Research On Vehicle Target Classification Method Based On Radar Echo

In modern informationized ground warfare,different vehicle targets generally serve different combat missions.Generally,the wheeled vehicles undertake the tasks of urban combat and transportation of materials due to the light-weight and maneuverability.The tracked vehicles often have fierce firepower and high protection,and they are responsible for field attack and fire suppression.The difference of these tasks determines that two vehicles have different degrees of threat.Therefore,the classification of these two targets is of great significance.Radar is a kind of equipment that uses radio to detect and measure the target.It has the characteristics of long-distance detection,all-day,all-weather,and has been widely used in military and civilian fields.Radar automatic target recognition technology provides strong technical support for occupying an active position in the battlefield.How to effectively improve the accuracy and reliability of target recognition is the focus of current radar automatic target recognition research.Based on the resolution of the radar system,this paper studies the vehicle targets classification methods under the high-resolution wide-band radar and the low-resolution narrow-band radar.The main contents of the paper can be summarized as follows:In the first part,aiming at the problem of narrow-band radar vehicle target classification,a target classification-rejection method based on random forests model is proposed.The target echo collected in the actual environment may have many unsatisfactory situation due to various factors,such as the radar fails to track the target,the echo signal-to-noise ratio is too low since the distance between the target and the radar is too far.The traditional method is difficult to classify effectively.After the ground clutter suppression based on CLEAN algorithm is applied to the target echo,a variety of features based on the micro-Doppler effect are extracted from the Doppler spectral domain,the time spectral domain and the characteristic spectral domain.The random forests model is used to achieve target classification.Meanwhile,the class posterior probability of target is used to evaluate the rejection threshold.The samples with lower confidence will be excluded automatically according to rejection threshold in the process of classification,which can greatly reduce the impact of unsatisfactory data to classification result.Experiments based on measured vehicle target data proved that this method can improve the classification performance compared with the traditional method.In the second part,a feature fusion method based on combined Canonical Correlation Analysis(CCA)is proposed and applied to the classification problem of wideband radar vehicles.Wideband signals contain a wealth of structural information,which can acquire a variety of features,and the effective fusion of these features can further improve the classification performance.Firstly,considering that traditional CCA has insufficient integration in multi-feature fusion,we combine the features in pairs and use the large singular value ratio threshold to control the fusion dimension.Secondly,a variety of one-dimensional signal domain features and two-dimensional image domain features are extracted from the target's High Resolution Range Profile(HRRP)and HRRP sequence images respectively.Finally,the proposed feature fusion method is used in the wideband radar target classification task.Experiments on MSTAR vehicle target data show that this method can improve the target classification accuracy rate compared with the traditional feature fusion methods.