FGDOA-Net: A Dual-Decoder Network With Frequency-Induced Detail-Guided Difference Aggregation for Remote Sensing Change Detection

Abstract

Remote sensing change detection (RSCD) aims to identify land surface changes from multitemporal remote sensing images and plays a critical role in applications, such as land monitoring and urban planning. Existing deep-learning-based RSCD methods often struggle to capture fine-grained details and to aggregate semantic features across scales, limiting change detection accuracy. This article proposes FGDOA-Net, a novel dual-decoder network that enhances detail sensitivity and difference modeling. The auxiliary decoder integrates a spatial-frequency fusion-based detail extraction module and an adaptive multiscale feature aggregation (AMFA) module to extract rich detail features. The primary decoder adopts the existing spatiotemporal feature fusion module and designs a novel difference feature optimization module to refine semantic difference representations under the guidance of detail features. AMFA is used in both decoders to adaptively fuse multiscale features and improve structural consistency. Extensive experiments on WHU-CD, SYSU-CD, EGY-BCD, and a challenging new dataset, CDTH-CD, show that FGDOA-Net outperforms state-of-the-art methods in accuracy and generalization while maintaining low computational complexity (8.64 GFLOPs).