摘要
为实现小麦粉品质的定量快速检测,基于近红外(Near infrared,NIR)光谱技术建立了小麦粉水分、灰分和面筋含量的偏最小二乘(Partial least squares,PLS)模型。通过分析和比较不同数据处理方法对建模结果的影响,筛选出最佳的样本划分、预处理和特征提取方法。水分含量模型采用随机选择法(Random selection,RS)结合卷积平滑法以及竞争自适应重加权法(Competitive adaptive reweighted sampling,CARS)进行数据处理,所建立的定量模型效果最好,交叉验证均方根误差(Root mean square error of crossvalidation,RMSECV)为0.0937,预测均方根误差(Root mean squared error of prediction,RMSEP)为0.0996,预测集决定系数(Coefficient of determination,R2)为0.9215,相对分析误差(Relative percent deviation,RPD)为3.57。灰分含量模型采用RS法结合一阶导数以及卷积平滑法的建模效果最好,RMSECV为0.0509、RMSEP为0.0731、R2为0.8718、RPD为2.79。面筋含量模型采用RS法结合归一化(Normalize)的建模效果最好,RMSECV为0.3589、RMSEP为1.1194、R2为0.7609、RPD为2.05。实验结果表明,经预处理和特征提取处理后,所建立的小麦粉品质定量快速检测模型稳定可靠,具有较高的预测精度。
In order to obtain a rapid quantitative detection of wheat flour quality,partial least squares(PLS)model were established based on near infrared(NIR)spectrum technology for the content of moisture,ash content and gluten content in wheat flour.By analyzing and comparing the influence of different data processing methods on the modeling results,the best methods of sample division,pretreatment,and feature extraction were selected.The accuracy of quantitative model of the moisture content model achieve the best by using random selection(RS)method,convolution smoothing method and competitive adaptive reweighted sampling(CARS)method.The cross validation root mean square error(RMSECV),the prediction root mean square error(RMSEP),the determination coefficient of the prediction set and the relative percent deviation(RPD)were 0.0937,0.0996,0.9215 and 3.57,respectively.RS method combined with first derivative and convolution smoothing method has the best modeling effect for ash content model,the RMSECV,RMSEP,R2 and RPD were 0.0509,0.0731,0.8718 and 2.79.RS method combined with normalize is the best method for gluten content model,the RMSECV,RMSEP,R2 and RPD were 0.3589,1.1194,0.7609 and 2.05,respectively.The experimental results show that after pretreatment and feature extraction,the established rapid quantitative detection models of wheat flour quality are stable and reliable with high prediction accuracy.
作者
孙晓荣
张晨光
刘翠玲
吴静珠
张善哲
李彦昊
SUN Xiaorong;ZHANG Chenguang;LIU Cuiling;WU Jingzhu;ZHANG Shanzhe;LI Yanhao(School of Artificial Intelligence,Beijing Technology and Business University,Beijing 100048,China;Beijing Key Laboratory of Big Data Technology for Food Safety,Beijing Technology and Business University,Beijing 100048,China)
出处
《食品科技》
北大核心
2023年第11期246-252,共7页
Food Science and Technology
基金
北京市自然科学基金项目(4222043)
2021年教育部高教司产学合作协同育人项目(202102341023)
2022年北京工商大学研究生教育教学改革专项(20220613)。
关键词
小麦粉
近红外光谱技术
光谱预处理
波长筛选
定量分析
wheat flour
near infrared spectroscopy technology
spectral pretreatment
wavelength selection
quantitative analysis
