遥感影像土地覆盖分类地理本体构建研究

发布时间：2018-05-16 12:49

本文选题：高分辨率遥感影像 + 土地覆盖分类　；参考：《江苏师范大学》2017年硕士论文

【摘要】：遥感影像土地覆盖分类是国际研究的热点,面向对象分类技术是遥感影像土地覆盖分类的主流且有潜力的手段,目前存在分类高度主观且规则难以重复应用的问题,主要是缺乏系统的概念化和形式化表达面向对象分类技术各元素的方法,地理本体凭借其概念化、明确性、形式化、共享性等特征在解决上述问题方面具有明显的优势。本文以地理国情普查中的土地覆盖二级类为例,从地理本体理论角度出发,研究知识驱动和数据驱动相结合的土地覆盖分类规则构建方法,研究土地覆盖分类地理本体建模方法,并开展了地理本体建模驱动的土地覆盖分类实验。具体研究内容及成果包括如下:(1)研究了知识驱动和数据驱动相结合的土地覆盖分类规则构建方法。依据高分辨率遥感影像面向对象分类过程,利用知识驱动的专家知识与数据驱动的决策树,构建土地覆盖分类规则,具体构建了面向地理国情普查中的土地覆盖二级类分类规则集。(2)构建了土地覆盖分类地理本体模型。利用知识工程方法及OWL语言构建了土地覆盖类型、影像对象特征、分类器的本体模型,利用推理机对土地覆盖分类本体模型进行推理,检验该地理本体模型的逻辑一致性。(3)研究了土地覆盖分类地理本体支持下的面向对象分类方法。将土地覆盖分类地理本体模型与面向对象分类的过程相衔接,实现土地覆盖分类地理本体支持下的面向对象分类。(4)开展了土地覆盖分类实验。选择地理国情普查中土地覆盖的13个二级类,包括人工草地、绿化林地、低矮独立房屋建筑、城市道路、库塘、旱地、河渠、多层及以上房屋建筑区、低矮房屋建筑区、泥土地表、公路、灌木林、果园。实验数据为德国波兹坦和西安临潼区的航片正射影像和WorldView-2高分辨率影像,实验平台为遥感影像智能解译工作站-FeatureStation_GeoEX、本体编辑软件-Protégé和数据分析软件-SPM(Salford Predictive Modeler)。实验表明,本文提出的方法能够实现遥感影像分类各个元素的客观建模,影像分类结果满足实际应用需求,领域知识可以共享,本文的研究方法具有一定的有效性和普适性,能够为遥感影像土地覆盖分类提供可共享、可扩展的地理本体。
[Abstract]:Land cover classification of remote sensing images is a hot topic in international research. Object-oriented classification technology is the mainstream and potential means of land cover classification of remote sensing images. At present, the classification is highly subjective and the rules are difficult to be applied repeatedly. There is a lack of systematic conceptualization and formal representation of all elements of object-oriented classification technology. Geographical ontology has obvious advantages in solving the above problems with its characteristics of conceptualization, clarity, formalization and sharing. This paper takes the land cover secondary category in the geography national situation census as an example, from the geographical ontology theory angle, studies the land cover classification rule construction method which combines the knowledge drive and the data drive. The modeling method of land cover classification geographic ontology is studied, and the land cover classification experiment driven by geographical ontology modeling is carried out. The main contents and results are as follows: 1) the method of constructing land cover classification rules based on knowledge driven and data driven is studied. According to the object oriented classification process of high resolution remote sensing image, using the expert knowledge driven by knowledge and the decision tree driven by data, the classification rules of land cover are constructed. In this paper, the land cover classification rules set. 2 (2) the geographical ontology model of land cover classification is built. The ontology model of land cover type, image object feature and classifier is constructed by using knowledge engineering method and OWL language, and the reasoning machine is used to infer the land cover classification ontology model. The object oriented classification method supported by the land cover classification geographic ontology is studied by checking the logical consistency of the geographical ontology model. The model of land cover classification geographic ontology is connected with the process of object oriented classification, and the object oriented classification supported by land cover classification geography ontology is realized. The experiment of land cover classification is carried out. Select 13 secondary categories of land cover in the survey of geographical conditions, including artificial grassland, greening woodland, low-rise independent house buildings, urban roads, ponds, dry land, river channels, multi-storey and above housing construction areas, low-rise housing construction areas, and low-rise housing construction areas. Soil, roads, shrubs, orchards. The experimental data are aerial orthophoto images and WorldView-2 high-resolution images from Poztan and Lintong district, Germany. The experimental platform is intelligent remote sensing image interpretation workstation -FeatureStationGeoEX. the ontology editing software -Prot 茅 g 茅 and the data analysis software -SPMN Salford Predictive Modeler. Experiments show that the proposed method can realize the objective modeling of each element of remote sensing image classification, the result of image classification can meet the needs of practical application, and the domain knowledge can be shared. The research method in this paper is effective and universal. It can provide shared and extensible geographical ontology for land cover classification of remote sensing images.
【学位授予单位】：江苏师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：P237

【相似文献】