Research On Neural Network Algorithm For Ballistic Target Recognition Based On Waveform Of Infrared Radiation

Ballistic missile has become an irreplaceable strategic deterrent weapon in the modern military struggle,with its advantages of long range and great power.Therefore,the study of ballistic missile defense technology for the maintenance of national security and stability is of vital importance.To achieve effective penetration,ballistic missile will release a great deal of baits during the midcourse flight process,which causes serious interference and influence on the midcourse anti-missile system.The identification of the warhead and baits in the process of penetration is the key-link in the anti-missile defense system,which is the process of accurately distincting the real target and interference target in the confrontation environment.Infrared Imaging Target System completes the identification by studying the optical intensity difference of different midcourse targets.However,the detector can only obtain a group of point targets with similar motion trajectories because of the performance limits.In this special case,the system can obtain nothing accurate information about the target mechanism but the law of the infrared energy radiated form the ballistic targets group.Considering the reason that the ballistic targets infrared radiation characteristics are susceptible to various factors,and the non-linear characteristic of the gray-time series of the target group is very prominent,so it's difficult for traditional method to obtain the complex relationship between the sequence change information and the target species.It's urgent to use neural networks to extract the target feature information contained in the gray-time series of the target group and simplify the complexity of the recognition system.With the background of the midcourse ballistic missile defense,the dissertation studies the neural network method for ballistic target recognition based on the infrared radiation waveform.The dissertation firstly analyzes the background and the significance of the research,summarizes the status quo of research on the midcourse ballistic target recognition technology and neural network algorithm and introduces both the research content and the structure arrangement.Secondly,according to the different geometrical and motion characteristics of the mid-stage trajectory,the dissertation solves the information of the surface temperature fields of the space targets by using the nodal thermal network method and obtains the infrared radiation characteristics of different ballistic targets,and then gets a large number of target gray time series.Aiming at the classification of the gray time series of ballistic targets,the dissertation draws support from the powerful adaptive ability of neural network by constructing the convolution neural network and the radial basis function neural network model.To learn the full study of the ballistic target group gray time series of different characteristics,the dissertation trains the models with a large-capacity training samples and then completes the task of accurately identificating the warhead and baits.In the aspect of the classification performance improvement of neural network,aiming at the problem that the nonlinear restricted classification ability of the Softmax classifier in the classifier of the convolution neural network,the support vector machines are used to replace the Softmax classifier in the convolution neural network;In order to solve the disadvantage that the model can't obtain the global optimal solution of the radial basis function neural network,RBF neural network identification method based on the Extreme Learning Machine algorithm optimization is proposed,it is ELM-RBF algorithm,which enhances the global optimization ability of the network,thereby improves the network recognition performance.Finally,the dissertation summarizes the full text,discusses the development trend of target recognition technology in the ballistic target defense system and puts forward some suggestions and ideas for the further research direction.The feasibility of the application of convolution neural network and the radial basis function neural network in the ballistic target identification is proved by simulation.At the same time,the experimental results show that the improved algorithm achieves the higher recognition rate on the test samples,and the size of the network is obviously superior to the other algorithms.