Robust Deep Compressive Sensing With Recurrent-Residual Structural Constraints

Existing deep compressive sensing (CS) methods either ignore adaptive online optimization or depend on costly iterative optimizer during reconstruction. This work explores a novel image CS framework with recurrent-residual structural constraint, termed as $\mathrm{R}^{2}$CS-NET. The notation="LaTeX">$\mathrm{R}^{2}$CS-NET first progressively optimizes the acquired samplings through recurrent neural network. cascaded residual convolutional network then fully reconstructs from optimized latent representation. As efficiently bridging optimization, integrates robustness of efficiency and nonlinear capacity learning methods. Signal correlation has been addressed architecture. nature further makes it an ideal candidate for color via leveraging channel correlation. Numerical experiments verify proposed design not only fulfills adaptation motivation, but also outperforms classic long short-term memory (LSTM) architecture in same scenario. overall demonstrates hardware implementation feasibility, leading generalization capability among existing benchmarks.