基于支撑集保护的回环匹配算法

针对部分压缩感知贪婪迭代类重构算法中误删正确支撑集元素的缺点，提出了一种基于支撑集保护的回环匹配算法（LM-P）。该算法依据最小残差内积初始化非受保护支撑集元素，然后依据观测向量在非受保护支撑集对应观测子矩阵上的投影，选择对应投影绝对值最大的元素添加到受保护支撑集，迭代获得受保护支撑集，从而重构原始信号。实验结果表明，对于非零值服从正态分布且稀疏度小于观测值一半数目的稀疏信号，LM-P算法的重构准确率超过86%；对于低信噪比稀疏信号，该算法的重构准确率能够维持在99%以上；与OMP、CoSaMP、SP和GPA算法相比，LM-P精确重构所需观测值数更少；此外，LM-P算法在二维图像信号的重构中也有较好性能。

Loopback matching algorithm with support set protection

There was a drawback of deleting right support elements in some greedy iterative reconstruction algorithms.To resolve this problem,loopback matching algorithm with support set protection (LM-P) was proposed.First,LM-P ini-tialized elements of non-protected support set based on minimum residual inner product.Second,it computed the projec-tions of observations on the observation sub-matrix corresponding to non-protected support set elements.Then,an ele-ment in non-protected support set with the largest projection was added to the protected support set.An alternative multi-plicative iteration method was employed to obtain the whole protected support set.As to reconstruct a sparse signal whose nonzero elements are normally distributed and the signal sparsity is less than half the number of measurements,experimental results show that the reconstruction accuracy of LM-P algorithm exceeds 86%.For sparse signals with small noise,the reconstruction accuracy of LM-P can maintain over 99 %.Compared with OMP,CoSaMP,SP and GPA algo-rithms,LM-P's observations are smaller.LM-P also has good performance for image reconstruction.greedy iteration;support set;sparse signal;LM-P