Application of Remote Sensing Foundation Models in the Intelligent Transformation of Government Services: A Case Study of Marine Disaster Management in Guangzhou

In the context of intensifying climate change and frequent marine disasters, coastal cities have put forward urgent needs for high-timeliness and high-reliability remote sensing information support. Traditional manual interpretation methods are difficult to cope with the processing pressure of massive amounts of high-resolution remote sensing images, and the existing large-scale remote sensing models rely on overseas technology ecology. There are problems such as data security, poor model interpretability, and disconnection from government processes, which limit their in-depth application in urban-level governance scenarios. In view of the above challenges, this study uses Guangzhou City as an empirical case to construct a basic remote sensing model for urban-level marine disaster management, and proposes an integrated embedded application framework of “ model-data-process ”. The core contribution of this research is: through the integration of large-scale self-supervision pre-training and task-oriented fine-tuning, a unified model with multi-task coordination capabilities is constructed; a lightweight adaptation mechanism is designed to realize the normalized operation of the model in the five major government processes of reclamation supervision, island supervision, disaster emergency response, map update, and rural housing change detection; a collaborative mechanism that takes into account daily supervision and emergency response is established to promote the transformation of AI capabilities from technical prototypes to governance infrastructure. The experimental results show that the average accuracy of the model can reach 70.44% in the target detection task, 82.03% in the overall accuracy in the object classification task, and 72.6% in the change detection task. The recognition accuracy of the change area is 72.6%, and the change map spots are generated within 6 hours after the disaster, which significantly improves the decision-making efficiency. This research provides a replicable and generalizable “ AI+government affairs ” technology path and practical paradigm for the intelligent governance of coastal cities.