China-PMF-10: a 10-m national map of plastic-mulched farmlands in China of 2020 using deep semantic segmentation

We provide China-PMF-10, which is the first remote sensing classification dataset for national-scale plastic-mulched farmlands (PMFs) of China in 2020 derived from Sentinel-2. To produce this dataset, we firstly organized 1°×1° grids covering cropland within the national extent. We then acquired pre-processed multi-temporal Sentinel-2 imagery via Google Earth Engine (GEE) and applied a random forest model to generate initial weak labels for deep learning training. Next, we integrated the active learning strategy and deep learning model to refine weak labels, enhance model robustness, and reduce false positives. Using the trained deep learning model, we mapped the national distribution of PMFs, producing the China-PMF-10 dataset with a high resolution of 10 meters. It shows that the estimated total PMFs area of China is about 67,000 km² in 2020. Validation indicates a high overall accuracy of about 86%. As the first publicly available spatial distribution dataset of China’s plastic-mulched farmlands, China-PMF-10 provides valuable data reference for agriculture, land use, remote sensing and environmental sciences researchers.

本研究提供China-PMF-10数据集，该数据集是首个基于Sentinel-2影像构建的2020年中国国家级尺度塑料地膜农田（plastic-mulched farmlands，PMFs）遥感分类数据集。为构建该数据集，研究团队首先在全国范围内布设覆盖耕地的1°×1°格网。随后通过谷歌地球引擎（Google Earth Engine，GEE）获取预处理后的多时相Sentinel-2影像，并采用随机森林模型生成深度学习训练所需的初始弱标签。接下来，研究团队结合主动学习策略与深度学习模型对弱标签进行精细化修正，以提升模型鲁棒性并降低误报率。借助训练完成的深度学习模型，研究团队完成了全国塑料地膜农田的空间分布制图，最终生成分辨率为10米的China-PMF-10数据集。该数据集显示，2020年中国塑料地膜农田总面积约为6.7万平方千米。精度验证结果表明，该数据集的总体精度约为86%。作为首个公开可获取的中国塑料地膜农田空间分布数据集，China-PMF-10可为农业、土地利用、遥感与环境科学领域的研究人员提供极具价值的数据参考。

搜集汇总

背景与挑战

背景概述

China-PMF-10是中国首个全国范围内塑料覆盖农田（PMF）的10米高分辨率遥感分类数据集，基于2020年Sentinel-2影像和深度学习技术生成，覆盖面积约67,000平方公里，总体准确率为86%。该数据集为农业、土地利用、遥感和环境科学研究提供了重要数据参考。