机电工程技术2025，Vol.54Issue(6)：23-27,50,6.DOI:10.3969/j.issn.1009-9492.2025.06.005
基于深度展开ISTA网络的动态路径选择的压缩感知图像恢复
Dynamic Path Selection for Compressive Sensing Image Recovery Based on Deep Unfolding ISTA Network
任重伟 1张雨晨2
作者信息
- 1. 西藏大学信息与科技学院,拉萨 850000
- 2. 西藏大学,拉萨 850000
- 折叠
摘要
Abstract
Due to the challenges of low computational efficiency and excessive resource consumption faced by traditional compressive sensing image recovery techniques when processing large-scale data,a novel approach based on deep learning frameworks is studied.This approach,known as Deep Unfolding Networks(DUNs),transforms the iterative process of traditional optimization algorithms into hierarchical structures of neural networks,thereby achieving efficient and accurate image reconstruction.However,recognizing the varying demands for computational resources across different stages of image recovery,a Dynamic Path Selection Network(DPCS-DUNs)method is proposed.A dynamic gradient descent control unit module and a dynamic proximal mapping path selection module are included in this method to adapt to diverse image characteristics.The effectiveness and superiority of the network are validated using the set11 dataset.Experimental results demonstrate that under various compression ratios,the PSNR and SSMI values obtained from this network outperform the recovery performance metrics of other networks at present.By adjusting the trade-off between complexity and efficiency,the network achieves a lightweight design,thus realizing computational lightness.These findings confirm the feasibility and high efficiency of the proposed network.关键词
深度展开网络/动态路径选择/压缩感知/图像恢复Key words
deep unfolding network/dynamic path selection/compressive sensing/image recovery分类
计算机与自动化引用本文复制引用
任重伟,张雨晨..基于深度展开ISTA网络的动态路径选择的压缩感知图像恢复[J].机电工程技术,2025,54(6):23-27,50,6.
