高光谱遥感图像具有特征(波段)数多、冗余度高等特点, 因此特征选择成为高光谱分类的研究热点。针对此问题, 提出了空间结构与光谱结构同时保持的高光谱数据分类算法。考虑高光谱图像的物理特性, 首先对图像进行加权空谱重构, 使图像的空间结构信息自动融入光谱特征, 形成空谱特征集; 对利用最小二乘回归模型保存数据集的全局相似性结构的基础上, 加入局部流形结构正则项, 使挑选的特征子集更好地保存数据集的内在本质结构; 讨论了窗口大小和正则参数对分类精度的影响。对Indian Pines、PaviaU和Salinas数据集的实验表明, 该算法得到的特征子集的总体分类精度达到93.22%、96.01%和95.90%。该算法不仅充分利用了高光谱图像的空间结构信息, 而且深入挖掘了数据集的内在本质结构, 从而得到更有鉴别性的特征子集, 相比传统方法明显提高了分类精度。

Hyperspectral remote sensing image contains the properties of much features(bands) and high redundancy, and the research of hyperspectral image classification focuses on feature selection. To overcome this problem, a hyperspectral image classification algorithm based on spatial and spectral structure preserving was proposed. Considering the physical characteristics of hyperspectral image, the weighted spatial and spectral reconstruction of the image was conducted firstly, in order to incorporate spatial structure information into the spectral feature set automatically, resulting in the spatial-spectral feature set. On the basis that the least square regression model uncovered the global similarity structure and the regularization term revealed the local manifold structure, the intrinsic structure of the spatial-spectral feature set was well preserved by the selected feature subset. The influence of window size and regularization parameter was also analyzed. The experiments on Indian Pines, PaviaU and Salinas datasets show that the classification accuracy of the proposed algorithm reaches 93.22%, 96.01% and 95.90% respectively. The proposed method not only makes full use of the spatial structure information of the hyperspectral image but also uncovers the intrinsic structure of the dataset, which contribute to select more discriminant feature subset and obtain higher classification accuracy compared with conventional methods.structure preserving