摘要
随着Wi-Fi设备的普及,Wi-Fi传感在医疗、家庭安全和监控等方面实现了广泛的应用,相对于传统摄像头,监控摄像头能够更好地保护使用者的隐私。目前,深度学习技术在人类活动识别领域的作用越来越重要,深度学习技术能从Wi-Fi数据包收集的信道状态信息(CSI)对事件进行分类。然而,大多数模型需要大量CSI数据提供支持,且模型在不同环境测试时的准确性较差。为了应对这个问题,提出了一个基于原型网络的跨环境信道状态信息人活动识别模型(HYHCPNet)。它通过利用已训练环境与未训练环境特征的差异,有效地提高了模型在不同环境中的准确率。实验结果表明,所提方案显著优于最先进的人体活动识别方法,实现了更高的识别精度和更少的训练时间。
With the popularity of Wi-Fi devices,in medical,home security,and surveillance fields,traditional cameras Wi-Fi sensing is widely used.Compared to traditional cameras,surveillance cameras better protect users’privacy.At present,the role of deep learning technology in human activity recognition is becoming increasingly important.Deep learning technology can classify events based on channel state information(CSI)collected from Wi-Fi packets.However,most models require a large amount of CSI data to provide support.These models have poor accuracy when tested in different environments.To address this issue,cross environment channel state information human activity recognition model based on prototypical network(HYHCPNet)was proposed.The accuracy of the model was effectively improved in different environments by utilizing the differences in features between the trained and untrained environments.Experiments show that the proposed scheme significantly outperforms state-of-the-art human activity recognition methods,achieving higher recognition accuracy and less training time.
作者
陆许明
黄缘昊
陈翔
LU Xuming;HUANG Yuanhao;CHEN Xiang(School of Electronic and Information Engineering,Wuyi University,Jiangmen 529020,China;School of Electronic and Information Engineering,Sun Yat-sen University,Guangzhou 510800,China)
出处
《通信学报》
北大核心
2025年第S1期121-127,共7页
Journal on Communications
基金
2022年度广东省教育厅广东普通高校重点领域专项基金资助项目(No.2022ZDZX1033)
2025年度五邑大学广东省百校联百县助力“百县千镇万村高质量发展工程”行动基金资助项目(No.BQW2025011)。
关键词
无线通信
信道状态信息
人体活动识别
少样本学习
wireless fidelity
channel state information
human activity recognition
few shot learning
