【摘要】 图像压缩感知（CS）重构方法旨在将采样过后的图像恢复为高质量图像。目前，基于深度学习的CS重构算法在重构质量及速度上性能优越，但在较低采样率时存在图像重构质量较差的问题。为此，提出一种基于多尺度注意力融合的图像CS重构网络，在网络中引入多个多尺度残差块提取图像不同尺寸的信息，并融合每个多尺度残差块的空间注意力与密集残差块的通道注意力，自适应地将局部特征与全局依赖性集成，从而提升图像重构质量。实验表明，所提算法在图像的PSNR、SSIM上均优于其他经典方法，重构性能更好。更多还原

【Abstract】 Image compressed sensing(CS) reconstruction method aims to restore the sampled image to a high-quality image. At present, CS reconstruction algorithm based on deep learning has superior performance in reconstruction quality and speed, but it has the problem of poor image reconstruction quality at low sampling rate. Therefore, an image CS reconstruction network based on multi-scale attention fusion is proposed. Multiple multi-scale residual blocks are introduced into the network to extract the information of different sizes of images, and the spatial attention of each multi-scale residual block and the channel attention of dense residual blocks are fused. The local features and global dependencies are adaptively integrated to improve the quality of image reconstruction. Experimental results show that the proposed algorithm is superior to other classical methods in PSNR and SSIM, and has better reconstruction performance.更多还原