Insulator and Spacer Dataset and Benchmark for Power Inspection

In the field of power inspection based on object detection, high-quality, multi-scenario inspection dataset is of great significance for accurate detection of power equipment. But due to various conditions, the currently open source power inspection dataset contains less image information, which limits the effect of the algorithm in actual operation. In order to achieve better object detection effect in power inspection, this paper establishes an insulator and spacer dataset, which improves the reliability and robustness of the algorithm in power inspection by introducing more image features and more complex background information. Experimental results show that the new dataset combined with YOLOV5 series algorithms can maintain high image recognition accuracy in the process of complex environment detection. In addition, we have made lightweight improvements to Yolov5, which reduces GFLOPs by 27%, detection time by 24% compared to YOLOV5n, achieving better balance in insulator and spacer dataset with 89.1 % accuracy on mAP50, and is more suitable for deployment in edge platforms.