Abstract: A detection method based on Yolov5 for identifying suface defect on bearing end was proposed to address the problems like low detection accuracy and poor reliability when using manual visual inspection approach. Firstly, in order to ensure there are sufficient sample data of bearing end surface defects, we present a data enhancement method by combining Mosaic and Copy-Pasting strategies. Then, on the strength of the superior target detection performance using YoLov5, a model was built based on YoLov5s. The performance of this method was assessed by employing image data of defects on bearing end surfaces collected in industrial field, which detection accuracy is high up to 94.7%.