Radar Target Recognition Based On Sparse Representation And Dictionary Learning

Radar target recognition technology is deeply studied as one of the key technologies of radar application.The use of high-resolution radar to detect the target can get more detailed information of the target structure.It is a hot research for many years to identify the target through the high-resolution radar signals.The typical high-resolution radar signals are one-dimens ional high resolution range profile and synthetic aperture Radar images.In the framework of the compressed sensing theory,this paper studies the target recognition method of high-resolution radar based on sparse representation and dictionary learning.The main research work is as follows:1.The sparse representation theory and dictionary learning theory are introduced.The sparsity of radar targets is studied and analyzed.Firstly,the sparse representation of the solution model,the solution algorithm and its applications are introduced.Then,the classic al dictionary learning algor ithm is introduced.The sparseness of one-dimensional high resolution range profile and synthetic aperture radar images are studied.2.A least mean square collaborative dictionary learning method is proposed in this paper.This method combines the collaborative representation model to solve the problem of s low processing of traditional dictionary learning methods.In this method,the dictionary learning algor ithm is used to update the sparse coefficient and compress the dictionary size.The dictionary is updated and the residual value is calculated through the regularized projection method,which improves the processing speed and accuracy of the target recognition.3.A dynamic kernel dictionary learning algorithm is presented.This method combines the kernel function model to solve the problem that nonlinear data is difficult to characterize.The method uses kernel function to transform the data into linear ly separable data and adaptively change the sparseness,which improves the processing ability of non-linear data and the accuracy of target recognition.4.A convolutional sparse coding and multi-classifier fusion method is proposed.This method combines the machine learning model to solve the problem that sparse representation has a poor performance on complex issues.The method uses convolutional dictionary learning to calculate the reconstruction error,and through the multi-classifier fusion method to achieve classification.The accuracy of target recognition is improved.