Dissertation Announcement for Yan Xu — 03/06/2019 at 1:00 PM
January 28, 2019
Dear Faculty, Graduate and Undergraduate Students,
You are cordially invited to my Ph. D. dissertation defense.
Dissertation Title: Efficient Analysis of Hyperspectral Remote Sensing Imagery
When: Wednesday, March 6, 2019, 1:00 pm
Where: Simrall 228
Candidate: Yan Xu
Degree: Ph.D., Electrical and Computer Engineering
Committee:
Dr. Jenny Q. Du
Professor of Electrical and Computer Engineering
(Major Professor)
Dr. James E Fowler
Professor of Electrical and Computer Engineering
(Committee Member)
Dr. Nicolas H. Younan
Professor and Department Head of Electrical and Computer Engineering
(Committee Member)
Dr. Qingmin Meng
Associate professor of Geosciences
(Committee Member)
Abstract
This dissertation develops new efficient algorithms for hyperspectral remote sensing imagery. Because of the large number of spatial pixels and high spectral dimensionality, it is necessary to develop efficient algorithms or conduct dimensionality reduction for fast processing.
An efficient probabilistic collaborative representation-based classifier (PROCRC) is proposed for hyperspectral image classification. Different types of spatial features including shape feature (i.e., extended multi-attribute feature), global feature (i.e., Gabor feature), and local feature (i.e., local binary pattern) are investigated. The original collaborative representation classifier (CRC) has limited classification performance. The Tikhonov regularized versions of CRC have excellent classification performance but their computational cost is high. The proposed framework offers superior classification performance but with much lower computational cost.
Nonlinear classification of multispectral imagery, using representation based classifications, is presented. The high spatial resolution of multispectral imagery provides merits for practical applications. However, due to rough spectral resolution, classification performance for multispectral imagery is often limited. Kernel based algorithms can be used to improve the classification performance, but they suffer from high computational cost. A simple nonlinear band ratio function is presented, which can yield better classification performance than the kernel method, while keeping the complexity low.
Fast kernel collaborative representation is also proposed for hyperspectral image classification. To reduce the computational burden of traditional kernel collaborative representation based classifiers, an explicit kernel mapping method is adopted to reduce the time complexity without decrease of classification accuracy.
Moreover, fast band selection techniques are built for hyperspectral target detection. Due to lack of training samples in a detection problem, it is more difficult than classification-purposed band selection. A simple yet effective objective function, called Maximum-Submaximum-Ratio (MSR) gauging target-background separation, is proposed for target detection. Efficient evolutionary searching methods such as firefly algorithm (FA) are adopted due to their high-dimensional space searching capability.
Best Regards,
Yan Xu