视觉注意机制下面向对象高分辨率遥感影像建筑物提取
发布时间：2019-02-15 10:00
【摘要】：随着航空航天技术的发展,人们获取的遥感影像数据的空间分辨率不断提高,同时获取难度大幅降低。在遥感数据日益普及的今天,利用影像处理与分析技术对高空间分辨率遥感影像中的感兴趣目标进行自动提取已经成为目前遥感领域研究的热点之一。建筑物作为与人类生活密切相关的主要人工地物,是城市发展的重要标志。目前,通过对高分辨率遥感影像进行建筑物自动提取已经成为建筑物信息获取的重要手段,而建筑物信息可用于进一步研究城市的扩张与发展、城市土地利用现状与变化、城市规划、城市热岛效应、人口估计与预测、灾害监测预警与评估等。 论文以基于信息理论的视觉注意机制和面向对象的影像分析方法为基础,采用自底向上的遥感影像低层特征提取和自顶向下的先验知识指导相结合的方式,提出了一种视觉注意机制下面向对象的高分辨率遥感影像建筑物提取框架。论文的主要研究工作包括以下几个方面： (1)遥感影像色彩一致性处理方法研究。 遥感影像中的不均匀光照现象是一种影像降质现象,会在一定程度上增加影像解译的难度。为了消除不均匀光照现象的影响,提高遥感影像建筑物提取的精度,论文对遥感影像色彩一致性处理方法展开了研究。针对传统的色彩一致性处理算法会导致遥感影像中地物颜色失真这一问题,论文提出了一种顾及色调保持的HSV彩色空间小波域增强处理算法。算法首先将输入影像变换到HSV彩色空间,然后对影像的亮度分量和饱和度分量进行小波分解,对小波系数采取低频分量抑制、次高频分量保持、高频分量增强的处理。在此基础上,针对薄云导致的光照不均匀现象,论文分析了薄云覆盖下的遥感影像成像机理,提出了对薄云覆盖影像的亮度分量和饱和度分量进行针对性处理的匀光算法。 (2)面向对象的高分辨率遥感影像多尺度分割方法研究。 面向对象的多尺度分割是面向对象的影像分析方法的前提和核心。如何描述遥感影像的低层特征,并结合多特征对影像进行面向对象的多尺度分割是现今面向对象分割算法的研究热点。论文首先分析了传统的影像分割方法以及面向对象的多尺度分割方法。然后,在比较了影像纹理特征描述方法的基础上,论文提出了一种基于时频分析的遥感影像纹理特征描述方法,并基于该方法构造了影像对象的纹理异质度的度量方法,采用自底向上的区域合并作为多尺度分割策略,提出了一种结合纹理特征的面向对象遥感影像多尺度分割算法。最后,论文从影像的边缘特征出发,提出了一种遥感影像的边缘强度描述方法以及边缘合并代价准则,并与异质度准则相结合,提出了结合光谱、形状、纹理和边缘特征的面向对象的多尺度分割算法。 (3)视觉注意机制下面向对象的建筑区提取方法研究。 建筑区提取可以作为建筑物提取中粗提取的过程,也可以将其提取结果作为建筑物的辅助场景信息。论文首先分析了高分辨率遥感影像中建筑区所特有的纹理表现,基于论文提出的多尺度多方向的纹理描述方法并针对建筑区的纹理特性进行建筑区纹理增强。然后以建筑区纹理特征作为主要低层特征,采用基于信息理论的视觉注意模型,模拟人类视觉系统感知环境的过程,提出了一种视觉注意机制下的建筑区显著指数(Built-up Areas Saliency Index,BASI)。最后基于面向对象的影像分析方法,对遥感影像进行顾及建筑区纹理特征的面向对象多尺度分割,再利用建筑区显著指数对高分辨率遥感影像进行建筑区提取。 (4)面向对象的高分辨率遥感影像阴影提取方法研究。 高分辨率遥感影像中的建筑物通常伴随有建筑阴影存在,这些阴影信息可作为高分辨率遥感影像建筑物提取的辅助判别依据。论文从阴影在高分辨率遥感影像中的特点出发,分析了阴影在亮度、色调、以及各光谱波段中的特性,并以此构建阴影对象特征知识库。然后采用面向对象的影像分析方法,在HSV彩色空间中对影像进行多尺度分割,最后根据特征知识库形成判别准则提取阴影。 (5)视觉注意机制下面向对象的建筑物提取方法研究。 在高分辨率遥感影像中,建筑物往往具有高亮度、强边缘、纹理明显等特征。论文结合自顶向下的先验知识与自底向上的遥感影像低层特征,提出了一种顾及建筑物特性的遥感影像特征空间变换方法,并结合论文提出的纹理特征和边缘特征描述方法,构成了高分辨率遥感影像建筑物提取框架的主要特征。然后通过构造建筑物光谱特征指数,提出了结合光谱和纹理特征的建筑物显著指数(Building Saliency Index, BSI)。通过对预处理后的高分辨率遥感影像进行多特征结合的面向对象多尺度分割,然后计算高分辨率遥感影像的建筑区显著指数BASI和建筑物显著指数BSI,并将阴影信息和建筑区信息作为辅助判别依据,在影像的多尺度对象集合中提取建筑物对象,最后对建筑物对象进行形态学平滑处理,实现了高分辨率遥感影像的建筑物提取。
[Abstract]:With the development of the aeronautics and astronautics technology, the spatial resolution of the acquired remote sensing image data is increasing, and the acquisition difficulty is greatly reduced. With the increasing popularity of remote sensing data, the automatic extraction of the object of interest in the high-space-resolution remote sensing image by using the image processing and analysis technology has become one of the hot spots in the field of remote sensing. As the main artificial ground feature which is closely related to human life, the building is an important symbol of urban development. At present, the automatic extraction of the high-resolution remote sensing image has become an important means of building information acquisition, and the building information can be used to further study the expansion and development of the city, the current situation and the change of the urban land use, the urban planning, the urban heat island effect, Population estimates and projections, disaster monitoring and early warning and assessment. On the basis of the visual attention mechanism based on the information theory and the object-oriented image analysis method, the paper adopts the combination of the low-level feature extraction of the remote sensing image and the top-down prior knowledge guidance. In this paper, an object-oriented high-resolution remote sensing image building extraction under the visual attention mechanism is presented. The main research work of the paper includes the following Aspect: (1) The color consistency of remote sensing image The non-uniform illumination in remote sensing image is an image degradation phenomenon, which can increase to a certain extent. In order to eliminate the influence of the non-uniform illumination, the accuracy of the extraction of the remote sensing image is improved, and the color consistency of the remote sensing image In order to solve the problem of the color distortion in the remote sensing image, the paper presents a small