摘要
黄花菜营养成分丰富,具有很高的食用、药用及经济价值,在中国产地众多。黄花菜的产地判别及可溶性蛋白质含量预测对黄花菜品质管理及农产品品牌建立和地方经济发展有着非常重要的意义。鲜黄花菜因含有多种生物碱不宜多食,市面上黄花菜多为干菜。基于近红外光谱建立了干制黄花菜产地判别模型及可溶性蛋白质含量预测模型。针对原有算法判别精确度不高及含量预测不准确等问题,对原有模型进行改进,通过结合不同的预处理方法及特征波长筛选算法有效地提高了模型的准确率。基于偏最小二乘判别分析(PLS-DA)、随机森林(RF)以及支持向量机(SVM),分别与多元散射校正(MSC)、标准正态变量变换(SNV)及Savitzky-Golay平滑(SG)三种预处理方式相结合建立干制黄花菜产地判别模型并比较模型的优劣,实验结果表明,使用PLS-DA结合MSC预处理方法进行产地判别效果最好,准确度达到93.33%,三产地的精确度与召回率均在85%以上,平均精确度达到91.9%,平均召回率为91.9%,说明该模型具有很好的准确性和稳定性,能够很好地进行干制黄花菜的产地判别。同时,使用偏最小二乘回归(PLSR)分别与多种预处理方法及无信息变量去除算法(UVE)、竞争性自适应重加权算法(CARS)和连续投影算法(SPA)三种特征波长筛选算法相结合,建立干制黄花菜可溶性蛋白质含量预测模型并进行预测结果对比,结果表明PLSR结合SG预处理及CARS特征波长筛选算法建立的模型预测效果最佳,决定系数值达到0.9815,预测均方根误差RMSEP为0.0214 g·kg^(-1),与原PLSR算法对比,值提高了0.12,RMSEP降低了0.0331 g·kg^(-1),该预测模型可很好地进行干制黄花菜可溶性蛋白质含量的预测。
Daylily is rich in nutrients and has high edible,medicinal,and economic value.It has many producing areas in China.The origin discrimination and soluble protein content prediction of daylilies are of great significance to the quality management of daylil ies,the establishment of an agricultural product brand,and the development of the local economy.Because fresh daylily contains a variety of alkaloids,it is not suitable to eat in large quantities.Therefore,most of the daylilies on the market are dried daylil ies.In this paper,the origin discrimination model and soluble protein content prediction model for dried daylily were established based on near-infrared spectroscopy.To address the issues of low discrimination accuracy and inaccurate content prediction in the original algorithm,the model was enhanced,resulting in a significant improvement in accuracy through the combination of various preprocessing methods and characteristic wavelength screening algorithms.In this study,Partial Least Squares Discriminant Analysis(PLS-DA),Random Forest(RF),and Support Vector Machine(SVM)were combined with Multiplicat ive Scatter Correction(MSC),Standard Normal Variate(SNV)and Savitzky-Golay smoothing(SG)respectively to establish the origin discrimination models of dried daylily and compare the model discrimination results.The experimental results show that PLS-DA combined with MSC has the best effect on origin discrimination,with an accuracy of 93.33%.The precision and recall of the three origins are all above 85%,with an average precision of 91.9%and an average recall of 91.9%.It demonstrates that the model exhibits good accuracy and stability,and can effectively distinguish the origin of dried daylilies.At the same time,Partial Least Squares Regression(PLSR)was combined with a variety of preprocessing methods and three characteristic wavelength screening algorithms:Unobserved Variable Elimination(UVE),Competitive Adaptive Reweighted Sampling(CARS)and Successive Projections Algorithm(SPA),respectively,to establish the prediction models of soluble protein content of dried daylily and compare the prediction results.The results show that the model established by PLSR,combined with SG and CARS,has the best predictive effect.The determination coefficient R2 reached 0.9815,and the Root Mean Square Error of Prediction(RMSEP)was 0.0214g·kg^(-1).Compared with the original PLSR,the R2 increased by 0.12,and the RMSEP decreased by 0.0331g·kg^(-1).This prediction model can well predict the soluble protein content of dried daylily.
作者
张雪莉
杨浩
李晨斐
孙一乐
刘宗霖
郑德聪
宋海燕
ZHANG Xue-li;YANG Hao;LI Chen-fei;SUN Yi-le;LIU Zong-lin;ZHENG De-cong;SONG Hai-yan(College of Agricultural Engineering,Shanxi Agricultural University,Taigu 030801,China;Dryland Farm Machinery Key Technology and Equipment Key Laboratory of Shanxi Province,Taigu 030801,China)
出处
《光谱学与光谱分析》
北大核心
2025年第9期2491-2495,共5页
Spectroscopy and Spectral Analysis
基金
国家重点研发计划子项目(2021YFD1600301-4)
大同市与山西农业大学市校合作科研项目(2020HXDTHH05)资助。
作者简介
张雪莉,女,1990年生,山西农业大学农业工程学院博士研究生,e-mail:zxl_gem@sxau.edu.cn;通讯作者:宋海燕,e-mail:yybbao@sxau.edu.cn。
