一种基于变步长的稀疏度自适应匹配追踪算法  

Improvement of Sparsity Adaptive Matching Pursuit Based on Variable Iteration Steps

作　　者：毕学霞[1] 尚振宏[1] 强振平[2] 刘辉[1] 

机构地区：[1]昆明理工大学信息工程与自动化学院,昆明650500 [2]西南林业大学计算机与信息学院,昆明650224

出　　处：《系统仿真学报》2014年第9期2116-2120,2125,共6页Journal of System Simulation

基　　金：国家自然科学基金项目(71161015);昆明市科技局项目(08S100310);中国科学院太阳活动重点实验室项目(KLSA201310)

摘　　要：压缩感知(Compressed Sensing,CS)理论突破了传统的采样定率,对于稀疏的或可压缩的信号,可同时进行采样和压缩。其中重构算法是压缩感知理论的研究热点之一,对采样过程的正确性验证有着重要意义。稀疏度自适应匹配追踪算法(Sparsity Adaptive Matching Pursuit,SAMP)在迭代过程中采用固定步长,容易导致过估计和欠估计。为解决该问题,根据相邻信号能量差的变化规律,在迭代过程中采用对数型的"变步长",即迭代的初始阶段步长增长速度较快,当能量差达到一定阈值时,则步长增长速度较慢,并设置双阈值严格控制逼近的精确度。实验表明,改进后算法提高了重构质量,尤其是在当采样率较低时,仍有较好的重构效果。Compressive Sensing （CS） theory hasbroken the limit of the sampling theorem, whichsamples and compresses at the same time if the signal iscompressive or sparse. In CS, the signal reconstruction is significant to verify the accuracy of sampling.Sparsity Adaptive Matching Pursuit （SAMP）a doptsfixed- step in the iteration causing under-estimated and over-estimated,therefore,according to the variation of the energy difference between adjacent signal, logarithmic ＂variable steps＂ during the iteration was used in the improved algorithm, namely, the step size increased quickly at the initial stage, and then slowed down when energy difference reached a certain threshold. At the same time, dual-threshold was set for controlling reconstruction accuracy.Experiments show that the improved algorithm improves the quality of reconstruction, especially when sampling ratio is low.

关 键 词：压缩感知 重构算法 稀疏度自适应 匹配追踪 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]