A Novel Approach for Classification and Detection of Apple Leaf Disease Using Enhanced RBVT-Net with Transfer Learning and YoloV7

Apples are a popular fruit worldwide, valued for their rich nutritional content and associated health benefits, such as reducing the risks for cancer, diabetes, and heart disease. However, apple production faces significant challenges from diseases and pests, which can lead to substantial losses for farmers. Early and accurate detection of apple diseases is crucial for effective disease management and improved yield, however the identification process is complicated by the similarity of symptoms across different diseases. Variations in symptom expression, such as leaf discoloration and spot patterns, further complicate disease diagnosis, making it challenging to distinguish between pathogens. This study utilizes a manually collected and expertly validated dataset of 8,000 apple leaf images from orchards in Himachal Pradesh and Uttarakhand, India, encompassing three common diseases: Marssonina leaf blotch, Alternaria leaf spot, and powdery mildew. A novel approach is proposed that utilizes a Transfer Learning (TL) Enhanced Residual BottleNeck Vision Transformer (RBVT-Net) model for the classification of Apple Leaf Disease (ALD) and disease detection, leveraging YOLOv7 for high-precision identification. Extensive experiments demonstrated the effectiveness of the proposed model, achieving a classification accuracy of 98.58% and a mean Average Precision (mAP) of 0.599 for disease detection using YOLOv7. These results underscore the potential of TL-enhanced models in aiding apple disease management, offering a promising solution to support farmers in early disease identification and better crop quality.