SAR及MODIS数据海面溢油监测方法研究

【作者】 石立坚；

【导师】 赵朝方；

【作者基本信息】 中国海洋大学 ， 海洋信息探测与处理， 2008， 博士

【摘要】 海洋溢油发生后,准确及时的监测溢油对于海洋环境保护具有重要意义。随着卫星遥感技术的高速发展,遥感已经成为监测溢油的最重要和最有效手段之一。本论文以海面溢油为研究对象,讨论了利用SAR和MODIS监测海面溢油的方法,重点在SAR图像中溢油现象识别、MODIS监测海面油膜厚度、基于GIS的遥感溢油监测系统和中国海溢油分布等方面进行研究。本文的主要研究结果归纳如下:1.系统介绍了SAR图像中识别溢油现象和疑似现象的全过程,主要包括图像预处理(辐射校正和斑点噪声滤除)、图像分割、特征提取、特征筛选和神经网络识别。在SAR图像预处理方面,比较了10种滤波方法,最后选择了窗口为3×3的增强Lee滤波对SAR图像进行滤除斑点噪声的处理。在SAR图像中油膜区域分割方面,将水平集方法与多尺度小波方法相结合,提高了计算效率,降低了算法的复杂度,同时对SAR图像中油膜区域的模糊边缘进行了有效监测。在识别溢油和疑似溢油方面,除灰度特征量外,引入图像纹理特征参量,并利用方差分析从计算的31个特征参量筛选出16个作为神经网络的输入,建立了区别溢油现象和疑似溢油现象的神经网络模型。研究结果表明,该模型识别精度达到83%,可以较好地识别溢油现象;纹理特征作为特征参量提高了溢油现象的识别精度;基于方差分析的特征参量筛选不仅简化了神经网络结构,而且提高了模型的识别精度。2.将模糊聚类方法FCM与图像纹理分析相结合用于对250m分辨率的MODIS图像中的溢油区域分割;将这种方法应用到2005年4月3日发生在大连海域“阿提哥”事件的MODIS图像中,较好的区分了溢油区域,并在此基础上,利用多时相的MODIS数据计算了油膜的漂移,结果与风场和潮汐数据一致。建立了一个简单的海面油膜的光学模型,并在此基础上对海面油膜的可见光波段的光学性质进行了定性分析:油膜与海水的反差主要依赖于油膜对海水向上辐亮度的吸收,离水辐亮度差异在海水光谱反射峰的波段油水反差最大,随着油膜厚度的增加,油膜的离水辐亮度减少。在传感器接收的大气顶辐亮度的图像中,油膜与海水的反差依赖于镜面反射和下表面向上辐亮度传输的平衡:当镜面反射的差异大于下表面向上辐亮度的差异时,油膜海水反差为正,反之为负。基于此,根据两次溢油事故的MODIS数据定性给出了油膜厚度的分布。3.设计了基于WebGIS和卫星数据的海面油污染监测系统,该系统是一个融合了Apache web服务器、Oracle数据库管理系统、Mapserver、PHP和卫星遥感数据处理系统的网络地理信息系统(WebGIS),主要包括环境信息收集模块、溢油信息提取模块和油膜扩散预测模块。基于这个系统和2002～2005年的SAR数据对东中国海的溢油事件进行了统计分析,污染比较严重的四个海区包括:黄海中部,纬度范围在32°～37°N;渤海海峡东部;长江口周围海域;纬度范围在28°～32°N的东海海域及台湾海峡。溢油主要是由船舶非法排污引起的,分布在主要航线的海区,另外还包括渔船作业产生的油膜、河口排放的工业污染形成的油膜。4.将WebGIS系统与FCM方法相结合建立了一个帮助农民用户产生定量施肥图的专家系统。基于此系统,进一步研究利用陆地卫星Landsat数据制定定量施肥图的可能性;系统输出结果通过产量或现场土壤属性的测量进行评价,研究表明可以利用作物生长茂盛时期(NDVI均值最高)的卫星NIR波段数据产生变量施肥图。更多还原

