Detection Method of Common Wheat Diseases Based on Improved YOLOv8

In order to improve the accuracy and efficiency of wheat pest detection, this study proposes a wheat disease detection model based on an improved YOLOv8. This model replaces the original YOLOv8 network's Neck structure with a Slim-Neck structure, reducing the impact of depthwise separable convolution on wheat disease feature extraction and fusion, improving accuracy while reducing model parameter quantity, and enabling the model to cope with detection scenarios. At the same time, the model introduces channel attention and spatial attention mechanisms (CBAM), which enhance the ability to capture wheat disease features and avoid interference from irrelevant information in the image. Finally, the model replaces YOLOv8's original bounding box loss function with MPDIoU, improving the level of detection and target localization. After testing, the improved model achieved a mean average precision (mAP) of 96.5%, with a parameter quantity of only 2.9M and a computation amount of only 7.4GFLOPs. Compared to the original YOLOv8n, YOLOv7-tiny, YOLOv6n, and YOLOv5s models, its mAP increased by 1.3, 2.0, 3.9, and 2.7 percentage points, respectively. This improved model relies on higher detection accuracy and fewer parameter quantity and computation amount, enabling real-time accurate detection of wheat pests and diseases.