超像素边缘约束HMRF的SAR图像溢油识别  

Oil Spill Recognition for SAR Images with Edge Constrained HMRF at the Super-Pixel Scale

作　　者：石雪 许文超 SHI Xue;XU Wenchao(College of Geomatics and Geoinformation,Guilin University of Technology,Guilin 541004,China;School of Civil and Surveying&Mapping Engineering,Jiangxi University of Science and Technology,Ganzhou 341000,China)

机构地区：[1]桂林理工大学测绘地理信息学院,桂林541004 [2]江西理工大学土木与测绘工程学院,赣州341000

出　　处：《航天返回与遥感》2025年第1期150-159,共10页Spacecraft Recovery & Remote Sensing

基　　金：广西自然科学基金项目(2022GXNSFBA035567)。

摘　　要：为了克服传统马尔可夫随机场模型在海洋溢油识别中对合成孔径雷达(Synthetic Aperture Radar,SAR)图像相干斑噪声高敏感性以及溢油边界识别模糊等问题,文章提出一种超像素尺度下边缘约束隐马尔可夫随机场(Hidden Markov Random Fields,HMRF)的SAR图像溢油识别算法(Edge-Corrected HMRF at the Super-Pixel Scale,SE-HMRF)。利用简单线性迭代聚类(Simple Linear Iterative Clustering,SLIC)对SAR图像进行超像素分割,克服SAR图像中相干斑噪声的影响。为了提高溢油识别的准确性,在超像素分割基础上构建HMRF描述图像的空间关系,根据贝叶斯定理将溢油识别问题转化为能量函数最小化问题;为了克服SLIC对溢油边缘过分割或欠分割,将溢油边缘信息引入到能量函数中约束溢油识别结果。为了验证本文提出算法对溢油识别的准确性,选取Sentinel-1卫星SAR图像进行对比实验,本文提出算法溢油识别结果的Kappa系数和概率兰德指数分别达到0.951和0.954,全局一致性误差仅为0.024,定性评价与定量评价的结果均优于对比算法,说明文章提出算法能够在保持识别效率的同时获得准确的溢油识别结果。To overcome the high sensitivity of the traditional Markov random field model to the speckle noise of synthetic aperture radar(SAR)images and the blurring of the oil spill boundary identification in marine oil spill identification,this study proposes an edge-constrained hidden Markov random fields(HMRF)at the superpixel scale algorithm(SE-HMRF)for oil spill recognition in SAR images.Super-pixel segmentation of SAR images using simple linear iterative clustering(SLIC)to overcome the effect of speckle noise in SAR images.To improve the accuracy,HMRF is constructed to describe the spatial relationship of the image based on super-pixel segmentation,and transform the problem into an energy function minimization problem by theorems.To overcome the over-segmentation or under-segmentation of oil spill edges by SLIC,the oil spill edge information is introduced into the energy function to constrain the oil spill identification results.To verify the accuracy of this study's algorithm for oil spill identification,Sentinel-1 satellite SAR images are selected for comparison experiments,and the Kappa coefficient and probability Rand index of the oil spill identification results of this study's algorithm reach 0.951 and 0.954,respectively,while the global consistency error is only 0.024,and the results of the qualitative and quantitative evaluations are both better than those of the comparison algorithms,indicating that this study's algorithm can maintain the identification efficiency while obtaining accurate oil spill identification results.

关 键 词：SAR图像 海洋溢油识别 超像素分割 边缘约束 隐马尔可夫随机场 

分 类 号：V19[航空宇航科学与技术—人机与环境工程] TP75[自动化与计算机技术—检测技术与自动化装置]