Semantic segmentation of satellite images for smart city development: a systematic literature review

Researchers and urban planners needs an enhanced strategies to handle their smart city development and this article is to provide a path for them about improving satellite driven urban intelligence, this Systematic Literature Review (SLR) presents technical advancements for smart city development needs, highlighting gaps in real-time processing and urban context generalization to offer a roadmap for researchers and planners to enhance satellite-driven urban intelligence. Image segmentation which is one of the image analysis technique that segments parts of the image as per the need of the application and semantic segmentation of satellite imagery specifically in this case is a pivotal task in remote sensing (RS) and Artificial Intelligence, leverages deep learning (DL) to classify each pixel, enabling applications from urban planning to disaster management. This SLR employs the SALSA framework (Search, Appraisal, Synthesis, Analysis) to systematically review 50 studies from January 2019 to September 2024, by exploring Deep Learning (DL) methodologies followed, datasets used, evaluation techniques, and contributions to smart city development. Searching papers from Scopus, IEEE Xplore, and Web of Science, we retrieved 1245 articles, finally refined to 50 via explicit criteria such as DL focus, satellite-specific, smart city relevance and peer-reviewed articles. Findings reveal that, U-Net and DeepLab V3 + dominate (IoU 0.8–0.90), driven by datasets like Sentinel-2 and DeepGlobe, supporting smart city goals such as intelligent mobility, environmental monitoring, and governance. However, challenges continue and future directions such as lightweight models, domain adaptation, and interdisciplinary impact studies are forwarded.