面向对象的SAR图像溢油信息提取研究

发布时间：2018-05-05 08:44

本文选题：SAR + 8-bit转换　；参考：《中国地质大学(北京)》2015年硕士论文

【摘要】：近年来,随着人们对石油的需求不断增加,海上运输业、石油开采业取得了快速发展,致使海上溢油事故发生的频率也呈现出不断增加的趋势。基于全天时、全天候的SAR技术能够穿透云和雾等优点,其目前已被广泛应用于海上溢油监测。本文采用面向对象的信息提取技术对SAR溢油图像进行了研究,成功提取出海上油膜信息,取得的主要成果归纳如下:(1)介绍了SAR系统工作原理和监测海面溢油原理,以及溢油现象在SAR图像上具有的表现特征。对实验数据进行去噪处理时,本文在不同的窗口大小下采用了七种常用的滤波方法对其进行斑点滤波并进行效果评价,最终选择了7×7窗口大小的增强Lee滤波方法。(2)在对SAR图像预处理过程中,引入了8-bit转换及灰度降级操作,保证了在不损失溢油信息的前提下压缩了数据量,从而提高了SAR图像处理的速度和效率。(3)在对SAR图像进行多尺度分割时,经过多次试验尝试,通过分析每种分割参数的影响并结合目视判读效果,最终选择了一组尺度参数为20、形状因子为0.2、紧致度因子为0.5的多尺度分割结果图像用于后续对溢油信息的进一步提取。(4)在提取溢油信息时,利用了能够区分油膜与疑似油膜目标的光谱特征、几何特征、纹理特征和物理特征,通过对这些特征做进一步的综合分析,构建了适用于溢油信息提取的模糊规则库,最后结合软件中提供的相应隶属度函数完成了对溢油信息的提取。(5)将基于像元的监督分类结果分别与面向对象的最邻近分类和模糊分类结果进行精度比较,其总体分类精度和Kappa系数分别为90.43%和0.7920、90.72%和0.8607、93.02%和0.8610。结果证明,采用面向对象的模糊分类方法可以更准确的完整的提取出海上溢油信息。
[Abstract]:In recent years, with the increasing demand for oil, the marine transportation industry and the oil mining industry have made rapid development, resulting in the frequency of oil spills on the sea also showing an increasing trend. Based on the advantages of all-day, all-weather SAR technology, which can penetrate clouds and fog, it has been widely used in offshore oil spill monitoring. In this paper, the object oriented information extraction technique is used to study the SAR oil spill image, and the oil film information is extracted successfully. The main results are summarized as follows: 1) the working principle of the SAR system and the principle of monitoring the oil spill on the sea surface are introduced. And the characteristics of oil spill on SAR images. When the experimental data is de-noised, seven common filtering methods are used in this paper under different window sizes to carry out speckle filtering and evaluate its effect. Finally, the 7 脳 7 window size enhanced Lee filtering method is chosen. In the process of SAR image preprocessing, the operation of 8-bit conversion and gray scale degradation is introduced to ensure that the data is compressed without losing the oil spill information. Thus, the speed and efficiency of SAR image processing are improved. When the SAR image is segmented at multiple scales, after many experiments, the influence of each segmentation parameter is analyzed and the effect of visual interpretation is combined. Finally, a group of multiscale segmentation images with scale parameters of 20, shape factor of 0.2 and tightness factor of 0.5 are selected for further extraction of oil spill information. The spectral features, geometric features, texture features and physical features are used to distinguish oil film from suspected oil film targets. Through further comprehensive analysis of these features, a fuzzy rule base suitable for oil spill information extraction is constructed. Finally, combined with the corresponding membership function provided in the software, the extraction of oil spill information is completed. Finally, the results of supervised classification based on pixel are compared with the results of object-oriented nearest neighbor classification and fuzzy classification, respectively. The overall classification accuracy and Kappa coefficient were 90.43% and 0.7920% 90.72% and 0.8607 793% and 0.8610%, respectively. The results show that the object-oriented fuzzy classification method can extract the oil spill information more accurately and completely.
【学位授予单位】：中国地质大学(北京)
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：X55;X87

【参考文献】