Signal processing algorithm development for land mine detection

探測地雷的信號處理算法之研發

Student thesis: Master's Thesis

View graph of relations

Author(s)

  • Chin Tao Thomas CHAN

Related Research Unit(s)

Detail(s)

Awarding Institution
Supervisors/Advisors
Award date15 Jul 2008

Abstract

As a result from numerous wars, millions of buried land mines are still left around the world. Ground penetrating radar (GPR) is an emerging tool that is used to detect these buried land mines as it allows for detection without causing harm to the environment. However, difficulties still exist as the signals are very susceptible to the ground surface response and the different conditions the land mines are buried in. Therefore, signal processing techniques are needed to increase the accuracy of land mine detection. In this thesis, an overview of existing methods is first presented. Two novel land mine detection algorithms are then devised. The algorithms are tested on both real and simulated ground penetrating radar data. The first algorithm is based on state space estimation using sequential Monte Carlo methods. In the state space formulation, the model space contains more than one model. A reversible jump Markov chain Monte Carlo move is used to explore the model space instead of relying on a discrete decision to choose the correct model. The results show that the algorithm is able to detect the land mine targets using this strategy and its performance is superior to the Kalman filter method which makes a discrete decision on which model to follow. The second algorithm involves estimating the background signal using a two sided linear prediction (LP) model. It is an extension to the one sided LP model such that a more accurate estimation of the background signal is made. The detection is then made by examining the residual signal which is obtained by subtracting the estimated background signal from the observed signal. Evaluation results based on the contour plots and the receiver operating characteristic curves show that the two sided LP model is superior in performance over existing background estimation methods including the one sided LP and the adaptive shifted and scale approaches.

    Research areas

  • Computer algorithms, Signal processing, Detection, Land mines