欢迎访问《中国农学通报》,

中国农学通报 ›› 2024, Vol. 40 ›› Issue (31): 133-138.doi: 10.11924/j.issn.1000-6850.casb2024-0452

• 食品·营养·检测·安全 • 上一篇    下一篇

基于近红外光谱数据的李子可溶性固形物含量预测模型研究

狄标1(), 林娟1, 刘现2()   

  1. 1 福建农林大学计算机与信息学院,福州 350003
    2 福建省农业科学院数字农业研究所,福州 350003
  • 收稿日期:2024-07-10 修回日期:2024-09-05 出版日期:2024-11-05 发布日期:2024-11-04
  • 通讯作者:
    刘现,女,1985年出生,福建福州人,助理研究员,硕士,研究方向:数字农业。通信地址:350003 福建省福州市鼓楼区华林路188号 福建省农业科学院数字农业研究所,Tel:0591-87869364,E-mail:
  • 作者简介:

    狄标,男,2000年出生,湖南汨罗人,硕士,研究方向:机器学习和智能计算。通信地址:350003 福建省福州市仓山区上下店路15号 福建农林大学智慧农林福建省高校重点实验室,Tel:0591-83789216,E-mail:

  • 基金资助:
    福建省科技厅自然科学基金资助项目“基于双分支长短期记忆神经网络的福建省碳排放峰值预测研究”(2022J01153); 2023年莆田市科技项目“基于深度学习及光谱成像技术的李子无损检测研究”(2023GM03)

Research on Prediction Model of Soluble Solid Content in Plums Based on Near-Infrared Spectroscopy Data

DI Biao1(), LIN Juan1, LIU Xian2()   

  1. 1 School of Computer Science and Technology, Fujian Agriculture and Forest University, Fuzhou 350003
    2 Digital Agriculture Research Institute, Fujian Academy of Agricultural Sciences, Fuzhou 350003
  • Received:2024-07-10 Revised:2024-09-05 Published:2024-11-05 Online:2024-11-04

摘要:

针对李子可溶性固体物含量预测方法存在前期数据预处理较复杂、预测精度不高、预测模型泛化能力不强的问题,提出针对李子品种的一维卷积神经网络(1D Convolutional Neural Network,1D-CNN)模型。首先针对300个三华李和三月李样本果品构建可溶性固体物含量的近红外光谱数据,并分别设计输入层、一维卷积层、池化层与全连接层和输出层等多层结构构建1D-CNN模型。模型决定系数为0.980、均方根误差为0.192,表现均优于支持向量回归、随机森林等传统机器学习方法,并且作为轻量级模型,具有建模过程简便、泛化能力强的特点,可满足实际场景需求。

关键词: 近红外光谱, 可溶性固形物, 一维卷积神经网络, 李子, 深度学习

Abstract:

To address the issues of complex data preprocessing, low prediction accuracy, and poor generalization capability of existing methods for predicting the soluble solid content (SSC) of plums, a one-dimensional convolutional neural network (1D-CNN) model specifically for plum varieties is proposed. Near-infrared (NIR) spectral data were constructed for 300 samples of Sanhua and Sanyue plums to predict their SSC. And the multi-layer structures such as input layer, one-dimensional convolution layer, pooling layer and full connection layer and output layer were designed to construct 1D-CNN model. Comparative experiments demonstrated that the model achieved a coefficient of determination (R2) of 0.980 and a root mean square error (RMSE) of 0.192, outperforming traditional machine learning methods such as support vector regression and random forest. As a lightweight model, it offers simplicity in the modeling process and strong generalization capability, making it suitable for practical applications.

Key words: near-infrared spectroscopy, soluble solids content, one-dimensional convolutional neural network, plum, deep learning