A Research On The Ultra-Wide Band SAR Detection Techniques Of Shallow Buried Target

Vehicle-borne Ultra - Wide Band Synthetic Aperture Radar (UWB SAR) has ground-penetrating capability to perform shallow buried targets, i.e. minefield or one landmine, quick detection over large areas, which is a safe and efficient detection method. In this thesis, shallow buried target detection information processing is consisted of three parts which are pre-processing, feature extraction and detection method. In this thesis, two important concepts: feature extraction and feature-based detection and their realization methods are proposed all together.In this thesis, all the data is developed, based on which, a novel Radio Frequency Interference (RFI) suppression method on the characteristic of the Rail-GPSAR region scan. In order to better the input for feature extraction and detection, a series of pre-processing methods will be taken, which are RFI, CFAR-Whitening and Energy Circle. Firstly, RIF can extract the frequency and aspect angle information in the target reflectivity function while maintains 2-dimensional high spatial resolutions. Then, based on the CFAR theory we better the image result. Ultimately, the Energy-circle is taken to find the important detection point.Feature extraction is the key of shallow buried objects detection. Every target samples will be transformed to a vector at the end, so in this thesis, we take the extraction operation based on two sides: samples structure and samples space. After comparison between the Structal chip and Structal range-cut, we introduce principal component analysis (PCA) and independent component analysis (ICA) data processing method which can extract the prime components and independent components respectively. Then KPCA and KICA method is introduced to extract more non-linear feature. Ultimately, Fisher discriminant information is defined to select optimal discriminant independent components.Based on the feature extraction methods we work over the detector which are: Generalized Likelihood Ratio Test (GLRT), deflection-optimal linear-quadratic (DOLQ) detector, and Support Vector Machine (SVM) detector. Results based on Rail-GPSAR data show that: SVM is more suitable to those small samples learning problems, i.e, landmine detection. In this thesis we impose a ICA&SVM detection method. The detection results of the real data show that the method has an effective landmine detection performance.