Improved YOLOX Lightweight Algorithm for Power System Component and Defect Detection

As one of the research focuses in the field of computer vision, object detection technology is crucial for ensuring the stable operation of the power grid by detecting power components such as insulators, dials, and silicone cylinders in the power system. Due to the small target size of defect areas in unmanned aerial vehicle aerial remote sensing inspection images, the model has poor robustness and insufficient accuracy. Therefore, an improved YOLOX-Nano power component fault recognition target detection strategy is proposed. At the same time, considering the slow detection speed, large volume, and the problem of neglecting important features and emphasizing non important features causing accuracy loss of the model, multiple attention mechanism modules are added to the original network structure to improve object detection accuracy and reduce model volume, making it suitable for embedded devices and mobile terminals, while also meeting the demand for real-time detection of input video streams. Experiments have shown that the optimized network in this paper can achieve good results in processing power inspection datasets in engineering practice. The improved YOLOX-Nano algorithm has an average detection accuracy of 97.73% in power inspection experiments, which is 7.82% higher than the original YOLOX-Nano algorithm and significantly improves the accuracy compared to algorithms such as YOLOv5. This proves the effectiveness of this method and has certain practical engineering significance.