YOLO-Based Real-Time Detection of Power Line Poles from Unmanned Aerial Vehicle Inspection Vision

Unmanned aerial vehicles (UAV)-based inspection has become an important approach for power line inspection after disaster. However, the current UAV-based inspection is still performed manually for damage assessments, which is not only time-consuming but also poor in accuracy. In this paper a real-time detection model based on YOLO deep learning algorithm is presented to detect the status of power line poles automatically from the UAV vision data after disaster. The data augmentation is performed for collapsed towers to solve the class imbalance problem. To improve the parameters of YOLO, K-means algorithm is used to cluster object frames of pole data. The experimental results show that the proposed model can effectively detect multi-scale towers in multiple environments. The Recall and Intersection-over-Union (IoU) of the improved YOLO are improved, with the mean average precision (mAP) on the test set of 94.09% and the average processing speed of 20 frames per second (FPS) after improving the parameters. Moreover, we tested the simplified YOLO with faster speed, and the average processing speed reaches 30 FPS.