【作者】 王欢；

【导师】 王建华；

【作者基本信息】 哈尔滨师范大学 ， 计算机科学与技术， 2014， 硕士

【摘要】 1972年，美国发射了首颗地球观测卫星—landsat-1，标志着人类进入了遥感时代的新纪元。随着遥感技术的迅猛发展和广泛应用，虽然遥感图像的分辨率已经不断提高但还是难以满足许多领域对于高分辨率遥感影像的需求，比如在遥感图像小目标（飞机、船只、桥梁等）识别当中，需要寻找的目标通常都是米级目标，利用现有的遥感图像，进行小目标分割精度偏低。提高遥感图像的分辨率可以通过改良硬件设备实现，但工艺水平越复杂精细，成本也会越高昂，并且成像设备改良的空间也是有极限的，人们开始尝试利用软件的方法解决这一问题，超分辨率重建技术在这个背景下应运而生。本文利用超分辨率重建技术提高现有遥感图像的分辨率，提出了一种基于的超分辨率重建方法，为目标分割提供包含更丰富有效特征的遥感图像源数据，将超分辨率重建技术运用于小目标识别，并通过实验成功提出小目标（塘坝）。本文对超分辨率重建技术关键技术做了较为深入的研究，主要包括以下几个方面：遥感图像配准，高精度的图像配准是完成超分辨率重建的基础，本文介绍了现在常用的图像配准技术，并分析了现有遥感配准技术的难点问题，针对小目标遥感图像的特殊性采用一种基于SIFT的多光谱遥感图像配准方法，通过实验验证本文方法可以快速、自动的完成高精度配准。遥感图像超分辨率重构，目前序列超分辨率算法主要分为频域类和空域类两类方法。本文概述了两类方法的主要算法原理和优点，并提出一种基于Hopfield神经网络的超分辨重建算法，在遥感图像超分辨率重建的结果中，往往图像的边界等细节容易产生模糊，而这些部分又包含较多重要的信息，该方法可以有效提高细节保护的能力，通过实验取得较好的超分辨效果。将超分辨率技术运用在小目标分割中，小目标分割的精度不仅依赖于目标分割方法，高分辨率的遥感图像作为源数据也是重要因素之一，高分辨率的遥感图像能够为目标识别提供更多有效特征，提高识别精度。本文将超分辨率技术运用于目标分割中，通过实验完了对小目标（塘坝）的目标识别。更多还原

【Abstract】 In1972,the United States launched the first earth observation satellite-landsat1,which marked the human entered the new era of remote sensing.With the rapiddevelopment of remote sensing technology and the widespread application,although theresolution of the remote sensing image has been improving but still difficult to meet theneeds of many areas for high resolution remote sensing image,such as during smalltarget in remote sensing image (planes,ships,Bridges,etc.) to identify,we need to find thetarget level is usually m and use of the existing remote sensing images for small targetsegmentation accuracy is low. Improve the resolution of remote sensing images can beimproved to hardware implementation,but the more elaborate have the craft level,costwill be high,and imaging equipment improved space is limited,people started to try themethod of using software to solve this problem,super-resolution reconstructiontechnology arises under the background of this historic moment.Based on super-resolution reconstruction technology improve the resolution of theexisting remote sensing images,this paper proposes a super-resolution reconstructionmethods based on,to provide target segmentation contains more abundant features ofremote sensing image data source effectively,the super-resolution reconstructiontechnology for small target recognition,and through the experiment (dam) small target isput forward successfully.In this paper,the super-resolution reconstruction of key technology to do a thoroughresearch,mainly includes the following aspects:Remote sensing image registration,the high accuracy of image registration is thebasis of Super-resolution reconstruction,this paper introduces the now commonly usedimage registration technique,and analyses the existing difficulties in registration ofremote sensing technology,direct at the particularity of the remote sensing image of thesmall target use a multispectral remote sensing image registration method based onSIFT,through those experiments in this paper,we find the method can fast andautomatically complete the high accuracy.Remote sensing image super-resolution reconstruction,the current sequencesuper-resolution algorithm is mainly divided into two categories,airspace andfrequency domain method. This paper summarizes the principle and advantages of the main algorithm of two methods,and put forward a kind of super resolutionreconstruction algorithm based on Hopfield neural network,in the remote sensingimage super-resolution reconstruction results,we often prone to detail such as theboundary of the image blur,and these part contains more important information,thismethod can effectively improve the ability of detail protection,through the experimentcan make better effect on super resolution.The super-resolution technology used in small target segmentation,thesegmentation accuracy of the small target segmentation not only depends on the targetmethod,high resolution remote sensing image as the source data is also one of theimportant factors,high resolution remote sensing images can provide more effectivefeature for target identification,improve the identification accuracy. In this paper,theapplication of super-resolution technique to target segmentation,which finished thesmall targets (dam) recognition through the experiment.更多还原