Rapid determination of rice protein content by near infrared spectroscopy

Abstract: A model was developed for the prediction of the protein content of rice samples with different grain sizes by near infrared spectroscopy (NIRS). The correlation coefficients between the model predicted values and the chemically determined values were O.94, and the mean standard deviation, mean relative error and mean absolute deviation of the predicted values were O.43, 2.1% and O.21, respectively, which solved the problem of the long detection time of the traditional method for the determination of protein.

 

The main nutrient in rice is protein, but high-protein rice is hard, light yellow, easy to deteriorate when stored, the rice is yellow-brown, so that the appearance and taste quality is reduced[1] . Therefore, the protein content is an important indicator for evaluating the quality of rice.

Protein is a complex nitrogen-containing organic compound, and the nitrogen content of different proteins is also different, so the Kjeldahl method is generally used for the determination of protein content, but it takes 7-8h to determine a sample with this method, and the detection time is long[2] . However, it takes 7~8h to measure a sample by this method, and the detection time is longer[2] . However, in general, the above methods require complicated pretreatment of the samples, cumbersome determination process, difficult to control the conditions, and large consumption of drugs, which is time-consuming and laborious. In this paper, near-infrared spectroscopy (NIRS) was used to determine the protein content of rice in order to shorten the detection time.

 

1 Test materials and methods

1.1 Sample collection

The samples used in this study were provided by the China Rice Research Institute (CRI) and covered typical varieties from major rice-producing regions in China. 97 samples were selected from 285 samples, including indica, japonica and glutinous rice varieties. The range of protein content was 7.6%~14.1%, and the wet basis water content was 9.8%.

 

1.2 Chemical analysis of test materials and determination of protein content

For the 97 samples, 3Og of rice was hulled and polished, and then milled in a cyclone mill (Tecator AB, Sweden) and sieved through a pore size of O.42mm to produce brown rice flour, which was packed into drying bottles and numbered for use.

Detection by Kjeldahl method. The sample is heated with concentrated sulphuric acid and a catalyst to nitrify the proteins, whereby the carbon and hydrogen are oxidized to carbon dioxide and water, and the organic nitrogen in the sample is converted to ammonia combined with sulphuric acid to form ammonium sulphate. Then distill with alkali to vaporize the ammonia, absorb it with boric acid and titrate with standard hydrochloric acid solution. The nitrogen content was determined from the standard acid consumption and multiplied by the conversion factor of nitrogen to protein to obtain the crude protein content. The frequency of protein distribution of rice samples is shown in Fig. 1.

 

1.3 Spectral Acquisition

The test materials and preparation methods used for spectral collection are described in the literature[3] . Before the spectral scanning, the refrigerated samples were dried in a drying oven at 75℃ for 12h, and then equilibrated in the spectrometer room for 24h after drying, so that the environmental conditions of the samples were consistent with those of the instrument. The resolution of WQF 4OON FT-NIR spectrometer (produced by Beijing Second Optical Instrument Factory) was adjusted to 4 cm- 1, and the number of scans was 64 times. In the working range of the spectrometer from 1OOOOcm-1 to 35OOcm-1, each sample was scanned 64 times, and the averaged spectral data were stored as the spectral data of the sample until the NIR spectra of all the samples were obtained.

 

1.4 Test methods

In the range of 1O24O~35OOcm-1, every 4cm-1 collects one data, each sample spectrum has 1685 data points, so the spectral data of all the samples will form a huge data array, which brings difficulties in analysis and processing, so the dimensionality reduction of the data array is a key step in the near-infrared spectral analysis technology. Therefore, the dimensionality reduction of the data array is one of the key steps in the near-infrared spectral analysis technique. The principal component regression method[4] is to select new variables with less dimensions from the wavelength space formed by the spectral data to replace the original variables on the premise of not losing the main spectral information. The ability to compress the independent variables is very high, which greatly reduces the raw data, improves the calculation speed, and eliminates the correlation between neighboring wavelengths, removes the noise, and solves the difficulty of analyzing due to the overlapping of spectral bands. In this experiment, principal component analysis (PCA) was used to establish a mathematical model between the protein content of the rice samples and the near-infrared (NIR) spectral data.

The 97 selected representative samples were divided into a calibration set and a correction set. The calibration set consisted of 60 samples to build the prediction model, and the correction set consisted of 37 samples to calibrate the model. Measure the near-infrared spectra of the 60 samples in the calibration set, and use principal component analysis to establish a quantitative mathematical model between the chemical measurements and the near-infrared spectral data; scan the near-infrared spectra of the 37 samples in the calibration set, and substitute the spectral values into the calibration numbers to obtain the predicted values of the samples, and use the correlation coefficients between the predicted values and the chemical measurements and the average relative error to measure the reliability of the model; if the calibration equations are stable and reliable, then the model can be used to calibrate the calibration equation. If the calibration equation is stable and reliable, the model can be used for the determination of unknown samples.

 

2 Spectral data processing

2.1 Analysis of protein content determination results of sample normalization spectra

Spectral normalization is the process of finding the highest and lowest coordinates of an entire spectrum, subtracting the lowest coordinate from all the data for that spectrum, and then taking the ratio of the maximum absorbance value to the difference between the highest and lowest coordinates as the spectral data for that spectrum. The purpose of this process is to make spectra with different coordinate limits comparable.

