计算机应用研究2024，Vol.41Issue(11)：3509-3514,6.DOI:10.19734/j.issn.1001-3695.2023.12.0635
双U型门控网络融合非局部先验的图像压缩感知重建方法
Dual U-shaped network with non-local prior fusion for image compressed sensing
摘要
Abstract
Aiming at the problem that insufficient detail processing ability and inadequate use of measurement values in cur-rent non-iterative network-based image compressed sensing methods,this paper proposed a novel dual U-shaped gated network(DUGN)for image compressed sensing.This network improved upon the existing U-shaped network structure and enhanced the learning ability in compressed sensing tasks.This paper proposed a measurements non-local fusion(MNLF)module based on cross-attention mechanism,MNLF integrated non-local prior information from measurements into the deep network and guided the network reconstruction.Furthermore,this paper proposed a window gated network(WGN)module as the basic blocks to enable the network to learn larger ranges of image information and enhance the ability to process details.Experimen-tal results demonstrate that compared with existing compressed sensing reconstruction methods,DUGN has higher PSNR and SSIM on Set11 and better performance in the authenticity of image reconstruction.关键词
图像压缩感知重建/非局部先验/U型网络/门控网络Key words
image compressed sensing/non-local prior/U-shaped network/gated network分类
信息技术与安全科学引用本文复制引用
林乐平,胡尚鋆,欧阳宁..双U型门控网络融合非局部先验的图像压缩感知重建方法[J].计算机应用研究,2024,41(11):3509-3514,6.基金项目
国家自然科学基金资助项目(62001133,61661017,61362021) （62001133,61661017,61362021）
广西科技基地和人才专项资助项目(桂科AD19110060) （桂科AD19110060）
广西高等学校千名中青年骨干教师培育计划资助项目 （）
广西自然科学基金资助项目(2017GXNSFBA198212) （2017GXNSFBA198212）
广西无线宽带通信与信号处理重点实验室基金资助项目(GXKL06200114) （GXKL06200114）
