Multispectral LiDAR Graph Segmentation Under Spatial And Adaptive Local Spectral Consistency Constraints

With the construction of digital cities,the rapid acquisition of accurate urban land cover information plays an important role in the process of urban digital management and planning.The classification of land cover information is an important means to carry out this process,so the high-precision land cover classification has important practical significance.Airborne Light Detection and Ranging(LiDAR),as a new type of sensor,can quickly and directly get three-dimensional coordinate information,to meet the needs of rapid urban informatization.In view of the traditional three-dimensional land cover classification method of airborne LiDAR data,due to the assumption that each land cover has internal spatial and global spectral consistency,it is unable to effectively distinguish the same objects with different spectra,foreign objects with the same spectrum and land covers with similar elevation.In this paper,as a new data source.A minimum spanning tree multi-spectral LiDAR land cover graph cutting algorithm with spatial and adaptive local spectral consistency constraints is proposed.First,radiometric calibration and multi-wavelength LiDAR point cloud fusion were performed on the multi-spectral LiDAR data to obtain the spatial location of the fused multi-spectral LiDAR point cloud and the corresponding single point cloud data with multi-spectral information.Then the weighted graph of single point cloud data is constructed.Then the weighted graph is cut by using the graph cutting algorithm under the constraint of spatial and adaptive local spectral consistency.Finally,the homogenous connected region after the map cutting is taken as the basic unit of processing.By extracting multi-dimensional features,the classification results of land cover are obtained by using RBF kernel support vector machine classifier,and the object-oriented land cover classification is realized.The algorithm fully considered the intensity of laser reflection by objectives to the influence of many factors such as material,the surface roughness of the internal present spectral characteristics of land cover,can make up for the various land cover internal statistics of the global spectrum of consistency hypothesis classification error caused by defects,more conducive to accurate and complete separation between the various land cover.The effectiveness and feasibility of the proposed algorithm are verified by experiments based on the point cloud data measured by Optech Titan airborne multi-spectral LiDAR system.The paper has thirty figures,six tables,and fifty-one references.