CDMFMamba: A Novel Remote Sensing Change Detection Framework Integrating MMEN and FAFFM Modules

Within remote sensing change detection (CD), both CNN and Transformer frameworks exhibit significant performance improvements. However, each of these architectures has inherent limitations: CNNs struggle to effectively capture broader spatial contextual information due to their limited receptive fields, while Transformer models involve high computational costs, making training and deployment expensive on large scale datasets. In this paper, we propose a model named CDMFMamba, we design a Multi - scale Mixture of Experts Network (MMEN) where a gating network assigns different inputs to specific experts, and their outputs are fused to integrate multi domain knowledge. This enhances the model’s generalization in complex remote sensing change detection. Also, a Frequency assisted Feature Fusion Module (FAFFM) is proposed. It extracts and enhances high and low frequency features to highlight image details and overall trends, offering the model more comprehensive features. Experiments on WHU - CD, LEVIR - CD, and LEVIR+-CD datasets show the model improves change detection accuracy, showing excellent performance and providing useful insights for future model design.