无线电工程2025，Vol.55Issue(9)：1894-1906,13.DOI:10.3969/j.issn.1003-3106.2025.09.017
基于WiFi的多用户行为识别技术
WiFi-based Multi-user Activity Recognition Technology
摘要
Abstract
Activity recognition is becoming an indispensable technological support in various facets of our daily lives.WiFi-based activity recognition systems have attracted significant interest for the widespread presence and non-invasive privacy preservation of WiFi devices.Nonetheless,existing WiFi-based activity recognition systems are primarily designed for single-user scenarios and are not suitable for multi-user environments.A novel deep learning framework,Multi-WiAR,is introduced,which enables multi-user activity recognition by isolating mixed signals of multiple users.Firstly,the number of users is identified by using the ResNet-18 model.Then,a separation algorithm based on correlation coefficients is proposed to transform the mixed signals of multi-user activities into single-user signals.Finally,a deep learning network,named WiSen-Net,is developed,which integrates Convolutional Neural Network(CNN)and advanced Res2Net units to realize the accurate recognition of user activities.The experimental results indicate that Multi-WiAR not only shows advantages of low latency and robustness but also achieves a high accuracy rate of 90.37%in triple-user scenarios.关键词
WiFi感知/多用户行为识别/盲源分离/信道状态信息/Res2NetKey words
WiFi sensing/multi-user activity recognition/blind source separation/channel state information/Res2Net分类
信息技术与安全科学引用本文复制引用
魏忠诚,董延虎,陈虹宇,陈炜,连彬,赵继军..基于WiFi的多用户行为识别技术[J].无线电工程,2025,55(9):1894-1906,13.基金项目
河北省自然科学基金面上项目(F2025402017) （F2025402017）
中央引导地方科技发展资金项目(246Z0308G) （246Z0308G）
邯郸市科学技术研究与发展计划项目(21422031288)General Program of Natural Science Foundation of Hebei Province(F2025402017) （21422031288）
Central Government-guided Local Science and Technology Development Funds(246Z0308G) （246Z0308G）
Project of Handan Science and Technology Research and Development Program(21422031288) （21422031288）
