High-Resolution Multi-Source Fusion Cropland Segmentation (HSR-MFCS) dataset from Gaofen (GF) satellites

We introduce the High Spatial Resolution Multi-Source Fusion Cropland Segmentation (HSR-MFCS) dataset, a binary segmentation benchmark distinguishing cropland from non-cropland. The dataset covers typical plain and hilly terrains, encompassing diverse cropland types and parcel configurations, ensuring strong representativeness and applicability. It consists of paired high-resolution optical and SAR imagery with corresponding binary cropland labels. Optical data provide spectral and textural information, while SAR data include HH and HV polarization modes to capture structural characteristics and improve cropland identification. Ground-truth annotations were generated from field surveys and manual labeling conducted by the Sichuan Provincial Land Preparation Center.

The dataset covers four study areas in Sichuan Province, China:

ZhongJiang-A and ZhongJiang-B (Zhongjiang County, Deyang City), located in a transitional zone between hilly and basin landscapes. The terrain is predominantly hilly, with fragmented and irregular cropland patches interspersed with forested land, showing strong spatial heterogeneity. The region has a subtropical humid monsoon climate, with rice, wheat, maize, rapeseed, and cash crops as the main cultivated species.

MeiShan-C and MeiShan-D (Dongpo District, Meishan City), situated in the Sichuan Basin, representing a typical plain agricultural zone. Cropland parcels are generally regular, though localized heterogeneity arises from mixed forest and construction land. The area also has a subtropical humid monsoon climate, supporting multiple cropping systems with rice, wheat, maize, vegetables, and horticultural crops.