Study On Soybean Quality Evaluation Technology Based On Near Infrared Spectroscopy

In order to explore the establishment of the optimal granularity of soybean quality indices（moisture,crude fat and protein）with Fourier Transform Near Infrared prediction model.Fourier transform infrared spectrometer and partial least squares（PLS）were applied,and treated with different spectra mathematical processing.Ninety different varieties of soybean samples were prepared to research the whole particles and pulverized particles（10 mesh,20 mesh,40 mesh,60 mesh and 80 mesh）,looking for soybean quality indices of the optimal modeling granularity.80 percentage（216）of the remaining 270 soybean samples were used as calibration samples to build the model.And the remaining 20 percentage（54）for external verification,validation set was used for the calibration models to predict the performance of evaluation,the result indicated:1.Ninety soybean samples were as the experimental materials,considering the modeling effect of soybean quality indices and the change of near infrared scanning sample size.According to the practical application,the near infrared model of the whole soybean grain moisture could be applied.The correlation coefficients of cross-validation(R_cv)is 0.971.For accurate determination of soybean each quality index,the near infrared model of the soybean must be processed through the crushing and screening.The best modeling effect was crushing over 40 mesh sieve and 60 mesh sieve of moisture,protein and crude fat,respectively.At this point,The correlation coefficients of cross-validation(R_cv)was 0.960,0.939 and 0.953,respectively.2.216 soybean samples were as the calibration set,comparison of different mathematical methods,determine the best modeling conditions of soybean quality indices ultimately,the abnormal values in the experiment were corrected simultaneously.The superiority of the evaluation model with internal cross validation by correcting the root mean square error of correction（RMSEC）,standard error of cross validation（SECV）and the correlation coefficients of cross-validation R_cv,the smaller the RMSEC and the SECV,the greater the R_cv.Model predictive performance is better when RMSEC and SECV are the smallest.The maximum of the correlation coefficients of cross-validation R_cv are soybean whole sample of moisture,soybean crush particles sieves 40 mesh of moisture,sieves 60 mesh of crude fat and sieves 60 mesh of protein,they are 0.964,0.965,0.949 and 0.941,respectively.3.The chemical measurements of validation set samples are comparing with the model predicted value,for the external validation of fifty-four soybean samples,the prediction coefficient of determination R² of near infrared detection model of whole grain soybean moisture content,soybean crush particles moisture content,crush particles soybean crude fat and crush particles soybean protein content are reaching 0.960,0.966,0.958 and 0.958,respectively.The above results showed that the Fourier transform-near infrared spectroscopy can be used to quickly detect part of the soybean quality.It could be used to detect the whole grain soybean moisture,but it need to crush and sieve a certain number mesh namely moisture,crude fat and protein crush and sieve 40 mesh,60 mesh and 60 mesh respectively when detect moisture,crude fat and protein precisely.