在煤矿等非金属矿井坍塌事故中，被困人员往往被厚重煤岩介质遮蔽，人体生命信号微弱且难以探测，因此亟需实现对被困人员生命状态的非接触式探测。超宽带（UWB）雷达凭借其强穿透性和高分辨率，为复杂遮蔽环境下的生命探测提供了新的技术手段。本文基于UWB雷达技术，首先剖析了UWB雷达信号的传播特性，开展了基于UWB雷达的动态人体信号探测方法研究，建立了适用于煤岩介质环境下人体目标识别的数学模型。基于多普勒效应、二维傅里叶变换和贝塞尔函数等理论工具，系统分析了人体运动等生命信号在雷达回波中的表现特征，揭示了生命体征信号在快时间与慢时间域中的分布规律，为后续信号提取和识别奠定了理论基础。

        其次，针对矿井救援场景中存在的强杂波干扰、信号混叠及背景噪声等问题，本文提出了一系列高效的雷达信号预处理方法，包括指数加权法、距离补偿和噪声抑制技术，以有效削弱由遮蔽物和多径效应引起的干扰。在此基础上，研究了EMD、EEMD及ICEEMDAN算法在雷达信号分解中的适用性，并通过仿真对比分析了三种算法在生命信号提取与杂波抑制方面的性能。结果表明，ICEEMDAN算法能够有效提升信号的提取精度，为矿井救援环境下的目标识别提供了可靠的技术支撑。

         此外，本文研究了相邻人体目标的回波信号特性，并发现当目标间距较小时，传统方法易导致信号混叠。针对这一问题，提出采用线性调频Z变换（CZT）进行频谱细化，以提高多目标检测的频率分辨率，并结合恒虚警（CFAR）检测技术，通过自适应门限调整提升低信噪比环境下的目标识别精度。为进一步优化识别效果，本文引入图像域后处理技术，利用形态学运算去除孤立虚警点并优化目标区域边界，从而有效提升UWB雷达在复杂介质环境下的多目标检测与分离能力。

         最后，本文搭建了UWB雷达实验平台，对系统在不同介质穿透条件下的人体动态目标探测能力进行了验证。实验采用400MHz带宽的UWB雷达系统，结合探测软件进行实时数据采集与信号处理。实验涵盖单目标运动、相邻双目标运动及多介质穿透等场景。结果表明，在穿透0.5m碎煤层及0.3m砖混墙的情况下，采用ICEEMDAN算法能够有效提取人体运动信号，并与理论预测一致；在相邻目标实验中，结合CZT频谱细化和CFAR检测方法，即使目标间距较小、运动频率相近，也能实现准确识别与分离；在多介质穿透实验中，分析了黄土、沙土和砖块等介质对信号传播的影响，发现松散介质对信号衰减较小，密实介质中信号衰减较大。实验结果验证了本文提出的信号处理方法在提升UWB雷达人体目标识别与分离能力方面的有效性，为矿山救援中被困人员的生命信息检测提供了坚实的技术支撑。

         In collapse accidents occurring in coal mines and other non-metallic underground mines, trapped individuals are often obscured by thick coal and rock media, making human vital signs extremely weak and difficult to detect. This creates an urgent need for non-contact detection methods to assess the life status of buried victims. Ultra-Wideband (UWB) radar, with its strong penetration capability and high resolution, offers a promising solution for life detection in such complex, obstructed environments.This study, based on UWB radar technology, first analyzes the propagation characteristics of UWB signals and develops a dynamic human signal detection method suitable for coal-rock environments. A mathematical model for human target identification under such media conditions is established. By leveraging theoretical tools including the Doppler effect, two-dimensional Fourier transform, and Bessel functions, the study systematically investigates how life-related signals such as human motion appear in radar echoes. It further reveals the distribution patterns of these signals in both fast-time and slow-time domains, providing a solid theoretical foundation for subsequent signal extraction and target recognition.

           To address the challenges posed by strong clutter interference, signal overlap, and background noise in mine rescue scenarios, a series of efficient radar signal preprocessing techniques are proposed, including exponential weighting, range compensation, and noise suppression. On this basis, the applicability of EMD, EEMD, and ICEEMDAN algorithms for radar signal decomposition is explored, and their performance in life signal extraction and clutter suppression is evaluated through simulation analysis. Results indicate that the ICEEMDAN algorithm significantly improves signal extraction accuracy, offering reliable technical support for target recognition in mine rescue environments.

             Furthermore, this study investigates the echo signal characteristics of adjacent human targets and identifies that signal overlapping frequently occurs when targets are closely spaced. To resolve this issue, the chirp Z-transform (CZT) is introduced to enhance spectral resolution for multi-target detection. Combined with constant false alarm rate (CFAR) detection and adaptive threshold adjustment, this method improves target recognition accuracy in low signal-to-noise ratio (SNR) conditions. To further refine detection results, image-domain post-processing techniques are employed. Morphological operations are applied to eliminate isolated false alarms and optimize target boundary definition, effectively enhancing the UWB radar’s capability in multi-target detection and separation under complex medium conditions.

          Finally, a UWB radar experimental platform was established to validate the system’s ability to detect dynamic human targets under different penetration scenarios. A 400 MHz bandwidth UWB radar system was used, coupled with dedicated software for real-time data acquisition and signal processing. The experiments covered scenarios including single target movement, adjacent dual-target movement, and multi-medium penetration. Results show that under conditions involving a 0.5 m thick coal fragment layer and a 0.3 m brick-concrete wall, the ICEEMDAN algorithm successfully extracted human motion signals consistent with theoretical predictions. In adjacent target experiments, the combination of CZT-based spectral refinement and CFAR detection enabled accurate identification and separation even when the targets were close together and had similar motion frequencies. In multi-medium penetration tests, the effects of loess, sand, and bricks on signal transmission were analyzed, revealing that loose media cause less signal attenuation, while dense media lead to greater attenuation. These results confirm the effectiveness of the proposed signal processing methods in improving UWB radar-based human target recognition and separation, offering a solid technical foundation for life signal detection in mine rescue operations.