Research Of Power Line Recognition Method Based On UAV Aerial Image

Power line is one of the core components have long been in the wild and distributed throughout the country in the transmission of national power networks.Power lines need to be over the mountains in many remote mountainous regions,and weather conditions have a great impact on power lines.If power lines are in an aging or depleted state for a long time,it would occur serious power line accidents.When workers inspection on power lines in complex areas such as mountains and forests,it has a large workload and a high risk.Compared with manual inspection,low cost and high efficiency UAV inspecting power line technology has developed rapidly.In the process of using UAV to inspect the power line,the key technology is the detection and fault recognition of power lines images.Therefore,this paper proposes an approach for power line detection and fault recognition from the UAV aerial power line images.This method requires preprocessing of the UAV aerial power line images,it includes apply median filtering and gaussian filtering to denoise the images,the last stage of preprocessing uses histogram equalization to enhance the images.After the preprocessing,the edge of image was detected by improved edge drawing algorithm.Then according to the power line distribution simplified model searches for the region boundary of the power line in the image,the power line was detected by the random Hough transform in the region of power line and the mechanism of false segment determination is used to remove false segments.This method effectively reduces the false detection rate and missed detection rate of power lines,and enhances the target detection and recognition rate of power lines.Finally,the aerial power line images of UAV inspection operations were used to establish the UAV aerial power line image database,and a fault recognition model based on the deep belief network was constructed.The DBN model is trained by a large number of samples and multiple iterations.Introduced the PReLU activation function instead of the Sigmoid activation function,and the Softmax classification layer is added to the output layer.Experiments show that the method proposed in this paper has higher fault recognition rate than BPNN,SVM and DBN-SVM methods.This paper has experimented and verified the image of UAV aerial power line provided by National Power Grid Nuri Group,the experimental result shows that this method can accurately detect and recognition the power line fault.