Rice product raw material quality prediction method and device based on deep feature fusion

Abstract

The invention provides a rice product raw material quality prediction method and device based on deep feature fusion, and relates to the technical field of food quality detection and control. The method. The method comprises the following steps: equally dividing raw materials of the rice product into a plurality of groups according to the weight, extracting the raw materials with the same weight from each group as samples to be detected, detecting cracks, yellowing and insect pests in the samples by using a Japan bamboo particle evaluation instrument, and weighing; meanwhile, the ground sample is equally divided into three parts, the content of protein, moisture and amylose in the sample is measured through a Kjeldahl method, a drying oven drying method and an iodine colorimetric method respectively, the absorbance of the sample in a specific wave band is recorded through an infrared spectrum imaging technology, a prediction model of the content of protein, moisture and amylose is established based on a partial least square method, and the prediction model is used for predicting the content of protein, moisture and amylose. And constructing a quality evaluation system of the rice product raw materials with the proportions of cracks, yellowing and insect pests, and dividing the quality of the rice product raw materials into different grades of excellent, good, qualified and poor according to different evaluation standards.

Description

Technical Field

[0001] The present invention relates to the technical field of food quality detection and control, and specifically to a method and device for predicting the quality of rice product raw materials based on deep feature fusion. Background Art

[0002] Rice products, a widely consumed food category worldwide, encompass various forms such as rice, rice noodles, rice cakes, and sushi, and are beloved by consumers across different regions. With the improvement of living standards and the growing concern for food safety and quality, higher standards are being placed on the quality control of rice product raw materials. During the rice product production process, the quality of the raw materials directly determines key attributes such as the product's taste, nutritional value, appearance, and shelf life. For example, high-quality rice raw materials can produce rice with a soft, glutinous texture and rich aroma. In addition to taste requirements, rice used in sushi also has strict requirements for rice grain integrity and color to ensure the appearance and quality of the sushi. For rice noodles, appropriate amylose and protein content can influence the rice noodles' toughness and taste, preventing problems such as breakage and mushy noodles during cooking.

[0003] Although some technologies and methods exist for testing rice product raw materials, many problems remain. Traditional testing methods often rely on manual sensory judgment, such as visually observing cracked and yellowed grains. This method is highly subjective, and different testers use different judgment criteria, resulting in poor accuracy and repeatability of test results. In terms of component content testing, some conventional methods are cumbersome and time-consuming. For example, traditional protein content determination methods may require complex chemical digestion and titration processes, which cannot meet the needs of rapid testing. Moreover, most existing testing technologies focus on single indicators and lack a comprehensive analysis and evaluation system for multiple indicators such as protein, moisture, amylose content, and defective particles. This makes it difficult to fully and accurately reflect the overall quality of rice product raw materials. Faced with the complex and diverse rice product raw materials, existing testing and evaluation methods are unable to quickly and accurately provide manufacturers with comprehensive raw material quality information. This leads to a lack of scientific basis for raw material procurement, production process adjustments, and other aspects of the company, affecting the quality stability and production efficiency of rice products.

[0004] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not form the prior art that is already known to a person of ordinary skill in the art. Summary of the Invention

[0005] The purpose of the present invention is to provide a method for predicting the quality of rice product raw materials based on deep feature fusion to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] The method for predicting the quality of rice product raw materials based on deep feature fusion includes the following steps:

[0008] Step 1: Samples were selected from the rice product raw materials using a stratified sampling method. Spectral data were collected using infrared spectroscopy imaging technology to record the characteristic wavelengths and absorbances related to protein, moisture, and amylose.

[0009] Step 2: Determine the protein content of the sample using the Kjeldahl method, the moisture content using the oven drying method, and the amylose content using the iodine colorimetric method. Construct a table of the relevant contents using the characteristic wavelengths and absorbances associated with protein, moisture, and amylose.

[0010] Step 3: Extract the characteristic wavelength and absorbance of the sample from the relevant content table, and establish a prediction model for protein, moisture, and amylose content based on the partial least squares method;

[0011] Step 4: Use the stratified sampling estimation method to count the yellowed and insect-damaged grains in the rice product raw materials, and use the prediction model to calculate the protein, moisture and amylose content, and construct the rice product raw material quality evaluation formula to predict the quality.

[0012] Furthermore, the rice product raw materials were divided into 20 equal portions according to mass. For each portion of rice product raw materials, 10 samples were selected by simple random sampling and divided into 20 groups.

[0013] Furthermore, infrared spectroscopy imaging technology was used to focus on 900 to 3000 cm-1 for 20 groups of rice product raw material samples. -1 Band, recording band 1000, 1500, 1600, 1700, 3200cm -1 The absorbance I λ , where λ = 1000, 1500, 1600, 1700 and 3200, and the characteristic wavelengths of proteins are set to 1500 and 1700 cm -1 The characteristic wavelengths of water are 1600 and 3200 cm -1 The characteristic wavelength of amylose is 1000 cm -1 .

[0014] Furthermore, each group of rice product raw material samples was ground and divided equally into three parts of mass A. The protein content of the samples was determined by the Kjeldahl method, the moisture content was determined by the oven drying method, and the amylose content was determined by the iodine colorimetric method.

