| Whether to obtain timely,accurate,complete and reliable military intelligence for strategies,battles,and tactics is the key to win the modern high-tech naval war.In naval battlefield,the important means to obtain military intelligence are the detection equipment distributed in various platforms such as vessels,aircrafts,submarines and satellites.The complicated electromagnetic environment leads information inaccurate,incomplete and uncertain.In order to obtain effective military intelligence,various information sources in command and control system must be fused to get more valuable information.Target comprehensive recognition is an important basis of situation assessment and other advanced fusion process;it is able to assist weapons to make reliable real-time tactical choices and weapon resources calls;it also can maximize operational effectiveness to meet the needs of auxiliary operations command decisions.Therefore,the research on the technology of marine target comprehensive recognition is very important.The decision-making level target comprehensive identification with advantages of convenient and flexible information processing,less information interaction from the systems and being capable of handling asynchronous information and so on is very popular in the comprehensive target identification field.However,the evidence theory information fusion technology has a strong advantage of the expression,measuring and combination of uncertain information.In order to resolve the problems in the reasonable characterization and effective combination for uncertain information,and utilize the target recognition information from various kinds of information sources comprehensively,this paper analyzes and researches three important problems in decision-making level comprehensive target identification based on evidence theory and proposed the corresponding solution.For the problems where the basic probability assignment function of marine target recognition is difficult to construct,according to the characteristics of the information source identification results,two basic probability assignment function construction methods respectively are proposed in this paper.The first method is based on the basic probability assignment function with Bayes reasoning and single element identification output,which describes the probability co-relationship between causal events by Bayes probability function to obtain the probability of the elements in the identification framework,and construct basic probability assignment by simple support function or consonant support function.The second method is based on the basic probability assignment function with Bayes reasoning and the sequenced element output which records the number of various types of elements of the sequence information with confusion matrix,and computes the similarity between elements of target recognition sequence and the elements in confusion matrix by a new similarity calculation method,constructs basic probability assignment function by simple support function or consonant support function with the probability of each possible target type by the identification training statistics and similarity calculations.The simulation results show that the above two methods are very flexible by which a higher quality of basic probability assignment function can be constructed when the prior probability can be obtained.For the problem that the conflicting identification results cannot be processed well by using the evidence theory,based on the evidence conflict measure standard and the evidence combination results evaluation standard,a kind of evidence combination method based on the testing discount repair array is constructed with the evidence subordinate degree concept and evidence discount thought,in which each tentative discount is constructed by the degree of not subordination The reliability of the evidence set and the convergence measure of the result can increase together in discount-testing early time,so the optimal convergence result can be obtained.And by setting optimization goals,the credibility of evidence set can be flexibly controlled.The experimental results and the related analysis indicate that this method is rational and effective to solve the confliction problems of target comprehensive recognition and obtain the high-reliability evidence set and can meet the needs for various decisions.Aiming at the problem that the high calculation complexity in the evidence theory and marine targets recognition real-time requirements are contradicted,we propose a new evidence combination approximation method to improve the marine target real-time recognition.By giving the allocation policy based on focal element information and the redistribution policy of the abandoned basic probability assignment,we propose a new evidence approximation calculation method.The simulation results show that it does not only reduce the number of focal elements,but also can retain as much information as possible contained in the original evidence.Above all,all kinds of methods proposed can effectively improve the comprehensive marine target recognition accuracy and the real-time performance,which can be applied as the basis of engineering applications. |