【Abstract】 Oil spill detection has important significance for the oceanic environmental protection. With the rapid development of the satellite remote sensing, remote sensing technique has become one of the important and effective tools in oil spill detection. This dissertation discussed the methods of the sea surface oil spill detection using Synthetic Aperture Radar (SAR) and MODIS data sets. The main researches focus on the oil spill phenomena recognition in the SAR images, oil film thickness estimation with MODIS data, oil spill detection system based on satellite data and GIS and oil spill distribution in the China Sea. The main results and conclusions of this dissertation are summarized as follows:1. The procedures are introduced to distinguish oil spill from the look-like phenomena in the SAR images, which include image preprocessing (radiation correction and speckle noise filtering), image segmentation, feature extraction, feature filtering and oil spill recognition based on Artificial Neural Network (ANN).After analyzing 10 different noise filtering methods, we chose the enhanced Lee filter with the window of size 3×3 pixels to reduce speckle noise. Oil spill region is segmented effectively by combining the Level Set and multi-level wavelet method, and the new segmented method improves the computation efficiency and decreases the computation complexity.To distinguish oil spills from the look-like phenomena, texture features of SAR images were extracted, in addition to the grey features, to be used as the inputs of ANN. The analysis of variance (ANOVA) was used to evaluate the importance of the features in distinguishing oil spills from the look-likes phenomena. The selected 16 features were used as the input of ANN. We found that 83% of the total test data were classified correctly, and it seems that the second-order statistic features based on co-occurrence matrix and features filtering with ANOVA improve the result of oil spills identification compared with the other methods.2. The fuzzy C-means (FCM) cluster algorithm with a texture feature analysis was developed to detect oil spill using.MODIS images in 250m resolution. Use of entropy based on the GLCM improved the efficacity of classification significantly, and the oil spills in the coastal region of Dalian port was clearly identified. The movement of oil spill estimated from two consecutive MODIS images was consistent with the ocean current estimated from the prevailing wind field and tidal data. Then we proposed a simple radiative transfer model of surface oil spill and analyzed the corresponding optical property in the visible band. The contrast between surface oil and background water depends on the absorption of upwelling radiance within the oil film. The upwelling radiance contrast owns a greatest value near the sea water reflectance peak and decreases as the oil film thickness increase. For the radiance received at the top of atmosphere (TOA) by the satellite sensor, the contrast between oil film and surrounding water depends on the balance between specular reflectance and the sub-surface upwelling radiance. The contrast of TOA radiance is positive when the contrast of specular reflectance is higher. Based on the above result and the MODIS data, the film thickness distribution of two oil spill accidents was analyzed qualitatively.3. An oil spill detection system is constructed by integrating Apache web server, Oracle database management system, Mapserver, PHP and satellite remote sensing data into the Web Geographic Information System(WebGIS). The system included the environmental information collecting model, oil spill information extracting model and oil spill drift forecasting model. Based on collecting SAR images from 2002 to 2005 and the WebGIS as a reference system, the oil spill distribution map for the East China Sea is presented. Four heavily polluted areas in the region have been delineated; they include the central part of the East China Sea, the eastern area of Bohai Strait, the marine area in the vicinity of the mouth of the Yangtze River and surrounding waters, and the southern part of the East China Sea including Taiwan Strait. A main source of oil pollution in the China Seas is the illegal discharges from passing ships, which are roughly distributed along main regional/international ship routes. Other sources include fish & oil waste caused by fishing operations and surface active materials with river runoff produced by onshore industrial enterprises.4. Combining the above WebGIS framework and FCM method, a system is constructed to help farmers to create management zones for variable rate applications of fertilizer. Based on the system, the further research is to examine how to use routine satellites data, such as Landsat,to create management zones. By comparing results with the management zones created with yield maps or soil survey, we will determine the optimal satellite images both temporally and spectrally that can be used to delineate appropriate management zones. NIR data from Landsat of the vegetation canopy flourishing season present the promising potential to replace yield data as the input of model to create the zone map of nitrogen fertilizer application.更多还原