近红外光谱波段优化在东北松子蛋白质定量检测中的应用

为了探究一种快速、无损与简便的东北松子品质检测方法,近红外光谱技术被应用到东北松子蛋白质无损检测研究中。利用偏最小二乘法建立带壳松子和去壳松仁的蛋白质定量分析模型,采用求导、多元散射校正、变量标准化校正、矢量归一化预处理方法优化模型,利用反向间隔偏最小二乘法、无信息变量消除法选取特征波段,建立全波段和特征波段下的偏最小二乘蛋白质预测模型。结果表明,带壳松子光谱经矢量归一化预处理方法后构建的模型最优,松仁光谱经变量标准化校正预处理方法后构建的模型最优;波段筛选能够优化模型质量,其中反向间隔偏最小二乘法的筛选结果最优,其带壳松子和松仁蛋白质模型校正集相关系数分别为0.9056和0.9383,验证集均方根误差分别为0.6670和0.5761。由此可知,经过优化后,模型的预测性能得到了提高,为带壳松子和松仁的蛋白质在线检测提供了一定的参考价值。

Application of Band Optimization of Near-infrared Spectra for Quantitative Detection of Proteins in Northeastern Pine Nuts

Near-infrared (NIR) spectroscopy was performed to develop a fast, nondestructive, and simple method to test the quality of Northeastern pine nuts. Using shelled and deshelled pine nuts, quantitative analysis models of proteins in the nuts were established using partial least squares (PLS) and the models were optimized by derivation, multiplicative scatter correction (MSC), standard normal variate (SNV), and vector normalization pretreatment. Backward interval partial least squares (BiPLS) and elimination of uninformative variables (UVE) were used to select characteristic bands to establish PLS protein prediction models with full wavelength and characteristic bands. The results showed that the models established after preprocessing with vector normalization and SNV exhibited optimal performance for deshelled and shelled pine nuts, respectively. The models were optimized by band selection and the optimum screening result was presented using BiPLS. The correlation coefficients (RC) of calibration subset of the protein models for deshelled and shelled pine nuts were 0.9056 and 0.9383, respectively. The root-mean-square error (RMSE) values of the validation subset were 0.6670 and 0.5761, respectively. Therefore, after optimization, the model prediction performance was improved, thus providing a reference point for online testing of proteins in deshelled and shelled pine nuts.