DFCNN-based Functional Semantic Recognition Of Urban Building By Integrating Remote Sensing Data And POI Data

The urban functional zones,as a special fundamental unit of the city,helps to understand the complex interaction between human space activities and environmental changes.With the rapid development of satellite and remote sensing technology,very high-spatial resolution remote sensing images can fully explore the spatiotemporal pattern of cities.However,traditional identification methods only rely on shallow features and cannot recognize complex and heterogeneous urban functional zones.In recent years,the upsurge of deep learning has made it widely used in urban recognition,and deep learning shows its ability to recognize and mine deep features.However,the high-resolution remote sensing image with complex mapspectrum structure has function of exploring the characteristics of geographic objects,the surrounding environment and the spatial distribution.On the contrary,it cannot describe the urban socio-economic functions and human space activities.Therefore,combining remote sensing data and social perception data to realize functional area analysis is an effective way to quickly and accurately comprehend the complex urban functional pattern.In response to the above problems,this paper organically integrated high spatial resolution remote sensing images and social perception data,carried out a deep learning method about physical semantics and social functional semantics of building,from bottom to top,and completed the conversion from objects to functional zones and from shallow features to deep semantics.Finally,complex urban functional zones were recognized of functional areas,and it provides convenience to the government to manage,plan and supervise the city.The work and results of this article are as follows:(1)Based on the image analysis of bottom-up at the object level,a deep learning model for urban functional zones recognition based on building semantics is constructed.The model fully mined the physical semantics and social functional semantics of buildings.The strategy of hierarchical scale estimation was adopted to avoid the blindness and subjectivity of scale parameter selection to a certain extent.It not only improves the accuracy of building extraction,but also effectively improves the efficiency of urban functional area identification.(2)There is diversity in the spectrum,shape and texture of buildings.Therefore,in regard to the problem of extracting geometric information of complex buildings,this paper considered the importance of depth features to the edge of buildings,and introduced the depth feature map into the segmentation and classification process to improve the effect of building edge recognition.(3)Very high-spatial resolution remote sensing image has a more complex graphspectrum structure,and deep semantics can more effectively realize the mining of city information.Aiming at the difficulty of understanding deep semantics and spatial information of features,and considering the limitations of traditional CNN,this paper proposed a new network with more and deeper features based on the “Inception module”---Deeper-Feature Convolutional Neural Network(DFCNN).Among them,the use of deep convolution module not only enhances the adaptability of the network,and expands the perception field of the network,but also can excavate the deep semantics of the building.Finally,DFCNN showed certain advantages in the recognition of urban functional zones.(4)In order to break the limitation of using only remote sensing images for semantic recognition of social functions,this paper adopts the strategy of organic integration of points of interest(POI)and high spatial resolution remote sensing images to effectively realize the complementary combination of physical semantics of buildings and social functional semantics,Showing a good effect in recognizing urban functional zones.