HSV color space, which takes into account the color tone, which can cause the color distortion of the ground objects in the remote sensing image. The method comprises the following steps of: firstly, transforming an input image into an HSV color space, and then carrying out small wave decomposition on the luminance component and the saturation component of the image, In this paper, the image forming mechanism of remote sensing image under the cover of thin cloud is analyzed, and the luminance component and the saturation component of the thin cloud cover image are put forward. a uniform light algorithm for sexual processing. (2) object-oriented high-resolution remote sensing Research on multi-scale image segmentation. The object-oriented multi-scale segmentation is oriented to The premise and the core of the image analysis method are as follows: how to describe the low-level features of the remote sensing image and the object-oriented multi-scale segmentation of the image in combination with the multi-feature is the present In this paper, the research focus of the object-oriented segmentation algorithm is discussed. The traditional image segmentation method is first analyzed. Then, on the basis of comparing the feature description of the image, the paper presents a method for describing the texture of the remote sensing image based on the time-frequency analysis, and based on the method, the paper presents a method to describe the texture of the remote sensing image. A method for measuring the texture heterogeneity of an object, using the region merging from the bottom up as the multi-scale segmentation strategy, presents a kind of surface combining the texture features. A multi-scale segmentation algorithm for remote sensing images of objects is presented. At last, from the edge of the image, a method for describing the edge strength of a remote sensing image and an edge combination cost criterion are presented, and the combined spectrum, shape, texture and edge are proposed. An object-oriented multi-scale segmentation algorithm for sign; (3) visual attention The method for extracting the object-oriented building area under the mechanism is as follows: the extraction of the building area can be used as the process of rough extraction in the extraction of the building, This paper first analyzes the texture performance of the building area in the high-resolution remote sensing image, and the multi-scale multi-direction texture description method based on the paper. A visual attention model based on information theory and a visual attention model based on information theory are used to simulate the perception environment of human vision system. Finally, based on the object-oriented image analysis method, the object-oriented multi-scale segmentation of the texture features of the building area is taken into account for the remote sensing image, and then the building area is used. the high-resolution remote sensing image is extracted by the saliency index. 4) The object-oriented high-resolution remote sensing image shadow extraction method is studied in this paper. The image information can be used as an auxiliary judgment basis for high-resolution remote sensing image building extraction. The paper analyzes the characteristics of the shadow in the high-resolution remote sensing image, and analyzes the brightness and the color tone of the shadow. and building a knowledge base of the shadow object by using the object-oriented image analysis method. line multi-scale segmentation, and finally, a discrimination criterion is formed according to the feature knowledge base to extract the cathode, Shadow. (5) The method of object-oriented building extraction under the mechanism of visual attention. In the high-resolution remote sensing image, the buildings often have the characteristics of high brightness, strong edge and texture. Combining the top-down prior knowledge and the lower-level feature of the bottom-up remote sensing image, a remote sensing system