Research And Application Of Local Feature Extraction In Synthetic Aperture Radar Imagery

SAR image feature extraction is the basic work of SAR image interpretation and application,which has caught the attention of domestic and foreign researchers and been one of the research focus in SAR image interpretation.In the SAR image interpretation practice processing,global features extraction of image and targets is usually not robust bacause there are some random variation and fuzzy in SAR image.It is a new idea to solve the above problems that local feature is extracted and matched in SAR imagery.Local feature extraction put forward important information to support SAR image automatic interpretation.At present,the extraction method of local feature in SAR image generally follows optical image processing method,which is incompatible for SAR image processing.It is necessary to study the local feature extraction algorithm according to the imaging characteristics and noise surrounding of SAR image.It will overcome the robustness lack of traditional global features methods and poor adaptability of local feature extraction robustness in optical image,which improve accuracy and timeliness image matching,target detection and recognition in SAR image.This thesis focuses on the local SAR image feature extraction,using theoretical analysis and experimental verification.The three parts to deeply study includes:(1)Extraction and application of sailent region and feature in SAR image,(2)Feature matching of target scattering center sets and application to vehicle target recognition in SAR image,(3)A novel method and application of SAR image local feature extraction.As the foundation of the whole thesis,the second chapter describes the theory of image local invariance,then analyzes the research foundation,status and performance analysis of SAR image local invariant feature extractions,according to SAR image basic characteristics and noise characteristics.The third chapter proposes a new detection method of salient region in SAR image based on noise characteristics analysis and human visual saliency theory.We propose a new local complexity metric,which is insensitive to speckle noise and can effectively describe the local intensity variation of SAR image.In addition,via incorporating a stable distribution distance measure,the self-dissimilarity metric is redefined.Using these two components,we construct the saliency metric and generate the salient map.Experimental results demonstrate the accuracy,robustness,and stability of our method for SAR images,indicating that this method is competent salient region extraction and high value target detection in the SAR image,with a good application prospects.In the fourth chapter,a method is proposed to sequential matching of target scattering centers in SAR image based on the theory of point pattern matching in computer vision and SAR image attribute scattering center model.Firstly,the feature of target in SAR image is extracted using attributed scattering center model.Then among all components in this featured,only position component and frequency influence factor is taken as matching feature,followed by sequential matching and identification of the target.Compared with other similar methods,the experimental results show better performance in accuracy and recognition.In the fifth chapter,a new method of SAR image local feature extraction based on local gradient ratio pattern histogram is proposed.Firstly,the gradient information extraction algorithm in SAR image is analyzed,followed by ROA operator,ROEWA operator and GR operator.And then Harris operator,Lo G operator and the improved Multi-Scale Harris operator are introduced.Followed by the introduction of the local binary pattern(LBP)and rotation invariant local binary pattern(RILBP),and the good performace of multi-scale local gradient ratio pattern histogram(MLGRPH)in SAR image and experimental verification of the rotationally invariant properties of MLGRPH.Finally,according to the process of SIFT and SAR-SIFT method,a local invariant feature extraction method in SAR image based on MLGRPH is proposed.Experiments using several sets of SAR data with different imaging time,different wavelengths,different polarization modes and different perspectives are cariied on classical sift method,SIFT-OCT method and the proposed method for analysis and comparison.The results show that method presented in this chapter has better performance than the former two methods,and has the potential to be improved.In the sixth chapter,the work of this thesis is summarized,and the future research in SAR image local feature extraction is suggested.