Population Spatialization Based on Convolutional Neural Network and Multi-Source Geospatial Big Data

Abstract

 This article adopts a more refined method to create a population spatial distribution model for Guizhou Province in 2021. Firstly, we divide the population data of Guizhou Province into different characteristic regions. Then, we use these regions as independent variables and population as dependent variables for analysis using convolutional neural network (CNN). In addition, we also introduced multi-source spatiotemporal data such as Points of Interest (POI), nighttime lighting, and Digital Elevation Model (DEM) for modeling to further enhance the accuracy and accuracy of the model. Through this method, we successfully generated the population spatial data of Guizhou Province in 2021 with a resolution of 1km. These data intuitively display the spatial distribution of the population in Guizhou Province, and have important reference value for policy makers, researchers, and planners.In addition, in order to verify the accuracy of our model, we evaluated and compared the accuracy of the model results. The results show that our model has high accuracy in expressing the spatial distribution of population, andcan accurately reflect the spatial distribution of population at different scales.In summary, this article successfully created a population spatial distribution model for Guizhou Province in 2021 using multiple spatiotemporal big data and multiple linear statistical regression methods, and evaluated and compared the accuracy of the model results. These results have important practical significance for understanding the population distribution, formulating land use planning and public policies in Guizhou Province.