A method for detecting the detachment of rubber sleeves in mother batches based on improved YOLOv10

In this paper, an improved multi-discriminator cyclic consistency generative adversarial network ( MD-CGAN ) is introduced to synthesize more abundant hidden danger area images with the hidden danger images of busbar rubber sleeve shedding in normal substations, so as to build a diversified data set and improve the model's ability to identify complex hidden dangers. The knowledge graph is introduced to associate the detected target features with entities such as equipment parameters and defect types, and the reasoning ability of the knowledge graph is used to generate hidden danger description and processing suggestions. At the same time, it innovates on the basis of YOLOv10 model, uses StarNet network as the backbone to replace the traditional network structure, and cooperates with the speed-optimized detection head LSCD, which significantly improves the detection speed and accuracy. In the experiment, the improved YOLOv10 and the reasoning scheme of introducing knowledge graph proposed in this paper are compared with the original detection scheme mAP in the busbar rubber sleeve shedding detection scene. It is improved by 7.7 %, which verifies the effect of the improved scheme.