Detection Method for Surface Micro Defects of Belt Wheel of Electromagnetic Clutch Based on Deep Learning
Targeting at the problems of low detection efficiency and poor accuracy caused by different shapes and small size of surface defects of electromagnetic clutch pulleys,the Pulley-YOLOv4 model is proposed by improving YOLOv4 for detection.An improved spa-tial pyramid pooling module is proposed,which can extract small defect features more effectively and locate them more accurately.A global attention mechanism is added between the neck network and the backbone feature extraction network so that the model can suppress unimportant features and focus on interesting targets.Experiments show that the average accuracy of the Pulley-YOLOv4 model is 98.54％,and the FPS index is 15.Combined with the requirements of actual production speed and accuracy,the Pulley-YOLOv4 pro-posed has obvious advantages in the detection of small defects on the surface of the pulley and meets the real-time requirements.
万方数据
维普
