Subspace Projection And Simultaneous Blind Signal Extraction Based Ground Bounce Removal Methods For Ground Penetrating Radar

There are about 110 million landmines buried in over 60 countries. These unexploded landmines are causing enormous economic problems and have posted a threat to the lives of thousands of innocent people. Thus landmine detection has become an emergent problem. For shallowly buried plastic landmines, the ground bounce (GB) is usually much stronger than the weak return from landmines and their arrival time is very close to each other, which makes it very difficult to detect landmines. So the GB removal is very important for the following tasks, such as target detection, imaging and recognition.Non-parametric methods, including spatial filtering, mean removal, median removal and others, are very simple in computation. However they are only effective to remove homogeneous GB. Parametric scattering model can be used to remove the GB effectively when there are distinct layers below the surface. Semi-parametric methods including SaS, ASaS and RLP, are suitable for rough surface, but their performance relies heavily on the choice of the reference clutter data.Subspace-based methods including PCA and ICA, have no special clutter model, so they can be used in a wider field. However the selection of Principal Components (PCs) and Independent Components (ICs) is a difficult problem. In this thesis, a modified subspace-based method is presented first. In this method, a fully automatic eigenimage based PC and IC selection strategy is introduced in combination with non-homogeneous detector (NHD). A SBSE based method is also proposed for the GB removal, which has much less number of components to extract, and hence less computational load can be reduced. Experimental results show that the proposed methods exhibit good performance.