极化SAR流冰区大块接触性浮冰分离方法研究

发布时间：2018-01-06 09:25

本文关键词：极化SAR流冰区大块接触性浮冰分离方法研究　出处：《大连海事大学》2017年硕士论文　论文类型：学位论文

【摘要】：全极化合成孔径雷达由于不受光照、云层和恶劣天气等的影响,具有全天时、全天候的特性,是目前对海洋浮冰进行监测的主要手段。流冰区数据的研究,对流冰流经区域的海上航运、海上生产、油气勘探和极地地区的海冰监测来说都是具有重要意义的,在这些领域都需要实时提供海冰的相关数据。但是不得不认识到的是,通过SAR得到的原始数据所包含的数据量过于庞大,所以对SAR图像进行解译离不开计算机的辅助,而解译的结果直接影响下一步研究的准确性。流冰区的浮冰具有大小不一、形状不规则、呈网状散落分布等特点。这些浮冰中有的大块浮冰之间相互挤压,有的碎冰之间紧紧挨在一起,还有的浮冰似断非段,加上大面积的冰皮的存在使得图像的分割的难度大大增加。基于流冰区海冰的特点,本论文提出了一个分割与分类相结合分离大块弱连接浮冰的方法,其研究内容及创新之处概括如下:首先,对碎冰和不稳定小浮冰的剔除。在滤波等噪声抑制处理的基础上,利用预处理后的图像先通过形态学的方法,选取多组窗口大小对海冰图像进行处理,用轮廓标记算法和面积条件剔除碎冰;然后利用凸包提取技术优化浮冰的周长信息,利用面积、圆形度、离心率等约束条件剔除不稳定的小块浮冰。然后,对大块弱连接性浮冰进行区分。先进行纹理特征的提取,利用样本特征散点图分析比较各个纹理特征的优劣;然后利用理想的纹理特征进行单特征分类和多特征线性融合分类,并进行比较分析。实验结果表明利用多特征融合的分类方式得到的图像分类准确度更高,稳定性更强。最后,利用格陵兰岛Fram海峡的两景图像进行了实验,并用第二景图像对实验结果进行了验证,证明了本文的实验方法具有可行性和稳定性。
[Abstract]:Full polarization synthetic aperture radar (ASAR), which is not affected by illumination, clouds and bad weather, has the characteristics of all-day and all-weather, which is the main method to monitor the sea ice. It is of great significance for maritime navigation, offshore production, oil and gas exploration and monitoring of sea ice in polar regions. There is a need to provide real-time data on sea ice in these areas, but it has to be recognized that the raw data obtained through SAR contains too large a large amount of data. Therefore, the interpretation of SAR images can not be separated from the computer, and the results of interpretation directly affect the accuracy of the next research. Some of the large pieces of ice in these ice floes squeeze each other, some pieces of ice close to each other, and some ice floes seem to break off. In addition, the existence of a large area of ice makes the image segmentation more difficult. Based on the characteristics of floating ice, this paper proposes a method to separate large pieces of weakly connected ice floes combined with segmentation and classification. The research contents and innovations are summarized as follows: first of all, remove the broken ice and unstable small ice floe. On the basis of filtering and other noise suppression processing, the pre-processed image is first through the morphological method. The image of sea ice is processed by selecting several groups of window sizes, and the ice fragments are eliminated by contour marking algorithm and area condition. Then using convex hull extraction technology to optimize the information of the circumference of the ice, using the area, roundness, centrifuge and other constraints to eliminate unstable small pieces of ice floe. Then. Firstly, the extraction of texture features is carried out, and the advantages and disadvantages of each texture feature are analyzed and compared by using the scatter plot of sample features. Then the ideal texture features are used for single feature classification and multi-feature linear fusion classification. The experimental results show that the classification method based on multi-feature fusion is more accurate. Finally, two scene images of Fram Channel in Greenland are used to carry out experiments, and the experimental results are verified by the second scene image, which proves the feasibility and stability of the experimental method in this paper.
【学位授予单位】：大连海事大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN958

【相似文献】