Defect Detection Photographed Images Based on Asynchronous Federated Learning

The integrity of the insulators on transmission lines directly affects the safety and reliability of power transmission. A method for insulator defect monitoring based on federated learning and YOLOv4 is proposed to better train models for insulator defect detection by utilizing distributed transmission line insulator data, which is constrained by communication and privacy. This method employs asynchronous federated learning and further scheduling algorithm design to simulate and optimize for real complex scenarios. The model training and performance testing are completed using the Chinese power line insulator dataset. Experimental results show that the detection model trained by federated learning has significant effects and achieves comparable performance with centralized YOLOv4. Furthermore, federation protects the security of power data belonging to different organizations, breaks industry data silos, and makes it possible to utilize more insulator data.