Image Splicing And External Damage Detection Of UAVs In Transmission Channels

China's high-voltage transmission channels are generally installed in the wild or on the steep terrain,which has caused great obstacles to the maintenance of transmission channels.With the development of aerial photography technology for drones,more and more drones have been utilized into the maintenance of transmission channels.However,due to the long distance between transmission towers,the performance of HD camera mounted on the drone is declined by the focal length,and the difficulty of collecting complete transmission channel images is raised.The images collected by the UAV tend to have great angle and brightness differences because of the affection of surroundings of channels which leads to a fact that it doesn't meet the requirements of line inspection.In order to obtain complete transmission channel images,an image mosaic method can be applied.The external loss information is obtained via detecting the panoramic view of the spliced transmission channel,which saves a lot of manpower and material resources,ensures the safety of the patrol personnel,improves the speed of the inspection and ensures the safe operation of the transmission channel.In this paper,the splicing fusion algorithm for the aerial image of the transmission channel and the external loss detection scheme for the spliced transmission channel map are discussed.The specific research contents are as follows:(1)Introduce the basic flow of image stitching,and the technology of image preprocessing is studied.The principle and process of geometric correction are introduced for the characteristics of UAV acquisition pictures.The characteristics of aerial image are affected by the surrounding environment,combined with transmission.The requirements of channel aerial image stitching are compared and analyzed.The advantages and disadvantages of several common preprocessing methods are compared and analyzed.The image preprocessing method suitable for this subject is selected.The feasibility of the preprocessing algorithm in this chapter is verified by algorithm simulation.(2)Aiming at the shortcomings of the existing feature detection algorithms in the transmission image of the transmission channel,by analyzing and comparing the advantages and disadvantages of the existing algorithms and the image characteristics collected in this subject,a FAST algorithm based on the salient map is proposed.Combined with the comparative analysis of the speed of simulation of several algorithms in the presence or absence of significant images,the algorithm has the advantages of high timeliness.(3)A RANSAC optimization algorithm(progressive uniform sampling method)is introduced for the timeliness requirements of transmission channel inspection.By comparing and analyzing the shortcomings of existing fusion algorithms,combined with the optimal suture algorithm,an improved weighted average fusion algorithm is proposed.The feasibility of the proposed algorithm and the traditional algorithm are compared and analyzed,and the superiority of the improved weighted average fusion algorithm is verified.(4)Aiming at the shortcomings of the model with high error rate in the traditional deep learning algorithm,an improved deep learning model is proposed,which can accurately detect the large construction vehicles appearing in the aerial image of the transmission channel.Combined with the detection data of the traditional model,it is proved that the improved deep learning model has a higher correct rate.