DETECTION AND EXTRACTION OF SEA MINE FEATURES USING CNN ARCHITECTURE
0 likes42 views
The document discusses the detection and extraction of features of sea mines using CNN architecture. It first provides background on deep neural networks, convolutional neural networks, and generative adversarial networks. It then summarizes previous literature on related topics like sonar target recognition, underwater image classification, pretext-invariant representation learning, and underwater mine detection using Mask RCNN. The paper proposes detecting sea mines in real-time using a more extensive dataset to train models like YOLO v3 for improved performance across variations in mines. It concludes by listing references used in the document.
![International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 09 Issue: 05 | May 2022 www.irjet.net p-ISSN: 2395-0072
© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 3881
DETECTION AND EXTRACTION OF SEA MINE FEATURES USING CNN
ARCHITECTURE
Sheethal S, Soundarya S, Vidhisha V, Kiran G, Dr A B Rajendra
Department of Information Science and Engineering
VidyaVardhaka College of Engineering, Mysuru
--------------------------------------------------------------------------***-----------------------------------------------------------------------
Abstract
The Conventional worry of navel asset is naval mines; these mines are stationary and were planted during war times and
now they have been working as a threat to naval ships, and submarines. Detection of those naval mines has been one of the
foremost risk-taking tasks, with modern technology various techniques are wont to detect these mines Using Ultrasonic
signals, Symbolic pattern analysis of side-scan sonar images but detection through image processing has been one of the
most challenging and efficient ones since it can solve the real-time problem with less error, the image classification model
like uses FRCNN(Fast Region Convolutional Neural Network) algorithm to classify the objects as mine or not. The cloud
platform is employed to watch the mine and as soon as the changes are observed the Android application will reflect the
changes.
Keywords: FRCNN, Neural network, Image processing, Deep Learning, ResNet, TensorFlow, Python
1. INTRODUCTION
Nonmilitary mines, often known as aquatic mines,
are used in combat. During a conflict, mines are used to
destroy naval assets. It's also used in the defense
industry, where mines operate as a border to protect the
country's marine territory. These mines prevent hostile
maritime assets from entering the unmarked territory.
The enemy must search the entire region for mines. The
opponent is forced to assault in an unmined place, where
the defense is ready for a fight. Unlike the older mines,
the ultramodern mines are detonated by pressing a
button. The discovery of underwater mines is critical in
ensuring that civilians are not endangered in any way.
Mines aid in securing high-altitude defense bases and
preventing the leakage of sensitive information.
Battlegroups will be able to pinpoint the exact location of
mines and avoid losses with the help of a reliable and
cost-effective method. The neural network's operation is
comparable to that of a mortal brain. It's used to show
the relationship between data across a computer system.
This artificial network is primarily based on Machine
Literacy. If a dependable and cost-effective technology
was utilized, battle groups would be able to find the
exact location of mines and save lives. The neural
network works in the same way that a human brain does.
It's a diagram that depicts the connections between data
in a computer system. Machine literacy is the foundation
of the artificial network. Mask RCNN creates an offer
about the region in the image where the object might be
present and then constructs bounding boxes and masks
at pixel position, as well as forecasts the object's class.
For producing point vectors from raw pictures, Mask
RCNN uses FPN as the backbone. For searching items in
regions, use RPN. Anchor boxes are utilized to align the
point vector with the position in the raw image, which
can then be compared to ground verity while discovering
and utilizing the idea of IoU value.
2. LITERATURE REVIEW
[1] The paper mentions the simplest way of
implementing various deep learning techniques, hence
the modifications need to be done to the techniques this
could be the major challenge of the paper. In this paper
modules and sub-modules used are CNN, Autoencoders,
Deep Belief Networks, and GAN. This paper provides
an overview of the simplest ways to implement target
recognition and hence not very efficient.
[2] In this paper, the author Huu-Thu Nguyen, Eon-Ho
Lee 1, and Sejin Lee specified sonar sensor needs and
challenges to auto-detect submerged human bodies
underwater. Sonar images need to be tested at different
levels of polarization and intensities, the target
background must be considered as there will be
scatterings and noises in the sonar image. The same
model needs to be retrained on sonar images of various
polarization and intensity.
[3] Self-Supervised Learning of Pretext-Invariant
Representations is to construct image representations
that are semantically meaningful via PIRL (Pretext
Invariant Representation Learning) that do not require
semantic annotations for a large training set of images.
