基于多测量动态聚类的压缩感知增强成像方法

Enhanced Compressive Imaging Approach Based on Multi-measurement and Dynamic Clustering

噪声环境下的稀疏信号重构可以转换为带约束的二次规划问题,通过正则化算法可以有效求解,而正则化参数l却是影响重构质量的重要因素。广义交叉验证(Generalized Cross-Validation,GCV)算法是噪声未知条件下估计l的有效算法,但当信噪比较低时却无法保证l收敛于全局最优,导致重构图像信杂比降低,甚至造成目标丢失。为实现低信噪比环境下稀疏信号的稳健重构,该文提出基于多测量动态聚类(Multi-Measurement Dynamic Clustering,MMDC)的压缩感知(Compressive Sensing,CS)增强成像方法。新方法首先对初始观测数据进行多次随机抽取,然后通过CS处理获得重构图像序列,最后利用动态聚类算法实现对原信号的稳健重构,在改善重构图像质量的同时也有效地抑制了杂波。另外,鉴于GCV计算量大及MMDC对l估计误差的不敏感,该文提出基于简化GCV算法的MMDC增强成像方法,仿真及实测数据的处理结果均验证了所提方法的有效性。

In noisy environments, signal reconstruction can be converted into the issue of bound constrained quadratic programming which can be resolved by the regularization programming algorithm, but the reconstruction quality depends heavily on the regularization parameter ？. Without any apriori knowledge of noise, the Generalized Cross-Validation （GCV） algorithm provides a suitable way for ？ estimation. But in low Signal-to-Noise Ratio （SNR） conditions, it is difficult for GCV to guarantee ？ perfect convergence at the global optimum, which results in the Signal-to-Clutter Ratio （SCR） of the reconstructed image declining and targets missing. For robust reconstruction in low SNR conditions, the enhanced compressive imaging approach based on Multi-Measurement and Dynamic Clustering （MMDC） is proposed in this paper. First, it extracts randomly the original measured data by multiple times. Second, it receives the image series by CS processing. Finally, it im-plements robust reconstruction by clustering the image series with DC algorithm. Both the simulated and ex-perimental results indicate that MMDC not only improves the reconstruction quality, but also receives effective clutter suppression. Due to the heavy computation of GCV and the insensitivity of MMDC to ？ estimation error, the MMDC based on a simplified GCV algorithm is also proposed in this paper.