Animal detection using pre trained model for all usages

Bangalore Institute of Technology
K.R. Road, V.V. Pura, Bengaluru.-560004.
Department of Computer Science & Engineering
2025-26
Presentation by:
Project id:22P33
1BI22CS106 POORVI H R
1BI22CS107 PRADEEPA ACHARYA
1BI22CS108 PRAGATI JAYARAM RATHOD
1BI22CS110 PRAJWAL S MADIVALR
Under the Guidance of
D R Nagamani
Assistant
Professor
Department of CSE, BIT
“Melanoma Spotter: A Hybrid Deep Learning Approach with VGG16 and
DenseNeT121”

Bangalore Institute of
Technology
Department of Computer Science &
Engineering
INTRODUCTION
• Melanoma is a dangerous form of skin cancer that originates from melanocytes, and early
detection greatly improves treatment success.
• Accurate diagnosis is challenging due to the visual similarity between melanoma and benign
skin lesions.
• Convolutional Neural Networks (CNNs) are effective in analyzing dermoscopic images and
identifying melanoma with high accuracy.
• This study uses VGG16 and DenseNet121 models to classify lesions as melanoma or non-
melanoma, supporting early diagnosis.

Technology
Engineering
LITERATURE SURVEY
1. Classifying Melanoma in ISIC Dermoscopic Images Using Efficient CNNs and Deep
Transfer
Author: Habeba Mahmoud et al.
Year Published: 2023
Journal/Conference: Traitement du Signal
Proposed Idea: The study applies deep transfer learning using CNN models to detect melanoma
in dermoscopic images. It compares various model architectures to determine the best performer.

Technology
Engineering
Methodology: Used the ISIC dermoscopic image dataset for training and evaluation. Employed
pre-trained CNN models such as ResNet-50 and VGG, which were fine-tuned for the specific task
of skin cancer classification. The models underwent training with labeled melanoma and non-
melanoma images to optimize their predictive performance
Drawbacks:
•Requires high computational resources due to the depth and size of CNN models.
•Pre-trained models may not effectively capture skin-specific features.

Technology
Engineering
2. Melanoma Skin Cancer Detection Using Recent Deep Learning Models
Authors: Takfarines Guergueb; Moulay A Akhloufi
Year: 2021
Journal/conference: 43rd
Annual International Conference of the IEEE Engineering in medincine
and biology society
Proposed idea: The proposed model is efficientnetb7 which out performs all other models it has
achieved an accuracy of 99.01%

Technology
Engineering
Methodology: Various deep learnig models like DenseNet, mobileNet, VGG, ResNet and 16
other models have been trained using a vast dataset which is obtained by combining ISIC 2017,
2018, 2019 datasets.
Drawbacks & Limitations:
•Imbalance in available dataset.

Technology
Engineering
3. A Hybrid CNN RF model for accurate and explainable skin lesion classification
Authors: Shiva Mehta, Aseem Aneja
Year: 2025
Journal: International Conference On Automation And Computation
Proposed Idea: This paper presents a hybrid model of ResNet50 and Random forest. The
ResNet50 is used for the feature extraction and the random forest is used for classifying into
different categories.

Technology
Engineering
Methodology: the methodology involves using the ISIC dataset, where CNN and Random forest
hybrid model with Grid Search CV are used to combine to give 89.33% accuracy.
Drawbacks:
•Less accuracy
•Consumes more for the cross validation

Technology
Engineering
4. A Comprehensive Analysis of Melanoma Skin Cancer Detection Using Machine
Learning and Deep Learning Algorithms
Authors: K. Muruganandam, S. Elangovan
Journal/Conference: International Research Journal of Modernization in Engineering
Technology and Science (IRJMETS)
Proposed model:
The authors proposed a hybrid framework integrating both Machine Learning (ML) and Deep
Learning (DL) techniques for early and accurate detection of melanoma skin cancer.

Technology
Engineering
Methodology: The methodology involves preprocessing dermoscopic images from the ISIC
dataset, followed by feature extraction using CNNs. These features are then classified using
both deep learning and traditional machine learning algorithms like SVM and KNN
Drawbacks / Limitations:
•Imbalanced dataset may bias the model.
•Deep learning model risks overfitting.
•High computational resources are needed.

