基于哨兵2号多光谱影像的水稻倒伏识别与分类  

Identification and Classification of Rice Lodging Based on Sentinel-2 Multispectral Image

作　　者：任志鹏[1] 高睿 王大庆 REN Zhi-peng;GAO Rui;WANG Da-qing(Heilongjiang Academy of Land Reclamation Sciences,Harbin 150000,China;School of Electrical and Information,Northeast Agricultural University,Harbin 150000,China;Heilongjiang Agricultural Reclamation Management Cadre College,Harbin 150000,China;Haikou University of Economic,Haikou 571127,China)

机构地区：[1]黑龙江省农垦科学院,哈尔滨150000 [2]东北农业大学电气与信息学院,哈尔滨150000 [3]黑龙江省农垦管理干部学院,哈尔滨150000 [4]海口经济学院,海口571127

出　　处：《节水灌溉》2022年第7期44-50,共7页Water Saving Irrigation

基　　金：黑土地保护与利用科技创新工程专项资助(XDA28000000)。

摘　　要：为加强对粮食生产区水稻倒伏面积、位置及严重程度的识别和监测,基于黑龙江红卫农场2019年9月22日的哨兵2号卫星多光谱遥感影像计算水稻的光谱反射率、植被指数以及图像纹理3种特征,利用决策树分类法对倒伏水稻进行识别和分类。首先根据现场调查和目视解译结果选定倒伏水稻样点,分析正常、轻度倒伏、中度倒伏、重度倒伏4种倒伏类型水稻的光谱反射率特征,发现在绿光、红光、红边3以及近红外1处存在较大差异。植被指数特征中,NDVI和RVI均随水稻倒伏程度加深而下降,GRVI、DVI和NDREⅠ则在水稻倒伏后逐渐增加。其中,不同倒伏类型水稻的DVI显示出了较大的差异。4种水稻倒伏类型在可见光波段的均值纹理特征差异显著,尤其是蓝光波段的纹理均值是区分不同倒伏类型的重要特征。基于对水稻倒伏敏感的特征量构建决策树,成功区分了正常、轻度倒伏、中度倒伏和重度倒伏4种倒伏类型,与实际倒伏面积对比的识别误差分别为5.33%、6.51%、10.25%和-7.75%,识别的准确度较高。In order to strengthen the detection of the area,location and severity of rice lodging in grain production areas,in this study,three main features of rice lodging in Hongwei Farm,including spectral reflectance,vegetation index and texture,were calculated,and the decision tree classifier was used to identify lodging rice based on Sentinel-2 multi-spectral images.Firstly,rice samples were selected based on field investigation and visual interpretation,and the spectral reflectance characteristics of four types lodging rice,including normal,mild,moderate and severe level,were analyzed.It wais found that there were great differences in green,red,red edge 3 and near infrared 1bands.Among vegetation index features,NDVI and RVI decreased with the deepening of rice lodging,while GRVI,DVI and NDREⅠincreased gradually after rice lodging.The DVI of rice with different lodging types showed great difference.The mean texture characteristics of four rice lodging types differed significantly in visible band,especially in blue band,which was an important feature to distinguish different lodging types.The decision tree was constructed based on the characteristics sensitive to rice lodging,and the four types of normal,mild,moderate and severe lodging were successfully distinguished.Compared with the actual lodging area,the identification errors were 5.33%,6.51%,10.25%and-7.75%,respectively.

关 键 词：水稻倒伏识别 多光谱影像 光谱特征 植被指数 纹理特征 

分 类 号：P407.8[天文地球—大气科学及气象学]