利用ICA算法进行全极化SAR影像滤波研究

Applications of ICA for Filtering of Fully Polarimetric SAR Imagery

提出了一种基于ICA算法的全极化SAR的滤波方法。首先对全极化SAR的四通道dB强度影像进行ICA变换,分离出信号及噪声分量,将噪声设为极小值,然后利用混合矩阵混合得到滤除噪声后的强度影像,结合原始相位信息,计算滤波后的散射矩阵。采用Foulum地物的EMISAR数据进行试验,并采用相干斑指数、均方差指数、边缘保持系数以及极化相关系数进行评价。结果表明,该方法不但能有效滤除影像的噪声,在边缘等细节信息保持上也具有较大优势。此外,滤波后影像虽散射特性发生改变,但维持了地物间的差异,在基于统计特性的精细分类、边缘信息及小目标检测方面仍具有较好的应用潜力。

A new class of filters for fully PolSAR image using ICA algorithm is proposed in this paper. This method, first transforms the four channel dB level intensity images by ICA,and gets four independent components, then sets noise components to the minimum, third utilizes the mixed matrix to calculate the four intensity image without noise, finally uses the original phase to work out the scattering matrix haven been filtered speckle. For verifying the validity of the method, the EMISAR data of Foulum is tested. Experiment results demonstrate that the method proposed can effectively filter speckle, and also has many advantages in preserving detail information. Even though polarization signatures of the scattering matrix after filtering have changed, differences and distinctive of objects are kept. Therefore, the proposed filter has large potential application in the fields of classification, edge and small target detection based on the statistical characteristics.