To achieve the highest single crop top-1 accuracy of all](https://image.slidesharecdn.com/irjet-v9i5795-221006074405-cdca2c36/85/DETECTION-AND-EXTRACTION-OF-SEA-MINE-FEATURES-USING-CNN-ARCHITECTURE-1-320.jpg)
![International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 09 Issue: 05 | May 2022 www.irjet.net p-ISSN: 2395-0072
© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 3882
self-supervised learners that use a single ResNet-50
model.
[4] In this paper, the author has specified more on
various techniques used for detecting sea mines. The
major challenge of this paper was images need to be
manually annotated and the overhead of labelling is very
large when dealing with huge datasets. In this paper,
they have used the Mask RCNN model which Region-
Based Convolutional Neural Network.
[5] in this paper, the methodology involves a Masked
RCNN module that comprises numerous convolutional
layers. In this referral paper, the dataset is a set of
images downloaded from the web hence a lot of pre-
processing, labeling, and augmentation is required.
[6] In this referral paper, Gabor Filter and K-means
clustering algorithm is used. Gabor Filter is used for
feature extraction, and the K-means Clustering algorithm
is used for segmentation. The accuracy and efficiency of
K-Means clustering are not accurate.
Table 1. Literature Review
[7] In this referral paper mainly two techniques are used,
Homomorphic, CLAHE, and Wavelet Filtering
Techniques
These techniques help to enhance the image quality and
removing the unwanted noise enhances the image
quality.
This survey paper also mentioned various other
techniques for image enhancement such as a median
filter for better quality and RGB for color level
stretching.
[8] In this referral paper they have specified image
denoising techniques which include Spatial Domain
Filter, Frequency Domain Filter, Mean Filter, Median
Filter, and Adaptive Filter. The paper does not consider
any hybrid filters which are more efficient in de-noising
images and for a given dataset, the right kind of filter
cannot be decided beforehand, we’ll have to implement
each of the filters on the dataset.
Title of the paper Authors of the paper Model/Sub Model used
A Review on Deep Learning-Based Approaches for
Automatic Sonar Target Recognition
Dhiraj Neupane and Jongwon Seok CNN, Autoencoders, Deep Belief
Networks, GAN
Study on the Classification Performance of
Underwater
Sonar Image Classification Based on Convolutional
Neural Networks for Detecting a Submerged
Human Body
Huu-Thu Nguyen, Eon-Ho Lee 1
and Sejin Lee
AlexNet, GoogleNet
Self-Supervised Learning of Pretext-Invariant
Representations
Ishan Misra, Laurens van der
Maaten
PIRL (Pretext Invariant
Representation Learning)
Underwater Mine Detection using Image Processing N Abhishek, Arjun, Bharathesh,
Kavitha K S, Prof. Manonmani S, Dr.
Shanta Rangaswamy
MaskRCNN model
Underwater Fish Detection Aditya Agarwal,Manonmani
S,Gaurav Rawal,Tushar
Malani,Navjeet Anand.
Masked RCNN
Image Segmentation Using Gabor Filter and
K-Means Clustering Method
Agyztia Premana,
Akhmad Pandhu Wijaya, Moch
Arief Soeleman.
Gabor Filter and K-means clustering
algorithm.
Comparative analysis of combining various
enhancement filtering techniques for underwater
images
Manonmani S, Dr. Shanta
Rangaswamy
Homomorphic, CLAHE and Wavelet
Filtering Techniques
Survey On Image Denoising Techniques Manonmani,Lalitha V.,Dr. Shanta
Rangaswamy.
Spatial Domain Filter, Frequency
Domain Filter, Mean Filter, Median
Filter and Adaptive Filter](https://image.slidesharecdn.com/irjet-v9i5795-221006074405-cdca2c36/85/DETECTION-AND-EXTRACTION-OF-SEA-MINE-FEATURES-USING-CNN-ARCHITECTURE-2-320.jpg)
![International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 09 Issue: 05 | May 2022 www.irjet.net p-ISSN: 2395-0072
© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 3883
3. BACKGROUND
3.1 DEEP NEURAL NETWORK
A deep neural network is an artificial neural
network (ANN) with several layers between the input
and output layers (DNN) as shown in Fig.1. Neurons,
synapses, weights, biases, and functions are all basic
components of neural networks, which come in a range
of forms and sizes.
Fig.1 Deep Neural Network Architecture
3.2 CONVOLUTIONAL NEURAL NETWORK
The CNN is a sort of neural network as shown in
Fig.2 that is quite similar to human vision and thought.
Various computer vision applications have become an
important part of this over time. CNN’s were constructed
for the first time in the 1980s. At the time, this neural
network was the best at recognizing manual digits. The
code reader has mostly been used or built to read zip
codes, pin codes, and other codes of a similar kind.
