论文部分内容阅读
由于高光谱数据的海量高维特征,对其进行降维处理成为高光谱遥感研究的一个重要问题。波段选择算法由于能够有效地保留原始数据的信息,在高光谱数据降维及后续的遥感识别与分类等方面具有明显的优越性。文章提出了一种基于正交投影散度(OPD)的波段选择方法,该方法继承了正交子空间投影(OSP)算法的特点,通过把原始数据投影到特征空间,实现感兴趣目标与背景噪声的分离;通过最大化光谱向量之间的相似性测度以及顺序浮动前向搜索(SFFS)算法,实现快速的波段选择。利用HYDICE和HY-MAP高光谱数据进行实验验证,并与其他传统波段选择算法,如光谱角度匹配、欧式距离、光谱信息散度和LCMV-BCC等进行对比,结果表明该算法在高光谱数据波段选择方面具有较好的适用性和鲁棒性,能够有效地应用于高光谱遥感数据的降维研究。
Due to the massive high-dimensional features of hyperspectral data, reducing the dimension of hyperspectral data becomes an important issue in hyperspectral remote sensing. The band selection algorithm has obvious superiority in reducing dimension of hyperspectral data and subsequent remote sensing identification and classification because it can effectively preserve the information of the original data. In this paper, a band selection method based on Orthogonal Projection Divergence (OPD) is proposed, which inherits the characteristics of the Orthogonal Subspace Projection (OSP) algorithm. By projecting the original data into the feature space, the target and background Noise separation; fast band selection by maximizing the measure of similarity between spectral vectors and the Sequential Floating Forward Lookup (SFFS) algorithm. The experimental data are validated by HYDICE and HY-MAP hyperspectral data, and compared with other traditional band selection algorithms such as spectral angle matching, Euclidean distance, spectral information divergence and LCMV-BCC. The results show that the proposed algorithm is effective in hyperspectral data band It has good applicability and robustness in selection and can be effectively applied to the dimensionality reduction research of hyperspectral remote sensing data.