Research On Feature Extraction And Classification Of Low-resolution Radar Moving Targets

With the rapid development of modern information technology,radar automatic target recognition has become a key direction of radar development.The low-resolution radar has limited echo information.Although its target recognition can only achieve simple classification of targets,its technology is mature,and most of the active radars in China are low-resolution radars,the radar system still has important military and civilian significance.In this paper,based on the classification and recognition of low-resolution radar moving targets,various features of the target are extracted,and the robust classification of the target is realized by training the classifier.The main work of this thesis is divided into the following aspects:1.According to the preprocessing flow of radar target recognition,the MTI and CLEAN algorithms are studied for clutter suppression.By comparing the clutter suppression effects of the two algorithms,the advantages and disadvantages of the two are analyzed,and the applicable scenarios are given.Then based on the principle of the micro-Doppler effect,the micro-motion of the wheel and the human body is analyzed.2.Based on the principle of micro-Doppler effect,the micro-motions of the wheel and the human body are analyzed,and then the multiple feature quantities are extracted according to the difference of Doppler spectrum shapes of different targets;the energy distribution of the target echo harmonics is utilized.Differently,the feature extraction method based on signal feature spectrum and CLEAN harmonic decomposition is proposed,and the related features such as ratio of maximum eigenvalue to sum of eigenvalues are extracted.After analyzing the principles,advantages and disadvantages of three time-frequency analysis methods,such as short-time Fourier transform,the features of the average time-frequency image entropy are extracted,and the measured data is used to discuss the distribution of the target based on each feature.3.In the actual situation,the low signal-to-noise ratio affects the classification effect of the above features,and thus the feature extraction method based on the target relative area and fractal dimension is proposed.Experiments show that the target relative area and fractal dimension can be effectively under low SNR conditions.Classification target.4.K-nearest neighbor classifier and support vector machine are used to classify the target using single feature quantity and combined feature quantity under different SNR conditions.The classification result shows that the classification result of support vector machine is higher than K nearest neighbor classifier.And the classification effect of a single feature can be improved by appropriately combining feature amounts.By comparing the classification effects of multiple sets of feature quantities,the classification effect of Doppler spectral entropy,time-frequency image entropy and timefrequency energy curve entropy combination is the best,the correct rate is 94.662%,and the correct rate is lower with the decrease of signal-to-noise ratio.It is also gradually reduced.The classification effect of RCS is not affected by the signal-to-noise ratio.Under the condition of low SNR,the classification effect of 100% correct rate can still be achieved,which proves the feasibility and reliability of the feature extraction method discussed in this paper.