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
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 meth-ods.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 elim-inated 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 accurately predict moisture,CP,OM,NDF,ADF,Ash,K,Ca,P,Mg,Fe,Mn and Zn contents in sunflower seed peel.[Chinese Journal of Animal Nutrition,2024,36(11):7335-7345]
万方数据
维普
