航空学报 > 2026, Vol. 47 Issue (12): 332927-332927   doi: 10.7527/S1000-6893.2025.32927

利用学习机制的多方法融合端到端证据建模

李思远1, 韩德强1(), DEZERT Jean2, 杨艺3   

  1. 1.西安交通大学 电信学部,西安 710049
    2.法国航空航天实验室 ONERA 信息处理与系统所,帕莱索 F -91761
    3.西安交通大学 航天航空学院,西安 710049
  • 收稿日期:2025-10-16 修回日期:2025-11-06 接受日期:2025-12-02 出版日期:2025-12-09 发布日期:2025-12-08
  • 通讯作者: 韩德强 E-mail:deqhan@xjtu.edu.cn
  • 基金资助:
    国家自然科学基金(62473304);国家自然科学基金(U22A2045)

Learning-based BBA modeling approach with multi-method fusion

Siyuan LI1, Deqiang HAN1(), Jean DEZERT2, Yi YANG3   

  1. 1.School of Automation Science and Engineering,Xi’an Jiaotong University,Xi’an 710049,China
    2.Department of Information Processing and Systems,The French Aerospace Lab-ONERA,Palaiseau F -91761,France
    3.School of Aerospace,Xi’an Jiaotong University,Xi’an 710049,China
  • Received:2025-10-16 Revised:2025-11-06 Accepted:2025-12-02 Online:2025-12-09 Published:2025-12-08
  • Contact: Deqiang HAN E-mail:deqhan@xjtu.edu.cn
  • Supported by:
    National Natural Science Foundation of China(62473304)

摘要:

Dempster-Shafer证据理论是一种用于不确定性建模与推理的理论框架,其中证据建模是关键环节之一。现有证据建模方法各有优劣,如能综合利用则有望达到更优的建模效果。显式地使用多种证据建模方法再融合的效率较低,因此提出了一种基于深度学习的多方法联合端到端证据建模方法。通过训练一个深度网络,学习从训练样本特征到作为广义训练标签的融合证据函数的映射关系,以此实现多方法联合端到端证据建模。在UCI数据集、遥感图像数据集上的实验结果表明:提出的证据建模方法相比于对比的单一证据建模方法,可以达到更优的分类性能。

关键词: 基本信度分配, 证据理论, 深度学习, 模式分类, 数据驱动

Abstract:

Dempster-Shafer evidence Theory (DST) is a theoretical framework for uncertainty modeling and reasoning, in which modeling the Basic Belief Assignment (BBA) constitutes a crucial and challenging part. The prevailing BBA determination methods have their own pros and cons, and the joint use of them is expected to provide a better BBA. However, explicitly using several BBA determination methods and combining the BBAs through a specific fusion rule is inefficient. To address this issue, we propose a learning-based BBA modeling approach with multi-method fusion. A deep network is trained which learns the mapping from the training samples to the comprehensive BBAs obtained by jointly using the prevailing BBA modeling methods as the generalized training labels. Experimental results on remote sensing image datasets and UCI datasets demonstrate that the proposed method outperforms the individual BBA modeling methods in terms of classification performance.

Key words: Basic Belief Assignment (BBA), evidence theory, deep learning, pattern classification, data-driven

中图分类号: