The Explore Of Forest Tree Species Discrimination Based On Hyperspectral Remote Sensing Data

With the continuous development of to hyperspectral technology, hyperspectral data have been widely used in all walks of life. Hyperspectral data in the application of forestry in our country was still in its infancy, and the research focused on the research on pixel level, which cannot satisfy the actual demand. How to extract forest tree species information from hyperspectral data fast and accuratly was the focus of this paper.Hyperspectral data with its advantage of the high spectral resolution would provide the high accuracy of identification of forest tree species. However, high degree of data redundancy would set back its application because of the hyperspectral data volume was large. The first derivative reflectance, second derivative reflectance and the continuum removed methods were used to resample the bands and get the features to identify the forest tree species.The classification of Linear Spectral Unmixing and Spectral Angle Mapping were based on the accuracy of the extracted endmember. In this paper, PP1was use to extract endmember of Mao bamboo, Ventricousinternode, masson pine, evergreen broad-leaf forest, broadleaved deciduous forest. The extracted endmembers were used to classify the forest species with the SAM and LSU.This article would discuss the forest tree species classification problem including the hyperspectral data dimension reduction, endmember extraction, classification methods of the maximum likelihood classification、spectral Angle mapping method and the SAM. hen comparing different characteristics among these ways,and run the linear decomposition of mixed pixels with the end member. The class accuracies were no less than65%and the Kappa coefficients were no less than0.45. The result showed that the bands of spcetral original with SAM would get the best accuracy, with the maximum likelyhood the second accuracy.The classification accuracies based on the Characteristic Bands of four methods with the SMA were less than the original bands. It was different than some articles conclusions. It showed that some methods of band extraction of species are not fully reflected the Canopy spectra! Characteristics. In this paper, the extraction bands of character are10bands, then the original bands were172, too few bands of Characteristics was the reason of accuracy decrease. In general, species discrimination with hyperspectral data would be feasible, SMA would be the good method to discriminate species. Feature band extraction would decrease the redundant information. But sometimes it would lose some information useful to discriminate species and decrease the classification accuracy.