MGFNet: a change detection network based on multilevel attention guidance and multiscale difference fusion

Deep learning has been used a lot to detect changes in remote sensing images. Nevertheless, contemporary models continue to encounter difficulties pertaining to the inefficiency of multiscale feature interaction and indistinct edge responses. These issues impede their capacity to accurately capture alterations against intricate backgrounds. In order to address the aforementioned issues, this paper proposes a novel change detection network, which is herein referred to as MGFNet. (Multiscale Gated Fusion Network), which is founded on multilevel attention-guided optimization and multiscale differential fusion. By dynamically adjusting the multilevel feature fusion through attention weights, MGFNet enhances the effectiveness of features within intricate backgrounds and significantly alleviates the phenomenon of hollow interiors in objects. In order to validate the efficacy of the model, experiments were conducted using public datasets. The findings demonstrate that MGFNet attains an enhancement of 1.93% in F1 score and 2.72% in IoU on the SYSU-CD dataset when juxtaposed with baseline methodologies. These enhancements indicate that MGFNet offers superior accuracy and performance gains in change detection tasks.