摘要
以西班牙Barrax地区的HyMap数据为例,利用K-L变换、波段指数、分段K-L变换、小波变换和可分性准则等常见的特征选择和提取方法,进行了特征提取和特征选择;然后,基于常用的监督分类法,通过各自的分类精度对各特征提取和特征选择法进行了比较和分析。实验结果表明使用可分性准则进行的特征选择时,基于波段指数、离散度平均值的分类精度相对较高;基于变换的特征选择中,K-L变换、分段K-L变换和小波变换的分类精度较高,三者基本上一致。而在监督分类方法的比较分析上,平行六面体法相对较差,最小距离法分类精度则相对较高,而最大似然法则比较稳定。
In this paper, a case study on the feature selection and extraction of the hyperspectral data of HyMap （128 bands） in Barrax, Spanish, is carried out based on K-L transform, segment K-L transform, wavelet transform, band index and some separability functions so as to select the optimal band combinations of the spectra of land covers. The original HyMap data were derived from Barrax test site in Spanish at 9：00 o＇clock, June 4, 1999, the air altitude was 3500 m, and the spatial resolving power was about 5 m. the spectral range of HyMap airborne imaging spectrometer was 400-2 480 nm, there were 128 wavebands except two with high vapor absorption near 1.4 μm and 1.9μm. The wavebands can be divided into 4 groups, i.e., the visible light, near infrared rays, shortwave infrared ray 1 and shortwave infrared ray 2, each waveband group includes 32 wavebands, and the waveband width is about 16 nm. The atmospheric correction and geometry correction of the test data are carried out, and the values of reflectivity of the ground objects are normalized. The classification accuracies of different methods of the feature selection and extraction are compared and analyzed based on the common supervised classification methods, such as the maximum likelihood, parallelepiped and minimum distance. The results show that the classification accuracy with feature selection based on band indexes and average dispersibility is high, and that with feature extraction based on K-L transform, segment K-L transform and wavelet transform is high. The compared and analyzed results show that the classification accuracy of the parallelepiped is relatively low, that of minimum distance is relatively high, and that of the maximum likelihood stable.
出处
《干旱区地理》
CSCD
北大核心
2006年第1期143-149,共7页
Arid Land Geography
基金
国家自然科学基金项目"水环境指标的高光谱响应模型研究"(40271048)
863项目(2002AA130010-1-4)
关键词
高光谱
特征选择和提取HyMap
小波变换
hyperspectrum
feature selection and extraction
wavelet transform
HyMap
classification
