WT-YOLOM: An Improved Target Detection Model Based on YOLOv4 for Endogenous Impurity in Walnuts

Since impurities produced during walnut processing can cause serious harm to human health, strict quality control must be carried out production. However, most detection equipment still uses photoelectric technology automatically sort heterochromatic particles, which is unsuitable for detecting endogenous foreign bodies with similar colors. Therefore, this paper proposes an improved YOLOv4 deep learning object algorithm, WT-YOLOM, in walnuts—namely, oily kernels, black spot withered and ground nutshells. In the backbone of model, a lightweight MobileNet module was used as encoder extraction features. The spatial pyramid pooling (SPP) structure pooling—fast (SPPF), model size further reduced. Loss function replaced more comprehensive SIoU loss. addition, efficient channel attention (ECA) mechanisms were applied after feature map improve model’s recognition accuracy. This compares speed accuracy WT-YOLOM algorithm Faster R-CNN, EfficientDet, CenterNet, algorithms. results showed that average precision different kinds walnuts reached 94.4%. Compared original reduced by 88.6%, 60.1 FPS, increase 29.0%. metrics significantly better than those comparative models efficiency walnuts.