Fig.2 A traditional CNN Architecture
3.3 GENERATIVE ADVERSARIAL NETWORK
A generative adversarial network (GAN) is a machine
learning (ML) model in which two neural networks
compete to improve the accuracy of their predictions as
shown in Fig.1. GANs are frequently unsupervised and
learn by playing a cooperative zero-sum game
Fig.3 Generative Adversarial Network
4. CONCLUSION
The detection of the sea mines can be done in real-
time or in real-time in order to be more effective in its
usage to the naval forces of the country. A more
extensive dataset that involves actual SONAR generated
images and contains a wide selection of mines of all
shapes, sizes, and specifications, images of different
visibilities, etc can be used to train a more
comprehensive detection model that detects accurately
across all these variations. Models may be trained using
different machine-learning techniques like YONO v3 and
a few more recent approaches which can lead to
improved performance and outcomes.
REFERENCES
[1] Dhiraj Neupane and Jongwon Seok, A Review on Deep
Learning-Based Approaches for Automatic Sonar Target
Recognition, MDPI Electronics, 2020.
[2] Huu-Thu Nguyen, Eon-Ho Lee 1 and Sejin Lee, Study
on the Classification Performance of Underwater Sonar
Image Classification Based on Convolutional Neural
Networks for Detecting a Submerged Human Body, MDPI
Sensors, 2019.
[3] Ishan Misra, Laurens van der Maaten, Self-Supervised
Learning of Pretext-Invariant Representations, arXiV,
2019.
[4] N Abhishek, Arjun, Bharathesh, Kavitha KS, Prof.
Manonmani S, Dr. Shanta Rangaswamy, “Underwater
Mine Detection using Image Processing”. 2020,
International Research Journal of Engineering and
Technology (IRJET).
[5] Aditya Agarwal, Manonmani S, Gaurav Rawal, Tushar
Malani, Navjeet Anand, “Underwater Fish Detection”,
2020, International Journal of Engineering Research &
Technology (IJERT).](https://image.slidesharecdn.com/irjet-v9i5795-221006074405-cdca2c36/85/DETECTION-AND-EXTRACTION-OF-SEA-MINE-FEATURES-USING-CNN-ARCHITECTURE-3-320.jpg)
![International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 09 Issue: 05 | May 2022 www.irjet.net p-ISSN: 2395-0072
© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 3884
[6] Agyztia Premana, Akhmad Pandu Wijaya, Moch
Arief Soeleman, “Image Segmentation Using Gabor
Filter Clustering Method”, 2017 International
Seminar on Application for Technology of
Information and Communication.
[7] Manonmani S, Dr. Shanta Rangaswamy,”
Comparative analysis of combining various
enhancement filtering techniques for underwater
images”, 2018, IEEE International Conference on
Control, Power, Communication and Computing
Technologies.
[8] Manonmani, Lalitha V., Dr. Shanta Rangaswamy.
“Survey On Image Denoising Techniques”, 2016,
International Journal of Science, Engineering and
Technology Research (IJSETR).](https://image.slidesharecdn.com/irjet-v9i5795-221006074405-cdca2c36/85/DETECTION-AND-EXTRACTION-OF-SEA-MINE-FEATURES-USING-CNN-ARCHITECTURE-4-320.jpg)
DETECTION AND EXTRACTION OF SEA MINE FEATURES USING CNN ARCHITECTURE
- 1. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 09 Issue: 05 | May 2022 www.irjet.net p-ISSN: 2395-0072 © 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 3881 DETECTION AND EXTRACTION OF SEA MINE FEATURES USING CNN ARCHITECTURE Sheethal S, Soundarya S, Vidhisha V, Kiran G, Dr A B Rajendra Department of Information Science and Engineering VidyaVardhaka College of Engineering, Mysuru --------------------------------------------------------------------------***----------------------------------------------------------------------- Abstract The Conventional worry of navel asset is naval mines; these mines are stationary and were planted during war times and now they have been working as a threat to naval ships, and submarines. Detection of those naval mines has been one of the foremost risk-taking tasks, with modern technology various techniques are wont to detect these mines Using Ultrasonic signals, Symbolic pattern analysis of side-scan sonar images but detection through image processing has been one of the most challenging and efficient ones since it can solve the real-time problem with less error, the image classification model like uses FRCNN(Fast Region Convolutional Neural Network) algorithm to classify the objects as mine or not. The cloud platform is employed to watch the mine and as soon as the changes are observed the Android application will reflect the changes. Keywords: FRCNN, Neural network, Image processing, Deep Learning, ResNet, TensorFlow, Python 1. INTRODUCTION Nonmilitary mines, often known as aquatic mines, are used in combat. During a conflict, mines are used to destroy naval assets. It's also used in the defense industry, where mines operate as a border to protect the country's marine territory. These mines prevent hostile maritime assets from entering the unmarked territory. The