单双季水稻遥感制图研究进展

Progress in single and double cropping rice mapping by using remote sensing

关键词:  单双季水稻  农作物遥感制图  多源遥感数据  样本  分类方法

Key words:  single and double cropping rice  crop type mapping  multisource remote sensing data  crop samples  classification methods

摘要:

【目的】 单双季水稻面积分布和时空变化监测对于农作物种植结构调整、国家粮食安全保障以及气候变化研究等具有重要意义。【方法】 文章概括了单双季水稻遥感提取的理论基础；采用文献综述的方法，从单双季水稻识别的遥感数据源、样本获取方法和模型构建等3个方面重点评述了不同遥感制图方法的特点及应用情况；总结出单双季水稻制图面临的挑战，并展望了未来单双季水稻遥感提取研究的发展方向。【结果】 （1）微波遥感影像是弥补多云雨区域光学遥感影像不足的重要数据源；（2）传统实地采样或目视解译难以满足大区域单双季水稻遥感制图和年际更新需求，农作物样本自动生成策略以及面向有限样本的制图方法在应对上述问题方面具有潜力。（3）现有单双季水稻遥感制图方法可分为基于关键物候特征的决策树模型、时间序列光谱相似度匹配、机器学习以及深度学习4类。【结论】 未来单双季水稻遥感提取将以大区域、长时序产品生产和动态监测为目标，重点围绕样本自动化获取、多源数据协同利用、高精度分类模型设计以及时空迁移策略构建等方面开展深入研究，以切实满足农业生产管理实际应用需求。

Abstract:

Purpose］ Timely and accurate information on the spatial distribution of single and double cropping rice is crucial for adjusting crop planting patterns，guaranteeing food security，and mitigating climate change.Method］ This paper reviewed and summarized relevant research on single and double cropping rice mapping using remote sensing technology over the past two decades. We analyzed the advantages，limitations，and applications of different methods from 3 perspectives：remote sensing data sources，sample generation methods，and classification models. Additionally，we highlighted existing challenges and provided insights into the future development of single and double cropping rice classification.Result］ （1）The literature review indicated that synthetic aperture radar（SAR）could better capture the key phenological information of different rice cropping patterns in cloudy or rainy regions，compared to optical images. （2）Traditional field surveys or visual interpretation could hardly meet the demands of mapping and inter-annual updating of single and double cropping rice at large scales. The automatic sample generation approach and classification methods for limited-sample scenarios held promise in addressing these challenges. （3）Current methods for mapping single and double-cropping rice could be grouped into 4 categories：phenology-based decision tree models，spectral similarity matching approaches，machine learning methods，and deep learning methods.Conclusion］ Future research on single and double cropping rice mapping will focus on generating large-scale and long-term datasets，monitoring spatio-temporal dynamics，and advancing automatic sample generation，multi-source data integration，high-precision classification models and transfer strategies to meet the diverse needs of agricultural applications.