Journal of Nuclear Agricultural Sciences

Figure/Table detail

Winter Wheat Yield Prediction Using UAV-Based Multispectral Time-Series Imagery

WAN Guoliang, WANG Yinting, LI Lei, WANG Duoxia, TAO Zhiqiang, SUN Haiyan

Journal of Nuclear Agricultural Sciences, 2026, 40(2): 413-422. DOI: 10.11869/j.issn.1000-8551.2026.02.0413

采集次序 Sequence	生长阶段 Growing stages	采集日期 Collection date
第1次1st time	拔节期	2023年3月30日
第2次2nd time	挑旗期	2023年4月16日
第3次3rd time	开花期	2023年4月30日
第4次4th time	灌浆前期	2023年5月13日
第5次5th time	灌浆中期	2023年5月22日
第6次6th time	灌浆后期	2023年5月31日
第7次7th time	蜡熟期	2023年6月5日

Table 1 Drone field data collection schedule

Other figure/table from this article

Fig.1 Test location map
Table 2 Multispectral vegetation indices used in the study
Fig.2 Flowchart of establishing yield estimation model with multispectral data
Fig.3 Correlation between multi-spectral bands and vegetation index and measured yield in the plot |r|
Note： Ⅰ： Elongation stage. Ⅱ： Flag leaf stage. Ⅲ： Flowering stage. Ⅳ： Early grain-filling stage. Ⅴ： Middle grain-filling stage. Ⅵ： Late grain-filling stage. Ⅶ： Wax-ripening stage.
Fig.4 Coefficient of determination （R²） of crop yield prediction at different growth stages using various machine learning models
Fig.5 Root RMSE and MAE of crop yield prediction at different growth stages using various machine learning models
Fig.6 Violin diagram of yield prediction values cross-validated by different machine learning models
Note： n.s.： Significance levels are P>0.05.
Fig.7 Scatter plots of predicted vs. observed yield values for each model
Note： A， B， C， D， E， F， and G represent the LSTM， RNN， RF， SVR， KNN， XGBoost， and RR models， respectively. The gray dashed line represents the 1∶1 line between the actual production and the model prediction， the red line is the fitted line for the observed and predicted values， RMSE and MAE units are t·hm^-2.
Fig.8 Scatter plots of predicted vs. observed yield values based on stratified sampling method
Note： A， B， C， D， E， F， and G represent the LSTM， RNN， RF， SVR， KNN， XGBoost， and RR models， respectively. The gray dashed line represents the 1∶1 line between the actual production and the model prediction， the red line is the fitted line for the observed and predicted values， RMSE and MAE units are t·hm^-2.
Fig.9 Comparison of prediction accuracy of machine learning models using random sampling and stratified sampling methods

模态框（Modal）标题

Figure/Table detail