[0015] The method for determining the protein content of samples by Kjeldahl method is:

[0016] Place a rice product raw material sample with a mass of A (in grams) into a dry Kjeldahl flask, add a mixture of 3g potassium sulfate and 1g copper sulfate and 20ml concentrated sulfuric acid, and continue heating until the liquid becomes blue-green, clear and transparent. Continue heating for 0.5 hour. After the Kjeldahl flask cools down, connect it to a Kjeldahl nitrogen determination distillation apparatus, add 20g / L boric acid solution and bromocresol green-methyl red mixed indicator to another conical flask, place the conical flask under the condenser of the distillation apparatus, and insert the lower end of the condenser below the liquid level of the boric acid solution. Add 400g / L sodium hydroxide solution to the Kjeldahl flask until the volume of the distillate, that is, the boric acid solution, reaches 100ml. Titrate the boric acid solution with 0.1mol / L hydrochloric acid standard solution. Stop when the solution changes from green to gray-red. At the same time,

[0017] Take another Kjeldahl flask, add a mixture of 3g potassium sulfate, 1g copper sulfate and 20ml concentrated sulfuric acid and continue heating until the liquid is blue-green, clear and transparent, and continue heating for 0.5 hour. After the Kjeldahl flask is cooled, it is connected to a Kjeldahl nitrogen determination distillation apparatus. Take another Erlenmeyer flask and add 20g / L boric acid solution and bromocresol green-methyl red mixed indicator. The Erlenmeyer flask is placed under the condenser of the distillation apparatus so that the lower end of the condenser is inserted below the liquid level of the boric acid solution. 400g / L sodium hydroxide solution is added to the Kjeldahl flask until the distillate, i.e., the volume of the boric acid solution reaches 100ml. The boric acid solution is titrated with 0.1mol / L hydrochloric acid standard solution until the solution becomes gray-red and stops. The volume of the hydrochloric acid standard solution consumed is recorded, i.e., the volume of the hydrochloric acid standard solution consumed in the blank test;

[0018] Calculation of nitrogen content:

[0019]

[0020] Where X represents the nitrogen content in the sample, c represents the concentration of the hydrochloric acid standard solution, V1 represents the volume of the hydrochloric acid standard solution consumed when titrating the sample, V0 represents the volume of the hydrochloric acid standard solution consumed in the blank experiment, m represents the mass of the sample, and 0.014 represents the molar mass of nitrogen;

[0021] Calculated protein content:

[0022] C p =X×6.25

[0023] Where C p Expressed as protein content;

[0024] The method for determining moisture content by oven drying is:

[0025] Select a weighing bottle with a mass of m1, place a rice product raw material sample with a mass of A into the weighing bottle, and record the total mass as m2. Place the weighing bottle containing the rice product raw material sample in an oven at 105°C for 3 hours. After drying, remove the weighing bottle and record the mass of the rice product raw material sample and the weighing bottle as m3. Calculate the moisture content:

[0026]

[0027] Where C w Expressed as moisture content;

[0028] The method for determining the content of amylose by iodine colorimetry is as follows:

[0029] Weigh 0.1000g of amylose standard, add 1ml of anhydrous ethanol, then add 9.0mL of 1mol / L sodium hydroxide solution, heat in a boiling water bath for 10 minutes, cool and transfer to a 100mL volumetric flask, dilute to the mark with 1.0mg / mL distilled water, and then pipette a set of volume data solutions starting from 0 and increasing in increments of 1.0mL to 10.0mL into a 100mL volumetric flask, dilute to the mark with distilled water, and dilute to a concentration starting from 0 and increasing in increments of 10.0μg / mL. Prepare a series of standard solutions with concentration data of 100 μg / mL to 100.0 μg / mL. Weigh the amylopectin standard and prepare a solution with a concentration of 1.0 mg / mL according to the preparation method of the above amylose standard solution. Then dilute it to a series of standard solutions with different concentrations according to the same gradient. Weigh 2.0 g of potassium iodide and dissolve it in a small amount of distilled water. Add 0.2 g of iodine and, after complete dissolution, dilute to 100 mL with distilled water. Store in a brown bottle. Weigh 40 g of 1 mol / L sodium hydroxide solution and dissolve it in distilled water. After cooling, dilute to 1000 mL.

[0030] A rice product raw material sample of mass A was placed in a 100 mL conical flask, 10 mL of anhydrous ethanol was added and stirred evenly, the mixture was soaked for 1 hour, filtered, and cooled to room temperature. The solution was then transferred to a 100 mL volumetric flask, diluted to the mark with distilled water, and shaken to obtain a rice product raw material sample solution. 2.0 mL of amylose standard solution of different concentrations was respectively drawn into a 25 mL colorimetric tube, 1.0 mL of iodine reagent was added to each solution, and the solution was diluted to the mark with distilled water. The solution was shaken and allowed to stand for 15 minutes. Distilled water was used as a blank control, and the absorbance of each standard solution was measured at a wavelength of 620 nm using a spectrophotometer. A standard curve was plotted with amylose concentration as the abscissa and absorbance as the ordinate to obtain the amylose regression equation.

[0031] 2.0 mL of the rice product raw material sample solution was drawn into a 25 mL colorimetric tube. Iodine reagent was added and the volume was adjusted according to the steps for drawing the standard curve. The solution was shaken and allowed to stand for 15 minutes to obtain a test rice product raw material sample. The absorbance of the sample solution was measured at a wavelength of 620 nm. A blank test was then performed according to the sample measurement steps without adding the sample, and the blank absorbance at a wavelength of 620 nm was measured.

[0032] At a wavelength of 620 nm, the difference between the absorbance of the rice product raw material sample solution and the blank absorbance was calculated, and the amylose concentration was obtained by comparing it with the amylose regression equation, and the dilution factor was calculated:

[0033]

[0034] Where n represents the dilution multiple;

[0035] The calculation formula for amylose content is:

[0036]

[0037] Where C Al Expressed as amylose content, c al It is expressed as the amylose concentration obtained from the amylose regression equation in ug / ml, and V is expressed as the solution volume of the test rice product raw material sample in ml.

