基于支持向量机和子空间划分的波段选择方法

高光谱图像具有较高的谱分辨力,从而能够更精确地描述地物目标特性。然而,其较大的数据量和较高的数据维给分析和处理带来很大的困难。高光谱图像间存在着大量的冗余信息,波段选择能够有效地去除冗余信息从而减少计算量。针对一类波段选择方法所选取的波段易于集中而造成信息冗余和信息损失的缺陷,提出一种基于支持向量机和子空间划分的波段选择方法。首先对支持向量机判决函数进行敏感度分析和对数据源进行子空间划分,然后结合敏感度分析结果和子空间划分结果来实现有效的波段选择。实验证明了这种方法的有效性。

Band selection method based on combination of support vector machines and subspatial partition

Hyperspectral image can describe ground objects' feature more accurately with its high spectral resolution power, but its great data size and high data dimension are very adverse to analysis and processing. There is great redundancy information in hyperspectral image and band selection can effectively remove it and reduce computational cost accordingly. Bands selected by some band selection methods tend to over-concentrate, which will lead to information redundancy and information loss. To overcome this disadvantage, a band selection method based on combination of support vector machines (SVMs) and subspatial partition is put forward. Band selection can be implemented on the basis of the sensitivity analysis of SVM discrimination function and the subspatial partition of data. The validation of this method is supported by experiments.