摘要
针对生鲜牛肉新鲜度的传统评价中,存在的参数单一、检测精度不足等问题。采用高光谱技术搭建无损检测系统,通过采集牛肉样本的光谱数据,参照国标方法测定其主要品质参数:pH值,总挥发性盐基氮(TVB-N),L^(*),a^(*),b^(*)值,并根据pH值和TVB-N将牛肉分为新鲜、次新鲜和腐败3个等级。采用不同的预处理方式建立各参数的偏最小二乘(PLSR)模型,确定SG+MSC为最优预处理方法。使用CARS和SPA特征波长筛选算法,分别建立PLSR、多元线性回归和人工神经网络预测模型。结果表明,SG+MSC-CARS-PLSR模型对牛肉品质参数的预测效果最佳,其中pH,TVB-N,L^(*),a^(*),b^(*)测试集的决定系数分别为0.9539,0.9660,0.9266,0.8683,0.9018,对应的测试集均方根误差分别为0.0857,1.3955,0.3756,0.3208,0.4555,试验验证分级准确率达到92.6%。研究为牛肉新鲜度品质评估提供有效的技术手段。
To address the limitations of traditional methods for assessing fresh beef freshness such as reliance on single parameters and insufficient detection accuracy,a non-destructive detection system was developed using hyperspectral imaging technology.Spectral data from beef samples were collected and key quality parameters including pH total volatile basic nitrogen TVB-N and color values L^(*),a^(*)and b^(*)were measured according to national standard methods.Based on pH and TVB-N values,beef samples were categorized into three freshness levels:fresh moderately fresh and spoiled.Partial least squares regression PLSR models were established for each parameter using different preprocessing methods identifying SG combined with MSC as the optimal preprocessing approach.Feature wavelength selection algorithms CARS and SPA were applied to develop PLSR multiple linear regression and artificial neural network prediction models.Results demonstrated that the SG+MSC-CARS-PLSR model achieved the best prediction performance for beef quality parameters with test set determination coefficients of 0.9539 for pH,0.9660 for TVB-N,0.9266 for L^(*),0.8683 for a^(*)and 0.9018 for b^(*)and corresponding root mean square errors RMSE of 0.0857,1.3955,0.3756,0.3208,and 0.4555,respectively.Experimental validation showed a classification accuracy of 92.6%.This study provides an effective technical approach for evaluating beef freshness quality.
作者
肖纵纵
蔡浩天
王炼
柳军
XIAO Zongzong;CAI Haotian;WANG Lian;LIU Jun(College of Artificial Intelligence,Nanjing Agricultural University,Nanjing 210095,China;Institute of Agricultural Facilities and Equipment,Jiangsu Academy of Agricultural Sciences,Nanjing 210014,China)
出处
《包装与食品机械》
北大核心
2025年第5期25-33,共9页
Packaging and Food Machinery
基金
国家重点研发计划项目(2022YFD2100500)。
关键词
牛肉
机器学习
高光谱成像技术
特征波长提取
无损检测
beef
machine learning
hyperspectral imaging technology
feature wavelength extraction
non-destructive detection
