基于水平集的SAR遥感图像分割的算法研究

发布时间：2018-07-26 13:23

【摘要】：合成孔径雷达(SAR)是一种高分辨的微波遥感相干成像雷达,在军事和国民经济等各个领域中都有着非常重要的作用。SAR遥感图像的分割是进行SAR遥感图像理解、解疑中基本且关键的技术之一。SAR遥感图像分割的目的就是把目标区域和背景区域分割开来,但由于SAR遥感图像中含有大量乘性相干斑噪声,且图像区域灰度分布不均匀,使得SAR遥感图像中目标物体边缘无法被精确定位,进而很难实现对SAR遥感图像精确且高效率的分割。如何快速而有效地实现SAR遥感图像的分割,是目前亟待解决的一个难题。随着SAR遥感图像研究的发展,水平集模型以其对曲线拓扑结构变化的良好适应能力和无需对噪声预处理的特性,受到国内外研究学者们的青睐。本文在总结和分析已有的基于水平集的SAR遥感图像分割方法的基础上,针对SAR遥感图像所具有的大量乘性相干斑噪声和灰度分布不均匀的特性,提出了两种融合区域信息和边缘梯度信息的水平集模型,对SAR遥感图像进行分割,主要工作如下:针对SAR遥感图像中目标边缘模糊和对目标边缘定位不正确的问题,提出了一种基于改进C-V模型的高分辨率SAR遥感图像的分割方法。该方法针对C-V模型不能分割灰度不均匀图像的缺点,以及该模型只利用区域信息而没有利用边缘梯度信息,从而造成分割后的目标物体虚假边缘较多的缺点,本文利用SAR遥感图像所特有的统计特性,提出了利用对均匀和不均匀区域都有很好拟合作用的G0分布函数,对图像进行拟合,解决对灰度分布不均匀图像分割不准确的问题,同时在C-V模型中引入改进的边缘指示函数,此边缘指示函数能够很好地去除SAR遥感图像中具有的乘性噪声、定位目标的边界、控制曲线的演化速率以及避免水平集函数的重新初始化。针对SAR遥感图像存在的灰度分布不均匀现象,提出了一种基于改进LIF模型的SAR遥感图像的分割方法。该方法是在LIF模型能较好地分割灰度不均匀图像的基础上,针对局部图像拟合(LIF)模型存在的对噪声敏感,以及在演化过程中易陷入局部极小值和边缘定位不准确的缺点,引入截断的基于线性最小均方误差的指数平滑滤波器来提高分割精度,同时引入结合了模糊C均值(FCM)和无限对指数滤波器的,基于梯度信息和全局区域信息的边缘检测函数,来避免陷入局部最优和边界定位不准的问题。利用人工合成的图像和真实的道路、湖泊以及舰船的高分辨率SAR遥感图像进行分割实验,对比已有的基于水平集的SAR遥感图像分割方法,证明了本文的两种改进水平集方法都能够在背景杂波下,很好地抑制乘性相干斑噪声,准确地定位目标物体的边缘轮廓,提高对SAR遥感图像的分割精度。
[Abstract]:Synthetic Aperture Radar (SAR) is a kind of high-resolution microwave remote sensing coherent imaging radar, which plays an important role in military and national economy. One of the basic and key techniques of SAR remote sensing image segmentation is to separate the target region from the background area. However, there are a lot of multiplicative speckle noises in the SAR remote sensing image and the gray distribution of the image region is not uniform. The edge of object in SAR remote sensing image can not be accurately located, and it is difficult to segment SAR image accurately and efficiently. How to quickly and effectively realize the segmentation of SAR remote sensing image is a difficult problem to be solved. With the development of SAR remote sensing image research, the level set model is favored by researchers at home and abroad because of its good adaptability to the curve topology change and no need for noise preprocessing. On the basis of summarizing and analyzing the existing SAR remote sensing image segmentation methods based on level set, this paper aims at the multiplicative speckle noise and uneven gray distribution of SAR remote sensing image. In this paper, two level set models for fusion of regional information and edge gradient information are proposed. The main work of segmentation of SAR remote sensing image is as follows: aiming at the problem of target edge blur and target edge location incorrectly in SAR remote sensing image, A high resolution SAR remote sensing image segmentation method based on improved C-V model is proposed. This method aims at the disadvantage that C-V model can not segment uneven grayscale image, and the model only uses the region information but not the edge gradient information, which results in more false edges of the target object after segmentation. In this paper, based on the statistical characteristics of SAR remote sensing images, a G0 distribution function, which can fit both uniform and non-uniform regions, is proposed to fit the images and to solve the problem of inaccurate segmentation of non-uniform gray-scale images. At the same time, an improved edge indicator function is introduced into the C-V model. The edge indicator function can remove the multiplicative noise in the SAR remote sensing image and locate the boundary of the target. The evolution rate of the control curve and the reinitialization of the level set function are avoided. Aiming at the uneven gray distribution of SAR remote sensing images, a method of SAR remote sensing image segmentation based on improved LIF model is proposed. This method is based on the fact that the LIF model can segment inhomogeneous grayscale images well, and aims at the disadvantages of local image fitting (LIF) model, which is sensitive to noise and easy to fall into local minimum value and inaccurate edge location in the evolution process. The truncated exponential smoothing filter based on linear minimum mean square error is introduced to improve the segmentation accuracy, and the edge detection function based on gradient information and global region information is introduced, which combines fuzzy C-means (FCM) and infinite pair exponential filter. To avoid the problem of local optimum and inaccurate boundary location. Using artificial synthetic image and real road, lake and ship high resolution SAR remote sensing image segmentation experiment, compare the existing SAR remote sensing image segmentation method based on level set. It is proved that the two improved level set methods in this paper can effectively suppress multiplicative speckle noise under background clutter, accurately locate the edge contour of the target object, and improve the segmentation accuracy of SAR remote sensing images.
【学位授予单位】：江苏科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP751

【参考文献】