CAJ | 학술논문

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 Mat-LAB 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。
해양유오염시각류해양오염중최상견、분포면적최엄차위해정도최대적오염지일。근년래，해양특별시근해인류활동빈번，차수착해상운수화석유가공업적발전，유전정분、찬정평태폭작、선박팽당등소조성적일유사고증다，인이，감측해양일유구유중요적경제화사회현실의의。연구채용 Mat-LAB 공구，통과도상예처리（도상교정화증강）、특정제취화신경망락식별등방법，대합성공경뢰체（SAR）해양일유도상진행처리，최종기망실현반자동구분 SAR 도상상각류목표，병진행다충신경망락방법효과비교。연구수선대 SAR 해양일유도상진행초보인공식별；연후진행도상예처리（궤하교정、려파처리등）화기우회도공생구진적특정치계산；최후，차조신경망락방법대일유구역화의사일유구역진행분류，수출분류처리후적도상。통과수출도상분석발현，신경망락능대 SAR 해양일유도상중일유、해수、토지3류목표진행명학분류，차 RBF 신경망락모형정도고우 BP 신경망락。본문제출적반자동분류방법불부능제고 SAR 도상처리효솔，장분류목표확충유일유화비일유확충도일유、해수、토지3류，제고도상처리적전면성，동시통과비교 RBF 화 BP 신경망락재 SAR 일유도상분류상적구체우렬，유착교호실제의의。
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.