中文摘要：

针对目前高光谱图像半监督降维算法中基于流形学习的开放性选择近邻参数问题,以及利用传统算法不能有效地获取标签数据的局部信息,提出了一种无需考虑近邻参数的半监督局部稀疏嵌入（SELSE）算法.该算法基于稀疏表示理论,通过求解范数优化问题构建稀疏系数图,并且利用有限的标签数据最大化类间信息,提取高光谱图像的特征.在AVIRIS高光谱遥感图像的Indian Pine数据集上进行仿真实验,结果表明所提出算法在分类精度和计算效率上都有所提高.

英文摘要：

Based on manifold learning for the current openness neighbor parameters to select the hyperspectral image semi-supervised dimensionality reduction,and local information using traditional algorithms cannot effectively get tag data,a semi-supervised local sparse embedding（SELSE）algorithms without considering the parameter for hyperspectral image feature extraction was proposed.Based on sparse representation theory,algorithm composition optimization problem by solving the norm,and used tag data to maximize the limited class information. With AVIRIS hyperspectral remote sensing images Indian Pine dataset in the simulation experiment,the results show that the algorithm could improve classification accuracy and computational efficiency to a certain degree.