An improved YOLOv5-based bird detection algorithm for transmission line
Tianyu Wang, Yongkang Zheng, MA Hai-jie, Chao Wang, Yong Liu, Xuxu Li · 2022 International Conference on Computer Engineering and Artificial Intelligence (ICCEAI) · 2022
In view of the large number of birds and dense small targets on the transmission line, this paper marks and makes a bird identification data set based on 20 common birds on the transmission line. The network model adopts yolov5s pre training model, which replaces the general convolution layer of the network structure with ghost lightweight convolution, replaces the bottleneck structure with ghostbottleneck module, and reduces the amount of calculation. The CBAM attention mechanism module has been added to the network layer to make the model more focused on regions of interest. The experimental results show that the mean average accuracy (map) of the improved yolov5 on the self-made bird data set is 94.2%,the detection effect is good, and the purpose of real-time identification can be achieved, which provides a reference for the prevention and control of bird related faults of transmission lines.