Global-aware siamese network for change detection on remote sensing images

Change detection (CD) in remote sensing images is one of the most important technical options to identify changes in observations in an efficient manner. CD has a wide range of applications, such as land use investigation, urban planning, environmental monitoring and disaster mapping. However, the frequently occurring class imbalance problem brings huge challenges to the change detection applications. To address this issue, we develop a novel global-aware siamese network (GAS-Net), aiming to generate global-aware features for efficient change detection by incorporating the relationships between scenes and foregrounds. The proposed GAS-Net, consisting of the global-attention module (GAM) and foreground-awareness module (FAM) that both learns contextual relationships and enhances symbiotic relation learning between scene and foreground. The experimental results demonstrate the effectiveness and robustness of the proposed GAS-Net, achieving up to 91.21% and 95.84% F1 score on two widely used public datasets, i.e., Levir-CD and Lebedev-CD dataset. The source code is available at https://github.com/xiaoxiangAQ/GAS-Net.