MANGO LEAF DISEASE DETECTION USING SWIN TRANSFORMER (SWINT): A DEEP LEARNING APPROACH

Abstract

Growing application of deep learning in agriculture has boosted the efficient and accurate techniques for plant disease detection. This is because there is a requirement to enhance the yield of the crops and avoid economic loss. Mango (Mangifera indica L.) is a fruit tree that is widely grown in the tropics. Its development is further endangered by diseases such as Anthracnose, Bacterial Canker, Cutting Weevil, Die Back, Gall Midge, Powdery Mildew, and Sooty Mould. In this paper, we present the deep learning and transfer learning-based mango leaf disease classifier using EfficientNetB0 and Swin Transformer Tiny models. The dataset contained 4,000 labeled mango leaf images and represented eight disease classes. Swin Transformer Tiny model, trained on Tensor Flow and pre-trained on ImageNet, was used. It achieved 99.44% training accuracy, 99.87% validation accuracy, and 99.87% test accuracy. EfficientNetB0 baseline model obtained 92.3% accuracy. It is observed in the results that the transformer models are efficient to detect subtle color and texture changes in leaves. We employed precision, recall, F1-score, and AUC-ROC to approximate the models' reliability. The observation is that Swin Transformer Tiny model functions with high-precision disease classification without any overfitting. This implies that it can be adapted to operate efficiently in actual-world smart farming conditions. The model is able to be used for facilitating early disease diagnosis of farmers and researchers from standard camera streams.