为了增强高光谱遥感图像的分类效果,提出基于谱聚类和稀疏表示的两级分类算法。利用谱聚类将待分类的像元及其邻域内所有的像元分成两类,利用联合稀疏表示模型确定按规则选取的其中一类的具体类别,并以该类别作为像元的类。该算法充分利用高光谱图像的光谱及空间信息,两级分类过程均考虑了噪声及区域边界对分类效果的影响。进一步利用空间信息对分类算法进行修正,即关联邻近像元的类别,平滑分类结果。数值实验表明,该算法的分类精度高、稳定性好、抗噪性强。

In order to improve classification effect of hyperspectral image, a classification algorithm with two levels is proposed based on spectral clustering and sparse representation. Pixels to be classified and its neighborhood pixels are divided into two parts by spectral clustering. The class of selected pixels is identified by the joint sparse representation model. This algorithm makes full use of hyperspectral image spectral and spatial information of hyperspectral images, and both of the two levels. Finally, the proposed algorithm is corrected with the spatial information, namely, neighboring pixels' class is associated and classification results is smoothed. Numerical experiments demonstrate that this algorithm has high classification accuracy, good stability and anti-noise performance.