Defect Detection of Aerial Insulator Based on Improved Lightweight YOLOv5

Real-time detection of insulators plays an important role in ensuring the stability of high-voltage power transmission and timely handling faults. For this reason, this paper proposes an aerial insulator defect detection method based on light-weight improved YOLOv5. Firstly, the ghost convolution was used to reduce the computational complexity in the YOLOv5 object detection network to improve the detection speed. Secondly, the inverted residual structure was used to improve the feature extraction network to extract richer target features, and the adaptive upsampling is combined to optimize the feature fusion network and improve the model detection accuracy. Finally, the α-IoU excitation factor was added to accelerate the convergence of the target frame loss value to improve the detection efficiency. Through the comparison of the experimental results, it can be seen that the average detection accuracy of the lightweight improved YOLOv5 model proposed in this paper is 93.6%, the detection speed is 68FPS, and the parameters of the model is 3.7M, which is better than other detection algorithms under the same conditions. The method in this paper can provide a reference for the detection of insulator defects.