Identifying up-to-date urban land-use patterns with visual and semantic features based on multisource geospatial data

Abstract

The cognition of up-to-date urban land-use patterns has important guiding significance for grasping the current status of urban development, promoting sustainable urban planning and social development. Despite the emergence of various methods for identifying land-use patterns with the advent of geographic big data and related technologies, several issues, such as the suitability of the data used for exploring land-use patterns and the provision of sufficient information to characterize the land functions, are worthy of further discussion. This paper establishes a framework that fuses visual and semantic features to identify the up-to-date urban land-use patterns based on multisource geospatial data. The proposed framework mainly consists of two components: the first component is a fine-grained land-use type identification model with deep learning networks that employ an Inception-based feature extractor to obtain visual features in remote sensing (RS) images and a BERT-based feature extractor to extract semantic features from human activity-related data; the second component is the evaluation of the land-use patterns at the parcel level from diverse perspectives. The experiments on real-world data demonstrate the effectiveness and superiority of the proposed method over alternative methods for identifying the up-to-date urban land-use patterns.