Satellite Image Semantic Segmentation

TL;DR

This paper introduces a Swin Transformer-based method for automatic semantic segmentation of satellite images into six classes, utilizing a new dataset from IGN open data, with publicly available trained models and results.

Contribution

It presents a novel application of Swin Transformer architecture for satellite image segmentation and provides a new dataset with benchmark results.

Findings

Achieved accurate segmentation results on the new dataset

Demonstrated strengths and limitations of the proposed method

Made dataset and trained models publicly accessible

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.