森林树种高光谱波段的选择

Band Selection from Hyperspectral Data of Forestry Species

高光谱是遥感技术发展的一个重要方向,也是地物识别的重要手段。本研究利用地物光谱仪对杉木、雪松、小叶樟树和桂花树4个树种进行高光谱数据测量,探索不同树种在不同波段上的识别能力。研究采用了逐步判别分析法和分层聚类法对实验数据进行数据分析。结果表明:逐步判别分析法选择的波段主要位于红、绿、蓝、和近红外区;分层聚类法选择的波段除了红、绿、蓝、和近红外波段外,还增加了蓝-绿边缘、绿-红边缘和红边区的波段。所选择的波段比原始波段在树种识别时具有更高的精度,最高识别精度达96.77%;边缘区波段对树种的识别有重要作用;用对数-微分变换处理较其他方法处理对树种识别有更好的效果。

Hyperspectal remote sensing is an important aspect of the development of remote sensing, and also an important means to identify the objects on the earth. In order to explore the ability to identify different tree species in different bands, the field spectrometry instrumentation was used to measure hyperspectral data of four different tree species, such as sha mu, xue song, zhang shu and gui hua in this research, and stepwise dlscriminant analysis and hierarchical clustering were used to analyse the experimental data. The result suggested that bands selected from stepwise discriminant analysis mainly lied along the blue, green, red and near-infrared bands. Using hierarchical clustering, in addition to the blue, green, red and near-infrared bands, the spectral bands along the bluegreen edge, green-red edge and red curves were selected. Bands selected had more accurate identification than the original ones in tree species discrimination. The most accurate identification is 96.77 %, and the bands lied along the edges had important information for discrimination of tree species. The spectral data, dealed with the transformation of logarithm and differential coefficient, could achieve better accuracy than others.