基于实测高光谱数据的树种分类

Tree Species Classification Based on Measured Hyperspectral Data

本文利用实测高光谱数据,结合光谱特征分析方法和全波段分析方法将15种树木进行有效分类。与多光谱数据对比,高光谱技术具有数据量多、波段窄等特点,根据这一特点高光谱数据可对树种进行更为精细的识别。其中,选择合适的分类波段对于提高树种分类精度非常重要。该文对于使用归一化、一阶微分、倒数等12种变换方法对原始光谱数据进行变换,分析不同变换下树种的光谱曲线,并使用光谱差异较大的特征区域及全波段两种角度树种进行识别分类。总体上,选取特征区域与全波段分类结果相比,在6种变换中,特征区域分类精度大于全波段分类精度,选取特征区域对于二阶微分变换中冬青分类有明显提高。其余6种变换中,全波段分类精度大于特征区域分类精度,对于低通滤波变换对于构树,选取特征区域与全波段相比分类精度较低。

Based on measured hyperspectral data,this study effectively classifies 15 tree species by analyzing spectral features and full band.Compared with multispectral data,hyperspectral technology has larger amount of data and narrow band,which helps hyperspectral data identify tree species more precisely.It is noted that it is very important to select the appropriate classification band for improving the classification accuracy of tree species.In this paper,12 kinds of transformation methods such as normalization,first-order differentiation and reciprocal are used to transform the original spectral data,analyze the spectral curves of tree species under different transformations,and identify and classify tree species by using the feature regions with large spectral differences and full band.In general,compared with the results of full band classification,the accuracy of feature region classification is higher than that of the band classification in nine kinds of transformations,and the selection of feature region can significantly improve the classification of holly in second-order differential transformation.In the other seven transformations,the accuracy of full band classification is higher than that of feature region classification.For low-pass filter transformation,for Broussonetia papyrifera,the accuracy of feature region selection is lower than that of full band classification.