Study On Radar Target Recognition Based On High Resolution Range Profile

The echo of the wideband radar signal is projected in the direction of its line of sight,and then the sum of the projection vectors is obtained,and the sum is a High Resolution Range Profile(HRRP).The echo is not only contains information about the shape of the target's scattering point,but also easy to acquire and process.Therefore,HRRP have been favored by many researchers in the field of radar target recognition.The time-frequency feature of the signal are a combination of time and frequency functions and shows the information about the signal changes with time and frequency.Common time-frequency feature include time-frequency representation and time-frequency distribution,and commonly used time-frequency representation tools include Short-time Fourier Transform,wavelet transform,etc.The commonly used method for obtaining time-frequency distribution of signals is Cohen's time-frequency distribution.This paper systematically studies the time-frequency characteristics of HRRP.The linear and nonlinear principal component analysis based on time-frequency characteristics of HRRP is studied.The dictionary learning algorithm based on image recognition is applied to the recognition of time-frequency features based on HRRP,and the recognition effect is a very good.The research content of this thesis and the innovations compared with the current related literature are as follows:1.Study of time-frequency representation of range profile by Short-time Fourier Transform and Wavelet Transform.Analysis of time-frequency distribution of range profile by Cohen's time-frequency distribution algorithm,then studied the instantly characteristics of time-frequency distribution and Hough transform.2.There is a problem that it is difficult to determine the number of intrinsic modal functions and the penalty parameters when the Variational Mode Decomposition(VMD)algorithm extracts the Intrinsic Mode Function(IMF)of range profile.Developed a VMD algorithm based on particle swarm optimization.3.Two-dimensional Principal Component Analysis(2DPCA)algorithm and Bilateral Two-dimensional Principal Component Analysis(B2DPCA)algorithm were used to dimensionality reduction of time-frequency analysis data for recognition,experimental results show that this method can indeed improve the recognition rate.However,these linear principal component analysis algorithms have nonlinear problems in data dimensionality reduction.For solve this problem,This paper introduces the Kernel function thought into the B2 DPCA algorithm then proposes a Kernel Bilateral Two-dimensional Principal Component Analysis(KB2DPCA)algorithm,this algorithm not only achieved better recognition results,but also reduced the amount of calculation.Experimental results prove that KB2 DPCA is better than PCA,2DPCA,and B2 DPCA,it is also proved that the extracted time-frequency features are indeed effective,especially the feature of the wavelet transform of the range image and the time-frequency distribution of the frequency edge features are most conducive to recognition and classification.4.Studying the basic idea of sparse feature based on dictionary learning and introducing the principle of constructing a dictionary based on K-SVD algorithm.Applying the Discriminant K-Singular Value Decomposition(D-KSVD)algorithm and Label Consensus K-Singular Value Decomposition(LC-KSVD)algorithm to the radar target recognition based on the features of time-frequency representation,time-frequency distribution,and instantly feature of time-frequency distribution.The experimental results prove the validity of these methods and further verify that the frequency edge features of the wavelet transform feature and time-frequency distribution are most conducive to recognition.