SAR影像海洋表面溢油检测方法与实现

海洋是地球的重要组成部分,它为人类提供了丰富的物质和宝贵的资源,每年海洋都承受着不同程度的侵害,其中油类污染是给海洋造成巨大危害的污染之一。而油类污染又主要来源于轮船破裂漏油以及油井平台或海底输油管道爆炸等。每次事故造成的直接经济损失达几百万至上千万不等,所以对海上溢油进行监测具有重要的意义。选用Envisat的ASAR数据进行海上溢油检测,介绍并分析了SAR图像溢油检测的一般步骤及其实现方法,通过采用单一阈值分割法、最大熵分割法和非监督分类法对影像进行目标检测,从而粗略地将影像区分为前景区域与背景区域,并结合影像的纹理特征进行分类。在纹理特征选取过程中,通过人工选取部分溢油区与非溢油区作为感兴趣区,在感兴趣区上分别统计SAR影像常用的纹理特征,并结合不同目标检测的结果以及原始影像进行基于BP神经网络的分类,得到了良好的效果。最后展望了SAR图像海洋溢油检测的发展方向。

Method and Implementation of Oil Spill Detection in SAR Image

Ocean is one of the significant parts of the earth,which provides abundant resources for human beings.However,pollutions take place at different levels every year,in which oil pollution plays an important role.Oil pollution is mainly caused by oil leak from streamers and well platform,as well as by explosion of submarine pipe lines.Each accident may cost a direct economic losses ranging from millions to billions.Thus it is meaningful to develop oil spill monitoring.This paper uses a SAR data of Envisat monitoring marine oil spill,introduces and analyzes general steps and implementation of SAR images oil spill detection,chooses mean,variance,equivalent looks and speckle reduction capability as filtering effect evaluation indicators.Through these indicators,this study evaluates seven filters as follows,Lee,Enhanced Lee,Frost,Enhanced Frost,Gamma,Local sigma and Bit error,to compare,analyze and then draw a conclusion that enhanced Lee filter is the most appropriate for the images in the research.Then does target detection to images with single threshold segmentation,maximum entropy segmentation and unsupervised classification,thus divides image area to foreground region and background region roughly,at the same time,classifies by texture features of images.In the processes of texture features selection,this paper manually selects some oil spill areas and non-oil spill areas as regions of interest,and statistics of general texture features including mean,variance,homogeneity,contrast,dissimilarity,second moment and correlation of texture features of SAR images in regions of interest.The result indicates that mean,homogeneity,dissimilarity and second moment have the maximum differences between oil film and non-oil film,and these four texture feature parameters are selected and combined with results of targets detection and original images classification by BP neural network.The SAR data of Cosmo-Skymed with one meter resolution are used as basic data to verify the accuracy.Verification shows that the highest accuracy combined with maximum entropy is 86%.The second accuracy combined with single threshold is 81%~84%,the Isodata accuracy combined with unsupervised classification is 75%,the accuracy just combined with original images is 70%,the lowest accuracy without any targets detection is 57%.So targets detection plays a significant role in marine oil spill.This paper finally defines maximum entropy method as the best marine oil spill detecting method,and puts forward the direction of marine oil spill detection with SAR images in the end.