地球信息科学学报 ›› 2014, Vol. 16 ›› Issue (2): 299-306.doi: 10.3724/SP.J.1047.2014.00299

• 遥感科学与应用技术 • 上一篇    下一篇

基于MODIS谱纹信息融合的海洋溢油检测方法

苏伟光1,2,3, 苏奋振2, 杜云艳2   

  1. 1. 中国科学院海岸带环境过程重点实验室(烟台海岸带研究所), 山东省海岸带环境过程重点实验室, 中国科学院烟台海岸带研究所, 烟台 264003;
    2. 中国科学院地理科学与资源研究所 资源与环境信息系统国家重点实验室, 北京 100101;
    3. 中国科学院大学, 北京 100049
  • 收稿日期:2013-08-28 修回日期:2013-09-22 出版日期:2014-03-10 发布日期:2014-03-10
  • 通讯作者: 苏奋振(1972- ),男,研究员,博士生导师,研究方向为海岸带遥感及海洋GIS。E-mail:sufz@lreis.ac.cn E-mail:sufz@lreis.ac.cn
  • 作者简介:苏伟光(1983- ),男,博士生,研究方向为海洋及海岸带遥感与应用。E-mail:suwg@lreis.ac.cn
  • 基金资助:

    国家科技支撑课题小卫星智能观测荒漠化和海岸带监测应用示范(2011BAH23B04);国家海洋公益性行业科研专项经费资助项目(201005011)。

MODIS Based Spectral and Texture Integration Oil Spill Detection Method

SU Weiguang1,2,3, SU Fenzhen2, DU Yunyan2   

  1. 1. Key Laboratory of Coastal Zone Environmental Processes, Yantai Institute of Coastal Zone Research(YIC), CAS, Shandong Provincial Key Laboratory of Coastal Zone Environmental Processes, YICCAS, Yantai Shandong 264003, China;
    2. State Key Laboratory of Resources and Environmental Information System, Institute of Geographic Sciences and Natural Resources Research, CAS, Beijing 100101, China;
    3. Graduate University of Chinese Academy of Sciences, Beijing 100049, China
  • Received:2013-08-28 Revised:2013-09-22 Online:2014-03-10 Published:2014-03-10
  • Contact: 10.3724/SP.J.1047.2014.00299 E-mail:sufz@lreis.ac.cn

摘要:

鉴于仅依赖光谱特征或纹理特征的传统溢油检测算法的信息检测精度较低的问题,本文提出了一种新的光学遥感数据的谱纹海面溢油检测方法。谱是光学遥感数据的油膜敏感波段图像,纹是利用灰度共生矩阵计算获得的图像纹理特征,将这些特征相结合,引入支持向量机方法(Support Vector Machine,SVM),建立谱纹海面溢油检测模型。本文以2006年渤海溢油事故为例,利用中等分辨率成像光谱仪MODIS的光学遥感数据对溢油进行检测,MODIS的第2波段为油膜敏感波段,所以,第2波段图像即为选取的谱特征,经过对各个纹理特征的分析得到,均值、对比和相关3个特征量可作为溢油提取的纹理特征。检测结果的总体精度达91.23%。试验结果表明,将MODIS图像的光谱特征和纹理特征相结合,可有效地对渤海海洋油膜信息进行检测,并具有很强的抑制噪声能力。

关键词: MODIS, SVM, 纹理, 光谱, 溢油

Abstract:

Marine oil spill detection has become a worldwide issue. Traditional oil spill detection algorithm only depended on the spectral or texture has low detection accuracy. This paper presents a new method of the oil spill detection based on spectral and texture from optical remote sensing data. The spectral feature is oil slick sensitive band of the optical remote sensing data and the texture feature is got by gray level co-occurrence matrix. The model used support vector machine method to establish the spectrum and texture oriented oil spill detection method with these features. This paper used the MODIS optical remote sensing data to detect the oil spill in China Bohai Sea in 2006. The oil spill information can be shown on this band clearly because the variance of the sea water spectrum is less than the contrast of oil-water spectrum in the MODIS band 2, so we called the MODIS band 2 is the oil slick sensitive band, which we selected as the spectral feature. We obtained eight texture eigenvalues of oil slick and sea by gray level co-occurrence matrix in the MODIS band 2. Then three texture eigenvalues include mean, contrast and correlation were selected by analysis. Based on the selected sample, the detection accuracy was up to 91.32% and the Kappa coefficient is 0.7125. In contrast with spectral and texture integration Maximum Likelihood method and SVM method only use the spectral feature, the detection method we proposed is superior to these methods. The results showed that the method to combine the spectral feature and textural feature of MODIS data can effectively extract the oil slick in the Bohai Sea, and has a strong ability to suppress noise.

Key words: MODIS, oil spill, SVM, spectral, texture