YOLOv8-Peaches: Lightweight Peach Surface Defect Detection Model

To enhance the accuracy and efficiency of peach surface defect detection while addressing suboptimal model performance and high complexity, this study proposes a lightweight improved model named YOLOv8-Peaches for precise identification of various surface defects including rot, wormholes, and damage. Specifically, the original C2f module is replaced with a Context Guided module to integrate local features with global contextual information, thereby improving dense defect detection capability; the downsampling operation is refined through an Adown module employing branch splitting and feature concatenation to reduce information loss while significantly lowering computational costs; and CIoU is substituted with Wise-IoU to dynamically allocate gradient gains, enhancing robustness against small and irregular defects. Extensive experiments conducted on our self-established Defective Surface Peaches Dataset (DSPD) demonstrate that the proposed model achieves 95.9% mAP@0.5—5.5% higher than the baseline YOLOv8n—while reducing Parameters and GFLOPs by 43.85% and 35.80%, respectively.