基于双路信息互补增强的压缩感知深度重构
Deep reconstruction for compressed sensing with dual-path information complementary enhancement
投稿时间:2020-08-07  修订日期:2020-08-07
DOI:
中文关键词: 压缩感知重构  深度学习  双路网络  信息互补增强
英文关键词: Compressed sensing reconstruction  deep learning  dual-path network  information complementary enhancement
基金项目:国家自然科学基金资助项目(61471400),中南民族大学中央高校基本科研业务费专项CZY19016
作者单位E-mail
熊承义 中南民族大学 电子信息工程学院 xiongcy@mail.scuec.edu.cn 
秦鹏飞 中南民族大学 电子信息工程学院  
高志荣 中南民族大学 计算机科学学院  
施晓迪 中南民族大学 电子信息工程学院  
刘川鄂 中南民族大学 电子信息工程学院  
摘要点击次数: 170
全文下载次数: 0
中文摘要:
      基于深度网络的方法在压缩感知重构中表现了良好潜能,而如何提升网络的深度特征表示能力是其中的一个关键问题。基于多通道网络能较好改善深度特征表示能力的发现,本文设计了一种双路信息互补增强的压缩感知深度重构网络,以更好地提升网络的重构性能。具体地,重构网络由两个并行通道组成,一个通道实现对图像原始分辨率的重构,而另一个通道实现对图像的降采样重构;两个通道的中间特征提取部分采用交叉融合技术,以实现信息提取的互补增强。此外,本重构网络也采用了测量矩阵的联合学习技术,以同时改进测量效率和重构效果。基于公开发布的标准测试集的大量实验验证了,本文方法在提升压缩感知图像的深度重构质量的有效性。
英文摘要:
      The deep network-based method has shown good potential in compressed sensing reconstruction, while how to improve the deep feature representation ability of the network is one of the key issues. Based on the finding that multi-channel networks can better improve the ability of deep feature representation, this paper proposes a two-channel information complementary enhancement based compressed sensing deep reconstruction network to better improve the reconstruction performance. Specifically, the proposed reconstruction network consists of two parallel channels, one of which realizes the reconstruction of the original resolution of the image, and the other one realizes the down-sampling reconstruction. The intermediate features extracted in the two channels are enhanced by using information fusion with attention mechanism. In addition, the measurement matrix is jointly learned to improve measurement efficiency and reconstruction effectiveness. Experiments based on the publicly standard test set validate the proposed method in term of in improving reconstructed image quality of deep networks.
View Fulltext   查看/发表评论  下载PDF阅读器
关闭