Method for detecting gelatinization characteristics of rice based on machine learning in combination with Fourier transform infrared spectroscopy

Abstract

The invention provides a method for detecting rice gelatinization characteristics based on machine learning in combination with Fourier transform infrared spectroscopy, which comprises the following steps: acquiring optical signals of different varieties of rice in a middle-infrared band and gelatinization characteristic values measured according to a traditional method, preprocessing the acquired optical signals by utilizing a smoothing (Savitzky-Golay) method; an evaluation model of rice gelatinization temperature, peak time, peak viscosity, minimum viscosity, cold glue viscosity, disintegration value, retrogradation value and subduction value is constructed based on machine learning and a characteristic wave band selection algorithm, and effective prediction of rice gelatinization characteristics is achieved only through optical characteristic parameters. According to the method, the gelatinization characteristics of the rice can be evaluated only by carrying out Fourier transform infrared spectrum acquisition on the rice, and technical support is provided for food processing enterprises to realize rapid, efficient and stable product quality monitoring. Therefore, a rapid, efficient and stable quality monitoring technical support is provided for food processing enterprises, so that the quality control in the production process is more convenient and accurate. Through the technology, an enterprise can adjust production parameters in time so as to ensure the consistency and excellent quality of products.

Description

Technical Field

[0001] The invention belongs to the technical field of spectral analysis and relates to the detection of rice gelatinization characteristic values ​​by applying Fourier transform infrared spectroscopy technology. Background Art

[0002] Rice is the main staple food for people in most parts of China and plays a dominant role. With the improvement of residents' living standards, consumers have higher requirements for food quality. The quality standard of edible rice is a comprehensive trait, which is generally divided into milling quality, appearance quality, cooking and taste quality, and nutritional quality. Among them, cooking and taste quality are directly related to consumers and determine the acceptability of rice. Traditional methods are very accurate in evaluating rice quality. However, these methods usually require complicated pretreatment, professional operation, and use a large amount of chemicals, making it difficult to achieve rapid detection of large-scale samples. In contrast, spectroscopy technology allows the spectral information of samples to be obtained with little or no damage to the samples. Therefore, spectroscopy technology is increasingly being used in food testing, and the use of vibrational spectroscopy technology to achieve real-time, rapid and accurate detection of rice quality has important scientific research value and application prospects.

[0003] Vibrational spectroscopy is an important means for rapid quality identification of agricultural products and processed foods. At present, common vibrational spectroscopy techniques related to rice include near infrared (NIR) spectroscopy, Fourier transform infrared (FTIR) spectroscopy, hyperspectral imaging (HIS), Raman spectroscopy, fluorescence spectroscopy (FS), terahertz spectroscopy and laser induced breakdown spectroscopy (LIBS). These spectral techniques have achieved the prediction of quality indicators such as moisture, starch, and freshness. However, accurate detection of rice quality characteristics (protein content, gelatinization characteristic value) based on vibrational spectroscopy has not been effectively achieved. The fundamental reason is that molecular spectra are mostly band spectra, and the spectral lines are easy to overlap, resulting in serious spectral interference, which needs to be improved by combining chemometrics. Different chemometric methods in some current studies have given unsatisfactory prediction results, which may be related to the small number of samples and the small differences between samples. Different spectral techniques have different principles and therefore have different prediction potentials in terms of rice quality attributes. The characteristic band screening method can greatly improve the accuracy of the model and help improve the generalization ability of the model. This will help promote the development of the rice industry. Summary of the invention

[0004] The purpose of the present invention is to provide a novel method for quickly detecting the characteristic value of rice gelatinization. By using Fourier transform infrared spectroscopy technology, the optical signal of rice is obtained and the characteristic parameters of rice gelatinization (gelatinization temperature, peak time, peak viscosity, minimum viscosity, cold gel viscosity, disintegration value, regeneration value, reduction value) are collected. -Golay) preprocessing, and the optical correlation system of rice gelatinization characteristic values ​​was constructed by using the continuous projection algorithm (SPA), competitive adaptive reweighted sampling algorithm (CARS), non-information variable elimination (UVE), interval combination optimization (ICO) feature band screening algorithm combined with partial least squares (PLS) and support vector machine (SVM) algorithm analysis, so as to realize the detection of rice gelatinization characteristics by rice optical properties.

