FTIR结合ELM对黑果腺肋花楸黄酮、多糖含量快速预测

Fast Prediction of Flavone and Polysaccharide Contents in Aronia Melanocarpa by FTIR and ELM

黑果腺肋花楸是继蓝莓后的又一小浆果,因其黄酮含量高于蓝莓受到关注,已获进入新资源食品名单,并在饮料行业中使用。黑果腺肋花楸黄酮、多糖是其果汁及果渣中的主要生物活性成分,也是影响其品质的重要因素。以中红外光谱技术结合化学计量学方法对黑果腺肋花楸黄酮、多糖含量进行预测,为建立简便、快捷的黑果腺肋花楸产品质量检测方法提供基础。采集15个产区共750份黑果腺肋花楸红外光谱数据,测量每份样品黄酮、多糖含量,采用K-S样本划分法按4∶1的比例将样本划分为校正集和验证集,并对分组后的光谱信息进行多元散射校正(MSC)、标准正态化(SNV)、平滑(SG)、一阶导数(FD)、二阶导数(SD)等光谱预处理,与原始光谱进行极限学习机(ELM)建模预测效果对比,确定最佳光谱预处理方法。采用竞争性自适应重加权算法(CARS)和连续投影算法(SPA)进行黑果腺肋花楸黄酮、多糖特征光谱波段选取,将2种方法选取的光谱数据结合偏最小二乘回归法(PLS)、极限学习机(ELM)、支持向量机(SVM)进行建模对比,选出预测效果最佳的算法模型。结果表明,7种光谱预处理方法中,MSC对原始光谱的处理效果最佳,在此处理下黄酮含量预测模型RPD值为6.2017,多糖含量预测模型RPD值为5.4473,预测模型的误差显著下降。经CARS、SPA提取特征光谱后,进行3种算法的建模结果对比,确定CARS-ELM为效果最佳的含量预测模型,其中黄酮含量预测模型的R_(C)为0.9972,RMSEC为0.0175,R_(P)为0.9912,RMSEP为0.0311,RPD为10.6315;而多糖含量预测模型中的R_(C)为0.9965,RMSEC为0.0173,R_(P)为0.9867,RMSEP为0.0337,RPD为8.6647。中红外光谱结合化学计量学方法,尤其是CARS-ELM模型能够更准确地对黑果腺肋花楸黄酮、多糖含量进行预测,此方法的开发为黑果腺肋花楸质量评价提供了快速、简便的方法。

Aronia melanocarpa is one kind of berrie richer in flavone than blueberry and has thus been approved as a new food resource largely used in the beverage industry.Flavone and polysaccharides have been revealed to be the main bioactive components in its fruit juice and pomace,affecting its quality.Therefore,their contents were predicted by infrared spectroscopy combined with chemometrics,which provided a basis for establishing a simple and rapid method for the quality detection of A.melanocarpa.A total of 750 infrared spectral data of A.melanocarpa from 15 production areas were collected,and their contents in flavone and polysaccharides were measured.The samples were divided into calibration set and validation set by K-S sample division method in the proportion of 4:1.The spectral information after grouping was pretreated by multiple scattering correction(MSC),standard normalization(SNV),smoothing(SG),first derivative(FD),second derivative(SD)and other spectral preprocessing,and the best spectrum preprocessing method was determined.The competitive adaptive reweighting algorithm(CARS)and continuous projection algorithm(SPA)were used to select the characteristic spectral bands of flavone and polysaccharides in A.melanocarpa.The spectral data selected by the two wave methods were combined with partial least square regression(PLS),limit learning machine(ELM)and support vector machine(SVM)for modeling and comparison,and the algorithm model with the best prediction effect was selected.The results showed that,MSC had the best effect on the original spectrum among the seven spectral pretreatment methods.Under this treatment,the RPD value of the flavone content prediction model was 6.2017,and 5.4473 for the polysaccharide content prediction mode,with the eror of the prediction model significantly decreased.After extracting the characteristic spectra by CARS and SPA,the modeling results revealed that the R_(c),R_(p),and RPD of the flavone content prediction model were respectively 0.9972,0.9912 and 10.6315,while they were 0.9965,0.9867 and 8.6647 respectively for the polysaccharide content prediction model.Therefore,infrared spectroscopy combined with chemometrics methods,especially the CARS-ELM model,can accurately predict the contents of flavone and polysaccharides in A.melanocarpa,and the development of this method provides a fast and simple method for its quality evaluation.