Application Of Hyperspectral Remote Sensing In Forest Tree Species Discrimination

The emergence of hyperspectral remote sensing technology will provide chance for solving problems of identifying forest tree species and tree species group precisely. The classification of tree species was analyzed and studied from two levels of leaves spectral data and hyperspectral remote sensing data.According to hyperspectral remote sensing data has lots of bands and large amounts of data,The paper maked use of repeatable double factor analysis of variance inspection,extraction and classification of features bands for leaf hyperspectral data. The hyperspectral original data of fifteen tree species leaves are checked differently by using repeatable double factor variance analysis, the result shows that the spectrum difference of different tree species leaves is remarkable. Using the tree species leaves hyperspectral data to identify the tree species is tremendous potential, there are too many bands,large amounts of data and big redundancy when identifying the tree species with hyperspectral remote sensing technology, the methods of derivative reflectance (the first derivative reflectance and second derivative reflectance)and the continuum removal were used to deal with the original spectral data of fifteen tree species leaves, analyze and compare curves of the original spectrum, the first derivative reflectance, the second derivative reflectance and the continuum removal of the different tree species, and the bands with bigger difference were selected to identify the different tree species. Then the Euclidean distance method was used to test the selective bands identifying different tree species, and the results showed that the selective bands could identify different tree species effectively. The most of the bands for identifying different tree species were near-infrared bands.The tree species at the study area were classified by using EO-1 Hyperion data. Firstly, hyperspectral data was reducted dimension by using the principal component analysis, the residential areas, agricultural land and water were classified and extracted with the maximum likelihood classification; secondly, according to the sample data, the best bands for classification of the tree species are extracted by using band index, subsection principal component analysis, derivative reflectance (the first and second derivative reflectance) and the continuum removal etc five methods, then the tree species were beening classified. the results show that the overall classification accuracy of the five methods were more than 85.0%, the Kappa coefficients are more than 0.8. Therefore, the extracted bands by the five methods could reflect the differences between different species, and could identify leizhu, evergreen broadleaved forest, deciduous broadleaved forest, bamboo, pinus massoniana, fir and miscellaneous bamboo etc seven tree species (group) well.Species classification problem could be solved properly by Hyperspectral remote sensing technique rather than Multi-spectrum remote sensing technique. Hyperspectral remote sensing technique can achieve high accuracy of classification and meet the need of forest resource survey. Feature bands can be well extracted by the methods of the band index, the subsection principal component analysis, derivative reflectance and the continuum removal, they could reduce hyperspectral dimension effectively, eliminate hyperspectral data redundancy, improve classification accuracy, the derivative reflectance and the continuum removal could increase the spectral differences between different species make a distinction between overlapping spectra.The five methods were effective methods for the reduction dimension of Hyperspectral remote sensing.