Construction of Terrain Information Extraction model in the Karst Mountainous Terrain Fragmentation Area Based on UAV Remote Sensing

Rapid extraction of land use information plays an important role in data support for land construction and development and natural resource management. For south China the Karst mountainous terrain, object type is complicated, and lead to a difficult scene classification and information extraction, this paper use of Rotor Drone acquisition topography broken zone optical remote sensing image and image matching point cloud data, the comprehensive utilization of two-dimensional spectral information in remote sensing and image matching point cloud data in three-dimensional space structure information. The object oriented rule multi-feature extraction method was used to establish the ground object information in the terrain fragmentation area, and to realize the classification of building land, cultivated land, forest land and other ground objects. At the same time, the maximum likelihood method and K-mean method were used to classify visible images in the study area, and the accuracy of the classification was compared with that of object-oriented rule extraction. The results show that: (1) the ground object information extraction based on three-dimensional point cloud data and two-dimensional visible remote sensing image has a high accuracy, with the overall accuracy of 91.8% and the Kappa coefficient of 0.89, which is simple and convenient for multi-factor extraction in complex terrain. (2) Compared with the single visible image for ground object extraction, the overall accuracy is improved by 22.91%, and the Kappa coefficient is improved by 0.29. The addition of image matching point cloud data increases the feature of height factor extraction, which is more conducive to the extraction of ground object and land category information in the terrain fragmentation area.