Blueberry pest and disease identification method based on the improved YOLOv11

Detecting pests and diseases on blueberry leaves poses significant challenges, particularly under complex environmental conditions where traditional methods often suffer from low identification accuracy and frequent instances of false positives and missed detections. To address these issues, this paper proposes an improved detection approach based on the YOLOv11 framework, specifically designed for identifying blueberry pests and diseases. The proposed method introduces the TS Self-Attention mechanism to enhance the C2PSA module, achieving linear computational complexity while improving feature representation capabilities. Then, a Gather-and-Distribute cross-scale feature fusion module is integrated into the neck of the network to strengthen the multi-scale feature aggregation capability of the FPN, thereby improving the model's adaptability to pest and disease features across varying scales. Furthermore, the AFGC-Attention mechanism—combining both channel and spatial attention—is employed to capture fine-grained details and effectively incorporate global contextual information. Extensive experiments were conducted on a self-constructed blueberry pest and disease image dataset. The results demonstrate that the improved model achieves a precision of 95.50%, a recall of 91.55%, and a mAP of 95.10%, representing improvements of 3.40%, 2.45%, and 4.65%, respectively, over the baseline YOLOv11 model. Compared with existing methods, this method not only achieves better detection performance, but also achieves an FPS of 276.2, meeting the requirements of real-time detection.