欢迎访问《麦类作物学报》编辑部网站

作者	单位
姜平涛，刘佳，罗一航，袁梦，夏艺，欧京瑞，王钰泽，闫子怡，吴婷婷，刘守阳，蔚睿，吴建辉，韩德俊	（1.西北农林科技大学农学院，陕西杨凌 712100 2.西北农林科技大学作物抗逆与高效生产全国重点实验室，陕西杨凌 712100 3.西北农林科技大学机械与电子工程学院，陕西杨凌 712100 4.南京农业大学前沿交叉研究院，江苏南京 210095）

摘要点击次数: 47

全文下载次数: 28

中文摘要:

为了解基于无人机多光谱传感器实现小麦籽粒灌浆速率（grain filling rate, GFR）的高通量无损监测效果，以565份小麦种质为材料，于2022和2023年，通过无人机采集灌浆期时序光谱植被指数（SVIs），并同步实测粒重。利用Logistic函数拟合SVIs与粒重动态变化过程，筛选出与灌浆速率密切相关的红边叶绿素指数（CIRE），基于随机森林回归（random forest regressor, RFR）、支持向量机回归（support vector regression, SVR）、线性回归（linear regression, LR）构建预测模型，并通过决定系数（r²）、均方根差（RMSE）评价其性能。结果表明，经Logistic拟合，两个生长季的SVIs与实测粒重动态曲线的r²分别达到0.87和0.92。CIRE的动态变化速率与灌浆速率呈显著正相关（r=0.40~0.43, P<0.05）。在预测中，RFR模型表现最佳，2022和2023年全数据集的r²分别为0.88和0.89，RMSE分别为0.27和0.32，显著优于SVR和LR模型。在特定品种数据集中，RFR模型同样具有较高的预测精度和泛化能力。因此，基于无人机多光谱数据，通过CIRE与RFR结合可实现小麦灌浆速率高通量监测，为小麦种质资源快速评价及抗逆性育种提供技术支持。

英文摘要:

To understand the effect of high-throughput non-destructive monitoring of wheat grain filling rate(GFR) based on UAV multi-spectral, 565 wheat germplasm accessions were used as experimental materials over two growing seasons(2022 and 2023). UAVs were used to collect temporal spectral vegetation indices(SVIs) during the grain filling period, along with synchronized measurements of grain weight. The logistic function was applied to fit the dynamic changes between SVIs and grain weight, and the red-edge chlorophyll index(CIRE) was identified as the SVI most closely related to GFR. Machine learning models, including Random Forest Regressor(RFR), Support Vector Regression(SVR), and Linear Regression(LR), were constructed to predict GFR, and their performance was evaluated using r² and RMSE. The results showed that the r² values of the logistic fitting dynamic curves between SVIs and measured grain weight reached 0.87 and 0.92, respectively. The dynamic change rate of CIRE exhibited a significantly positive correlation with GFR(r=0.4-0.43, P<0.05). Among the models, RFR demonstrated the best predictive performance, with r² values of 0.88 and 0.89 and RMSE values of 0.27 and 0.32 for the full datasets in 2022 and 2023, respectively, significantly outperforming SVR and LR. Similarly, RFR achieved higher prediction accuracy and generalization capability in specific genotype datasets. Therefore, based on UAV multi-spectral data, the wheat grain filling rate was monitored by combining CIRE and RFR with high-precision and high-throughout. This method provides essential technical support for the evaluation of wheat germplasm resources and breeding for stress resistance.

查看全文查看/发表评论下载PDF阅读器

关闭