小米硒含量近红外预测模型的建立与评价

为实现对多样本小米硒含量的快速检测,以93份遗传背景不同的小米样品为研究对象,将样品分为校正建模集(样本容量n=51)和外部验证集(n=42),利用丹麦生产的NIRSTMDS2500台式近红外光谱仪采集光谱信息,通过标准正态变化(SNV)、卷积平滑(Detrend)等光谱预处理方法和偏最小二乘法(PLSR)建模方法建立脱壳谷子-小米总硒含量的测定模型,用工作流调用模型实现小米总硒含量的快速检测;采用国家标准规定的方法分别测定小米总硒含量,以此作为小米总硒含量预测模型的化学参比值。结果表明:小米总硒含量内部交叉验证的相关系数为84.5%;校正集均方根误差和验证集均方根误差分别为0.039 6和0.089 2,说明小米总硒含量的近红外预测值接近化学参比值;性能偏差比为5.478,大于美国谷物化学家协会和国际谷物科技协会等提出的质量控制标准,本研究建立的模型中预测集和建模集标准误差的比值为1.073 0;因此采用PLSR建立模型具有较高的预测精度且稳健程度较高,可实现对小米总硒含量的快速检测。

Establishment and evaluation of near-infrared prediction model for selenium content in millet

In order to achieve rapid detection of selenium content in multiple samples of foxtail millet, 93 foxtail millet samples with different genetic backgrounds were used as the research object, and the samples were divided into a calibration modeling set(sample size n=51)and an external verification set(n=42). The NIRS^TMDS2500 desktop near-infrared spectrometer produced in Denmark was used to collect spectral information, and the SNV and Detrend spectral preprocessing method and PLSR(Partial Least Squares)modeling method were used to establish the determination model of the total selenium content of shelled foxtail millet through variable standardization. The workflow call the model was used to realize the rapid detection of the total selenium content of foxtail millet; the total selenium content of foxtail millet was determined by by the national standard, and the results were used as the chemical reference value of the prediction model of the total selenium content of foxtail millet. The result showed that: The internal cross-validation correlation coefficient of the total selenium content of foxtail millet is 84. 5%; The root mean square error of the calibration set and the root mean square error of the verification set are 0. 039 6 and 0. 089 2 respectively, indicating that the near-infrared predicted value of the total selenium content of foxtail millet is close Chemical reference value; The performance deviation ratio is 5. 478, which is greater than the quality control standards proposed by the American Association of Cereal Chemists and International Association for Cereal Science and Technology. The ratio of the standard errors of the prediction set and the modeling set in the model established in this study is 1. 073 0, which is less than 1. 2. Therefore, the use of PLSR to establish a model has a high prediction accuracy and a high degree of robustness, which can achieve rapid detection of the total selenium content of foxtail millet.