Class
Article
Graduation Year
2017
College
College of Engineering
Department
Electrical and Computer Engineering Department
Faculty Mentor
Todd K. Moon
Presentation Type
Poster Presentation
Abstract
Compressive sensing (CS) is one of the evolving areas in signal acquisition and reconstruction with many applications including the study of brain activities, recovery of multi-band signals, separating the foreground and background components from the collection of noisy frames of a video recording, reconstruction of hand-written digits, taking images using one-pixel camera, and so forth. It is a promising technique in processing compressible or sparse signals by requiring far few samples than the well-known Nyquist rate. Sparse signals have very few non-zero elements. In CS the goal is to efficiently measure and then reconstruct the signal under the assumption that the underlying signal of interest is sparse but the number and location of the non-zeros are unknown. Here, we provide some of our recently proposed algorithms in this area using Bayesian approach. Bayesian learning models are powerful and flexible to incorporate the prior knowledge on the characteristics of the underlying signals. We evaluate the performance of our proposed algorithms compared to other existing algorithms in terms of the detection and false-alarm rate via receiver operating curves (ROC) on the synthetically generated data. We also illustrate the performance based on some real-world data.
Location
North Atrium
Start Date
4-13-2017 1:30 PM
End Date
4-13-2017 2:45 PM
New Bayesian Compressive Sensing Algorithms for Sparse Signal Recovery
North Atrium
Compressive sensing (CS) is one of the evolving areas in signal acquisition and reconstruction with many applications including the study of brain activities, recovery of multi-band signals, separating the foreground and background components from the collection of noisy frames of a video recording, reconstruction of hand-written digits, taking images using one-pixel camera, and so forth. It is a promising technique in processing compressible or sparse signals by requiring far few samples than the well-known Nyquist rate. Sparse signals have very few non-zero elements. In CS the goal is to efficiently measure and then reconstruct the signal under the assumption that the underlying signal of interest is sparse but the number and location of the non-zeros are unknown. Here, we provide some of our recently proposed algorithms in this area using Bayesian approach. Bayesian learning models are powerful and flexible to incorporate the prior knowledge on the characteristics of the underlying signals. We evaluate the performance of our proposed algorithms compared to other existing algorithms in terms of the detection and false-alarm rate via receiver operating curves (ROC) on the synthetically generated data. We also illustrate the performance based on some real-world data.