Development of a plastic landmine detection system based on CMRF model

本研究的目的是开发一种基于新提出的理论，即复值神经网络，以类脑方式自适应处理空间和频域干涉纹理的杀伤人员塑料地雷可视化雷达系统。该系统由三部分组成：(1)密集集成宽带天线手持设备；(2)采用复值自组织映射（CSOM）的图像聚类处理模块；(3)基于复值联想记忆的地雷类标识。(1)密集集成宽带天线手机：我们开发了一种新的天线元件，即壁式ltsa，它具有更小的孔径和更宽的带宽。该天线阵列使我们能够以相敏方式以高空间分辨率二维捕获地下反射。(2)基于复值自组织映射（complex-value self-organizing map， CSOM）的图像聚类处理模块：我们还开发了基于CSOM的自适应纹理分类器…More。反射在空间（二维）和频率（一维）域中（共三维）具有复杂的振幅信息。CSOM自适应地考虑复合振幅纹理对三维图像进行分割。该模块成功地将塑料地雷区与其他土壤区（例如石头和金属碎片）区分开来。(3)基于复值联想记忆的地雷类别识别：在CSOM模块中，我们需要在分割区域中识别塑料地雷类别。这个识别模块在两个阶段处理特征向量。首先，该模块在记忆的教师数据中选择最接近测试观测的特征向量集。然后，确定各个特征向量与教师数据中的特征向量的对应关系。在开发的系统中，我们发现在实验室实验中，塑料地雷的可视化性能接近金属地雷的金属探测器的可视化性能。但是，我们在观察中也发现，其性能依赖于土壤、湿度等环境条件。因此，在下一步，我们将在实验室之外进行现场测试，但是，在几乎与地雷场相同的场地上。在我们的计划中，我们在下一学年开始实地试验。少

This study was carried out with the purpose of developing an antipersonnel plastic landmine visualization radar system that deals with spatial-and frequency-domain interferometric texture adaptively in a brainlike manner based on the newly proposed theory, i.e., complex-valued neural networks. The developed system consists of three parts : (1)densely integrated wideband antenna handset, (2)iAmage clustering processing module employing complex-valued self-organizing map (CSOM), and (3)landmine-class identifier based on complex-valued associative memory.(l)Densely integrated wideband antenna handset :We have developed a new antenna element, i.e., walled-LTSA, which has a smaller aperture and a wide bandwidth. This antenna array enables us to capture subsurface reflection two-dimensionally with a high spatial resolution in a phase-sensitive manner.(2)Image clustering processing module employing complex-valued self-organizing map(CSOM) :We have also developed an adaptive texture classifier … More based on the CSOM. The reflection has complex-amplitude information in space (two-dimension) and frequency (one-dimension) domains (three-dimension in total). The CSOM segment the three-dimensional image by taking into account the complex-amplitude texture in an adaptive manner. The module successfully distinguishes a plastic landmine area from other soil areas including, e.g., stones and metal fractions.(3)Landmine-class identifier based on complex-valued associative memory :We have to identify plastic landmine class(es) among the segmented areas in the CSOM module. This identifying module deals with the feature vectors on double stages. First, the module chooses a feature-vector sets, in the memorized teacher data, nearest to those in the test observation. Then, it determines the correspondence of respective feature vectors to those in the teacher data.In the developed system, we found that in the laboratory experiment the visualization performance for plastic landmines is near to that of the metal detectors for metal landmines. However, we also find that the performance is dependent on the conditions of soil, humidity, and other environment in the observation. Therefore, in the next step, we are going to have a field test out of the laboratory, but instead, on the field almost identical to that of landmine field next. In our plan, we start the field test in the following academic year. Less