Research On Radar Hrrp Noise-robust Recognition Methods

High resolution range profile(HRRP)has the advantages of easy acquisition and simple processing,and thus HRRP is an important means to realize radar automatic target recognition(RATR).At present,the HRRP samples used in many theories and methods are obtained through the cooperative target flight test or electromagnetic simulation calculation,and thus HRRP samples usually have a high signal-to-noise ratio(SNR).However,in practical engineering applications,the recognition targets are mostly long-distance targets,and the SNR of the obtained HRRP test samples is often low.At this time,if the template library established under the condition of high SNR is directly used to identify the low SNR test sample,the SNR mismatch exists between the training sample and the test sample,resulting in a decrease in recognition performance.In addition,for non-cooperative goals,high SNR training samples are also difficult to obtain.If the recognition model is trained directly by the low SNR sample,the recognition performance will be seriously affected by the SNR mismatch.In view of the above problems,the thesis has carried out research on HRRP noise robust identification methods.The main work contents are summarized as follows.1.The existing noise robust identification methods are summarized,which are mainly divided into the following three categories:(1)echo enhancement: denoising pre-processing of the received echo data,using the data after noise reduction for identification;(2)feature correction or noise robust feature extraction: extracting features with noise robustness in the training phase or correcting the proposed features,so that the extracted features are not sensitive to noise,and then store the proposed features into the template library for use in recognition;(3)model adaptive correction: using high SNR training sample learning statistical model,the model parameters are corrected according to the SNR of the test sample,and finally the classifier is used for identification.2.The noise robust identification method based on noise matching is studied.The method firstly adds artificial noise(complex gaussian white noise)of different powers to the high SNR training samples to obtain training samples with different SNR.Then,in the training phase,the corresponding recognition template library is established by using different SNR training samples.In the test phase,the SNR of the test sample is estimated and the template matching the SNR is selected for identification.The simulation results verify that the method can effectively improve the recognition performance of the recognition system for low SNR HRRP.3.The noise robust identification method based on adaptive identification of statistical recognition model parameters is studied.Taking the classical factor analysis(FA)and complex factor analysis(CFA)statistical recognition models as examples,for the two typical application scenarios of "high SNR HRRP database low SNR HRRP recognition" and "low SNR HRRP database identification",the corresponding model parameter correction methods are proposed respectively.Simulation experiments verify that the model parameter adaptive correction method can effectively improve the performance of the recognition system under these two application scenarios.4.The method of noise robust identification based on deep neural network is proposed.Directly train the recognition network under low SNR conditions to achieve end-to-end identification of low SNR samples.Simulation experiments show that the proposed method can effectively improve the recognition performance of the recognition system for low SNR HRRP.