Status and Trends in Deep-learning-assisted Compressed Sensing Reconstruction and Its Applications
The classic Nyquist sampling theorem requires a sampling frequency greater than twice the maximum signal frequency for reconstructing the original signal.Compressed sensing method can break through this limitation when coping with signals with sparsity,significantly reducing sampling frequency and improving efficiency of data collection and analy-sis,and therefore in recent years has been regarded as a new sparse signal processing framework and found wide applica-tions in fields such as image reconstruction and sparse signal reconstruction.Moreover,rapid development of deep learning technologies enables the combination of compressed sensing with various types of neural networks,thus achieving both improved reconstruction accuracy and reduced reconstruction time consumption.This review conducts extensive literature survey,elaborates basic technological concepts,and summarizes prevailing compressed sensing algorithms and primary applicative aspects of deep compressed sensing,with respect to deep-learning-assisted compressed learning.It also critical-ly discussed current problems and future development trends.
deep learningcompressed sensingimage reconstructionsignal reconstructiontransfer learning
万方数据
维普
