目标识别中面向时域不确定信息的证据推理研究

The need for timely and accurate processing of large amounts of uncertain and possibly incomplete data from multiple dissimilar sources is felt in target recognition, which is foundational to homeland security. Under this background, methodologies for temporal uncertain information fusion are studied in this project to enhance the credibility of target recognition, which plays an important role in target recognition based on information fusion. The main scientific problems in this research are the determination of basic probability assessment (BPA) based on incomplete information, reliability evaluation for temporal uncertain information and the foundation of the information transportation model between neighboring time nodes. By solving these problems, we can get BPA rationally based on temporal uncertain information. The temporal uncertain information also can be evaluated and fused effectively. Thus, scientific decision can be made. Under the assumption of information conservation, the BPA functions can be obtained based on interval evidence theory and intuitionistic fuzzy evidence theory. Based on reliability self-assessment model and importance mutual-assessment model, the reliability and importance of temporal evidence can be dynamically evaluated, respectively. The influence of time and a priori knowledge is also considered in the comprehensive evaluation of temporal uncertain information. Then the information transportation model between neighboring time nodes is founded. After proposing sequential evidence reasoning method based on cooperative fusion, we can get the dynamical decision making model aiming at process awareness. The influence of time, experts’ knowledge and decision maker’s psychological character are all considered in reliability evaluation, information fusion and decision making. Therefore, the recognition algorithm has certain ability of situation awareness. The expect results of this project can enrich the theory of information fusion. Meanwhile, it can provide novel strategy for target recognition, which is of great theoretical significance and practical value.

本项目以目标融合识别为背景，在证据理论框架内对时域不确定信息融合方法开展研究，以增强识别结果的可信性。通过突破非完备条件下的基本概率赋值生成、时域不确定信息动态可靠性评估以及时域不确定信息间的信息传递等关键问题，实现时域不确定信息的合理建模、综合评估和有效融合。项目将基于信息守恒原理，在非完备条件下研究基于区间证据和直觉模糊证据的基本概率赋值生成方法；在综合考虑时间因素和先验知识的基础上，通过建立时域不确定信息可靠性自评估模型和重要性互评估模型，对时域不确定信息进行综合评估；尝试建立不同时间节点间的信息传递模型，研究基于协同融合的序贯证据推理，实现面向过程感知的动态决策。通过融入先验知识、时间因素和主观因素来探索其对时域不确定信息融合的影响机理，提升融合方法智能性和鲁棒性。本项目预期成果在丰富信息融合相关理论的同时，将为目标融合识别技术研究提供新思路和新方法，具有重要的理论意义和应用价值。

本项目对目标识别中的时域不确定信息融合方法进行了深入研究，主要从信息建模、信息评估、信息融合三个方面提出了针对性的解决方案。首先针对信息的统一描述问题，分别在证据理论和直觉模糊集框架内对不确定信息测度进行了研究，提出了有效的距离测度、相似度测度、不确定性测度和知识测度，为不确定信息的统一建模奠定了基础，为信息评估提供了工具。然后，在直觉模糊框架内，对证据可靠性评估问题进行了研究，结合实时可靠性评估和相对可靠性评估，提出了不确定信息综合评估方法，为不确定信息自适应融合提供了支撑。最后，分别提出了考虑决策者时序偏好的时域证据融合方法、基于可靠性评估的时域证据推理方法和基于协商策略的时域证据自适应融合方法，探索了决策者主观因素以及时间因素对时域证据融合的影响规律，通过实验验证了所提方法的有效性，这些方法可以充分体现时域不确定信息处理的动态性特点，可以有效提升融合结果的科学性。