Improved YOLOv7-Based Detection Method for Small-Scale Operational Targets in Distribution Networks

To address the problems of poor detection accuracy of small-scale operating targets and easy missed or false detection of occluded targets in distribution network scenarios, this paper proposes an improved YOLOv7 algorithm. We optimized the preset anchor boxes based on the K-means++ algorithm, introduced deformable convolution (DCNv2) into the feature extraction network to replace the ordinary convolution, and optimized the model post-processing using the NWD-NMS candidate box screening algorithm, improving the algorithm's detection performance for small-scale operating targets and occluded targets in the complex background of the distribution network. On this basis, this paper constructs a dataset containing multiple operating targets such as drop-out fuses, rod insulators and lightning arresters, and expands the dataset using methods such as image angle transformation, image noise processing and saturation processing. After experimental comparison, the method proposed in this paper has improved the detection performance of operating targets in distribution network scenarios, and the average detection accuracy has been improved by 10.9%.