欢迎访问《生态学杂志》官方网站,今天是

生态学杂志 ›› 2026, Vol. 45 ›› Issue (11): 3841-3854.doi: 10.13292/j.1000-4890.202611.017

• 技术与方法 • 上一篇    下一篇

遥感产品辅助样本自动解译的水稻种植区提取

冯雪1,2,3,徐源1,2,3,杨会巾1,2,3*,李宁1,2,3,赵建辉1,2,3,吴文福1,2,3   

  1. (1河南大学计算机与信息工程学院, 河南开封 475004; 2河南大学河南省大数据分析与处理重点实验室, 河南开封 475004; 3河南大学河南省空间信息处理工程研究中心, 河南开封 475004)
  • 出版日期:2026-11-10 发布日期:2026-11-19

Extraction of rice planting areas based on remote sensing product-assisted automatic sample interpretation.

FENG Xue1,2,3, XU Yuan1,2,3, YANG Huijin1,2,3*, LI Ning1,2,3, ZHAO Jianhui1,2,3, WU Wenfu1,2,3   

  1. (1School of Computer and Information Engineering, Henan University, Kaifeng 475004, Henan, China; 2Henan Key Laboratory of Big Data Analysis and Processing, Kaifeng 475004, Henan, China; 3Henan Province Engineering Research Center of Spatial Information Processing, Kaifeng 475004, Henan, China).

  • Online:2026-11-10 Published:2026-11-19

摘要: 大区域水稻种植面积提取对保障国家粮食安全和可持续发展具有重要意义。针对年际农作物轮作导致历史样本标签不匹配问题和产品精度限制带来的水稻样本噪声,提出一种遥感产品辅助样本自动解译方法(RSPASI),结合机器学习完成水稻种植面积快速提取。该方法以现有水稻产品生成候选样本,基于目标年份Sentinel-1垂直发射/水平接收(VH)极化后向散射系数时序,通过迭代自组织数据分析技术(ISODATA)聚类候选水稻样本,并结合数量筛选和方差约束完成样本净化,实现历史样本标签向目标年份的迁移;随后,融合Sentinel-1和Sentinel-2时序特征,分别构建RSPASI-RF、RSPASI-SVM和RSPASI-KNN模型,实现黑龙江、广东和湖南3个不同水稻种植模式区的水稻种植面积快速提取。结果表明,时序VH后向散射特征在3个研究区的重要性均居首位,贡献占比分别为48.09%、55.30%和74.39%。RSPASI-RF在3个研究区均获得最高分类精度,其总体精度(OA)分别为96.53%、93.61%和95.24%,Kappa系数分别为0.882、0.845和0.905;其中,在种植景观较复杂的湖南研究区,其OA较RSPASI-SVM和RSPASI-KNN分别提高2.58%和9.95%。产品对比结果显示,RSPASI-RF在黑龙江研究区的OA较You-RiceMap和Shen-RiceMap分别提高1.24%和5.14%,在广东研究区的OA和Kappa系数较Pan-RiceMap分别提高5.62%和0.177。研究表明,样本数量与时序方差联合约束能够提高历史产品样本的置信度,RSPASI-RF可适用于不同水稻种植模式区域的高精度制图,为遥感辅助的大区域水稻种植面积快速提取提供了新的技术方案。


关键词: 水稻种植面积提取, 样本自动解译, 时间序列Sentinel-1/2影像, 遥感产品

Abstract: The extraction of large-scale rice planting area plays a vital role in ensuring national food security and sustainable development. To address the issue of sample label mismatch caused by interannual crop rotation, as well as the problem of sample noise in rice plantation data due to the limited accuracy of remote sensing products, we proposed a method termed Remote Sensing Products for Automatic Sample Interpretation (RSPASI) and integrated it with machine-learning algorithms for rapid extraction of rice planting areas. The method first generates candidate samples from currently available products. Candidate rice samples are then clustered using the Iterative Self-Organizing Data Analysis Technique (ISODATA) based on target-year Sentinel-1 vertical transmit/horizontal receive (VH) polarized backscatter coefficient time series. Sample-size screening and variance constraints are subsequently combined to purify the candidate samples, thereby transferring historical sample labels to the target year. Sentinel-1 and Sentinel-2 time-series features are then integrated with random forest (RF), support vector machine (SVM), and K-nearest neighbors (KNN) to construct the RSPASI-RF, RSPASI-SVM, and RSPASI-KNN models, respectively, for rapid extraction of rice planting areas in Heilongjiang, Guangdong, and Hunan, which represent distinct rice cropping systems. The results showed that the time-series VH backscatter feature ranked first in importance across all the three study areas, contributing 48.09%, 55.30%, and 74.39% in Heilongjiang, Guangdong, and Hunan, respectively. RSPASI-RF achieved the highest classification accuracy across all the three study areas, with overall accuracy (OA) values of 96.53%, 93.61%, and 95.24% and Kappa coefficients of 0.882, 0.845, and 0.905, respectively. In Hunan, which has a more heterogeneous agricultural landscape, the OA of RSPASI-RF exceeded those of RSPASI-SVM and RSPASI-KNN by 2.58% and 9.95%, respectively. Product comparisons showed that the OA of RSPASI-RF in Heilongjiang was 1.24% and 5.14% higher than those of You-RiceMap and Shen-RiceMap, respectively. In Guangdong, RSPASI-RF outperformed Pan-RiceMap by 5.62% in OA and by 0.177 in the Kappa coefficient. These findings demonstrate that jointly constraining sample size and temporal variance can improve the confidence of samples derived from historical products. RSPASI-RF enables high-accuracy mapping across regions with different rice cropping systems and provides a new technical approach for rapid extraction of large-scale rice planting areas assisted by remote sensing products.


Key words: rice planting area extraction, automatic sample interpretation, time series Sentinel-1/2 imagery, remote sensing product