ATTENTION-GUIDED HYBRID SVM–VISION TRANSFORMER FRAMEWORK FOR ACCURATE DETECTION OF APPLE LEAF DISEASES IN INDIAN ORCHARDS
Piyush Sharma, Devi Prasad Sharma, Sulabh Bansal
Department of Information Technology, Manipal University Jaipur, India
Abstract: Apple cultivation in India is frequently affected by fungal and bacterial infections that are hard to detect in early stages, often leading to serious yield losses. Reliable automated diagnosis can help farmers take timely action, but many existing deep learning models struggle to capture both fine-grained local symptoms and broader visual patterns on leaves. This study proposes an attention-guided hybrid learning framework that combines Support Vector Machines (SVM) with Vision Transformers (ViT) for precise apple leaf disease classification. The model integrates the strong decision boundaries of SVM with the global feature learning capability of ViT, while an optimized attention mechanism emphasizes disease-relevant regions in the images. This design helps the system focus on subtle texture variations and lesion patterns that are typically missed by conventional convolution-based networks. Experiments were conducted on a dataset of Indian apple leaf images containing angular leaf spot, bean rust, and healthy samples. Performance was evaluated using cross-validation and benchmarked against several established deep learning models. The proposed hybrid model achieved an accuracy of 98.7%, precision of 98.5%, recall of 98.3%, and F1-score of 98.4%, outperforming comparison models that remained below 95% accuracy. The attention maps also provide visual insight into the model’s decision process, improving transparency. The results suggest that the proposed framework can serve as a reliable and interpretable tool for early disease detection in precision agriculture applications.
Keywords: Apple Fruit disease, Hybrid ML model, ViT, SVM, CNN, Prediction
VOLUME 10 ISSUE 02 2026: 162 – 191