that takes into account the characteristics of buildings is proposed. image feature space transform method and its texture feature in combination with that present pap The method of edge characterization constitutes the main feature of the extraction frame of high-resolution remote sensing image. Then, by constructing the spectral characteristic index of the building, a building with spectral and texture features is proposed. Object-oriented multi-scale segmentation based on multi-feature combining of pre-processed high-resolution remote sensing images, and then a significant index of the building area of high-resolution remote sensing images (BASI) and finally, the building object is extracted in the multi-scale object set of the image,
【学位授予单位】：武汉大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：TP751
本文编号：2423221
[Abstract]:With the development of the aeronautics and astronautics technology, the spatial resolution of the acquired remote sensing image data is increasing, and the acquisition difficulty is greatly reduced. With the increasing popularity of remote sensing data, the automatic extraction of the object of interest in the high-space-resolution remote sensing image by using the image processing and analysis technology has become one of the hot spots in the field of remote sensing. As the main artificial ground feature which is closely related to human life, the building is an important symbol of urban development. At present, the automatic extraction of the high-resolution remote sensing image has become an important means of building information acquisition, and the building information can be used to further study the expansion and development of the city, the current situation and the change of the urban land use, the urban planning, the urban heat island effect, Population estimates and projections, disaster monitoring and early warning and assessment. On the basis of the visual attention mechanism based on the information theory and the object-oriented image analysis method, the paper adopts the combination of the low-level feature extraction of the remote sensing image and the top-down prior knowledge guidance. In this paper, an object-oriented high-resolution remote sensing image building extraction under the visual attention mechanism is presented. The main research work of the paper includes the following Aspect: (1) The color consistency of remote sensing image The non-uniform illumination in remote sensing image is an image degradation phenomenon, which can increase to a certain extent. In order to eliminate the influence of the non-uniform illumination, the accuracy of the extraction of the remote sensing image is improved, and the color consistency of the remote sensing image In order to solve the problem of the color distortion in the remote sensing image, the paper presents a small HSV color space, which takes into account the color tone, which can cause the color distortion of the ground objects in the remote sensing image. The method comprises the following steps of: firstly, transforming an input image into an HSV color space, and then carrying out small wave decomposition on the luminance component and the saturation component of the image, In this paper, the image forming mechanism of remote sensing image under the cover of thin cloud is analyzed, and the luminance component and the saturation component of the thin cloud cover image are put forward. a uniform light algorithm for sexual processing. (2) object-oriented high-resolution remote sensing Research on multi-scale image segmentation. The object-oriented multi-scale segmentation is oriented to The premise and the core of the image analysis method are as follows: how to describe the low-level features of the remote sensing image and the object-oriented multi-scale segmentation of the image in combination with the multi-feature is the present In this paper, the research focus of the object-oriented segmentation algorithm is discussed. The traditional image