Research On Credibility Assessment Method For Embedded Target Classification Decision System

Whether in the military field or in the civilian field,the embedded target classification decision-making system has a very wide range of applications.An accurate decision can only be made if the target classification is accurate.Only by making accurate decisions can people be assured of using intelligent systems to assist in combat and life.Therefore,the reliability of the embedded target classification decision-making system becomes particularly important.But at present,there is not a complete system for the credibility evaluation of the embedded target classification decision-making system,the evaluation indicators are relatively scattered,and there is no evaluation platform that is convenient for users to manage.Therefore,this research aims to establish a set of credibility evaluation system for the target classification decision-making system,study the credibility evaluation method,build a credibility evaluation software platform,and then make the target classification decision-making system more secure and credible.Aiming at the actual needs and project requirements of the reliability evaluation of the embedded target classification decision-making system,this topic designs the overall scheme of the research,and proposes several evaluation models for the reliability evaluation.The research tasks in this paper are classification algorithms and decisionmaking algorithms,and the corresponding data sets are image data and vector data,respectively.The credibility evaluation models established for data characteristics include data evaluation model,generalization capability evaluation model,robust capability evaluation model and autonomous capability level evaluation model.A visual credibility evaluation software platform is built.The platform mainly includes a test case generation module and a credibility evaluation module.A large-scale sample library and algorithm library are established for the concrete realization of the reliability evaluation of the embedded target classification decisionmaking system.Since the credibility evaluation platform needs to be universal,the subject has established a test sample library and an algorithm library,and conducted a lot of experiments on this basis to verify the universality of the platform.The specific research and realization of credibility evaluation is divided into generalization evaluation,robustness evaluation and autonomous ability level evaluation according to the result characteristics.For each evaluation,the corresponding evaluation method is studied.In addition,this paper also builds the credibility evaluation software platform of the embedded target classification decision-making system.This subject uses a variety of data processing methods to generate generalization test cases and robustness test cases of the target classification decision-making system,and uses the data set as the input of the algorithm to obtain the results of network classification or decision-making.Then the results are sent into the established credibility evaluation model to get the corresponding evaluation index results,and finally the whole system is comprehensively evaluated through the index comprehensive analysis evaluation model.After the basic implementation,the evaluation model is embedded in the evaluation software platform through Py Qt5 as a technical tool,thus completing the overall process of the credibility evaluation of the embedded target classification decision-making system that can be analyzed in multiple dimensions.Aiming at the functional requirements and efficiency requirements of the credibility evaluation platform,firstly,the test environment is built,and the credibility evaluation system is built.Then,the software platform is comprehensively tested and the results are analyzed comprehensively.The test results show that all indicators completely cover the preset system technical requirements and meet all functional indicators required for reliability assessment.At the same time,from the user’s point of view,the trustworthy software platform has a good visualization effect,does not require excessive hardware configuration requirements,is very user-friendly,and meets the design requirements.