基于多源数据和深度学习的城市边缘区判定

城市边缘区的定量分析及判定,对城市发展评价和规划,或是城市空间结构研究都具有重要意义。然而现有研究的边缘区判定指标选择过于单一,判定结果过于破碎,城市预设边界、水体及城市绿地对边缘区判定结果干扰大。针对上述问题,从自然、人口、社会经济的视角出发,以遥感影像、人口数据、POI大数据为数据基础,结合深度学习技术,构建基于多源数据和深度学习的城市边缘区判定方法,进行广州市城市边缘区判定及城市结构空间分布特征分析。结果表明：① 此方法能将城市划分为核心区-边缘区-外缘区,判定结果不会受到预设边界范围的影响,且消除了城市内部水体和城市绿地所造成的破碎化;② 城市边缘区与路网耦合良好;③ 广州市的城市核心区空间分布合理。综上所述,此方法能有效检测城市边缘地带,且结果符合实际情况,能为城市规划、政府决策提供参考。

The definition of urban fringe based on multi-source data and deep learning

With the development of the economy, most cities will expand continuously to the surrounding areas, thus leading to the emergence of urban fringe areas with both urban and rural characteristics. The urban fringe area, located between urban and rural areas, is the most intense area of urban land use change and one of the most likely areas for urban construction land expansion in the future. How to identify urban fringe accurately and quantitatively is of great significance for urban planning and sustainable land use. However, most existing methods about the delineation of urban fringe area is just based on one or one type of indicators, and the judgment result is too fragmented to reflect the continuity of the urban spatial structure. What's more, the urban preset boundary range, the water body and the urban green space have great interference with the judgment results of urban fringe. In view of the above problems and from multi-perspective of nature, population and social economy, this paper defines urban fringe based on deep learning and multi-source data (remote sensing image, population density and POI big data). Furthermore, the proposed method has been used to detect the urban fringe area of Guangzhou city in our experiments. The results show that: (1) This method can divide the city into urban core area, urban fringe and rural area accurately without the impact of the preset boundary range. Eventually, this way can eliminate the fragmentation caused by the internal water and green space of urban areas. (2) The results of urban fringe area are well coupled with the road network. Network distribution of the urban core area is densest, followed by the urban fringe area. (3) The spatial distribution of urban core area of Guangzhou from the experiments is reasonable and consistent with the actual situation. All in all, the proposed method can consider comprehensively multi- perspective factors and detect urban fringe effectively, thus can provide better guidance for formulation of policies for urban development, such as urban planning, sustainable development, and urban statistical analysis.