Change map of flood detection using attention mechanism assisted quantum convolutional neural network utilising SAR and optical images

Climate change has intensified the frequency of extreme weather events, with floods posing a significant threat to environmental, economic, and societal stability. Accurate and timely flood detection is therefore essential for effective disaster response. By leveraging the complementary strengths of Sentinel-1 SAR (all-weather capability) and Sentinel-2 optical imagery (high spatial resolution), two quantum-assisted attention-based models, SA-QCNN and CA-QCNN, are proposed for generating flood change maps using pre- and post-event images. SA-QCNN computes attention within a single modality, while CA-QCNN enhances cross-domain feature fusion by using SAR features as keys, optical features as queries and values. Experiments on the Ombria flood dataset demonstrate that CA-QCNN achieves 96.88% overall accuracy and 91.2% F1-score, surpassing SA-QCNN’s 92.1% accuracy and 86.12% F1-score. These results confirm the advantages of quantum-assisted attention mechanisms for integrating multimodal satellite data, highlighting the robustness of CA-QCNN in capturing cross-domain features and underscoring the potential of quantum computing to advance flood detection in remote sensing.