Abstract:

Apple leaf black rot is a common and destructive disease that severely affects apple quality and yield. To address the poor sensitivity to small targets, background clutter, and low efficiency of traditional identification methods, this study proposed an improved YOLOv11−based detector. A C3K2 module was introduced into the backbone to enhance multi−scale feature modeling; a C2PSA attention module was appended after the SPPF block; and the detection head adopted depthwise separable convolutions together with Distribution focal loss (DFL) and CIoU loss. The improved model achieved an mAP of 99.5%, a recall of 99.7%, and an F1−score of 99.6%, outperforming YOLOv8 by 3.2 percentage points and reaching 48 frames·s⁻¹. Ablation experiments showed that combining C3K2, C2PSA and depthwise separable convolutions raised mAP from 93.1% to 95.2%. The proposed method ensures high−precision detection of small black rot lesions on apple leaves while markedly improving real-time performance, and has strong practicality and deployment value.

Key words: Apple black rot, YOLOv11, Object detection, Depthwise separable convolution