| 基于无人机多光谱数据的水稻LAI反演与应用 |
| |
| 引用本文： | 王靖,彭漪,刘小娟,莫佳才,梁婷.基于无人机多光谱数据的水稻LAI反演与应用[J].中国农业大学学报,2021,26(12):145-156. |
| |
| 作者姓名： | 王靖 彭漪 刘小娟 莫佳才 梁婷 |
| |
| 作者单位： | 武汉大学 遥感信息工程学院, 武汉 430079;武汉大学 生命科学学院, 武汉 430072 |
| |
| 基金项目： | 国家自然科学基金项目(41771381) |
| |
| 摘 要： | 为比较和验证不同叶面积指数(LAI)机器学习模型迁移后的稳定性，以无人机获取的湖北鄂州与海南水稻田的多光谱影像数据为研究对象，使用8 种植被指数经验模型与3 种机器学习方法对鄂州试验田的水稻LAI进行反演，并推广至海南试验区。结果表明:1)非线性模型在鄂州试验数据的建模集测试中精确度较优，其中归一化红边差值(NDRE)非线性模型验证的变异系数(CV)=31.05%，但推广至海南试验区后精确度下降严重(验证集CV=74.90%)，可移植性差；2)线性模型和机器学习模型在模型移植后表现出较优的稳定性，其中梯度提升回归(GBR)二波段模型在鄂州数据建模集CV=28.91%，在海南数据验证集CV=26.58%；3)增强植被指数(EVI2)线性拟合模型在鄂州数据建模集CV=33.78%，在海南数据验证集CV=27.90%。最后使用EVI2构建的经验模型，对各生育时期2 种不同水稻(珞优9348和丰两优4号)的LAI进行预测，结果表明在相同氮水平下，珞优9348表现为氮高效品种。
|
| 关 键 词： | 水稻 叶面积指数 植被指数 氮高效 |
| 收稿时间： | 2021/4/13 0:00:00 |
Inversion and application of rice LAI based on UAV multispectral data |
| |
| WANG Jing,PENG Yi,LIU Xiaojuan,MO Jiacai,LIANG Ting.Inversion and application of rice LAI based on UAV multispectral data[J].Journal of China Agricultural University,2021,26(12):145-156. |
| |
| Authors: | WANG Jing PENG Yi LIU Xiaojuan MO Jiacai LIANG Ting |
| |
| Institution: | School of Remote Sensing Information Engineering, Wuhan University, Wuhan 430079, China; School of Life Sciences, Wuhan University, Wuhan 430072, China |
| |
| Abstract: | In order to compare and verify the stability of different leaf area index(LAI)machine learning estimation models after migration, the multispectral image data of rice fields in Ezhou, Hubei Province and Hainan Province obtained by unmanned aerial vehicle(UAV)were used for. The rice LAI of Ezhou experimental field was estimated by using eight vegetation index empirical model and three machine learning methods, and then validated in Hainan experimental area. The results showed that: 1)The accuracy of the nonlinear model was better in the modeling set test of Ezhou experimental data. The coefficient of vatiation(CV)of using nonlinear fitting model was 31. 05%. However, the accuracy of the NDRE decreased seriously when it was extended to Hainan experimental area(CV=74. 90%), and the portability was poor. 2)Linear model and machine learning model showed better stability after model validation on other dataset. The CV of GBR two-band model in Ezhou data modeling set was 28. 91%, and that of Hainan data validation set was 26. 58%. 3)The CV of EVI2 linear fitting model in Ezhou data modeling set was 33. 78%, and that in Hainan data validation set was 27. 90%. Finally, the empirical model constructed by EVI2 was used to predict LAI of two different rice varieties(Luoyou 9348 and Fengliangyou 4)at different growth stages. The results showed that Luoyou 9248 was a nitrogen efficient variety under the same nitrogen level. |
| |
| Keywords: | rice leaf area index vegetation index nitrogen efficient |
|
| 点击此处可从《中国农业大学学报》浏览原始摘要信息 |
| 点击此处可从《中国农业大学学报》下载免费的PDF全文 |
|
