Grape Leaf Black Rot Detection Based on Lightweight Network

In order to improve the detection of grape leaf black rot and facilitate deployment on mobile devices, this paper introduced a lightweight network detection model called SM-YOLO v4 based on YOLO v4 architecture. The model adopts MobileNet V1 as the backbone feature extraction network, effectively reducing the number of model parameters. Meanwhile, deep separable convolutional blocks are introduced in the feature fusion part of the model to reduce computational complexity. Furthermore, an SPP (Spatial Pyramid Pooling) module is added to enhance the feature extraction process, enabling the fusion of local and global features. The input data is preprocessed using super-resolution image enhancement techniques to enhance the pixel values and improve the model's recognition accuracy. Experimental results show that the SM-YOLOv4 detection model achieves an evaluation metric (P) of 94.85% and a FPS (Frames Per Second) of 1.25s, with a reduction of 80% in model parameters compared to the original model.The experiments demonstrate that the improved model can accurately detect grape leaf black rot with strong generalization, real-time capability, and lightweight characteristics, providing valuable insights for deployment on mobile devices.