Research On Synthetic Aperture Radar Image Target Recognition Based On Feature Extraction And Hypergraph

Synthetic Aperture Radar(SAR)is a kind of high-precision earth observation device.Due to its advantages of not being limited by time,climate and altitude,it plays an irreplaceable role in many aspects such as natural disaster monitoring,classification of military sports targets and so on.In recent years,SAR image target recognition has gradually become one of its important applications,excellent classification accuracy has become an important criterion to judge the efficiency of target recognition.However,with the acquisition of SAR images in large quantities,the characteristics of high-dimensional imaging and the influence of speckle noise become one of the main obstacles to target recognition in SAR image.Although SAR image target recognition methods have been promoted to a certain extent with the in-depth research of more and more researchers at home and abroad.Due to the influence of environment and observation targets,the SAR image data are complex and changeable,there are still many problems in these methods.For example: considering data object global structure unitarily,poor target robustness and low efficiency.To solve the above problems,the research work of this paper is as follows:1)In order to solve the problem of ignoring the local structure of image data in the existing linear SAR image target recognition methods,a method based on hypergraph and Laplacian Eigenmaps(LE)is proposed.First,introducing category information of samples,LE and Linear Discriminant Analysis(LDA)are combined to extract features of SAR images with the objective of minimizing intra-class differences and maximizing inter-class differences,so as to reduce the feature dimensionality of SAR images.Then,considering the speckle noise of SAR images,hypergraph is used to describe the multivariate or even more complex relationship between targets in SAR images to improve the recognition accuracy.The experimental results of a large number of data on Moving and Stationary Target Acquisition and Recognition(MSTAR)database show that the proposed method has high discriminant performance and recognition ability for SAR images.Compared with other methods based on subspace feature extraction,the proposed method has different degrees of improvement.2)Aiming at the problem of poor target robustness and low efficiency,a method of SAR image target recognition based on Robust Kernel Principal Component Analysis(RKPCA)is proposed.First,the RKPCA is used to preprocess the SAR image data.Robust data sets are obtained by eliminating outliers several times.Then,the typical subspace learning methods and the algorithms based on convolutional neural networks are respectively applied to the new data set for experiments.The experimental results of different training time and parameters show that the effectiveness of proposed method,which not only make the recognition accuracy better,but also improves the training efficiency.