Research On Information Extraction Technology Of Power Corridor Based On Airborne LiDAR Data

Traditionally, the inspection of power corridor system mainly relies on human resource or helicopters equipped with optical camera, but these methods may consume more material and financial resources. At the same time, because of the limitation of human eye recognition rate and the spatial location accuracy of image, the information obtained by these methods is not accurate enough. Resent years, the airborne LiDAR(Light Detection and Ranging) technology has got more and more attention, which can obtain the 3D spatial information of ground sence. With the development of LiDAR system and the decrease of flight cost, there is a broader space for airborne LiDAR used in the inspection of power corridor. In this paper, based on the 3D point cloud obtained by airborne LiDAR, according to the characteristics of LiDAR data and different objects in the scene, the feature extraction processing before LiDAR point classification is first explained. And then a sparse representation classifier(SRC) is used for classification, in which a deep study is carried out on the nonlinearity of the classification problem and the heterogeneity of different features. And next, a joint sparse representation classifier based on multiple kernel learning(MKL) is proposed for LiDAR point classification. Last, there is a further analysis on the classification results, to get the security status of power corridor.There are mainly four parts in this article: the feature extraction processing of LiDAR point data, the SRC combined with joint sparse representation and kernel method, the SRC based on multiple kernel learning, and the state analysis of power corridor based on classification results. The detail description is as following:First of all, the feature extraction processing for point cloud classification has been studied. For the discrimination of different types of ground objects, the effective feature information needs to be analyzed and get first, which contains the single point feature and the multiscale neighboring feature. A pre-processing opearation is carried out for the adaptability of SRC method, and the feature information for classification is prepared.Next, the SRC method for point cloud classification has been studied. Based on the original SRC, we introduce the joint sparse representation and the kernel method into the classification problem. The kennel method has a good effect on the nonlinear problem of high dimensional data classification, and the joint sparse representation can make full use of the similarity of the labels of neighboring points. For the heterogeneity problem of different features of point cloud data, we proposed a SRC method based on multiple kernel learning(MKL). This method can evaluate the importance of different features, and the classification performance has a obvious improvement comparing with the single kernel SRC.Last, based on the results of LiDAR point cloud classification, we make some analysis on different objects in the scene, including powerline, ground, buildings and trees. By computing the distance between powerline and other ground objects, we could confirm whether the powerline is affected according the electric safety standard.