Abstract

Farmers typically lack knowledge of diagnosing and controlling different apple diseases. However, some apple diseases have visual symptoms that the eye can diagnose, but their diagnosis is time-consuming and costly. Earlier detection of infections and crop health issues can enable control of fruit diseases through effective management approaches. A proposed solution is to design an automatic disease diagnosis system using image processing techniques. This paper presents a low-cost method for apple disease diagnosis using a neural network and classifying fruit into four classes: scab, bitter rot, black rot, and healthy. This method involves different models (VGG-16, VGG-19, and ViT) and feature extraction to tackle this problem; we introduce the Vision Transformer image detection (ViT) Model to classify apple diseases. Initially, preprocessing was performed using an image processor, feature scaling, feature extraction, feature normalization, data augmentation, and hyperparameter tuning. Finally, the classification section compares ViT with other models, such as the Visual Geometry Group – 16 layers (VGG-16) and the Visual Geometry Group – 19 layers (VGG-19). The average accuracy of VGG-16 for disease classification is 97.03%. The average accuracy of the Visual Geometry Group – 19 layers (VGG-19) for disease classification is 95.5%. The proposed ViT model can significantly improve the accuracy of apple fruit disease classification. The average accuracy of disease classification achieved with the current proposed techniques was 99.17%.