Image Compression Sensing Depth Reconstruction Based on Inverted Residual Blocks
With the combination of deep learning and compressed sensing theory,more stacked and deeper convolu-tional network architectures have been proposed,but the existing networks generally use the traditional residual structure to solve the problem of network performance degradation,and the introduced attention mechanism cannot effectively perceive the global information of images.Therefore,this paper introduces a reciprocal residual structure and an efficient multi-scale attention mechanism to capture more context information and thus better reconstruct images.Through a large number of comparative experiments,it can be seen that the network framework proposed in this paper achieves better performance indicators and visual effects compared with the existing deep learning compressed sensing reconstruction algorithms.
万方数据
维普
