3、示模型为x=Ψθ,所以传感矩阵A=ΦΨ;而有些文献中稀疏模型为θ=Ψx,而一般Ψ为Hermite矩阵(实矩阵时称为正交矩阵),所以Ψ-1=ΨH (实矩阵时为Ψ-1=ΨT),即x=ΨHθ,所以传感矩阵A=ΦΨH,例如沙威的OMP例程中就是如此。1、gOMP重构算法流程:文档可编辑word格式整理版2、广义正交匹配追踪(gOMP)MATLAB代码(CS_gOMP.m) 本代码完全是为了保证和前面的各算法代法格式一致,可以直接使用该实验室网站提供的代码[2]压缩包中的islsp_EstgOMP.m。[plain] viewplaincopy1.function [ theta ] = C
4、S_gOMP( y,A,K,S ) 文档可编辑word格式整理版1.%CS_gOMP Summary of this function goes here 2.%Version: 1.0 written by jbb0523 @2015-05-08 3.% Detailed explanation goes here 4.% y = Phi * x 5.% x = Psi * theta 6.% y = Phi*Psi * theta 7.% 令 A = Phi*Psi, 则y=A*theta 8.% 现在已知y和A,求theta 9.% Re
5、ference: Jian Wang, Seokbeop Kwon, Byonghyo Shim. Generalized 10.% orthogonal matching pursuit, IEEE Transactions on Signal Processing, 11.% vol. 60, no. 12, pp. 6202-6216, Dec. 2012. 12.% Available at: http://islab.snu.ac.kr/paper/tsp_gOMP.pdf 13. if nargin < 4 14. S = ro
6、und(max(K/4, 1)); 15. end 16. [y_rows,y_columns] = size(y); 17. if y_rows
7、初始化残差(residual)为y 24. for ii=1:K%迭代K次,K为稀疏度 25. product = A'*r_n;%传感矩阵A各列与残差的内积 26. [val,pos]=sort(abs(product),'descend');%降序排列 27. Sk = union(Pos_theta,pos(1:S));%选出最大的S个 28.