[0038] Furthermore, a univariate linear regression equation is constructed:

[0039] y=a+bx

[0040] Where y represents absorbance, x represents amylose concentration, a represents intercept, and b represents slope;

[0041] By experimentally measuring the absorbance of different concentrations of amylose standard solutions, a series of corresponding data points (x1, y1), (x2, y2), (x 11 , t 11 ), recorded as (x i ,y i );

[0042] Calculate the mean value of amylose concentration:

[0043]

[0044] Where, It is expressed as the mean value of amylose concentration;

[0045] Calculate the average absorbance of different concentrations of amylose standard solutions:

[0046]

[0047] Where, It is expressed as the average absorbance of standard solutions of amylose at different concentrations;

[0048] Calculate the slope of a linear regression equation:

[0049]

[0050] Compute the intercept of a univariate linear regression equation:

[0051]

[0052] Furthermore, based on the extracted 20 sets of data, each set of data contains the absorbance of 5 characteristic wavelengths, and the dimension of the absorbance data matrix is set to 20×5:

[0053]

[0054] Where X represents the absorbance data matrix;

[0055] Set the dependent variable matrix to 20×3:

[0056]

[0057] Where Y represents the dependent variable matrix;

[0058] Calculate the mean of columns 1, 2, and 3 of the dependent variable matrix Y:

[0059]

[0060] Where, and They are respectively represented as the average values of the 1st, 2nd and 3rd columns of the dependent variable matrix Y, and h is represented as the row index of the matrix;

[0061] Calculate the standard deviation of columns 1, 2, and 3 of the dependent variable matrix Y:

[0062]

[0063] Where s1, s2 and s3 represent the standard deviations of the first, second and third columns of the dependent variable matrix Y.

[0064] Standardize the absorbance data matrix X and the dependent variable matrix Y:

[0065]

[0066] Where g hjRepresented as original data, where j represents the column index of the matrix, j∈(j dy ,j dr ), j dy Expressed as the column index of the absorbance data matrix X, j dr Represented as the column index of the dependent variable matrix Y, Expressed as the average value of the jth column, s j Expressed as the standard deviation of the jth column, the standardized results are substituted back into the original absorbance data matrix X and dependent variable matrix Y;

[0067] The mean value of column j is:

[0068]

[0069] The standard deviation of the jth column is

[0070]

[0071] Calculate the covariance matrix of the absorbance data matrix X and the dependent variable matrix Y:

[0072]

[0073] Where Cov(X, Y) represents the covariance matrix of the absorbance data matrix X and the dependent variable matrix Y;

[0074] Assume that the process matrix with dimension q×q is:

[0075] B=YY T

[0076] Where B represents the process matrix;

[0077] The process 2 matrix with dimension p×p is:

[0078] C=X T YY T X

[0079] Where C represents the process two matrix;

[0080] Compute eigenvalues and eigenvectors:

[0081] Since CW=λ′W, we have:

[0082] (C-λ′I)W=0

[0083] Where λ′ is the eigenvalue, W is the eigenvector, and I is the identity matrix with dimension p×p;

[0084] Construct the characteristic equation of the process two matrix:

[0085] det(C-λ′I)=0

[0086] Where det(C-λ′I) is the characteristic equation of the process 2 matrix, and p eigenvalues λ′1, λ′2, …, λ′ are calculated. p ;

[0087] For each eigenvalue λ i′ , solve the homogeneous linear equations (C-λ′ i′ I) W = 0, get the corresponding eigenvector W i′ ;

[0088] Compare p eigenvalues λ′1, λ′2, …, λ′ p The size of the eigenvalue is found, which is recorded as λ′ max , and its corresponding eigenvector is recorded as W1;

[0089] After that, calculate the number of principal components to be extracted:

[0090] Assuming the number of principal components is k, the variance contribution rate of the i′th principal component is:

[0091]

[0092] Where, v i′ Expressed as the variance contribution rate of the i′th principal component;

[0093] The calculation formula for the cumulative variance contribution rate of the first k principal components is:

[0094]

[0095] Starting from the first principal component, calculate the cumulative variance contribution rate v1, v2, ..., v p , when it reaches or exceeds 0.85 for the first time, k is the number of principal components determined;

[0096] Compute the first component of the absorbance data matrix X:

[0097] t1=X·W1

[0098] Where X represents the first component of the absorbance data matrix X;

[0099] Calculate the regression coefficient of the dependent variable Y on t1:

[0100]

[0101] Where C1 represents the regression coefficient of the dependent variable Y on t1;

[0102] Calculate the first component of the dependent variable matrix Y:

[0103] u1=t1C1

[0104] Where u1 represents the first component of the dependent variable matrix Y;

[0105] Calculate the residual matrix for the absorbance data matrix X:

[0106] E1=X-t1P1

[0107] Where E1 represents the residual matrix of the absorbance data matrix X, P1 represents the first load vector, and the calculation formula is:

[0108]

[0109] Calculate the residual matrix of the dependent variable matrix Y:

[0110] F1=Y-t1C1 T

[0111] Where F1 is the residual matrix of the dependent variable matrix Y;

[0112] Repeat the above steps for the residual matrix E1 of the absorbance data matrix X and the residual matrix F1 of the dependent variable matrix Y until the kth (K>1) pair of components t k and u k :

[0113] Calculate E k-1 , F k-1 The covariance matrix of :

[0114]

[0115] Where Cov(E k-1 , F k-1 ) is expressed as F k-1 , F k-1 The covariance matrix of

[0116] Calculate the residual matrix E at the kth iteration k-1 and F k-1 The cross-covariance characteristic matrix of :

[0117] D=(E k-1 )TF k-1 (F k-1 )TE k-1

[0118] Where D is the cross-covariance feature matrix;

[0119] Perform eigendecomposition on it and select the eigenvector corresponding to the maximum eigenvalue as W k , calculate E k-1 The kth component of:

[0120] t k =E k-1 W k

[0121] Calculate F k-1 t k The regression coefficient of :

[0122]

[0123] Where C k Indicated as F k-1 t k The regression coefficient of

[0124] Calculate F k-1 The kth component of:

[0125] u k =t k C k

[0126] Calculate the residual matrix of the k-th extracted component:

[0127] E k =E k-1 -t k P k T

[0128] Where, E k It is expressed as the residual matrix of the k-th extracted component, P k is represented as the kth load vector, where

[0129]

[0130] Calculate the residual matrix 2 of the k-th extracted component:

