单幅遥感影像超分辨率重建及分类研究

发布时间：2018-06-12 08:09

本文选题：超分辨率重建 + 超松弛迭代　；参考：《中南林业科技大学》2014年硕士论文

【摘要】：图像超分辨率重建是指利用同一场景的不同观测角度、不同观测时间或不同传感器获取的低分辨率图像序列重建出一幅或多幅高空间分辨率图像的技术。本论文对单幅图像的超分辨率重建算法和遥感图像的分类进行了研究。在单幅图像超分辨率重建的研究过程中,提出了K均值结合局部线性嵌入的重建方法。由于图像分块后,很多图像块是类似的,对分块图像先进行聚类,分成不同的类,重构时根据相似性分别从不同类中抽取一个或几个图像块,然后利用局部线性嵌入对分类后的图像块进行重构,可提高重构速度及效果。分别对普通数字图像和遥感图像运用插值法、局部线性嵌入算法,以及K均值结合局部线性嵌入算法进行多次实验,证明K均值结合局部线性嵌入重建方法重建结果较理想。在遥感图像的分类研究中,提出基于超松弛迭代的模糊K均值算法。在模糊K均值算法中引入超松弛迭代算法,在不牺牲分类精度的情况下,很大程度上提高了收敛速度。该算法适合于对大幅面遥感影像的分类。
[Abstract]:Image super-resolution reconstruction is a technique to reconstruct one or more high-spatial resolution images by using low-resolution image sequences obtained from different observation angles, different observation times or different sensors of the same scene. In this paper, the super-resolution reconstruction algorithm of single image and the classification of remote sensing image are studied. In the process of super-resolution reconstruction of single image, a reconstruction method based on K-means and local linear embedding is proposed. Because many image blocks are similar after the image is divided into blocks, the block images are first clustered into different classes, and one or more blocks are extracted from different classes according to the similarity of reconstruction. Then local linear embedding is used to reconstruct the classified image blocks, which can improve the reconstruction speed and effect. The interpolation method, local linear embedding algorithm and K-means combined with local linear embedding algorithm are used to reconstruct ordinary digital images and remote sensing images respectively. It is proved that the reconstruction results of K-means combined with local linear embedding are satisfactory. In the research of remote sensing image classification, a fuzzy K-means algorithm based on overrelaxation iteration is proposed. The overrelaxation iterative algorithm is introduced into the fuzzy K-means algorithm, which improves the convergence speed to a great extent without sacrificing the classification accuracy. This algorithm is suitable for classification of large format remote sensing images.
【学位授予单位】：中南林业科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP751

【参考文献】