Research On The Detection Method Of Main Components Of Corn Stalk Based On Near Infrared Spectroscopy Technology

Corn straw resource is the most abundant crop straw resource in China,which is often used as the main raw material for industrial and biomass energy development.The content of cellulose and lignin,the main component of corn straw,often affects the capacity of industrial and biomass energy.Therefore,it is very important to detect the cellulose and lignin content of corn straw rapidly.At present,chemical methods such as van's method are mainly used to detect cellulose and lignin content in plants.Chemical methods not only measure long period,but also destroy the internal structure of samples.Near infrared spectroscopy(NIRS)has been widely used in agriculture,medicine,petroleum and other fields in recent decades due to its advantages of fast,nondestructive and simple operation.Therefore,this study attempts to use near-infrared spectroscopy to measure the cellulose and lignin content of corn straw quickly.The main contents of this paper are as follows.(1)The prediction model of corn straw cellulose content was established.The near-infrared spectrum data of corn straw samples were collected,and the normal form method(van Soest)measures the cellulose content of the corresponding samples,and the abnormal samples were eliminated by PCA-MD,MCCV and odxy.The spectra after the abnormal samples were pretreated by savitzky-Golay(S-G)smoothing and orthogonal signal correction(OSC)and X-Y distance The combined sample division method(SPXY)divides the sample set in a ratio of 2:1.In this paper,the joint interval partial least squares(SIPLS),backward interval partial least squares(BIPLS)and simulated genetic annealing algorithm(GSAA)are used to select characteristic wavelengths.For the selected characteristic wavelength points,the corresponding pls and SVM models are established.The performance of cellulose PLS model is better than that of SVM model.Considering the input dimension,distribution characteristics and performance of the model,SIPLS-GSAA-PLS model has more advantages.The input of the model is reduced from 1845 dimension of the original spectral data to 32 dimension.The input variables are The evaluation parameters R_P,RMSEP and RPD of the model are 0.9414,0.8568 and 2.9501,respectively.Experimental results show that SIPLS-GSAA feature selection method can greatly reduce the input of the model and improve the performance of the model.(2)The prediction model of lignin content in corn straw was established.Firstly,MCCV is used to eliminate abnormal samples.After SPXY and SG+OSC preprocessing,the characteristic wavelengths are selected by SIPLS,BIPLS and GSAA,and then the PLS and SVM models are established.Through comparison,in the lignin model,the performance of nonlinear SVM model is better than that of linear PLS model,the input of SIPLS-SVM model is reduced to 160dimensions,and the evaluation parameters R_P,RMSEP and RPD of the model are 0.9590,0.3431and 3.5499,respectively.The experimental results show that the wavelength selected by SIPLS method is more suitable for SVM model.In this study,the influence of five characteristic wavelength selection algorithms,BIPLS,SIPLS,GSAA,BIPLS-GSAA and SIPLS-GSAA,on the model was discussed.Finally,the near-infrared spectrum detection models of corn straw cellulose and lignin were established,which provided theoretical basis for the rapid detection of corn straw,and also provided some new ideas for the comprehensive utilization of crop straw.