LDIRNet: ceramic bearing end face defect detection based on lightweight dynamic inverted residual network

This paper proposes a specialized network model, Lightweight Dynamic Inverted Residual Network (LDIRNet), for the detection of surface defects in ceramic bearings. The model integrates the iRMB module to improve feature extraction accuracy and reduce computational costs. In addition, the DySample sampling algorithm is used in the study, which significantly improves processing speed and reduces memory requirements while maintaining high efficiency. By optimizing the lighting conditions and using composite light source illumination technology, the probability of missed detections and false alarms can be reduced. Combined with the LDIRNet detection network, the detection performance of surface defects in ceramic bearings is effectively enhanced. Experimental results show that LDIRNet effectively improves the detection accuracy of surface defects in ceramic bearings, with an mAP0.5 value of 75.2% on the CBS-DET dataset, an increase of 2.4%. The model parameters are reduced, and detection speed is improved. Compared to existing detection models, this research method demonstrates significant improvements in detection accuracy, model size, and computational efficiency, especially in terms of model parameters and detection speed, proving its potential application value in practical industrial scenarios.