Fine-Grained Grass Segmentation Dataset

The dataset was created using high-resolution (8 m) satellite imagery from the Gaofen series (Gaofen-2 and Gaofen-6), captured in 2019 over Maduo County, China, located in the Yellow River source area. This region is known for its high-altitude, alpine grasslands, and complex terrain, with coordinates between 33°50'–35°40' N latitude and 96°50'–99°20' E longitude.

We collected two 13,872 × 13,150-pixel images from Gaofen-6 and two 7,300 × 6,905-pixel images from Gaofen-2, all containing red, green, and blue spectral bands. These images provide critical information for fine-grained grass extraction.

Labeling was assisted by the X-AnyLabeling tool, supplemented with manual refinements to ensure high accuracy. Grassland coverage was classified into five levels, based on national grassland survey standards:

• Low coverage (<10%)
• Medium-low coverage (10%–25%)
• Medium coverage (25%–50%)
• Medium-high coverage (50%–75%)
• High coverage (>75%)

The final dataset comprises 1,151 pairs of 256×256 patches, split into training and testing sets with an 8:2 ratio. This dataset, with its detailed and accurate labeling, is a valuable resource for advancing remote sensing applications, particularly in ecologically sensitive and high-altitude regions like the Yellow River source area.