A Defects Detection System for Substation Based on YOLOX

To improve the intelligent supervision level of substation, a defects images dataset facing to substation scenario was built by collecting and labeling a huge number of images about substation equipment defects. Then a transfer learning model based on the YOLOX model was trained by adjusting the model training parameters. Finally, a model with 87.4 % mean average precision and affordable speed (about 0.07 second per image) was constructed. And the experiments results proved that this model can detect the preset defects of substation equipment accurately in acceptable speed according to the visible image obtained from monitoring equipment, which means it has satisfying application potential in future.