基于非负特征值分解的极化SAR子空间分解滤波

Subspace Decomposition Filtering Based on NNED for Polarimetric SAR

极化SAR子空间分解滤波的优势在于能很好地保持极化信息,然而斑点噪声抑制效果与边缘、点目标信息的保持能力却有待提高。针对这一问题,提出了一种基于非负特征值分解(NNED)的极化SAR子空间分解滤波。对于每一个像素点,首先计算其参数向量协方差矩阵的特征值与特征向量,进而得到各个特征子空间;然后,以散射机制相似度最小化为标准,利用NNED选取分离信号子空间与噪声子空间的最优阈值;最后根据信号子空间得到滤波后的结果。实测极化SAR实验表明,相比于同类算法,所提出的算法能有效地抑制斑点噪声并且能很好地保持边缘、点目标信息。

Although subspace decomposition filtering for polarimetric SAR can keep the polarimetric information very well, it is necessary to enhance its despeckling performance and capability of keeping edge-point information. For the problem, a subspace decomposition filtering based on nonnegative eigenvalue decomposition（NNED） was proposed. For each pixel,firstly we calculated eigenvalues and eigenvectors of the parameter vector covariance matrix, and then each eigen-subspace was obtained, secondly we used the NNED to select the optimal threshold for separating signal subspace from noise subspace, and the selecting criterion is the minimal similarity measurement of scattering mechanisms, thirdly the filtered result was produced by the signal subspace. The real-POLSAR-data experiment shows that compared with other congener algorithms, the proposed algorithm can efficiently suppress speckle noise and keep the edge-point infor- mation very well.