基于压缩感知的SAR图像分割研究

发布时间：2018-02-21 06:15

本文关键词： SAR图像分割压缩感知 Curvelet变换图像特征 FCM聚类　出处：《西安电子科技大学》2014年硕士论文　论文类型：学位论文

【摘要】：高分辨合成孔径雷达(Synthetic Aperture Radar,SAR)突破了外界条件影响的局限,是一种能产生高分辨率遥感图像的相干系统,具有全天候、全天时、多波段、多极化工作方式、可变侧视角和强透射性等特点,广泛应用在军事领域和国民经济。而SAR图像分割是SAR图像自动处理的基础,也是SAR图像自动理解和解释的基本关键技术之一。压缩感知理论打破了传统采样过程中信号采样速率必须达到信号带宽两倍以上才能精确重构原始信号的乃奎斯特采样定理,它将信号的采样与压缩同时进行,利用信号在变换基上可以稀疏表示的先验知识,用比香农采样少的多的观测值重构原始信号,大大降低了信息存储、处理和传输的成本。本文将压缩感知理论应用于SAR图像分割,提出两种基于压缩感知的SAR图像分割算法,并取得比较好的分割结果,具体内容和工作如下。(1)提出了一种基于压缩感知和Curvelet变换的模糊C均值(Fuzzy C-Means,FCM)聚类分割算法。首先,利用压缩感知理论中的正交匹配追踪(OMP)重构算法对SAR图像进行去噪处理并进行重构获得重构图。其次,再利用Curvelet变换在提取图像细节方面的优点,先对重构图进行Curvelet变换,再对子带系数进行Curvelet逆变换,获得细节增强的增强图。而后,利用FCM聚类算法在图像分割上的优势,将重构图和增强图融合后,得到的融合图是去噪后并且细节已经增强的融合图像,再把融合图像用FCM聚类算法进行分割。最后通过实验结果比较,在真实SAR图像上进行验证OMP_Curvelet_FCM算法的性能。(2)提出了一种基于稀疏表示和图像特征的FCM聚类分割算法。在K-SVD冗余字典进行图像稀疏表示的基础上,引入SAR图像的统计特征,并用特征构造特征字典。本算法提出一种新的稀疏表示方法,用联合字典进行稀疏表示,联合字典是有特征字典和K-SVD冗余字典相乘构造而成。在用联合字典进行稀疏表示的过程中进行去噪,并用FCM聚类算法进行分割。最后通过实验结果比较,在真实SAR图像上进行验证基于稀疏表示和图像特征的FCM聚类分割算法的有效性。
[Abstract]:High resolution synthetic Aperture radar (SAR) has broken through the limitation of external conditions and is a coherent system that can produce high-resolution remote sensing images. It has all-weather, all-weather, multi-band, multi-polarization operation mode. Variable side view and strong transmission are widely used in military field and national economy. SAR image segmentation is the basis of SAR image automatic processing. Compression sensing theory breaks the Quast sampling theorem that the signal sampling rate must be more than twice the signal bandwidth in the traditional sampling process in order to reconstruct the original signal accurately. The signal is sampled and compressed at the same time, using the prior knowledge that the signal can be expressed sparsely on the transform basis, and the original signal is reconstructed with a much smaller sample than Shannon, which greatly reduces the information storage. In this paper, we apply compression sensing theory to SAR image segmentation, propose two SAR image segmentation algorithms based on compression perception, and obtain better segmentation results. The main contents and work are as follows: (1) A fuzzy C-means fuzzy C-Means-FCM-based clustering segmentation algorithm based on compressed perception and Curvelet transform is proposed. Using the orthogonal matching tracking (OMP) reconstruction algorithm in compression perception theory, the SAR image is de-noised and reconstructed to obtain the reconstructed image. Secondly, using the advantages of Curvelet transform in extracting the image details, the recomposition is first transformed by Curvelet. Then the subband coefficients are inversely transformed by Curvelet to obtain the enhanced image of detail enhancement. Then, using the advantage of FCM clustering algorithm in image segmentation, the reconstructed and enhanced images are fused. The fusion image obtained is the fused image which has been de-noised and the details have been enhanced. Then the fused image is segmented by FCM clustering algorithm. Finally, the experimental results are compared. This paper presents a FCM clustering segmentation algorithm based on sparse representation and image features. Based on the sparse representation of K-SVD redundant dictionary, the statistical features of SAR images are introduced. In this algorithm, a new sparse representation method is proposed, which is represented by a joint dictionary. The joint dictionary is constructed by multiplying characteristic dictionaries and K-SVD redundant dictionaries. In the process of sparse representation of joint dictionaries, denoising is carried out and segmented by FCM clustering algorithm. Finally, the experimental results are compared. The validity of the FCM clustering segmentation algorithm based on sparse representation and image features is verified on real SAR images.
【学位授予单位】：西安电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN957.52

【相似文献】