欢迎访问《农学学报》,

农学学报 ›› 2026, Vol. 16 ›› Issue (5): 71-78.doi: 10.11923/j.issn.2095-4050.cjas2025-0034

• 农业工程 农业机械 生物技术 食品科学 • 上一篇    下一篇

基于机器学习及多源遥感技术的区域水稻种植面积提取方法

樊冰1,2(), 王训诗3, 赵香玲3, 马良1,2, 黄乾1,2, 李福林1,2, 曾庆飞1   

  1. 1 山东省水利科学研究院, 济南 250014
    2 山东省水资源与水环境重点实验室, 济南 250014
    3 山东省海河淮河小清河流域水利管理服务中心, 济南 250100
  • 收稿日期:2025-02-18 修回日期:2025-06-19 出版日期:2026-05-20 发布日期:2026-05-15
  • 通讯作者:
    赵香玲,女,1980年出生,高级工程师,研究方向:建设管理规划设计。通信地址:250100 山东省济南市华阳路30号 省海河淮河小清河流域水利管理服务中心。
  • 作者简介:

    樊冰,男,1982年出生,高级工程师,硕士,主要研究方向:地图制图学与地理信息、遥感技术。通信地址:250014 山东省济南市历下区历山路125号 山东省水利科学研究院,E-mail:

  • 基金资助:
    山东省重点研发计划“大型喷灌装备关键技术研发与应用课题三”(重大科技创新工程2022CXGC2020707); 山东省水利科学研究院自选课题“无人机雷达技术在山丘区水利测绘中的研究与应用”(SDSKYZX202105); 山东省水利科技计划“湖东滞洪区洪水演进三维仿真及人员避险转移风险模拟、安全监测技术研究与应用”(HDZHQ-KY2020)

A Method for Extracting Regional Rice Planting Area Based on Machine Learning and Multi-Source Remote Sensing Technology

FAN Bing1,2(), WANG Xunshi3, ZHAO Xiangling3, MA Liang1,2, HUANG Qian1,2, LI Fulin1,2, ZENG Qingfei1   

  1. 1 Water Resources Research Institute of Shandong Province, Jinan 250014
    2 Key Laboratory of Water Resources and Environment of Shandong Province, Jinan 250014
    3 Haihe River, Huaihe River and Xiaoqinghe River Basin Water Conservancy Management and Service Center of Shandong Province, Jinan 250100
  • Received:2025-02-18 Revised:2025-06-19 Online:2026-05-20 Published:2026-05-15

摘要:

为精确监测水稻的分布与规模,满足水稻产量快速评估、农业水管理及决策支持需求,本研究基于国产卫星数据源构建了一套适用于县级区域的水稻种植面积快速识别策略。通过融合多时相高分二号多光谱遥感图像与合成孔径雷达(SAR)数据,结合归一化植被指数NDVI分析,进一步提取在VV极化模式下各种地物的后向散射特性,生成时序特征曲线。应用Savitzky-Golay滤波降噪技术优化数据质量,并采用随机森林分类模型提取水稻种植空间分布信息,系统揭示了水稻回波散射特性及其光谱动态变化规律。实验表明,本方法针对性地设计的水稻识别方案,提取水稻种植区域完整率精度达到92.65%,质量92.82%,优于常规国外卫星识别方法的精度。研究验证了国产卫星在农作物精细识别中的技术优势,为粮食安全监测、农田管理优化及灾害损失评估提供了高效解决方案。

关键词: 机器学习, 雷达, 多光谱, 多源遥感, 区域水稻, 归一化植被指数, 随机森林

Abstract:

To accurately monitor the distribution and scale of rice cultivation, address the needs for rapid assessment of rice yields, agricultural water management, and decision support, this study utilizes domestic satellite data to develop a rapid identification strategy for county-level adaptive rice planting areas. It conducts an in-depth analysis of the scattering characteristics and spectral changes of rice echoes. Combining multi-temporal high-resolution images from the Gaofen-2 satellite with normalized vegetation index (NDVI) and other information, the study initially identifies rice planting areas. Based on this, synthetic aperture radar (SAR) images are used to extract the backscattering characteristics of various ground objects under VV polarization mode, generating time-series feature curves. Savitzky-Golay filtering technology is applied to reduce noise interference. Finally, a random forest classification model is used to extract spatial distribution information of rice planting. This study has specifically designed a rice recognition scheme, achieving a completeness accuracy of 92.65% and a quality of 92.82%, surpassing the precision of conventional foreign satellite recognition methods. The research verifies the technical advantages of domestic satellites in fine crop identification, and provides an efficient solution for food security monitoring, farmland management optimization and disaster loss assessment.

Key words: machine learning, radar, multispectral, multi-source remote sensing, regional rice, normalized vegetation index, random forest

中图分类号: