A Modelling and Evaluation Framework for EuroCrops-Driven Sentinel-2 Crop Segmentation
Researchers have developed a new framework for segmenting crops using Sentinel-2 satellite imagery, driven by EuroCrops parcel data. This pipeline harmonizes annotations and image data to create aligned pairs for training. A U-Net model trained on this dataset achieved a mean IoU of 0.7665 on an internal test split, demonstrating the value of multi-scale spatial representations over traditional methods. However, the study also revealed significant performance drops when evaluated on external datasets, highlighting challenges in transferring models across different annotation protocols and spatial coverages. AI
IMPACT This framework could improve agricultural monitoring and yield prediction by enabling more accurate crop segmentation from satellite imagery.