独立分量分析在多光谱遥感图像分类中的应用

多光谱遥感图像反映了不同地物的光谱特征,其分类是遥感应用的基础。但是在多光谱遥感波段图像中存在不同地物对应着相同的灰度,即异物同谱的问题。独立分量分析算法对未知的源信号的混合信号进行估计,可以获得相互独立的源信号的近似。独立分量分析算法利用了信号的高阶统计信息,对于多光谱遥感图像而言,算法去除了波段图像之间的相关性,获得的波段图像是相互独立的。但是独立分量分析算法有一个缺点,即计算量太大,影响了在多光谱遥感图像分类上的应用。文章对独立分量分析的一种快速算法FastICA进行改进,减少了计算量,提高了算法的有效性。在性能相当的情况下,改进FastICA算法能有效地减少算法的计算量。由于FastICA算法是线性ICA算法,对于非线性混合的光谱信号的估计存在一定误差,因此应用BP神经网络的非线性特性对其进行自动分类。在同原始遥感图像的BP神经网络分类结果进行比较,结果表明独立分量分析算法能提高多光谱遥感图像的分类的正确率。

Remote Sensing Images Classification Based on Independent Component Analysis

The multi-spectral remote sensing images reflect the spectral features of diverse surface features,and the their classification is the base of remote sensing applications.The classification faces the problem that differs surface features have the same spectral features.Independent Component Analysis(ICA)algorithm can estimate the indepen-dent source signals that are mixed by unknown mode,and the source signals are unknown,too.The ICA algorithm uses the high-order information of signals;to multi-spectral remote sensing images,ICA algorithm not only removes the corre-lation of images,but also obtains the new band images that are mutual independent.But the computational complexity of ICA is too big,and it influences the application of ICA in remote sensing field.Modified FstICA algorithm is approached based FastICA algorithm,a fast algorithm of ICA.By modifying on iterate approach of estimating separation matrix of ICA,Modified FastICA algorithm reduces the computational complexity and improves the effectiveness of ICA.With the correspondent separation performance,Modified FastICA algorithm can effectively decrease the amount of computation.Because FastICA algorithm is a linear ICA algorithm,and it has some mistake that the estimation of spectral signals which mixed non-linearly,BP Neural Network is used to automatic classify for non-linearly characteristic.Compared to classification result of source remote sensing images,the ICA pretreatment can improve the classification accuracy.