Research On Target Extraction In High-resolution ISAR Imagery

Inverse Synthetic Aperture Radar(ISAR)can carry out high-resolution,long-range,all-day and all-weather imaging of moving ship targets.It is a very important strategic defense method.Target extraction,as the preparatory work for target recognition,is the prerequisite for image target interpretation,which has a direct impact on the subsequent target feature description and even classification recognition.ISAR images are usually aimed at the detection and identification of a single moving target,which requires higher target details and contour integrity.However,in practical applications,high-resolution ISAR images often have problems such as blurred target contours,broken or loose target areas,and severe interference from fringe noise.This undoubtedly adds difficulty to the subsequent target recognition work.Therefore,this article aims to study the method of high-resolution ISAR image ship target extraction.The core method of target extraction in this paper is the Constant False Alarm Rate(CFAR)algorithm.Before CFAR detection,the optimal statistical model of ISAR image background clutter needs to be obtained.The parametric model modeling method was used to analyze several commonly used clutter statistical distribution models and parameter estimation methods.The modified mean square error test method was used to test the goodness of fit of the fitting results of several distributions.The experimental results of the measured ISAR images show that the gamma distribution and Weibull distribution have the best fitting effect on the ISAR image background distribution.Subsequently,the basic principles of CFAR are analyzed,and the global CFAR detection that is more suitable for ISAR images is selected.The gamma distribution is used as the ISAR image background distribution model to realize the extraction of ISAR image targets based on pixel-level CFAR.Aiming at the unique horizontal fringe noise interference of ISAR image,this paper analyzes its causes and solutions.The morphological calculation method is used to post-process the binary map after CFAR detection,to a certain extent,to suppress noise and fill the holes inside the target.The segmentation accuracy and residual images are used to evaluate the target extraction results.The experimental results show that when using the traditional pixel-level CFAR algorithm to extract high-resolution ISAR targets,the noise suppression effect and target integrity can not meet the requirements.Morphological operations are uncertain and will lose the original information.In order to solve the above problems,this paper uses the CFAR detection method in units of image blocks in combination with the surrounding information of pixels.The Markov Random Field(MRF)and superpixel methods are used to achieve ISAR image over-segmentation.Aiming at the problem that the traditional simple linear iterative cluster(SLIC)superpixel generation method is sensitive to multiplicative noise of ISAR image,this paper introduces the gray average value ratio to improve the SLIC method.The simulated ISAR images before and after adding multiplicative noise are used for comparison and verification.The results show that the improved SLIC superpixel segmentation method based on the gray average ratio is robust to multiplicative noise of ISAR images.The image over-segmentation results of MRF and improved SLIC method are used respectively,and the gamma distribution is used as the background distribution model for global CFAR detection.The experimental results of the measured ISAR images show that the area-based CFAR detection method has higher extraction accuracy than the traditional pixel-level CFAR detection target,and the superpixel-based CFAR has a shorter running time than the MRF-based CFAR.Aiming at the problem of disconnection of the target caused by the weak local scattering characteristics of the ISAR image,based on the traditional target clustering algorithm,this paper proposes a target superpixel clustering and connectivity method,which is used as a post-processing link in superpixel CFAR Binary image after detection.According to the actual situation of the target on the sea,the target superpixel clustering is achieved.The center line of the hull is extracted by Radon transform,and the position and shape of the filling area are determined around the center line and the target area information.