Predicting Nutritional Components of Chicken Sausage Using Principal Component Analysis and Partial Least Squares Regression
DING Shuxian, ZHANG Xinyu, CAI Mengran, ZHOU Hui, XU Baocai, WANG Zhaoming
1. School of Food and Biological Engineering, Hefei University of Technology, Hefei 230601, China; 2. Anhui Fuliji Roast Chicken Group Co. Ltd., Suzhou 232101, China
摘要采用近红外光谱技术结合化学计量学方法实现鸡肉火腿肠水分、脂肪及蛋白质含量预测。制备鸡肉火腿肠120 根，在4 000～10 000 cm－1波段采集其近红外光谱数据。为去除光谱噪声，采用Savitzky-Golay平滑预处理对光谱数据进行降维，剔除6 个异常样本，采用114?个样本进行建模分析。样品集按7∶3划分，随机选取80 个样本作为校正集，其余34 个样本作为预测集。基于主成分分析（principal component analysis，PCA）降维后，采用偏最小二乘回归（partial least squares regression，PLSR）法分别构建鸡肉火腿肠水分、脂肪和蛋白质定量模型。结果表明，水分含量PLSR模型的预测集决定系数（determination coefficient of prediction，R2p）为0.914、预测集均方根误差（root mean square error of prediction，RMSEP）为0.673、相对分析误差（ratio of prediction to deviation，RPD）为2.468；脂肪含量PLSR模型的R2p为0.929、RMSEP为0.068、RPD为2.699；蛋白质含量PLSR模型的R2p为0.873、RMSEP为0.504、RPD为2.048。综上，近红外光谱技术结合PCA与PLSR具有较好的预测能力，可实现鸡肉火腿肠主要营养成分的快速无损检测。
Abstract：Near-infrared (NIR) spectroscopy combined with chemometrics was used to predict the moisture, fat and protein contents of chicken sausage. The NIR spectra of 120 chicken sausages were collected in the wavenumber range of 4 000–10 000 cm-1. Savitzky-Golay (SG) smoothing was used for dimensionality reduction to eliminate spectral noise, removing six abnormal samples. Finally, the NIR spectra of the remaining 114 chicken sausages were used for modeling. The samples were divided into two sets at a ratio of 7:3. In total, 80 samples were randomly assigned into the calibration set, and the other 34 samples were served as the prediction set. After dimensionality reduction by principal component analysis (PCA), partial least squares regression (PLSR) was used to establish quantitative prediction models for the moisture, fat and protein contents of chicken sausage. The results showed that the coefficient of determination of prediction, root mean square error of prediction and ratio of prediction to deviation of the prediction models were 0.914, 0.673 and 2.468 for the moisture content, 0.929, 0.068 and 2.699 for the fat content, and 0.873, 0.504 and 2.048 for the protein content, respectively. Therefore, the combination of NIR spectroscopy with PCA and PLSR has good predictive ability and enables rapid and nondestructive detection of the major nutrients in chicken sausage.
2026, Vol. 40 
