基于Stacking框架的蓝莓硬度预测可溶性固形物含量及维生素C模型构建

Construction of blueberry hardness prediction soluble solid content and vitamin C model based on Stacking framework

中文关键词:  Stacking框架  添加多项式特征  蓝莓  理化指标

英文关键词:Stacking framework  adding polynomial features  blueberry  physical and chemical indicators

中文摘要:

      目的 探索一种基于蓝莓的质量和硬度快速、无损预测其可溶性固形物含量(soluble solid content, SSC)与维生素C (vitamin C, VC)含量的方法, 为蓝莓化学成分的预测提供一种新思路。方法 通过对蓝莓质量、硬度与SSC、VC相关性分析后, 建立基于一维特征的质量和硬度预测SSC与VC模型。其次, 对硬度添加多项式特征做升维处理, 同一维进行相同研究。最后, 对比Stacking框架与单一模型、及添加多项式特征的预测效果。结果 一维特征条件下, 基于Stacking框架的硬度预测SSC与VC的决定系数(coefficient of determination, R2)分别为0.873、0.875, 预测效果优于质量与单模型预测; 多维特征条件下, 硬度添加到3次方时预测SSC效果最佳, R2为0.889; 添加到12次方时预测VC含量效果最佳, R2为0.890, 预测效果均好于一维特征。结论 Stacking框架结合添加多项式特征在蓝莓硬度快速、无损预测其SSC及VC方面具有良好的潜力, 为蓝莓品质检测提供新途径。

英文摘要:

      Objective To explore a fast and nondestructive method to predict the soluble solid content (SSC) and vitamin C (VC) content of blueberry based on its mass and hardness, and provide a new idea for the prediction of chemical composition of blueberry. Methods Through the correlation analysis of blueberry mass, hardness with the SSC and VC, a prediction model of the SSC and VC based on one-dimensional features of mass and hardness was established. Secondly, polynomial features were added to the hardness to do the dimensional lift and the same study was carried out in the same dimension. Finally, the prediction effects of Stacking framework with a single model and the addition of polynomial features were compared. Results Under one-dimensional feature condition, coefficient of determination (R2) of the hardness prediction of SSC and VC based on Stacking framework was 0.873 and 0.875, respectively, which was better than the prediction of mass and single model; under the condition of multi-dimensional features, when hardness was added to the 3rd power, it was the best to predict the SSC, R2 was 0.889; when hardness was added to the 12th power, it was the best to predict the content of VC, R2 was 0.890, which was better than one-dimensional features. Conclusion The Stacking framework combined with the addition of polynomial features has good potential for rapid and nondestructive prediction SSC and VC based on blueberry hardness, which provides a new way for blueberry quality detection.