Farmland Shelterbelts in China

This dataset provides high-resolution remote sensing imagery and pixel-level annotation samples for farmland shelterbelt extraction in Jilin Province, China. The dataset was developed using multispectral imagery from the Jilin-1 JL1GF02 satellite constellation acquired in late May and early June 2020. Three multispectral scenes were selected from representative farmland shelterbelt regions, with acquisition dates of May 27, May 28, and June 7, 2020. The selected imagery has a spatial resolution of 3 m and includes four multispectral bands covering the blue (450–510 nm), green (510–580 nm), red (630–690 nm), and near-infrared (770–895 nm) spectral ranges.

Farmland shelterbelt samples were generated through visual interpretation of Google Earth Pro imagery and subsequently validated using high-resolution unmanned aerial vehicle (UAV) imagery and field survey data. The validated vector samples were rasterized and precisely co-registered with the corresponding Jilin-1 imagery to achieve pixel-level spatial consistency between the input imagery and annotation masks. Both imagery and labels were then divided into paired patches of 256 × 256 pixels to preserve the fine-scale linear characteristics of farmland shelterbelts.

A quality-control procedure was applied during dataset preparation. Image-label pairs were assessed for co-registration accuracy and manually inspected, and samples with a misalignment greater than one pixel were discarded. The patches were further screened according to radiometric and textural characteristics to remove samples with abnormal radiometric values or insufficient texture variability. The resulting dataset contains 1,893 image-label pairs, including 1,325 training samples, 379 validation samples, and 189 testing samples, following a random 7:2:1 split.

This dataset is intended to support research on farmland shelterbelt extraction, remote sensing image segmentation, fine-scale ecological feature mapping, and deep-learning-based monitoring of linear shelterbelt structures. It can also serve as a benchmark dataset for developing and evaluating remote sensing methods for farmland shelterbelt mapping under heterogeneous agricultural landscape conditions.