Multi-source Information Fusion Based On Data-driven

Multi-source information fusion can be defined as theories and methods whichare used to research and make full use of the indeterminate multi-source information.It has broad application prospects in air or land traffic control, process industrysupervision, military command system, battlefield surveillance and other applicationareas. Most traditional information fusion methods depend on system models, wherecertain simplification will be introduced. However, with increased complexity ofapplications, these models tend to be inadequate and show bias to the real situation.Even precise models are just impossible to build up in some cases. This paper focuseson the research of multi-source information fusion using the thought of data-driven.The main work and achievements are as follows:First of all, an overview on present research situation of information fusion andits inadequateness is presented. Also certain fundamental theories including someinformation fusion methods at different levels and several common target trackingalgorithms are introduced.Secondly, due to drawbacks of traditional multi-source information fusion, thispaper proposes a data-driven based information fusion method, which is divided intotwo different implementation ways, named dynamic data-driven information fusionand information fusion based both on data-driven and model-driven respectively. Theanalysis of the characteristics and applicability has been given as well.Thirdly, the research of an information fusion method, in which data-driven andmodel-driven are associated, is theoretical practical detailed analyzed. And thenchoose acoustic vehicle classification, where a data-driven feature set and amodel-based feature set are combined, show that the performance of informationfusion is improved due to a compensation deficiency for model-based approaches.The feasibility and advantages to introduce data-driven ideas into information fusionare verified.Finally, an application of the joint tracking and classification in the basis of priorJTC framework using multiple model particle filter algorithm is realized, whichprovides credible tracking and accurate classification. However, this technique hasintrinsical weakness. To solve it, a probabilistic argumentation system joint tracking and classification algorithm(PAS-JTC) is proposed based on data-driven. ThePAS-JTC can obtain more reliable type discrimination so as to promote better tracking.The data-driven based JTC brings new development to traditional joint tracking andclassification.