Application of Graph Convolutional Neural Networks and multi-sources data on urban functional zones identification, A case study of Changchun, China

• GCNN was used for improved urban functional zones (UFZs) identification. • The study integrates street view imagery to extract critical ground scene features. • The study achieves robust UFZs identification accuracy, 96 % on the test set. • It enhances residential and industrial UFZs identification by 8 % and 16 %. Urban functional zones (UFZs) identification is integral to comprehensive city management. However, current methods often fail to effectively leverage multiple data sources to enhance identification accuracy and overlook the spatial interconnections between neighboring units. In this study, we employed a Graph Convolutional Neural Networks (GCNNs) model to consolidate information from adjacent units and enhance the accuracy of UFZs identification. We specifically integrated street view imagery into our methodology to extract features from ground scenes. Furthermore, we conducted a co-occurrence analysis, correlating these visual features with socio-economic characteristics. With Changchun as a case study, the results indicate that 1) our proposed framework exhibits robust performance, achieving an accuracy of 96 % on the test set and a visual interpretation accuracy of 78 %; 2) the integration of street view imagery effectively addresses gaps in social sensing data features. Notably, the inclusion of ground scene features bolster the identification accuracy of residential and industrial areas by approximately 8 % and 16 %, respectively; 3) relative to other frequently utilized classification models, the graph convolutional model enhances the accuracy of UFZs identification by 11.2 %-16.6 %. Consequently, our framework effectively identifies UFZs, offering innovative methods and substantial data support for governmental bodies and urban planning authorities.