[0005] The benefits of the present invention are as follows: (1) the variation law of the characteristic value of rice gelatinization and the optical property is explored, and a high-throughput optical database of rice is constructed; (2) a prediction model of the characteristic value of rice gelatinization by Fourier transform infrared spectroscopy is developed, and a characteristic band screening method is used to digitally evaluate the characteristic value of rice gelatinization; (3) a method for rapidly predicting the characteristic value of rice gelatinization is developed, and the detection steps of the characteristic value of rice gelatinization are simplified. Compared with the traditional determination of the characteristic value of rice gelatinization, the present invention saves time, reduces labor, and significantly reduces the determination cost; (4) only by rapidly characterizing the optical property of rice, the gelatinization property of rice is rapidly predicted, so that food enterprises can achieve rapid, efficient, and stable product quality control. BRIEF DESCRIPTION OF THE DRAWINGS

[0006] Figure 1 :Flowchart of the present invention

[0007] Figure 2 : Schematic diagram of rice varieties

[0008] Figure 3 :(a) Box plot of gelatinization characteristic values; (b) Frequency distribution of gelatinization temperature; (c) Frequency distribution of peak time

[0009] Figure 4 :Results of different feature screening algorithms in the mid-infrared band (a) SPA feature wavelength extraction process; (b) feature bands selected by the SPA algorithm; (c) CARS feature wavelength extraction process; (d) feature bands extracted by the CARS algorithm

[0010] Figure 5 :Results of different feature screening algorithms (a) UVE stability diagram; (b) Feature bands extracted by UVE algorithm; (c) Box plot of ICO algorithm root mean square error as a function of iteration number; (d) Feature bands extracted by ICO algorithm DETAILED DESCRIPTION

[0011] A method for detecting rice gelatinization characteristics based on machine learning combined with Fourier transform infrared spectroscopy, the specific implementation method is as follows:

[0012] 1. Experimental Materials and Methods

[0013] Rice sample preparation: 155 rice varieties were sown in mid-to-late May. Mature seeds were harvested in batches from mid-September to early October, 40 days after heading. Varieties with similar maturity were selected, and the collected seeds were naturally air-dried and placed indoors for one month of post-ripening. They were then husked and ground to obtain brown rice and polished rice. The rice grains were ground into fine powder using a vibrating ball mill GT300 and passed through a 100-mesh sieve for later use.

[0014] RVA spectrum determination: According to the methods in the National Standard of the People's Republic of China GB / T 24852-2010 and the AACC (American Association of Cereal Chemists) regulations (1995-61-02), the RVA-TecMaster rapid viscosity analyzer developed and produced by Perten of Sweden and the supporting analysis software TCW3 were used to measure the viscosity change curve over time and obtain the characteristic values ​​of the RVA spectrum. Turn on the rapid viscosity analyzer and preheat for 30 minutes, open the TCW3 control software, and load the set rice rapid analysis program. Measure (25±0.1)mL of water that meets the requirements of GB / T6682, use a 1 / 10,000 balance to accurately weigh (3.00±0.01)g of refined rice flour, transfer all the experimental materials to the matching aluminum sample cylinder, and quickly stir it until the sample is completely dispersed. Place the stirrer in the aluminum sample cylinder, transfer it to the copper Haff base, plug the connector into the card slot, and ensure that the cylinder is stably connected to the rapid viscosity analyzer. After ensuring that conditions permit, press down the motor cap of the instrument and run the test program. After the test is completed, the RVA spectrum and the 8 characteristic values ​​therein are exported, namely peak viscosity, minimum viscosity, cold gel viscosity, peak time, gelatinization temperature, disintegration value, reduction value and recovery value. Among them, the units of peak viscosity, minimum viscosity, cold gel viscosity, disintegration value, reduction value and recovery value are expressed in centipoise (cP), and the measurement results are retained in integers; the unit of peak time is minute (min), and the measurement result is accurate to 0.01; the unit of gelatinization temperature is degree Celsius (℃), and the measurement result is accurate to 0.01.

[0015] Fourier transform infrared spectroscopy and spectral data acquisition: Fourier transform attenuated total reflection infrared spectroscopy (ATR-FTIR) uses Nicoleti S10 spectrometer (Thermo Fisher Scientific, USA), which is equipped with attenuated total reflection (ATR) accessories and OMNIC software (Thermo Fisher Scientific Inc., USA). The design mode of software sampling workflow facilitates the management and control of the spectrometer. During the collection, a small amount of rice powder was placed on the surface of the ATR crystal window of the sample tray, and the pressure tower was rotated to press the powder downward to make it in good contact with the ATR crystal. The scanning number was 32 times, and the wave number scanning range was 525-4000cm-1 Resolution 4cm -1 , and background spectra were collected every 1 h.

