Research on method for extracting vegetation information based on hyperspectral remote sensing data

When extracting vegetation distribution information from the hyperspectral resolution images of the remote sensing satellites, the influences of mixed pixels and training sample sizes should be considered in order to improve the accuracy of vegetation information extraction. Taking the north part of Guangzhou City as a test area, the linear spectral mixed model (LSMM) and support vector machine (SVM) were applied to classify Hyperion image and evaluate litchi classification. The results were compared with the randomly selected grid points in QuickBird image with 1m spatial resolution. The accuracy of litchi classification reached 85.7%. The corresponding accuracy of Spectral Angle Mapping (SAM) was 74.3%. Results show that the integration of LSMM and SVM can improve the extraction accuracy of vegetation distribution comparing with other traditional spectral extraction methods of remote sensing.