Doctoral Dissertations

Abstract

"This dissertation addresses the problem of anomaly detection in spatial data. The problem of landmine detection in airborne spatial data is chosen as the specific detection scenario. The first part of the dissertation deals with the development of a fast algorithm for kernel-based non-linear anomaly detection in the airborne spatial data. The original Kernel RX algorithm, proposed by Kwon et al. [2005a], suffers from the problem of high computational complexity, and has seen limited application. With the aim to reduce the computational complexity, a reformulated version of the Kernel RX, termed the Spatially Weighted Kernel RX (SW-KRX), is presented. It is shown that under this reformulation, the detector statistics can be obtained directly as a function of the centered kernel Gram matrix. Subsequently, a methodology for the fast computation of the centered kernel Gram matrix is proposed. The key idea behind the proposed methodology is to decompose the set of image pixels into clusters, and expediting the computations by approximating the effect of each cluster as a whole. The SW-KRX algorithm is implemented for a special case, and comparative results are compiled for the SW-KRX vis-à-vis the RX anomaly detector. In the second part of the dissertation, a detection methodology for buried mine detection is presented. The methodology is based on extraction of color texture information using cross-co-occurrence features. A feature selection methodology based on Bhattacharya coefficients and principal feature analysis is proposed and detection results with different feature-based detectors are presented, to demonstrate the effectiveness of the proposed methodology in the extraction of useful discriminatory information"--Abstract, page iii.

Advisor(s)

Agarwal, Sanjeev, 1971-

Committee Member(s)

Roe, Robert P.
Wunsch, Donald C.
Madria, Sanjay Kumar
Moss, Randy Hays, 1953-

Department(s)

Electrical and Computer Engineering

Degree Name

Ph. D. in Electrical Engineering

Sponsor(s)

Night Vision Laboratory. Countermine Division

Publisher

University of Missouri--Rolla

Publication Date

Fall 2007

Journal article titles appearing in thesis/dissertation

  • Buried mine detection using co-occurrence texture features

Pagination

x, 135 pages

Note about bibliography

Includes bibliographical references (pages 122-134).

Rights

© 2007 Spandan Tiwari, All rights reserved.

Document Type

Dissertation - Open Access

File Type

text

Language

English

Subject Headings

Kernel functionsLand mines -- Detection -- Mathematical modelsLand mines -- DetectionMines (Military explosives) -- Detection

Thesis Number

T 9953

Print OCLC #

794777350

Electronic OCLC #

779946944

Download

Share

 
COinS