Insulator Defect Detection Based on improved YOLOv5

As an important device for line support and electrical insulation in power system, Insulators play a pivotal role in the high-voltage lines of the power grid. To overcome the shortcomings of insulator aerial photography at present, this paper proposes a deep learning model based on improved YOLOV5 to realize insulator defect detection. Firstly, perform a data enhancement on the existing data set, thereby increasing the number of samples in the database, and making the generalization of the model reach a higher level. Secondly, the Triplet Attention modules introduced into the backbone network of YOLOV5, Improve the feature extraction of the network structure by establishing a certain spatial relationship between the residual transformation and the rotation operation. Thirdly, the Dense Block module is added to alleviate that problem of gradient disappearance and reduce the phenomenon of over-fitting, so as to achieve a good balance among accuracy, speed and storage space. Finally, By comparing the improved network structure with the existing two-stage network model and one-stage network model in terms of test speed, model accuracy and recall rate, the experimental results show that the accuracy rate of the improved network structure reaches 94.5%, and the recall rate reaches 93.1%. which has better detection effect for fuzzy insulators and minimal insulator faults.