🤖 AI Summary
This study addresses challenges in crop semantic segmentation using Sentinel-2 imagery and EuroCrops field annotations, including label heterogeneity, domain shift, and limited cross-regional generalization. To tackle these issues, the authors develop a configurable data processing and evaluation framework that harmonizes multi-source vector labels into aligned multispectral image–mask pairs. They train a four-level U-Net architecture with Group Normalization to segment ten crop classes, leveraging all ten Sentinel-2 spectral bands and optimizing a composite loss function combining class-weighted cross-entropy and Dice loss. Experimental results show a mean Intersection over Union (mIoU) of 0.7665 and pixel accuracy of 0.8693 on an internal test set. The first systematic cross-regional evaluation reveals strong transfer performance for dominant crops like maize and wheat, yet highlights limited generalization for minority classes and single-temporal inputs.
📝 Abstract
This work presents a configurable pipeline for generating semantic-segmentation-ready agricultural datasets from Sentinel-2 imagery and EuroCrops parcel-level annotations. The workflow transforms heterogeneous vector crop annotations into aligned multispectral image--mask pairs through label harmonization, Sentinel-2 product selection, spatial alignment, rasterization, patch extraction, quality filtering, and class-aware sample selection. The generated dataset contains 67,337 patches from five European countries and uses a reduced taxonomy of ten crop classes plus background.
A four-level U-Net with Group Normalization was trained using 10 Sentinel-2 spectral bands and a composite loss combining class-weighted cross-entropy and Dice loss. On the internal EuroCrops-based test split, the model achieved a mean Intersection over Union (mIoU) of 0.7665, a pixel accuracy of 0.8693, and a mean class accuracy of 0.9072. Compared with spectral and spatial-context Random Forest baselines, the U-Net showed the importance of learned multi-scale spatial representations for crop segmentation.
External evaluation was performed on unseen Belgian EuroCrops subsets, DACIA5, and PASTIS. The results show a clear performance gap under external and cross-dataset evaluation, especially for benchmarks with different taxonomies, annotation protocols, spatial coverage, or temporal organization. The model transfers more reliably to dominant and taxonomically aligned classes such as maize and wheat, while performance remains limited for several minority classes and for the adapted single-date PASTIS setting. These findings highlight both the potential and the limitations of using EuroCrops-derived supervision for Sentinel-2 crop segmentation under realistic domain shifts.