[0131] F k =F k-1 -t k C k T

[0132] Where, F k It is represented as the residual matrix 2 of the k-th extracted component;

[0133] Construct the composition matrix:

[0134] T cf =[t1, t2, ..., t k ]

[0135] Where, T cf Represented as a component matrix;

[0136] Construct the regression coefficient matrix:

[0137] C hg =[C1, C2, ..., C k ] T

[0138] Where C hg Expressed as a regression coefficient matrix;

[0139] Construct a standardized regression model:

[0140]

[0141] Due to t i′ =XW i′ , set W′=[W1,W2,…,W k ]

[0142] The standardized prediction model is:

[0143]

[0144] The normalized prediction model Y′ is denormalized, and the prediction model of protein is:

[0145]

[0146] Where y h1 Represented as a prediction model for proteins;

[0147] The prediction model for moisture is:

[0148]

[0149] Where y h2 It is expressed as a prediction model of moisture;

[0150]

[0151] Where y h3 Expressed as a prediction model for amylose.

[0152] Further, the rice product raw material is divided into N new The same weight WT of rice raw materials were taken from each group as the test samples. The cracks, yellowing and insect-damaged grains in each sample of rice raw materials were identified using the Japanese Satake Grain Rating Instrument. The identified cracked grains, yellowing grains and insect-damaged grains were weighed separately and their respective weights wT were recorded. id , where id = 1, 2, ... N new ;

[0153] Calculate the proportion of cracked kernels, yellowed kernels, and insect-damaged kernels in each sampling group:

[0154]

[0155] Where p id Expressed as the percentage of cracked, yellowed, and insect-damaged grains in group id;

[0156] Calculate the proportion of cracked, yellowed, and insect-damaged rice grains in the raw materials for rice products:

[0157]

[0158] Where, P total It is expressed as the proportion of cracked grains, yellowed grains and insect-damaged grains in the raw materials of rice products.

[0159] Furthermore, for N new The method for predicting the content of the τth species (τ=1 represents albumin, τ=2 represents protein, and τ=3 represents amylose) is as follows:

[0160] The extracted bands are 1000, 1500, 1600, 1700, and 3200 cm -1 The absorbance is constructed with dimension N new ×5 absorbance data matrix X to be detected new :

[0161] The absorbance data matrix X to be detected new To perform standardization:

[0162]

[0163] Where h new Represented as row index, j new Represented as a column index, It is represented as the hth matrix of the absorbance data to be detected after normalization new OK, J new Column data, get the standardized absorbance data matrix X * new ;

[0164] The calculation dimension is N new The projection matrix of the new sample ×k in the direction of the extracted k principal components:

[0165] M new =X * new W′

[0166] Where M new Represented as the projection matrix of the new sample in the direction of the extracted k principal components;

[0167] Among them, the projection of the new sample in the direction of the principal component after standardization is:

[0168]

[0169] It is represented as the projection of the new sample in the direction of the principal component after standardization, l new Represents the index of the principal component, where l new =1,2,…,k, Indicated as X * new Middle h new Row j new Elements of the column index, Denote the jth new Row 1 new Elements of column index;

[0170] Z new =M new (C hg )^

[0171] Where Z new Indicated as N new ×3 matrix, (C hg )^ represents the estimated value of the regression coefficient matrix with dimension k×3;

[0172] Calculate the hth new The content-normalized intermediate prediction value of a new sample at τ:

[0173]

[0174] Where, Expressed as h new The standardized intermediate predicted value of the τth content of the samples to be tested, c lτ Expressed as (C hg )^ in the first new The element in row and column τ;

[0175] Calculate the hth new The formula for the standardized predicted value of the τth content is obtained after principal component projection and regression coefficient calculation of the samples to be tested:

[0176]

[0177] Where Y hτ ′ represents the hth new The standardized predicted value of the content of the τth component in the samples to be tested, Expressed as h new The residual error of the sample to be tested on the prediction of the content of the τth component is 0;

[0178] Substitute into the prediction model, then the h new The prediction formula for the content of the τth component of a new sample is:

[0179]

[0180] Where, Expressed as h new The predicted value of the content of the τth component of a new sample, Y sτ Expressed as the standard deviation of the τth column in the dependent variable matrix Y, Expressed as the mean value of the τth column in the dependent variable matrix Y;

[0181] The protein content of rice product raw materials is:

[0182]

[0183] Where dω1 represents the protein content of rice product raw materials;

[0184] The moisture content of rice product raw materials is:

[0185]

[0186] Where dω2 represents the moisture content of the rice product raw material;

[0187] The amylose content of rice product raw materials is:

[0188]

[0189] Where dω2 represents the amylose content of the rice product raw material.

[0190] Furthermore, a quality assessment formula for rice product raw materials was constructed based on the protein, moisture, and amylose content of the rice product raw materials and the proportion of cracked grains, yellowed grains, and insect-damaged grains:

[0191]

[0192] Where Q is the quality evaluation formula for rice product raw materials, ε1, ε2, and ε3 are weight coefficients, and ε1+ε2+ε3=1, and β is the adjustment parameter;

[0193] The quality thresholds are set to Q1, Q2 and Q3, and Q1>Q2>Q3>0. When Q>Q1, the quality of the rice product raw material is evaluated as excellent; when Q1≥Q>Q2, the quality of the rice product raw material is evaluated as good; when Q2≥Q>Q3, the quality of the rice product raw material is evaluated as qualified; when Q3≥Q, the quality of the rice product raw material is evaluated as poor.

[0194] Compared with the prior art, the present invention has the following beneficial effects:

[0195] The present invention incorporates key indicators of rice product raw materials, such as protein, moisture, amylose content, as well as cracked grains, yellowed grains, and insect-infested grains, into a unified evaluation system. Through scientific testing processes and professional equipment, it is possible to comprehensively and accurately grasp the various characteristics of the raw materials, avoiding the limitations of traditional single-indicator testing. The content prediction model constructed based on the partial least squares method has powerful data processing capabilities and can deeply analyze the complex nonlinear relationship between the protein, moisture, and amylose content in rice product raw materials and numerous influencing factors, thereby accurately assessing the degree of influence of each factor on protein, moisture, and amylose content, providing strong support for accurate prediction. By strictly controlling the quality of raw materials, it provides a stable and reliable raw material base for the production of rice products, ensuring that the product maintains stable high quality in terms of taste, nutritional value, appearance, and shelf life.

