| The monitoring of urban surface elements refers to the use of remote sensing,big data and other technical means to grasp the type,area,scope,distribution and change of urban natural resources and human geographical elements,and provide data support for land and space planning and management,urban planning and comprehensive management,regional decisionmaking and management.At present,the monitoring data sources of urban surface elements are mainly high-resolution remote sensing images.Common extraction methods include mainstream methods such as machine learning,object-based image analysis(OBIA)and deep learning models.There are the following difficulties in the extraction of urban surface elements:(1)The distribution of urban features is complex,and the classification system of land use and land cover is often confused in the classification system.The common extraction methods are mainly based on a single data source of remote sensing images.The accuracy of automatic extraction is low,and manual visual interpretation is time-consuming and laborious.(2)The definition of each type is vague,the semantic expression is unclear,and there is a lack of conceptual and formal expression of urban surface elements.(3)The monitoring of urban surface elements has not yet provided a widely applicable definition of the monitoring content.In the face of different industries and different monitoring emphases,different classification systems are often defined according to the needs,and the classification systems are not compatible with each other.Different classification systems have different definitions for the description of the same name feature type.In the construction of surface element monitoring,it is often necessary to reselect a large number of high-precision samples and then reclassify them,resulting in repeated construction costing manpower and material resources.In view of the above problems,this paper proposes an ontology-based urban surface feature extraction method.This method first performs semantic analysis and disassembly on the definition and description of ground object types under different classification systems.According to the elements contained in the EAGLE(EIONET Action Group on Land monitoring in Europe)matrix,the definition of ground object types in the original classification system is compared.The EAGLE matrix is used to extract the ontology primitives to form a new semantic description as a ’ component ’ for constructing urban surface elements.Using multi-source data such as high-resolution remote sensing images,time series images,and DEM,multi-source data features are extracted,and rules between features and ontology primitives are established to extract ontology primitives from multi-source data.Secondly,the Web Ontology Language(OWL)is used to construct the ontology model of different land cover classification systems and feature information based on multi-source data extraction.At the same time,the definition description of the processed land cover type is used as the judgment basis for the classification of subsequent urban land elements.Through the knowledge of remote sensing experts,the Semantic Web Rule Language(SWRL)is transformed into the rules of subsequent feature classification that can be recognized by the computer.This paper takes the forest and grass cover and planting land in the urban surface elements as an example to study.Under the background of the current era of achieving the ’ double carbon’ goal and green city,the accurate extraction and segmentation of urban vegetation is concerned by more and more researchers.Taking Beijing II image,Beijing II multi-temporal image and DEM data as the source data,the study area is Sujiatuo Town and Shangzhuang Town in Haidian District,Beijing.For the planting land and forest and grass coverage under the basic urban geographical conditions monitoring content,and the corresponding feature types in the finer resolution observation and monitoring of global land cover(From＿GLC)monitoring and classification system,the classification and extraction experiment of urban surface elements is carried out.The final results of the experiment show that the overall accuracy of the classification results of the regional geographical conditions monitoring system in Shangzhuang Town is 86 %,and the Kappa coefficient is 76.68 %.The overall accuracy of the classification results of the From＿GLC system is 86.6667 %,and the Kappa coefficient is77.72 %.The overall accuracy of the classification results of the regional geographical conditions monitoring system in Sujiatuo Town is 93.2886 %,and the Kappa coefficient is83.04 %.The overall accuracy of the classification results of the From＿GLC system is93.2886 %,and the Kappa coefficient is 83.04 %.When applying the two classification systems to classify the same area,the problem of repeated construction is avoided.Experiments show that the research method in this paper meets the practical application requirements of urban surface element monitoring,and provides a new method with scalability,adaptability and objectivity.This method can also be extended to other types of urban surface elements monitoring. |
