上海科技大学知识管理系统

We explore the application of $\ell_1/\ell_q^p$ regularization as an alternative to conventional $\ell_1/\ell_p$ techniques in sparse signal recovery. Our findings reveal that the $\ell_1/\ell_q^p$ approach provides robust sparsity-inducing property. This study introduces an efficient Difference of Convex (DC) algorithm tailored for this non-convex and non-smooth model. Analyzing the KL property and using the Restricted Isometry Property (RIP) as a theoretical benchmark for our regularization, we demonstrate its efficacy in sparse signal reconstruction across varying recovery scenarios, outperforming classical approaches in both accuracy and adaptability.