Change detection method based on style transfer for heterogeneous image feature alignment
Jiandi Wu, Jiannan Cai, Shuwen Yang · International Journal of Digital Earth · 2026
Heterogeneous images often contain significant feature discrepancies caused by sensor differences, imaging conditions, and geometric distortions, leading to feature-space inconsistencies that degrade change detection performance. To address this issue, this paper proposes a style transfer–based change detection framework for feature alignment. A style-aware global–local network is developed to reduce information loss and structural distortion during style transfer. By integrating multi-scale linear attention, spatially adaptive feature modulation, and style–content feature attention, it enhances feature representation, preserves structural structure, and improves feature-space consistency. The style-transferred images are used to extract high-quality difference images, reducing heterogeneous feature discrepancies. A histogram-gradient thresholding strategy is applied for reliable sample selection, followed by a convolutional wavelet neural network for accurate change classification. Experimental results demonstrate the effectiveness and superiority of the proposed method in heterogeneous image change detection.