Research On UAV Autonomous Line And Insulator Identification And Location Method

With the large-scale construction of UHV and the plan of “Nine Crossing and Nine Straight”,China's power industry has gradually entered the forefront of the world.However,this poses a higher challenge to the safety maintenance of power systems.Most of the current power line inspections rely on traditional manual methods.The labor volume is not only heavy but also inefficient and subject to geographical location.In recent years,UAV patrol technology has always been a hot spot for power grids,and its application to power patrol is also the trend of smart grids.Therefore,based on the research of this problem,this paper proposes a self-propelled line tracking method for UAVs.At the same time,a method based on deep learning insulator detection is proposed in the process of line inspection,which automatically identifies and locates the insulators in the aerial image.In order to solve the effects of complex and variable power line background and uneven illumination distribution in aerial image,the paper firstly preprocesses the aerial image,and selects homomorphic filtering to enhance the image quality of the power line to improve the recognizability,and then adopts Wiener,mean and median.Three filtering methods reduce the noise of the aerial power line image.It is verified that the median filtering signal to noise ratio is the highest,and the mean square error is the best denoising performance.Finally,the minimum error threshold segmentation is performed on the aerial power line.On this basis,the Canny operator with higher positioning accuracy is used to complete the edge detection.At the same time,the small-area interference noise in the edge detection image is eliminated by mathematical morphology.The experimental results show that the power line contour information has good recognition effect after pre-treatment.In order to solve the problem that the traditional Hough transform method,Radon transform method takes a long time to extract linear lines,and needs to constantly update the threshold of the power line to be detected,and can not locate the endpoint,a sub-pixel linear time line extraction algorithm LSD is selected.The algorithm first generates a linear support domain and performs a rectangular fitting on this basis.Finally,the straight line is verified according to the probability value of the aligned points in the rectangular domain.The test results show that the LSD algorithm used in this paper can accurately locate the end position of the power line,the integrity of the extracted power line is good,and there is no need to adjust the parameters too much and the line is strong in real time.Then,in order to realize the autonomous navigation line of the UAV,the Kalman filtering idea is used to predict the range ofthe line slope of the LSD extracted power.The ROI area is established and the power line extraction is completed in the area to shorten the overall detection time.Finally,Kalman is given in the paper.The simulation results of the filter tracking verify the feasibility of the method.In order to solve the defect that the feature extraction of aerial insulators in traditional algorithms relies on image segmentation and detection is not universal,a method based on Faster R-CNN insulator identification is proposed.The ZFNet and VGG16 network models are selected in the algorithm to extract the characteristics of aerial insulators.At the same time,the Loss curves of ZFNet and VGG16 models in the Faster R-CNN training process are visualized,and the feasibility of the two models is verified.Finally,the test performance of the two models is verified.The test results show that the average accuracy of the ZFNet model is 0.75,and the average accuracy of the VGG16 model is 0.89.Both of them can identify and locate the aerial insulators in complex environments,and the VGG16 recognition effect is comprehensively compared.Slightly better than ZFNet.At the same time,the VGG16 model motion blur and the insulator image under different illumination are tested.The test results show that the VGG16 model has good robustness as a whole.The insulator identification and location method proposed in this paper has certain universality and practical value.This paper has 74 pictures,5 tables,and 72 references.