空-谱协同正则化稀疏超图嵌入的高光谱图像分类

传统依据图嵌入的高光谱图像维数约简算法多数仅利用光谱信息表征像元间单一关系,忽视了数据间的多元几何结构。本文提出了一种面向高光谱图像分类的空-谱协同正则化稀疏超图嵌入算法(SSRSHE)。该算法首先利用稀疏表示揭示像元之间的相关性,自适应选择近邻,并构建稀疏本征超图和惩罚超图,以有效表征像元间的复杂多元关系,并进行正则化处理。然后利用遥感图像空间一致性原则,计算局部空间邻域散度来保持样本局部邻域结构,并引入样本总体散度来保持高光谱数据的整体结构。在低维嵌入空间中,尽可能使类内数据聚集、类间数据远离,提取鉴别特征用于分类。在Indian Pines和PaviaU高光谱遥感数据集上试验结果表明,本文算法总体分类精度分别达到86.7%和 92.2%。相比传统光谱维数约简算法,该算法可有效改善高光谱图像地物分类性能。

Using spatial-spectral regularized hypergraph embedding for hyperspectral image classification

In recent years, many graph embedding methods were developed for dimensionality reduction (DR) of hyperspectral image (HSI), while these methods only use spectral information to reveal a simple intrinsic relation and ignore complex spatial-spectral structure in HSI. A new DR method termed spatial-spectral regularized sparse hypergraph embedding (SSRSHE) is proposed for the HSI classification. SSRSHE explores sparse coefficients to adaptively select neighbors for constructing the regularized sparse intrinsic hypergraph and the regularized sparse penalty hypergraph. Based on the spatial consistency property of HSI, a local spatial neighborhood scatter is computed to preserve local structure, and a total scatter is computed for global structure of HSI. Then, the optimal discriminant projection is obtained by possessing better intrinsic data compactness and penalty pixels separability, which is beneficial for classification. The experimental results on Indian Pines and PaviaU hyperspectral data sets show that the overall classification accuracies respectively reach 86.7% and 92.2%. The proposed SSRSHE method can effectively improve classification performance compared with the traditional spectral DR algorithms.