[0196] The present invention further provides a device for predicting the quality of rice product raw materials based on deep feature fusion, wherein the device is used to perform the above-mentioned method for predicting the quality of rice product raw materials based on deep feature fusion, comprising:

[0197] A spectral data acquisition module is used to select samples from rice product raw materials using a stratified sampling method, collect spectral data using infrared spectral imaging technology, and record characteristic wavelengths and absorbances related to protein, moisture, and amylose;

[0198] The content determination module is used to determine the protein content of the sample using the Kjeldahl method, the moisture content using the oven drying method, and the amylose content using the iodine colorimetric method. The characteristic wavelengths and absorbances associated with protein, moisture, and amylose are used to construct a table of related contents.

[0199] The model building module is used to extract the characteristic wavelength and absorbance of the sample from the relevant content table and establish a prediction model for protein, moisture and amylose content based on the partial least squares method;

[0200] The quality prediction module is used to use the stratified sampling estimation method to count the yellowed grains and insect-damaged grains in rice product raw materials, and use the prediction model to calculate the protein, moisture and amylose content, and construct the rice product raw material quality evaluation formula to predict the quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0201] Figure 1 Schematic diagram of the overall method flow of the present invention;

[0202] Figure 2 It is a schematic diagram of the overall system module of the present invention. DETAILED DESCRIPTION

[0203] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to specific embodiments.

[0204] It should be noted that, unless otherwise defined, the technical or scientific terms used in the present invention should have the usual meanings understood by people with ordinary skills in the field to which the present invention belongs. The "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative position relationships. When the absolute position of the object being described changes, the relative position relationship may also change accordingly.

[0205] Example:

[0206] See also Figure 1 , the present invention provides a technical solution:

[0207] The method for predicting the quality of rice product raw materials based on deep feature fusion includes the following steps:

[0208] Step 1: Samples were selected from the rice product raw materials using a stratified sampling method. Spectral data were collected using infrared spectroscopy imaging technology to record the characteristic wavelengths and absorbances related to protein, moisture, and amylose.

[0209] Use precise weighing equipment to divide the rice product raw materials into 20 equal parts by mass. The mass of each part is exactly the same to ensure the consistency and fairness of sampling. For example, if the total mass of the rice product raw materials is 100 kg, then the mass of each part is 5 kg. For each part of the rice product raw materials, 10 samples are selected by simple random sampling. The 10 samples are divided into one group, so a total of 20 groups of samples are obtained.

[0210] 1500cm -1 There are a large number of amide bonds in protein molecules. In the infrared spectrum, the NH bending vibration and CN stretching vibration of the amide bond are coupled to produce a characteristic absorption peak, which is called the amide II band and can be used to characterize the presence and content of proteins. -1Nearby, the stretching vibration of the carbonyl group (C=O) in the protein will produce an absorption peak, called the amide I band. Due to the interaction between different amino acid residues in the protein molecule and factors such as the secondary structure of the protein, the position of the amide I band will change. It is one of the important characteristic wavelengths of the protein.

[0211] 1600cm -1 Near 3200cm, the HOH bending vibration in water molecules will produce an absorption peak. This is because the water molecules will cause changes in the molecular dipole moment in this vibration mode, thereby absorbing infrared light of the corresponding wavelength. This is a characteristic absorption wavelength of water. -1 Nearby: The OH stretching vibration in water molecules will produce a strong absorption peak. Due to the hydrogen bonding between water molecules, the absorption peak of OH stretching vibration is broadened and the position is shifted, 3200cm -1 There is obvious absorption at this wavelength, which can be used as an important characteristic wavelength for detecting moisture;

[0212] 1000cm -1 Amylose is a linear polysaccharide composed of glucose units connected by α-1,4-glycosidic bonds. In the infrared spectrum, amylose is -1 There is a relatively unique absorption peak, which is related to the structure of the glucose unit in the amylose molecule and the vibration characteristics of the bond. This absorption peak can be used for the qualitative and quantitative analysis of amylose.

[0213] Therefore, infrared spectroscopy imaging technology was used to focus on 900 to 3000 cm-1 for 20 groups of rice product raw material samples. -1 Band, resolution set to 8cm -1 , recording bands 1000, 1500, 1600, 1700, 3200cm -1 The absorbance I λ , where λ = 1000, 1500, 1600, 1700 and 3200, and the characteristic wavelengths of proteins are set to 1500 and 1700 cm -1 The characteristic wavelengths of water are 1600 and 3200 cm -1 The characteristic wavelength of amylose is 1000 cm -1 .

[0214] Step 2: Determine the protein content of the sample using the Kjeldahl method, the moisture content using the oven drying method, and the amylose content using the iodine colorimetric method. Construct a table of the relevant contents using the characteristic wavelengths and absorbances associated with protein, moisture, and amylose.

[0215] The Kjeldahl method, based on the relatively stable nitrogen content of protein, converts the organic nitrogen in the sample into inorganic nitrogen and then performs quantitative analysis to calculate the protein content. This method, with its high accuracy and precision, is a classic method for protein determination and is widely recognized and applied. The oven drying method, based on the principle of water evaporation under certain temperature and time conditions, removes water from the sample and calculates the moisture content based on the difference in mass before and after drying. Amylose forms a specific blue complex with iodine, the color of which is proportional to the amylose content. The iodine colorimetric method utilizes this characteristic to calculate the amylose content by colorimetrically measuring absorbance. This method has a high specificity for amylose and can accurately determine the amylose content in rice product raw materials without interference from other polysaccharides.

