Studies On The Methods And Application Of Multi-Sources Information Fusion In Uncertain Reasoning

Along with the information sources developing to the direction of diversification and uncertainty, the technique of multi-sources information fusion has been an international research hotspot in the fields of data processing and analysis. At the same time, this kind of multi-resource information perhaps is uncertain, imprecise, un-perfect, even is fuzzy, inconsistent and alterable with time, so in the processing of information fusion, we need to research the denotation and reasoning methods of uncertain information unavoidably. In the paper, just from this point of view, we will introduce 4 kind well-rounded models of imprecise reasoning, then process the uncertain information and apply to the information fusion with these models.In the paper, the models of uncertain reasoning are focused, such as the reasoning model of Bayes probability, Reliability theory, D-S evidence theory and Neural Network. When these models deal with material uncertainty, there are different from basic thought and method in detail, but the reasoning has the same hypostasis and they have the same form of structure, such as the same description of knowledge' uncertainty, the same description of evidence's uncertainty, the same update algorithm of uncertainty and so on. So, we will introduce the difference of these four kinds reasoning models in details, then give some examples to validate these models, finally, we will try our best to describe the same of these models in the paper.In the probability model, we provide the case of client evaluation in the bank credit and utilize the some kinds of probabilities from statistics analysis and field experts to reasoning and information fusion. In the reliability model, based on many papers, we use a subsection function to unite the probability model and reliability model in some degree, explain that if you want to obtain the solid foundation of mathematics, you will lose the point of model which is simple and easy to operate. In the D-S evidence model, we provide two examples, the first case is client evaluation in the bank credit with D-S model, which can be compared with the model of probability. The second case is identification and classification in medical images, which use to explain the applied method. In the Neural Network method, we use the classification in remote-sensing image to explain that the information fusion method can improve the precision of classification.By the experience, we can draw the conclusion that every models of uncertain reasoning in the information fusion have different characters, if these models are considered form different applied areas, the conclusion must be acquired differently. In the paper, some theory discussion and method research is useful and there still have a lot work to do.