高分辨率遥感影像居民地提取方法研究

发布时间：2018-08-11 17:11

【摘要】：摘要：居民地是人类居住和进行各种日常活动的中心场所。在高分辨遥感影像中,居民地通常由密集的建筑物群,内部的绿地,以及周边的道路交通网所构成。准确有效地获取居民地的实时信息,在“城镇化”建设、数字城市、城市规划、土地利用及GIS系统更新等多个领域都具有重要的现实及经济意义。为此,本文充分利用高分辨率遥感影像上居民地特有的局部特征,进行了居民地提取和分类方面的探索和尝试。总体而言,本文主要包括以下三个方面的研究工作： (1)基于边缘密度特征的高分辨率遥感影像居民地提取利用影像上居民地与非居民地的边缘密度特征差异提取居民地,首先对影像进行滤波平滑预处理,其次检测影像上的边缘特征并将其拟合成直线段,然后计算影像上像素点到所有边缘直线段的空间距离,最后利用高斯函数量化边缘密度并最终阈值分割提取居民地。该方法是一种全自动的居民地提取方法,有效地避免了由于人为因素对提取结果带来的影响,提高了居民地提取的精度。 (2)基于Gabor特征的高分辨率遥感影像居民地提取居民地内部除了包含丰富的边缘特征之外,同时也具有密集的角点特征,而在角点处往往会出现明显的灰度梯度和曲率变化,Gabor滤波对于这种变化具有较强的响应。据此,本文首先利用Gabor滤波对影像做多尺度多角度的变换,其次检测滤波图像上的Gabor特征并优化,最后构建特征的空间投票矩阵并结合OStu阈值分割方法提取居民地。该方法同样是一种非监督的提取方法,相对于基于边缘密度特征的提取方法,在运行效率和居民地提取精度上都有一定的提高。 (3)高分辨率遥感影像城镇及乡村居民地监督分类利用以上两种方法可以有效地提取高分辨率遥感影像上的居民地,但是未能对城镇和乡村居民地进行更进一步的分类,因此不能准确地体现出城乡之间的发展变化信息。为此,本文在以上两种居民地提取方法的理论基础上,充分利用城镇和乡村居民地的边缘特征及Gabor特征的分布差异,发展了一种城镇及乡村居民地监督分类方法。首先设计了五种可以体现城乡居民地边缘特征和Gabor特征分布差异的分类规则,然后构建训练样本集对各类规则进行学习,最后通过大量的测试样本验证以上规则的分类精度。由于该方法只是作为城镇居民地与乡村居民地初级分类的一个探索,中间过程还不够完善,因此所取得的分类精度有限,但是该方法具有独创性意义。
[Abstract]:Absrtact: residential land is the central place for human beings to live and carry out various daily activities. In high resolution remote sensing images, residential land is usually composed of dense buildings, inner green space, and surrounding road traffic network. It is of great practical and economic significance to obtain the real time information of residents' land accurately and effectively in the fields of "urbanization" construction, digital city, urban planning, land use and GIS system update. Therefore, this paper makes full use of the local characteristics of residents in high resolution remote sensing images, and explores and tries to extract and classify residents. Overall, This paper mainly includes the following three aspects of research work: (1) based on the edge density characteristics of high-resolution remote sensing images of residents to extract and use the image of the edge density of residents and non-residents of the special density Differential extraction of residential land, First, the image is processed by filtering smoothing, then the edge features of the image are detected and synthesized into straight line segments, and then the spatial distance between pixels on the image and all edge line segments is calculated. Finally, the edge density is quantized by Gao Si function and the final threshold segmentation is used to extract the resident land. This method is a fully automatic extraction method for residents, which effectively avoids the influence of human factors on the extraction results. (2) the high resolution remote sensing image based on Gabor features not only contains rich edge features, but also has dense corner features. But there are obvious grayscale gradient and curvature change at the corner point. Gabor filter has strong response to this change. Based on this, this paper firstly uses Gabor filter to transform the image from multi-scale and multi-angle, then detects and optimizes the Gabor features on the filtered image. Finally, the spatial voting matrix of the feature is constructed and the residential area is extracted by using the OStu threshold segmentation method. This method is also an unsupervised extraction method, compared with the edge density feature extraction method. (3) the high resolution remote sensing image can be effectively extracted from the urban and rural areas by using the two methods of urban and rural land supervision and classification. (3) the two methods mentioned above can be effectively used to extract high resolution. (3) the two methods mentioned above can be effectively used to extract high resolution from urban and rural areas. The inhabitants of the remote sensing image, However, there is no further classification of urban and rural land, so it can not accurately reflect the information of development and change between urban and rural areas. Therefore, on the basis of the above two land extraction methods, this paper develops a land supervision classification method for urban and rural residents by making full use of the distribution differences between the edge characteristics and Gabor characteristics of urban and rural residential land. Firstly, five kinds of classification rules are designed, which can reflect the difference between urban and rural residents' marginal features and Gabor characteristics. Then, the training sample set is constructed to study the rules. Finally, a large number of test samples are used to verify the classification accuracy of the above rules. Because this method is only an exploration of the primary classification of urban and rural areas, and the intermediate process is not perfect, the accuracy of classification is limited, but the method is of original significance.
【学位授予单位】：中南大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：P237

【参考文献】