报告人: Xiao Fu (Oregon State University)
时 间: 2018年7月13日 10:30—11:30
地 点: 理科楼 LA106
摘 要:：Hyperspectral images (HSIs) are special images that are measured at a large number of wavelengths, which contain rich spectral information for analytics. Hyperspectral imaging finds a large variety of applications in diverse domains such as agriculture, mining, outer space exploration, and food/medicine security. Because of hardware limitations, HSIs are known to have good spectral resolution but rough spatial resolution. On the other hand, another type of images, namely, multispectral images (MSIs), is usually with fine spatial resolution but the spectral resolution is low. Hyperspectral super-resolution (HSR) is a technique that aims at fusing a hyper spectral image and a (co-registered) multispectral image to produce a super-resolution image that admits high spatial and spectral resolutions. Existing algorithms are mostly based on joint low-rank factorization of the matricized HSI and MSI. Such a framework is effective to some extent, but several challenges remain. First, it is unclear whether or not the super-resolution image is identifiable in theory under this framework, while identifiability usually plays an essential role in parameter estimation. Second, most algorithms assume that the degradation operators from the super-resolution image to the HSI and MSI are known or can be easily estimated -- which is hardly true in practice.
In this talk, I will introduce a novel coupled tensor decomposition method that can effectively circumvent these issues. Leveraging modern tensor algebra, the proposed approach guarantees the identifiability of the super-resolution image under realistic conditions. The proposed method can work even without knowing the spatial degradation operator, which could be hard to accurately estimate in practice. Simulations using real hyperspectral image data are employed to demonstrate the effectiveness of the proposed approach.
报告人简介:Xiao Fu, an Assistant Professor in the School of Electrical Engineering and Computer Science, Oregon State University, Corvallis, OR, United States. He received his Ph.D. degree in Electronic Engineering from The Chinese University of Hong Kong (CUHK), Hong Kong, 2014. He was a Postdoctoral Associate in the Department of Electrical and Computer Engineering, University of Minnesota, Minneapolis, MN, United States, from 2014-2017. His research interests include the broad area of signal processing and machine learning, with a recent emphasis on tensor and matrix factorization. He received a Best Student Paper Award at ICASSP 2014, and co-authored a Best Student Paper Award at IEEE CAMSAP 2015. He was recognized as the ``outstanding postdoctoral scholar’’ by University of Minnesota in 2016.