Abstract: To address the potential threats to power grid safety posed by construction changes in surrounding structures (e.g.,factories,residences,temporary buildings),this study proposes an efficient and reliable change detection method.A multi-level attention-driven Mamba building change detection network is designed.Adopting an encoder-decoder architecture,the encoder extracts multi-level features based on the Visual Mamba framework.The decoder enhances feature expression in regions of interest and multi-scale information fusion through a feature enhancement module (FEM)and a hierarchical feature fusion module (HFFM),enabling automatic identification of building changes in dual-phase remote sensing images.Comparative and ablation experiments on synthetic datasets demonstrate superior performance across multiple metrics compared to existing mainstream methods.The approach significantly improves detection accuracy for small targets and buildings in complex backgrounds,exhibiting enhanced change recognition capability and robustness.In real-world power line scenarios,the proposed method accurately identifies newly constructed and demolished buildings with clear change boundaries and high localization precision.Integrating the Mamba architecture with attention mechanisms effectively enhances remote sensing change detection performance,providing an efficient and reliable technical approach for monitoring construction changes in buildings surrounding power grid lines.

Key words: remote sensing images, change extraction, state space, attention mechanism, power grid security