Research On Evaluation And Test System Of Deep Recognition And Decision Network

Intelligent recognition and decision-making equipment has a wide and important application in military and civil fields,so its reliability and security are particularly important.Before being put into production,it is necessary to test the two key systems of recognition and decision-making strictly.At present,different intelligent recognition systems have its own specificity and lack of uniform technical specifications.Today,the performance of recognition algorithm tends to be saturated,especially at the application level,the recognition accuracy can not be used as a single index to evaluate different recognition systems.For the intelligent decision system,it is necessary to evaluate the difference between the expected results and the actual results,so as to judge whether the decision meets the requirements.Therefore,this paper studies and implements a evaluation and test system of deep recognition and decision network,which aims to establish a unified system and method of technical specifications as well as standards,to comprehensively evaluate the intelligent recognition and decision system,and then improve its credibility,reliability and security.Aiming at the requirements of the evaluation and test system of deep recognition and decision network,this paper completes overall scheme design of the application architecture and the technical architecture.What's more,this paper designs and analyzes the key technologies.The system adopts browser-server architecture,which is convenient for users to reach through the browser.The main task objects are deep recognition network and intelligent decision network,which provide two functions of algorithm test and data evaluation corresponding to two specific steps of intelligent equipment recognition and decision respectively.Four key technologies are put forward: the research of test case generation method based on data model distribution,the construction of largescale integrated test sample database,the establishment of benchmark algorithm database,and the construction of comprehensive evaluation system and the research of data evaluation method based on three character analysis.Aiming at the characteristics of large amount of data,large specificity of networks,lack of technical standards and simple evaluation index,combined with the technical principle of target recognition based on deep learning,the specific technical scheme and key technologies of the evaluation and test system of deep recognition and decision network are realized.The test case generation based on data model distribution is realized by using the method of independent marking and multi sampling with combination enumeration.XML file is used to realize effective organization of image data and objectives as well as the construction of database and large-scale integrated sample library.this paper duplicate and integrate typical depth recognition algorithms build benchmark algorithm library.In order to realize the comprehensive performance evaluation of the deep recognition and decision network,a comprehensive evaluation system of the deep recognition network and a method of data evaluation of the deep decision network are proposed.Aiming at the stability and reliability requirements of the deep recognition and decision network evaluation system,this paper tests and analyzes the system comprehensively.The function test results show that the evaluation and test system of deep recognition and decision network has the ability to test the recognition and decision network,to evaluate the performance of different algorithms and networks,to evaluate the consistency,robustness and vulnerability of decision system based on the expected results and the actual results,and to provide a good visual effect for the result display.The performance test results show that the system has certain robustness and compression resistance,and meets the design requirements.