Research On Key Intelligent Technologies Of Special Missile-borne Recorde

The successful application of artificial intelligence in various fields has provided new ideas for the intelligent development of the bullet-borne recorder,which is a product of the development of weapon testing technology and weapon testing system under the background of informationization of national defense.In the process of some weapons tests,the "non-intelligence" of the bullet-borne recorder causes great trouble to the testing process.In the test process,the total test length is long while the target data appears at random time and the duration is short,most of the data collected and saved by the bullet-borne recorder are invalid data,and these invalid data will do damage to power supply and storage system of the Recorder,as well as the post-processing process after the data is read back,which is the significance of realizing the "intelligence" of the Recorder.The Recorder with intelligent capability can classify and interpret the collected data during the test,reject invalid data,and classify and store the valid target data.In this paper,CNNs with the ability to extract high-level abstract features from low-level data is selected as the intelligent method,and FPGA,a mobile chip commonly used in weapon test systems,is selected as the platform for intelligent implementation.The subject of this paper,FPGA-based CNNs deployment and application is proposed.Firstly,the layer basis and algorithmic advantages of the advanced feature extraction capability of CNNs are investigated,and the feasibility of deploying CNNs on FPGA platforms is analyzed by combining the computational characteristics of FPGAs and HLS,a large-scale circuit design technology on FPGAs.Secondly,the type of CNN model and network structure parameters are determined according to the task objectives,a suitable hardware experimental platform is selected,and the implementation of 1D-CNN on the hardware platform is designed.Finally,the existing experimental data are used to create a dataset for training and validation of the network model,to determine the effectiveness of this paper and point out the shortcomings of the research,and to make an outlook on the following research methods and objectives.The experiment proves that the work done in this paper provides a reliable solution for the intellectualization of the bullet-borne recorder and is of an important reference value for the development of weapon testing systems in the context of informationization of national denfense.