Establishment And Application Of Portable Near Infrared Spectrometer Data Model For Alfalfa Hay

Objective:The traditional method for quality classification of loquat cannot quickly classify loquat hay quality,which is not conducive to market transactions.Under such circumstances,how to quickly detect the quality of defective products and classify them in the process of trading has become an urgent need for people.The near-infrared spectroscopy technology is now in sight,and it has the characteristics of being fast,accurate,capable of detecting multiple indicators at the same time,and being easy to operate,and can solve problems.This paper intends to explore the feasibility of the application of spectral models based on portable near-infrared instrumentation in the analysis of alfalfa hay quality in Shihezi area on the basis of predecessors.At the same time,crude protein(CP),acid detersive fiber(ADF),neutral detergent fiber(NDF),relative feed value(RFV),and roughage grading index(GI)in the near-infrared prediction model of alfalfa hay in Shihezi area has been established.It provides the basis for the research of near-infrared spectroscopy in Shihezi area.Methods:(1)To determine the content of CP,ADF and NDF in 524 samples,and to calculate the RFV and GI values,a certain number of samples were taken in proportion.The number of calibration set samples for the CP,ADF,NDF,RFV,and GI near-infrared models of the alfalfa hay were selected to be 138,138,166,144,and 166,respectively.The number of samples for the validation set was 69,69,83,71,and83,respectively.Spectral and chemical anomalies were rejected by leverage,concentration residuals,and Hotelling T2.Finally,the number of calibration and validation sets for each index's near-infrared model was determined.(2)Determine the best modeling method for each index through different preprocessing,different modeling regression methods,and different bands.(3)Compare calibration set standard error(RMSEC),correction relative coefficient(R_c),correction determination coefficient(R_c~2),verification set standard error(RMSEP),verification relative coefficient(R_p),and verification determination coefficient(R_p~2)of each model The relative analysis error(RPD)measures the accuracy and accuracy of the model.(4)The accuracy of the two models is analyzed by comparing the correlation coefficient between the prediction model and the chemical value of the certified near-infrared instrument and the CP,ADF and NDF models established in this experiment.Result:(1)The best pretreatment method of the CP model in the experiment is the baseline+smoothing treatment;the best pretreatment of the NDF model is multi-dimensional scattering correction(MSC),and the best pretreatment method for the ADF,RFV,GI model is not deal with.The best modeling method for NDF is principal component regression(PCR);the best modeling method for CP,ADF,RFV,and GI models is partial least-squares regression(PLS).All indicators using the full spectrum(1100~1800nm)modeling best.(2)The verification root mean square error(RMSEP),verification relative coefficient(R_p),and relative analysis error(RPD)values of the established CP model were 1.1243,0.9558,and 3.9773,respectively;the RMSEP and verification determination coefficients(R_p~2)and RPD of the established ADF model were obtained.The RMSEP,R~2_p and RPD values were 1.5654,0.9150,and 3.5935,respectively.The RMSEP,R_p~2,and RPD values of the NDF models were 1.7834,0.9326,and 3.1541,respectively;the RMSEP,R_p~2,and RPD values of the RFV models that were built were 12.7513,0.9292,3.5388;The RMSEP,R_p~2,and RPD values of the constructed GI model are 1.1999,0.9591,and 4.7629,respectively.The above indicators indicate that the CP,ADF,NDF,RFV,and GI models built can be accurately predicted.(3)Among the CP,ADF,and NDF models established in this experiment,the correlation coefficients between the external verification set results and the chemical values predicted by the experiment were 0.8801,0.9023,and 0.8826,respectively.Compared with the portable near infrared self-contained models,The correlation coefficient(0.8617,0.8707,0.8721)between the validation set prediction results and the chemical values is higher.Therefore,it can be determined that the model built in this experiment can improve the accuracy of the portable near-infrared model.Conclusion:The near-infrared model of CP,ADF,NDF,RFV and GI in alfalfa hay established in this experiment can be applied to actual measurements,and the accuracy of the portable near-infrared model is improved to some extent.