基于双注意力机制的FMCW雷达人体行为识别  

Human behavior recognition with FMCW radar based on dual attention mechanism

作　　者：卓智海[1] 祝文胜 王双龙 ZHUO Zhihai;ZHU Wensheng;WANG Shuanglong(School of Information&Communication Engineering,Beijing Information Science&Technology University,Beijing 100101,China)

机构地区：[1]北京信息科技大学信息与通信工程学院,北京100101

出　　处：《北京信息科技大学学报（自然科学版）》2024年第5期58-66,共9页Journal of Beijing Information Science and Technology University（Science and Technology Edition）

摘　　要：为了提高调频连续波(frequency modulated continuous wave,FMCW)雷达人体行为识别的分类精度和泛化性能,提出了一种基于双注意力机制的特征融合方法。通过设置阈值,对距离-时间谱图和微多普勒谱图中的有效频谱进行提取、拼接后送入AlexNet和VGG16神经网络来提取特征;加入空间注意力和改进的通道注意力模块,丢弃冗余信息,以增强对重要信息的关注,获取更感兴趣的特征进行特征融合分类。实验结果表明,该方法对6种日常人体行为的平均识别准确率高达97.0%。To improve the classification accuracy and generalization performance of frequency modulated continuous wave(FMCW)radar-based human behavior recognition,a feature fusion method based on dual attention mechanism was proposed.Through threshold setting,the effective spectrum in the range-time spectrogram and micro-Doppler spectrogram was extracted and spliced,and then sent to AlexNet and VGG16 neural networks to extract features.Subsequently,spatial attention and enhanced channel attention module were introduced to discard redundant information,enhancing focus on crucial details to gather more interesting features for feature-fusion classification.The experimental results demonstrate that this method attains an impressive average recognition accuracy of 97.0%for six daily human behaviors.

关 键 词：调频连续波雷达 特征融合 通道注意力 空间注意力 人体行为识别 

分 类 号：TN957.51[电子电信—信号与信息处理]