The Research On The Online Learning Of Radar Target Recognition Algorithm

Radar is a basic radio detection device,which has the ability of all-weather,day/night and long range detection capabilities.Therefore,radar plays an important role in military and civilian fields.With the development of radar technology,radar automatic target recognition technology has been widely concerned in the world.Most of the traditional radar automatic target recognition algorithms are designed and worked in the batch learning model.However,in practice,we often encounter the excessive data set and the dynamic increased data set,which leads to a large amount of time and space resources for batch learning.It poses a huge challenge to the traditional radar automatic target recognition algorithms.This thesis mainly focuses on the online learning of radar automatic target recognition algorithms.The main contents of the thesis are summarized as follows:1.According to the target classification problem of military and civilian vehicles,the method,based on feature extraction and Support Vector Machine(SVM)online learning algorithm,realized the classification of synthetic aperture radar(SAR)images of military and civilian vehicles.We first use the improved Spectrum Parted Linked Image Test(SPLIT)algorithm to perform scattering center detection,scattering center feature extraction and scattering center classification on the target SAR image.And then the scattering center type distribution feature of military and civilian vehicle targets can be obtained by counting the scattering center's types.At last,we use SVM online learning algorithm to classify military and civilian vehicle targets.Experiments show that the feature extracted by this method have robust recognition performance.It also shows that the SVM online learning algorithm not only ensures the correct classification rate,but also improves the learning efficiency and reduces the required storage space.2.For the problem of spatial target recognition,the data acquisition of spatial targets has periodicity and discontinuity,so long-term observation and accumulation are needed.The traditional spatial target recognition methods need to save all observation data,and all data needs to be re-learned when new data is acquired,so it takes a lot of time and space resources.In view of the above shortcomings,this paper proposes an online Adaptive Gaussian Classifier(AGC)based on statistical model and online learning method.The method firstly constructs an initial recognition template by using the existing high resolution range profile (HRRP)data of the spatial target.And when the new data is acquired,the new data is used to update and expand the recognition template online.The experimental results show that the recognition performance of online AGC is not much different from that of traditional AGC,and the required time and space resources are less,and the learning efficiency is higher.