农业机械学报2025，Vol.56Issue(9)：536-546,11.DOI:10.6041/j.issn.1000-1298.2025.09.044
基于机器学习的玉米SPAD值影像分期反演与水氮优化利用研究
Staging Inversion of Maize SPAD Values and Optimal Utilization of Water and Nitrogen Based on Machine Learning
摘要
Abstract
By using unmanned aerial vehicle remote sensing technology combined with multiple machine learning algorithms,the maize SPAD values with different water and nitrogen treatments in the Yellow River irrigation area can be inverted,which can provide precise irrigation and fertilization diagnosis for maize planting.Three irrigation quotas(W1:150 m3/hm2,W2:225 m3/hm2,W3:300 m3/hm2)and three nitrogen application rates(N1:140 kg/hm2,N2:210 kg/hm2,N3:280 kg/hm2)were set,and water nitrogen treatment experiments were carried out.Farmland image data were obtained by using the multispectral camera of unmanned aerial vehicles(UAVs),and 16 spectral variables were obtained.Spectral variables with high correlations with SPAD values were selected at each growth stage of maize to construct machine learning algorithms such as ridge regression,support vector machine,random forest,and partial least square method.The models were compared and screened,and the screened models were used to predict the SPAD values of different treatments.The predicted SPAD values were used to fit the equation with the irrigation and fertilization amounts to diagnose the expected optimal water and fertilizer dosages at each growth stage.The research results showed that the maize SPAD value reached its peak during the filling period.Among them,W2N2 was 9.14％higher than W1N2,17.22％higher than W3N2,10.9％higher than W2N1,and 35.4％higher than W2N3.Combining the measured SPAD values in three important periods of maize,by comparing the determination coefficient R2 of the modeling set and the validation set of different machine learning models,it was found that the random forest model showed the best predictive performance,and the spatial distribution map of chlorophyll content in farmland was expanded by using the predicted values of the RF model.The regression equations between the average SPAD value of nine treatments prediction points and the corresponding irrigation and fertilizer application amounts were established.Through optimization,it was obtained that at the jointing stage of maize,the optimal irrigation amount was 547 m3/hm2 and the optimal fertilizer application amount was 42.98 kg/hm2.During the large trumpet stage of maize,the optimal irrigation amount was 450 m3/hm2,and the optimal fertilizer application amount was 63 kg/hm2.During the tasseling stage,the optimal irrigation amount was 498 m3/hm2,and the optimal fertilization amount was 73.29 kg/hm2.During the grouting period,the optimal irrigation water amount was 493.5 m3/hm2,and the optimal fertilizer application amount was 43.4 kg/hm2.Based on the combined method of unmanned aerial vehicle remote sensing technology and machine learning algorithms,the maize SPAD value was successfully inverted,and the irrigation and fertilization situation in farmland was revealed,providing a feasible solution for precise irrigation and fertilization in farmland.关键词
玉米SPAD值/无人机遥感/机器学习/灌水施肥管理Key words
maize SPAD value/UAV remote sensing/machine learning/irrigation and fertilization management分类
农业科技引用本文复制引用
苗世龙,李仙岳,史海滨,闫建文,丁世杰,苗平..基于机器学习的玉米SPAD值影像分期反演与水氮优化利用研究[J].农业机械学报,2025,56(9):536-546,11.基金项目
国家自然科学基金项目(52369008)、内蒙古自治区水利科技项目(NSK202201)、内蒙古自治区科技计划项目(2022YFHH0039)和一流学科科研专项(YLXKZX-NND-022) （52369008）
