To accurately and precisely analyze the growth and health status of kiwifruit trees, a leaf hyperspectral dataset was constructed using kiwifruit trees in the Guanzhong Plain of Shaanxi Province. The dataset was divided based on the stochastic method and the Kennard-Stone method, and the characteristic bands of the samples were extracted using the competitive adaptive reweighted sampling (CARS), principal component analysis (PCA), and iteratively retains informative variables (IRIV) algorithms. Multiple linear regression (MLR), ridge regression (RR), partial least squares regression (PLSR), support vector regression (SVR) and random forest regression (RFR) were used to establish an intelligent detection model for leaf chlorophyll content. Comparative analysis of the models showed that the CARS-RR model based on the 81 feature bands extracted by the CARS algorithm had the best prediction effect, with an R² of 0.86 and an RMSE of 2.71 on the validation set. Therefore, the proposed intelligent detection model can detect the chlorophyll content of kiwifruit trees based on the spectral information in a nondestructive manner. Furthermore, it can analyze the overall health status of the orchard, providing decision-making support for subsequent refined orchard management.

为准确实时分析猕猴桃树的生长健康状况，以陕西关中平原猕猴桃树为研究对象，构建叶片高光谱数据集；基于随机法和Kennard-Stone方法划分数据集，采用竞争自适应重加权采样算法（CARS）、主成分分析法（PCA）和迭代保留信息变量算法（IRIV）提取样本的特征波段；进而采用多元线性回归（MLR）、岭回归（RR）、偏最小二乘回归（PLSR）、支持向量回归（SVR）和随机森林回归（RFR）等方法建立叶片叶绿素含量智能检测模型。模型对比分析表明，基于CARS算法所提取的81个特征波段建立的CARS-RR模型预测效果最好，验证集上的R²为0.86，RMSE为2.71。因此，提出的智能检测模型能够基于光谱信息无损检测猕猴桃树叶绿素含量，进而分析果园整体健康状况，为后续果园精细化管理提供决策支撑。