BN-YOLO: a lightweight method for bird's nest detection on transmission lines
Yunjie Xiang,Congliu Du,Yan Mei,Liang Zhang,Yutong Du,Aoxing Liu
DOI: https://doi.org/10.1007/s11554-024-01577-8
IF: 2.293
2024-11-17
Journal of Real-Time Image Processing
Abstract:Bird's nest on power transmission lines pose a threat to the safe operation of power equipment and may even affect the stability of the entire power system. To address the challenge that traditional methods face in achieving accurate real-time detection in complex environments, this paper proposes the BN-YOLO bird's nest detection method based on the YOLOv8s baseline model, enhancing its suitability for real-time applications in intricate settings. First, we replaced the original backbone network of YOLOv8s with a lightweight FasterNet module, thereby reducing computational burden and improving network performance. Second, the feature fusion network was redesigned, incorporating the efficient multi-scale attention module (EMA) to optimize feature fusion capabilities across different scales. Subsequently, we proposed a lighter and faster C2f structure by substituting the standard convolution in the C2f structure with partial convolution (PConv). Finally, Wise-IOUv3 was utilized as the regression loss function to mitigate the effects of low-quality annotations and accelerate network convergence. Experimental results demonstrate that on our self-constructed transmission line bird's nest dataset, the bird's nest detection method we proposed achieved a 97.73% score. Compared to the original YOLOv8s, the mean average precision (mAP) increased by 2.19%, and the detection speed improved from 61 FPS to 83 FPS. In addition, our method outperforms other mainstream object detection algorithms, such as SSD, DETR, and RT-DETR, providing higher detection efficiency while maintaining high accuracy. These results confirm that the proposed method can effectively detect bird's nest targets in complex environments and fulfill the requirements for real-time inspection. This research not only enhances the accuracy and efficiency of bird's nest detection in power transmission lines but also offers a novel solution for the safe operation and maintenance of smart grids, which holds significant practical application value.
computer science, artificial intelligence,engineering, electrical & electronic,imaging science & photographic technology