近红外光谱技术结合改良偏最小二乘法和反向传播神经网络预测葵花籽皮营养成分含量  

Prediction of Nutrient Contents of Sunflower Seed Peel by Near-Infrared Reflectance Spectroscopy Technology Combined with Modified Partial Least Square and Back Propagation Neural Network

作　　者：李欣荣 李飞[1] 翁秀秀[1] 刘保仓 邓晓裕 王新基 史艳丽 郭涛 王力[1] 李钰 李开栋 李建栋 田多湖 LI Xinrong;LI Fei;WENG Xiuxiu;LIU Baocang;DENG Xiaoyu;WANG Xinji;SHI Yanli;GUO Tao;WANG Li;LI Yu;LI Kaidong;LI Jiandong;TIAN Duohu(Key Laboratory of Grassland Livestock Industry Innovation,Ministry of Agriculture and Rural Affairs,State Key Laboratory of Grassland Agri-Ecosystems of Lanzhou University,College of Pastoral Agriculture Science and Technology,Lanzhou University,Lanzhou 730020,China;Aksu Taikun Feed Limited Liability Company,Aksu 842008,China;Minqin County Animal Husbandry and Veterinary Workstation,Wuwei 743000,China;College of Animal Science and Technology,Gansu Agricultural University,Lanzhou 730070,China;Minqin County Bureau of Agriculture and Rural Areas Animal Husbandry and Veterinary Station of Chongxing Town,Wuwei 733399,China;Minqin County Tongze Agricultural Limited Liability Company,Wuwei 733399,China;Minqin County Vocational Secondary Vocational School,Wuwei 748299,China)

机构地区：[1]兰州大学草地农业科技学院,兰州大学草地农业生态系统国家重点实验室,农业农村部草牧业创新重点实验室,兰州730020 [2]阿克苏泰昆饲料有限责任公司,阿克苏842008 [3]民勤县畜牧兽医工作站,武威743000 [4]甘肃农业大学动物科学技术学院,兰州730070 [5]民勤县农业农村局重兴镇畜牧兽医站,武威733399 [6]民勤县同泽农业有限公司,武威733399 [7]民勤县职业中等专业学校,武威748299

出　　处：《动物营养学报》2024年第11期7335-7345,共11页CHINESE JOURNAL OF ANIMAL NUTRITION

基　　金：自治区科技支疆计划(2022E02140);甘肃省重点研发计划(20YF8NH158);甘肃省重点研发国际合作类——中荷联合攻关奶牛饲料效率与牛奶品质调控关键技术集成(22YF7WA011)。

摘　　要：本研究旨在利用近红外光谱(NIRS)技术结合不同化学计量学方法建立葵花籽皮营养成分含量的预测模型。采集101份葵花籽皮样品,测定水分、粗蛋白质(CP)、有机物(OM)、中性洗涤纤维(NDF)、酸性洗涤纤维(ADF)、酸性洗涤木质素(ADL)、粗灰分(Ash)、钾(K)、钙(Ca)、磷(P)、镁(Mg)、铁(Fe)、锰(Mn)、锌(Zn)和铜(Cu)含量。通过主成分分析(PCA)剔除异常值后,利用KS算法将剩余样品分为定标集和验证集,利用NIRS技术结合改良偏最小二乘法(MPLS)和反向传播神经网络(BPNN)分别建立葵花籽皮营养成分含量预测模型。结果表明:1)葵花籽皮中水分、NDF、ADF、Ash、Mg、Fe和Mn含量的预测决定系数(RSQ)为0.88~0.99,验证相对分析误差(RPD)为2.82~8.36,利用MPLS和BPNN模型定标结果较好,且预测准确性较好,能够用于实际测量。2)葵花籽皮中K和Zn含量的MPLS模型的PRD分别为2.75和2.44,而BPNN模型的PRD分别为1.76和1.69,K和Zn含量可利用MPLS模型进行实际预测。3)葵花籽皮中CP、Ca和P含量的BPNN模型的RSQ分别为0.9、0.89和0.83,而MPLS模型的RSQ分别为0.75、0.62和0.71,CP、Ca和P含量可通过BPNN模型进行实际预测。4)葵花籽皮中ADL和Cu含量的MPLS和BPNN模型的RSQ为0.30~0.68,RPD为1.03~1.79,预测结果不可用于实际预测。综上所述,利用NIRS技术结合MPLS和BPNN建立的预测模型能够准确预测葵花籽皮中水分、CP、NDF、ADF、Ash、K、Ca、P、Mg、Fe、Mn和Zn含量。The purpose of this study was to establish prediction models of nutrient contents of sunflower seed peel by near-infrared reflectance spectroscopy(NIRS)technology combined with different chemometric methods.A total of 101 sunflower seed peel samples were collected,and the contents of moisture,crude protein(CP),organic matter(OM),neutral detergent fiber(NDF),acid detergent fiber(ADF),acid detergent lignin(ADL),crude ash(Ash),potassium(K),calcium(Ca),phosphorus(P),magnesium(Mg),iron(Fe),manganese(Mn),zinc(Zn)and copper(Cu)were determined.After the abnormal values were eliminated by principal component analysis(PCA),the remaining samples were divided into calibration set and verification set by Kennard-Stone algorithm,and the prediction models of nutrient content of sunflower seed peel were established by NIRS technology combined with modified partial least square method(MPLS)and back propagation neural network(BPNN),respectively.The results showed as follows:1)the coefficient of determination for validation(RSQ)of moisture,NDF,ADF,Ash,Mg,Fe and Mn contents in sunflower seed peel were 0.88 to 0.99,the ratio of performance to deviation for validation(RPD)were 2.82 to 8.36,the calibration results and the prediction accuracy were good by using MPLS and BPNN models,so it could be used in practical measurement.2)The PRD of MPLS model for K and Zn contents of sunflower seed peel were 2.75 and 2.44,while those of BPNN model were 1.76 and 1.69,K and Zn contents could be predicted by MPLS model in practical measurement.3)The RSQ of BPNN model for CP,Ca and P contents in sunflower seed peel were 0.9,0.89 and 0.83,while those of MPLS were 0.75,0.62 and 0.71.CP,Ca and P contents could be predicted by BPNN model in practical measurement.4)The RSQ and RPD of MPLS and BPNN models for ADL and Cu contents in sunflower seed peel were 0.30 to 0.68 and 1.03 to 1.79,so the predicted results could not be predicted in practical measurement.In summary,the prediction models established by NIRS technology combined with MPLS or BPNN can accur

关 键 词：葵花籽皮 近红外光谱技术 改良偏最小二乘法 反向传播神经网络 预测模型 

分 类 号：S816.17[农业科学—饲料科学]