AN ENVIRONMENTALLY FRIENDLY DEFECT DETECTION METHOD FOR SMALL FITTINGS OF TRANSMISSION LINES BASED ON FASTER R-CNN

With the in-depth development of artificial intelligence (AI), big data and other fields, more and more people apply this technology to power systems to save energy and protect the environment. Concentrating on the issue that the locking pin fault has an effect on the normal operation of transmission lines and even leads to the unsafe operation of lines, an improved detection method of locking pin fault is proposed based on UAV image and Faster R-CNN theory. The proposed algorithm is based on the basic framework of Faster R-CNN. ResNext101 is exploited as feature extraction network. In what follows, three improved methods, Deformable Convolutional Networks V2 (DCNv2), Feature Pyramid Networks (FPN) and Online Hard Example Mining (OHEM) are utilized to optimize the algorithm. The experimental results show that the developed algorithm can significantly improve the detection accuracy of lock pin fault compared with other object detection algorithms such as Cascade R-CNN, Grid R-CNN and RetinaNet. We believe that the development of artificial intelligence will make a huge contribution to environmental protection.