Research On Radar Unknown Target Discrimination Using High Resolution Range Profile Base On Dictionary Learning

In traditional radar target recognition based on high resolution range profile,the target to be identified must participate in model training in order to be correctly identified.However,in practical applications,the target to be identified may be an unknown target that has not participated in the training process,and will be mistakenly identified as a known target.Therefore,it is necessary to identify the library attributes of the target to be identified and confirm whether it is a known target or an unknown target before performing traditional target recognition.Therefore,this thesis studies the unknown target discrimination method based on dictionary learning.The main contents are as follows.1.An unknown target discrimination method based on parameter optimal regularization K-order singular value decomposition dictionary learning is proposed.This method effectively solves the ill-posed of sparse coding problem by adding a sparse coding regular term to the conventional K-order singular value decomposition dictionary learning optimization problem.At the same time,the L-curve is used to automatically set the optimal regularization parameter,which avoids the shortcoming of fixed regularization parameter in the conventional regularized K-order singular value decomposition dictionary learning.This method improves the performance of dictionary learning and further rises the discriminant rate.The simulation results show that when the signal-to-noise ratio is 15dB,the discriminant rate of this method is 9.11%,3.8%and 3.48%higher than that of Gaussian kernel support vector data domain description,conventional K-singular value decomposition dictionary learning and regularized Ksingular value decomposition dictionary learning.2.An unknown target discrimination method based on multi-kernel fusion dictionary learning is proposed.This method realizes the nonlinear sparse representation of target data by introducing the kernel method into dictionary learning.Meanwhile,multiple kernel functions are fused according to the weighted rule of discriminant contribution to improve the learning ability of the dictionary and effectively improve the discriminant performance.The simulation results show that when the signal-to-noise ratio is 5dB,the discrimination rate of this method is improved by 4.83%and 3.66%respectively compared with the discriminative methods such as polynomial kernel dictionary learning and Gaussian kernel dictionary learning.3.An unknown target discrimination method based on multi-kernel fusion discriminative dictionary learning is proposed.This method combines the multi-kernel fusion method with the discriminant dictionary,so that the sparse coefficients have strong discriminative ability,and increase the nonlinear reconstruction performance of the dictionary,thereby improving the discrimination rate of unknown targets.The simulation results show that when the signal-to-noise ratio is 5dB,the discriminant rate of this method is improved by 2.5%and 3.08%respectively compared with the discriminative methods such as polynomial kernel discriminative dictionary learning and Gaussian kernel discriminative dictionary learning.