Development of high performance classification algorithm for ground penetrating radar systems used in landmine detection

在这项研究中，我们开发了一种用于地雷探测的探地雷达系统的高性能分类算法。地雷探测过程分为两个步骤。第一步是寻找阶段，所有类型的埋藏物都在这里。第二个阶段是识别阶段，然后使用参考数据区分地雷和石块和其他物体。在本研究中，我们考虑了识别阶段。1.环境杂波去除与特征提取为了去除探地雷达数据中的杂波，提出了一种基于字典匹配的杂波去除方法。对于目标分类，我们提出了两种基于隐马尔可夫模型和似然比测试的分类算法。3.改进了数据生成的数值方法。为了评估其分类性能，需要使用FDTD方法生成的数据集进行蒙特卡罗模拟。由于生成数百个模拟数据需要较大的计算时间和存储量，因此我们对数值方法进行了改进。4.性能评估我们以接收机工作特性曲线的形式给出了分类性能，结果表明，即使是埋在粗糙地面下浅层的地雷，也获得了良好的分类性能。还应利用通过现场试验获得的实际探地雷达数据进行性能评估。这一重要的研究问题目前正在调查中。

In this research, we have developed a high performance classification algorithm for ground penetrating radar(GPR) systems used in landmine detection. A process of landmine detection is divided into two steps. The first step is the find stage, where all the various types of buried objects are located. The second stage, the identification stage, then differentiates landmines from stones and other objects using reference data. In this research, we have considered the identification stage. Results of this research are summarized as follows :1.Ground clutter removal and feature extractionIn order to remove clutters from the GPR data, we have proposed a method based on a Matching Pursuits with a dictionary whose elements are deformed incident pulses. After the removal of the clutters, we have extracted three kinds of target features related to wave correlation, energy ratio, and signal arrival time from the residual signals.2.High performance classification algorithmFor target classification, we have proposed two classification algorithms based on a theory of hidden Markov models and a likelihood ration test.3.Improvement of numerical method for data generationIn order to evaluate its classification performance, a Monte Carlo simulation using dataset generated by an FDTD method is required. Since it takes large computation time and storage capacity for generating hundreds of simulation data, we have improved the numerical method.4.Evaluation of performanceWe have presented the classification performance in the form of receiver operating characteristics curves and have shown that good classification performance has been obtained, even for landmines buried at shallow depths under rough ground surfaces. Performance evaluation using actual GPR data obtained through field experiments should also be undertaken. This important research problem is currently under investigation.