Vegetable Disease Identification Algorithm Based on Improved Swin Transformer

The background of vegetable disease images is complex, with the diseased areas exhibiting minimal contrast against the backdrop, making the precise identification of foliar diseases in vegetables a significant challenge in agricultural production. This study aims to address the insufficiencies in recognition accuracy inherent to traditional identification models by implementing enhancements based on the Swin Transformer model. The paper proposes the design of a local perception module that, by expanding the size of the perceptual field, enhances the capability for local feature extraction to optimize model recognition accuracy. Additionally, the Poly Loss function is employed to address the issue of data imbalance within the vegetable disease dataset. Experimental results indicate that the model proposed herein consistently outperforms other recognition models in comparative assessments, achieving an accuracy rate of 97.14%, which represents an approximate 3.6% improvement over the baseline model. The model holds promising prospects for application in identifying foliar diseases of vegetables.