FISTA-Net: Learning a Fast Iterative Shrinkage Thresholding Network for Inverse Problems in Imaging

Inverse problems are essential to imaging applications. In this letter, we propose a model-based deep learning network, named FISTA-Net, by combining the merits of interpretability and generality Fast Iterative Shrinkage/Thresholding Algorithm (FISTA) strong regularization tuning-free advantages data-driven neural network. By unfolding FISTA into architecture FISTA-Net consists multiple gradient descent, proximal mapping, momentum modules in cascade. Different from FISTA, matrix can be updated during iteration operator network is developed for nonlinear thresholding which learned through end-to-end training. Key parameters including step size, value scalar training data rather than hand-crafted. We further impose positive monotonous constraints on these ensure they converge properly. The experimental results, evaluated both visually quantitatively, show that optimize different tasks, i.e. Electromagnetic Tomography (EMT) X-ray Computational (X-ray CT). It outperforms state-of-the-art methods exhibits good generalization ability over other competitive learning-based approaches under noise levels.