基于回归卷积神经网络的近红外光谱苹果脆片品质评价方法研究  

Evaluation Method of Apple Chips Quality by Near Infrared Spectroscopy Based on Regressive Convolutional Neural Network

作　　者：方明明 刘静[1] FANG Mingming;LIU Jing(School of Information Engineering,Shanghai Maritime University,Shanghai 201306,China)

机构地区：[1]上海海事大学信息工程学院,上海201306

出　　处：《食品科技》2020年第7期303-308,316,共7页Food Science and Technology

基　　金：国家自然科学基金项目(31701515)。

摘　　要：近红外光谱分析作为一种便捷且成功的非破坏性的光谱方法,在苹果脆片品质检测上具备独特的优势。针对近红外光谱样品数量少、非线性、高维等数据特点,文章提出了一种基于回归卷积神经网络的近红外光谱苹果脆片品质评价方法。为了验证算法的有效性,经多次验证实验,得出苹果脆片水分、总糖、总酸的最佳模型的RMSE_P分别为0.0916、0.0623、0.1338,相关系数R_P分别为0.9459、0.9251、0.9116,并与传统的建模方法诸如PLS、BP、LSSVM进行比较。实验结果表明,卷积神经网络用于苹果脆片品质近红外光谱分析具有更好的稳定性和泛化能力。将深度学习方法引入到苹果脆片品质近红外光谱分析领域中,为近红外光谱分析提供了一种有效的新思路,该方法同样可以推广到其他领域的近红外光谱检测中。As a convenient and successful non-destructive spectroscopy method,near-infrared spectroscopy has unique advantages in apple chips quality detection.Aiming at the characteristics of a small number of samples,non-linearity,and high-dimensional data in the near-infrared spectrum,this paper proposes a method for evaluating apple chips quality in near-infrared spectroscopy based on regression convolutional neural network.In order to verify the effectiveness of the algorithm,the paper has repeatedly verified the experiments and found that the RMSEP of the best models of apple chips moisture,total sugar,and total acid are 0.0916,0.0623,and 0.1338,and the correlation coefficients R are 0.9459,0.9251,and 0.9116,respectively.Compared with traditional modeling methods such as PLS,BP,and LSSVM,the experimental results show that the convolutional neural network has better stability and generalization ability for apple chips quality near-infrared spectral analysis.The deep learning method is introduced into the field of apple chips quality near-infrared spectroscopy,which provides an effective new idea for near-infrared spectroscopy analysis.This method can also be extended to near-infrared spectroscopy detection in other fields.

关 键 词：卷积神经网络 近红外光谱 回归 苹果脆片 

分 类 号：TS255.36[轻工技术与工程—农产品加工及贮藏工程]