摘要
针对传统室内定位算法未考虑指纹数据非欧几里德特征的问题,提出一种基于图卷积网络(graph convolutional neural network,GCN)双层特征提取的Wi-Fi指纹室内定位算法(DuGCNLoc)。该算法在接入点(access point,AP)层面通过设计邻接矩阵建立图结构;在参考点(reference point,RP)层面,使用K近邻(K-nearest neighbor,KNN)选取邻近节点构建子图,并通过GCN分别对图结构特征提取,位置预测由全连接层(fully connected layer,FC)完成。实验结果表明,所提算法在自建数据集和公共数据集上的定位性能均优于传统算法,实现了平均定位误差为0.85 m的精度。
To address the issue that traditional indoor localization algorithms overlook the non-Euclidean characteristics of fingerprint data,a Wi-Fi fingerprint based on double-layer feature extraction graph convolutional network(GCN)indoor localization algorithm(DuGCNLoc)is proposed.The proposed algorithm constructs a graph structure at the access point(AP)level by designing an adjacency matrix.At the reference point(RP)level,it builds subgraphs by selecting neighboring nodes using the K-nearest neighbor(KNN)algorithm and extracts graph structural features through a GCN.The location prediction is achieved by a fully connected layer(FC).The experimental results indicate that the proposed algorithm outperforms traditional algorithms in localization performance on both self-constructed datasets and public datasets,achieving an average localization error of 0.85 meters.
作者
康晓非
梁琪悦
李雨玫
KANG Xiao-fei;LIANG Qi-yue;LI Yu-mei(College of Communication and Information Engineering,Xi’an University of Science and Technology,Xi’an 710600,China)
出处
《计算机工程与设计》
北大核心
2025年第8期2157-2162,共6页
Computer Engineering and Design
基金
国家自然科学基金项目(61801372)。
关键词
室内定位
位置指纹
图结构
邻接矩阵
图卷积网络
最近邻算法
接收信号强度
indoor localization
location fingerprinting
graph structure
adjacency matrix
graph convolutional network
K-nearest neighbor
received signal strength
