Research On Detection Of Soybean Meal Quality By NIR Based On PLS-GRNN

Soybean meal is a by-product of soybean oil extraction after proper drying and heat treatment.It is mainly used for livestock feed,as well as fertilizer and supplementary material for food production.Its nutritional value is determined by its quality.The existing soybean meal quality detection methods include chemical analysis and chromatographic analysis,etc.,which have many problems,such as excessive use of toxic chemical reagents,complex operation and long analysis time,etc.,which cannot meet the needs of rapid detection and regulation of actual production lines.Near infrared spectroscopy(NIR)has the characteristics of fast,efficient and it is easy to realize on-line analysis.Therefore,this study proposed a multi-component soybean meal quality detection method based on near infrared spectroscopy analysis,in order to be used for on-line detection and control of product quality.Firstly,449 soybean meal samples were collected from soybean oil processing line.The chemical values of water,protein and fat in soybean meal were determined by national standard method,and the near-infrared diffuse reflectance spectra of the samples were obtained by synchronous scanning.The abnormal samples were removed by Mahalanobis distance method.The selection of decomposition scale,wavelet base and threshold in wavelet denoising is discussed.Compared with other denoising methods,wavelet denoising effect is found to be the best.Two machine recognition methods,Kennard Stone and SPXY,were used to carry out sample diversity.PLS model was established to compare the diversity effect and determine the best sample diversity of different components.Secondly,in order to explore the NIR absorption characteristics of soybean meal components,IPLS was used for feature extraction from 4000-10000cm-1,and the characteristic absorption bands of water,protein and fat are respectively 4904-5200cm-1 4304-4600cm-1 and 4304-4600cm-1,to eliminate redundant information of the spectrum,reduce the computational complexity of the model and improve the prediction performance.Finally,the nonlinear modeling method based on PLS-GRNN is discussed to solve the nonlinear response problem in NIR spectrum analysis.In order to reduce the input variables of the network,reduce the network scale and improve the operation speed,PLS was used to reduce the dimension of the spectral data,and the main factor score was extracted as the input variable of GRNN.Through cross-validation cycle method optimizing network parameter—smoothing factor(spead),the PLS-GRNN prediction model of multi-component content of soybean meal was established.Compared with the classical PLS linear model and traditional nonlinear BP model,the PLS-GRNN model was found to have better effect.,and the Coefficient of Determination of Prediction(R2)of water,protein and fat reached respectively 0.9769,0.9402 and 0.9111,the Root Mean Square Error of Prediction(RMSEP)are 0.0912,0.3834 and 0.1134,and the relative standard deviations of the prediction(RSD)are 0.79%,0.83%and 8.53%.Although the prediction effect of fat is less effective,it is also within the available range of the evaluation criteria of the model.The results showes that PLS-GRNN based on near infrared spectroscopy is feasible for soybean meal quality detection,and could be used for quality control in the actual production process.