基于FMCW雷达的异常行为智能识别检测技术  

Intelligent recognition and detection of abnormal behavior based on FMCW radar

作　　者：孙泽钰 宫新保[1] SUN Ze-yu;GONG Xin-bao(Department of Electronic Engineering,Shanghai Jiaotong University,Shanghai 200240,China)

机构地区：[1]上海交通大学电子工程系,上海200240

出　　处：《信息技术》2025年第2期74-79,共6页Information Technology

摘　　要：为了实现在注重隐私保护场景下的人体异常行为检测,文中基于调频连续波(FMCW)雷达,提出了一种融合残差神经网络(Resnet)和对比学习的异常行为检测系统。该方法通过残差神经网络和对比学习方法对雷达信号的运动特征谱图进行关键特征极子获取,以此作为异常行为识别的判决依据。实验结果表明,该系统对实测数据的检测率达到了98%,对虚警情况有较好的抑制作用,且具有较好的泛化性能。In order to detect abnormal human behavior in scenarios that focus on privacy protection,this paper proposes an abnormal behavior detection system that combines residual neural network(Resnet)and contrastive learning based on frequency modulated continuous wave(FMCW)radar.This method obtains the motion feature spectrum of the original signal through spectral analysis method,and then uses the residual neural network and contrastive learning method to obtain the key feature poles of the motion feature spectrum,and uses a post-processing decision mechanism to identify abnormal behavior.Experimental results show that the system achieves a detection rate of 98%on measured data,has a good suppression effect on false alarms,and has good generalization performance.

关 键 词：异常行为检测 调频连续波雷达 残差神经网络 对比学习 

分 类 号：TN957.52[电子电信—信号与信息处理] TP181[电子电信—信息与通信工程]