Monkeypox Lesions Detection With a Lightweight Hybrid Model

Project Code :TCMAPY2140

Objective

This project presents a lightweight hybrid deep learning approach for detecting Monkeypox lesions from skin images. Four models are explored: ConvNeXt-Tiny with CBAM, EfficientNetV2-S with SE and Deformable Convolution, RegNetY with Coordinate Attention, and a VGG16 baseline. The models are trained on a six-class dataset including Monkeypox, Chickenpox, Measles, Smallpox, Normal, and Unknown cases. Attention mechanisms and positional awareness are integrated to enhance feature extraction while reducing computational cost. A Flask-based web application enables user registration, login, image classification, and logout.The proposed system demonstrates an end-to-end flask based implementation pipeline. Overall, the work highlights efficient and accurate lesion classification suitable for dermatological applications.

Abstract

The rapid and accurate detection of skin lesions is critical for effective disease management. This study proposes a lightweight hybrid approach for detecting Monkeypox lesions from skin images. The project explores four models: ConvNeXt-Tiny combined with CBAM, EfficientNetV2-S enhanced with SE and Deformable Convolution, RegNetY integrated with Coordinate Attention, and a VGG16 baseline model. These models are trained and evaluated on a dataset consisting of six classes: Chicken_Pox, Measles, Monkey_Pox, Normal, Small_Pox, and unknown. The hybrid architectures incorporate attention mechanisms and positional awareness to improve feature extraction while maintaining computational efficiency. A web-based interface developed using Flask allows users to register, log in, classify images, and log out. The results demonstrate that the hybrid models outperform the baseline VGG16 in accuracy, sensitivity, and specificity. This research provides a foundation for developing lightweight models for lesion classification, emphasizing optimized performance and reduced computational complexity. The project also demonstrates an integrated pipeline for deployment in a user-friendly interface. This approach can be extended to other dermatological applications where accurate classification of skin lesions is essential.

Keywords: Monkeypox, Skin Lesion Detection, ConvNeXt-Tiny, CBAM, EfficientNetV2-S, SE, Deformable Convolution, RegNetY, Coordinate Attention, VGG16

NOTE: Without the concern of our team, please don't submit to the college. This Abstract varies based on student requirements.