谐波能量谱特征向量的高光谱影像Bayes分类

Bayes classification for hyperspectral image based on energy spectral characteristic vectors obtained by harmonic analysis

对于高光谱影像存在高维非线性、数据冗余多、纯训练样本难以提取等不足,引入频率域空间的谐波分析(harmonic analysis,HA)理论并提出了一种高光谱影像的HA-Bayes监督分类方法。该方法在保持高光谱数据空—谱特性不变的情况下,从光谱维角度分析不同分解层的影像光谱谐波特征,将高光谱影像变换成由谐波能量谱组成的频率域特征矢量信息。通过建立谐波能量谱特征向量的先验知识,实现Bayes准则下谐波能量谱特征矢量信息判别与分类,最终实现高光谱影像分类。将此方法应用到ROSIS高光谱影像分类时获得的分类总体精度达85.5%,Kappa系数也达到了0.812。进一步实验也证明了频率域的谐波分析在高光谱遥感影像特征提取与分类方面具有更好的优势和潜力。

Considering the disadvantages of hyperspectral image as high-dimensional nonlinear,data redundancy and hard extractability to pure training samples,this paper proposed a supervised HA-Bayes classification method for the hyperspectral image using the frequency domain harmonic analysis( HA) theory. The method could analyze the harmonic characteristics of the image spectra by decomposing pixel-spectra into different levels with HA from the perspective of spectral dimension,maintaining the unchanged space-spectral characteristics of hyperspectral data and transforming the hyperspectral image into the characteristic vector information that were consisted of harmonic energy spectra. Next,it used the Bayes criteria to discriminated the vector information according to the established prior knowledge from training samples,so as to realize the hyperspectral image classification finally. At the same time,it applied the HA-Bayes classifier to a ROSIS hyperspectral image,and the overall accuracy of classification reached 85. 5% with the Kappa coefficient up to 0. 812. Further experiments also prove the HA in frequency domain,which has better advantages and potential in feature extraction and classification for hyperspectral remote sensing image.