论文部分内容阅读
为解决高光谱图像中高维数据和有标记训练样本不足的矛盾导致“维度灾难”问题,提出一种无监督的基于流形学习的波段选择(MLBS)方法。首先通过流形学习方法,得到原始数据的流形嵌入映射;然后通过LASSO优化过程,运用顺向坐标下降算法,得到原始波段对每个流形结构维度的贡献度;最后统计每个波段的贡献度,选取贡献度大的波段形成波段子集。用真实的AVIRIS高光谱图像对算法进行仿真实验的结果表明,本文方法在小样本下的高光谱地物分类识别问题上具有良好的效果。
In order to solve the problem of “dimensionality disaster” caused by the contradiction between the high dimensional data in hyperspectral image and the shortage of labeled training samples, an unsupervised band selection (MLBS) method based on manifold learning is proposed. Firstly, the manifold embedded map of the original data is obtained by the manifold learning method. Then, the contribution of the original band to each manifold structure dimension is obtained through the LASSO optimization process and the forward coordinate descent algorithm. Finally, the contribution of each band Degree, select the contribution of a large band to form a subset of the band. The simulation results of the real AVIRIS hyperspectral image show that the proposed method has a good effect on the classification and identification of hyperspectral features under small samples.