基于频率分集阵列的有耗介质目标探测方法研究

In the fields of geological exploration, urban construction and disposal of landmines or abandoned military explosives, there are needs for the hidden multi-targets detection in layered media. By changing the carrier frequency emitted by each element, frequency diverse array(FDA) provides the beam pattern which has the characteristics of range-angle-dependent, periodic bending and time-varying, and automatically realizes multi beam scanning to detect the targets. In this project, according to the application scenarios of ground penetrating and through-the-wall radar, the transmission and reception mechanism of FDA of near field is explored, and a lossy dielectric beam pattern model is established which guides the principle and performance analysis of the relevant parameter selection. In the light of the characteristics of the FDA beam pattern covering the whole distance domain, the attribute and distribution characteristics of the measured target and layered medium are speculated, and the energy focusing method of the transmitting beam is proposed, so as to ignore the influence of the interlayer medium. Based on the characteristics of FDA beam scanning, combined with the principle of inverse imaging and sparse decomposition, the algorithm of target imaging and location is proposed to get high resolution range and angle information. The theoretical deduction and experimental construction are completed to support near field transmitting and receiving system easy for hardware implementation, and effectively solve multi-target detection problems in lossy layered medium.

在地质勘探、城市建设以及战事遗留地雷或废弃军事未爆物排除等领域都存在对分层介质中多目标或隐蔽目标探测的需求。频率分集阵列基于波束形成理论，通过改变各阵元载波频率，产生依赖角度、距离，呈现周期弯曲且时变特征的波束方向图，自动进行多波束扫描实现目标探测。本项目结合探地和穿墙应用场景，研究频率分集阵列（Frequency Diverse Array,FDA）发射和接收机制，建立有耗介质波束方向图模型，给出相关参数选择的原则和性能分析；根据FDA波束方向图覆盖整个距离域的特性，推测出被测目标和分层介质的特征，提出发射波束能量聚焦方法，从而忽略中间层介质影响；基于FDA波束扫描特征，结合逆成像、稀疏分解原理，提出目标成像和定位算法，获取高分辨率距离角度信息。完成有关理论推演和实验搭建，为构建易于硬件实现的近场发射接收系统提供支撑，有效解决有耗分层介质中多目标探测难题。

频率分集阵列（FDA）雷达以其距离和角度依赖优势获得了广泛关注。现阶段对频率分集FDA雷达的研究主要集中在远场，本项目研究其在近场尤其探地雷达环境中的应用。. 项目研究了频率分集阵列雷达应用于地下目标的定位和识别，提出了对称双频偏FDA-MIMO雷达波束扫描地下目标定位方法，获得了被遮挡目标的高精度定位结果；为消除单频偏下距离和角度耦合，提出了FDA-BP神经网络、FDA-CNN神经网络目标定位方法，直接获得目标的距离、角度信息。. 研究多层介质情况下探地雷达压缩感知三维成像，提出了多层介质场景近似等效为三层并基于二分法求取折射点方法，大幅增加了成像方法在不同场景中的适用性，使用堆叠-拆分方式构造字典，实现了多层介质下目标三维成像。为有效重构层状介质相对介电常数和厚度等参数，提出了基于重采样粒子群算法的层状介质反演方法，通过比例选择法对粒子进行重采样并复制表现好的粒子，提升了反演精度、稳定性和实时性。. 为提升地下目标定位及成像精度，针对传统算法受发射信号脉宽、波形等因素影响，无法直观反应埋藏物体位置问题，提出了压缩感知卷积后向投影方法；提出了复块稀疏贝叶斯步进频率探地雷达成像算法、频率分集阵列FDA离网探测方法，结合复高斯尺度混合模型拓展复数域应用，获得了埋藏物体的准确位置。. 为有效实现地下目标分类，提出了卷积神经网络CNN融合特征热力图感兴趣区域目标分类算法、改进一阶全卷积目标检测算法等，实现了真实探测环境下目标分类与识别。