Dissertation Announcement for Deok Han
12/01/15 at 1:00 PM
November 17, 2015
Dear faculty, graduate and undergraduate students,
You are cordially invited to my Ph.D. dissertation oral defense.
Dissertation Title: STATISTIAL ANALYSIS OF RADAR AND HYPERSPECTRAL REMOTE SENSING DATA
When: Tuesday, December 1, 2015, 1:00 pm
Where: Simrall 228
Candidate: Deok Han
Degree: Ph.D., Electrical and Computer Engineering
Committee:
Dr. Qian Du
(Major Professor)
Dr. Tung Lung Wu
(Minor Professor)
Dr. Nicolas H. Younan
(Co major professor)
Dr. James V. Aanstoos
(Committee Member)
Dr. Farshid Vahedifard
(Committee Member)
Abstract:
In this dissertation, three related studies are proposed for radar and hyperspectral remote sensing applications using statistical techniques. The first study is to explore the relationship between synthetic aperture radar (SAR) backscatter and in situ soil property for levee monitoring. Utilizing remote sensing techniques to extract soil properties can facilitate several engineering applications for large-scale monitoring and modeling purposes. This study presents results of a series of statistical analyses which were performed to investigate potential correlations between three independent polarization channels of radar backscatter and various physical soil properties. Polarimetric synthetic aperture radar (PolSAR) imagery from an airborne SAR instrument, the UAVSAR, was used along with an extensive set of in situ soil properties measured over the study area of the lower Mississippi river. The results showed weak but considerable correlation between the cross-polarized (HV) radar backscatter coefficients and most of these properties.
The second study aims to find effective statistical features for levee slide classification. Detrimental damage may occur to the region of failure by levee slides. Usually levee stretches alongside a river, and remote sensing data like SAR images are popular to monitor levee condition. However its SAR data becomes very large making it hard to monitor quickly because of high computation cost and large memory requirement.
Therefore, time-efficient method to monitor levee conditions is necessary. Support vector machine (SVM) has been applied to many applications successfully. However, SVM does not work well on original SAR images with three magnitude bands requiring spatial feature extraction. Gray level co-occurrence matrix (GLCM) is one of the most common methods for extracting textural information from grey-scale images, but it may not be practically useful for a big data in terms of calculation time. In this study, very efficient feature extraction methods with spatial lowpass filtering are proposed to use, including a weighted average filter and a majority filter in conjunction with a nonlinear band normalization process. Feature extraction with these filters yielded comparable results to GLCM with much lower computational cost.
The third study is focused on the case when only a small number of ground truth labels are available for classification. To overcome the difficulty of not having enough training samples, a semisupervised method is proposed. The major idea is to expand confident ground truth using a relationship between existing labeled data (ground truth) and other unlabeled data. A fast self-learning algorithm as one of semisupervised approaches was presented in this study. Reliable unlabeled data are chosen from SVM output using majority voting (MV) and weighted majority voting (WMV), and those are added to labeled data to build a better SVM classifier. The results showed MV and WMV can select reliable unlabeled data for a self-learning processing, and WMV offers better performance than MV.
Best regards,
Deok Han