测绘学报 ›› 2023, Vol. 52 ›› Issue (5): 738-747.doi: 10.11947/j.AGCS.2023.20220013

• 大地测量学与导航 • 上一篇    下一篇

联合SVM和HMM的水上/水下导航场景感知模型构建

朱锋1,2, 罗科干2, 陈惟杰2, 刘万科2, 张小红1,2   

  1. 1. 湖北珞珈实验室, 湖北 武汉 430079;
    2. 武汉大学测绘学院, 湖北 武汉 430079
  • 收稿日期:2022-01-18 修回日期:2023-03-10 发布日期:2023-05-27
  • 作者简介:朱锋(1989-),男,特聘副研究员,研究方向为多传感器集成与多源信息融合。E-mail:fzhu@whu.edu.cn
  • 基金资助:
    国家重点研发计划(2020YFB0505803);国家自然科学基金(42104021);湖北省科技重大项目(2021AAA010);湖北珞珈实验室专项(220100005)

Hybrid SVM and HMM based navigation context awareness models for overwater and underwater mixed scene

ZHU Feng1,2, LUO Kegan2, CHEN Weijie2, LIU Wanke2, ZHANG Xiaohong1,2   

  1. 1. Hubei Luojia Laboratory, Wuhan 430079, China;
    2. School of Geodesy and Geomatics, Wuhan University, Wuhan 430079, China
  • Received:2022-01-18 Revised:2023-03-10 Published:2023-05-27
  • Supported by:
    The National Key Research and Development Program of China (No. 2020YFB0505803);The National Natural Science Foundation of China (No. 42104021);The Science and Technology Major Project of Hubei Province (No. 2021AAA010);The Special Fund of Hubei Luojia Laboratory (No. 220100005)

摘要: 导航场景感知是智能化PNT的重要特征,更是实现多场景无缝导航定位的基础。本文聚焦水上/水下导航场景,考虑电磁波的衰减程度差异将其细分为水上、浅水、深水3类场景,利用支持向量机(support vector machine,SVM)进行场景分类与识别,在此基础上,引入隐马尔可夫模型(hidden Markov model,HMM)表达导航场景切换,进一步提升场景识别可靠性。本文分别构建了基于结果联合(SVM-HMM1)及基于概率联合(SVM-HMM2)的水上/水下导航场景感知模型。实测分析表明,两种模型能够实现高精度场景感知,SVM-HMM1与SVM-HMM2识别准确率分别为91.36%与95.11%;与单一的HMM和SVM模型相比,联合模型在结果分类与识别上更为稳定,准确率提升约为0.95%~8.46%。

关键词: 智能PNT, 导航场景感知, 水上/水下导航场景, 支持向量机, 隐马尔可夫模型

Abstract: Navigation context awareness is not only an important feature of intelligent PNT (positioning, navigation and timing), but also the basis for realizing multi-scene seamless navigation and positioning. This paper focuses on the overwater and underwater scene, which is divided into three kinds of subdivide scenes: overwater, shallow water and deep water according to the change of GNSS signal characteristics, using support vector machine (SVM) for scene classification and recognition. On this basis, hidden Markov model (HMM) is introduced to express navigation scene switching to further improve the reliability of context awareness. This paper constructs two kinds of context awareness models based on result combination (SVM-HMM1) and probability combination (SVM-HMM2). The recognition accuracy of SVM-HMM1 and SVM-HMM2 are 91.36% and 95.11%, respectively. Compared with HMM and SVM, the combined models are more stable in result classification and recognition, and the accuracies are improved by about 0.95%~8.46%.

Key words: intelligent PNT, navigation context awareness, overwater and underwater scenes, support vector machine, hidden Markov model

中图分类号: