An Improved Method for the Foreign Object on the Power Line Detection Based on Sample Simulation

In the transmission line intelligent inspection object detection, the imbalance of defect categories results low detection rates for the few data categories in deep learning algorithm, such as the foreign objects on the power line, which causes short circuits and fires in the transmission line potentially. In this paper, a sample simulation method based on the 3D modeling and the generative adversarial network is proposed to generate foreign objects simulated samples. Compared with the result with the only ground truth samples as training dataset, the detection rate of the foreign objects category with the addition of the simulated samples in the training dataset increased from 74.70% to 81.22% based on the YoloV5 object detection model. The experiment results indicate that the samples simulation method benefits the power line defect detection in the intelligent inspection.