基于KMM与超像素的SAR海面暗斑分割算法  

Method of dark spot segmentation with combination of K multiplicative model and superpixel from SAR Imagery of sea surface

作　　者：胡桂香[1] 李宁[1] 邢艳肖 

机构地区：[1]中国科学院电子学研究所,北京100190

出　　处：《国外电子测量技术》2016年第6期101-108,共8页Foreign Electronic Measurement Technology

摘　　要：提出了一种基于K乘性模型(K multiplicative model,KMM)与超像素分割算法的无监督的SAR海面暗斑检测算法。KMM算法采用矩估计法进行参数估计,应用于海表面含暗斑的感兴趣区域(region of interest,ROI)检测。通过对感兴趣区域生成的超像素进行聚类来完成最终暗斑的边缘分割。在边缘回撤率、遗漏误差、交叉误差上有较大改善,机载SAR的平均误差率为10%,而TerraSAR的平均误差率低于2%。在计算效率上也有较大提高,平均0.53s内可处理104个像素点。实验基于高分辨率L波段机载SAR数据以及TerraSAR-X数据进行,实验结果表明该方法具有较强的鲁棒性以及较高的计算效率。This paper proposes an unsupervised approach for detecting regions of interest（ROI）based on the K multiplicative model（KMM）and superpixel segmentation in SAR imagery.KMM,with method of moments parameter estimation,mainly aims at detecting the dark-contain ROI on the sea surface.And the superpixel-based segmentation accomplishes edge extraction of ROI through clustering superpixels of ROI.The approach in this paper significantly outperforms adherence to boundary,achieving the error rate of less than 10%for airborne SAR imagery,and less than 2%for TerraSAR imagery.This approach also performs well in the computational efficiency,processing 104 pixels in 0.53 seconds on average.Experiments,accomplished by resorting to high resolution of the L-band airborne synthetic aperture radar（SAR）data and the TerraSAR-X data,demonstrate strong robustness and high effectiveness based on the proposed approach.

关 键 词：K乘性模型 暗斑检测 合成孔径雷达 超像素 

分 类 号：TN911.73[电子电信—通信与信息系统]