Technology
Engineering
5. Detection and optimization of skin cancer using deep learning
Authors: S. Balambigai et al.
Journal/Conference: Journal of Physics: Conference Series, Volume 2318, Paper 012040
Published by IOP Publishing
Proposed idea: The input dermoscopic images are pre-processed before training a CNN model.
The model classifies images into seven skin lesion types, including melanoma and nevus.
Random search is used to optimize CNN hyper parameters for better performance

Technology
Engineering
Methodology: The methodology includes collecting and pre-processing breast cancer data,
selecting key features, and applying classification algorithms like Decision Tree, Naive Bayes,
KNN, and SVM. Model performance is evaluated using standard metrics such as accuracy,
precision, recall, and F1-score
Drawback :
•Hyper-parameter tuning is very tedious.
•Optimization process consumes more time.
•Difficult to achieve best model.

Technology
Engineering
6. Advanced Melanoma Detection with InceptionResNetV2 using Skin Lesion Images
Author: Aditya Kumar; Leema Nelson
Year of publication: 2025
Conference/journal: International conference on Multi-Agent systems for collaborative
intelligence
Proposed idea: This work developed a deep-learning model based on InceptionResNetV2
architecture to meet the the demand for accurate and quick melanoma diagnosis.the model was
trained and evaluated using a carefully curated dataset of 13879 skin lesion images.

Technology
Engineering
Methodology: They have used the mcid(melanoma cancer image dataset) dataset. It uses softmax
activation function for multi-class classification and achieved an accuracy of 95.05%
Drawbacks:
•Limited dataset diversity may hinder model generalization across different skin types
•Performance may degrade with poor-quality or inconsistent images.

Technology
Engineering
7. A Study on the Application of Machine Learning and Deep Learning Techniques for Skin
Cancer Detection
Author: Hritwik Ghosh, Irfan Sadiq Rahat, Sachi Nandan Mohanty, J. V. R. Ravindra, Abdus
Sobur
Journal/conference: The International Journal of Computer and Systems Engineering
Year Published:2024
Proposed idea: The paper proposes a comparative analysis of machine learning and deep
learning techniques for skin cancer detection. It aims to identify the most effective approach by
evaluating their performance on dermoscopic image datasets.

Technology
Engineering
Methodology: The study explores both machine learning and deep learning methods for skin
cancer detection.ML models rely on manually extracted features from dermoscopic images.
DL models like CNNs are trained end-to-end to learn patterns and classify skin lesions.
Drawbacks:
•Deep learning needs large, labeled datasets to perform well.
•Traditional ML requires manual feature extraction, limiting accuracy and scalability.

Technology
Engineering
8. Skin Cancer Detection using transfer learning model and ensemble approach to
enchancd diagnostic accuracy
Author: Khan, Hee-cheol Kim et al.
Year:2025
Journal/conference: 27th
international conference on advanced communication
Proposed idea: This work has proposed the hybrid model combining inception, xception,
VGG19, resnet50.

Technology
Engineering
Methodology: The study explores both machine learning and transfer learning methods for skin
cancer detection. It obtained an accuracy of 94.50% using ISIC dataset.
Drawbacks:
•Ensemble of multiple models increases computational complexity
•Training time is significantly longer for hybrid architectures.

Technology
Engineering
9. DeepSkinNet: A Deep Learning Model for Skin Cancer Detection
Author: A P Abhiram, S M Anzar et al.
Year Published:2022
Journal: IEE Explore
Proposed idea: DeepSkinNet model is proposed and tested. It is compared with contemporary
models such as AlexNet, VGG-16, InceptionV3. The proposed model achieves better
accuracy than other models .

Technology
Engineering
Methodology: The methodology involves using the HAM10000 dataset, where images are
preprocessed with resizing, normalization, and data augmentation techniques. ReLu is the
activation function used.
Drawbacks:
•Data imbalance
•High computational requirements
•Lack of interpretability in CNN decisions

Technology
Engineering
10. Innovative computer-aided techniques for early detection of melanoma using
dermoscopic image analysis
Authors: G. Ravikumar, Susanta Kumar Satpathy
Conference: International Conference on electronics and renewable systems(ICEARS)
Year of publication: 2025
Proposed Idea: The proposed model involves preprocessing dermoscopic images to enhance
quality, segmenting lesions, extracting key features, and classifying them using a deep learning
model. It achieves 94% accuracy, 92% sensitivity, and 96% specificity in melanoma detection.

Technology
Engineering
Methodology: The proposed methodology uses dermoscopic images from the ISIC 2020
dataset and applies preprocessing, segmentation, feature extraction, and deep learning
classification to detect melanoma. Data enhancement techniques like resizing, noise reduction,
and augmentation improve accuracy and robustness.
•Faces challenges with class imbalance between benign and malignant cases.
•Lacks validation in real-world clinical settings.
•High computational requirements restrict deployment on low-resource devices.

