| 刘斌*,黎晨*,史秀纺** ***.基于联邦学习的多源机会信号融合指纹定位[J].高技术通讯(中文),2025,35(5):502~512 |
| 基于联邦学习的多源机会信号融合指纹定位 |
| Federated learning based fingerprinting localization fusing multiple signals of opportunity |
| |
| DOI：10. 3772 / j. issn. 1002-0470. 2025. 05. 006 |
| 中文关键词: 指纹定位； 机会信号； 深度学习； 注意力机制； 联邦学习 |
| 英文关键词: fingerprinting localization, signals of opportunity, deep learning, attention, federated learning |
| 基金项目: |
| 作者 | 单位 | | 刘斌* | (*浙江工业大学信息工程学院杭州 310023）
(**南京理工大学计算机科学与工程学院南京 210094)
(***浙江三锋实业股份有限公司金华 321300) | | 黎晨* | | | 史秀纺** *** | |
|
| 摘要点击次数: 55 |
| 全文下载次数: 114 |
| 中文摘要: |
| 目前基于联邦学习的指纹定位研究主要基于单一信号，未能充分利用空间中广泛存在的无线信号。本文提出了一个基于联邦学习的多源机会信号融合指纹定位系统。首先，设计了一个基于两级注意力机制的深度学习模型融合多种机会信号，包括单个机会信号特征提取和多种机会信号融合，这个模型作为用户本地模型；然后，在模型聚合阶段，考虑到用户之间的数据异质性会降低模型训练性能，提出基于用户本地数据的采样密度和信息熵来确定本地模型在聚合过程中的权重；最后，基于实际采集的数据开展了实验验证，结果表明融合多种机会信号可以有效提高定位精度，并且在用户数据非独立同分布的场景下本文所提出的联邦学习聚合算法优于传统的联邦学习聚合算法。 |
| 英文摘要: |
| Existing researches on federated learning based fingerprinting localization are mainly based on single kind of wireless signal, which have not taken full advantage of the ubiquitous wireless signals. This paper proposes a federated learning based fingerprinting localization system fusing multiple signals of opportunity (SOP). First, a two-level attention based deep learning model is designed to fuse multiple SOP. Specifically, it includes single SOP feature extraction and multiple SOP feature fusion, and this model serves as a local model for users. Then, during the model aggregation process, considering that the heterogeneity of data among users will reduce model training performance, the weight of each local model is determined based on the sampling density and information entropy of the user’s local data. Finally, the experimental verification is carried out based on actual collection of data. Extensive experimental results show that the fusion of multiple SOPs can effectively improve the localization accuracy and the proposed federation learning aggregation algorithm outperforms the traditional algorithms in scenarios where the user data are not independent identically distributed. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
