Multi Plant Disease Diagnosis: Leveraging Vision Transformers and Transfer Learning

Plant diseases greatly affect agricultural productivity resulting in economic losses and reduction of crop production. Early and precise identification is crucial for good crop managementbecause traditional disease testing methods are generally expensive, time-consuming and lessefficient. This paper proposes a hybrid deep learning framework of Transfer Learning andVision Transformer (ViT) for plant diseases’ classification. We study 5 pre-trained models,ResNet50, MobileNetV2, EfficientNetB0, VGG19 and DenseNet121 on diverse plant diseasedatasets such as tomato, apple, maize, grape and potato leaves. The performance of thesemodels in extracting discriminative characteristics for disease identification is compared inthe study. Experimental results show that the maximum accuracy of classification is 99.40%of the ResNet50 based Vision Transformer model, which exceeds the other pre-trained models. The proposed method provides a reliable and efficient solution to the problem of detection of plant leaf diseases and contributes to the increase of agricultural output and protection of crops.