融合不同成熟度的苹果可溶性固形物预测模型研究

Soluble solid predictive models of apples with different maturity levels

中文关键词:  苹果  成熟度  可溶性固形物  特征变量提取  近红外光谱检测

英文关键词:apple  maturity  soluble solid  characteristic variables extraction  near-infrared spectroscopy

中文摘要:

      目的 为实现对不同成熟度的苹果可溶性固形物的预测, 建立普适性强的混合分析模型。方法 选取甘肃静宁241个不同成熟度的苹果作为研究对象, 利用近红外光谱采集系统获取苹果漫反射光谱信息, 并对苹果可溶性固形物含量进行测定。利用S-G卷积平滑、多元散射校正(multiplicative scatter correction, MSC)、以及标准正态变量变换(standard normal variable transformation, SNV)等预处理方法结合竞争自适应加权算法(competitive adaptive reweighted sampling, CARS)、随机蛙跳(random frog, RF)算法提取苹果可溶性固形物的特征变量, 然后利用偏最小二乘回归(partial least squares regression, PLSR)和支持向量机(support vector machine, SVM)算法建立分析模型。结果 对比发现, 采用RF选取的特征波长变量数更少且预测精度优于CARS, 原始波长点由1251个减少到55个, MSC-RF-PLSR建立的模型预测结果最好, 其预测相关系数r和预测均方根误差分别为0.906和0.744。结论 采用近红外光谱方法构建的苹果可溶性固形物混合分析模型可以实现对苹果不同成熟度的预测, 为建立适用于不同成熟度苹果的可溶性固形物便携设备提供理论依据。

英文摘要:

      Objective To establish a model for achieving the prediction of soluble solid of apples with different maturity levels by a universally applicable mixed analysis. Methods In this study, 241 apples with different maturity levels in Jingning, Gansu were selected as the research object. Near-infrared spectroscopy was used to acquire diffuse reflectance spectra of apples and apple soluble solid content was measured. Pretreatment methods such as SG-Smooth, multiple scatter correction (MSC), and standard normal variable transformation (SNV) were combined with competitive adaptive weighting algorithm (CARS) and random frog leaping (RF) algorithm to extract characteristic variables of apple soluble solid, then the partial least squares regression (PLSR) and support vector machine (SVM) algorithms were used to establish the analytical model. Results It could be found that the number of feature wavelength variables selected by RF was fewer and the prediction accuracy was better than that of CARS. The number of original wavelength points was reduced from 1251 to 55. The model established by MSC-RF-PLSR had the most predictive results, its prediction correlation r and prediction root mean square error were 0.906 and 0.744. Conclusion The results show that the apple soluble solid hybrid analysis model constructed by near-infrared spectroscopy can realize the prediction of apple with different maturity levels and provide the theoretical basis for the establishment of soluble solid portable equipment suitable for apples of different maturity levels.