Recognition Method of Aerial Insulator Defects Based on Deep Learning

Abstract

Aiming at the problem of low accuracy of aerial photography insulator defect identification during UAV patrol, a method of insulator defect identification and positioning based on improved YOLOv3 target detection model is proposed. This method uses asymmetric convolutional block (ACB) as the structural block of the convolutional neural network model in the training phase. On the basis of not increasing the inference time, the depth of neural network learning is improved. At the same time, the feature map combination of the pyramid multi-scale detection model is improved, and the distance index parameters of the K-means ++ algorithm are optimized, which is used for target prior frame clustering. The experimental results show that the improved method proposed in this paper improves the recognition accuracy of small defective targets. Without reducing the speed of inference, the average accuracy is increased from 84.5% to 91.9%, which meets the requirements of intelligent inspection of UAVs for overhead transmission lines.