segmentation method is first analyzed. Then, on the basis of comparing the feature description of the image, the paper presents a method for describing the texture of the remote sensing image based on the time-frequency analysis, and based on the method, the paper presents a method to describe the texture of the remote sensing image. A method for measuring the texture heterogeneity of an object, using the region merging from the bottom up as the multi-scale segmentation strategy, presents a kind of surface combining the texture features. A multi-scale segmentation algorithm for remote sensing images of objects is presented. At last, from the edge of the image, a method for describing the edge strength of a remote sensing image and an edge combination cost criterion are presented, and the combined spectrum, shape, texture and edge are proposed. An object-oriented multi-scale segmentation algorithm for sign; (3) visual attention The method for extracting the object-oriented building area under the mechanism is as follows: the extraction of the building area can be used as the process of rough extraction in the extraction of the building, This paper first analyzes the texture performance of the building area in the high-resolution remote sensing image, and the multi-scale multi-direction texture description method based on the paper. A visual attention model based on information theory and a visual attention model based on information theory are used to simulate the perception environment of human vision system. Finally, based on the object-oriented image analysis method, the object-oriented multi-scale segmentation of the texture features of the building area is taken into account for the remote sensing image, and then the building area is used. the high-resolution remote sensing image is extracted by the saliency index. 4) The object-oriented high-resolution remote sensing image shadow extraction method is studied in this paper. The image information can be used as an auxiliary judgment basis for high-resolution remote sensing image building extraction. The paper analyzes the characteristics of the shadow in the high-resolution remote sensing image, and analyzes the brightness and the color tone of the shadow. and building a knowledge base of the shadow object by using the object-oriented image analysis method. line multi-scale segmentation, and finally, a discrimination criterion is formed according to the feature knowledge base to extract the cathode, Shadow. (5) The method of object-oriented building extraction under the mechanism of visual attention. In the high-resolution remote sensing image, the buildings often have the characteristics of high brightness, strong edge and texture. Combining the top-down prior knowledge and the lower-level feature of the bottom-up remote sensing image, a remote sensing system that takes into account the characteristics of buildings is proposed. image feature space transform method and its texture feature in combination with that present pap The method of edge characterization constitutes the main feature of the extraction frame of high-resolution remote sensing image. Then, by constructing the spectral characteristic index of the building, a building with spectral and texture features is proposed. Object-oriented multi-scale segmentation based on multi-feature combining of pre-processed high-resolution remote sensing images, and then a significant index of the building area of high-resolution remote sensing images (BASI) and finally, the building object is extracted in the multi-scale object set of the image,
【学位授予单位】：武汉大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：TP751
【引证文献】
相关博士学位论文 前2条
1 康一飞;光学遥感卫星影像云检测方法及应用[D];武汉大学;2018年
2 李政;基于无人机高分影像的空心村建筑物信息获取关键技术研究[D];西南交通大学;2018年
相关硕士学位论文 前3条
1 党涛;基于多层次分割分类模型及其特征空间优化的高分辨率遥感影像城市建筑物提取研究[D];兰州大学;2018年
2 张静敏;基于超级像素的居民地自动提取研究[D];东华理工大学;2017年
3 朱姝;面向对象的高分遥感影像建筑物提取方法研究[D];成都理工大学;2017年
，本文编号：2423221
本文链接：https://www.wllwen.com/guanlilunwen/gongchengguanli/2423221.html
