Pox Detection using image dataset

 Pox diseases, including Chickenpox, Measles, Smallpox, and Monkeypox, pose challenges in clinical diagnosis due to their similar dermatological manifestations. Accurate classification of these conditions is essential for early detection and appropriate medical intervention. This research presents a deep learning-based approach for automated pox detection using dermal images categorized into six classes: Chicken_Pox, Measles, Monkey_Pox, Normal, Small_Pox, and Unknown. A dataset consisting of 4099 images for training and 1104 images for evaluation was utilized in this study.The framework employs MobileNet for feature extraction, leveraging its lightweight architecture to efficiently process skin lesion images. Three model architectures were developed and trained using different optimizers—Adam, SGD, and RMSprop: (1) a convolutional architecture using InceptionNet, (2) a hybrid MobileNet-LSTM model to capture both spatial and temporal features, and (3) a Deep Belief Network (DBN) model that utilizes a generative approach for learning deep representations. Each model was evaluated under varied optimization strategies to understand the impact of training dynamics on classification performance.This study aims to provide a structured comparison of these deep learning models in the context of pox classification using visual cues. The proposed work emphasizes the design, training, and comparative evaluation of models tailored for medical image classification, contributing to the growing body of research in AI-assisted disease detection.

Keywords:
Pox Detection, MobileNet, InceptionNet, Deep Belief Network, LSTM, Convolutional Neural Network, Skin Lesion Classification, Image Analysis, Optimizers, Medical Image Processing