Aiming at the problems of high missed detection rate of small targets and multi-targets in the detection of industrial bearing surface defects by deep learning models in complex environments, and insufficient model feature fusion, an improved YOLOX industrial defect target detection model was proposed.