Research On Radar HRR Target Recognition Based On Statistical Modeling

High range resolution complex echo is the coherent summation of projection vectors of the complex echoes from target scatterers along the radar line of sight(LOS),which contains target size,scatterers dist ribution,etc.,thereby high range resolution com plex echo has advantages in obtaining and computation.So it gradually becomes attractive in radar target recognition field.Various kinds of statistical recognition m ethods based on Bayes theory are presented recently.But the presented statistical recognition methods mainly use high range resolution profile which is the m odulus of high range resoluti on complex echo,then it will resu lt that the recognition performance of radar tar get recognition system is limited.Actually,the phase of high range resolution complex echo sample as well as amplitude,also reflects the target structure infor mation.In order to utilize phase inform ation to im prove radar tar get recognition performance of high range resolution complex echo,this article takes adaptive Gaussian classifier(AGC)model and factor analysis(FA)model for instance to generalize Gaussian statistical models to complex domain to build complex m odels.Furthermore,to enhance the recognition perform ance in lo w signal-to-noise ratio(SNR)conditions,a noise-robust modification algorithm is introduced.It solves the mismatch problem between complex Gaussian models and noisy test signals.Traditional statistical recognition m ethods ba sed on Bayes theory are the m aximum correlation coefficient(MCC),principal com ponents analysis(PCA),the above AGC and FA model,etc,.When the training samples are enough,it can achieve good performance to use these statistical recognition m ethods for m odeling.But it can’ t achieve good performance with small training data size.A multi-task factor analysis(MTL-FA)model is proposed to characterize the FFT-magnitude feature of complex HRRP and the inspiring recognition performance is achieved with sm all training data size in Journal Electronics and Information Technology.However,the MTL-FA model as well as the other traditional statistical recognition methods,is unsuperv ised m odel and does no t utilize the label information which is benefit for recognition.For that reason,this article introduces a label aided factor analysis(LA-FA)model.The LA-FA model combines the label inf ormation with m ulti-task learning under Bayes fram e,a nd im proves the recog nition perform ance with small training data size.