Author(s): Chen Zhe; Xiang Daxiang; Zhao Jing; Li Jingwei; Jiang Ying; Wu Yibang; Wen Xiongfei
Linked Author(s):
Keywords: Remote sensing; Flood detection; Data fusion; SAR; Multi-spectral
Abstract: Floods threats to people’s lives and property. Monitoring the spatial and temporal extents of flood water is vital for water resource management. In recent years, timely and accurate flood detection products derived from satellite remote sensing imagery are becoming effective methods of responding flood disaster. Based on remote sensing technology, researchers have done a lot of work in flood detection. It is proved that using image time series and data fusion techniques to increase the accuracy of flood detection is promising. Combining optical and microwave satellite date can help to increase spatial and temporal resolution. Although the identification of temporary water body in flood disasters mainly rely on multi- temporal change detection methods, this type of approach generally requires a pair of images acquired before and after a flood event, which was greatly limited due to the mandatory demand for satellite imagery before disasters. This article is to propose a deep learning-based fusion approach using SAR and optical images for improving temporary flood water mapping. A hybrid CNN-SVM model was used in deep learning process, which utilized pre-trained Convolutional Neural Network (CNN) as the feature extraction to improve the accuracy of Support Vector Machine (SVM) model. The approach is tested over Wuhan in China for 2016. Result shows that our approach provides better accuracy for mapping flood area compared to traditional approaches.
Year: 2024