Feature Extraction And Classification Of Ship Objects From SAR Images

With the increasing development of synthetic aperture radar(SAR),the resolution of SAR image is becoming higher.Thus,the SAR image contains much information of the object,and more efforts should be made on the interpretation of the image.As an important step in the interpretation of the SAR image,the extraction and classification of the object have been paid attention.The paper focuses on the extraction and classification of the ship object from the SAR image on the background of SAR image application in marine surveillance.Firstly,several classical segmentation algorithms for SAR image are analyzed and compared.Then geometric features are extracted from the segmented image,which include azimuth angle and minimum enclosing rectangle(MER).The algorithm are proposed to deal with the interference of scattering center and the influence of sea clutter,which are the method of azimuth angle estimation based on the maximum entropy of projection curve and the method of MER extraction based on the maximum filling ratio.Statistic results indicate that the proposed method can accurately extract the azimuth angle and the MER of the ship object.After that,the paper studies on the local structure feature extraction of the cargo ship in SAR image.With the object region being projected to the main axis,a distribution curve is obtained.Considering the periodic structure of the cargo ship,this paper proposes two method to estimate the period from the noisy periodic signal,i.e.the method based on multiple autocorrelation and F test,and the method based on generalized likelihood ratio test(GLRT).From the periodic extraction of local structure in the cargo ship,one can discriminate cargo ship from other ship objects efficiently.Afterwards,we design and implement a software which is capable of distinguish cargo ship from oil tanker on the background of the ship object classification application.The experiment results show that the average percentage of correct discrimination for cargo ship from oil tanker is 88.46%.Finally,since the complex SAR image contains much information about the object of interest,this paper applies the method of compressed sensing(CS)to extract the scattering centers from such data.Based on the extracted parameters of these scattering centers,the proposed maximum entropy method is used to estimate the azimuth angle of the ship object,and the estimation performance is compared with that of the estimator resulting from the intensity image.Furthermore,we also discussed the possibility of jointly analyze between the projection curve of scattering center and that of the intensity image along the main axis.