【作者】 金添；

【导师】 周智敏；

【作者基本信息】 国防科学技术大学 ， 信息与通信工程， 2007， 博士

【摘要】 机载或车载超宽带合成孔径雷达(Synthetic Aperture Radar,SAR)能够实现大区域浅埋目标的快速探测,具有安全和高效的优点。本文结合实际系统研究了超宽带SAR浅埋目标成像与检测中存在的若干理论和技术问题。传统超宽带SAR信息处理将成像与检测割裂开来,限制了最终检测性能的提高。因此本文提出了“成像和检测一体化框架”的思想,包括面向检测的成像和基于成像的检测。在此基础上,提出了一种面向检测的时频表示成像算法(Time-Frequency Representation Image Formation,TFRIF),该方法也可有效解决基于成像的检测这一技术难题,具有实际应用价值。针对传统成像算法会引起浅埋目标定位误差和图像散焦的问题,提出了修正波前重构(Modified Wavefront Reconstruction,MWR)和浅地表BP(SubsurfaceBack-Projection,SBP)两种浅埋目标成像算法,具有较高的聚焦和定位精度;针对实际工作条件下无法获取埋设深度等先验信息,提出了图像域折射和色散影响校正(Image Domain Refraction and Dispersion Correction,IDRDC)的浅埋目标聚焦和定位方法,该方法可有效解决无先验信息时的不同埋设深度和土壤环境中浅埋目标的聚焦和定位问题。为了进一步提高图像质量,提出了基于二维频域支撑区特性的射频干扰(Radio Frequency Interference,RFI)抑制技术和适合前视系统的地距平面多视配准技术,分别可在降低系统复杂度的同时保证RFI抑制性能和有效提高前视系统多视处理相干斑噪声抑制中的配准效率。目前制约超宽带SAR浅埋目标探测实用化的主要问题是虚警太多,因此提取有效特征和设计合适鉴别器是提高浅埋目标检测性能的关键。本文研究了金属地雷和未爆物特征提取技术,提出了基于图像域的金属地雷双峰特征增强算法,在此基础上,提出了基于空间-波数分布(Space-Wavenumber Distribution,SWD)的金属地雷四维散射函数估计及其特征选择方法,提取了包含双峰特性及方位不变性的特征向量;并利用SWD和Hu不变矩,提取了未爆物的多方位特征。对于浅埋目标鉴别这种小样本学习和一类分类问题,考虑到目标和杂波误判风险不同以及埋设环境多样性等因素,提出了模糊超球面支持向量机(FuzzyHyoerSphere Support Vector Machine、FHS-SVM)浅埋目标鉴别算法,并将证据框架理论应用于高斯核FHS-SVM超参数优化,有效降低了检测结果的总体误判风险,提高了不同探测环境下金属地雷和未爆物的鉴别性能。在此基础上,提出用描述未爆物散射多方位特性的隐马尔可夫模型核替换高斯核函数,进一步改善了FHS-SVM对未爆物的鉴别性能。更多还原

【Abstract】 Air- or vehicle-borne ultra-wideband Synthetic Aperture Radar (SAR) can perform quick detection of shallow buried targets over large areas, which has the advantages of safety and efficiency. In this thesis, some theoretical and technical problems in ultra-wideband SAR shallow buried targets imaging and detection are studied on real systems.In the traditional UWB SAR information processing, imaging and detection are separated, which limits the improvement of final detection performance. Therefore, in this thesis, the theory of the imaging and detection integrated framework is proposed, which includes the detection oriented imaging and the imaging based detection. Based on this framework, a detection oriented Time-Frequency Representation Image Formation (TFRIF) is proposed, which can also solve the technical problem of imaging based detection and has applicable value in practice.For the problem of locating error and imaging focusing of shallow buried target caused by traditional image formations, two shallow buried target image formations, the Modified Wavefront Reconstruction (MWR) and the Subsurface Back-Projection (SBP) algorithms, are proposed, which have higher focusing and locating precision. According to the problem of no priori information of buried depth etc. in practical operating conditions, an Image Domain Refraction and Dispersion Correction (IDRDC) method for shallow buried targets focusing and locating is proposed, which can solve the problem of focusing and locating multiple targets with different buried depths in various soil environments without priori information.In order to improve the image quality further, a Radio Frequency Interference (RFI) suppression method on the region of support characteristic in the 2-dimensional frequency domain and a ground-plane multi-look registration technique for forward-looking systems are proposed, which can ensure RFI suppression performance while reduce system complex and improve the registration efficiency of the multi-look processing to suppress speckle noise in forward-looking systems, respectively.Nowadays, the major challenge limiting the practical use of UWB SAR in shallow buried targets detection is the too many false alarms. Therefore, efficient feature extraction and suitable discriminator design are key points to improve the shallow buried targets detection performance. In this thesis, the feature extraction technique for metallic landmines and unexploded ordnances is studied. For a metallic landmine target, a double-peak feature enhancement algorithm in the image domain and a Space-Wavenumber Distribution (SWD) based 4-dimensional scattering function estimation and associated feature selection method are proposed, which can extract the feature vector with the double-peak and aspect-invariance characteristics. For an unexploded ordnance target, based on the SWD and the Hu moment invariants, the multi-aspect feature is extracted.For the small sample learning and the one-class classification problems as shallow buried targets discrimination, considering the factors of the misclassification risk difference between targets and clutter and the buried environment diversity, the Fuzzy HyperSphere Support Vector Machine (FHS-SVM) shallow buried target discrimination algorithm is proposed. Furthermore, the evidence framework is applied to the problem of the Gaussian kernel FHS-SVM hyperparameters optimization, which can reduce the total misclassification risk of detection result and improve the discrimination performance for both metallic landmines and unexploded ordnances in varying detection environments. Based on the FHS-SVM, the hidden Markov model (HMM) kernel, which describes the multi-aspect characteristic of unexploded ordnance, is used to replace the Gaussian kernel to improve the FHS-SVM discrimination performance for unexploded ordnances further.更多还原