针对传统高光谱影像特征提取算法大多仅考虑光谱信息或提取空间信息不够精细的问题,提出了一种监督空间正则化流形鉴别分析(SSRMDA)算法,以提高遥感地物的分类性能。该算法首先利用样本数据的标签信息构建谱域类内图和类间图,以揭示高光谱数据潜在的非线性流形结构;然后构建空域类内图,并将空间信息以正则化方式与光谱信息融合,实现谱-空信息的有效融合,并可在低维空间内使类内数据更加聚集,增强嵌入数据的可分性。在Indian Pines和Washington DC Mall数据集上的实验表明,所提算法的总体分类精度分别为91.58%和96.67%,说明所提算法有效提升了地物分类能力,尤其在小样本下的优势更为明显,更有利于实际应用。

Traditional feature extraction algorithms consider only spectral information in the hyperspectral image (HSI) and cannot extract fine spatial information. To solve this problem, this paper proposes a supervised spatially-regularized manifold discriminant analysis (SSRMDA) algorithm to improve the classification performance of ground objects in the HSI. The SSRMDA algorithm firstly constructs a spectral-domain intraclass image and an interclass image by using the label information of training samples, which reveals the potential nonlinear manifold structure of hyperspectral data. Based on that, a spatial-domain intraclass image is constructed, and it combines the spectral information of HSI by regularization to realize the effective fusion of spectral-spatial information. In low-dimensional space, the intraclass data in low dimensional space becomes more clustered and the separability of embedded features is enhanced. Experiments on the Indian Pines and Washington DC Mall datasets show that the overall classification accuracy of the SSRMDA algorithm reaches 91.58% and 96.67%, respectively, which denotes that the proposed algorithm effectively improves the classification ability of ground objects. Compared with other feature extraction algorithms, the proposed algorithm is effective in practical applications, especially when a small number of training samples are available.