CAM?SwinUNet A Hybrid Attention?Driven Transformer for Camouflage vs. Non?Camouflage Classification

This project aims to develop an intelligent and secure web-based image classification system capable of detecting whether an uploaded image contains camouflage or non-camouflage objects. The system utilizes deep learning techniques by training and comparing two advanced architectures, EfficientNet-B0 and CAM-SwinUNet, using the CAMO-COCO V1.0 dataset to evaluate their performance in terms of validation accuracy and F1-score. To improve model generalization and prediction reliability, data augmentation and transfer learning techniques are incorporated during training. In addition to the machine learning component, the project includes the development of a secure Flask web application that supports user registration, login authentication, and session management. User account details and prediction records are stored in a MySQL database with proper relational constraints to ensure data integrity and security. The application also provides a dedicated history page where users can review their complete prediction history. Furthermore, all user inputs, including usernames, email addresses, and passwords, are validated on both the frontend and backend to enhance application security and ensure reliable user interaction.