Research On The Detection Technology Of Acid Detergent Fiber Content In Corn Stalk Based On NIRS

As the main source of herbivorous livestock,the determination of ADF content in corn stalk is of great significance for the breeding of maize plant and the selection of maize varieties and the increase of edible value of livestock.Based on the detailed analysis of the mechanism of ADF chemical functional groups in corn stalks,this paper proposed the use of near infrared spectroscopy(NIRS)to quickly and non-destructively detect the content of ADF in corn stalks.After obtained 569 corn stalks from the planting base of the Hulan Campus of Heilongjiang University,NIRS DS2500 spectral analyzer and F12A automatic cellulose meter were severally used to acquire the diffuse reflectance spectra within the range of400-2500nm and the corresponding ADF content.In order to build a high accuracy model,the study used mahalanobis distance method to eliminate the abnormal 31samples.For enhancing the spectral features and reducing noise,first derivatives(lst),second derivative(2nd),standard normal variable(SNV),multiplicative scatter correction(MSC),moving smoothing(MA)and its combinations were selected to preprocess spectra.Then,at the ratio of 2:1,the Sample set partitioning based on joint X-Y distance(SPXY)algorithm was using to divide the samples into calibration set and test set.In addition,to reduce data dimension and improve modeling efficiency,the study applied correlation coefficient method(CCM)and principal component analysis(PCA)method to eliminate irrelevant wavelengths,then the characteristic wavelengths were extracted by using of the CCM and PCA in series.Finally,partial least squares regression(PLSR)model,back propagation neural network(BP-NN)model,radial basis function neural network(RBF-NN)model,support vector regression(SVR)model,the PLSR-BPNN model of PLSR combining with BP-NN,the PLSR-RBFNN model of PLSR combining with RBF-NN were built respectively.By comparing the coefficient of determination(R~2),root mean square error(RMSEP),relative analysis(RPD),relative standard deviation(RSD),the PLSR-BPNN model was determined to detect the ADF content in corn stalk.Moreover,the verification results of the optimal model showed that the NIRS technology could be applied to accurately detect the content of ADF in corn stalks,which had positive significance for the development of agricultural and animal husbandry.