Technology
Engineering
11. Enhancing Skin Cancer Detection: An Integrated Approach Using Deep Learning and
Metaheuristic Algorithms
Authors:Safia Mohamed, Zaher Al Aghbari, Ahmed M. Khedr
Conference: 2nd International Conference on Advanced Innovations in Smart Cities (ICAISC)
Year of publication:2025
Proposed Idea: Various models combined deep learning with metaheuristics for skin cancer
detection, using CNNs, PSO, and optimization algorithms to improve feature selection,
segmentation, and classification, achieving high accuracy across datasets like PH2 and ISIC

Technology
Engineering
Methodology: The proposed method uses the PH2 dataset and applies preprocessing steps like
hair removal, image normalization, and data augmentation. Features are extracted using a fine-
tuned ResNet-50 model, and Particle Swarm Optimization (PSO) selects the most relevant
features. Finally, classification is performed using K-Nearest Neighbors (KNN)
•model is limited by its small dataset size
•lack of external validation
•High computational costs

Technology
Engineering
12. Melanoma Skin Cancer Detection Using InvolutionNet
Authors: Shubham Chaudhary, Vishal Gupta, Kapil Sharma
Conference: Control Instrumentation System Conference (CISCON)
Proposed Idea: This study proposes InvolutionNet model for melanoma classification. The
model showcases competitive or superior performance compared to existing CNN architectures
proposed in the literature.

Technology
Engineering
Methodology: The proposed method uses the custom dataset and addresses the limitations
of CNNs. The proposed InvolutionNet model, uses an involution operation. This operation
enables our model to address the shortcomings of the CNN model. The InvolutionNet model
achieves an overall accuracy of 87.0%.
•model is limited by its small dataset size
•lack of external validation

Technology
Engineering
13. An Intelligent System For Skin Cancer Detection Using Deep Learning Techniques
Authors: Vani Rajasekar, Nirmala Devi K, Sathya K, Santhosh A
Conference: International Conference on Advances in Data Engineering and Intelligent
Computing Systems (ADICS)
Proposed Idea: The proposed work provides a reliable method for classifying skin cancer using
convolutional neural networks. The CNN model uses deep learning to automatically identify
complex patterns and features and for the accurate and automated classification of skin lesions
into benign and malignant categories.

Technology
Engineering
Methodology: The dataset used in this research is the "Skin Cancer MNIST HAM10000"
dataset. the suggested CNN-based method demonstrates impressive accuracy, achieving an
overall accuracy rate of approximately 98.10%.
•Limited dataset

Technology
Engineering
EXISTING SYSTEMS
• Visual Inspection: Relies on doctors' judgment to identify skin cancer but can be subjective,
inaccurate, and inaccessible in remote areas.
• Biopsy: An invasive, costly method with risks like pain, scarring, or disease spread.
• Dermoscopy: A non-invasive method requiring expert skill and costly equipment, limiting its
use in some areas.
• AI/ML Systems: These use algorithms for automated skin lesion detection but require large
datasets, struggle with low-quality data, and lack interpretability due to model complexity.

Technology
Engineering
PROBLEM STATEMENT
To Develop an accurate and efficient skin cancer detection system using VGG16 and
DenseNet121 to improve early detection.
Input: Image of a skin lesion (dermatoscopic images).
Output: Classification of the lesion as melanoma and non melanoma with confidence score and
optional visual explanation.

Technology
Engineering
OBJECTIVES
• To build a reliable and fast skin cancer detection model.
• To minimize errors, delays, and access gaps in skin cancer diagnosis
• To implement an automated system for skin cancer detection.
• To design an accurate and efficient skin cancer detection system.
• To integrate skin cancer scanning using AI integrated tools like VGG16 and DenseNet121.

Technology
Engineering
Proposed System
Hybrid Model: The system will combine two deep learning models, VGG16 and DenseNet121,
to enhance melanoma detection accuracy.
VGG16: Known for its depth and ability to capture complex features, VGG16 utilizes small
convolutional filters to learn detailed patterns from dermoscopic images.
DenseNet121: DenseNet121 improves on feature extraction by using dense connections,
promoting feature reuse and better gradient flow through the network.

