High Resolution Range Profile Sequences Recognition Based On Temporal Restricted Boltzmann Machine

High Resolution Range Profile(HRRP)is widely used in radar target recognition for its easy acquisition and no motion compensation.However,there are many problems to be solved in the traditional HRRP recognition algorithm.First of all,most of the traditional recognition methods only use a single HRRP for recognition,rarely convert one-dimensional range profile into a sequence of targets for recognition,which makes the space-time correlation between adjacent frames of HRRP sequence not be utilized and is a loss of information and a waste of resources.Secondly,because of the HRRP's pose sensitivity,it is difficult for traditional methods to extract robust features.Finally,the performance of traditional methods in low SNR environment is very poor,and can not adapt to the actual environment of radar.In recent years,under the background of big data,the radar target recognition method based on "deep learning" has achieved great success.Compared with the traditional methods of computer vision,the deep learning method can make more effective use of a large number of training data and get a higher level of expression.This paper focuses on the existing problems in these three areas of HRRP recognition,and proposes effective solutions.The research results will provide new ideas and support for HRRP sequence recognition.The first chapter is the introduction,which describes the research background and significance of this paper,summarizes the main research methods and progress of target recognition,analyzes the main types of target feature extraction and target recognition based on HRRP and classification algorithm based on HRRP,and points out the advantages and disadvantages of each method.Finally,the restricted Boltzmann machine model and its application in target recognition are briefly reviewed.Potential advantages of restricted Boltzmann machine in sequential target recognition are pointed out,which points out the direction of research.In the second chapter,the temporal model is used to analyze and identify HRRP sequences,and the temporal and spatial correlation between HRRP sequences is utilized.Aiming at the problem that traditional recognition methods can not effectively use sequence information,this chapter regards HRRP sequence as one-dimensional sequence data on the range unit and directly takes it as input for recognition task.The temporal restricted Boltzmann machine model is applied to HRRP sequence recognition task,which makes the spatial-temporal correlation between adjacent frames of HRRP sequence be effectively utilized.In the third chapter,the RTRBM model is applied to the HRRP sequence recognition task to model the high-dimensional data.In order to solve the problem of complex temporal restricted Boltzmann machine model and large amount of computation,this chapter applies the recursive temporal restricted Boltzmann machine model to HRRP sequence recognition,and solves this problem and achieves a higher recognition rate,for the current state of the recurrent temporal restricted Boltzmann machine model is only related to the node state of the previous time.Unlike the temporal restricted Boltzmann machine model with time series constraints,the current state of the model is related to the state of the previous m times.In the fourth chapter,an attention-based RTRBM model is proposed and applied to HRRP sequence recognition tasks,which improves the recognition rate and has strong anti-noise ability.Aiming at the problem that traditional methods are difficult to deal with high-dimensional data and low signal-to-noise ratio data,this chapter introduces the attention mechanism into the recurrent temporal restricted Boltzmann machine model and proposes a recurrent temporal restricted Boltzmann machine model based on the attention mechanism.The model uses the recurrent temporal restricted Boltzmann machine model to extract the sequence information of HRRP and store it in the hidden layer vector.Attention mechanism can assign more weight to the features which contribute more to the recognition.This makes the model pay more attention to the features which are more separable.It solves the problem of high-dimensional data processing and makes the proposed model more suitable for low signal-to-noise ratio environment,which is more similar to the actual working environment of radar.The fifth chapter is summary and prospect.This chapter summarizes the main work and innovations of this paper,and prospects the recognition of HRRP sequence under the condition of small sample and non-uniform sampling,and the application of deeper network and attention mechanism to HRRP sequence tasks.