基于稀疏时间-频率-调频率表示的雷达目标微动特征提取

The micro-motion feature plays an important role in the field of radar target detection and recognition. Most of the existing micro-motion feature extraction methods are based on the micro-Doppler model, and the time-frequency representation (TFR) is the main signal analysis tool, which is facing with a lot of challenges in the processing of complicated scenarios and complicated micro-motion targets. The main-body interference of the non-rigid body target, the unknown micro-motion forms, the coupling of the translation and the micro motion and the combination of these cases, lead to more complicated modulation properties of the radar echoes of the micro-motion targets. As a result, the performance of the micro-motion features extraction decreases dramatically. In addition, the TFR cannot simultaneously satisfies the requirements of the representation accuracy and the resolution of multiple components, which has limitations on processing overlapped multicomponent micro-motion signals. To solve the above problems, we have to explore novel modulation laws and develop novel signal representation tools. Aiming at solving the aforementioned problems, this project will focuses on the following three directions: 1) time-frequency-frequency-rate modulation models for radar targets with typical micro-motions, to uncover novel micro-motion modulation laws; 2) multicomponent signal representation based on sparse time-frequency-frequency rate representation, to improve the representation performance of the time-varying multicomponent micro-motion signals; 3) micro-motion discrimination and feature extraction based on the time-frequency-frequency rate sequence, to achieve robust micro-motion feature extraction with the novel modulation properties and signal representation method.

雷达微动特征在目标探测与识别领域中具有重要的作用。已有微动特征提取方法主要建立在微多普勒模型的基础上，以时频表示作为主要分析工具，在处理复杂观测场景及复杂微动目标上面临诸多挑战。非刚体目标主体回波的干扰、微动形式的未知性、平动与微动相耦合以及这些情况的组合，使微动目标雷达回波呈现出更加复杂的调制特性，降低微动特征提取性能。另外，时频表示难以同时满足表示精度与多分量分辨的要求，在处理交叉的多分量微动信号上性能明显不足。雷达微动特征提取亟需挖掘新的调制规律，研发新的信号处理工具。本项目针对上述问题进行研究，包括三方面科学问题：一是建立典型微动目标时间-频率-调频率调制模型，发掘新的微动调制规律；二是突破基于稀疏时间-频率-调频率表示的多分量信号表征，提升时变多分量微动信号的表征能力；三是实现基于时间-频率-调频率序列的微动形式辨识与特征提取，依托新的调制规律和信号表示方法获得鲁棒的微动特征。

雷达微动特征提取是空间目标探测识别的重要手段。目前，以时频分析为代表的非平稳信号处理工具难以同时满足表示精度与多分量信号分辨的要求，在处理交叉多分量信号上性能明显不足。针对上述问题，本项目以稀疏时间-频率-调频率表示为核心，从微动调制规律挖掘、多分量信号表征和微动形式辨识及特征提取三个层面开展系统性的研究。.首先，建立典型微动目标时间-频率-调频率调制模型，通过理论推导和仿真分析给出了典型平动和微动以及弹道目标和直升飞机在三维时间-频率-调频率空间中的具体调制模式。在三维空间中，各散射中心分量信号不发生交叉，且信息相比时频平面增加了一个维度，将提供更多的目标信息。.为实现多分量微动信号的高精度表征，本项目建立了以短时稀疏表示为核心的稀疏时间-频率-调频率表示框架。根据雷达微动目标信号特点，提出了基于改进匹配追踪、改进语义线检测以及YOLOX网络检测的多种瞬时频率-瞬时调频率联合估计方法。仿真表明，相比传统方法，稀疏时间-频率-调频率表示获得的估计精度显著提升。.进一步，本项目提出了K均值聚类、基于模型和全局最近邻关联、基于一阶连续性条件重组、基于卡尔曼滤波的瞬时估计值和形状特征的时序关联4种多分量信号关联方法，能够在低信噪比、非正弦微动等复杂场景中实现多分量瞬时频率和调频率的鲁棒估计，获取目标散射点级特征。.最后，基于获得的瞬时频率和瞬时调频率估计值，提出了序列重构误差和序列特征融合方法实现了目标微动形式辨识与特征提取，验证了稀疏时间-频率-调频率表示用于目标精细信息获取和识别的潜力和价值。.项目研究内容的突破，对进一步提升复杂观测场景及复杂微动目标的表征与处理能力具有重要意义。