地面实测地物光谱可提供细致的光谱信息, 表现同种地物不同理化特性和不同种类地物光谱的微小差异, 使利用光谱进行地物识别成为可能。 使用美国HR-768型地物光谱仪, 在塔里木河下游和吐鲁番沙漠植物园实测胡杨、 柽柳、 梭梭和沙拐枣高光谱数据, 利用包络线去除、 一阶微分和二阶微分法对原始光谱进行变换处理, 使用马氏距离法确定所测树种原始光谱和变换光谱的差异显著波段, 利用逐步判别法检验所选差异波段的识别效果。 结果表明: 马氏距离法可准确确定树种识别的最佳波段, 且上述4树种光谱识别波段大多位于近红外区。 原始光谱、 包络线去除、 一阶微分和二阶微分四种光谱对4树种的识别精度分别为: 85%, 93.8%, 92.4%和95.5%;可见, 原始光谱经变换处理可提高树种的识别精度。 但不同研究对象、 不同光谱处理方法, 提高识别精度的效率不同。 研究结果将为大尺度高光谱遥感影像用于荒漠植物分类与生境监测和评价提供依据。

The hyperspectral reflectance of Populus euphratica, Tamarix hispida, Haloxylon ammodendron and Calligonum mongolicum in the lower reaches of Tarim River and Turpan Desert Botanical Garden was measured by using the HR-768 field-portable spectroradiometer. The method of continuum removal, first derivative reflectance and second derivative reflectance were used to deal with the original spectral data of four tree species. The method of Mahalanobis Distance was used to select the bands with significant differences in the original spectral data and transform spectral data to identify the different tree species. The progressive discrimination analyses were used to test the selective bands used to identify different tree species. The results showed that The Mahalanobis Distance method was an effective method in feature band extraction. The bands for identifying different tree species were most near-infrared bands. The recognition accuracy of four methods was 85%, 93.8%, 92.4% and 95.5% respectively. Spectrum transform could improve the recognition accuracy. The recognition accuracy of different research objects and different spectrum transform methods were different. The research provided evidence for desert tree species classification, monitoring biodiversity and the analysis of area in desert by using large scale remote sensing method.