Research on the Identification of Tomato Leaf Diseases Based on Multi-scale Feature Fusion in EfficientNetV2-S

Plant diseases usually manifest on leaves. By judging the state of tomato leaves, we can diagnose the disease in time and take corresponding control measures, thereby effectively reducing crop damage from diseases. Therefore, this article proposes a new tomato leaf disease recognition method. This method uses BiFPN to fuse the three different scale features extracted by EfficientNetV2-S network model, which promotes the transfer and fusion of useful information between different stages of the network, so as to improve the performance of the model. At the same time, the global attention mechanism (GAM) is introduced into the fusion process to capture global features, obtain more abundant feature information. Then, the Encoder module of Vision Transformer is inserted after each GAM attention mechanism to enhance the features and help improve the richness and abstraction ability of the extracted features. The proposed model has achieved an identification accuracy of 97.30%, and its loss convergence is faster, with significant improvement in macro-average and weighted average. The ablation experiments show that the model performs best on various indicators, surpassing the baseline model and fusion model in this paper, greatly improving the performance of image classification and recognition tasks. This method effectively improves the recognition ability and detection accuracy of disease areas, and can achieve high-precision recognition in strong interference environments, with good robustness and adaptability. At the same time, it solves the problems of weak generalization ability, low precision, and low computational efficiency in disease recognition.