Foreign Object Detection of Transmission Lines Based on Faster R-CNN

Abstract

The object detection method based on RCNN network model has good mobility and robustness compared with the traditional methods. Classical foreign object detection algorithms for transmission line, such as SIFT and ORB feature matching algorithms. These methods have low recognition accuracy for edge blurred images and complex background images. In view of the above deficiencies, this paper constructs a transmission line training data set based on the characteristics of the collected transmission line images, and trains the Faster R-CNN model to detect the falling objects, kites, balloons and other foreign objects in the transmission lines. The experimental results show that compared with the traditional object recognition method, Faster R-CNN not only overcomes the instability of manual extraction features, but also improves the accuracy of foreign object detection in transmission lines. It can realize the detection of foreign objects in transmission lines in complex scenes.