Wavelet-enhanced cross-domain fusion network for remote sensing change detection
摘要
To fully exploit spatial information and address the issue of false detections in abrupt grayscale transition regions of remote sensing images, we propose WECF-Net (Wavelet-Enhanced Cross-Domain Fusion Network). The model operates in two main stages. In the first stage, spatial features are decomposed in the frequency domain using energy-based wavelet transformation to obtain low-frequency components that capture global structures and high-frequency components that preserve local details. A hierarchical enhancement strategy is employed to mitigate grayscale shift interference. In the second stage, cross-domain alignment of frequency-domain features is achieved via spatial-aware projection, enabling the effective integration of frequency information while preserving the rich semantic representation of spatial features. Experimental results show that the model achieves an F1-score of 91.34