Research On Key Techniques In Synthetic Aperture Radar Target Recognition

Synthetic aperture radar(SAR)could work in 24-hour a day and all-weather and has the capability of penetrating the earth surface.Consequently,it has very important meanings in battlefield surveillance,anti-ballistic missile and strategic warning,etc.As a key step in SAR image interpretation,automatic target recognition(ATR)plays an important role in the acquisition of battlefields reports.From 1980 s,researchers from all over the world have conducted a large amount of researches on SAR ATR and made great progresses.With the enhancement of SAR acquisition abilities and SAR image resolution,the interpretation of massive SAR images makes it a necessary work to improve the performance of ATR techniques.Therefore,it has important meanings to study SAR ATR techniques.By deeply analyzing the preceding works on SAR target recognition,this dissertation tries to improve the ATR performance from two aspects,i.e.,feature extraction and classifier design.The main contributions of this dissertation are as follows:Firstly,this dissertation comprehensively reviews the development of SAR ATR and demonstrates the importance of studying SAR ATR methods.Second,this dissertation provides detailed descriptions of the two key techniques in SAR target recognition,i.e.,feature extraction and classifier design.The state of the art of the feature extraction is presented and the strengths and defections of these features are analyzed.Several typical classifiers use in SAR target recognition are listed and their principles are briefly introduced.These analysis provide some advisable ways for us to improve SAR ATR performance.Thirdly,this dissertation proposes a SAR target recognition method based on joint sparse representation of multi-level dominant scattering images.The multi-level dominant scattering images are used as the basic features to describe SAR targets.They can describe the spatial distribution as well as the relative intensities of the target scattering centers in detail.Furthermore,they can reflect the target geometrical shape to some extent.Therefore,they can provide strong discriminability for SAR target recognition.The joint sparse representation based on Bayesian multi-task learning is employed to jointly classify the multi-level dominant scattering images for target recognition.The joint sparse representation can make use of individual discriminability of different levels of dominant scattering images while exploiting the inner-correlation between different levels.Therefore,the propose method can effectively improve the target recognition performance.Experiments are conducted on several typical recognition scenes on MSTAR(Moving and Stationary Target Acquisition and Recognition)dataset to validate the effectiveness of the propose method.Lastly,this dissertation proposes a SAR target recognition method based ondecision fusion of complementary features.PCA(Principal Component Analysis)features,target contour and peaks are used to describe the global characteristics,geometrical shape and local properties of SAR targets.The three features depict the SAR target from different aspects with good complementarity thus providing more information for target recognition.As for different types of features,suitable classification schemes are designed for target recognition.Then the three decisions are fused based on D-S evidence theory for robust target recognition.Experiments are conducted on several typical recognition scenes on MSTAR dataset to validate the effectiveness of the propose method.