基于HRRP特征的墙后静止人体目标识别关键技术研究

In UWB through wall human detection technology which has extensive applied prospect in domain of searching and rescuing in disaster, anti-terrorism and etc, because it is difficult to identify static human target behind the wall under the condition of micro-Doppler feature is not obvious, feature extraction of high resolution range profile (HRRP) become a key technology to solve this problem. Therefore, we will study: firstly, analyze the signal reception process based on electromagnetic scattering theory, and then establish the UWB through wall radar signal model; secondly, based on the established model, transform wall clutter mitigation into blind separation and clustering of scattering center signal, and then using independent component analysis and k-means clustering algorithm for radar signals processing to eliminate the influence of wall environment on feature extraction of HRRP; lastly, based on the established model, transform HRRP feature extraction into scattering center estimation, by designing a dictionary, the problem of scattering center estimation is sparsely represented, and then use compressed sensing for signal after wall clutter mitigation to reconstruct parameters of target scattering center. The research of this project will improve the recognition framework of static human target behind the wall based on HRRP feature and be as a basic research for application.

在灾害搜救、反恐等领域广泛应用的超宽带穿墙探测技术中，由于在人体微多普勒特征不明显条件下难以识别墙后静止人体目标，提取表征目标散射特性的高分辨率距离像(HRRP)特征成为解决该问题的一个关键技术。对此，本项目首先基于电磁散射理论分析信号接收过程，建立超宽带穿墙雷达信号模型；其次，根据所建立的信号模型，将墙体杂波抑制问题转化为多散射中心回波信号的盲分离和聚类问题，利用独立成分分析和k-means聚类算法对雷达接收信号进行处理以消除墙体环境对HRRP特征提取的影响；最后，根据所建立的信号模型，将HRRP特征提取问题转化为多散射中心参数重构问题，通过设计一字典矩阵，将该参数重构问题引入稀疏表示理论框架之下，通过压缩感知算法对墙体杂波抑制后的信号进行处理以重构目标散射中心参数。本项目的研究将完善基于HRRP特征的墙后静止人体目标识别框架，为应用研究奠定基础。

在该基金项目的资助下，我们研究了复杂目标在超宽带信号照射下的电磁散射特性，建立了超宽带雷达信号模型，获得了雷达信号与多散射中心回波的解析关系；根据所建立的超宽带穿墙雷达信号模型，研究了墙体杂波抑制新方法；根据建立的目标回波模型，研究了高分辨率距离像特征提取新方法；此外，基于HRRP特征，构建训练和测试数据集，研究了利用机器学习算法测试目标识别的准确性。取得主要研究成果如下：利用时域有限差分方法对穿墙雷达探测场景进行建模仿真，并采用盲分离方法对回波数据进行处理，实验结果表明该方法能够在无任何先验信息条件下达到墙杂波抑制效果；通过设计字典矩阵，将HRRP特征提取问题引入到稀疏表示理论框架之下，利用压缩感知算法实现HRRP特征提取，实验结果表明HRRP作为特征的可行性和有效性，该特征的稀疏性可以大大减少目标识别中训练和测试数据量；将目标识别问题重建为回归估计问题，将HRRP作为特征，利用机器学习算法对该回归问题进行训练和测试，实验结果表明将HRRP作为特征无论是在预测精度还是在算法鲁棒性上都优于将目标散射信号幅度和时延作为特征。研究成果已发表SCI学术论文4篇，EI论文12篇，中文核心期刊1篇，申请专利1项。