Abstract:

In compressive sensing reconstruction algorithms,the multi-path matching pursuit algorithm improves the possibility of obtaining the global optimal solution by searching multiple paths,but a large number of redundant paths will cause a serious drop in performance.To solve this problem,a multi-path matching pursuit reconstruction algorithm based on adaptive acceleration is proposed.First,the number of generated child branches is controlled by setting the threshold,optimizing the strategy of the original algorithm in allocating the number of paths evenly,so that the parent branches with a strong coherence traverse more child branches and atoms with a low coherence are restricted from being assigned to new paths.Second,by using the reconstruction residuals generated by the first path,a new pruning criterion is designed to perform secondary screening on candidate paths,thus reducing computational expenses.Finally,under an ideal state,the proposed algorithm derives the restricted isometry property condition to accurately reconstruct the signal,and presents the signal-to-noise ratio limit for the accurate reconstruction of the signal in the presence of noise interference.Simulation results show that in the reconstruction experiments for one-dimensional and two-dimensional signals,the proposed algorithm effectively improves the reconstruction efficiency compared to the multi-path matching pursuit algorithm,while ensuring a high reconstruction accuracy.

Key words: compressed sensing, multi-path matching pursuit, coherence, restricted isometry property