Gully Erosion Dataset

An gully erosion dataset was constructed by using high-resolution optical imagery. The imagery covers key agricultural regions within Lishu County (Siping City), Nong'an County (Changchun City), Qianguo County and Fuyu City (Songyuan City) in Jilin Province, China. Data acquisition occurred during periods of low vegetation cover (April-May and October-November) between 2020 and 2023 to optimize gully visibility. The study areas are predominantly characterized by chernozem, known for their susceptibility to soil erosion.
Integrating findings from field surveys and high-resolution remote sensing interpretation, the identified gully erosion features were categorized into three primary types: (1) ephemeral gullies (small-scale, transient channels formed by concentrated overland flow, often recurring seasonally); (2) permanent gullies (deep-seated, stable channels with near-vertical sidewalls and irreversible morphology); and (3) gullies in complex environments (features whose detection is hindered by vegetation, terrain complexity, or anthropogenic disturbance). An initial dataset of gully images was constructed. All images were standardized to 512×512 pixels for uniformity. Using the LabelMe software, annotations were created, marking gully pixels in red and background pixels in black. To enhance dataset scale and diversity for robust model training, data augmentation was performed using spatial-level enhancements (rotation, flipping) and pixel-level enhancements (blurring, noise addition, contrast adjustment) for each gully type.