利用近红外光谱技术的葵花籽品质分析  

Quality Detection of Sunflower Seed in Near Infrared Spectroscopy

作　　者：李壮壮 吴琼水[1] 黄莎 LI Zhuang-zhuang;WU Qiong-shui;HUANG Sha(School of Electronic Information,Wuhan University,Spectral Imaging Laboratory,Wuhan 430072,China)

机构地区：[1]武汉大学电子信息学院,光谱成像实验室,武汉430072

出　　处：《科学技术与工程》2020年第10期4061-4065,共5页Science Technology and Engineering

基　　金：国家科技支撑计划(2011BAF02B02)。

摘　　要：葵花籽作为中国需求量很大的油料作物,常用的化学检测方法虽然准确性高,但是时间久,破坏样本完整性,难以实现快速检测。为了研究近红外光谱法快速无损检测葵花籽中蛋白、脂肪及水分含量的准确性,使用454份葵花籽样本近红外光谱数据及蛋白质、水、脂肪三个含量信息为测试对象,随机选取其中383份作为测试集,71份作为验证集,对测试集使用不同的预处理方法之后分别进行PLS(partial least-square method)和BP神经网络建模,并通过验证集对模型进行预测分析。结果发现:①PLS模型预测中发现小波变换的预处理方法对蛋白质含量的预测最好,小波变换的预处理方法对水分含量的预测最好,标准化的预处理方法对脂肪含量的预测最好;②BP神经网络预测中一阶导数+均值中心化对蛋白质含量的预测最好,一阶导数对水分含量的预测最好,一阶导数+标准化对脂肪含量的预测最好。比较两种神经网络模型的预测结果,PLS模型预测精度要高于BP神经网络模型。Sunflower seeds are largely demanded in China.Although the chemical detection methods used are accurate,the time required is long,and the sample is destroyed.To search for a quick,intact,and accurate method,a new method in near-infrared spectroscopy for rapid detection of protein,fat,and moisture content in sunflower seeds was developed.In the test,454 sunflower seeds were used and the near-infrared spectroscopic data of protein,water,and fat content were analyzed,of which 383 were used as test sets,and 71 were used as validation sets.(partial least-square method,PLS)and BP neural networks were modeled separately using different pre-processing methods for the test set.The model prediction results were compared against the verification set.Results show that,in the PLS model,the wavelet transform pretreatment method had the best prediction of protein content,the wavelet transform pretreatment method had the best prediction of water content,and the standardized pretreatment method had the best prediction of fat content;in the BP neural network model,the first derivative+mean centralization was the best for protein content prediction,the first derivative was the best for moisture content prediction,and the first derivative+standardization was the best for fat content prediction.Compared the prediction results of the two models,the prediction in PLS model was more accurate than BP neural network model.

关 键 词：近红外光谱技术 葵花籽品质测定 PLS BP神经网络 预处理方法 

分 类 号：TS210.1[轻工技术与工程—粮食、油脂及植物蛋白工程]