Machine Learning-Based Inversion Modeling of Fertilizer Application Rate Using Tomato Leaf Color
摘要
为实现设施番茄施肥量的快速无损诊断,探索叶片颜色特征与施肥量的定量关系,以番茄品种"美人尖"为对象,设置T1(67.5 kg/hm²)、T2(90 kg/hm²)、T3(112.5 kg/hm²)和T4(135 kg/hm²)4个施肥梯度,利用X-Rite Color Checker Classic 24色卡校正采集叶片数字图像,提取并构建13种颜色特征参数(R、G、B、R+G-B、G-R、G+R、(G-R)/(G+R)、R²、G²、B²、RG、RB、GB),基于随机森林算法建立施肥量预测模型.结果表明:随施肥量增加,叶片R、B通道值呈总体下降趋势,G通道呈先增强后减弱的单峰变化,复合颜色特征(RB、RG、G²等)对施肥量的解释力显著优于单一R、G、B通道;随机森林模型能有效捕捉颜色特征与施肥量的非线性关系,预测精度(R²=0.520,RMSE=17.1 kg/hm²)明显优于多元线性回归(R²=0.495,RMSE=18.621 4 kg/hm²).研究证实了基于RGB图像的番茄叶片颜色特征可用于施肥量快速、无损诊断,为设施番茄精准施肥管理提供了技术依据.
To achieve rapid,non-destructive diagnosis of fertilizer application rates in greenhouse tomatoes and to explore the quantitative relationship between leaf color characteristics and fertilizer rates,this study utilized the tomato cultivar Meirenjian as the research subject.Four fertilizer treatments were established:T1(67.5 kg/hm²),T2(90 kg/hm²),T3(112.5 kg/hm²),and T4(135 kg/hm²).Digital images of tomato leaves were acquired using an X-Rite Color Checker Classic 24-color chart for color calibration.Thirteen color characteristic parameters(R,G,B,R+G-B,G-R,G+R,(G-R)/(G+R),R²,G²,B²,RG,RB,GB)were extracted and constructed,and a fertilizer rate prediction model was established based on the random forest algorithm.Results indicated that as fertilizer application rates increased,R and B channel values exhibited an overall decreasing trend,while G channel values displayed a unimodal pattern of initial enhancement followed by reduction.Composite color characteristics(such as RB,RG and G²)demonstrated significantly greater explanatory power for fertilizer rates compared to individual R、G、B channels.The random forest model effectively captured the nonlinear relationships between color characteristics and fertilizer rates,achieving prediction accuracy(R²=0.520,RMSE=17.1 kg/hm²)significantly superior to multiple linear regression(R²=0.495,RMSE=18.621 4 kg/hm²).This study confirms the feasibility of utilizing tomato leaf color characteristics derived from RGB images for rapid,non-destructive diagnosis of fertilizer application rates,providing a technical foundation for precision fertilization management in greenhouse tomato cultivation.