Fusion of high spatial resolution hyperspectral image and LiDAR data for buildings and trees extraction
High spatial resolution hyperspectral image and LiDAR (Light Detection And Ranging) data play an important role in feature extraction applications.In this paper,the combination of hyperspectral image and LiDAR data,both at 0.5 m spatial resolution,were used to extract buildings and trees in Anyang urban areas.Normalized digital surface model (nDSM) was generated from the difference of LiDAR derived digital surface model (DSM) and digital terrain model (DTM).Normalized Difference Vegetation Index (NDVI) was computed from hyperspectral image using NIR (near infrared) and R (red) bands.Minimum noise fraction (MNF) transformation was used to reduce the dimension of the hyperspectral image.Ruled-based object-oriented feature extraction method was used to extract buildings and trees from the combination of nDSM and NDVI data.Support Vector Machine (SVM) algorithm based supervised classification was applied to the MNF transformed hyperspectral image for different types of trees extraction.Results showed that buildings were extracted accurately,and tilted-roofed buildings could be discriminated from nDSM slope image.Four types of trees were classified with the overall accuracy of 87.81 percent and Kappa coefficient of 0.79.
Yanfang Dong Pang Yong Dian Yuanyong
Institute of Earthquake Science, China Earthquake Administration, Beijing100036, China Institute of Forest Resources Information Techniques, Chinese Academy of Forestry, Beijing 100091, C Institute of Forest Resources Information Techniques, Chinese Academy of Forestry, Beijing 100091, C
国际会议
北京
英文
71-80
2013-10-09（万方平台首次上网日期，不代表论文的发表时间）