基于高分六号多光谱数据的油茶林遥感提取研究  

Study on Remote Sensing Extracting Camellia oleifera Forest Based on GF-6 Multispectral Imagery

作　　者：高霞霞 汪天颖 陈磊士 刘思华 帅细强 谢傲[1,2] GAO Xiaxia;WANG Tianying;CHEN Leishi;LIU Sihua;SHUAI Xiqiang;XIE Ao(Hunan Key Laboratory of Meteorological Disaster Prevention and Reduction,Changsha,Hunan 410118;Hunan Meteorological Research Institute,Changsha,Hunan 410118;Dongting Lake National Climate Observatory,Yueyang,Hunan 414000,China)

机构地区：[1]气象防灾减灾湖南省重点实验室,湖南长沙410118 [2]湖南省气象科学研究所,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》2024年第12期117-125,共9页Guizhou Agricultural Sciences

基　　金：湖南省自然科学基金重大专项“中小尺度强对流天气系统及其衍生灾害多源卫星遥感模型研究”(2021JC0009);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);湖南省气象局短平快科研课题“基于多时相卫星遥感数据的油茶分布提取研究”(XQKJ22B028);中国气象局创新发展专项“油茶产量形成关键期主要气候影响因子研究”(CXFZ2022J055)。

摘　　要：【目的】探究基于高分六号多光谱数据的油茶林遥感提取方法,为油茶林的高精度识别及监管监测提供参考。【方法】基于高分光学遥感影像分类提取技术,利用GF-6 WFV多光谱影像和植被指数,构建基于随机森林结合光谱特征、红边指数特征等多特征的分类模型,对比分析油茶林和其他地物4个时相的光谱特征、类别可分离性及植被指数特征,评估模型提取精度。【结果】基于随机森林的8波段分类加入优选植被指数后,总体分类精度达95.63%,较原分类精度提高2.96百分点;油茶林制图精度和用户精度分别达93.46%和96.51%,分别提升2.60百分点和3.67百分点。红边指数NDVI_(750)、IRECI能提高模型区分油茶林与其他地类信息能力,从而提升对油茶林的提取精度。【结论】GF-6的宽幅8波段多光谱影像宏观上具备有效识别油茶林的能力,其中红边波段、黄波段、蓝波段和紫波段在油茶林精准识别中发挥重要作用。选用加入红边指数NDVI_(750)、IRECI和植被指数NDVI、RVI后的随机森林8波段多特征分类模型可有效提高识别油茶林地斑块精度。【Objective】The remote sensing extraction method for Camellia oleifera forests based on GF-6 multispectral imagery was explored to provide references for high-precision recognition and monitoring of C.oleifera forests.【Method】Based on the classification extraction technology of high-resolution optical remote sensing imaging,GF-6 WFV multispectral images and vegetation indices were used to construct a classification model on the basis of random forest combined with spectral features and red edge index features,and then the spectral features,category separability and vegetation index features of C.oleifera forest and other land cover in four phases were compared and analyzed to evaluate the extraction accuracy of the model.【Result】After adding the optimal vegetation index to the 8-band classification based on random forest,the overall classification accuracy reached 95.63%,an increase of 2.96 percentage points compared to the original classification accuracy.The mapping accuracy and user accuracy of C.oleifera forest reached 93.46% and 96.51%,respectively,with an increase of 2.60 percentage points and 3.67 percentage points,respectively.The red edge indices NDVI_(750) and IRECI could improve the model's ability to distinguish C.oleifera forests from other land types,thereby enhancing the accuracy of extracting C.oleifera forests.【Conclusion】GF-6's wide 8-band multispectral images have the ability to effectively identify C.oleifera forests macroscopically,among which the red-edge band,yellow band,blue band and purple band play an important role in the accurate recognistion of C.oleifera.The use of a random forest 8-band feature classification model adding red edge index NDVI_(750) and IRECI,and vegetation index NDVI and RVI can effectively improve the accuracy of identifying C.oleifera forest patches.

关 键 词：油茶 高分六号 红边指数 多时相 随机森林 农业遥感 

分 类 号：S794.4[农业科学—林木遗传育种]