Research On SAR Target Recognition With Three Dimensional Electromagnetic Part Model

Synthetic aperture radar(SAR)is an active microwave imaging radar,which use the technology of the pulse compression and synthetic aperture to achieve high spatial resolution in range and crossrang respectively.Compared to the optical and infrared sensors,SAR is able to penetrate vegetation and is not affected by weather and illumination,which makes it can provide all-weather,all-day long investigation on interested target.Therefore,it has been an indispensable means for information acquisition in the areas of remote sensing.However,due to the coherent imaging principle,the sensed SAR data is sensitive to target's pose,configuration,radar parameters and background noise.Subtle changes may lead to some quite different scattering phenomenon.This factors make it hard for the existing method to be used in the actual battlefield.The three dimensional parametric electromagnetic part model(3D PEPM)characterizes the scattering of a target through a set of concise parameters and provides clear physical attributions,which makes it a good candidate for SAR target recogniton.This paper focuses on the research of SAR target recognition based on 3D PEPM,emphasizing how to use 3D PEPM to increasing the SAR automatic target recognition(ATR)performance effectively in the complex environment.The research achivements of this dissertation are listed as follows:1.After the expounding of the research status and challenges in SAR target recognition,the basic problems in 3D PEPM based SAR target recognition are discusses.The development of the parametric electromagnetic model are analysis systematically.The building of parametric electromagnetic model is discussed as two levels,i.e.,how to characterize the scattering of scatterer parametricly and how to characterize the scattering of the whole complex target using the parametric scatterer model.The problems and the existing solutions in SAR target recognition based on 3D PEPM is discussed.2.A 3D PEPM based SAR target recognition framework is proposed.The whole framework is divided into four part:model projection,similarity measurement,azimuth search and decision.This framework tries to make use of the prior information contained in 3D PEPM.Various techniques can be used flexibly to facilitate the target recognition task within this framework.3.For the core problem,i.e.,the similarity measurement between the target model and the test data,the third,the fourth and the fiftch chapter puts forward three methods for similarity measurement respectively.The feasibility and effectiveness of the framework and the proposed methods are verified by experimental results.(1)In the fourth chapter,an attributed scattering center matching method for target recognition is studied.The scattering center feature is relatively stable and can be used as descriptor for target recogntion.First,the scattering centers that are extracted from the test data and the ones predicted from 3D PEPM are matched.Secondly,similarity between 3D PEPM and SAR data is evluatated based on the matching result.Finally,the test target is recognized according to the similarities between 3D PEPMs and the data.(2)In the fifth fourth chapter,the 3D PEPM driven SAR target reocognitoin method is proposed.In this method,the target is recognized according to the similarity between model and data.Here,the similairy is achieved by evaluating the exsitance of each scatterer through a hypothesis testing approach.And similarities from all scatterers are synthesized as a whole similarity between 3D PEPMs and the data.This model driven method is pertinent to the model target which makes it robust with noises and is adaptive to partial occlusion.(3)In the sixth chapter,a 3D PEPM based SAR ATR method using Convolution Neural Network(CNN)is studied.CNN learns hierarchic features from massive training data automatically.It has shown the state-of-the-art power in many classification and recognition tasks.The 3D PEPM is used to simulate the scattering of the target both in the standard operation condition(SOC)and different extended operation condition(EOCs)including part missing,part rotation and different background.Then these simulated images are used as the training data set for CNN.In this method,3D PEPM is manipulated to simulate the scattering of target in varied operation conditions which provides a potential solution for the current ATR problem in the real world.