摘要
本文提出一种基于RFID传感标签和深度置信网络(Deep belief networks,DBN)的人体活动识别技术。首先,设计了一种无源RFID(Radio frequency identification,RFID)传感标签,人体加速度信号存入传感器数据区,标签序列号和硬件版本组成了标签ID可以进行唯一标识。然后从数据中提取初始特征,采用滑动窗口技术对特征进行进一步处理,使其更具鲁棒性,有利于快速的人体活动识别。最后,利用这些特征训练DBN,寻找最优DBN结构,实现人体行为识别。在一个可穿戴传感器数据集上进行实验,仿真结果表明,所设计的传感标签最小灵敏度约为-17 dBm,对应在2 W的阅读器功率下传感标签最大工作距离为10.5 m;所提出的DBN算法优于其他算法,也极大地提高了识别准确率。
This paper presents a human activity recognition technology based on RFID sensor tag and deep belief networks(DBN).Firstly,a passive radio frequency identification(RFID)sensor tag is designed.The human acceleration signal is stored in the sensor data area.The tag ID can be uniquely identified by the tag serial number and hardware version.Then the initial features are extracted from the data,and the sliding window technology is used to further process the features to make them more robust and conducive to rapid human activity recognition.Finally,these features are used to train DBN,find the optimal DBN structure and realize human behavior recognition.The simulation results show that the minimum sensitivity of the designed sensor tag is about-17 d Bm,and the maximum working distance of the sensor tag is 10.5 m at 2 W reader power.The proposed DBN algorithm is superior to other algorithms and greatly improves the recognition accuracy.
作者
阳丽
邓芳明
YANG Li;DENG Fangming(School of Information and Computer Engineering,Pingxiang University,Pingxiang Jiangxi 337000,China;School of Electrical Engineering and Automation Engineering,East China Jiaotong University,Nanchang Jiangxi 330013,China)
出处
《电子器件》
CAS
北大核心
2021年第5期1274-1280,共7页
Chinese Journal of Electron Devices
基金
国家自然科学基金项目(51767006)
江西省重点研发计划项目(20181BBE50019)
江西省应用研究培育计划项目(20181BBE58015)
江西省教育厅科学技术项目(GJJ170378)。
关键词
射频识别
深度置信网络
传感标签
加速度
人体活动识别
radio frequency identification
deep belief networks
sensing label
acceleration
human activity recognition