基于超级像素的居民地自动提取研究

发布时间：2018-03-02 08:21

本文关键词： 居民地提取超级像素最大相似性区域合并　出处：《东华理工大学》2017年硕士论文　论文类型：学位论文

【摘要】：近年来,随着人口的不断增加,农村居民地的规模也在不断变化,如何及时准确的监测农村居民地的变化,成为亟待解决的问题。由于利用高分辨率影像提取居民地具有实地测量难以比拟的优势,其可以不受地形、海拔等各种自然条件限制。然而由于农村居民地包含道路、树木、建筑物等一系列信息,而且影像分辨率越高,其纹理、颜色信息更加丰富,虽然可将农村居民地的细节信息表现的更加清楚,但其光谱差异随之增大,传统基于光谱分类的居民地提取方法无法有效的从高分辨率影像上提取居民地。为了能够有效的从高分辨率影像中提取居民地,本文提出一种基于超级像素的全自动提取方法。该方法首先利用保拓扑的超级像素方法ETPS将影像分割成一系列具有相似特征且不破坏影像边界信息的超级像素;然后计算各超级像素的颜色直方图,并用超级像素颜色直方图的巴氏系数(Bhattacharyya Coefficient,BC)表征超级像素间相似度;最后,基于最大相似性的区域合并(MSRM)准则在地名坐标的引导下将超级像素合并为居民地区域和背景区域。利用提出方法,对0.5米和1米高分辨率卫星影像进行了居民地提取实验。结果表明,该方法可提取具有一定规模的居民地,验证了提出方法的可行性和有效性。本文的目的是探索利用超级像素方法和最大相似性的区域合并方法用于高分辨率影像居民地自动提取的可能性和存在的问题。论文首先分析比较几种典型的超级像素分割算法,然后选取比较好控制的超级像素算法作为居民地提取的方法。在居民地提取过程中,采用地名调查的地名坐标作为种子,以最大程度地减少人工交互的需要,尽可能提高居民地提取的自动化。同时,利用超级像素的梯度直方图改进了相似性测度,从背景和目标竞争性出发改进了区域合并的流程。由于时间的关系,本论文着重研究居民地自动提取的流程,验证基于超级像素和区域合并方法的居民地自动提取的可行性。今后可以在区域合并过程中加入更多的影像信息,并利用建筑物显著性等特征来细化居民地的初始范围选择等方面进一步深入研究。
[Abstract]:In recent years, with the continuous increase of population, the scale of rural residential land is also constantly changing, how to timely and accurately monitor the changes of rural residential land, It is an urgent problem to be solved. Because the use of high-resolution images to extract resident land has the advantage of being incomparable to field surveying, it can not be restricted by various natural conditions, such as topography, elevation, etc. However, because rural residents include roads, trees, and so on, A series of information, such as buildings, and the higher the resolution of the image, the richer the texture and color information. Traditional methods based on spectral classification can not effectively extract resident land from high-resolution images. In this paper, an automatic extraction method based on super-pixel is proposed. Firstly, a topology-preserving super-pixel method (ETPS) is used to segment the image into a series of superpixels with similar characteristics without breaking the edge information of the image. Then the color histogram of each super pixel is calculated, and the similarity between the super pixels is characterized by the Pasteur coefficient of the super pixel color histogram and the Bhattacharyya coefficient of the super pixel color histogram. Finally, Based on the maximum similarity, the MSRM) criterion merges super pixels into residential and background areas under the guidance of geographical names coordinates. The experiments of extracting 0.5 and 1 meters high resolution satellite images have been carried out. The results show that the proposed method can be used to extract residential land of a certain scale. The feasibility and effectiveness of the proposed method are verified. The purpose of this paper is to explore the possibility and problems of using the super-pixel method and the maximum similarity region merging method for automatic extraction of high-resolution image residents. Firstly, several typical super pixel segmentation algorithms are analyzed and compared. Then the super-pixel algorithm which is well controlled is selected as the method of inhabitants' extraction. In the process of inhabitants' extraction, the toponymic coordinates of toponymic survey are used as seeds to minimize the need of artificial interaction. At the same time, the similarity measure is improved by using gradient histogram of super pixels, and the process of region merging is improved based on the competition of background and target. This paper focuses on the process of automatic extraction of residents' land, and verifies the feasibility of automatic extraction of residents' land based on the method of super pixel and region merging. In the future, more image information can be added to the process of region merging. Further research is carried out by using the characteristics of building significance to refine the initial range selection of residential land.
【学位授予单位】：东华理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：P237

【参考文献】