基于迁移学习的WiFi RSSI与地磁融合室内指纹定位方法  

Indoor fingerprint positioning method based on transfer learning with the fusion of WiFi RSSI and geomagnetic

作　　者：鹿海州 罗业超 邵尉 晋军 刘杨 冯梦婷 邹航 LU Haizhou;LUO Yechao;SHAO Wei;JIN Jun;LIU Yang;FENG Mengting;ZOU Hang(Army Engineering University of PLA,Nanjing 210007,China)

机构地区：[1]陆军工程大学,江苏南京210007

出　　处：《通信与信息技术》2025年第2期57-60,共4页Communication & Information Technology

摘　　要：为了提升基于WiFi的室内定位精度,提出了一种基于迁移学习的WiFi接收信号强度指示(Received Signal Strength Indication,RSSI)与地磁融合的室内指纹定位方法,构建了一种具有时间和空间特性的位置指纹,采用两级分步定位,逐步提高定位精度。首先将位置指纹数据转换为灰度图像。然后利用卷积神经网络(Convolutional Neural Networks,CNN)进行空间特征的提取,实现粗定位。最后基于迁移学习的方法经过CNN处理后迁移至长短期记忆网络(Long Short-Term Memory Network,LSTM)进行时间特性的提取,进一步提升定位精度。实测结果表明,采用的包含时间和空间特性的位置指纹相比于只包含空间特性的位置指纹,定位的平均绝对误差(Mean Absolute Error,MAE)减小了12.9%。迁移学习使得定位MAE相比于迁移前减小了5.9%。融合地磁数据后,相比于仅使用WiFi RSSI进行定位,MAE减小超过30%。To improve WiFi-based indoor positioning accuracy,an indoor fingerprint positioning method based on transfer learning that integrates WiFi received signal strength indication(RSSI)and geomagnetism is proposed.A position fingerprint with temporal and spatial traits is constructed,and a two-level stepwise positioning is adopted to gradually enhance accuracy.Firstly,the position finger⁃print data is transformed into grayscale images,and then the convolutional neural network(CNN)is used for extracting spatial features for coarse positioning.Finally,transfer learning is applied to further enhance accuracy by transferring the processed data to the long short term memory network(LSTM)for extracting temporal characteristics.The results show that including both temporal and spatial character⁃istics in the position fingerprint leads to a 12.9%reduction in mean absolute error(MAE)compared to including only spatial characteris⁃tics.Transfer learning also results in a 5.9%reduction in MAE,and integrating geomagnetic data decreases MAE more than 30%com⁃pared to using only WiFi RSSI for positioning.

关 键 词：指纹定位 迁移学习 WiFi RSSI 地磁 融合 

分 类 号：TN915[电子电信—通信与信息系统]