摘要
针对传统奶粉品质检测方法的主观性和滞后性问题,该研究利用奶粉的加工条件和颗粒形态构建了一个基于图像分析技术和人工智能算法的奶粉品质软测量模型,用于准确、实时地预测速溶全脂奶粉的分散性和溶解性这2个重要品质性能。利用显微数码摄像头和图像处理技术,获取了奶粉颗粒的形态参数。同时,利用重采样技术解决了奶粉工厂原始数据集中的数据不平衡问题。根据实验获取的奶粉颗粒形态参数和奶粉工厂提供的奶粉生产时的加工条件数据,选用偏最小二乘模型和人工神经网络模型构建了速溶全脂奶粉分散性和溶解性的软测量模型,并利用原始数据验证了所构建模型的准确性。结果表明,用于预测奶粉分散性和溶解性所构建的偏最小二乘模型的Q^(2)和R^(2)分别为Q^(2)=0.72,R^(2)=0.94和Q^(2)=0.85,R^(2)=0.95,而用于预测奶粉分散性和溶解性所构建的人工神经网络模型的R^(2)分别为0.97和0.96。模型的良好性能证明,这些模型可以准确、实时地预测奶粉的分散性和溶解性,并为奶粉品质的在线检测提供了新的方法。
To address the subjectivity and latency issues inherent in traditional methods of powdered milk quality assessment,this study established a soft sensor model for the quality of powdered milk.This model,constructed based on image analysis technology and artificial intelligence algorithms,utilized the processing conditions and particle morphology of powdered milk.It enabled the accurate and real-time prediction of two crucial quality attributes,including the dispersibility and solubility of instant whole milk powder.With the aid of digital microscopes and image processing techniques,morphological parameters of milk powder particles were obtained.Concurrently,the issue of data imbalance within the original dataset from milk powder factories was rectified by employing resampling techniques.Using the morphological parameters obtained from the experiments and the processing condition data provided by the milk powder plant,this study constructed soft sensor models for dispersibility and solubility based on partial least squares and artificial neural network models.The constructed models were then validated for their accuracy using the original data.Results showed that the partial least squares model,constructed for predicting dispersibility and solubility,had Q^(2)and R^(2)values of 0.72 and 0.94,and 0.85 and 0.95,respectively.Additionally,the artificial neural network model designed for the same purpose yielded R^(2)values of 0.97 and 0.96.The outstanding performance of these models proves their ability to predict the dispersibility and solubility of milk powder accurately and in real time and introduces a new approach for the online quality assessment of instant whole milk powder.
作者
丁浩晗
谢祯奇
田嘉伟
辛星
王震宇
DING Haohan;XIE Zhenqi;TIAN Jiawei;XIN Xing;WANG Zhenyu(Science Center for Future Foods,Jiangnan University,Wuxi 214000,China;School of Artificial Intelligence and Computer Science,Jiangnan University,Wuxi 214000,China;Jiaxing Institute of Future Food,Jiaxing 314005,China)
出处
《食品与发酵工业》
CAS
CSCD
北大核心
2024年第10期273-281,共9页
Food and Fermentation Industries
基金
国家重点研发计划“食品营养与安全关键技术研发”重点专项(2022YFF1101100)
中央高校基本科研业务费专项资金资助(JUSRP123053)
跨境网络空间安全教育部工程研究中心2023年度开放课题“东南亚重点人物画像数据库构建及其在跨境食品供应链领域的应用研究”(KJAQ202304007)。
关键词
速溶全脂奶粉
图像处理
品质检测
偏最小二乘法
人工神经网络
instant whole milk powder
image processing
quality testing
partial least squares
artificial neural network
