The traditional Sparse Recovery (SR) algorithms are unsuitable for signal reconstruction of Multiple Measurement Vectors (MMV) for the following two reasons

one is the high computing burden

and the other is that the presented algorithms are not used to the case when MMV are arbitrary sparse structure.To solve the problems

a novel fast sparse recovery algorithm is proposed.Firstly

the Matrix Smoothed L0-norm (MSL0) algorithm is adopted to reconstruct the MMV of arbitrary sparse structure and estimate the initial support.Secondly

using the relationship between the sparse level and measurement number

the pre-selection support is obtained from choosing the initial support.Thirdly

the final support is gotten with Bayesian Group Testing (BGT) method.And finally

the MMV is reconstructed precisely via the final support.The proposed algorithm makes full use of high efficiency of the MSL0 and redundancy support elimination ability of the BGT.The algorithm can not only reconstruct MMV of arbitrary sparse structure more efficiently

but also has higher reconstructed accuracy and better robustness.ISAR imaging experiments based on real data show the validity of the proposed algorithm.