enemy must search the entire region for mines. The opponent is forced to assault in an unmined place, where the defense is ready for a fight. Unlike the older mines, the ultramodern mines are detonated by pressing a button. The discovery of underwater mines is critical in ensuring that civilians are not endangered in any way. Mines aid in securing high-altitude defense bases and preventing the leakage of sensitive information. Battlegroups will be able to pinpoint the exact location of mines and avoid losses with the help of a reliable and cost-effective method. The neural network's operation is comparable to that of a mortal brain. It's used to show the relationship between data across a computer system. This artificial network is primarily based on Machine Literacy. If a dependable and cost-effective technology was utilized, battle groups would be able to find the exact location of mines and save lives. The neural network works in the same way that a human brain does. It's a diagram that depicts the connections between data in a computer system. Machine literacy is the foundation of the artificial network. Mask RCNN creates an offer about the region in the image where the object might be present and then constructs bounding boxes and masks at pixel position, as well as forecasts the object's class. For producing point vectors from raw pictures, Mask RCNN uses FPN as the backbone. For searching items in regions, use RPN. Anchor boxes are utilized to align the point vector with the position in the raw image, which can then be compared to ground verity while discovering and utilizing the idea of IoU value. 2. LITERATURE REVIEW [1] The paper mentions the simplest way of implementing various deep learning techniques, hence the modifications need to be done to the techniques this could be the major challenge of the paper. In this paper modules and sub-modules used are CNN, Autoencoders, Deep Belief Networks, and GAN. This paper provides an overview of the simplest ways to implement target recognition and hence not very efficient. [2] In this paper, the author Huu-Thu Nguyen, Eon-Ho Lee 1, and Sejin Lee specified sonar sensor needs and challenges to auto-detect submerged human bodies underwater. Sonar images need to be tested at different levels of polarization and intensities, the target background must be considered as there will be scatterings and noises in the sonar image. The same model needs to be retrained on sonar images of various polarization and intensity. [3] Self-Supervised Learning of Pretext-Invariant Representations is to construct image representations that are semantically meaningful via PIRL (Pretext Invariant Representation Learning) that do not require semantic annotations for a large training set of images. To achieve the highest single crop top-1 accuracy of all
- 2. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 09 Issue: 05 | May 2022 www.irjet.net p-ISSN: 2395-0072 © 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 3882 self-supervised learners that use a single ResNet-50 model. [4] In this paper, the author has specified more on various techniques used for detecting sea mines. The major challenge of this paper was images need to be manually annotated and the overhead of labelling is very large when dealing with huge datasets. In this paper, they have used the Mask RCNN model which Region- Based Convolutional Neural Network. [5] in this paper, the methodology involves a Masked RCNN module that comprises numerous convolutional layers. In this referral paper, the dataset is a set of images downloaded from the web hence a lot of pre- processing, labeling, and augmentation is required. [6] In this referral paper, Gabor Filter and K-means clustering algorithm is used. Gabor Filter is used for feature extraction, and the K-means Clustering algorithm is used for segmentation. The accuracy and efficiency of K-Means clustering are not accurate. Table 1. Literature Review [7] In this referral paper mainly two techniques are used, Homomorphic, CLAHE, and Wavelet Filtering Techniques These techniques help to enhance the image quality and removing the unwanted noise enhances the image quality. This survey paper also mentioned various other techniques for image enhancement such as a median filter for better quality and RGB for color level stretching. [8] In this referral paper they have specified image denoising techniques which include Spatial Domain Filter, Frequency Domain Filter, Mean Filter, Median Filter, and Adaptive Filter. The paper does not consider any hybrid filters which are more efficient in de-noising images and for a given dataset, the right kind of filter cannot be decided beforehand, we’ll have to implement each of the filters on the dataset. Title of the paper Authors of the paper Model/Sub Model used A Review on Deep Learning-Based Approaches for Automatic Sonar Target Recognition