论文部分内容阅读
针对传统的基于特征提取的高光谱图像分类算法大多只考虑光谱信息而忽略空间信息的问题,提出了一种基于空谱半监督局部判别分析(S3 ELD)和空谱最近邻(SSNN)分类器的高光谱图像分类算法。该算法结合高光谱图像的空间一致性,在利用标记样本的判别信息保持数据集可分性的基础上,定义空间近邻像元散度矩阵来保存像元的空间近邻结构,提出基于空谱距离的相似性度量并将其应用于局部流形结构的发现和SSNN的构建。S3 ELD算法不仅能揭示数据集的局部几何关系,而且增强了光谱域同类像元和空间域近邻像元在低维嵌入空间的聚集性。结合SSNN进行分类,进一步提升了分类精度。利用PaviaU和Salinas数据集进行的实验结果表明,S3 ELD算法的总体分类精度分别达到了92.51%和96.29%;与现有几种算法相比,该算法能更有效地提取出判别特征信息,并达到更高的分类精度。
Traditional hyperspectral image classification algorithms based on feature extraction mostly only consider the spectral information and neglect the spatial information. Based on the semi-supervised local discriminant analysis (S3 ELD) and the space-spectrum nearest neighbor (SSNN) classifier Hyperspectral image classification algorithm. Based on the spatial consistency of hyperspectral images, this algorithm uses the discriminant information of labeled samples to maintain the separability of data sets and defines the spatial nearest neighbor scatter matrix to preserve the spatial neighbor structure of the pixels. Based on the spatial distance Similarity measure and apply it to the discovery of local manifold structure and the construction of SSNN. The S3 ELD algorithm can not only reveal the local geometric relationship of data sets, but also enhance the clustering of similar pixels in the spectral domain and the neighboring pixels in the spatial domain in the low-dimensional embedded space. Combined with SSNN for classification, to further improve the classification accuracy. Experimental results using PaviaU and Salinas datasets show that the overall classification accuracy of S3 ELD algorithm is 92.51% and 96.29%, respectively. Compared with several existing algorithms, this algorithm can extract discriminant characteristic information more effectively Achieve higher classification accuracy.