A multi-scale feature fusion framework for urban functional zone identification using automatic ensemble learning method

In the process of the continuous expansion of global urbanization, the identification and classification of urban functional zones (UFZs) are essential for accurately mapping the internal organization of cities and scientifically planning urban layout patterns. Multi-source remote sensing data are significant for identifying the distribution patterns of UFZ and achieving sustainable urban development. However, the current research mainly uses remote sensing images and point of interest (POI), although the effect of the area of interest (AOI) has great potential, its role has been overlooked. In addition, multi-scale features are difficult to integrate. To solve these problems, in this study, a multi-scale feature fusion framework is proposed for identifying the UFZ distribution. This study integrates five types of features extracted from GF-2 hyperspectral images, POI, AOI, SDGSAT-1 nighttime light images, and building data. Based on these datasets, a plot ratio-enhanced nightlight index (PRENI) was proposed to identify UFZs more accurately and efficiently. The overall accuracies of the UFZs in Beijing, Chengdu and Shanghai were 90.70, 91.23 and 90.75%, respectively, confirming the effectiveness and robustness of the proposed method. In addition, a comparative analysis was performed against a deep learning–based UFZ classification approach using only GF-2 imagery, which further demonstrates the superior performance of the proposed method in terms of structural rationality and computational efficiency. This study provides a scientific basis for supporting accurate urban planning and development policies to achieve cities’ sustainable development goals.