Speaker: Yifei Lou, University of North Carolina at Chapel Hill
Title: "Hyperspectral Unmixing via Graph Regularizations, Active Learning, and Spectral Bundles"
Abstract:
Hyperspectral unmixing (HSU) is a powerful technique for determining the material composition of each pixel in a hyperspectral image, which typically contains hundreds of spectral channels. In this talk, I will present three approaches for identifying the pure spectra of individual materials (endmembers) and estimating their proportions (abundances) at each pixel. The first approach is based on graph total variation (gTV) regularization and uses the alternating direction method of multipliers (ADMM) together with the Merriman–Bence–Osher (MBO) scheme to obtain a model solution. The second approach incorporates active learning to strategically select training pixels, achieving substantial performance gains with minimal supervision. Finally, I will discuss a group‑sparsity framework in which each material is represented by a set of spectral signatures, known as endmember bundles, with each group corresponding to a specific material. Extensive experiments on both synthetic and real hyperspectral datasets demonstrate the effectiveness and advantages of these methods.
Bio:
Yifei Lou holds a joint position in the department of mathematics in the College of Arts & Sciences and the School of Data Science and Society. She served as a faculty member in the mathematical sciences department at the University of Texas at Dallas from 2014 to 2023, first as an assistant professor and then as an associate professor. She received her Ph.D. in applied math from the University of California, Los Angeles (UCLA) in 2010. After graduation, she was a postdoctoral fellow at the School of Electrical and Computer Engineering at the Georgia Institute of Technology, followed by another postdoctoral training at the department of mathematics, University of California, Irvine from 2012-2014. Lou received the National Science Foundation CAREER Award in 2019.
Her research lies in the intersection of computational mathematics and data sciences. Specifically, she focuses on signal/image recovery from a limited number of measurements, where “limited” refers to the fact that the amount of data that can be taken or transmitted is restricted by technical or economic constraints. In this scenario, additional information and reasonable assumptions are often required to recover useful information from the insufficient amount of data. Over the years, she has developed efficient computational tools for data-driven applications ranging from phase retrieval to hyperspectral imaging and seismic data completion by exploiting sparsity, low-rankness and tensor structures to analyze high-dimensional data.
In person attendance at this seminar is only open to Columbia University affiliates. External guests are welcome to attend remotely. Please contact [email protected] if you need the Zoom link for this seminar.
