as the Cause of Light and Gravity—On the Excitation of Electricity in Glass—His Reflecting Sextant invented before 1700—His Reflecting Microscope—His Prismatic Reflector as a Substitute for the small Speculum of Reflecting Telescopes—His Method of varying the Magnifying Power of Newtonian Telescopes—His Experiments on Impressions on the Retina CHAPTER XVIII. His Acquaintance with Dr. Pemberton—Who edits the Third Edition of the Principia—His first Attack of ill Health—His Recovery—He is taken ill in consequence of attending the Royal Society—His Death on the 20th March, 1727—His Body lies in state—His Funeral—He is buried in Westminster Abbey—His Monument described—His Epitaph—A Medal struck in honour of him—Roubiliac’s full- length Statue of him erected in Cambridge— Division of his Property—His Successors 284 CHAPTER XIX. Permanence of Newton’s Reputation—Character of his Genius—His Method of Investigation similar to that used by Galileo—Error in ascribing his Discoveries to the Use of the Methods recommended by Lord Bacon—The 292
- 83. Pretensions of the Baconian Philosophy examined—Sir Isaac Newton’s Social Character—His great Modesty—The Simplicity of his Character—His Religious and Moral Character—His Hospitality and Mode of Life— His Generosity and Charity—His Absence—His Personal Appearance—Statues and Pictures of him—Memorials and Recollections of him Appendix, No. I.—Observations on the Family of Sir Isaac Newton 307 Appendix, No. II.—Letter from Sir Isaac Newton to Francis Aston, Esq., a young Friend who was on the eve of setting out on his Travels 316 Appendix, No. III.—“A Remarkable and Curious Conversation between Sir Isaac Newton and Mr. Conduit.” 320
- 84. L I F E OF SIR ISAAC NEWTON.
- 85. Welcome to our website – the perfect destination for book lovers and knowledge seekers. We believe that every book holds a new world, offering opportunities for learning, discovery, and personal growth. That’s why we are dedicated to bringing you a diverse collection of books, ranging from classic literature and specialized publications to self-development guides and children's books. More than just a book-buying platform, we strive to be a bridge connecting you with timeless cultural and intellectual values. With an elegant, user-friendly interface and a smart search system, you can quickly find the books that best suit your interests. Additionally, our special promotions and home delivery services help you save time and fully enjoy the joy of reading. Join us on a journey of knowledge exploration, passion nurturing, and personal growth every day! ebookbell.com