基于近红外光谱模型转移的牛奶蛋白检测方法研究

Research on online dairy product quality detection method based on near-infrared spectral model transfer

中文关键词:  近红外光谱，模型转移，乳制品，在线检测

英文关键词:Near-infrared  spectroscopy, model  transfer, dairy  products, on-line  detection

中文摘要:

      目的 研究基于近红外光谱模型转移的牛奶蛋白检测方法。方法 分别采用实验室与在线检测近红外光谱仪采集生产过程中原料奶样品的近红外光谱, 研究斜率截距法(slope/bias, S/B)、分段直接标准化(piecewise direct standardization, PDS)算法、Shenk’s方法在不同仪器测量光谱之间模型转移应用, 优化模型参数, 提高实验室仪器建立的校正模型应用于在线光谱仪器的预测精度。结果 经过Shenk’s算法转移, 主从机的光谱平均差异降低为0.0075, 光谱校正率达到98.95%。利用模型转移方法与偏最小二乘模型结合, 将实验室分析光谱仪建立的模型用于生产在线光谱仪测量光谱预测, 显著提高了牛奶中蛋白质含量预测准确度, 不同仪器之间模型预测相对均方根误差从5.52%下降到2.03%。结论 本研究的方法实现了实验室分析与在线检测仪器测量光谱及定量分析模型转移共享, 为近红外在线检测的智能化改进提供了基础。

英文摘要:

      Objective To study the online dairy product quality detection method based on near-infrared spectral model transfer. Method The near-infrared spectroscopy of raw milk samples during production were collected using laboratory and online detection near-infrared spectrometers respectively. The slope/bias (S/B), piecewise direct standardization (PDS) algorithm, and Shenk’s method were studied to transfer models between different instrument measurement spectra, optimize model parameters, and improve the prediction accuracy of the calibration model established by laboratory instruments applied to online spectral instruments. Results After the Shenk’s algorithm transfer, the average spectral difference between the master and slave machines was reduced to 0.0075, and the spectral correction rate reached 98.95%. By combining model transfer method with partial least squares model, the model established by the laboratory analysis spectrometer was used to predict the measurement spectrum of the production online spectrometer, significantly improving the accuracy of protein content prediction in milk. The relative root mean square error of model prediction between different instruments decreased from 5.52% to 2.03%. Conclusion The method of this study achieves the transfer and sharing of laboratory analysis and online detection instrument measurement spectra and quantitative analysis models, providing a foundation for the intelligent improvement of near-infrared online detection.