DETECTION AND THE PERFORMANCE OPTIMIZATION OF SHALLOW BURIED OBJECTS WITH MICROWAVE AND IMAGE ANALYSIS APPROACH

Abstract

The broad objective of this dissertation is to develop fusion approach using Electromagnetics and signal processing techniques to improve the detectability of buried metallic objects and dummy anti-personnel mines. This objective can be again classified in two parts. In the first part the algorithm for ground based scatterometer, to detect the shape and the depth of shallow buried object is proposed while in the second part we propose an algorithm for radar images, for detection of buried objects. With the ground based scatterometer data, first the shape reconstruction algorithm is developed, which is based on two approaches; (a)Image processing and (b)Support Vector Machine Classification (SVM-C) and performance of both of these were compared, while, for the depth estimation, we employ Support Vector Machine Regression(SVM-R) as an optimization technique. The support vector machine is a learning algorithm that can be seen as an alternative technique for linear, polynomial, Radial Basis function and sigmoid classifiers. In this technique the principle of structural risk minimization (SRM) is employed that bounds the maximum generalization error. Firstly with the scatterometer data, the mask is generated that represents the shape of the buried utility and the scattered power over the masked region is used to train for the depth of the buried utility. The SVM Regression is employed for the estimation of the functional dependence of the scattered power and the depth of the buried utility. The experimental setup consists of transmitter and receiver system mounted on the stand over the sand pit and when operated it moves over it. In this system, single pyramidal horn antenna is used for transistor and receiver. This system can be used for detection of shallow buried objects. In the satellite based system, the image processing algorithm is proposed which employs the unsupervised classification. This algorithm does not require any a priori information about the region. This algorithm is based on the statistics derived from the electrical properties of the sand, dummy landmines and the metallic sheets. In present dissertation, we compared the performance of K-Means and the Iso-data clustering algorithms for identifying the region of interest (position of the buried utilities) in the radar images.