面向可解释压缩感知的算法展开综述  

Survey on algorithm unrolling for interpretable compressed sensing

作　　者：曾春艳[1] 余琰 王志锋[2] 夏诗言 ZENG Chunyan;YU Yan;WANG Zhifeng;XIA Shiyan(School of Electrical and Electronic Engineering,Hubei University of Technology,Wuhan 430068,China;Department of Artificial Intelligence Education,Central China Normal University,Wuhan 430079,China)

机构地区：[1]湖北工业大学电气与电子工程学院,湖北武汉430068 [2]华中师范大学人工智能教育学部,湖北武汉430079

出　　处：《华中科技大学学报（自然科学版）》2022年第11期35-43,共9页Journal of Huazhong University of Science and Technology(Natural Science Edition)

基　　金：国家自然科学基金资助项目(61901165,62177022,61501199);信息化与基础教育均衡发展省部共建协同创新中心研究资助项目(xtzd2021-005);中央高校基本科研业务费专项资金资助项目(CCNU20QN013)。

摘　　要：针对压缩感知可解释性的问题,首先回顾了基于深度学习的压缩感知方法发展历程及存在问题,阐述了算法展开方法的研究背景和重要意义,并对主流的算法展开网络及特点进行分析总结;然后根据传统迭代算法,将算法展开网络进行分类,并选择三个具有代表性的算法,分别为迭代收缩阈值算法、近似消息传递算法和交替方向乘子法,概述其相应的原理,分析这三种算法的网络展开方式、网络特点和实验仿真效果;最后从具体展开方式的设计、展开网络的理论分析和展开网络在资源受限平台下的实施这三方面,探讨了可解释性压缩感知领域中算法展开技术存在的问题和下一步研究的方向.To address the issue of compressed sensing interpretability,firstly the development history and existing problems of compressed sensing methods based on deep learning were reviewed,the research background and importance of algorithm unrolling methods were described,and the mainstream algorithm unrolling networks and their characteristics analysis were summarized.Then,according to the traditional iterative algorithms,the algorithm unrolling network was classified,and three representative algorithms were selected to outline their corresponding principles,which were iterative shrinkage and threshold algorithm,approximate message passing algorithm and alternating direction method of multipliers,and the network unrolling methods,network characteristics and experimental simulation effects of these three algorithms were analyzed.Finally,the existing problems and next research directions of algorithm unrolling techniques in the field of interpretable compressed sensing were discussed from three aspects:the design of specific unrolling methods,the theoretical analysis of unrolling networks,and the implementation of unrolling networks under resource-constrained platforms.

关 键 词：算法展开 压缩感知 信号重建 可解释性 深度学习 

分 类 号：TN911.73[电子电信—通信与信息系统]