Swin Transformer-based Disease Identification Model for Apple Plants

Diseases may cause substantial loss to the productivity of commercial crops like apple, thereby directly affecting global food security. The prompt and precise identification of diseases is vital for fruitful plant disease management. However, conventional manual disease detection approaches are tedious, time-consuming in nature, and unsuitable for large-scale applications. In the present study, we have introduced a lightweight deep learning model designed to identify the diseases of apple plants using RGB images. Our proposed approach leverages the concept of shifted window-based transformer network that is inspired by Swin transformer networks. To create a representative dataset, we collected real-in-field images of apple plants from several apple farms using mobile devices. Our proposed network achieved a significant classification accuracy of approx. 98.13% on the testing dataset, indicating its substantial efficacy. We also conducted a comprehensive performance analysis comparing our proposed network with the state-of-the-art (SOTA) CNN architectures, which demonstrates the superiority of our proposed network.