基于合成孔径雷达回波信号的海洋溢油监测方法研究

Research on offshore petroleum oil spilling detection using SAR echo signal

海洋油污染是各类海洋污染中最常见、分布面积最广且危害程度最大的污染之一。近年来,海洋特别是近海人类活动频繁,且随着海上运输和石油加工业的发展,油田井喷、钻井平台爆炸、船舶碰撞等所造成的溢油事故增多,因而,监测海洋溢油具有重要的经济和社会现实意义。研究采用MatLAB工具,通过图像预处理(图像校正和增强)、特征提取和神经网络识别等方法,对合成孔径雷达(SAR)海洋溢油图像进行处理,最终期望实现半自动区分SAR图像上各类目标,并进行多种神经网络方法效果比较。研究首先对SAR海洋溢油图像进行初步人工识别;然后进行图像预处理(几何校正、滤波处理等)和基于灰度共生矩阵的特征值计算;最后,借助神经网络方法对溢油区域和疑似溢油区域进行分类,输出分类处理后的图像。通过输出图像分析发现,神经网络能对SAR海洋溢油图像中溢油、海水、土地3类目标进行明确分类,且RBF神经网络模型精度高于BP神经网络。本文提出的半自动分类方法不仅能提高SAR图像处理效率,将分类目标扩充有溢油和非溢油扩充到溢油、海水、土地3类,提高图像处理的全面性,同时通过比较RBF和BP神经网络在SAR溢油图像分类上的具体优劣,有着较好实际意义。

Oil spilling is one of the major sources for in marine pollutions,which are widely distributed and can bring cause terrible significant environmental damages.In recent years,due to the increase in offshore human activities and development of petroleum processing industries,oil spill accidents are also increasing,which are mostly caused by well blowouts,explosions of drilling platforms and ship collisions.Therefore,monitoring oil spilling has impor-tant significance in both economical and social aspects.As an all-weather high-resolution active microwave imaging sensor,Synthetic Aperture Radar （SAR）can greatly improve the resolution of images and the accuracy of fore-casts,and thus takes an important role in oil spill monitoring.This paper aims to realize the semi-automatic identi-fication of various targets on SAR images.We have conducted a convincing contrast of different neural networks, using Matlab as the tool through image preprocessing （image correction and enhancement）,feature extraction and neural network recognition.First,oil spilli images are preliminarily manually identified,followed by image prepro-cessing （such as geometric correction,filtering,etc.）and feature extraction based on gray level co-occurrence ma-trix.Then,two types of neural networks,namely RBF and BP ,are introduced to classify the oil spill area and other suspected areas.Finally,the processed images are analyzed,indicating the capability in classifying oil,sea water,and land targets.The results reveal that the outputs from the RBF neural network are more accurate compared to those from the BP neural network.