Abstract: Aiming at the problem of detecting defects in voltage balancing rings and insulators in mining area transmission lines, this paper proposes the PSI-YOLOv9 algorithm based on the improvement of YOLOv9. This algorithm builds the PRepNCSPELAN4 module by combining partial convolution PConv and RepNCSP structure, enhancing the feature extraction ability and reducing the number of parameters; it adopts the Shape＿IoU loss function to optimize the accuracy of bounding box regression. Experiments show that the mAP of the improved model increases from 82.2% to 86.7%, the number of parameters decreases from 50.71 MB to 41.99 MB, the FPS increased from 6.04 fps second to 6.44 fps, and the FLOPs decreased from 236.7 G to 202.2 G, achieving higher precision and more lightweight deployment requirements, and is suitable for the detection tasks of four types of targets: normal voltage balancing rings, tilted voltage balancing rings, damaged insulators, and insulator flashover.