| Land cover refers to the synthesis of various material types and their natural attributes and characteristics on the earth’s surface.Its data is very important for global climate change,sustainable development and so on.In recent years,with the rapid development of remote sensing imaging technology and classification algorithm,a variety of free and high-resolution remote sensing images emerge in endlessly,which greatly promotes the research and production of global or large area land cover products.However,due to different data sources,there are different resolutions and classification systems,which cannot meet needs of data sharing and interoperability.Integration is a simple,effective and low-cost method to generate products with higher accuracy or to meet the needs of users.By quantifying the advantages and disadvantages of each source data,we can concentrate the advantages of each product to produce new land cover products or land cover products to meet certain needs.At present,the integration research of land cover products mainly solves the following problems:(1)different land cover products have some differences in classification system due to the different research purposes and significance;(2)the differences in the definition of various concepts.In the past,the integration of land cover products is to analyze the classification system of each data source directly in the integration process,and transform the corresponding legend through the relevant land cover product translation system.This method lacks the description of the concept of land cover products,because for different land cover products,even the concept with the same name,the semantics will be very different.Therefore,it is a main trend in the field of land cover product integration that the classification system of land cover products can realize knowledge sharing and reuse in the field of land cover integration through unified conceptualization and formal objective description.It is also a main research direction of scholars at home and abroad.In view of this direction,this paper takes multi-source land cover remote sensing products as an example,and proposes an ontology-based land cover product integration method,which mainly considers from two aspects of model layer and data layer.The schema layer mainly describes the semantics of the product classification system from different sources through ontology.Based on the hybrid ontology method,the concepts in multiple local ontologies are connected and compared through the shared vocabulary with EAGLE matrix elements as the shared vocabulary.The similarity between local ontology concepts is obtained by ontology mapping algorithm,and the comprehensive concept similarity between heterogeneous land cover products is obtained by combining ontology mapping based on concepts,attributes and instances,so as to realize the integration of pattern layer.On the other hand,the integration of data level considers the accuracy of the product itself,and uses the geostatistical Kriging interpolation method to obtain the local accuracy of the source product by collecting and interpreting the ground verification points.Finally,integration model is developed combined the semantic similarity and local accuracy.Taking NLCD2011(National Land Cover Database 2011)and FROM-GLC-Seg2010(Fine Resolution Observation and Monitoring of Global Land Cover Segmentation 2010)as source products,the second level class refinement of Globeland30 land cover product as an example,the forest types of Globeland30 in the United States are divided into broad-leaved forest,coniferous forest and mixed forest.The experimental results show that the second integrated model has the highest precision,and the precision and overall precision of broad-leaved forest,coniferous forest and mixed forest are 82.6%,72.0%,60.0% and 76.3% respectively.Compared with the first integrated model,the precision is increased by 1.2%,1.4%,11.7% and 1.0% respectively.Compared with the previous integration results of land cover products,the accuracy of secondary forests such as broad-leaved forest and coniferous forest is improved by 1.2% to10.0%,and the overall accuracy of products is improved by 1.0% to 7.3%.It can be seen from the results that compared with using other coding methods or translation systems to achieve information exchange between land cover products,the final land cover product integration effect is better by using ontology to achieve knowledge sharing and interoperability of classification technology.This method can also be extended to other types of surface features and land cover products. |