[0016] Spectral data set: 465 spectra in total (155 varieties × 3 repeated measurements).

[0017] 2. Spectral preprocessing and model construction

[0018] use R2023a (The MathWorks, USA) was used to construct support vector machine (SVM) and partial least squares (PLS) models. Before building the model, 465 spectral data were smoothed (Savitzky-Golay) preprocessed. The calibration set and validation set were divided into a ratio of 3:1 using the random principle. For the calibration set of the model: 155 varieties of rice, including 349 mid-infrared spectra. For the validation set of the model: 155 varieties of rice, including 116 mid-infrared spectra. In addition, PLS and SVM regression models were constructed based on the results of continuous projection algorithm (SPA), competitive adaptive reweighted sampling algorithm (CARS), non-information variable elimination (UVE), and interval combination optimization (ICO) feature band screening. The root mean square error (RMSEV) of the validation set and the prediction accuracy (R c 2 ) and the prediction accuracy of the validation set (R v 2 ) was used to evaluate the prediction performance of the model.

[0019] 3. Analysis of changes in rice physical and chemical quality characteristics

[0020] like Figure 2 As shown in the figure, there is no significant difference in the appearance of rice among all varieties. However, its physical and chemical quality changes significantly ( Figure 3). As shown in Table 1, for the 8 characteristic values ​​of the RVA spectrum, the peak viscosity is as high as 5298cP, which reflects the expansion force of the starch molecules of this variety; the lowest viscosity appears after most of the starch granules are broken, reflecting the ability of the starch to resist shear at high temperature. The distribution of the lowest viscosity ranges from 1181cP to 3964cP, which shows that the shear resistance of different varieties of rice starch is significantly different; the cold gel viscosity reflects the ability of the starch paste to form a gel after cooling. The cold gel viscosity of different varieties of rice increases from 1798cP to 7548cP. The higher the cold gel viscosity, the stronger the ability of the starch to form a gel; the disintegration value (BDV) is the difference between the highest viscosity and the lowest viscosity. The standard deviation of the disintegration value of 155 portions of rice in this sample is 542.45, and the difference between species is significant. The reduction value is the difference between the cold gel viscosity and the peak viscosity. In the process of rice starch determination, it reflects the softness and hardness of the rice texture. It can be seen that some varieties have negative reduction values, indicating that their apparent amylose content is low, resulting in a hard taste of the rice. As Figure 3 As shown in the figure, the gelatinization temperature of most varieties is concentrated above 75℃. The gelatinization temperature is proportional to the cooking time and the amount of water absorbed, which is affected by the structure of amylopectin and the crystal characteristics of starch molecules. The peak time distribution range is 4.06-6.33min, which is roughly normal distribution.

[0021] Table 1 Changes in physical and chemical indicators of different rice varieties

[0022]

[0023] 4. Comparative analysis of the results of predicting rice gelatinization characteristic values ​​based on machine learning combined with characteristic band selection algorithm

[0024] In addition, the results of selecting mid-infrared spectral data based on different characteristic band selection algorithms are as follows: Figure 4 As shown in Figure 1, Figure a shows that as the number of variables increases, the root mean square error of the SPA algorithm gradually decreases until the number of variables reaches 10, and the model error no longer decreases. As shown in Figure b, a total of 10 characteristic wavelengths are selected. Figure c shows the feature selection process of the CARS algorithm. The cross-validation root mean square error first decreases and then increases with the increase in the number of sampling times, reaching the lowest value around 450 times. Finally, 119 characteristic bands are selected as shown in Figure d.

[0025] The results of UVE are as follows Figure 5As shown in a and b, it can be seen from Figure a that the red part is random noise and the blue part is the actual variable. By comparing the t value of the actual variable and the random variable, it is determined which variables have important contributions to the model. It can be seen from the figure that there is a clear distinction between the actual variable (blue line) and the random variable (red line), indicating that the actual variable in the figure has more significant contribution bands, and finally 976 characteristic bands were selected. The results of ICO are shown in Figures c and d. By the fifth iteration, the root mean square error of the cross-validation set began to rise and was close to the number of variables finally retained, and finally 1081 characteristic bands were screened out.

