Research on Real-time Detection of Insulator Fault Based on UAV Aerial Photography

In order to solve the concerns of the upper computer of the UAV’s limited storage capacity and the challenge of balancing the detection speed and accuracy of prior models, an upgraded real-time target identification approach ideal for insulator defect detection is created. The algorithm is based on the YOLO v7 detection network and uses Wise-IOU to optimize the loss function, ODConv is used in place of the original convolution module to improve feature extraction, channel pruning, and γ coefficient fine-tuning is performed on the model, the network accuracy, speed, and deployment ability are all generally improved. When compared to the original YOLO v7 method, the enhanced algorithm has an average accuracy improvement of 4.4%, a speed improvement of 28.9%, and a volume reduction of 57.7% when tested on the self-created insulator defect data set.