Establishment Of Alfalfa Hay Quality Parameters Model Based On Portable Near Infrared Spectrometer

Alfalfa trade focuses on graded sales and quality pricing.Traditional quality evaluation methods are time-consuming and costly,so an analytical method that can quickly detect and grade alfalfa hay in the process of negotiation is urgently needed.The near infrared spectroscopy has the advantages of fast detection speed,no sample preparation,simultaneous detection of multiple chemical components,and online analysis,etc.,which has been applied in the agricultural field.However,the traditional desktop near-infrared spectrometer is only suitable for use in the laboratory environment,and it takes a certain amount of time to transmit samples and obtain results,while the portable near-infrared spectrometer is convenient to carry and can detect samples immediately,which is more suitable for on-site trading.Therefore,on the basis of previous studies,this paper intends to explore the feasibility of applying the prediction model based on portable near infrared spectrometer in the analysis of alfalfa hay quality.Meanwhile,partial least square method is used to initially establish the near infrared prediction model of DM,CP,NDF,ADF and RFV of alfalfa hay in Gansu region.It provides basic data for the application and research of portable near infrared spectrometer in alfalfa quality evaluation.In this experiment:1)DM,CP,NDF,ADF and RFV of 493 alfalfa hay samples collected in Gansu were determined and calculated;2)All samples are divided into correction set samples and prediction set samples by using Kennard-stone algorithm;3)Monte Carlo cross-validation algorithm was used to eliminate abnormal samples in the correction set;4)Different pretreatment methods were used to conduct PLS modeling for different quality parameters using the full spectrum segment(950～1650 nm),and the best modeling method for each index was determined;5)By comparing the Slope(Slope),calibration standard deviation(SEC),calibration determination coefficient(R2c),cross validation standard deviation(SECV),cross validation determination coefficient(R2(CV)),prediction standard deviation(SEP),prediction determination coefficient(R2p)and relative analysis error(RPD)of the constructed alfalfa hay component models.Measure the accuracy of the model.The results are as follows:1)The optimal pretreatment method for DM model in this experiment was SNV treatment,Slope,SEC,R~2c,SECV,R~2(CV),SEP,R~2p and RPD were 1.098,0.035,0 0.220,0.996 and 4.241,respectively.2)The optimal pretreatment method for CP model was SNV+Detrend treatment,Slope,SEC,R~2c,SECV,R~2(CV),SEP,R~2p and RPD were 1.027,0.209,0.990,0.272,0.981,0.285,0.986 and 7.193,respectively.3)The optimal pretreatment method of NDF model is no treatment,Slope,SEC,R~2c,SECV,R~2(CV),SEP,R~2p and RPD are1.004,0.654,0.975,0.783,0.960,1.147,0.970 and 3.389,respectively.4)The optimal pretreatment method of ADF model is SNV treatment,Slope,SEC,R~2c,SECV,R~2(CV),SEP,R~2p and RPD are 0.950,0.365,0.990,0.433,0.983,0.663,0.984 and 5.430,respectively.5)The optimal pretreatment method of RFV model is SNV treatment,Slope,SEC,R~2c,SECV,R~2(CV),SEP,R~2p and RPD are 1.033,2.158,0.981,2.521,0.972,5.679,0.944 and 2.770,respectively.In conclusion,the NIR models of DM,CP,NDF and ADF of alfalfa hay in Gansu province established in this study can be applied to actual measurement and can be accurately predicted,but the near infrared model of RFV can be applied to actual measurement but can only be roughly estimated.