Related Experiment Video
Updated: Jun 3, 2025
04:48
Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
Published on: July 5, 2024
367
Adaptive Memory-Augmented Unfolding Network for Compressed Sensing.
Mingkun Feng1, Dongcan Ning1, Shengying Yang1
1School of Information and Electronic Engineering, Zhejiang University of Science and Technology, Hangzhou 310023, China.
Sensors (Basel, Switzerland)
|January 8, 2025
Summary
This study introduces an adaptive memory-augmented unfolding network for compressed sensing (AMAUN-CS). The novel approach enhances feature capture and information dependency, outperforming existing methods with lower training complexity.
Area of Science:
- Signal Processing
- Machine Learning
- Computer Vision
Background:
- Deep unfolding networks (DUNs) are popular for compressed sensing (CS) due to interpretability and performance.
- Existing DUNs often suffer from high parameter counts and feature information loss during iterations.
Purpose of the Study:
- To propose a novel adaptive memory-augmented unfolding network for compressed sensing (AMAUN-CS).
- To address limitations of current DUNs, specifically parameter count and feature loss.
Main Methods:
- Integration of an adaptive content-aware strategy into the proximal gradient descent (PGD) algorithm.
- Extension of AMAUN-CS to AMAUN-CS+ incorporating a memory storage mechanism for cross-stage information dependency.
Main Results:
- AMAUN-CS adaptively captures adequate features without losing interpretability.
- AMAUN-CS+ effectively develops deep information dependency across cascading stages.
- AMAUN-CS model surpasses advanced methods on benchmark datasets.
Conclusions:
- AMAUN-CS offers improved performance in compressed sensing.
- The proposed network achieves superior results with lower training complexity compared to existing methods.
