Bushing Surface Defect Detection Method Based on Improved YOLOX

A small target defect detection approach based on multi-attention feature fusion is suggested based on YOLOX in order to address the issues of high missed detection rates of tiny targets and insufficient fusion of model features in deep learning models for bushing surface defect detection. In the backbone network, the Res2NetBlock module with finer-grained feature extraction is introduced, and the CoT self-attention mechanism is embedded to strengthen the regional features of hidden small targets and lower the missed detection rate. Coordinate attention is embedded in the feature fusion stage to further increase the receptive field of the model. According to the experimental findings, when compared to the original YOLOX algorithm, the mAP of the improved algorithm on the bearing bushing surface defect dataset is increased by 4.04%, and the recognition rate of small targets is significantly improved.