A Deeping-Learning-Based Foreign Object Inspection System Design for Overhead Power Transmission Lines

Overhead power transmission lines are widely used to transmit electrical energy for generation plants to distribution substations. Foreign objects such as kites, kite string or balloons may be accidently twisted on the power lines, which may lead to power outages if there are not removed. This paper proposed a foreign object inspection system based on deep learning for power transmission lines to overcome the shortages such as poor operation efficiency, high cost and time consuming of ordinary inspection methods. The system is consisted of inspection equipment and ground control station. The inspection equipment travels on the ground line of power transmission lines, and the deep learning algorithm is embedded in the software installed on the computer of the ground station, which can be used to control the inspection equipment remotely. The YOLOv3 algorithm is used to detect the foreign object, and the model is trained and tested in the field. Experimental results show the inspection system can inspect the lines semi-autonomously and the algorithm is capable of inspecting the foreign objects with 25fps in speed and 84% in accuracy, which helps in regular inspection task of the power transmission lines.