Transmission line abnormal target detection algorithm based on improved YOLOX

Zhongqin Bi, Lina Jing, Chao Sun, Meijing Shan, Wei Zhong

Abstract: The detection of anomalous targets in transmission lines is an important research topic for industrial applications and power grid construction. However, due to the complexity of anomalous targets in the natural environment, existing target detection algorithms have problems such as false and missed detection. To improve the detection performance of anomalous targets, we make the following improvements based on the YOLOX algorithm. First, we assign weights to important target features, extend the perceptual domain of small targets, and enhance their nonlinear representation. Second, the residual network structure is optimized to obtain the key information of the target. Finally, a feature enhancement network with an attention mechanism is proposed to enhance the visibility of anomalous targets in the feature map. The experimental results show that the detection accuracy of the detection model in this paper reaches 82.7% and 91.1% for high-voltage tower bird nest and power line insulator targets, respectively, and the detection speed can reach 72.4 FPS. Compared with the YOLOX network, our method has better detection performance.