Study On Ship Detection And Discrimination In Spaceborn Synthetic Aperture Radar Imagery

Spaceborne SAR(Synthetic Aperture Radar)is widely used in ship surveillance due to its wide swath and independence on climate and weather.Researches on maritime ship detection and discrimination in spaceborne SAR imagery have a significant impact on military and civilian applications.This dissertation focuses on the current key problems in ship detection and discrimination in spaceborne SAR imagery.Ship detection,region of interest(ROI)segmentation and discrimination are discussed in detail in this dissertation.In the context of ship ROI detection,preprocessing is researched in this dissertation firstly.Constant false alarm rate(CFAR)is an effective ship detection method.However,the performance will degrade at low signal-clutter rate(SCR)imagery.In order to improve the performance,this dissertation proposes an approach based on spatial information.It calculates the local gray density information.Then this information is combined with the original image to get a combined image.Finally a CFAR method is applied to the combined image.Experimental results based on ENVISAT and TerraSAR-X data demonstrate that the proposed method can achieve better performance in the case of low SCR in different resolution imagery.In addition,pixel values of ships and the background are different and ships can be regarded as sparse compared to the background.Therefore,this dissertation proposes a block prescreening based ship detection algorithm.This algorithm employs a classifier to decide which block may contain ships and then makes a detection only on these blocks.Experimental results based on RS-3 data demonstrate that the proposed algorithm can achieve better performance.In the context of ROI segmentation and discrimination feature extraction,this dissertation has discussed a variety of segmentation methods firstly.According to the performance of each method,2D-Otsu is chosen as the ROI segmentation method in this dissertation.Then conventional features are discussed in this dissertation.The traditional standard deviation can be easily influenced by the area ratio of the ship target in the candidate chip.Thus traditional standard deviation may not distinguish ship targets and false alarms.Aiming at this problem,this dissertation proposes a novel feature called modified standard deviation.Experiments on TerraSAR-X show that the novel feature presents better separability and stability.In the context of ship ROI discrimination based on feature selection,different feature selection methods are analyzed elaborately.Then this dissertation proposes an innovative method combining the Sequential Floating Forward Selection(SFFS)with Support Vector Machine(SVM).Feature selection is applied to TerraSAR-X data and the discrimination performance is compared in feature groups by different feature selection methods.Experimental results can achieve better classified performance while selecting fewer features.