面向智能手机GNSS接收信号视距/非视距分类的机器学习方法

Machine Learning Methods for Line-of-Sight/Non-Line-of-Sight Classification of GNSS Signals Received by Smartphones

ES评分 0

DOI 10.14075/j.jgg.2025.01.005
刊名
Journal of Geodesy and Geodynamics
年,卷(期) 2025, 45(12)
作者
作者单位

1 上海交通大学感知与导航研究所, 上海市东川路800号, 200240;
2 武汉科技大学计算机科学与技术学院, 武汉市黄家湖西路10号, 430065;
3 武汉大学卫星导航定位技术研究中心, 武汉市珞喻路129号, 430079

摘要
现有的GNSS接收信号分类研究主要针对专业级接收机,通过提取信号特征将其区分为视距(LOS)和非视距(NLOS)信号,以实现高精度定位。然而,这些研究中所采用的信号特征和分类方法并不一定适用于智能手机中低成本GNSS芯片的信号特性。本文提出一种基于机器学习的智能手机GNSS接收信号LOS/NLOS分类方法。该方法结合了载噪比、仰角、伪距残差、归一化伪距残差以及伪距残差百分比5个关键指标,并融合基于深度学习的天顶视图语义分割和卫星标定技术,构建精确的训练数据集。在典型城市场景下对多种机器学习模型进行评估,结果表明,所提出的方法在不同型号智能手机上的平均分类准确率达到85.8%。
Abstract
Existing research on GNSS signal classification primarily targets professional-grade receivers, distinguishing between line-of-sight(LOS) and non-line-of-sight(NLOS) signals through the extraction of signal features to achieve high-accuracy positioning. However, the signal features and classification methods used in these studies may not be applicable to the signal characteristics of low-cost GNSS chips in smartphones. This paper proposes a machine learning-based method for classifying smartphone GNSS reception signals into LOS and NLOS. The method combines five key indicators: carrier-to-noise ratio, elevation angle, pseudorange residuals, normalized pseudorange residuals, and pseudorange residual percentage. It also integrates deep learning-based zenith view semantic segmentation and satellite calibration techniques to construct an accurate training dataset. Evaluations of various machine learning models in typical urban scenarios show that the proposed method achieves an average classification accuracy of 85.8% across different smartphone models.
关键词
智能手机; GNSS信号分类; 机器学习; 视距; 非视距
KeyWord
smartphone; GNSS signal classification; machine learning; line-of-sight; non-line-of-sight
基金项目
页码 1246-1252
  • 参考文献
  • 相关文献
  • 引用本文

贾祥龙, 王辰宇, 王牌. 面向智能手机GNSS接收信号视距/非视距分类的机器学习方法 [J]. 大地测量与地球动力学. 2025; 45; (12). 1246 - 1252.

  • 文献评论

相关学者

相关机构