基于多任务学习的多人无设备定位方法  

Multi-Person Device Free Localization Method Based on Deep Learning

作　　者：刘天蒙 杨海效 李旋 吴虹[1,3] Liu Tianmeng;Yang Haixiao;Li Xuan;Wu Hong(College of Electronic Information and Optical Engineering,Nankai University,Tianjin 300350,China;Tianjin Key Laboratory of Optoelectronic Sensor and Sensing Network Technology,Tianjin 300350,China;Engineering Research Center of Thin Film Optoelectronics Technology,Tianjin 300350,China)

机构地区：[1]南开大学电子信息与光学工程学院,天津300350 [2]光电传感器与传感网络技术重点实验室,天津300350 [3]光电子薄膜器件与技术研究所,天津300350

出　　处：《南开大学学报(自然科学版)》2025年第1期41-48,共8页Journal of Nankai University(Natural Sience)

基　　金：卫星导航系统与装备技术国家重点实验室开放基金(CEPNT-2021KF-13)。

摘　　要：无设备定位(device free localization,DFL)的研究多为单人定位.针对多人DFL研究少、精度低的问题,提出了一种基于ZigBee接收信号强度指示(received signal strength indication,RSSI)与多任务学习的多人DFL方法.首先介绍了ZigBee无线传感网络的搭建,并分析了ZigBee RSSI数据包的预处理方法,然后介绍了算法设计:使用基于多任务学习框架,构建以ZigBee RSSI为特征的神经网络,两个任务分别为人数预测与位置预测,通过在线阶段的实验验证,计算得到定位任务的分类准确率为97.68%.提供了一种多人DFL的创新思路,在较少网络参数量的情况下达到了较高的定位精度.Research on device free localization(DFL)mostly focuses on localization of single target.Aiming at the problems of few researches and low accuracy of multi-person DFL,a multi-person DFL method was proposed based on ZigBee Received Signal Strength Indication(RSSI)and multi-task learning.The construction of ZigBee wireless sensor network is first introduced,and analyzes the preprocessing method of ZigBee RSSI packet,and then introduces the algorithm design:A neural network characterized by ZigBee RSSI was constructed based on the multi-task learning framework.The two tasks were number of people prediction and location prediction respectively.Through the experimental verification in the online stage,the classification accuracy of the positioning task was calculated to be 97.68%.An innovative idea of multi-person DFL is presented,which achieves high positioning accuracy with fewer network parameters.

关 键 词：无设备定位 深度学习 ZIGBEE 

分 类 号：TN926.23[电子电信—通信与信息系统]