YOLOv8-ACCW: Lightweight Grape Leaf Disease Detection Method Based on Improved YOLOv8

Grape black root, black measles, and blight are three common grape leaf diseases that significantly impact grape yield. However, current research lacks real-time detection methods for grape leaf diseases, which cannot ensure the healthy growth of grape plants. To improve the accuracy of grape leaf disease detection and enable easy deployment of the model on mobile devices, this study proposes a lightweight grape leaf disease detection method based on improved YOLOv8.Firstly, the AKConv module is employed to enable arbitrary sampling of targets of various sizes, replacing traditional convolutional (Conv) modules, thereby reducing model parameters and enhancing disease detection. Secondly, the Coordinate Attention (CA) mechanism is introduced at the end and neck of the backbone network, embedding positional information into channel attention to strengthen feature extraction capabilities and suppress irrelevant feature interference. Next, the lightweight Content-aware Reassembly of Features (CARAFE) module is introduced to improve the model’s ability to extract important features. Lastly, the Wise-IoU (Weighted Interpolation of Sequential Evidence for Intersection over Union) boundary loss function replaces the original loss function, enhancing the network’s bounding box regression performance and detection accuracy for small target diseases. The experimental results on a self-constructed dataset demonstrate that the improved YOLOv8-ACCW exhibits strong detection capabilities for small target disease regions. In the identification of grape leaf lesions, the model achieved F1 scores, mAP50, and mAP50-95 values of 92.4%, 92.8%, and 73.8%, respectively. Compared to the original algorithm, these metrics represent improvements of 3.1%, 3.1%, and 5.6%, respectively. The model’s parameter size is only 2.8M, and its computational cost is merely 7.5G, reflecting reductions of 6.6% and 8.5%, respectively. The algorithm’s detection speed reaches 143 FPS, meeting the requirements for real-time detection and enabling the rapid and accurate identification of grape leaf diseases. Through comparison with other mainstream object detection algorithms, the effectiveness and superiority of this method have been verified. This advancement can provide references for the deployment and application of mobile detection equipment such as grape leaf disease detection robots. It offers a valuable pathway to enhance the grape industry in Guizhou and ensure its healthy development.