Target Recognition Method, A Combination Of Evidence Theory And Neural Networks

With the rapid development of information technology, many kinds of multi-sensor systems facing with complex application background are emerging. The processing of uncertain information has been a research hotspot. Evidence theory, as one of the effective inference methods processing uncertain information, is widely applied to multi-sensor information fusion system. It is applicable only when evidence is independent. But the evidence that needs processing is generally dependent and inconsistent in practice. Therefore, research and improvement on evidence theory is an important subject. The thesis combines evidence theory with neural network technology to deal with the target identification problem in multi-sensor system.The main work and achievements are as follows:Firstly, the existing evidence combination approaches are analyzed. On the basis of research on conflict extent between focal element of different evidences, a combination algorithm of conflict evidence is proposed. The simulation results show that the proposed method is ideal in composition effect and results.Secondly, a kind of target identification system, which can solve the fusion and identification problems of conflict information, is proposed based on evidence theory.The principle and the steps are presented. The simulation results prove its feasibility.Thirdly, considering that it is difficult to obtain Basic Probability Assignment, a method based on neural network technology is presented. It can get over the subjectivity of existing methods that depend on the experience of experts.Finally, the steps of target identification are designed using combination of evidence theory and neural network technology. The network topology of neural network identificator based on evidence theory and the learning formula of network are given. The simulation result testifies the feasibility and validity of the method.The research work of the dissertation is helpful for the development of multi-sensor information fusion and has certain academic meanings and project application values.