基于模糊C均值聚类的高分辨率合成孔径雷达图像舰船目标检测  

Ship Target Detection in High Resolution SAR Images Based on FCM

作　　者：张临杰[1] 张晰 郎海涛[3] 

机构地区：[1]中国海洋大学数学科学学院,山东青岛266100 [2]国家海洋局第-海洋研究所,山东青岛266061 [3]北京化工大学应用物理系,北京100029

出　　处：《中国海洋大学学报（自然科学版）》2018年第3期134-140,共7页Periodical of Ocean University of China

基　　金：山东省自然科学基金项目(ZR2013FQ026); 海洋公益性行业科研专项经费项目(201505002); 国家自然科学基金项目(11371333); 中央高校基本科研业务费专项(201362033,201564019)资助

摘　　要：给出了一个基于模糊C均值聚类(FCM)的舰船目标检测算法。其中心思想是将SAR图像中的各像素点灰度值视为样本集,然后利用FCM算法对该样本集聚类,并利用聚类结果计算图像分割全局阈值。与目前流行的恒虚警率(CFAR)检测算法相比,所给检测算法具有参数少、计算量与图像大小成正比、舰船轮廓保持良好的特点,为高分辨率SAR图像舰船目标检测提供了一个新的选择。Ship detection is of great significance for both military and civilian applications.Synthetic aperture radar(SAR)with all-day,all-weather,ultra-long-range characteristics,has been used widely.The constant false alarm rate(CFAR)methods are popular for target detection,work efficiently when the background clutter can be modeled easily.However,in recent years,the resolution of SAR images is improved greatly due to the rapid development of the SAR sensor.With the improvement of resolution,the background clutter get more complicated,and the contour of targets is clearer too.When we make use of the CFAR methods,modeling of the background clutter is more difficult,and more computation time is needed for keeping the contour of ship targets.In this situation,new detection method is needed for high resolution SAR images.The objective of this paper is to give a new ship detection method for high resolution SAR images.This new detection method is based on Fuzzy c-mean clustering method(FCM).In this method,gray values of SAR image pixels are regarded as sample set firstly.Then cluster this sample set into four clusters with FCM method.After that,aglobal threshold is calculated basing on the clustering result,and potential ship targets are detected with this global threshold.Finally,like other detection methods,morphological method is used for further screening.Numerical experiment was carried out to compare the proposed method with two-parameter CFAR method.A group of full-polarization SAR images were used.The quality factor of two-parameter CFAR method is 0.26~0.77 when the size of target window is 10×10,0.23~0.63 when the size of target window is 2×2.The computation time is 25.23~25.48 s,566.59~569.93 sseparately.The quality factor of the proposed method is 0.48~0.92,and the computation time is 44.95~47.05 s.Also,the contour in the proposed method is the best.Comparing to two-parameter CFAR method,the given method has merits as follows:few parameters,computation cost is proportional to image size,ship contour can be kept

关 键 词：舰船目标检测 FCM 高分辨率SAR 图像分割 舰船轮廓 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]