特征波长筛选结合PCA-LSSVM 对甜瓜叶片SPAD值的预测

【目的】利用光谱技术对定量估测大田甜瓜冠层叶片叶绿素含量，为田间的水肥调控以及田间管理提供理论依据。【方法】采用一阶求导对400～1 100 nm的叶绿素可见近红外反射光谱数据进行预处理，对于冗余的光谱数据，先分别使用特征筛选中的竞争性自适应重加权采样法(CARS)、遗传算法(GA)、蒙特卡罗无信息变量消除法(MC-UVE),再分别与主成分分析(PCA)特征提取算法融合；分别建立极限学习机(ELM)、支持向量机(SVM)、最小二乘支持向量机(LS-SVM)对甜瓜叶片SPAD定量预测模型。【结果】单一的特征筛选下，最优预测模型为CARS+SVM,校正集相关系数为0.903 5,预测集相关系数为0.893 1;特征筛选和特征提取融合下，最优的预测模型为GA+PCA+LSSVM,校正集相关系数0.955 8,预测集相关系数为0.939 7。【结论】优化后的模型可用于定量分析的使用，精准测定甜瓜叶片叶绿素含量。

Prediction of SPAD Value in Melon Leaves by Characteristic Wavelength Screening Combined with PCA-LSSVM

【Objective】 This project aims to use quantitative estimation of chlorophyll content in cantaloupe canopy leaves by spectral technique to provide theoretical basis for water and fertilizer control and field management. 【Methods】 The first derivative was used to preprocess the visible and near infrared reflectance spectra of chlorophyll in the range of 400 to 1,100 nm. Firstly, competitive adaptive weighted sampling (CARS), genetic Algorithm (GA) and Monte Carlo information-free variable elimination (MC-UVE) were used in feature selection, and then they were fused with Principal Component Analysis (PCA) at the same time. Considering that different models might produce different prediction results, the limit learning machine (ELM), the support vector machine and the least square support vector machine (LSSVM) were established to predict SPAD of muskmelon leaves quantitatively.【Results】 The results showed that the optimal prediction model was CARS+SVM, correlation coefficient of correction set was 0.903,5, correlation coefficient of prediction set was 0.893,1 under the single feature selection and fusion of feature selection and feature extraction. The optimal prediction model was GA+PCA+LSSVM, the correlation coefficient of calibration set was 0.955,8, and the correlation coefficient of prediction set was 0.939,7.【Conclusion】 The optimized model can be used for the quantitative analysis to achieve the accurate determination of chlorophyll content in muskmelon leaves.