ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Download
Publications Copernicus
Download
Citation
Articles | Volume V-3-2020
ISPRS Ann. Photogramm. Remote Sens. Spatial Inf. Sci., V-3-2020, 263–268, 2020
https://doi.org/10.5194/isprs-annals-V-3-2020-263-2020
ISPRS Ann. Photogramm. Remote Sens. Spatial Inf. Sci., V-3-2020, 263–268, 2020
https://doi.org/10.5194/isprs-annals-V-3-2020-263-2020

  03 Aug 2020

03 Aug 2020

SOCIAL INFORMATION FUSED URBAN FUNCTIONAL ZONES CLASSIFICATION NETWORK

W. Lu, C. Tao, Q. Ji, and H. Li W. Lu et al.
  • School of Geosciences and Info-Physics, Central South University, Changsha 410083, China

Keywords: Urban Functional Zones, Point of Interest, Remote Sensing, Classification

Abstract. Fast-changing cities need efficient management. Accurate classification of urban functional zone (UFZ) can provide important reference for cities management. Remote sensing imagery (RSI) is large scale, high resolution and fast update, which can provide massive data for UFZ extraction. However, UFZ are more concerned with social attributes such as industrial production and commercial activities, while images can only provide visual features, which is not enough for an elaborate UFZ classification. To solve this problem, in this paper, we combine RSI and point of interest (POI) data together for UFZ classification, and propose a Social Information Fused Urban Functional Zones Classification Network (SIF-Net). For RSI, we simply use a Xception CNNs network extract the visual information. For POI data, we first build a coarse heatmap for each type of POI (e.g. retail, apartment…), and then combine them as a POI tensor. Afterward, we use a channel attention module (CAM) based CNN model to fuse heatmaps from each type of POI, and then build a fine distribution of UFZ as the social information. Finally, we fuse the visual information extracted from RSI and social information extracted from POI by concatenating them. By fusing this two complementary information, our method makes up for the shortcomings of extracting UFZ based on RSI and general CNNs only. Compared with current state-of-the-art methods, experiments show that the proposed SIF-Net can significantly improve the UFZ classification result.