Research on Lemon Surface Defect Detection Based on YOLOv8-SGN Model

Aiming at the problems of low recognition precision and high missing rate of small targets in the complex background of traditional lemon surface defect detection methods, a YOLOv8-SGN detection model was proposed in this study. SCConv module was introduced to replace the standard convolution layer, effectively reducing the redundant calculation of the model and enhancing the feature expression ability. GAM module was embedded before the detection head, and the attention of defect area was enhanced through channel-space dual-dimensional feature re-calibration. Experimental verification based on four data sets including highquality, rot, crack and spot shows that: The precision of the improved model was as high as 97.1%. Ablation experiments confirmed that the collaborative optimization of SCConv and GAM modules effectively increased the recall rate to 98.8%, providing a lightweight solution for intelligent sorting of lemon quality.