基于时间卷积网络的CSI动作识别  

Temporal Convolutional Network based activity recognition using CSI

作　　者：阴庚雷 丁文超 张俊宝[1] YIN Genglei;DING Wenchao;ZHANG Junbao(School of Computer Science, Zhongyuan University of Technology, Zhengzhou 450007, China)

机构地区：[1]中原工学院计算机学院,河南郑州450007

出　　处：《中原工学院学报》2020年第5期59-65,共7页Journal of Zhongyuan University of Technology

基　　金：国家自然科学基金(U1504614);河南省教育厅科学技术研究重点项目(14B520055)。

摘　　要：随着WiFi感知技术的发展,基于WiFi信道状态信息(Channel State Information,CSI)的动作识别在未来物联网时代具有良好的发展前景。目前,基于CSI的动作识别研究大部分是提取统计特征进行分类,并未深层次挖掘跨时段的动作特征关系。针对这个问题,提出了基于时间卷积网络(Temporal Convolutional Network,TCN)的CSI动作识别。首先提取原始CSI数据的幅值,之后进行异常点去除、去噪和方差提取;然后通过加入注意力机制的TCN网络提取序列特征,而无需手动提取特征;最后采用Softmax分类器实现动作识别分类。使用WiAR数据集进行验证,在该数据集上动作识别的准确率平均达到了90%。With the popularization of WiFi sensing technology,activity recognition based on WiFi Channel State Information(CSI)has great potential for development in the future of Internet of Things.At present,most of the researches on CSI-based activity recognition extract statistical features for classification,and do not deeply explore the relationship of activity features across time periods.To solve this problem,activity recognition based on CSI using Temporal Convolutional Network(TCN)is proposed.First,the amplitude of the original CSI data is extracted,then the CSI amplitude is subjected to data processing including abnormal point removal,denoising and variance extraction.Then the sequence feature extraction is carried out through the TCN network with attention mechanism,without the need of manual extraction of features.Finally,the Softmax classifier is used to recognize the activity.The WiAR dataset is used for verification,and the average accuracy of activity recognition on this dataset reaches 90%.

关 键 词：信道状态信息 时间卷积网络 动作识别 注意力机制 

分 类 号：G642[文化科学—高等教育学]