基于HMM核超球面支持向量机的超宽带SAR未爆物检测

Ultra-Wideband Synthetic Aperture Radar Unexploded Ordnance Detection Using HMM Kernel Hypersphere Support Vector Machine

利用机载超宽带合成孔径雷达(UWB SAR)探测地下未爆物(UXO)具有安全和高效的优点.UXO检测分为预筛选和鉴别.预筛选从大面积SAR图像中提取若干怀疑目标,而鉴别则将这些怀疑目标分成UXO和杂波从而降低虚警.本文提出隐马尔可夫模型(HMM)核的超球面支持向量机(HS-SVM)UXO鉴别器.HS-SVM基于结构风险最小原理并利用核特征空间中的超球面区分UXO和杂波能够解决小训练样本集和无典型杂波样本两个问题.此外将描述UXO多方位特征的HMM作为HS-SVM核函数进一步提高了UXO的鉴别性能.实测数据处理结果表明,HMM核HS-SVM优于HMM和高斯核HS-SVM等UXO鉴别器.

Using air-home Ultra-WideBand Synthetic Aperture Radar （UWB SAR） to detection underground unexploded ordnance （UXO） has the advantages of safety and efficiency. UXO detection is composed of prescreening and discrimination. Prescreening is to extract several suspected targets from SAR imagery of wide areas and discrimination is to classify these suspected targets into UXO and clutter to reduce false alarms. In this paper, the Hidden Markov Model （HMM） kernel HyperSphere Support Vector Machine （HS-SVM） UXO discdminator is proposed. HMM kernel HS-SVM employs the structural risk minimization theory and uses hypersphere in kernel feature space to classify UXO and clutter, which can solve the two problems of a small training set and without typical clutter samples. In addition,the HMM, which describes the UXO multi-aspect feature,is used as the kernel function of HS-SVM can improve the UXO discrimination performance further. The field data processing discrimination results show that HMM kernel HS-SVM outperforms the HMM and the Gaussian kernel HS-SVM in UXO discrimination.