面向对象的高分辨率SAR影像变化检测方法研究

发布时间：2018-04-15 00:20

本文选题：面向对象 + 多尺度分割　；参考：《山东科技大学》2017年硕士论文

【摘要】：遥感影像变化检测是通过对比同一地区不同时相的影像数据进行地球环境和安全监测的技术。由于合成孔径雷达(Synthetic Aperture Radar,SAR)全天时全天候的成像模式,相较于光学遥感有明显优势,在变化检测领域中发挥重要作用。高分辨率SAR影像变化检测技术被广泛应用在农业调查、城市扩展监测、自然灾害评估等多个领域,为科学分析和决策提供保障和服务,已经成为国家发展规划的重要研究内容。目前SAR影像非监督变化检测方法主要分为两类:像素级变化检测和对象级变化检测。像素级变化检测方法对不同影像间辐射度差别、噪声等因素比较敏感,并且像素级变化检测方法忽略了目标的整体性。而对象级变化检测方法以图斑为影像分析的处理单元,更符合人类认知事物的思维逻辑,并且可综合利用对象的强度信息、空间信息和纹理信息。为此,本文以面向对象的思想为研究基础,充分利用SAR影像丰富的细节信息,以提高SAR影像的变化检测精度为目的进行深入研究,提出了基于图斑对象均值对数比的变化检测算法和灰度直方图和LBP纹理直方图的G统计量距离自适应加权的变化检测方法。本文主要工作和创新点如下:1.总结变化检测方法的研究现状并分析存在的问题,介绍SAR影像特征和常用的SAR影像变化检测预处理方法,分析常用变化检测方法及其优缺点,介绍SAR影像变化检测精度评价方法。2.归纳总结SAR影像分割方法,重点介绍了分形网络演化分割方法。本文在SAR影像分割前采用非局部三边滤波方法,相比传统增强Frost滤波,取得了更好的分割结果。3.提出了一种基于图斑对象均值对数比的变化检测算法。传统的对数比值法以单个像素或者像素的中心邻域为基本处理单元,在高分辨率SAR影像进行变化检测时,像素级对数比值法存在大量检测虚警,邻域均值对数比值法损失细节信息,检测结果边缘模糊。本文采用面向对象的思想,SAR影像分割后进行矢量联合,以合并后的图斑为基本处理单元,计算各图斑的灰度均值,图斑均值进行对数比值运算得到差异图,该算法可避免斑点噪声的干扰,同时保留了变化区域的边缘信息。4.提出了灰度直方图和LBP纹理直方图的G统计量距离自适应加权的变化检测方法。针对均值对数比值法仅仅依靠SAR影像的灰度信息,在对建筑物进行变化信息检测时,SAR影像中建筑物由于高度的影响存在阴影现象,此时该算法检测效果不佳的特点。本文采用面向对象的方法,以图斑对象为基本的处理单元,计算灰度直方图和LBP纹理直方图,用G统计量计算直方图距离,将灰度直方图G统计量距离和LBP纹理直方图G统计量距离进行自适应加权确定差异图,采用CFAR算法对差异图进行阈值分割得到变化二值图。
[Abstract]:Change detection of remote sensing image is the earth's environment and safety monitoring through the image data of the same area in different time contrast technology. Because the synthetic aperture radar (Synthetic Aperture, Radar, SAR) imaging mode of all-weather and all day long, compared with optical remote sensing has obvious advantages, play an important role in the field of change detection of high resolution changes. SAR image detection technology is widely used in agricultural research, city extension monitoring, many areas of natural disaster assessment, provide support and services for the scientific analysis and decision-making, has become an important research content in the national development plan. The current SAR image detection method of supervision changes are mainly divided into two categories: pixel level change detection and object level change detection. Detection method of pixel level change on different images of different radiation, noise and other factors more sensitive, and the pixel level change detection method is ignored The overall target. While the object level change detection method for patch image analysis processing unit, more in line with the logic of human cognition, and intensity information of comprehensive utilization of objects, spatial information and texture information. Therefore, based on the object-oriented thought as the research basis, make full use of the rich details of the SAR image the changes of SAR images to improve the detection accuracy for the purpose of in-depth research, put forward the G statistic distance map spot changes mean logarithmic ratio detection algorithm and gray histogram and LBP texture histogram adaptive detection method based on weighted changes. This paper mainly work and innovation are as follows: 1. summarize the research status change detection method and analyze the existing problem, the pretreatment method of SAR features and SAR image change detection and analysis of common change detection methods and their advantages and disadvantages are introduced. SAR image change detection method for accuracy evaluation of.2. induction methods of SAR image segmentation, introduces the segmentation method of fractal net evolution. Based on the SAR image segmentation by non local three edge filtering method, compared with the traditional enhanced Frost filter has better segmentation results of.3. proposed a map object changes than the logarithmic mean the detection algorithm based on logarithmic ratio method. The traditional neighborhood center to a single pixel or pixel as the basic processing unit, change detection in SAR images with high resolution, pixel level log ratio method has a large number of false alarm detection, neighborhood average logarithmic ratio method the loss of detail information, fuzzy edge detection results. This paper uses the object oriented thinking SAR image segmentation after vector combination, combined with patches as the basic processing unit, the calculation of gray mean patch, patch mean logarithmic ratio. The difference is, this algorithm can avoid the interference of speckle noise, while preserving the edge information.4. change area detection method is proposed gray histogram and LBP texture histogram G statistic distance adaptive weighted changes. According to mean logarithmic ratio method only depend on the gray information of SAR image, the change detection of building SAR, image building due to a high degree of influence of the shadow phenomenon, the characteristics of poor detection effect of the algorithm. This paper uses the method of object oriented, patch object is the basic processing units, calculate the gray histogram and LBP texture histogram, histogram distance calculated by G statistics, the gray histogram and LBP texture histogram statistics from G the G statistic distance by adaptive weighted CFAR algorithm is used to determine the difference map segmentation obtained two value image of difference map.

【学位授予单位】：山东科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：P237;TP751

【参考文献】