Determination Of Nutrient Values Of Wheat By Near Infrared Spectroscopy

In recent years,the application of wheat in livestock and poultry diets has gradually increased.The nutritional and physical properties of wheat vary greatly due to factors such as variety,origin,processing method,and storage method.Traditional wet chemistry methods for determining nutrient content in wheat involve the use of multiple reagents and complex procedures that make on-site rapid detection challenging,and are prone to significant variability due to human factors.Therefore,there is a need to develop a fast,non-destructive,and precise detection method to accurately predict the nutrient content in wheat.Such predictions can be used to formulate diets rations that are tailored to meet the specific nutritional requirements of poultry and enabling precision feeding.This study aims to investigate the feasibility of using near-infrared spectroscopy reflection technology for nutritional assessment of wheat.A total of 165 wheat samples from different regions and varieties in China(Shaanxi,Hebei,Liaoning,Jiangsu,Henan,Shandong,and Zhejiang)were selected,with 124 samples used to establish near-infrared calibration models for moisture,crude protein,crude ash,calcium,total phosphorus,and amino acids,and the models were cross-validated.The remaining 41 samples were used as an external validation set to test the stability and adaptability of the Hanon N500 portable near-infrared analyzer and the BRUKER Fourier transform infrared spectrometer calibration and verification models.In Experiment 1,the reference values for moisture,crude protein,crude ash,calcium,and phosphorus content in wheat were determined using wet chemical methods.The results showed that there was significant variation in the nutritional composition of different types of wheat,with coefficients of variation ranging from 9.72%to 29.21%for conventional nutrients,including crude protein,crude ash,calcium,and total phosphorus.The coefficient of variation for moisture was 9.72%,while for all amino acids it ranged from 16.97%to 30.21%.In Experiment 2,the Hanon N500 portable near-infrared analyzer was used to establish calibration models for moisture and crude protein in wheat.The coefficient of determination for calibration(R~2c)were 0.95 and 0.92,respectively,while the coefficient of determination of cross-external validation(R~2cv)were 0.94 and 0.88.The coefficient of determination for validation(R~2v)were 0.89 and 0.79,and the ratio of performance to deviation for validation(RPDv)were 3.02 and 2.2,respectively.The model showed good prediction capability for moisture,making it suitable for practical analysis.However,the prediction capability for crude protein was only moderate and could only be used for rough screening of samples.The calibration models for crude ash,calcium,and total phosphorus were not yet suitable for actual prediction.For essential amino acids,the R~2v and RPDv for isoleucine were 0.89 and 2.99,respectively,and for histidine they were 0.87 and 2.75,indicating that the model had good prediction capability and could be used for practical analysis.However,the prediction capability for valine and lysine was only moderate and could only be used for rough screening of samples.The calibration models for threonine,methionine,leucine,phenylalanine,and arginine showed poor prediction performance and could not be used for actual measurement.For non-essential amino acids,the prediction performance of the models for aspartic acid,serine,glutamic acid,glycine,alanine,cysteine,tyrosine,and proline were all unsatisfactory.In Experiment 3,the R~2c of wheat moisture and crude protein established by BRUKER desktop Fourier transform infrared spectrometer were 0.88 and 0.94,R~2cv were 0.86 and 0.93,R~2v were 0.81 and 0.86,and RPDv were 2.33 and 2.98,respectively.The crude protein prediction ability was good and could be used for practical analysis.Moisture is only moderately predictive.For essential amino acids,the R~2v of threonine,valine,isoleucine,leucine and arginine were 0.86,0.89,0.93,0.89 and 0.92.The RPDv of threonine,valine,isoleucine,leucine and arginine were 2.69,3.01,3.82,2.96 and 3.63,respectively,indicating that the model had good predictive capability and could be used for practical analysis.The R~2v and RPDv of methionine were 0.79 and 2.26,respectively.The prediction capability for methionine was only moderate,while the prediction performance of the models for phenylalanine,lysine,and histidine were unsatisfactory.For non-essential amino acids,the R~2v of aspartic acid,serine,glutamic acid,glycine,alanine and tyrosine were 0.84,0.91,0.93,0.89,0.90 and 0.91.The RPDv were 2.51,3.32,3.69,3.00,3.18 and 3.28,respectively,indicating that the model had good prediction capability and could be used for practical analysis.The prediction capability for cysteine and proline was only moderate.In summary,the model established by a portable near-infrared analyzer can be used to predict moisture content and two essential amino acids,while the model established by a desktop near-infrared analyzer can be used to predict crude protein and five essential amino acids.Compared with desktop analyzers,portable near-infrared analyzers have the advantages of small size,portability,and lower cost,making them more suitable for small feed companies.However,their spectral range is relatively narrow,and the collected spectral information is not comprehensive enough,resulting in a relatively worse modeling effect for amino acids.The performance of the instrument still needs further optimization.