基于DPLS+PLS算法的大米直链淀粉含量快速检测模型构建
Development of a Rapid Detection Model for Rice Amylose Content Using the DPLS+PLS Algorithm
摘要
利用近红外光谱技术可以实现对大米中直链淀粉含量的快速无损检测,但传统偏最小二乘法(Par-tial Least Squares,PLS)准确性有待提高.本研究提出一种利用 DPLS+PLS 构建预测模型,即先利用判别式偏最小二乘法(Discriminant Partial Least Squares,DPLS)性分类,再对每类数据建立偏最小二乘法定量检测.结果表明,大米样品原始光谱经过 SNV+Detrend+一阶求导(7 点平滑)预处理后,利用 DPLS+PLS 建立的预测模型性能良好,验证集 R2p 为0.9986、SEP 为0.26,RPD 为 7.11.相比传统单一 PLS 预测模型,本研究构建的模型提升了直链淀粉含量预测精度与稳定性,具有良好的推广应用前景.
Near-infrared(NIR)spectroscopy offers a rapid,non-destructive method for determining amylose con-tent in rice;however,the predictive accuracy of conventional Partial Least Squares(PLS)models requires im-provement.This study proposed a hybrid Discriminant Partial Least Squares and Partial Least Squares(DPLS+PLS)prediction mode to address this limitation.The approach utilizes DPLS for initial qualitative classification of samples,followed by the development of distinct PLS regression models for quantitative determination within each i-dentified category.After preprocessing the raw spectra of the rice samples with standard normal variate transforma-tion(SNV),detrending,and a first-order derivative(7-point smoothing),the DPLS+PLS model exhibited good performance.The validation set yielded a coefficient of determination(R2p)of 0.9986,a standard error of predic-tion(SEP)of 0.26,and a residual predictive deviation(RPD)of 7.11.Compared to a conventional PLS model,the model developed in this study demonstrated improved accuracy and stability in predicting amylose content and showed good potential for wider application.
年，卷(期)：
2026, 55(4)
分类号：
TS210.7
机标分类号：
TS231
TP393.08
O657.3
在线出版日期：
2026年9月15日
基金项目：
2023B0202060002:广东省重点领域研发计划项目:广东省粮食BZ2023001:物资储备保障中心横向合作项目，