[0216] Each group of rice product raw material samples is divided equally into 3 parts of mass A after grinding. This is because rice product raw materials may have differences in particle size, component distribution, etc., and even after grinding, it is difficult to ensure complete uniformity. Dividing each sample equally can ensure to the greatest extent that the samples used for different index measurements come from the "same" sample with the same or similar components, structure and characteristics, and reduce the fluctuation of results caused by differences in the samples themselves. For example, if equal division is not performed, the sample used for protein measurement may contain more particles with high protein content, while the sample used for amylose measurement may contain fewer such particles, which will make the measurement results unable to truly reflect the situation of the entire rice product raw material. The protein content of the samples is determined by the Kjeldahl nitrogen method, the moisture content is determined by the oven drying method, and the amylose is determined by the iodine colorimetric method.

[0217] The method for determining the protein content of samples by Kjeldahl method is:

[0218] Place a rice product raw material sample with a mass of A (in grams) into a dry Kjeldahl flask, add a mixture of 3g potassium sulfate and 1g copper sulfate and 20ml concentrated sulfuric acid, and continue heating until the liquid becomes blue-green, clear and transparent. Continue heating for 0.5 hour. After the Kjeldahl flask cools down, connect it to a Kjeldahl nitrogen determination distillation apparatus, add 20g / L boric acid solution and bromocresol green-methyl red mixed indicator to another conical flask, place the conical flask under the condenser of the distillation apparatus, and insert the lower end of the condenser below the liquid level of the boric acid solution. Add 400g / L sodium hydroxide solution to the Kjeldahl flask until the volume of the distillate, that is, the boric acid solution, reaches 100ml. Titrate the boric acid solution with 0.1mol / L hydrochloric acid standard solution. Stop when the solution changes from green to gray-red. At the same time,

[0219] Take another Kjeldahl flask, add a mixture of 3g potassium sulfate, 1g copper sulfate and 20ml concentrated sulfuric acid and continue heating until the liquid is blue-green, clear and transparent, and continue heating for 0.5 hour. After the Kjeldahl flask is cooled, it is connected to a Kjeldahl nitrogen determination distillation apparatus. Take another Erlenmeyer flask and add 20g / L boric acid solution and bromocresol green-methyl red mixed indicator. The Erlenmeyer flask is placed under the condenser of the distillation apparatus so that the lower end of the condenser is inserted below the liquid level of the boric acid solution. 400g / L sodium hydroxide solution is added to the Kjeldahl flask until the distillate, i.e., the volume of the boric acid solution reaches 100ml. The boric acid solution is titrated with 0.1mol / L hydrochloric acid standard solution until the solution becomes gray-red and stops. The volume of the hydrochloric acid standard solution consumed is recorded, i.e., the volume of the hydrochloric acid standard solution consumed in the blank test;

[0220] Calculation of nitrogen content:

[0221]

[0222] Where X represents the nitrogen content in the sample, c represents the concentration of the hydrochloric acid standard solution, V1 represents the volume of the hydrochloric acid standard solution consumed when titrating the sample, V0 represents the volume of the hydrochloric acid standard solution consumed in the blank experiment, m represents the mass of the sample, and 0.014 represents the molar mass of nitrogen;

[0223] Calculated protein content:

[0224] C p =X×6.25

[0225] Where C p Expressed as protein content. Protein is a biological macromolecule composed of amino acids connected by peptide bonds. Generally speaking, the nitrogen content in protein is relatively stable, averaging about 16%. This is a statistical average obtained after analyzing and measuring a large number of proteins from different sources and types.

[0226] The method for determining moisture content by oven drying is:

[0227] Select a weighing bottle with a mass of m1, place a rice product raw material sample with a mass of A into the weighing bottle, and record the total mass as m2. Place the weighing bottle containing the rice product raw material sample in an oven at 105°C for 3 hours. After drying, remove the weighing bottle and record the mass of the rice product raw material sample and the weighing bottle as m3. Calculate the moisture content:

[0228]

[0229] Where C w Expressed as moisture content;

[0230] The method for determining the content of amylose by iodine colorimetry is as follows:

[0231] Weigh 0.1000g of amylose standard, add 1ml of anhydrous ethanol, then add 9.0mL of 1mol / L sodium hydroxide solution, heat in a boiling water bath for 10 minutes, cool and transfer to a 100mL volumetric flask, dilute to the mark with 1.0mg / mL distilled water, and then pipette a set of volume data solutions starting from 0 and increasing in increments of 1.0mL to 10.0mL into a 100mL volumetric flask, dilute to the mark with distilled water, and dilute to a concentration starting from 0 and increasing in increments of 10.0μg / mL. Prepare a series of standard solutions with concentration data of 100 μg / mL to 100.0 μg / mL. Weigh the amylopectin standard and prepare a solution with a concentration of 1.0 mg / mL according to the preparation method of the above amylose standard solution. Then dilute it to a series of standard solutions with different concentrations according to the same gradient. Weigh 2.0 g of potassium iodide and dissolve it in a small amount of distilled water. Add 0.2 g of iodine and, after complete dissolution, dilute to 100 mL with distilled water. Store in a brown bottle. Weigh 40 g of 1 mol / L sodium hydroxide solution and dissolve it in distilled water. After cooling, dilute to 1000 mL.

[0232] A rice product raw material sample of mass A was placed in a 100 mL conical flask, 10 mL of anhydrous ethanol was added and stirred evenly, the mixture was soaked for 1 hour, filtered, and cooled to room temperature. The solution was then transferred to a 100 mL volumetric flask, diluted to the mark with distilled water, and shaken to obtain a rice product raw material sample solution. 2.0 mL of amylose standard solution of different concentrations was respectively drawn into a 25 mL colorimetric tube, 1.0 mL of iodine reagent was added to each solution, and the solution was diluted to the mark with distilled water. The solution was shaken and allowed to stand for 15 minutes. Distilled water was used as a blank control, and the absorbance of each standard solution was measured at a wavelength of 620 nm using a spectrophotometer. A standard curve was plotted with amylose concentration as the abscissa and absorbance as the ordinate to obtain the amylose regression equation.

