A modular vision transformer and machine learning framework for scalable multi-crop plant disease classification

Plant disease classification is essential for precision agriculture, yet existing deep learning approaches often face challenges in scalability, generalization, and computational complexity across diverse crop species. This study presents a two-phase hybrid framework for multi-crop plant disease classification that combines deep feature extraction with machine learning-based classification. In the first phase, four pretrained CNN models (VGG16, ResNet50, InceptionV3, and MobileNetV2) are evaluated as feature extractors on a 21-class dataset comprising tomato, chilli, and potato diseases. In the second phase, a Vision Transformer (ViT) is employed to extract global feature representations from a 25-class dataset containing diseases from seven crop species, which are subsequently classified using machine learning classifiers. Experimental results show that MobileNetV2-SVM achieves the highest performance among CNN-based configurations with an accuracy of 89.44%, while the ViT-SVM framework attains 98% accuracy for multi-crop disease classification. The findings demonstrate the effectiveness of transformer-derived feature representations for generalized plant disease recognition and highlight the potential of hybrid deep feature learning frameworks for precision agriculture applications .