A Deep Learning Benchmark Dataset for Cropland Parcel Boundary Extraction from Multi-Source Optical Remote Sensing Imagery

We provide a standardized dataset of diversified parcel labels based on multi-source remote sensing imagery. The parcels included in this dataset are sourced from four countries: the Netherlands, Denmark, Spain, and China, encompassing over 140,000 cropland parcel. The average parcel size across different study areas ranges from 0.05 hectares to 10 hectares.