[0233] 2.0 mL of the rice product raw material sample solution was drawn into a 25 mL colorimetric tube. Iodine reagent was added and the volume was adjusted according to the steps for drawing the standard curve. The solution was shaken and allowed to stand for 15 minutes to obtain a test rice product raw material sample. The absorbance of the sample solution was measured at a wavelength of 620 nm. A blank test was then performed according to the sample measurement steps without adding the sample, and the blank absorbance at a wavelength of 620 nm was measured.

[0234] At a wavelength of 620 nm, the difference between the absorbance of the rice product raw material sample solution and the blank absorbance was calculated, and the amylose concentration was obtained by comparing it with the amylose regression equation, and the dilution factor was calculated:

[0235]

[0236] Where n represents the dilution multiple;

[0237] The calculation formula for amylose content is:

[0238]

[0239] Where C Al Expressed as amylose content, c al It is expressed as the amylose concentration obtained from the amylose regression equation in ug / ml, and V is expressed as the solution volume of the test rice product raw material sample in ml.

[0240] Among them, the steps for constructing the amylose regression equation are:

[0241] Construct a univariate linear regression equation:

[0242] y=a+bx

[0243] Where y represents absorbance, x represents amylose concentration, a represents intercept, and b represents slope;

[0244] By experimentally measuring the absorbance of different concentrations of amylose standard solutions, a series of corresponding data points (x1, y1), (x2, y2), (x 11 , t 11 ), recorded as (x i ,y i );

[0245] Calculate the mean value of amylose concentration:

[0246]

[0247] Where, It is expressed as the mean value of amylose concentration;

[0248] Calculate the average absorbance of different concentrations of amylose standard solutions:

[0249]

[0250] Where, It is expressed as the average absorbance of standard solutions of amylose at different concentrations;

[0251] Calculate the slope of a linear regression equation:

[0252]

[0253] Compute the intercept of a univariate linear regression equation:

[0254]

[0255] Construct a relevant content table, in which the row headers include sample number, protein characteristic wavelength and corresponding absorbance, water characteristic wavelength and corresponding absorbance, and amylose characteristic wavelength. According to the grouping order of the samples, number each group of samples from 1 to 20 and fill in the first column of the table. For each group of samples, fill in the characteristic wavelengths of protein, water, and amylose and their corresponding absorbance in the corresponding row.

[0256] Step 3: Extract the characteristic wavelength and absorbance of the sample from the relevant content table, and establish a prediction model for protein, moisture, and amylose content based on the partial least squares method;

[0257] When studying protein, moisture, and amylose content, multiple influencing factors and variables are often involved. Partial least squares can effectively handle the complex relationships between multiple independent variables and multiple dependent variables, comprehensively considering the impact of various factors on content, unlike some traditional methods that may only handle a single variable or simple linear relationship;

[0258] However, in actual measurements, there is a high correlation between the independent variables used to predict protein, moisture and amylose content, which is a multicollinearity problem. The partial least squares method can eliminate or reduce the collinearity effect between variables by extracting principal components, making the model more stable and accurate, and able to accurately evaluate the true contribution of each variable to the content.

[0259] Extract the characteristic wavelengths and absorbances of the samples from the relevant content table. Based on the 20 sets of data extracted, each set of data contains the absorbances of 5 characteristic wavelengths, and set the dimension of the absorbance data matrix to 20×5:

[0260]

[0261] Where X represents the absorbance data matrix;

[0262] Set the dependent variable matrix to 20×3:

[0263]

[0264] Where Y represents the dependent variable matrix;

[0265] Calculate the mean of columns 1, 2, and 3 of the dependent variable matrix Y:

[0266]

[0267] Where, and They are respectively represented as the average values of the 1st, 2nd and 3rd columns of the dependent variable matrix Y, and h is represented as the row index of the matrix;

[0268] Calculate the standard deviation of columns 1, 2, and 3 of the dependent variable matrix Y:

[0269]

[0270] Where s1, s2 and s3 represent the standard deviations of the first, second and third columns of the dependent variable matrix Y.

[0271] Standardize the absorbance data matrix X and the dependent variable matrix Y:

[0272]

[0273] Where g hj Represented as original data, where j represents the column index of the matrix, j∈(j dy ,j dr ), j dy Expressed as the column index of the absorbance data matrix X, j dr Represented as the column index of the dependent variable matrix Y, Expressed as the mean value of the jth column, s j Expressed as the standard deviation of the jth column, the standardized results are substituted back into the original absorbance data matrix X and dependent variable matrix Y;

[0274] The mean value of column j is:

[0275]

[0276] The standard deviation of the jth column is

[0277]

[0278] Calculate the covariance matrix of the absorbance data matrix X and the dependent variable matrix Y:

[0279]

[0280] Where Cov(X, Y) represents the covariance matrix of the absorbance data matrix X and the dependent variable matrix Y;

[0281] Assume that the process matrix with dimension q×q is:

[0282] B=YY T

[0283] Where B represents the process matrix;

[0284] The process 2 matrix with dimension p×p is:

[0285] C=X T YY T X

[0286] Where C represents the process two matrix;

[0287] Compute eigenvalues and eigenvectors:

[0288] Since CW=λ′W, we have:

[0289] (C-λ′I)W=0

[0290] Where λ′ is the eigenvalue, W is the eigenvector, and I is the identity matrix with dimension p×p;

[0291] Construct the characteristic equation of the process two matrix:

[0292] det(C-λ′I)=0

[0293] Where det(C-λ′I) is the characteristic equation of the process 2 matrix, and p eigenvalues λ′1, λ′2, …, λ′ are calculated. p ;

[0294] For each eigenvalue λ i′ , solve the homogeneous linear equations (C-λ′ i′ I) W = 0, get the corresponding eigenvector W i′ ;