Dhiraj Neupane and Jongwon Seok CNN, Autoencoders, Deep Belief Networks, GAN Study on the Classification Performance of Underwater Sonar Image Classification Based on Convolutional Neural Networks for Detecting a Submerged Human Body Huu-Thu Nguyen, Eon-Ho Lee 1 and Sejin Lee AlexNet, GoogleNet Self-Supervised Learning of Pretext-Invariant Representations Ishan Misra, Laurens van der Maaten PIRL (Pretext Invariant Representation Learning) Underwater Mine Detection using Image Processing N Abhishek, Arjun, Bharathesh, Kavitha K S, Prof. Manonmani S, Dr. Shanta Rangaswamy MaskRCNN model Underwater Fish Detection Aditya Agarwal,Manonmani S,Gaurav Rawal,Tushar Malani,Navjeet Anand. Masked RCNN Image Segmentation Using Gabor Filter and K-Means Clustering Method Agyztia Premana, Akhmad Pandhu Wijaya, Moch Arief Soeleman. Gabor Filter and K-means clustering algorithm. Comparative analysis of combining various enhancement filtering techniques for underwater images Manonmani S, Dr. Shanta Rangaswamy Homomorphic, CLAHE and Wavelet Filtering Techniques Survey On Image Denoising Techniques Manonmani,Lalitha V.,Dr. Shanta Rangaswamy. Spatial Domain Filter, Frequency Domain Filter, Mean Filter, Median Filter and Adaptive Filter
- 3. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 09 Issue: 05 | May 2022 www.irjet.net p-ISSN: 2395-0072 © 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 3883 3. BACKGROUND 3.1 DEEP NEURAL NETWORK A deep neural network is an artificial neural network (ANN) with several layers between the input and output layers (DNN) as shown in Fig.1. Neurons, synapses, weights, biases, and functions are all basic components of neural networks, which come in a range of forms and sizes. Fig.1 Deep Neural Network Architecture 3.2 CONVOLUTIONAL NEURAL NETWORK The CNN is a sort of neural network as shown in Fig.2 that is quite similar to human vision and thought. Various computer vision applications have become an important part of this over time. CNN’s were constructed for the first time in the 1980s. At the time, this neural network was the best at recognizing manual digits. The code reader has mostly been used or built to read zip codes, pin codes, and other codes of a similar kind. Fig.2 A traditional CNN Architecture 3.3 GENERATIVE ADVERSARIAL NETWORK A generative adversarial network (GAN) is a machine learning (ML) model in which two neural networks compete to improve the accuracy of their predictions as shown in Fig.1. GANs are frequently unsupervised and learn by playing a cooperative zero-sum game Fig.3 Generative Adversarial Network 4. CONCLUSION The detection of the sea mines can be done in real- time or in real-time in order to be more effective in its usage to the naval forces of the country. A more extensive dataset that involves actual SONAR generated images and contains a wide selection of mines of all shapes, sizes, and specifications, images of different visibilities, etc can be used to train a more comprehensive detection model that detects accurately across all these variations. Models may be trained using different machine-learning techniques like YONO v3 and a few more recent approaches which can lead to improved performance and outcomes. REFERENCES [1] Dhiraj Neupane and Jongwon Seok, A Review on Deep Learning-Based Approaches for Automatic Sonar Target Recognition, MDPI Electronics, 2020. [2] Huu-Thu Nguyen, Eon-Ho Lee 1 and Sejin Lee, Study on the Classification Performance of Underwater Sonar Image Classification Based on Convolutional Neural Networks for Detecting a Submerged Human Body, MDPI Sensors, 2019. [3] Ishan Misra, Laurens van der Maaten, Self-Supervised Learning of Pretext-Invariant Representations, arXiV, 2019. [4] N Abhishek, Arjun, Bharathesh, Kavitha KS, Prof. Manonmani S, Dr. Shanta Rangaswamy, “Underwater Mine Detection using Image Processing”. 2020, International Research Journal of Engineering and Technology (IRJET). [5] Aditya Agarwal, Manonmani S, Gaurav Rawal, Tushar Malani, Navjeet Anand, “Underwater Fish Detection”, 2020, International Journal of Engineering Research & Technology (IJERT).
- 4. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 09 Issue: 05 | May 2022 www.irjet.net p-ISSN: 2395-0072 © 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 3884 [6] Agyztia Premana, Akhmad Pandu Wijaya, Moch Arief Soeleman, “Image Segmentation Using Gabor Filter Clustering Method”, 2017 International Seminar on Application for Technology of Information and Communication. [7] Manonmani S, Dr. Shanta Rangaswamy,” Comparative analysis of combining various enhancement filtering techniques for underwater images”, 2018, IEEE International Conference on Control, Power, Communication and Computing Technologies. [8] Manonmani, Lalitha V., Dr. Shanta Rangaswamy. “Survey On Image Denoising Techniques”, 2016, International Journal of Science, Engineering and Technology Research (IJSETR).