Deep Semantic Segmentation for Rapid Extraction and Spatial-Temporal Expansion Variation Analysis of China’s Urban Built-Up Areas

Abstract

Changes in the spatial expansion of urban built-up areas are of great significance for the analysis of China’s urbanization process and economic development. Nighttime light data can be used to extract urban built-up areas in a large-scale and long-time series. In this article, we introduced the UNet model, a semantic segmentation network, as a base architecture, added spatial attention and channel attention modules to the encoder part to improve the boundary integrity and semantic consistency of the change feature map, and constructed an urban built-up area extraction model—CBAM_UNet. Also, we used this model to extract urban built-up areas from 2012 to 2021 and analyzed the spatial and temporal expansion of China’s urban built-up areas in terms of expansion speed, expansion intensity, expansion direction, and gravity center migration. In the last decade, the distribution pattern of urban built-up areas in China has gradually changed from “center” to “periphery-networked” distribution pattern. It reveals a trend from agglomeration to the dispersion of urban built-up areas in China. It provides a reference for China’s urban process and its economic development.