Object detection of UAV power line inspection images based on federated learning

Power lines are widely distributed and scattered in power system. The inspection images collected by various power supply companies are existing in the form of data fragments and island. The data sharing is sometimes forbidden because of privacy protection. This means a single data owner (e.g. a regional power supply company) cannot always train good generalized models when dealing with inspection images by deep learning based methods. In this work, we propose a federated learning method to detect the insulators in power line inspection images which have different data distribution from different power companies by Unmanned Aerial Vehicles (UAVs). The experiment results show that through participating in the federated training a single power company can effectively improve the detection accuracy and generalization of the object detection model.