The near infrared spectral data of each sample were normalized, and then a prediction model was established by principal component analysis to obtain the prediction results shown in Table 1, and the near infrared spectral prediction results of the protein content of the O. 15mm grain size semolina samples are shown in Fig. 2.

 

The correlation coefficients between the predicted values of NIR spectroscopy and the chemically determined values of O. 15 mm grain size of semolina reached O. 94, which was a very good linear correlation. The standard deviation of the prediction and the mean relative error were relatively small, and the number of principal components needed to reach the above index was only seven. The predicted values of NIR spectroscopy for the three grain sizes of brown rice also showed good linear correlation with the chemically determined values, and their predicted standard deviations, mean relative errors and mean absolute deviations were not much different from those of semolina with the same grain sizes, which were O.47, 2.6% and O.24, respectively, and the predicted principal components were the same as those of semolina. The linear correlation between the predicted and analyzed values of protein content of rice samples was highly significant in the F-test (Table 2).

 

2.2 Analysis of the results of protein content determination by second-order derivative spectra of samples

During the spectral scanning process, the baseline shift and rotation of the scanned spectra may occur due to differences in instrumental parameter settings, sample loading methods, etc. By baseline shift, we mean a fixed deviation of the apparent absorbance value from the true value at any wavelength in the spectrum. The so-called baseline shift refers to a fixed deviation between the apparent absorbance value and the true value at any wavelength in the spectrum, while the baseline rotation refers to the change in spectral error that is linearly related to the wavelength. These will bring errors to the analytical results, and the second-order derivatives of differential spectra can be used to effectively eliminate the baseline shift and rotation of the spectrum. In order to compare the prediction effect of the model constructed from the normalized spectrum, the second-order derivative spectrum of the sample near-infrared spectrum is used for mathematical modeling.

 

The normalized spectrum was transformed into a second-order derivative spectrum, and the number of smoothing points was 21. Principal component regression analysis was performed on rice samples of three grain sizes (O. 15 mm, O. 2O mm, O. 28 mm) and two types (polished rice and brown rice), and the results are listed in Table 3.

 

It is clear from Tables 3 and 4 that the effect of the second-order derivative in removing the translations and rotations of the spectral baseline greatly eliminates the effect of differences in grain size, which can be seen from the fact that the standard deviation of the prediction, the mean absolute deviation, and the mean relative error, and the R2 of the evaluation indexes do not show significant variations in the values between different grain sizes, and the correlation coefficients are around O.9. This indicates that the second-order derivative spectrum can eliminate the effect of differences in grain size on the determination results. This shows that the second-order derivative spectrum can eliminate the effect of grain size differences on the results, so the second-order derivative spectrum should be the preferred spectral parameter based on the results of NIR spectroscopy for the prediction of the protein content of rice.

The predictive effect of the models was tested by F-test, which proved that the correlation between the predicted values of the models and the corresponding chemical measurements of the rice protein content was highly significant at the confidence level of α=O.O5, and the results are listed in Table 4.

 

3 Conclusion

(1)  The NIR spectra of three samples with different grain sizes (O.15 mm, O.20 mm and O.28 mm) and two different types of samples (polished rice and brown rice) were analyzed by NIR spectral analysis, and the NIR spectral data of the samples were calibrated and corrected for modeling. The calibration and calibration modeling of the NIR spectral data of the samples were carried out. The smaller the particle size of the samples, the better the predictive performance of the model. The correlation coefficient between the predicted and analyzed values of the refined rice model was O. 94, and the standard deviation, mean relative error and mean absolute deviation of the predictions were O.43, 2.1% and O.21, respectively.

 

(2) The calibration and calibration modeling of the near-infrared spectral data of the above samples were carried out by using the second order derivative as a spectral parameter. The results showed that the second order derivative spectrum was more effective than the normalized spectrum in predicting rice protein content in the case of inhomogeneous grain size. The results showed that the second-order derivative spectrum was more effective than the normalized spectrum in predicting the protein content of rice with uneven grain size. The second-order derivative spectrum was used to simplify the milling process by substituting the O.20 mm sample for the O.15 mm sample.

 

References:

[1] Mo Huidong.  Improvement of rice quality in China.   Chinese Agricultural Science , 1993, 26(3):8~24

[2] Liu JX , Wu SY , Fang RM . Application and prospect of near infrared spectroscopy in agricultural products. Journal of Jiangsu University of Science and Technology , 1998, 19(3): 1~5

[3] Liu J X , Wu S Y , Fang R M . Neural network based on near-infrared spectroscopy for the prediction of rice straight-chain amylose content. Journal of Agricultural Machinery , 2OO1, 32(2): 55~57

[4] Xu Lu. Methods of chemometrics. Beijing :Science Publishing House , 1995. 143~154

[5] Liu JX. Rice quality testing based on near infrared spectroscopy: [Doctoral dissertation]. Zhenjiang :Jiangsu University of Science and Technology , 1999. 66-72