UAV Transmission Line Inspection Algorithm Based on Cross-scale Feature Fusion and Attention Mechanism
Xuan Yang, Chaoxu Mu, Xiaoyu Zhang, Zhenhuan Ding, Changyin Sun · 2023
Insulators play an important role in transmission lines and almost 81.3% of power accidents in the power transmission system are caused by insulator defects. Hence, insulator identification is an important task in the process of UAV inspection. However, due to the interference of complex environment, it is difficult to identify multiple targets or small targets with the normal detection of UAV. In this paper, a YOLOv7 insulator recognition algorithm based on cross-scale feature fusion and attention mechanism is proposed. First, the YOLOv7 model framework is built. In order to reduce the interference brought by the complex inspection environment, the attention mechanism is introduced to improve the contrast between the detection target and the background. Secondly, the FPN+PAN structure in the original framework is converted into BiFPN structure, which increases the ability of cross-scale feature fusion. Secondly, in the neck part, Focus layer is selected to replace MP module, which preserves the integrity of feature extraction. Then, the original SPPCSPC structure of the space pooling layer is replaced by SPPFCSPC structure to improve its computing speed. Finally, a case study is implemented to verify the accuracy and timeliness of the proposed algorithm. The results show that the proposed algorithm can achieve good results in insulator image recognition, with an average accuracy of 97.8%.