基于WiFi信号分离多用户入侵检测  

WiFi-based Multi-Person Intrusion Detection via Signal Separation

作　　者：王萍 梁修胜 张振亚[1,2] 刘皓然 WANG Ping;LIANG Xiusheng;ZHANG Zhenya;LIU Haoran(Anhui Province Key Laboratory of Intelligent Building and Building Energy Saving,Anhui Jianzhu University,230601,Hefei,Anhui,China;School of Electronics and Information Engineering,Anhui Jianzhu University,230601,Hefei,Anhui,China)

机构地区：[1]安徽建筑大学智能建筑与建筑节能安徽省重点实验室,安徽合肥230601 [2]安徽建筑大学电子与信息工程学院,安徽合肥230601

出　　处：《淮北师范大学学报(自然科学版)》2025年第1期62-67,共6页Journal of Huaibei Normal University:Natural Sciences

基　　金：安徽省高校学科拔尖人才学术资助项目(gxbjZD2021067,gxyq2022030);安徽省重点实验室主任基金项目(IBES2022ZR01,BES2024ZR02);安徽建筑大学校级科研项目(2021QDZ08)。

摘　　要：针对现有入侵检测解决方案大多集中于单人场景,而忽视多人场景问题,设计一种多用户被动入侵检测框架WiMD(WiFi-based multi-person passive intrusion detection)。框架采用经过训练的级联双向长短时记忆网络-全连接(bidirectional long-short term memory network-fully connected,BLSTM-FC)神经网络模型实现从去噪后的多用户信号中分离出单人信号,在此基础上利用小波近似系数进行特征提取,将分离出的信号应用于多个二分类器进行入侵检测判断。结果表明,WiMD平均检测准确率可达96.46%,框架可以有效解决合法人员数量较少的特定安全区域入侵检测问题。In view of the fact that existing intrusion detection solutions mainly focus on single-person scenarios while neglecting multi-user scenarios,a multi-user passive intrusion detection framework called WiMD(WiFi-based multi-person passive intrusion detection)is developed.A trained cascaded bidirectional long-short term memory network-fully connected(BLSTM-FC)neural network model is employed to separate single-person signals from multi-user signals after denoising.On this basis,feature extraction is conducted using wavelet approximation coefficients.The separated signals are then applied to multiple binary classifiers for intrusion detection judgments.Experimental results show that the average detection accuracy of WiMD can reach 96.46%.This framework effectively addresses intrusion detection problems in specific secure areas with a limited number of authorized personnel.

关 键 词：信号分离 双向长短时记忆网络 特征提取 入侵检测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]