Abstract:

To address the degraded detection performance of insulated glove wearing status caused by small hand regions in complex distribution network operation scenarios, a small-object-oriented detection algorithm, termed LN (LLC (Light-weight, Light, Coordinate) +NWD(Normalized Wasserstein Distance)), -YOLO is proposed.First, an LLC module was designed by integrating lightweight spatial pyramid pooling, a receptive field fusion attention mechanism, and coordinate convolution, thereby enhancing spatial perception through multi-scale feature aggregation and coordinate embedding.Subsequently, the Mixup data augmentation strategy was introduced to improve model robustness, and a NWD loss function is incorporated to optimize small-object detection.Finally, the effectiveness of LLC module was validated through controlled experiments to determine the optimal baseline, followed by ablation and comparative experiments on the proposed method.The results demonstrate that the proposed algorithm achieves a detection accuracy of 90.1%, representing a 2.0% improvement over the baseline, with a detection speed of 56 frames per second and a memory footprint of 15.7 MB, meeting the requirements for accuracy, real-time performance, and edge-device deployment in distribution network operation scenarios.

Key words: distribution network operations, insulating gloves, wearing status, detection algorithm, receptive field