兴趣面数据和随机森林方法的城市功能区划分

Urban functional area division based on AOI data and random forest method

针对传统城市功能区研究探索了利用兴趣点、遥感影像等数据进行城市功能区自动划分的方法,但分类精度仍有待进一步提高的问题,该文提出一种基于兴趣面数据和随机森林算法的城市功能区自动划分方法,首先对城市道路网络数据运用形态学运算方法自动构建交通分析区,然后从其区域内的兴趣区等多源数据中提取4类41项特征,最后结合地理国情普查数据和人工识别选取476个样本数据,构建基于随机森林分类器的城市功能区识别模型。结果表明,随机森林模型能够准确地识别城市功能区,识别模型预测精度为87.67%,支持向量机模型精度为83.71%,未引入兴趣面数据的模型精度为78.72%。利用上述模型对北京市内4 112个单元进行功能区识别,准确率达81.37%。该研究方法有较好的重复性和推广性,有助于对城市功能区进行定量、客观的研究。

Aiming at the problem that the classification accuracy still needed to be further improved when the traditional urban functional area research explored the method of automatic urban functional area classification using data such as points of interest and remote sensing images, this paper proposed an automatic urban functional area classification method based on point of interest data and random forest algorithm. Firstly, the traffic analysis area was automatically constructed by applying morphological operation method to urban road network data.Secondly, four categories of 41 features were extracted from the area of interest within its area. Finally, 476 samples were selected by combining the geographic census data and manual recognition to build a random forest classifier-based urban functional area recognition model. The results showed that the random forest model could accurately identify urban functional areas, and the prediction accuracy of the identification model was 87.67%,the accuracy of the support vector machine model was 83.71%,and the accuracy of the model without introducing point of interest data was 78.72%. The accuracy of functional area identification using the above models for 4 112 units within Beijing was 81.37%. The method of this study had good repeatability and generalizability, which helped to conduct quantitative and objective research on urban functional areas.