鄱阳湖5种典型植被高光谱特征波段选择与光谱分类识别

Hyperspectral Characteristic Band Selection and Spectral Classification of Five Typical Vegetation in Poyang Lake

光谱特征波段的选取是植被高光谱分类识别的重要基础之一。利用鄱阳湖5种典型植被的实测高光谱数据,在对数据进行预处理和分析的基础上,提出了一种基于均值极差阈值法的光谱特征波段选择方法,并利用马式距离-光谱角法对不同植被种类进行识别。结果表明:所提方法有效提取了植被间的光谱特征波段,分别为1111~1132nm、1466~1522nm和1577~1750nm,三个波段全部位于红外区域;在光谱特征波段范围内,利用马氏距离-光谱角法可对不同植被类型进行有效识别,其中南荻的光谱分类精度最高,灰化薹草的光谱分类精度最低,为84%,总体分类精度为91%,分类效果较好。

The selection of spectral characteristic band is one of the important basis of plant hyperspectral classification. On the basis of measured hyperspectral data of five typical vegetation in Poyang Lake and data preprocessing and analysis, a method of spectral characteristic band selection based on the average and range threshold method is proposed, and the Mahalanobis distance-spectral angle method is used to identify the species of different vegetation. The results show that the proposed method effectively extracts the spectral characteristic band of the vegetation, the band is 1111-1132 nm, 1466-1522 nrn, and 1577-1750 nm, respectively, and all of them are located in the infrared region. In the spectral characteristic band, the Mahalanobis distance-spectral angle method can effectively identify different vegetation types, the spectral classification accuracy of Triarrhena is the highest, the accuracy of Cynodon is 84%, and the overall classification accuracy is 91%, which shows that the classification effect is good.