Super-Resolution Water Body Extraction Based on MF-SegFormer

Tianyi Zhang, Weibin Li, Xihui Feng, Yi Ren, Chenhao Qin, Wenbo Ji, Xin Yang

Abstract: Surface water body (WB) as one of the world’s most critical natural resources, plays a significant role in forming and sustaining life. Therefore, accurate extraction of WB is particularly important. However, how to make full use of existing low-resolution images and labels to achieve accurate WB extraction on super-resolution images is a challenge. This study proposed a new method for super-resolution WB extraction on Landsat 8 OLI images based on MF-SegFormer. WB extraction and analysis were performed on test datasets at 30 m and 15 m resolutions, respectively. Precision, recall, F1-score, and mIoU are used as evaluation metrics. The results showed that the Multiscale Fusion SegFormer (MF-SegFormer) network has the best extraction results, and the constructed WB extraction model performed well on the 15 m super-resolution test dataset. In areas with rich details, our constructed WB extraction model demonstrates better performance in extracting from super-resolution images. This study provides technical support for further super-resolution WB research.