Radar High Resolution Range Profiles Target Recognition Based On Time Series Model

With the development of radar technology,the resolution of radar is smaller than the target size.In the scatterers model,the echo signal describes the distribution of scatterers along the radar line of sight,which contains a lot of discernible information such as size and structure.Radar high-resolution range profile is widely used in the field of radar automatic target recognition due to its easy to obtain and low computation.The key problem is how to extract and recognize the HRRP data effectively.In this paper,two kinds of time series models based on Recurrent Neural Network are proposed,and the problem of target recognition based on HRRP is discussed.The main contents of this paper include the following three parts:The first part mainly introduces the research status of radar target recognition technology at home and abroad and the development history of deep learning,and introduces the overall content of the article and the details of the data used in the experiment.Then,based on the scattering point model,the basic properties of HRRP echo signal are analyzed,and the problems of target-aspect,time-shift and amplitude-scale sensitivity which have great influence on recognition are analyzed,and the solutions are proposed respectively.Then two traditional classification methods,SVM and LDA,are introduced.Both of them regard HRRP as a whole and ignore the temporal correlation between range cells.Finally,the recognition performance of these two methods is obtained through experiments,and the experimental results are analyzed.The second part mainly considers the correlation information between the range cells in the HRRP data,and divides the HRRP data into the form of sequence by the sliding window method.After that,the basic principle of Recurrent Neural Network and the method of parameter updating is introduced,and the problem of gradient disappearance is discussed.Considering the important role of the target area in recognition,an algorithm based on attention mechanism and bi-directional recurrent network model is proposed.The algorithm can automatically extract the features of HRRP data,and give more weight to the features of the target area,so as to improve the ability of data feature extraction.Finally,the validity of the proposed model is verified by the measured data,and the key parameters in the model are discussed.In the third part,considering that HRRP data is non-stationary signal,the relation between different range cells is not always constant,while the classical Recurrent Neural Network model uses the same weight to extract features at different times,which makes it difficult to learn the weight matrix.In this paper,the weight of Recurrent Neural Network is improved from two-dimensional matrix to three-dimensional weight tensor,and the data segments are clustered by Gaussian Mixture Model.From the weight tensor,the corresponding weight matrix is selected according to the clustering results,so that the input data segments with similar distribution form are modeled by the same weight matrix.The optimal number of clustering elements is calculated automatically by Dirichlet Process.Each weight matrix can model a part of HRRP data,which reduces the difficulty of model parameter learning.Finally,the validity of the model is verified by the measured data,and parameters which affect the performance of the model are discussed.