基于分段行列2D-PCA的高光谱图像数据降维方法

Data Dimension Reduction Method for Hyperspectral Images Based on Segmented Column-and-line 2D-PCA

针对传统二维主成分分析(2D-PCA)方法不能直接应用于高光谱图像数据降维的不足,提出一种基于分段行列2D-PCA的降维方法。利用高光谱图像波段间的相关系数进行波段子空间划分,在各子空间内通过旋转构建新的数据模型,以2D-PCA方法提取其行、列主成分信息,经过图像重建得到行、列主成分图像,对各波段子空间的行、列主成分图像进行小波分解,按照不同规则融合低频、高频系数,再通过小波逆变换得到降维后的图像。实验结果表明,与PCA和分段PCA方法相比,该方法在保证降维图像质量的前提下可缩短运算时间,提高高光谱图像的降维效率。

Traditional 2-Dimensional Principal Component Analysis（ 2D-PCA） method cannot be used to reduce the dimension of hyperspectral image data, so this paper proposes a dimension reduction method based on segmented columnandline 2D-PCA. The whole bands of hyperspecral images are divided into different groups according to the correlation coefficients between them. In each group,the hyperspectral data are revolved to two different directions and reconstructed as two new data models. The features of line and column principal components of new data models are extracted separately by 2D-PCA. Line and column principal component images are obtained by image reconstruction. Wavelet decomposition is carried out on the obtained imanges,and the low frequency coefficients and the high frequency coefficients are fused with different rules. At last,the dimension reduced image is gotten by wavelet inverse transformation. Experimental results show that compared with classical PCA and segmented PCA, segmented column-line 2D-PCA in this paper reduces operation time while keeping good data dimension reduction performance,improving efficiency of dimension reduction for hyperspectral images.