WKICA算法在多光谱遥感特征提取中的应用研究

多光谱遥感图像具有高度冗余、非线性化的特点。对其进行有效的特征提取是完成精确目标分类的前提。基于核方法强大的非线性映射能力,提出一种小波核独立成分分析WKICA算法。WKICA算法结合小波核函数近似正交,能够对复杂非线性信号进行逐步逼近的优势,有效提高了核独立成分分析KICA算法的求解精度。将其应用在多光谱遥感图像特征提取中,并对预处理条件下的图像采用不同的监督分类方法进行分类实验。实验结果表明,在WKICA算法预处理条件下,能够更有效的滤除图像噪声,相比较传统的特征提取算法,能够获取更加精确的图像分类效果。

Applied research of multispectral remote sensing image feature extraction using WKICA algorithm

The effective feature extraction of multispectral remote sensing image which is characterized by high redundancy and nonlinearity is the premise of accurate target classification. Based on the powerful nonlinear mapping ability of kernel learning, one algorithm of kernel independent component analysis based on wavelet kernel is proposed. Combining with the advantages of wavelet kernel which is approximately orthogonal and able to represent complex nonlinear signals gradually and precisely, WKICA promotes the precision of KICA effectively. WKICA is then applied to the feature extraction of multispectral remote sensing images, and implement target classification experiment to the images using different supervised classification methods. Experimental results show that, the approach of feature extraction of multispectral remote sensing image with the WKICA algorithm can separate noise more effectively, also the accuracy of image target classification is higher under the same condition.