Computer Engineering and Applications ›› 2019, Vol. 55 ›› Issue (17): 1-8.DOI: 10.3778/j.issn.1002-8331.1903-0437

Previous Articles     Next Articles

Survey of Compressed Sensing Reconstruction Algorithms in Deep Learning Framework

ZENG Chunyan, YE Jiaxiang, WANG Zhifeng, WU Minghu   

  1. 1.Hubei Key Laboratory for High-efficiency Utilization of Solar Energy and Operation Control of Energy Storage System, Hubei University of Technology, Wuhan 430068, China
    2.Department of Digital Media Technology, Central China Normal University, Wuhan 430079, China
  • Online:2019-09-01 Published:2019-08-30



  1. 1.湖北工业大学 太阳能高效利用及储能运行控制湖北省重点实验室,武汉 430068
    2.华中师范大学 数字媒体技术系,武汉 430079

Abstract: Compressed Sensing(CS) technology is a milestone in the field of signal processing, which samples signals far less than Nyquist frequency, and reconstructs the original signals with high probability. In recent years, the advantages of deep learning technology in feature extraction and pattern classification provide new ideas for CS. Data-driven method is adopted in deep learning-based compressed sensing reconstruction algorithm, which reduces the reconstruction time by an order of magnitude, and the reconstruction accuracy is comparable or higher. This paper focuses on the deep learning-based compressed sensing reconstruction methods, considering the traditional reconstruction methods, and divides them into three categories:prior knowledge-based, pure data-driven, mixed prior knowledge-driven and data-driven. The characteristics of typical algorithms, network structure and key steps are analyzed. Finally, three kinds of algorithms are analyzed and summarized, and the research prospects of deep learning technology applied to compressed sensing are prospected.

Key words: compressed sensing, reconstruction algorithms, deep learning, data driven

摘要: 压缩感知技术以远小于奈奎斯特频率采样信号,并高概率重建原信号,是信号处理领域里程碑式的进展。近年来深度学习在特征提取与模式分类方面的优势给压缩感知技术提供了新的思路,基于深度学习的压缩感知重建算法采用数据驱动的方式,在重建时间上有数量级的降低,且重建精度具有可比性或更高。重点综述基于深度学习的压缩感知重建方法,综合考虑传统重建方法,并分为基于先验知识、纯数据驱动、混合先验知识与数据驱动的三类,分析了典型算法的特点、网络结构、关键步骤。最后分析与总结,展望了深度学习技术应用于压缩感知的研究前景。

关键词: 压缩感知, 重建算法, 深度学习, 数据驱动