中国农业科学2025，Vol.58Issue(18)：3632-3647,16.DOI:10.3864/j.issn.0578-1752.2025.18.005
无人机多光谱参数与Shapley分析融合的青贮玉米生物量估算
Multispectral Unmanned Aerial Vehicle Parameters Combined with Machine Learning to Predict Silage Maize Biomass
摘要
Abstract
[Objective]Aboveground biomass is an important indicator of crop growth,in order to explore the accuracy difference between single spectral parameter model and fusion of different growth stage models in silage maize above ground biomass(AGB)estimation,this study aimed to compare the effects of unmanned aerial vehicle(UAV)multispectral feature parameters and model fusion methods on silage maize AGB estimation modeling accuracy,so as to improve the accuracy of silage maize biomass monitoring in Ningxia,and to provide a feasible technological solution for silage maize biomass dynamic monitoring.[Method]DJI UAV M300 RTK equipped with M600 Pro multispectral camera was used to acquire multispectral image data of silage maize at each growth stage under six different nitrogen levels,and the relationship between the spectral reflectance and vegetation index and the change of biomass of silage maize in the upper part of the ground under different treatments were analyzed.The data set of silage maize in the whole life cycle were classified into the nutrient growth stage data set and reproductive growth stage data set,and the correlation analysis of the two different growth stage data sets were carried out.The multispectral vegetation index with high degree of correlation was selected as the input of modeling data,and the AGB estimation model of silage maize at different growth stages were constructed by using machine learning methods,such as Random Forest(RF)and Convolutional Neural Network(BP).The model was optimized by using the Gray Wolf Optimization Algorithm,and finally optimizing the model by using the Shapley Analysis.The optimized AGB estimation model of different growth stages of silage maize was combined to obtain the AGB estimation model of silage maize with multi-spectral change characteristics for the whole reproductive period.[Result]The division of two different growth stages could improve the correlation between silage maize biomass and multispectral vegetation index,in which the green chlorophyll vegetation index(GCVI)improved with the highest value,and the absolute values of the correlation reached 0.61 and 0.64;the accuracy of the RF model after the combination using Shapley analysis was relatively high,with R2 of 0.89 and root mean square error(RMSE)of 1.31 kg·m-2;The RF model optimized by Gray Wolf algorithm with the Shapley combination had the highest accuracy with R2 of 0.92 and RMSE of 1.11 kg·m-2.[Conclusion]In this study,screening the optimal spectral parameters at each growth stage of silage maize and integrating multi-stage modeling using Shapley analysis could effectively improve the accuracy of silage maize AGB prediction model.关键词
青贮玉米/地上部生物量/无人机/多光谱/Shapley分析/机器学习Key words
silage maize/above ground biomass/UAV/multispectral/Shapley analysis/machine learning引用本文复制引用
韩林蒲,马纪龙,齐勇杰,高嘉琪,谢铁娜,贾彪..无人机多光谱参数与Shapley分析融合的青贮玉米生物量估算[J].中国农业科学,2025,58(18):3632-3647,16.基金项目
国家自然科学基金(32360432) （32360432）
