| 龚智,刘江凡,郭新华,邵精翠,李朝阳.机器学习算法在苹果叶绿素估算中的应用[J].干旱地区农业研究,2026,(3):275~284 |
| 机器学习算法在苹果叶绿素估算中的应用 |
| Application of machine learning algorithms in estimating chlorophyll in apples |
| |
| DOI：10.7606/j.issn.1000-7601.2026.03.26 |
| 中文关键词: 苹果 叶绿素 多光谱 机器学习 |
| 英文关键词:apple chlorophyll multispectral machine learning |
| 基金项目:新疆生产建设兵团财政科技计划项目（2022BC009） |
|
| 摘要点击次数: 266 |
| 全文下载次数: 12 |
| 中文摘要: |
| 为探讨机器学习算法在南疆地区苹果冠层叶绿素含量估算中的适用性并构建覆盖苹果全生育期的叶绿素最优估算模型，以成龄苹果树为对象，采集各生育期无人机多光谱影像及冠层叶绿素含量数据并计算多种植被指数，进而结合实测叶绿素含量构建估算模型。系统对比偏最小二乘回归（PLSR）、随机森林（RF）、支持向量回归（SVR）、极端梯度提升（XGBoost）、套索回归（LASSO）、岭回归（RR）、神经网络（NN）及逐步回归（SR）的估算精度，筛选各生育期最优模型及覆盖全生育期的最佳模型。结果表明：（1）植被指数与叶绿素含量相关性良好，其中归一化植被指数（NDVI）、优化土壤调节植被指数（OSAVI）、比值植被指数（RVI）和土壤调节植被指数（SAVI）与叶绿素含量的相关系数均大于0.6；（2）幼果期以RF模型精度最高（R2=0.626，RMSE=0.129）；果实膨大期以XGBoost模型表现最优（R2=0.720，RMSE=0.144）；果实着色期以RF模型精度最高（R2=0.667，RPD=1.733）；（3）果实膨大期RF与XGBoost模型精度相当，但RF模型在全生育期均表现稳定且精度较高，被确定为覆盖苹果全生育期的最优叶绿素估算模型。 |
| 英文摘要: |
| To explore the applicability of machine learning algorithms in estimating apple canopy chlorophyll content in the Southern Xinjiang region and construct an optimal chlorophyll estimation model covering the entire growth period of apples, this study used mature apple trees as subjects, collected UAV multispectral imagery and canopy chlorophyll content data at various growth stages, calculated multiple vegetation indices, and constructed estimation models in combination with measured chlorophyll content. The study systematically compared the estimation accuracy of Partial Least Squares Regression (PLSR), Random Forest (RF), Support Vector Regression (SVR), Extreme Gradient Boosting (XGBoost), Least Absolute Shrinkage and Selection Operator regression (LASSO), Ridge Regression (RR), Neural Network (NN), and Stepwise Regression (SR) to screen the optimal model for each growth stage and the best model covering the entire growth period. The results showed that: (1) Vegetation indices exhibited good correlation with chlorophyll content, with correlation coefficients between NDVI, OSAVI, RVI, SAVI, and chlorophyll content all exceeding 0.6. (2) During the young fruit stage, the RF model achieved the highest accuracy (R2=0.626, RMSE=0.129); during the fruit expansion stage, the XGBoost model performed best (R2=0.720, RMSE=0.144); during the fruit coloring stage, the RF model achieved the highest accuracy (R2=0.667, RPD=1.733). (3) Although the RF and XGBoost models showed comparable accuracy during the fruit expansion stage, the RF model demonstrated stable and high accuracy throughout the entire growth period and was therefore identified as the optimal chlorophyll estimation model covering the entire growth period of apples. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
| | |