Research On Space Target Recognition Recurrent Neural Network Algorithm Based On Infrared Radiation Time Series

Space target recognition is the key technology for the terminal guidance information processing of missile defense system.As the main identification means,infrared target recognition must identify the warhead target accurately and quickly from a large number of warheads,heavy decoys,light decoys,debris and other target groups.The identification process of the infrared targets is to analyze and process the data information acquired by the infrared sensors,and to grasp the physical characteristics(such as temperature,radiation,trajectory,mass and structure)and motion characteristics(such as flight speed,orbit and micro-motion)to describ the target group comprehensively.According to the feature difference of different targets,designed classification algorithm is used to distinguish the key targets and discriminate the space targets.The paper focuses on the following three aspects:First,the infrared characteristics of the spatial targets are analyzed.The physical characteristics,such as the geometry and temperature of the space targets,and the motion characteristics,such as flight orbit and micro-motion,are analyzed.According to the difference and the ability to distinguish the space target,the infrared radiation intensity time seires is established based on the measurement analysis and simulation.The sequence model simplifies the characterization and description of spatial targets with comprehensive and unified one-dimensional time series data.Second,the recurrent neural network space target classifier is designed.RNNs can comprehensively utilize the infrared radiation intensity sequences,and classify and identify various types of spatial targets such as warheads,heavy decoys,light decoys and missile decomposers,thus reducing the dependence on feature extraction methods and constructing an effective target classification and recognition system.Experiments on common time series data and simulated radiation series data show that RNNs can discriminate time series more quickly and effectively than many classical algorithms.Thirdly,the extended recurrent structure algorithm is designed for space target recognition.It is necessary to focus on the task requirements of recognizing the axisymmetric warhead targets and the non-axisymmetric interference targets.Consisting of convolution operation,random projection and bidirectional transmission,BICORN-RNN is proposed to improve the learning ability of complex data and to deal with the limited scenes.The simulation results demonstrate that BICORN-RNNs can effectively classify and identify spatial targets in the case of data length variation,noise pollution,data loss and overlap,which provides a reference for the design of target recognition system in free flight phase of missile defense system.