Cross-modal change detection flood extraction based on convolutional neural network

• DOI: 10.1016/j.jag.2023.103197
• Publication Date: 2023-01-20T23:26:58Z
• Authors: Xiaoning He; Shuangcheng Zhang; Bowei Xue; Tong Zhao; Tong Wu
• Keywords: Feature (linguistics)

Flood events are often accompanied by rainy weather, which limits the applicability of optical satellite images, whereas synthetic aperture radar (SAR) is less sensitive to weather and sunlight conditions. Although remarkable progress has been made in flood detection using heterogeneous multispectral SAR there a lack publicly available large-scale datasets more efforts required for exploiting deep neural networks detection. This study constructed pre-disaster Sentinel-2 post-disaster Sentinel-1 mapping dataset named CAU-Flood containing 18 plots with careful image preprocessing human annotation. A new convolutional network (CNN), cross-modal change (CMCDNet), was also proposed images. The employs encoder-decoder structure performs feature fusion at multiple stages gating self-attention modules. Furthermore, overcomes misalignment issue during decoding embedding alignment module upsampling operation. CMCDNet outperformed SOTA methods terms accuracy achieved an intersection over union (IoU) 89.84%. codes at: https://github.com/CAU-HE/CMCDNet.