基于LLE低维嵌入的高光谱影像分类

Classification of Hyperspectral Images Based on LLE Lower Dimensionality Embedding

针对高光谱数据中内在的非线性流行结构,分析了LLE低维嵌入算法的基本原理,给出了该算法的计算步骤。介绍了模糊ISODATA分类算法的基本思想,在计算目标函数中,利用测地距离代替欧氏距离,对模糊ISODATA分类算法进行改进。利用两套PHI高光谱影像数据,在LLE低维嵌入结果上实现了ISODATA分类实验。结果表明:LLE低维嵌入后的数据能够降低ISODATA影像分类的迭代次数与计算时间,提高分类的效率;与原始ISODATA分类算法相比,改进的ISODATA分类算法能够更好地挖掘类别之间的自组织关系,提高分类的可靠性。

According to the nonlinear manifold structures that exists in hyperspctral images inherently,basic principles of LLE lower dimensionality embedding algorithm were analyzed,and then,the computing process of the algorithm was given.Principles of fuzzy ISODATA classification algorithm were given.For the ISODATA classification algorithm,Geodesic Distance was proposed to replace the Euclid Distance to compute target function.Two PHI hyperspectral image datasets were used,and ISODATA classification experiments were done on the LLE lower dimensionality embedding data.Experimental results proved that LLE lower dimensionality embedding data could reduce the iteration and computing time of ISODATA classification,meanwhile,compared with original ISODATA classification algorithm,the improved algorithm could mine the self-organizing relationships between different classes to improve the reliability of classification results.