High-Voltage Transmission Line Foreign Object and Power Component Defect Detection Based on Improved YOLOv5

With the outstanding performance of deep learning in the field of computer vision, the automatic visual detection of foreign bodies in transmission lines and electrical equipment defects by inspection robots and Unmanned Aerial Vehicles based on neural networks has become an attractive topic in smart grid. However, in practical application scenarios, small-sized target defects pose a great challenge to the detection accuracy of existing mainstream deep learning detection networks with limited perceptual fields. To solve the above problems, the paper proposes a detection model of YOLOv5 transmission line inspection image. Firstly, the key target images under different backgrounds and attitudes are collected and preprocessed. Specifically, in order to improve the perception ability of networks for small-sized targets, the K-means clustering algorithm is used to optimize the size of the anchor box, which effectively improves the fitting ability of the key target features. Then, recursive gated convolution is used as the backbone network to improve the ability to extract key target features. Finally, considering the concealment of small-scale features, the space-to-depth convolution module is added to the neck network to realize down-sampling and retain all the information in the channel dimension. In addition, a feature prediction layer is added to optimize the scale of network detection, and a Simple Parameter-Free Attention Module is added to further optimize the characterization of network features. The experimental results show that the accuracy and recall of the proposed network are 96.8% and 93.3%, respectively. The average detection accuracy reaches 97.1%, which is 3.8% higher than that of YOLOv5 network, 5.2% higher than YOLOv6 and 1.0% higher than that of YOLOv7 network. The proposed method significantly improves the detection performance of critical targets and defects of high-voltage transmission lines.