结合地块信息分类的复杂山地水稻种植区域提取:以湘西永顺为例  

Extraction of Complex Mountainous Rice Planting Areas Combined with Plot Information Classification:Taking Yongshun,Xiangxi as an example

作　　者：陈磊士 汪天颖 谢佰承[1,2] 郑仲帅 帅细强 CHEN Leishi;WANG Tianying;XIE Baicheng;ZHENG Zhongshuai;SHUAI Xiqiang(Hunan Meteorological Research Institute,Changsha,Hunan 410118;Hunan Key Laboratory of Meteorological Disaster Prevention and Reduction,Changsha,Hunan 410118;Dongting Lake National Climate Observatory,Yueyang,Hunan 414000,China)

机构地区：[1]湖南省气象科学研究所,湖南长沙410118 [2]气象防灾减灾湖南省重点实验室,湖南长沙410118 [3]洞庭湖国家气候观象台,湖南岳阳414000

出　　处：《贵州农业科学》2025年第2期134-142,共9页Guizhou Agricultural Sciences

基　　金：国家重点研发计划项目“长江中下游水稻高温热害监测评估及预警预测研究”(2022YFD2300203);湖南省气象局创新发展重点专项“湖南省作物智能观测与农用天气预报释用系统建设”(CXFZ2023-ZDZX02);湖南省气象局创新发展专项“第三代杂交稻高产栽培气候适宜性研究”(CXFZ2023-FZZX30)。

摘　　要：【目的】结合监督分类地块信息建立水稻连续遥感物候曲线,为复杂山地水稻种植区域提取提供理论依据和技术指导。【方法】利用多光谱和SAR影像建立多维度空间特征,开展多特征结合的机器学习监督分类稻田地块信息提取(传统监督分类法);基于非监督分类空间聚类算法,利用3年连续多光谱和SAR影像,耦合构建水稻完整连续遥感物候曲线,在稻田地块信息的基础上筛选出精确的一季稻物候特征地块,实现高精度的复杂山地水稻种植区域提取(传统监督分类+长时序物候曲线分类法),并比对分析2种分类方法水稻种植区域的提取精度。【结果】非监督分类长时序物候信息的分类精度优于仅使用监督分类算法的稻田地块分类结果,二者结合的方法错分、漏分情况均少于传统监督分类算法,传统监督分类+长时序物候曲线分类法的总体分类精度为94.60%,Kappa系数为0.89,优于传统监督分类法(精度为89.20%,Kappa系数为0.82),对湘西永顺县一季稻种植遥感提取面积较2021农业统计年鉴数据仅高出1.8%,分类结果和高清底图可基本契合,符合湘西地区复杂山地一季稻的种植分布特征,降低了研究区下垫面破碎、耕地分布分散、可用影像较少等问题对水稻遥感分类的不利影响。【结论】传统监督分类+长时序物候曲线分类法优于传统监督分类方法,可筛选出精确的一季稻物候特征地块,可有效发挥在线云平台的数据与算力优势,建立结合地块信息分类的复杂山地水稻种植区域提取模型。【Objective】Combined with supervised classification plot information,the continuous remote sensing phenology curve was constructed,which provided theoretical basis and technique guidance for the planting area of rice in complex mountainous.【Method】The multi-dimensional spatial features was established by using multi-spectral and SAR images to carry out multi-feature machine learning supervised classification for extracting plot information of rice-planting field.Based on the spatial clustering algorithm of unsupervised classification,the three years of continuous multi-spectral and SAR images was used to construct a complete continuous remote sensing phenology curve of rice,and screen accurate plot with one-season rice phenological characteristics,which achieved high-precision extraction of complex mountainous rice-planting areas,and compare the extraction accuracy of rice-planting areas by using the combined method and the traditional method.【Result】The classification accuracy of long-term phenological information with unsupervised classification was better than the classification results of rice field plots using only supervised classification algorithms.The combination of the two methods had fewer wrong classification and missing classifications than the traditional supervised classification algorithm.The overall classification accuracy of combining traditional supervised classification with classification of long-term phenological curve was 94.60%and the Kappa coefficient was 0.89,which was better than the accuracy of 89.20%and Kappa coefficient of 0.82 of the traditional supervised classification method.The extracted planting area of one-season rice in Yongshun County of western Hunan by remote sensing was only 1.8%higher than the data from yearbook of agricultural statistics in 2021.The classification results and the high-definition base map could be basically consistent,which was in line with the planting distribution characteristics of one-season rice in the complex mountainous areas of western

关 键 词：水稻 地块信息 农业遥感 物候信息 作物识别 复杂山地 

分 类 号：S127[农业科学—农业基础科学] TP79[自动化与计算机技术—检测技术与自动化装置]