Research On Transmission Line Bird Nest Detection Algorithm Based On Faster R-CNN

With the rapid development of China’s economy,the scale of the power grid is also constantly expanding,and the distribution of overhead transmission lines is very extensive.In some mountainous areas with complex terrain,many birds will nest on transmission lines,causing short circuit faults in the power system,and easily causing safety hazards such as line trips and equipment damage.Because the bird nest targets on overhead transmission lines are small and difficult to identify,manual detection method is time-consuming and laborious,with low accuracy.Therefore,this paper adopts the deep learning method to detect the bird ’s nest in the aerial pictures of transmission line inspection,and makes corresponding algorithm improvements.The GPS information carried by the pictures is used to accurately locate the bird ’s nest,so as to achieve high-precision and high-efficiency automatic detection.Firstly,the bird nest pictures of transmission lines with GPS information are collected by DJI UAV,and the pictures are screened.With the help of common picture processing methods and some public data sets,the data sets are expanded,which lays a foundation for subsequent experiments.Secondly,by comparing and analyzing the commonly used target detection algorithms,Faster R-CNN is selected as the basic algorithm.Because the target detected in this paper is the bird nest on the overhead transmission line,the target is relatively small and the feature is not very obvious.The bird nest detection method based on Faster R-CNN algorithm can detect the bird nest,but the stability and positioning accuracy of the algorithm detection are low,so the Faster R-CNN algorithm needs to be improved.Considering the size of network parameters and algorithm learning ability,the feature extraction network is replaced by Res Net50 network with residual structure.The SKNet mechanism network is added to the feature extraction network to extract pictures feature information efficiently and quickly,so that the network focuses on learning channels related to bird nest features.ROI Align is used to replace ROI Pooling in the original algorithm to improve the accuracy of algorithm detection.Finally,under the same network training strategy,the improved Faster R-CNN algorithm is compared with the unimproved algorithm and other commonly used deep learning algorithms.In order to verify the robustness of the improved algorithm,the influence of different illumination conditions on the experimental results is discussed.The experimental results show that the improved detection algorithm in this article performs well in terms of the mean average precision m AP and F1 score of comprehensive evaluation indicators.The m AP value has increased from 89.61% before the improvement to 92.53%,an increase of 2.92%;The F1 score has increased by 0.05.When the light intensity is different,the improved Faster R-CNN algorithm in this paper can still accurately locate and identify the bird nest of the transmission line,which has a certain degree of robustness.Compared with SSD and YOLOv4 algorithm,the improved algorithm has an m AP increase of 4.79% and 0.67%,respectively.The experimental results verify the feasibility of the improved method proposed in this paper,and it has important application value for the detection of bird nests on transmission lines.