穿墙雷达多维参数人体姿态识别方法

现有的人体姿态识别方案大多数是从单一的角度来考察人体的姿态特征,但是仅采用距离像很难体现人体关节的位置信息,仅提取微多普勒特征有时会覆盖掉径向速度不明显的特征.为此,本文首先利用慢时间-距离像和慢时间-微多普勒谱图构建出人体姿态的三维张量数据集,扩展了人体姿态的特征维度,然后采用改进型瓶颈残差模块构成的神经网络提高了人...

Human Posture Recognition with Multi Dimensional Parameter for Through the Wall Radar

In order to improve the generalization performance of radar target recognition model based on convolutional neural network (CNN), this paper introduces the deep adaptation networks (DAN) method into the high resolution range profile (HRRP) target recognition. In order to further improve the performance of DAN method, the mixed kernel MMD is proposed to replace multi-kernel MMD (MK-MMD) in DAN, and the MMD loss function based on the mixed kernel function is designed. In this paper, the sea clutter obeying Rayleigh distribution is used to interfere with target domain data. In the network model, one-dimensional CNN is used to extract features. The mixed kernel function DAN is used to reduce the difference of feature distribution between source domain and target domain. The experimental results show that compared with the conventional transfer learning method and DAN method, this method can improve the recognition rate of target domain data by about 15% under the influence of sea clutter. It greatly improves the generalization and robustness of the model.