摘要
随着WiFi基础设施的广泛部署,使得WiFi成为继RSSI(接收信号强度指标)之后强大的无线传感介质。通过对CSI(信道状态信息)的收集和处理,可以逐步描摹人类活动。然而,这种WiFi感知的解决方案也伴随着识别规模扩大、识别准确率极度下降,以及CSI信息固有的与域(环境和方向)相关的问题。论文提出了一种基于残差网络的深度学习WiFi感知方法。首先,通过两个设计的残差块自动提取预处理后的CSI信号特征,解决了细腻度手势特征提取困难和特征不足的问题。其次,网络结构简单,识别单个用户仅需0.59389ms。最后,我们在两种环境中验证了模型的有效性。在简单的办公环境和复杂的标准实验室场景下,2人~8人的识别准确率分别达到97.86%~100%和97.2%~99.35%。在跨域识别方面,论文提出的模型对预处理后的信号域敏感度较低,在复杂的实验室环境下,15人的分组仍然可以达到92%以上。
With the extensive deployment of WiFi infrastructure,it makes WiFi gradually become a powerful wireless sensing medium after RSSI(received signal strength indicator). Through the collection and processing of channel state information(CSI),human activities can be gradually described. However,this solution of WiFi sensing is also accompanied by problems such as the expansion of recognition scale and the extreme decline of recognition accuracy,as well as problems related to domain(environment and direction)inherent in CSI information. In order to solve these problems,this paper proposes a deep learning WiFi sensing method based on residual network. Firstly,the preprocessed CSI signal features are automatically extracted by two designed residual blocks,which solves the problem of difficulty and insufficiency of fine gesture feature extraction. Secondly,the network structure is simple,and it only takes 0.59389ms to identify a single user. Finally,the effectiveness of the model is verified in two environments.In a simple office environment and a complex standard laboratory scene,the recognition accuracy of 2~8 people reaches 97.86%~100% and 97.2%~99.35% respectively. In terms of cross domain recognition,the model proposed in this paper is less sensitive to the domain of the preprocessed signal,and the grouping of 15 people can still reach more than 92% in a complex laboratory environment.
作者
郭梦丽
许勇
何昕
陈锡敏
李凤妹
方群
GUO Mengli;XU Yong;HE Xin;CHEN Ximin;LI Fengmei;FANG Qun(School of Computer and Information,Anhui Normal University,Wuhu 241002)
出处
《计算机与数字工程》
2022年第10期2252-2258,共7页
Computer & Digital Engineering
基金
国家自然科学基金面上项目(编号:62072004)
国家自然科学基金青年基金项目(编号:61702011)资助。
关键词
WiFi感知
信道状态信息
残差网络
非接触式人体识别
Wi Fi sensing
channel status information
residual network
non-contact human body recognition
