SAR图像目标分割活动轮廓模型及其算法的研究

发布时间：2018-03-08 12:09

本文选题：活动轮廓　切入点：梯度流　出处：《天津理工大学》2017年硕士论文　论文类型：学位论文

【摘要】：合成孔径雷达(Synthetic Aperture Radar,SAR)由于具有不受光照、时间及气候等客观条件影响的优点,使得它广泛应用于很多领域。然而因为SAR系统成像机理的特殊性和不同物体表面后向散射的多样性,使得SAR图像的处理工作变得极其困难。而SAR图像中固有的斑点噪声更加剧了图像中目标边缘的模糊程度。SAR图像的目标边缘检测是SAR图像处理领域中的重要研究内容,其轮廓提取精度的高低直接影响着图像解译的质量。因此,设计一种高效、高精度的SAR图像轮廓提取算法是非常有必要的。而现存的SAR图像分割方法往往含有算法复杂度高、噪声抑制能力差、提取精度低等缺点。为了解决以上问题对实际应用造成的困扰,本文开展了以下研究工作:1)首先使用Gamma分布来描述SAR图像的统计特性,其次,针对Ayed模型算法复杂度低而抑制噪声能力差、局部高斯分布拟合能量(local Gaussian distribution fitting energy,LGDF)模型算法复杂度高而抑制噪声能力强的特点,结合两者的优点,提出了基于区域的活动轮廓分割模型,该模型抑制噪声能力强且算法复杂度低。最后通过合成和真实SAR图像的实验结果验证了提出方法的有效性。2)针对现存方法处理SAR图像尖锐边缘和弱边缘能力较差的问题,在上述研究内容的基础上,提出了基于边缘-区域的活动轮廓模型。该模型包括两项:边缘-区域项和正则项。边缘-区域项由Gamma模型和梯度项模型导出,主要用于处理斑点噪声和牵引曲线朝着期望位置运动;而正则项不仅能够规范演变曲线的形状,同时也具有较强地平滑作用从而避免小的孤立区域的出现。并且利用梯度下降流方法来最小化所提出的能量泛函。实验结果表明该模型不仅对轮廓初始化位置不敏感,而且相比于其他方法,具有收敛速度快及良好的处理非均质区域能力。
[Abstract]:Synthetic Aperture radar (SAR) has the advantage of not being affected by objective conditions such as light, time and climate. However, because of the particularity of imaging mechanism of SAR system and the diversity of backscattering from different objects, It is very difficult to process SAR image, and the inherent speckle noise in SAR image intensifies the blur of target edge. The detection of target edge in SAR image is an important research content in the field of SAR image processing. The precision of contour extraction directly affects the quality of image interpretation. Therefore, it is necessary to design an efficient and high-precision SAR image contour extraction algorithm. However, the existing SAR image segmentation methods often contain high complexity. In order to solve the problems caused by the above problems in practical application, the following research work is carried out in this paper: firstly, Gamma distribution is used to describe the statistical characteristics of SAR images, and secondly, In view of the low complexity and poor noise suppression ability of the Ayed model algorithm, the local Gaussian distribution fitting energy model is characterized by high complexity and strong noise suppression ability, which combines the advantages of the two algorithms. A region-based active contour segmentation model is proposed. The model has strong noise suppression ability and low algorithm complexity. Finally, the experimental results of synthetic and real SAR images verify the effectiveness of the proposed method. Based on the above research, an active contour model based on edge-region is proposed. The model includes two terms: edge-region term and regular term. The edge-region term is derived from Gamma model and gradient term model. Mainly used to deal with speckle noise and traction curves moving towards the desired position, and the canonical term not only can normalize the shape of the evolution curve, It also has a strong smoothing effect to avoid the appearance of small isolated regions, and the gradient descent flow method is used to minimize the proposed energy functional. The experimental results show that the proposed model is not only insensitive to the initial position of the contour, And compared with other methods, it has fast convergence speed and good ability to deal with heterogeneous regions.
【学位授予单位】：天津理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN957.52

【相似文献】