Research On Target Region Rapid Positioning And Target Detection In SAR Images

With the development of remote sensing technology, the resolution of synthetic aperture radar(SAR) image has been greatly improved. In order to quickly and effectively locate the interestedtargets in SAR images, the research on target detection needs to be done. In this thesis, a new mixturestatistical distribution model is proposed based on the studies on statistical modeling for clutter andtarget in the SAR images. With the help of the mixture statistical model, the research works on targetregion rapid positioning and target detection are conducted. The main research works are listed as thefollows.Firstly, focusing on the land clutter and metallic target, a variety of goodness of fit test methodsare applied to test the degrees of fitting of the often used statistical models for modeling clutter andtarget in the SAR images, and the best statistical model for fitting land clutter and the best statisticalmodel for fitting metallic target are discovered.Secondly, a mixture statistical model is proposed in this thesis. This mixture model is able tomodel the SAR data collected from clutter region, target region, or the region containing both clutterand target. Also, through fitting a SAR image using the mixture model, we can obtain the proportionsof clutter region and target region as well as the statistical properties of the clutter data and target data.The parameter estimation method is studied for the mixture model, and the parameter estimationmethod is tested through experiments, and the results show that the method is able to accuratelyestimate the parameters of the mixture model.Thirdly, target region rapid positioning method is proposed based on the property that themixture model can provide the proportions of clutter region and target region. Coarse to fine regiondivision is applied to a SAR image to partition the image into regions, and then the mixture model isused to model the data from each region, and based on the parameter estimation results, the clutterregions are excluded and the target regions are then located.Finally, based on the property that the mixture model can provide the statistical properties of theclutter data and target data, the generalized likelihood ration test (GLRT) target detection method isdesigned. Experiments are conducted, and the results show that the proposed GLRT method has lowerfalse alarm rate and higher correct detection rate than the often used CFAR detection method, and theGLRT method is comprehensively superior to the CFAR method.The research works carried in this thesis involve not only the theoretical researches for the SAR image statistical modeling and target detection but also the specific design methods, and the researchworks of this thesis should be useful for the related academic studies as well as the practicalapplications.