Research On Polarimetric SAR Target Image Simulation And Recognition Technology

As an important microwave imaging system,Synthetuc Aperture Radar(SAR)has attracted extensive attention due to its characteristics of all-day,all-weather and high resolution.Especially in the past 20 years,SAR systems have also been widely used.SAR can actively detect ground targets on satellites,airplanes and other carriers,and has certain penetrating capability.SAR has important applications in resource exploration,environment detection,military investigation and so on.Compared with the traditional SAR system,polarimetric SAR echo signal contains the polarization characteristics information of the target,which can be used to extract the scatterring type information of the irradiated target and has greater application value in ground object classfication and target recognition.We know that ground object classification and target recognition based on SAR images need relatively complete training data to support algorithm research.If polarization SAR images of targets are acquired by actual measurement,it takes a lot of manpower,material resources and financial resources.Image simulation of polarmetric SAR targets is a simple and effective technical approach to solve this problem.Based on this,this thesis focuses on the research of the image simulation method of polarmetric SAR targets and compiles simulation software,aiming at providing research conditions for target recognition technology based on polarimetric SAR.On this basis,this thesis has carried out some research on feature extraction and target recognition based on simulated target data.The main research contents are as follows:In the first part of the thesis,the basic theory of polarimetric SAR is primarily introduced,including the concept,classification and expression of polarized electromagnetic waves.And then,the expression of polarization characteristics of targets is described.And further more,the theory of SAR imaging is introduced.The content above lays a theoretical foundation for the image simulation and recognition of polarimetric SAR targets.In the second part,the high resolution and fast imaging simulation of polarimetric radar is mainly studied.While simulating the image of SAR targets according to the echo,echo is generated on the basis of the reflection characteristics of the target model and the beamforming algorithm,and then the echo is processed by a SAR imaging algorithm and finally the final image is obtained.The high-resolution SAR imaging algorithm based on ray tracing is proposed in this thesis.According to the radar parameters set,the two-dimensional polarized SAR image can be generated directly from the 3D model.Since the original ray tracing algorithm has a lot of unnecessary operations in the application process,an adaptive bounding volume hierarchy structure partition method is proposed.The original 3D model is divided according to the tree structure,which can quickly determine the location of the point electromagnetic wave irradiated on the target,which greatly improves the speed and efficiency of polarimetric SAR target image simulation.Meanwhile,it can provide the data base for the research of polarimetric SAR target recognition.The third part mainly describes the study of target recognition for SAR images.This thesis introduces the method of feature extraction based on Principle Component Analysis(PCA),Deep Belief Networks(DBN)and Convolutional Neural Networks(CNN).In order to research the recognition of extracted features,Support Vector Machine(SVM)is used to classify different targets.For the sake of improvement of the recognition accuracy,the method of decision fusion is used in this thesis.Firstly,the principles of support vector machine(SVM)and three feature extraction methods are introduced,and the idea of decision fusion is put forward.Then,the target feature extraction and recognition method are validated on the polarimetric SAR target image dataset and the MSTAR dataset.And,it is proved that the image simulation method proposed in this thesis can be used for target recognition.