[0026] The results of predicting the gelatinization characteristic values ​​of rice based on the PLS and SVM algorithms combined with the characteristic band selection algorithm are shown in Table 2. The prediction effect of the gelatinization characteristic values ​​of rice based on the CARS-SVM model is significantly better than that based on the CARS-PLS model. In addition, among all the models for predicting gelatinization characteristic values, the CARS-SVM model is more effective in evaluating the cold gelatin viscosity (R v 2 =0.8548, RMSEV = 1067.1862), minimum viscosity (R v 2 =0.8531, RMSEV = 515.8572), reduction value (R v 2 =0.8599,RMSEV=933.1614), recovery value (R v 2 =0.8967, RMSEV=610.1677) and gelatinization temperature (R v 2 =0.9654, RMSEV=3.5137) all obtained excellent results, indicating that the prediction of rice gelatinization characteristic values ​​based on machine learning combined with feature screening algorithm has commercial application potential.

[0027] Table 2 Prediction of rice gelatinization characteristic values ​​based on PLS and SVM algorithms combined with characteristic band selection algorithm

[0028]

[0029] In summary, the prediction results of rice gelatinization characteristic values ​​based on machine learning technology combined with characteristic band selection algorithm show that by selecting a small number of key bands for modeling, not only the complexity of the model is greatly simplified, but also the prediction accuracy is significantly improved. This result shows that the use of optimized characteristic band screening strategy can accurately and quickly predict the gelatinization characteristic values ​​of rice. This study provides important support for the application of Fourier transform infrared spectroscopy technology in the rapid and non-destructive detection of rice gelatinization characteristic values, demonstrates its great potential and broad application prospects in actual detection, and helps to improve the detection efficiency in grain processing and quality control.

[0030] The foregoing has shown and described the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the above embodiments do not limit the present invention in any form, and all technical solutions obtained by means of equivalent replacement or equivalent transformation fall within the protection scope of the present invention.

Claims

1. A method for detecting rice gelatinization characteristics based on machine learning combined with Fourier transform infrared spectroscopy, comprising the following steps: S1. 155 rice varieties were sown in mid-to-late May. The mature seeds were harvested in batches from mid-September to early October, 40 days after heading. The varieties with similar maturity were selected, and the collected seeds were naturally air-dried and placed indoors for one month of post-ripening. Then, the rice was husked and ground to obtain polished rice. The rice grains were ground into fine powder using a vibrating ball mill and passed through a 100-mesh sieve for later use. S2. When collecting Fourier transform attenuated total reflection infrared spectroscopy (ATR-FTIR), a small amount of rice powder was placed on the surface of the ATR crystal window of the sample plate. The pressure tower was turned downward to press the powder to make it in good contact with the ATR crystal. The scanning number was 32 times, and the wave number scanning range was 525-4000cm -1 Resolution 4cm -1 , and background spectra were collected every 1 h. S3, RVA spectrum determination: According to the method in the National Standard of the People's Republic of China GB / T 24852-2010 and the AACC (American Association of Cereal Chemists) regulations (1995-61-02), the viscosity change curve over time is measured to obtain the characteristic values ​​of the RVA spectrum. Preheat the rapid viscosity analyzer for 30 minutes, load the rice rapid program, weigh water and refined rice flour into the sample tube, stir evenly and place it into the instrument, and start the test program. After the test is completed, the RVA spectrum and the 8 characteristic values ​​therein are derived, namely peak viscosity, minimum viscosity, cold gel viscosity, peak time, gelatinization temperature, disintegration value, reduction value and recovery value. S4. Construct support vector machine (SVM) and partial least squares (PLS) models. Before building the model, the 465 spectral data were smoothed (Savitzky-Golay) preprocessed. The calibration set and validation set were divided into a 3:1 ratio using the random principle. For the calibration set of the model: 155 varieties of rice, including 349 mid-infrared spectra. For the validation set of the model: 155 varieties of rice, including 116 mid-infrared spectra. The root mean square error (RMSEV) of the validation set and the prediction accuracy (R c 2 ) and the prediction accuracy of the validation set (R v 2 ) was used to evaluate the prediction performance of the model. S5. In addition, PLS and SVM regression models are constructed based on the results of continuous projection algorithm (SPA), competitive adaptive reweighted sampling algorithm (CARS), non-information variable elimination (UVE), and interval combination optimization (ICO) feature band screening. The root mean square error (RMSEV) of the validation set and the prediction accuracy (R c 2 ) and the prediction accuracy of the validation set (R v 2 ) was used to evaluate the prediction performance of the model.

Images