高分辨率遥感影像耕地多尺度提取

发布时间：2019-02-10 18:44

【摘要】：摘要：耕地是人类生存发展过程中不可或缺的宝贵资源,实时、动态、精确地掌握耕地信息是人民生活保障、国家进步发展的重要基础。高分辨率遥感影像的出现为大范围、精确、高效地提取耕地信息提供了可能,然而高分辨率遥感影像中耕地复杂多样难以从某单一尺度精确提取,一直以来这都是遥感界的一个研究难题。因此,本文将在高分辨率遥感影像耕地多尺度分析上做了一点较为有实用意义的研究与探索。总体概括来说,本文开展如下几个部分的研究工作：首先,对分水岭分割算法进行简要的介绍,分析了该算法的优劣性；对遥感中尺度的含义、内容、方法进行了阐述；对遥感影像的光谱、形状、纹理特征的提取方法及其实用性进行分析。其次,将物理学中场论方法引入到遥感影像分割中,提出一种多层次场聚类的分割方法。通过分水岭分割得到地物对象,通过这些对象的分布关系构建聚类场,然后通过逐层聚类的方法进行合并,达到较优的分割效果。在含有大片耕地的高分辨率遥感影像上进行实验验证,取得了较好的分割结果。然后,丘陵地区地势高低起伏、耕地形状大小不一,传统人工测量方法获取耕地信息费时费力,而自动化的方法提取丘陵耕地的效果又不十分令人满意的情况。针对这个难点,本文提出一种丘陵耕地多尺度提取方法。在利用各向异性多尺度构建方法生成多尺度梯度影像的基础之上,通过分水岭算法获得不同尺度的耕地分割结果,之后利用GS评价准则从多尺度分割结果中选取最优的耕地边界识别结果。对比实验表明本文方法在丘陵耕地的自动化提取中有一定的优势。最后,城郊耕地是城市居民生活物资的保障,城市化进程的加快使其日益受到被侵占的威胁。高分辨率遥感影像中,城郊地区地物类别多样、分布不均且光谱混杂,该区域耕地提取难度较大。为此,本文将建筑物区域进行剔除,在此基础之上针对耕地占主导地位的地物进行多尺度分析得到较好的耕地提取结果。该算法主要有如下三个部分：利用改进的Harris角点特征对城郊建筑物进行提取,大致得到建筑区与非建筑区的划分；通过多尺度方法对非建筑区耕地为主导的地物分析,得到耕地最佳分割结果；通过设定规则提取形状规整的耕地与建筑区中非耕地,使用支持向量机进行分类得到耕地结果。
[Abstract]:Absrtact: cultivated land is an indispensable and precious resource in the process of human survival and development. It is an important foundation for people's living guarantee and national progress to grasp the information of cultivated land in real time, dynamic and accurate. The appearance of high-resolution remote sensing images makes it possible to extract cultivated land information accurately and efficiently. However, it is difficult to accurately extract cultivated land from a single scale in high-resolution remote sensing images. This has always been a research problem in remote sensing. Therefore, this paper will do a bit of practical research and exploration on the multi-scale analysis of cultivated land in high resolution remote sensing images. In general, the following parts of the research work are carried out in this paper: firstly, the watershed segmentation algorithm is introduced briefly, and the advantages and disadvantages of the algorithm are analyzed, the meaning, content and method of remote sensing mesoscale are expounded. The extraction method and practicability of spectral, shape and texture features of remote sensing image are analyzed. Secondly, the physical field theory method is introduced into remote sensing image segmentation, and a multi-level field clustering segmentation method is proposed. The ground objects are obtained by watershed segmentation, and the clustering field is constructed by the distribution of these objects, and then the clustering method is combined to achieve a better segmentation effect. The experimental results of high resolution remote sensing images with large area of cultivated land are verified and good segmentation results are obtained. Then, the terrain in hilly area is up and down, the shape of cultivated land is different, the traditional manual measurement method takes time and effort to obtain the information of cultivated land, and the effect of automatic method to extract cultivated land in hilly area is not very satisfactory. In view of this difficulty, this paper presents a multi-scale extraction method for hilly cultivated land. On the basis of using anisotropic multi-scale construction method to generate multi-scale gradient image, the watershed algorithm is used to obtain the results of farmland segmentation with different scales. Then the optimal cultivated land boundary recognition results are selected from the multi-scale segmentation results by using the GS evaluation criterion. The comparative experiments show that this method has some advantages in automatic extraction of hilly farmland. Finally, suburban cultivated land is the guarantee of living materials for urban residents, and the acceleration of urbanization makes it increasingly threatened by encroachment. In the high-resolution remote sensing image, there are many kinds of ground objects in the suburban area, which are unevenly distributed and the spectrum is mixed, so it is difficult to extract cultivated land in this region. For this reason, this paper culls the building area, and on the basis of this, carries on the multi-scale analysis to the cultivated land which occupies the dominant position, obtains the better cultivated land extraction result. The algorithm consists of the following three parts: the improved Harris corner feature is used to extract the suburban buildings, and the division between the building area and the non-building area is obtained roughly; Through multi-scale analysis of cultivated land in non-construction areas, the optimal segmentation results of cultivated land are obtained, and the cultivated land results are obtained by using support vector machine (SVM) to extract cultivated land with regular shape and non-cultivated land in construction area.
【学位授予单位】：中南大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP751

【参考文献】