Design And Implementation Of Urban Land Use Sample Database From Satellite GF-2 Imagery

Typical urban feature elements are important basic geographic information in urban planning,construction and management,which can provide basic data support for urban land use surveys,urban refined management,and spatial planning.At present,deep learning technology has been widely used in the extraction of feature elements such as urban buildings and roads based on remote sensing images.Among them,the feature elements samples input to the convolutional network model for training are the key to determining the success or failure of information extraction based on deep learning.One of the factors.The rapid development of domestic high-resolution satellites provides more available data resources for the extraction of feature elements based on high-resolution images.Aiming at the current lack of sample data sets of typical features of cities based on domestic GF-2 satellites,this paper designs a classification system for typical features of cities specifically serving the housing and construction industry,including buildings,roads,green spaces,water bodies and bridges.And based on the identifiable features of GF-2 images,the city buildings are classified into two sub-categories,which lays the foundation for the conversion of buildings from feature classification to functional classification.On this basis,a set of implementable sample collection plan and database construction process was designed,and automatic sample storage was realized based on Arc Py.In order to better manage and use the sample data set,this paper develops a visual management platform for the sample library of typical urban features based on the B/S development architecture,combined with open source tools such as Geo Server and Open Layers,and achieves 405 kilometers of grid sample data in 20 cities across the country.Centralized management.The large-scale and highly flexible sample library constructed in this paper will help convolutional neural networks to train high-precision segmentation results of typical feature elements.