Buried Small Target Imaging And Detection With Surface Penetrating Radar

Being capable of penetrating medium and sensing the abrupt change of electromagnetic prosperities,the Surface Penetrating Radar(SPR)is a promising remote sensing modality for subsurface target detection.This dissertation focuses on the detection of Shallow Buried Small Target(SBST)with SPR.The term ”shallow buried” refers to that the typical target depth is about a few wavelengths at the central operational frequency or less.And a target is treated as small target when its size is comparable to or smaller than the wavelength at the central operational frequency.Actually,the need for detection of SBST is urgent because many countries and regions across the world is contaminated by Antipersonnel Mines,which are typical and lethal SBSTs.The efficiency of APM detection with SPR is plagued by the high False Alarm Rate(FAR)because APM gives weak responses,which are tend to be masked and distorted by the heavy clutter and noise.Aiming at improving the detection performance of SPR,this dissertation proposes a solution for the SBST detection problem.First,the electromagnetic scattering characteristics of the SBST is discussed.The backscattering response model for SBST is derived from source-type integral representations of the scattered field.Based on the model,we investigate the differentiation effect of SBST's response in time domain and the amplitude modulation of spectrum in frequency domain,which will be used as the target feature for target discrimination.Second,the method for SBST imaging is presented.Detection based on the raw data is difficult because of the weak scattering nature of SBST.However,the target response can be focused and enhanced by the coherent imaging.Therefore,we present a two step strategy for imaging of the SBST under a rough surface.Specifically,the wavefield extrapolation is utilized to compensate the phase shift caused by the air layer and the phased screen approximation is used to cancel the phase distortion cause by the rough surface.Third,the target detection in presence of strong clutter and noise is studied.Taking advantage of low-rank and sparse structure of SPR image,we firstly adopts an efficient Robust Principal Component Analysis(RPCA)technique to extract the target image.Then,thresholds are applied to the extracted image to detect target and reject false alarms.Furthermore,we propose a novel target detection method taking advantage of the low-rank and sparse structure in the multidimensional data.We firstly created a multidimensional image tensor using subband SPR images that are computed from the band-pass filtered GPR signals,such that differences of target response between subbands can be captured.Then,exploiting the low-rank and sparse property of the image tensor,we use the recently proposed Tensor Robust Principal Analysis to remove clutter by decomposing the image tensor into three components: a low-rank component containing the clutter,a sparse component capturing the target response,and noise.Finally,target detection are accomplished by applying thresholds to the extracted target image.At last,the SBST discrimination is studied.Utilizing the backscattering model of SBST,we investigate the time-frequency representation of SBST repsonse and find out the underlying mechanism for the peak and valley of the instant frequency curve,which is then adopted as the target feature.Based on the Synchrosqueezing Transform(SST),a feature extraction approach is proposed.Then an Support Vector Machine(SVM)is utilized to classify the detection alarms in to target and non-target.