Efficient Wheat Disease Identification

Wheat is an essential crop that requires careful monitoring to prevent yield losses caused by diseases. Efficient identification of wheat diseases can enhance analysis and support corrective measures. This project presents a system for accurate wheat disease identification using advanced deep learning techniques. Four models are implemented: CNN-ViT hybrid, Swin-Tiny + ViT hybrid, Lightweight Swin-SHARP, and Swin-SHARP + ViT with attention fusion. Each model leverages feature extraction and attention mechanisms to capture both local and global patterns in leaf images. The dataset used contains images of wheat leaves affected by yellow rust and healthy leaves. Images are pre-processed and augmented to improve model generalization. The system is deployed using a web interface built with HTML, CSS, and JavaScript for front-end and Flask for back-end. Users can register, log in, upload leaf images, and receive classification results. The models are evaluated based on accuracy, precision, recall, and F1-score to identify the most efficient approach. This framework aims to reduce computational overhead while maintaining high classification performance. The proposed system provides an accessible solution for automated disease recognition, enabling effective analysis and decision-making.

Keywords: Wheat, Disease Identification, CNN-ViT, Swin Transformer, ViT, Attention Fusion, Deep Learning, Leaf Images, Classification, Hybrid Models