Study On Rapid Detection Of Wheat Parameters By Near Infrared Spectroscopy

To nondestructively analysis quality of the wheat in wheat grain, we put forward a method based on the near-infrared reflectance spectroscopy combined with A variety of qualitative and quantitative algorithm. The near-infrared reflectance spectra of 160 wheat samples were acquired and divided into calibration sets and prediction sets randomly. After calibration validation, we can test other ingredients of wheat with the best model.The outcome goes as followings:1 Establishing a quantitative analysis model of neutral network for wheat protein based on NIR.Firstly, the NIR reflectance spectra of 160 samples should be pretreated by different methods such as wavelet algorithm, detrended algorithm, 1nd Derivative,principal component analysis, we can see from the results that the best testing model was the one which combined principal component analysis with detrended algorithm. The correlation coefficient was 0.98 and the RMSEP was 0.26%. The other mathematical model of linear results are also good enough to prove the neutral network works well, the ratio R is between 0.95 and 0.98,while the RMSEP is between 0.26% and 0.30%.2 Establishing the quantitative analysis model of partial least for wheat protein based on NIR.Establishing the algorithm mathmatical model combined with different methods such as wavelet algorithm, detrended algorithm, 1ndDerivative, 2nd Derivative, cross-validation-leave one out to be used to detect the spectrum of wheat protein. A conclusion could be drawed from that the model combined cross-validation with wavelet method is best. the ratio R is 0.92,the RMSEP is 1.71%.the other mathmatical didn’t work well enough.3 Comparing the results come from partial least model and neutral network model.The conclusion shows that the prediction ability of neutral network model is obviously higher than partial least model. It indicates that the neutral network mathmatical model is more suitable for dealing with non-linear issue.4 Detecting the water and gluten content of wheat based on the existed model.The results show that wheat gluten gets the best ratio R 0.98 and RMSEP0.24% by using detrended algorithm, while wheat water gets the best R 0.96 and 0.31% by using wavelet algorithm.