Abstract
In this paper, we propose a method for the automatic semantic segmentation of satellite images into six classes (sparse forest, dense forest, moor, herbaceous formation, building, and road). We rely on Swin Transformer architecture and build the dataset from IGN open data. We report quantitative and qualitative segmentation results on this dataset and discuss strengths and limitations. The dataset and the trained model are made publicly available.
本页仅发布可检索的书目信息与作者摘要。HTML 正文、PDF 和 LaTeX 源文件是独立阅读资源，不作为搜索引擎落地内容。打开页面后会按浏览器语言自动翻译标题与摘要，并写入缓存。