Abstract
In the face of bi-temporal image style inconsistency and complex land surface environment, most existing change detection methods focus solely on spatial domain modeling while neglecting frequency domain information. Building upon the foundation that frequency domain features effectively characterize image styles, we propose a cropland change detection method based on spatial-frequency domain synergy and Scale Change Awareness (SFSCANet). Specifically, features are first extracted in the spatial domain, then transformed into the frequency domain. The Frequency Domain Feature Interaction Module (FDFIM) filters stylistic noise from features to unify image styles. After returning to the spatial domain, a Transformer performs global modeling, followed by the Scale Change Awareness Module (SCAM) to refine local feature details, enhancing the model’s ability to detect cropland in complex scenes. Experiments demonstrate that our method outperforms 12 state-of-the-art change detection (CD) methods across four datasets.