Study Of Wide-band Radar One Dimensional Range Profile Recognition Method Based On Recurrent Neural Network

Wide-band radar one dimensional range profile is the projection of three-dimensional structure of the target in the direction of radar line of sight.It can represent the structural characteristics of the observed target.It has several advantages such as simple obtaining condition,easy acquisition,less computation and few storage consumption,and it is an important feature for recognition of the observed target.Therefore,the problem of target recognition based on wide-band radar one dimensional range profile has always been a research hotspot in the domain of radar automatic target recognition(RATR).In recent years,recurrent neutral network(RNN)has obtained higher performance rather than the traditional methods in natural language processing and speech recognition based on sequence data.Wide-band radar range profile is used as sequence data representing the target characteristics.Its recognition problem can be well combined with the recurrent neural network model.Therefore,it is of great theoretical significance and application value to research the target recognition method of wide-band radar one dimensional range profile based on recurrent neural network.The main research works of the thesis are as follows:1.The imaging principle of wide-band radar one dimensional range profile is researched,the characteristics of wide-band one dimensional radar range profile are deeply analyzed,and the corresponding preprocessing methods are given in the thesis.Then,some common feature extraction methods in radar one dimensional range profile target recognition are summarized,and the general process of radar range profile target recognition and two traditional target recognition methods which are widely used in radar range profile target recognition are introduced.2.A wide-band radar one dimensional range profile target recognition method based on recurrent neural network is studied.Firstly,the model structure of the recurrent neural network model and the working principle of its computing unit are summarized,and two improved structures of the recurrent neural network unit are introduced.Then,according to the working principle of different computing units,the gradient back-propagation formula corresponding to each computing unit is deduced,which is combined with the parameter optimization algorithm of the recurrent neural network model to form the training algorithm of the recurrent neural network model.Finally,the workflow of target recognition system based on recurrent neural network is established,and the radar range profile target recognition system is realized according to the workflow.The recognition performance and noise robustness of target recognition system are tested and analyzed in the measured dataset,and the performance is compared with that of traditional target recognition methods.The experimental results show that the wide-band one dimensional radar range profile target recognition method based on recurrent neural network surpasses the traditional target recognition method in recognition performance and noise robustness.3.The structure design of complex recurrent neural network and its designing method are studied.Firstly,the regularization method of neural network model is introduced into the recurrent neural network,and the specific implementation process is given.Then,from the perspective of context synthesis of sequence information and depth extraction of sequence information features,three complex network structures for target recognition are designed,which are bidirectional recurrent neural network,multi-dimensional recurrent neural network and multi-dimensional and bidirectional recurrent neural network,respectively,using the structural flexibility of the recurrent neural network model.Finally,combined with the design of target recognition model and the regularization method of the model,the recognition performance and noise robustness of various improved recurrent neural network structures are tested and analyzed in the measured dataset.The experimental results show that a reasonable structure of recurrent neural network is helpful to improve the performance of target recognition system.