Study on the fine-grained classification of urban green spaces by integrating GF-2 satellite imagery with open map data
Urban green space was an important carrier of urban ecological benefits.Highly accurate spatial monitoring target detec-tion and attribute classification of green space provided data support for optimizing urban ecological spatial structure,maintaining urban ecological balance and building the"carbon-neutral"city.This study improved the classical U-Net algorithm and apply it to GF-2 multi-spectral remote sensing image classification.Furthermore,drawing upon landscape ecology theory and utilizing open map data re-presented by POI and OSM,this paper performed multidimensional fine-grained classification of green patches.For the method pro-posed,this paper selected a rectangular area at the intersection of Futian District and Luohu District in Shenzhen City as the sample for validation.The results showed that the ASPP+SFAM fused U-Net network model proposed in the study matches the identified urban green space boundaries and the real urban green space boundaries more closely than the U-Net and U-Net3+models.The overall classification accuracy of the model was 90.87％,which was 11.13％and 7.39％better than the U-Net and U-Net3+models,re-spectively.Meanwhile,to address the problem that the texture features of remote sensing images could not directly classify the social at-tributes of urban green space in small areas,this study used the attribute information contained in POI data,the topological relationship between urban green space and OSM road network,and the landscape morphology index.Finally,this paper have realized the refine clas-sification of urban green space with four dimensions of functional classification,type characteristics,service scope and morphological characteristics.
urban green spacesclassificationU-Net networkPOIOSM
万方数据
维普
