Abstract—Change detection plays a crucial role in various fields such as environmental monitoring and disaster management. In recent years, deep learning-based approaches have significantly improved detection accuracy and efficiency. However, challenges such as multi-scale feature misalignment, ineffective fusion strategies, and inconsistent cross-scale semantics still hinder their deployment in complex real-world scenarios, especially in geological hazard monitoring. To address these issues, we propose a novel change detection framework that integrates channel attention and non-local context modeling into the feature extraction stage to enhance channel discrimination and global dependency learning. In the decoding phase, we introduce a flow-guided feature alignment and fusion module, which estimates optical flow fields and performs adaptive warping to reduce temporal feature discrepancies and improve alignment accuracy. In addition, multi-level feature fusion and semantic consistency refinement are employed to better capture subtle and sparse changes. Extensive experiments on the public LEVIR-CD dataset and a newly constructed Three Gorges rock mass dataset demonstrate that our method achieves state-of-the-art performance in terms of both accuracy and computational efficiency. Moreover, the framework exhibits strong robustness in complex terrains while maintaining a lightweight design, showing great potential for practical applications in geological disaster monitoring, early warning, and risk-informed decision-making. Quantitatively, our method achieves an F1-Score of 90.0% and Intersection over Union (IoU) of 81.82% on the LEVIR-CD dataset, surpassing existing methods such as ChangeStar by 0.7% and 1.16%, respectively.
 
Keywords—change detection, feature extraction, feature alignment, deep learning, feature fusion, attention mechanism, geological hazard monitoring


Copyright © 2025 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).