关键词: 驴肉, 脂肪, 蛋白质, 近红外, 支持向量回归

Abstract: Donkey meat is of high nutrition value due to its low content of fat and high of protein. Forty fresh donkey meat samples from different individuals and parts were selected. Near infrared(NIR)diffuse reflection spectra with spectral range of 4 000-12 500 cm^-1 were collected. Reference data of fat and protein in donkey meat samples were determined by Soxhlet extraction and Kjeldahl method. Two spectral compression methods of principal component analysis(PCA)and Partial least squares regression(PLSR)were used to compress spectra data of intact and minced samples. Support vector regression(SVR)calibration models were developed with principal component scores(PCs)by PCA and latent variables(LVs)by PLSR respectively, which performances were compared with PLSR models. The optimal models were obtained by SVR with PCs decomposed from spectra of minced samples for fat, and with LVs from spectra of - DOI:10.3969/j.issn.1000-1565.2019.01.007基于支持向量回归的驴肉脂肪和蛋白质近红外检测模型优化牛晓颖¹,邵利敏²,焦慎江¹,李晓灿¹,赵志磊¹(1.河北大学 质量技术监督学院,河北 保定 071002;2.河北农业大学 机电工程学院,河北 保定 071001)摘 要:驴肉的脂肪含量低、蛋白质含量高,是一种营养价值较高的食用肉类.选择了40个不同个体和不同部位的驴肉鲜肉样品,采集了样品在4 000~12 500 cm^-1光谱的近红外漫反射光谱,并使用索氏提取法和凯氏定氮法分别检测了样品的脂肪和蛋白质参考数据.分别使用主成分分析和偏最小二乘回归对肉块和肉糜2种类型的样品光谱数据进行了压缩,结合支持向量回归算法分别建立了驴肉脂肪和蛋白质的定量模型,并与偏最小二乘回归模型进行了性能比较,发现肉糜光谱使用主成分分析降维结合支持向量回归算法所建立的驴肉脂肪模型,以及肉块光谱使用偏最小二乘回归降维结合支持向量回归算法所建立的驴肉蛋白质模型定量结果最优,其交叉验证均方根误差和相对预测误差分别达到了0.058%、14.69以及0.111%、14.39.结果表明,近红外光谱结合主成分分析或偏最小二乘回归降维以及支持向量回归算法所建立的模型预测精度较高,可对驴肉的脂肪和蛋白质含量进行可靠的检测.关键词:驴肉;脂肪;蛋白质;近红外;支持向量回归中图分类号:TS251.7 文献标志码:A 文章编号:1000-1565(2019)01-0035-06Optimization of near infrared models to predict fat and protein in donkey meat based on support vector regressionNIU Xiaoying¹, SHAO Limin², JIAO Shenjiang¹, LI Xiaocan¹, ZHAO Zhilei¹(1. College of Quality and Technical Supervision, Hebei University, Baoding 071002, China; 2. College of Mechanical and Electrical Engineering, Hebei Agricultural University, Baoding 071001, China)Abstract:Donkey meat is of high nutrition value due to its low content of fat and high of protein. Forty fresh donkey meat samples from different individuals and parts were selected. Near infrared(NIR)diffuse reflection spectra with spectral range of 4 000-12 500 cm^-1 were collected. Reference data of fat and protein in donkey meat samples were determined by Soxhlet extraction and Kjeldahl method. Two spectral compression methods of principal component analysis(PCA)and Partial least squares regression(PLSR)were used to compress spectra data of intact and minced samples. Support vector regression(SVR)calibration models were developed with principal component scores(PCs)by PCA and latent variables(LVs)by PLSR respectively, which performances were compared with PLSR models. The optimal models were obtained by SVR with PCs decomposed from spectra of minced samples for fat, and with LVs from spectra of - 收稿日期:2017-04-23 基金项目:国家自然科学基金资助项目(31201430);河北省自然科学基金资助项目(C2016201092) 第一作者:牛晓颖(1980—),女,河北清河人,河北大学副教授,博士,主要从事农产品及食品无损检测技术研究.E-mail: xiaoyingniu@126.com第1期牛晓颖等:基于支持向量回归的驴肉脂肪和蛋白质近红外检测模型优化intact samples for protein. The Root Mean Square Error and ratio of prediction to deviation of cross validation were 0.058%, 14.69 for fat; and 0.111%, 14.39 for protein. The results show that fat and protein in donkey meat can be accurately predicted by NIR and SVR with PCA or PLSR as spectral compression methods.

Key words: donkey meat, fat, protein, near infrared(NIR), support vector regression(SVR)