The Study On Identification Of Forest Tree Species Based On Hyperspectral Image

The development of remote sensing technology provides the possibility to extract the information of forest resources quickly and efficiently. Dominant tree species information is the core index of forest resources survey, and it is the focus of remote sensing technology application. Compared with high resolution image, hyperspectral image has rich band information which can detect the subtle differences in the spectra of different tree species and make it possible to identify and classify forest tree species. In this study, the experimental area is in Culai mountain, Shandong province, and the experimental data contains Hyperion EO-1 hyperspectral image and measured spectral data. The study focus research on the method of selecting characteristic band and classifying tree species, it also researched the application potential of spectral angle classification.High spectral data have serious redundancy and it’s hard to distinguish spectrum between tree species, so the research transformed spectral data, selected characteristic bands, identified and validated tree species. Experiments show that the Transformation Analysis method can effectively enhance the spectral difference between species, Feature Band Selection method can not only reduce the dimension, but also identify tree species efficiently. After First order differential transformation, the characteristic bands selected by Markov Distance method get the highest accuracy at 96.3%, the selected characteristic bands are concentrated in the visible and near infrared region.Aiming at the effect of dimension reduction method of feature selection and feature extraction to the classification of tree species, based on the above results, the study made dimensionality reduction of Hyperion image, used Maximum Likelihood method, Support Vector Machine method and improved Spectral Angle method to verify tree species classification accuracy. As a result, the three classification methods can achieve the accuracy no less than 74.22%, kappa coefficient is no less than 0.67, which can meet the production requirements. This indicates that Feature Selection method can not only greatly reduce the amount of computation, but also retain characteristic information of image spectrum which have more potential on improving the classification accuracy of tree species.Further, the accuracy of the improved Spectral Angle Classification method is improved by nearly 5% compared with the original method and kappa coefficient increased by nearly 0.07 which can identify and classify tree species efficiently. Improved Spectral Angle Classification method can effectively improve the classification accuracy of tree species and classification method based on spectral information has great potential in hyperspectral remote sensing classification applications.It can be seen that the use of hyperspectral data for tree species classification and identification is feasible. Characteristic band analysis, selection and classification by using spectral information of forest tree species can improve reduced dimension speed, classification speed and classification accuracy for hyperspectral remote sensing image.