提出了一种基于加权空-谱距离(WSSD)的相似性度量方法, 并将其应用到最近邻分类器(KNN)中, 导出了一种新的高光谱图像分类算法。 该算法利用高光谱图像的物理特性, 通过引入空间窗口和光谱因子这两个参数来挖掘出图像中的空间信息与光谱信息, 利用空间近邻点对中心像元进行重构。在最大限度减少图像冗余信息的基础上, 增大了同类像元间的相似性以及异类像元间的差异性, 获得了更为有效的鉴别特征, 从而更好地实现了数据间的相似性度量。基于Indian Pines和PaviaU高光谱数据集进行了实验, 结果表明: 将提出的WSSD-KNN算法应用于高光谱图像分类时, 其分类精度高于其他算法, 总体分类精度分别达到了91.72%和96.56%。由于算法较好地融合了图像中的空间-光谱信息, 提取出了更为有效的鉴别特征, 故不仅有效地改善了高光谱数据的地物分类精度, 而且可在训练样本较少时, 保持较高的识别率。

A spatial consistency measurement method based on the Weighted Spatial-Spectral Distance(WSSD) is proposed and applied to the K Nearest Neighbor(KNN) classifier, and a new hyperspectral image classification algorithm is obtained. On the basis of the physical characters of hyperspectral images, the proposed algorithm combines both spatial window and spectral factor to obtain the spatial information and spectral information , and uses the spatial nearest points to reconstruct the center point and to reveal the local spatial structure. With effectively reducing the redundant information in the image, this algorithm increases the consistency of the same kinds pixels and the difference of the different kinds pixels and obtains extract discriminating features, so it implements the consistency measurement between the data points. The experiments were performed on the Indian Pines and PaviaU hyperspectral data sets. Experiment results show that the WSSD-KNN algorithm has better classification accuracy than other algorithms when it is applied to the classification of hyperspectral image, and the overall classification accuracies reach 91.72% and 96.56%, respectively. With the spectral information, spatial information and extract discriminating features, the proposed algorithm effectively improves ground object classification accuracy of hyperspectral data and has better recognition ability in less train samples.