Research On Landmine Target Detection Techniques Based On Airborne SAR Image

Airborne Ultra-wideband Synthetic Aperture Radar(UWB SAR) has the ability of penetrating soil or foliage and imaging in high resolution. With the characteristic of long operating distance, being less affected by operating condition and richer aspect-frequency dependent information of targets, airborne UWB SAR is becoming a promising method of landmine detection in large areas. Based on the analysis of the electromagnetic scattering characteristic of landmine target in airborne UWB SAR, key techniques in landmine target detection are discussed in this article. The major work can be listed as follows:Firstly, the problem of feature extraction of landmine in airborne UWB SAR is investigated. The method of physical optics is exploited to analyze the double-hump characteristic of landmines in time domain and applicability of common landmine characteristics is then discussed. Because of the wide aspectual angle and bandwidth of UWB SAR, aspect-frequency dependent information of landmine is then extracted.Secondly, the problem of prescreening in landmine detection is investigated. By analyzing Constant False Alarm Rate(CFAR) method, we conclude that CFAR method lacks full use of target information. Aiming at sovle the problem of high false alarm rate in landmine detection, Azimuth Scattering Entropy(ASE) is used to measure the aspectual invariance of landmines and then adopted as detection feature. Fusion detection algrithom which exploits the characteristics of aspectual invariance and local contrast is then proposed. The results of experimental data show that fusion detection has better detection performance than CFAR method.Finally, the problem of landmine discrimination is investigated. Support Vector Machine(SVM) algorithm is first introduced and HyperSphere SVM(HS-SVM) is used as the discriminator in landmine detection. Based on further analysis of target's aspect-frequency dependence, Singular Value Decomposition(SVD) is used to extract structural information of aspect-frequency dependence. Singular value character vector is input in HS-SVM to discriminate landmine.The results of experimental data demonstrate the effectiveness of the proposed method.