采用分段行-列核2DPCA的高光谱图像降维

摘要

Two-dimensional principal component analysis (2DPCA) is successfully applied to a hyperspectral image, and is less time-consuming with better dimensionality reduction performance. Based on the two-dimensional principal component method, the segmented row-column kernel 2DPCA algorithm is realized to excavate non-linear information. The dimension reduction effect of the row-column 2DPCA method, the segmented row-column 2DPCA method, and the row-column kernel 2DPCA method are compared and analyzed. The hyperspectral image is divided into five subspaces by correlation, and the kernel 2DPCA of rows and columns is realized by transforming the data structure. Finally, the row and column results are merged to obtain dimensionality reduction results. The reduced dimension results show that the segmented row-column kernel 2DPCA method has the highest figure definition and edge intensity at a higher information retention rate. The pixel line graph indicates that the proposed method can distinguish the different features better, and is well suited to classification and target recognition.%二维主成分方法计算时间少,降维效果好,被成功应用到高光谱图像降维中.基于二维主成分方法,为挖掘高光谱图像的非线性信息,实现了分段行-列核2DPCA方法的降维,并对比分析了行-列2DPCA方法、分段行-列2DPCA方法和行-列核2DPCA方法的降维效果.利用相关性将高光谱图像划分为5个子空间,通过转换数据结构来实现行和列的核2DPCA变换,最后将行和列结果进行融合得到降维结果.降维结果表明,在较高信息保持率情况下,分段行-列核2DPCA方法具有最高的图像清晰度和边缘强度.不同地物像元像素折线图表明,分段行-列核2DPCA方法能更好地区分不同地物,可以很好地应用于地物分类和目标识别.