Research on the Method of Artificial Intelligence for Identifying Urban Land-Use Types Based on Areas of Interest (AOI) and Multi-Source Data

Abstract

Urban land-use types, a fundamental aspect of urban planning, land management, and the effective utilization of spatial resources, are exhibiting increasing complexity. Efficient and scientific identification of large-scale urban land-use types has become a major challenge in urban research. To address this, the present study adopts a functional structure-based perspective and integrates commercial AOI data, POI data, nighttime light data, and population distribution data to classify land use. Departing from existing data weighting algorithms, this research applies artificial intelligence techniques, utilizing the categorical information of AOI data as labels. Through supervised deep learning, urban land-use types are refined into nine major categories and 21 subcategories across cities of different scales and locations. Compared to SVM, RF, and MLP models, the XGBoost model achieved the highest accuracy in classifying urban construction land (weighted avg F1 score = 0.87). Furthermore, by comparing the AOI data with real-world test datasets, the accuracy and granularity of land-use classification were significantly enhanced. Finally, this AI model, combined with remote sensing imagery and transportation network data, was used to generate a land-use map for the target city, offering insights into the generalizability of AI models in urban land-use classification.