| 引用本文: | 李宗鹏,李连豪,陈 震,等.基于Stacking法的无人机光谱遥测冬小麦产量[J].灌溉排水学报,2021,(8):50-56. |
| LI Zongpeng,LI Lianhao,CHEN Zhen,et al.基于Stacking法的无人机光谱遥测冬小麦产量[J].灌溉排水学报,2021,(8):50-56. |
|
| 摘要: |
| 【目的】精确、高效地预测作物产量。【方法】以冬小麦为研究对象，利用无人机搭载多光谱相机，获取抽穗期、开花期和灌浆期的多光谱图像数据。根据多光谱波段选取对产量敏感的14种植被指数，并优选出与产量极显著相关的13种植被指数；基于优选出的植被指数分别建立各生育期的MLR、PLSR、SVM和Cubist产量估算初级模型进行对比分析，并利用Stacking方法集成初级学习器模型分别建立各个时期MLR和Cubist次级产量估测模型。【结果】随着冬小麦生长阶段的发展，各植被指数与产量的相关性逐渐增大，在灌浆期达到最大值0.67；对比4个初级学习器模型精度，Cubist模型在抽穗期、开花期和灌浆期的估产精度均为最高；利用Stacking方法构建的次级学习器模型以Cubist模型的估产效果最佳，MLR和Cubist模型的估产精度在各个时期均得到了提升。【结论】基于Stacking方法融合估产模型能够显著提升冬小麦的产量估算精度，为今后的估产研究提供参考。 |
| 关键词: 多光谱；植被指数；Stacking；模型 |
| DOI：10.13522/j.cnki.ggps.2021073 |
| 分类号: |
| 基金项目: |
|
| Estimating Winter Wheat Yield Using UAV Remote Sensing Imageries and Stacking Method[ |
|
LI Zongpeng, LI Lianhao CHEN Zhen, CHENG Qian, XU Honggang, PANG Chaofan
|
|
1. Henan Agricultural University, Zhengzhou 450000, China;
2. Institute of Farmland Irrigation, Chinese Academy of Agricultural Sciences, Xinxiang 453002, China
|
| Abstract: |
| 【Background and objective】Predicting potential crop yield based on crop traits at different growth stages is required in crop breeding and food safety management, but is challenging because of the spatial heterogeneity of crop traits and yield. The traditional yield prediction method is point-based, which is tedious and laborious, and the rapid development in unmanned aerial vehicle (UAV) and remote sensing technologies has started to break this barrier. The aim of this paper is to investigate the applicability of UAV and remote sensing, as well as their accuracy in predicting potential crop yield.【Method】The experiment was conducted in a winter wheat field. Remote sensing imageries of the crop at heading, flowering and filling stages were taken using a multispectral camera mounted in a drone. Using the multispectral bands derived from the imageries, 14 cropping indexes postulated to affect wheat yield were calculated. Based on the optimized vegetation indexes, different primary models including MLR, PLSR, SVM and Cubist were established for each growth stage to predict the eventual yield. We compared and analyzed all models and reconstructed the MLR and Cubist models for each growth stage using the Stacking method.【Result】As the wheat grew, the correlation between the vegetation index and the wheat yield increased, peaking at the filling stage with a correlation coefficient of 0.67. Comparison of the four primary models revealed that the Cubist model using data at heading, flowering and filling stages was most accurate to predict the potential wheat yield. Reconstructing the primary models using the Stacking method improved the accuracy of all models, with the Cubist model being the most accurate.【Conclusion】This study proves that fusing the primary models using the Stacking method can significantly improve their accuracy for predicting wheat yield. The methods and results in this paper have implications for predicting yield of other crops. |
| Key words: multispectral imageries; vegetation index; stacking; method |
