Research on the Image Model of Substation UAV Inspection Based on the Improved YOLO Algorithm

With the gradual expansion of China's power grid, the traditional substation inspection method can no longer meet the practical needs. In view of the defects of YOLO algorithm in target identification, such as low accuracy, false detection and poor timeliness, the paper integrates the lightweight network and attention mechanism, improves the YOLO algorithm, and constructs a lightweight image model of UAV inspection in substation. The results show that the method to improve the loss function and non-maximum suppression of the YOLO algorithm is effective, with its target identification accuracy improving more than 5.37% and the average accuracy by 2.14%. The detection accuracy and detection time of the YOLO model integrating the lightweight network and attention mechanism are improved, and the frame rate is more than 20.67% higher than that of the traditional model, realizing the purpose of the lightweight model. The research has important application value in the inspection of small UAV in substation.