基于超像素的SAR图像海岸线检测算法

发布时间：2018-07-06 18:14

本文选题：SAR图像 + 超像素　；参考：《大连海事大学》2017年硕士论文

【摘要】：合成孔径雷达(Synthetic Aperture Radar,SAR)图像以全天候成像等优势在海域动态监测、自动导航等领域有广泛的应用,其中海岸线检测是动态监测海岸线变化的一个解决手段。然而在SAR图像中,由于相干斑、海风和陆地环境的复杂性等因素,使得海岸线检测具有较大难度。本文从超像素角度出发,利用区域合并、三重马尔科夫随机场(TMF),重点研究SAR图像中海、陆区域存在陆、海类似小区域且海陆对比度较低时的海岸线检测算法。本文主要工作如下:(1)给出了一种基于超像素的区域合并改进海岸线检测算法。当海面或者陆地不均匀时,已有的区域合并海岸线检测算法容易出现无法合并的小区域,且需要人为设定合并阈值的问题,本文首先给出了一种改进的超像素算法,通过构建一个改进的局部窗,可以有效的解决传统矩形窗中因含有边缘导致计算出的特征模糊的问题,通过该局部窗构建一个相似性描述子,使提取出的特征更加精确,超像素的边缘贴合度更高。然后以超像素为基元,又给出一种改进的区域合并准则,该准则同时考虑超像素的像素均值、相对大小和统计量信息,再根据邻域信息得到确定局部阈值的方式,解决了已有算法中需要人为设置阈值的问题。通过对SAR图像的实验,验证了算法的有效性。(2)给出了一种基于超像素的三重马尔科夫随机场(TMF)改进海岸线检测算法。针对已有算法边缘贴合度不高的问题,首先给出一种基于Gamma分布的超像素算法,假设均匀局部邻域服从Gamma分布,通过置信区间确定左右门限形成超像素以提高边缘贴合度。然后以超像素为基元,给出一种改进的TMF算法,构建了一种基于改进辅助场的能量函数以解决已有TMF算法的阈值依赖性问题。通过对SAR图像的实验,验证了算法的有效性。(3)给出了一种基于概率的超像素三重马尔科夫随机场(TMF)改进海岸线检测算法。针对海陆对比度较低的问题,首先给出一种统计量以增强海陆对比度,其次给出一种改进超像素算法以解决边缘贴合度不高的问题。最后给出一种基于概率的TMF改进算法以解决已有算法未考虑同类超像素相关性问题,以一个概率向量表示当前超像素属于海或陆的概率,由此构建一个基于概率的改进能量函数,替代已有TMF中的Ising模型。通过对SAR图像的实验,验证了算法的有效性。
[Abstract]:Synthetic Aperture radar (SAR) images have been widely used in sea area dynamic monitoring, automatic navigation and so on with the advantage of all-weather imaging. Among them, coastline detection is a solution to dynamically monitor the changes of coastline. However, in SAR images, it is difficult to detect coastline because of the complexity of speckle, sea breeze and land environment. From the perspective of super-pixel, using region merging and triple Markov random field (TMF), this paper focuses on the coastline detection algorithm in SAR images where there are small land and sea similar areas and low contrast between land and sea. The main work of this paper is as follows: (1) an improved shoreline detection algorithm based on super-pixel region merging is proposed. When the sea surface or land is not uniform, the existing algorithms for detecting the combined shoreline are prone to small areas which can not be merged and need to set the merging threshold artificially. In this paper, an improved super-pixel algorithm is proposed. By constructing an improved local window, we can effectively solve the problem of feature ambiguity caused by the edges in the traditional rectangular window, and construct a similarity descriptor through the local window to make the extracted features more accurate. The edge of the superpixel fits more closely. Then an improved region merging criterion is presented, which considers the pixel mean, relative size and statistic information of the super-pixel, and then obtains the method of determining the local threshold according to the neighborhood information. The problem of artificial threshold setting in existing algorithms is solved. The effectiveness of the algorithm is verified by experiments on SAR images. (2) an improved coastline detection algorithm based on super-pixel triple Markov random field (TMF) is proposed. In order to solve the problem that the edge sticker degree of the existing algorithms is not high, a super-pixel algorithm based on Gamma distribution is proposed. Assuming that the uniform local neighborhood obeys Gamma distribution, the left and right threshold is determined by confidence interval to form super-pixels to improve the edge sticker degree. Then an improved TMF algorithm based on the superpixel is presented and an energy function based on the improved auxiliary field is constructed to solve the threshold dependence problem of the existing TMF algorithm. Experiments on SAR images show that the algorithm is effective. (3) A probabilistic super-pixel triple Markov random field (TMF) improved shoreline detection algorithm is proposed. In order to solve the problem of low contrast between land and sea, a statistic is presented to enhance the contrast between land and sea, and an improved super-pixel algorithm is proposed to solve the problem of low edge matching. Finally, a probabilistic TMF improved algorithm is proposed to solve the problem that the existing algorithms do not consider the similar super-pixel correlation problem. A probability vector is used to represent the probability of the current super-pixel belonging to the sea or land. An improved energy function based on probability is constructed to replace the Ising model in existing TMF. The effectiveness of the algorithm is verified by experiments on SAR images.
【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】：TN957.52

【相似文献】