Technology
Engineering
Model Combination: The hybrid model combines VGG16 and DenseNet121 to enhance
feature extraction and improve melanoma classification from dermoscopic images.
Training Plan: The hybrid model will be trained on a large, labeled dermoscopic dataset.
Transfer learning will be used to leverage pre-trained weights for faster training and better
results.
Expected Outcome: The proposed hybrid model is expected to provide higher accuracy in
classifying melanoma vs. non-melanoma lesions by integrating the feature extraction
capabilities of VGG16 and DenseNet121.

Technology
Engineering
Sample Dataset
Three general datasets which can be used for melanoma skin cancer detection system are:-
•HAM10000 : The HAM10000 dataset contains 10,015 dermatoscopic images of pigmented
skin lesions across 7 categories, used to train and evaluate machine learning models in
dermatology.
• PH2 : The PH2 dataset is a dermoscopic image dataset containing 200 high-resolution images
of melanocytic lesions, including common nevi, atypical nevi, and melanomas. It is widely used
for evaluating algorithms in skin lesion segmentation and classification

Technology
Engineering
• ISIC 2017: ISIC 2017 dataset is part of a skin lesion analysis challenge and includes 2,750
dermoscopic images for tasks like lesion segmentation, dermoscopic feature detection, and
disease classification.

Technology
Engineering
Sample Dataset

Technology
Engineering

Bangalore Institute of Technology
Engineering
PROJECT PLANNING
Gantt Chart

Technology
Engineering
Applications
• Early Diagnosis & Screening
• Telemedicine & Remote Consultation
• Dermatology Clinics & Hospitals
• Research & Medical Advancements
• Integration with Smart Wearables

Technology
Engineering
References
[1] H. Mahmoud, A. A. Ewees, A. Hefnawy, M. A. Ismail and M. A. El-dosuky, "Classifying Melanoma in ISIC
Dermoscopic Images Using Efficient CNNs and Deep Transfer," Traitement du Signal, vol. 40, no. 2, pp. 399–407,
2023.
[2] J. Qadir, S. Anwar, M. A. Khan, M. T. Mahmood and A. Hussain, "Melanoma Skin Cancer Detection Using
Ensemble of Machine Learning Models Considering Deep Feature Embeddings," Procedia Computer Science, vol.
227, pp. 1144–1151, 2024.
[3] Y. Lu, K. Wang, L. Zhang, X. Li and M. Chen, "Deep Learning Algorithms for Melanoma Detection Using
Dermoscopic Images: A Systematic Review and Meta-Analysis," Journal of Investigative Dermatology, vol. 144,
no. 2, pp. 251–263, 2024.

Technology
Engineering
References
[4] K. Muruganandam and S. Elangovan, "A Comprehensive Analysis of Melanoma Skin Cancer Detection Using
Machine Learning and Deep Learning Algorithms," International Research Journal of Modernization in
Engineering Technology and Science (IRJMETS), vol. 4, no. 5, pp. 1422–1429, 2022.
[5] S. Balambigai, M. Karthikeyan, R. Sangeetha and V. R. Manikandan, "Detection and Optimization of Skin
Cancer Using Deep Learning," Journal of Physics: Conference Series, vol. 2318, no. 1, p. 012040, 2022, IOP
Publishing
[6] P. M. Kulkarni, Y. M. Kotecha, A. Jaiswal and H. M. Kulkarni, "Melanoma Skin Cancer Detection Using Deep
Learning and Classical Machine Learning Techniques," International Journal of Environmental Research and
Public Health, vol. 18, no. 10, p. 5479, 2021.

Technology
Engineering
References
[7] H. Ghosh, I. S. Rahat, S. N. Mohanty, J. V. R. Ravindra and A. Sobur, "A Study on the Application of
Machine Learning and Deep Learning Techniques for Skin Cancer Detection," 2024.
[8] N. Nanthini, D. Aishwarya, A. Simon and N. B. Vishnupriya, "Detection of Melanoma Skin Cancer Using
Deep Learning," in Proceedings of the International Conference on Smart Electronics and Communication
(ICOSEC), Tirunelveli, India, 2022, pp. 1434–1439.
[9] Kajal Kathuria, Anita Sahoo, Chakesh Kumar Jain, “Prior Detection of Melanoma Skin Cancer Using Deep
Learning”, 2024

Technology
Engineering
THANK YOU

Animal detection using pre trained model for all usages

More Related Content

Similar to Animal detection using pre trained model for all usages (20)

Recently uploaded (20)

Animal detection using pre trained model for all usages