Neural-symbolic integrated framework for dynamic environments: advanced semantic understanding of remote sensing data in sustainable urban planning

This study aims to develop a neural-symbolic integrated framework that elevates the semantic understanding of remote sensing data for sustainable urban planning. Conventional methodologies remain deficient in five interlocked areas: advanced semantic representation, interpretability, knowledge fusion, dynamic adaptation, and causal reasoning; to date, no single solution has simultaneously closed these gaps in rapidly evolving urban contexts. While existing neural-symbolic systems show promise, they are overwhelmingly trained on static data and falter when policies or land-use patterns shift. To address these shortcomings, we fuse a Vision Transformer for multi-scale feature extraction with a Prolog rule engine for logic verification, embed federated and incremental learning for real-time adaptation, and employ multimodal data fusion coupled with NSGA-II multi-objective optimization to balance ecological protection and urban development. Extensive experiments reveal classification accuracies of 93.4% for “unsuitable expansion areas” and 90.7% for “ecological protection zones”, an 88.44% planning-rule compliance rate, and robust adaptability and interpretability under continuous environmental change. The proposed framework furnishes an efficient, explainable, and evolvable solution for remote-sensing data analytics, offering direct decision support for sustainable urban planning and ecological monitoring.