结合循环和卷积结构:用于稀疏信号恢复的深度压缩感知网络

Zhongjun Liu, Jun Zhang, Luhua Wang
{"title":"结合循环和卷积结构:用于稀疏信号恢复的深度压缩感知网络","authors":"Zhongjun Liu, Jun Zhang, Luhua Wang","doi":"10.1145/3579654.3579718","DOIUrl":null,"url":null,"abstract":"With the powerful learning ability of neural networks, compressive sensing methods based on deep neural networks have shown great potential. However, most existing methods treat the neural network as a black box, which weakens the regularization of \"known\" signal priors. In order to solve this problem, a class of compressed sensing methods based on deep unfolding has been proposed in recent years. By mapping the iterative optimization algorithm in the classical model-based methods onto networks, the method based on deep unfolding integrates the advantages of both, making the network interpretable while greatly reducing the time complexity. In this paper, to solve the problem of sequential sparse reconstruction, we propose a novel model based on deep unfolding, dubbed DW-FISTA. The whole model can be divided into two modules, namely the mapping module and the refine module. The first module maps the iterative process of the fast iterative shrinkage-thresholding algorithm (FISTA) to an interpretable recurrent neural network composed of fixed phases. Compared with other ISTA models based on deep unfolding, our model has a better global convergence rate. In the second module, we use convolution filters and nonlinear activation functions to refine the reconstructed signal by taking the immediate reconstruction results of the mapping module as input. Experimental result shows that the proposed DW-FISTA model outperforms existing state-of-art models in sparse sequence reconstruction and can ensure a faster convergence rate.","PeriodicalId":146783,"journal":{"name":"Proceedings of the 2022 5th International Conference on Algorithms, Computing and Artificial Intelligence","volume":"54 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"2022-12-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"Combining Recurrent and Convolution structures: A Deep Compressed Sensing Network for Sparse Signal Recovery\",\"authors\":\"Zhongjun Liu, Jun Zhang, Luhua Wang\",\"doi\":\"10.1145/3579654.3579718\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"With the powerful learning ability of neural networks, compressive sensing methods based on deep neural networks have shown great potential. However, most existing methods treat the neural network as a black box, which weakens the regularization of \\\"known\\\" signal priors. In order to solve this problem, a class of compressed sensing methods based on deep unfolding has been proposed in recent years. By mapping the iterative optimization algorithm in the classical model-based methods onto networks, the method based on deep unfolding integrates the advantages of both, making the network interpretable while greatly reducing the time complexity. In this paper, to solve the problem of sequential sparse reconstruction, we propose a novel model based on deep unfolding, dubbed DW-FISTA. The whole model can be divided into two modules, namely the mapping module and the refine module. The first module maps the iterative process of the fast iterative shrinkage-thresholding algorithm (FISTA) to an interpretable recurrent neural network composed of fixed phases. Compared with other ISTA models based on deep unfolding, our model has a better global convergence rate. In the second module, we use convolution filters and nonlinear activation functions to refine the reconstructed signal by taking the immediate reconstruction results of the mapping module as input. Experimental result shows that the proposed DW-FISTA model outperforms existing state-of-art models in sparse sequence reconstruction and can ensure a faster convergence rate.\",\"PeriodicalId\":146783,\"journal\":{\"name\":\"Proceedings of the 2022 5th International Conference on Algorithms, Computing and Artificial Intelligence\",\"volume\":\"54 1\",\"pages\":\"0\"},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2022-12-23\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"Proceedings of the 2022 5th International Conference on Algorithms, Computing and Artificial Intelligence\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/10.1145/3579654.3579718\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"Proceedings of the 2022 5th International Conference on Algorithms, Computing and Artificial Intelligence","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1145/3579654.3579718","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 0

摘要

由于神经网络强大的学习能力,基于深度神经网络的压缩感知方法显示出巨大的潜力。然而,大多数现有方法将神经网络视为一个黑箱,这削弱了“已知”信号先验的正则化。为了解决这一问题,近年来提出了一类基于深度展开的压缩感知方法。基于深度展开的方法将经典的基于模型方法中的迭代优化算法映射到网络中,结合了两者的优点,使网络具有可解释性,同时大大降低了时间复杂度。为了解决序列稀疏重建问题,本文提出了一种基于深度展开的新模型DW-FISTA。整个模型可分为两个模块,即映射模块和细化模块。第一个模块将快速迭代收缩阈值算法(FISTA)的迭代过程映射为固定相位组成的可解释递归神经网络。与其他基于深度展开的ISTA模型相比,我们的模型具有更好的全局收敛速度。在第二个模块中,我们将映射模块的即时重构结果作为输入,使用卷积滤波器和非线性激活函数对重构信号进行细化。实验结果表明,所提出的DW-FISTA模型在稀疏序列重建方面优于现有的先进模型,并能保证更快的收敛速度。
本文章由计算机程序翻译,如有差异,请以英文原文为准。
Combining Recurrent and Convolution structures: A Deep Compressed Sensing Network for Sparse Signal Recovery
With the powerful learning ability of neural networks, compressive sensing methods based on deep neural networks have shown great potential. However, most existing methods treat the neural network as a black box, which weakens the regularization of "known" signal priors. In order to solve this problem, a class of compressed sensing methods based on deep unfolding has been proposed in recent years. By mapping the iterative optimization algorithm in the classical model-based methods onto networks, the method based on deep unfolding integrates the advantages of both, making the network interpretable while greatly reducing the time complexity. In this paper, to solve the problem of sequential sparse reconstruction, we propose a novel model based on deep unfolding, dubbed DW-FISTA. The whole model can be divided into two modules, namely the mapping module and the refine module. The first module maps the iterative process of the fast iterative shrinkage-thresholding algorithm (FISTA) to an interpretable recurrent neural network composed of fixed phases. Compared with other ISTA models based on deep unfolding, our model has a better global convergence rate. In the second module, we use convolution filters and nonlinear activation functions to refine the reconstructed signal by taking the immediate reconstruction results of the mapping module as input. Experimental result shows that the proposed DW-FISTA model outperforms existing state-of-art models in sparse sequence reconstruction and can ensure a faster convergence rate.
求助全文
通过发布文献求助,成功后即可免费获取论文全文。 去求助
来源期刊
自引率
0.00%
发文量
0
×
引用
GB/T 7714-2015
复制
MLA
复制
APA
复制
导出至
BibTeX EndNote RefMan NoteFirst NoteExpress
×
提示
您的信息不完整,为了账户安全,请先补充。
现在去补充
×
提示
您因"违规操作"
具体请查看互助需知
我知道了
×
提示
确定
请完成安全验证×
copy
已复制链接
快去分享给好友吧!
我知道了
右上角分享
点击右上角分享
0
联系我们:info@booksci.cn Book学术提供免费学术资源搜索服务,方便国内外学者检索中英文文献。致力于提供最便捷和优质的服务体验。 Copyright © 2023 布克学术 All rights reserved.
京ICP备2023020795号-1
ghs 京公网安备 11010802042870号
Book学术文献互助
Book学术文献互助群
群 号:604180095
Book学术官方微信