Study On Detection Of The Protein Content Based On Near-infrared Spectroscopy In Miscanthus Sinensis

Miscanthus sinensis is a low input,high biomass yielding perennial woody grass that is adaptable to a wide range of environments.Given these characteristics,Miscanthus has huge potential to be used as a bio-energy crop.Miscanthus is also used as forage grass in animal husbandry,where protein content is an important indicator of livestock feed quality.This study thus evaluated near infrared spectroscopy as a non-destructive analysis method for protein content determination in Miscanthus.Samples of Miscanthus from four countries(China,Japan,South Korea,and USA)were used in this study.Data from Kjeldahl chemical determination in combination with near infrared spectroscopy was used to create an ideal NIR calibration model of protein content.The main results are as follows:(1)Protein contents of 235 samples from the four countries were compared.The study found that Miscanthus sinesis samples from China had the highest proportion of protein content at 1.318%followed by samples from Japan with 1.1234%.Miscanthus sinensis samples from South Korea and USA had lower proportions of protein content at 0.9215%and 0.8572%respectively.From this we can conclude that Chinese Miscanthus is more suitable for use as livestock feed due to its relatively higher protein content.(2)Seven kinds of pretreatment methods were compared to build a partial least squares(PLS)model of full spectrum.The results showed that the optimal pretreatment methods in Protein content were 1st Derivative and Baseline.The R-Square of cross validation(R_c~2)and R-Square of prediction(R_p~2)of 1st Derivative pretreatment are 0.962 and 0.947,respectively.The SEC and SEP values were 0.080 and 0.097,respectively and thus lower than original spectrum.The R_c~2 and R_p~2 of Baseline pretreatment are 0.954 and 0.949,respectively.The SEC and SEP values were0.087 and 0.095,respectively and lower than the original spectrum.(3)Protein contents of 324 samples were built full spectrum,based on linear method of partial least squares(PLS),nonlinear method of support vector machine(LS-SVM),radial basis function neural network(RBF-NN)and RBF＿LSSVM were used to perform models.The nonlinear models based on RBF-NN was optimum,with R_c~2 and R_p~2 are 0.9942 and 0.9508,respectively.The results of these models showed that all of the nonlinear method models in the protein content were better than those using linear method.(4)The characteristic spectrum is based on the methods of CARS-MLR and SPA-MLR,with the number of characteristic spectrum are 84 and 35,respectively.Nevertheless,The effect of CARS-MLR and SPA-MLR is not better than four methods of full spectrum.The method of CARS-MLR with R_c~2 and R_p~2 are 0.8826 and 0.7518,respectively.The method of SPA-MLR with R_c~2 and R_p~2 are 9014 and 0.8630,respectively.In summary,NIRS can provide a reference model with the character of fast and accurate.Miscanthus sinensis combine with NIRS testing protein content will have a good application in practice with the high value of feeding.