【作者】 成艳；

【导师】 张建中；

【作者基本信息】 厦门大学 ， 信号与信息处理， 2008， 硕士

【摘要】 探地雷达成像是最直接和最有效的探地雷达目标检测与识别技术,探地雷达成像方法一直是探地雷达信号处理方法的研究热点。目前,偏移成像及合成孔径成像等方法较成熟,但均不能重建地下目标物体的电性参数,从而不能对目标物体进行分类识别。逆散射成像能重建地下目标物体的电性参数,但逆散射成像是个非线性问题,而非线性逆散射成像算法计算量巨大,难以适应实际应用。因此研究计算量小的线性逆散射成像算法具有非常重要的实用价值。本论文针对近似无损和有损介质,并考虑到探地雷达应用中一些实际因素,研究并实现了线性逆散射成像算法,使之达到能够在现场进行处理的程度。首先,研究了基于探地雷达数值模拟软件GprMax的正演模拟相关技术,包括激励源、离散步长及吸收边界条件等的选取原则;通过对多个模型的散射数据进行模拟,分析了影响散射数据质量的因素,也为后面逆散射成像算法提供了所需要的散射数据。其次,根据衍射层析成像理论,并考虑到空气和土壤分界面的情况,提出了一种无损或近似无损介质的三维探地雷达线性逆散射成像算法。在此基础上,从目标函数的空间谱域角度分析了探地雷达工作频率及水平测线对算法的影响,并对算法频率步长的选取进行了优化,使得在保证算法成像效果的同时,最大限度地避免对冗余数据的采集和处理,从而提高了算法计算效率,节约了大量数据采集时间。对仿真数据和实测数据的成像例子表明该算法能对目标物体进行快速且准确的重建,且取优化频率步长时计算效率更高。最后,基于一阶born近似、并矢格林函数及奇异值分解技术,提出了一种有损介质的线性逆散射成像算法。该算法考虑到了天线辐射类型和空气土壤界面的情况,使得算法更符合实际应用环境;采用渐进近似方法对描述目标函数与频域散射数据关系的线性算子进行近似,避免了离散线性算子时对多个积分进行计算,使得离散步骤更简单,且大大减少了计算量,进一步提高了算法的计算效率。对仿真数据和实测数据的成像结果表明该算法能对目标物体进行快速且较好的重建。更多还原

【Abstract】 Ground penetrating radar (GPR) is a nondestructive testing technique of buried targets developed in these decades. It has been widely used in many fields such as demining, archaeology and civil engineering. The imaging technique of ground penetrating radar is the most promising data processing technique for GPR application. So far, the migration imaging algorithms are more mature and they can reconstruct the location and shape of buried object accurately, but they can not reconstruct the dielectric and conductive properties of the buried objects. Both non-linear and linear inverse scattering algorithms can reconstruct the dielectric and conductive properties of the buried objects. But non-linear methods have heavy computational burden, so they are not suitable for engineering application. Thus, it is highly desirable and valuable to do research on linear inverse scattering imaging algorithms. The research work in this thesis aims to improve the performance of GPR linear inverse scattering algorithms for lossless and lossy soil respectively.Firstly, the forward modelling is investigated based on GprMax which is a numerical modelling tool of GPR. The main work of this part includes the following three aspects: The research on excitation source, discretization steps and absorbing boundary conditions, which are techniques relating to the forward modelling based on GprMax; The research on the input and output file of GprMax; The analysis of the factors that influence the quality of scattered field data according to the basic theory and the numerical modelling results of serveral models, the purpose of which is to get high-quality scattered field data by selecting the right parameters and methods when collecting or modelling the data.Secondly, a three-dimensional linear GPR inverse scattering imaging algorithm which takes the planar air-soil interface into account is derived for lossless soil. The first Born Approximation and the dyadic Green function for a two-layer medium are used to get a linear forward model, the forward model is then inverted using Fourier transform. Basing on the above work, the influences of the frequecy diversity and observation line are analyzed with reference to the spatial spectral of the unknown object function, and then an optimal frequency step for the algorithm is determined, which can ensure nonredundancy in the data, thereby enhancing the computational effectiveness of the algorithm and economizing a large amount of time for collecting data. Numercial examples show that the algorithm can reconstruct the buried object rapidly and accurately.Finally, a three-dimensional linear GPR inverse scattering imaging algorithm based on the first Born Approximation, the dyadic Green function for a two-layer medium and singular value decomposition (SVD) is derived for lossy soil. The algorithm takes the radiation patterns and the planar air-soil interface into account, which makes it accord with the practical application of GPR better. Meanwhile, the asymptotic approximation is used to achieve the linear relationship between the unknown object function and the spectrum of the scattered field, which makes it avoid evaluating several integrals when discretizing the linear operator, thereby making it much easier to implement the discretizing step and reducing the computational burden, thus enhancing the computational effectiveness of the algorithm. Numerical examples show that the algorithm can reconstruct the buried object well.更多还原