LA_YOLOx: Effective Model to Detect the Surface Defects of Insulative Baffles

In the field of industry, defect detection based on YOLO models is widely used. real detection, method insulative baffles artificial detection. The work efficiency this method, however, low because depends absolutely human eyes. Considering excellent performance YOLOx, an intelligent YOLOx proposed. First, we selected a CIOU loss function instead IOU by analyzing characteristics baffles. addition, considering limitation model resources in application scenarios, lightweight We replaced YOLOx’s backbone with backbones (MobileNetV3 and GhostNet), used Depthwise separable convolution conventional convolution. This operation reduces number network parameters about 42% compared original network. However, mAP it decreased 0.8% model. Finally, attention mechanism introduced into feature fusion module to solve problem, called LA_YOLOx. final value LA_YOLOx reaches 95.60%, while 95.31%, which proves effectiveness