MT4AFE: A Deep Learning Benchmark Dataset for Agricultural Field Extraction

This paper introduces MT4AFE (Multi-Task Learning for Agricultural Field Extraction), a new large-scale dataset dedicated to the semantic segmentation of agricultural fields. MT4AFE consists of high-resolution Sentinel-2 Level-3A satellite images and comprehensive reference data on agricultural fields from the French graphical parcel registry (RPG). The dataset contains 132,222 paired image-label patches, each measuring 256 × 256 pixels, derived from multiple Sentinel-2 tiles across France. It covers a wide diversity of landscapes, environmental settings, and agricultural conditions. MT4AFE has been curated to provide researchers with a ready-to-use dataset to develop and assess deep learning (DL) methods for agricultural field delineation. The dataset offers not only pixel-level segmentation labels but also field boundary annotations and distance maps, making it particularly well-suited for multi-task learning approaches. MT4AFE contributes to advancing the state-of-the-art in agricultural remote sensing by offering a publicly available resource for reproducible research. The dataset's diverse coverage, with three different years of Sentinel-2 Level-3A data, will help further machine learning research in agricultural monitoring and improve data-driven methods for field delineation. The dataset is available at: https://doi.org/10.5281/zenodo.15395167.