What deep learning can (and can’t) see: a systemic review assessing applicability for quantifying qualities of urban vitality

Abstract Deep learning (DL) methods are an emerging tool that can be well-suited to uncover the complexity of urban vitality at a greater scale and depth. However, the use of DL in the field, even beyond vitality, remains underexplored, with no established state of the art to inform further research. This paper assesses DL’s potential as a research tool, critically examining its current status, benefits, and challenges in urban vitality studies. Through a systematic review, we analysed 208 publications that examine qualities under the dimensions of the built environment and urban activity, tying big data resources and DL methods to the study of urban vitality. A thematic analysis was conducted, identifying research contributions, big data types, resources, and DL architectures. Our findings indicate that DL shows significant promise for urban vitality research due to its scalability, multimodal capability, and fine-grained analysis, particularly of visual data. While it is well-established in analysing the built environment, its applications in human activity and spatio-temporal dynamics are only emerging. Multi-source data use is growing, enhancing robustness, but challenges such as geographic bias and high data requirements persist. Notably, over half of studies originate from China, limiting the generalizability of findings to contexts with different data availability, urban morphologies, and socio-cultural dynamics. Nonetheless, this synthesis serves as a critical assessment of DL’s possible future role in urban vitality studies.