雷达目标的主体或它的一部分组件在进行平动的同时，如果有振动、旋动、自转或摇摆等微运动时，雷达回波会受到相应的、在多普勒中心之外附加的时变频率调制，这种现象被称为是微多普勒效应(micro-Doppler effect)。本文围绕提高基于雷达微多普勒分析的非规则微动目标分类识别性能这一课题，主要研究非规则微动目标的特征提取、数据融合、分类器设计等内容，提升了人体步态分类识别系统和动态手势识别系统的性能。论文的主要工作和创新点如下。 1）以人体步态为例研究了一类非规则微动目标的特征提取、数据融合和分类器设计的问题，提出了一种基于双波段雷达的人体步态分类识别算法。该方法根据人体步态信号在时频平面分布的特点提取经验特征，数据融合后输入到SVM分类器中识别目标类型。该方法提高了目标识别系统对人体步态的分类识别准确率，在小样本情况下识别性能良好且优于仅用单雷达传感器使用的情况。 2）以动态手势为例研究了另一类非规则微动目标的特征提取和分类器设计问题。根据动态手势信号在时频平面分布的稀疏性提出了一种基于稀疏恢复技术的多观测角度的动态手势分类识别算法。将双波段雷达提取的稀疏时频特征数据融合后用k-平均聚类方法聚类得到中心时频轨迹，并用改进的Hausdorff距离分类器对动态手势进行分类识别，提高了动态手势识别率。该方法分析了在不同观测角度和不同雷达位置设置下的动态手势分类识别性能，提出了一种可以提高动态手势分类识别系统稳定性的双波段雷达与手势位置设置方法。 3）以三维动态手势为例研究了基于干涉信息的非规则微动目标分类识别问题。本文使用一发多收的多基地雷达来观测三维动态手势目标。利用雷达传感器多个接收天线之间的相对位置关系，引入通道信号之间的干涉信息进行动态手势识别。干涉信息与使用传统方法提取到的特征进行数据融合，同时获得目标径向运动和水平方向、垂直方向的运动信息，实现了使用多基地雷达传感器对三维动态手势的分类识别，扩大了动态手势的应用范围。
When the radar target or a part of its components are moving with micro motions, such as vibration, rotation, rotation or swing, the radar echos will be modulated by the corresponding time-varying frequency out of the Doppler center, which is called micro-Doppler effect. This thesis is focused on improving the recognition performance of radar irregular moving targets using micro-Doppler analysis and the main study contents include feature extraction, data fusion and classifier design of irregular moving targets. In this thesis, the performances of human gait classification system and dynamic gesture recognition system are improved. The main work and contributions of this thesis are outlined as follows: 1) Feature extraction, data fusion and classifier design for a kind of irregular moving targets are studied by using human gait classification as an example. The human gait classification algorithm based on dual-band radar sensors is proposed. Based on the characteristics of the time-frequency distribution of human gait signals, this method extracts the empirical features and input the fused features to SVM classifier to identify the target type. This method improves the accuracy of human gait classification in the radar target recognition system. In the case of small sample, the recognition performance is good and better than that of using only single radar sensor. 2) Feature extraction and classifier design of another kind of irregular moving target are studied by using dynamic gesture as an example. The dynamic gesture recognition algorithm with angular diversity based on sparse recovery technology using the sparsity of dynamic gesture signal in time-frequency domin is proposed. After fusing the sparse time-frequency features extracted by dual-band radar sensors, K-means clustering algorithm is used to extract the center time-frequency trajectory, and the modified Hausdorff distance classifier is used to recognize the dynamic gestures, which improves the recognition accuracy of dynamic gestures. This method analyzes the recognition performance of dynamic gestures under different observation angles and different radar sensors position settings, and proposes a dual-band radar sensors and gestures position settings, which can improve the stability of dynamic gesture recognition system. 3) The recognition of irregular moving target based on interference information is studied by using the three-dimensional dynamic gestures as an example. In this paper, a multistatic radar with one transmitter and multiple receivers is used to observe the three-dimensional dynamic gestures. By using the relative position between multiple receiving antennas of the radar sensor, the interference information between different channel signals is introduced to identify the three-dimensional dynamic gestures. The motion information of the target in radial direction, horizontal direction and vertical direction is obtained by fusing the interference information and the features extracted using traditional methods, which can expand the applications of the three-dimensional dynamic gestures.
