The Loess Plateau is an important ecological barrier and an agricultural production base in China, with its arable land resources playing a key role in regional ecological balance and economic development. This dataset is based on high-resolution optical remote sensing imagery (GF-2) from 2022 to 2023, and was generated by combining manual visual interpretation with deep learning algorithms to produce a sample dataset of arable land in the Yanhe River Basin. The dataset covers single-band grayscale (panchromatic) images, multiband RGB (multispectral) true-color (3,2,1) synthetic image samples, and multiband RGB true-color synthetic fusion image (panchromatic fused with multispectral imagery) samples, with spatial resolutions of 4 m, 2 m, and 2 m, and a patch size of 256×256 pixels, respectively. The dataset includes both manually annotated and augmented samples, totaling 3,000, with a data volume of about 648 MB, and file formats, including tif,.tfw and txt. After data quality control, the dataset shows high data quality, with a Kappa coefficient above 0.8, overall accuracy of 89.64%, precision of 92.13%, and recall of 88.51%, indicating high data quality. This dataset provides high-quality data support for dynamic monitoring of arable land resources in the Yanhe River Basin, assessment of ecological effects, and the land-use planning, offering a scientific basis for regional ecological research and land resource management. Moreover, this dataset can also support research in geography, ecology and soil and water conservation, providing data resources and reference materials for related scientific studies.
Shi et al. (Sun,) studied this question.