论文部分内容阅读
针对高光谱遥感数据特征提取方法的研究,提出了一种新的监督近邻重构分析(Supervised Neighbor Reconstruction Analysis,SNRA)算法。该方法首先利用同一类别的近邻数据点对各数据点进行重构;然后在低维嵌入空间中保持该重构关系不变,尽可能地分离开非同类数据点,并利用总体散度矩阵来约束数据间的相关性;最后求解得到一个最佳投影矩阵,进而提取出鉴别特征。SNRA算法不仅保持了同类数据的局部结构而且增强了非同类数据的可分性,同时减少了数据的冗余信息。在Indian Pine和KSC高光谱遥感数据集上的实验结果表明:提出的方法能更好地揭示出高光谱遥感数据的内在特性,提取出更有效的鉴别特征,改善分类效果。
Aiming at the research of hyperspectral remote sensing data feature extraction, a new supervised Neighbor Reconstruction Analysis (SNRA) algorithm is proposed. In this method, each data point is reconstructed by using the nearest neighbor data points of the same class. Then, the reconstruction relation is kept unchanged in the low-dimensional embedding space, the non-homogeneous data points are separated as much as possible, and the overall dispersion matrix is used Constraining the correlation between the data; Finally, an optimal projection matrix is obtained and the discriminant features are extracted. The SNRA algorithm not only maintains the local structure of similar data but also enhances the separability of non-homogeneous data and reduces the redundant information of data. Experimental results on Indian Pine and KSC hyperspectral remote sensing datasets show that the proposed method can better reveal the inherent characteristics of hyperspectral remote sensing data, extract more effective discriminant features and improve the classification results.