Vision-based Corrosion Detection Of High Voltage Transmission Line

The prosperity of the social economy has led to the rapid development of the power economics.More and more high-voltage transmission lines have been built in far and wide span,which demands transmission lines should be long enough and reliable.Using unmanned aerial vehicle(UAV)has become an important means to protect the safety of high voltage transmission lines,which means UAV can shoot the key components from a very close distance along the transmission lines to obtain aerial images,and then apply computer vision technique to deal with these images,so as to realize the automatic diagnosis of transmission lines.As one of the key detection tasks,the detection of metal rust defects in high-voltage transmission lines is of great importance.An algorithm framework based on segmentation of corrosion area detection is proposed for aerial images in this paper,which presents an improved semi-interactive segmentation algorithm under the analysis of image characteristics of foreground objects of transmission lines in different background.In this paper,the image of a set of similar scenes is extracted by means of cooperative segmentation,and the color image is corroded in the foreground image after segmentation.The experimental results show that proposed corrosion detection algorithm can identify the corrosion area of the transmission lines and improve the maintenance efficiency.The main works of this paper are as follows:First,a semi-interactive metal-image segmentation method based on GrabCut is proposed.For a set of images of similar scenes of UAV aerial photography,the author marks one of the less pre-background pixels to construct mixed Gaussian model and then to manually traverse the pixel values of other images in the group,and finally to calculate the Gaussian model probability to achieve segmentation of foreground and background.The author collects more than four thousand pieces of images of high-voltage transmission lines,and divides them into two categories:simple background and complex background,and marks the target area Ground Truth manually,so as to evaluate and analyze computational efficiency and accuracy of the results from two aspects.The author selects ten-group images of the similar scene,each set of 50 images,from simple and complex background,to detect algorithm based on these samples,which achieves 90%accuracy,and the processing time of each image was 0.5ms.Second,a method of image segmentation based on collaborative segmentation is proposed.In order to overcome the shortcomings of semi-interaction and segmentation which still need human interaction,the images taken by the drone around the same high-voltage transmission tower are similar,which can be seen as a series of continuous video sequence.Therefore,the method of hardware image segmentation based on video collaborative segmentation is proposed.Based on the inter-video and intra-foreground coherence and the temporal and spatial continuity,a target-based common selection graph model is constructed.Finally,the model is optimized to get the final segmentation results,which divides from the perspective of energy function optimization to obtain more accurate segmentation results.Third,corrosion defect detection based on segmentation images is proposed.Based on the results of image segmentation,we obtain the foreground image of the target,and transform the RGB image model into the YCrCb image model from the pretreatment,then analyze the change rule of the color value of the Cr channel,get the best segmentation threshold through calculation,so as to extract the corroded area.In the experiment,50 samples are selected,among which there were 20 rusty images.The accuracy of identifying corroded images and recognizing the corroded areas can reach 88.4%.The accuracy of the algorithm for detecting corroded defects directly without segmentation increases by over 20%.