[0295] Compare p eigenvalues λ′1, λ′2, …, λ′ p The size of the eigenvalue is found, which is recorded as λ′ max , and its corresponding eigenvector is recorded as W1;

[0296] After that, calculate the number of principal components to be extracted:

[0297] Assuming the number of principal components is k, the variance contribution rate of the i′th principal component is:

[0298]

[0299] Where, v i′ Expressed as the variance contribution rate of the i′th principal component;

[0300] The calculation formula for the cumulative variance contribution rate of the first k principal components is:

[0301]

[0302] Starting from the first principal component, calculate the cumulative variance contribution rate v1, v2, ..., v p , when it reaches or exceeds 0.85 for the first time, k is the number of principal components determined;

[0303] Compute the first component of the absorbance data matrix X:

[0304] t1=X·W1

[0305] Where X represents the first component of the absorbance data matrix X;

[0306] Calculate the regression coefficient of the dependent variable Y on t1:

[0307]

[0308] Where C1 represents the regression coefficient of the dependent variable Y on t1;

[0309] Calculate the first component of the dependent variable matrix Y:

[0310] u1=t1C1

[0311] Where u1 represents the first component of the dependent variable matrix Y;

[0312] Calculate the residual matrix for the absorbance data matrix X:

[0313] E1=X-t1P1

[0314] Where E1 represents the residual matrix of the absorbance data matrix X, P1 represents the first load vector, and the calculation formula is:

[0315]

[0316] Calculate the residual matrix of the dependent variable matrix Y:

[0317] F1=Y-t1C1 T

[0318] Where F1 is the residual matrix of the dependent variable matrix Y;

[0319] Repeat the above steps for the residual matrix E1 of the absorbance data matrix X and the residual matrix F1 of the dependent variable matrix Y until the kth (K>1) pair of components t k and u k :

[0320] Calculate E k-1 , F k-1 The covariance matrix of :

[0321]

[0322] Where Cov(E k-1 , F k-1 ) is expressed as F k-1 , F k-1 The covariance matrix of

[0323] Calculate the residual matrix E at the kth iteration k-1 and Fk-1 The cross-covariance characteristic matrix of :

[0324] D=(E k-1 ) T F k-1 (F k-1 ) T E k-1

[0325] Where D is the cross-covariance feature matrix;

[0326] Perform eigendecomposition on it and select the eigenvector corresponding to the maximum eigenvalue as W k , calculate E k-1 The kth component of:

[0327] t k =E k-1 W k

[0328] Calculate F k-1 t k The regression coefficient of :

[0329]

[0330] Where C k Indicated as F k-1 t k The regression coefficient of

[0331] Calculate F k-1 The kth component of:

[0332] u k =t k C k

[0333] Calculate the residual matrix of the k-th extracted component:

[0334] E k =E k-1 -t k P k T

[0335] Where, E k It is expressed as the residual matrix of the k-th extracted component, P k is represented as the kth load vector, where

[0336]

[0337] Calculate the residual matrix 2 of the k-th extracted component:

[0338] F k =F k-1-t k C k T

[0339] Where, F k It is represented as the residual matrix 2 of the k-th extracted component;

[0340] Construct the composition matrix:

[0341] T cf =[t1, t2, ..., t k ]

[0342] Where, T cf Represented as a component matrix;

[0343] Construct the regression coefficient matrix:

[0344] C hg =[C1, C2, ..., C k ] T

[0345] Where C hg Expressed as a regression coefficient matrix;

[0346] Construct a standardized regression model:

[0347]

[0348] Due to t i′ =XW i′ , set W′=[W1,W2,…,W k ]

[0349] The standardized prediction model is:

[0350]

[0351] The normalized prediction model Y′ is denormalized, and the prediction model of protein is:

[0352]

[0353] Where y h1 Represented as a prediction model for proteins;

[0354] The prediction model for moisture is:

[0355]

[0356] Where y h2 It is expressed as a prediction model of moisture;

[0357]

[0358] Where yh3 Expressed as a prediction model for amylose.

[0359] Step 4: Use stratified sampling estimation method to count yellowed and insect-damaged grains in rice product raw materials, and use prediction models to calculate protein, moisture, and amylose content, and construct a rice product raw material quality assessment formula to predict quality;

[0360] Divide the rice product raw materials into N new After the rice is divided into groups, the same weight WT of rice raw materials is extracted from each group as the sample to be tested. This ensures that the extracted samples can evenly cover the entire rice product raw material group and avoid the fact that some parts of the raw materials are not detected due to sampling bias, so that the test results can more accurately reflect the quality status of the overall raw materials. The cracks, yellowing and insect-damaged grains in each sampled rice raw material are identified using the Japanese Satake Grain Rating Instrument. The identified cracked grains, yellowing grains and insect-damaged grains are weighed separately and their respective weights wT are recorded. id , where id = 1, 2, ... N new In the food industry, there are strict standards and specifications for the quality of rice raw materials. Such testing can ensure that the raw materials used by companies meet the relevant standards and avoid reputation losses due to raw material quality issues.

[0361] Calculate the proportion of cracked kernels, yellowed kernels, and insect-damaged kernels in each sampling group:

[0362]

[0363] Where p id It is expressed as the percentage of cracked grains, yellowed grains and insect-damaged grains in group id;

[0364] Calculate the proportion of cracked, yellowed, and insect-damaged rice grains in the raw materials for rice products:

[0365]

[0366] Where, P total It is expressed as the proportion of cracked grains, yellowed grains and insect-damaged grains in the raw materials of rice products.

[0367] For N new The method for predicting the content of the τth species (τ=1 represents albumin, τ=2 represents protein, and τ=3 represents amylose) is as follows:

[0368] The extracted bands are 1000, 1500, 1600, 1700, and 3200 cm -1 The absorbance is constructed with dimension N new ×5 absorbance data matrix X to be detected new :

[0369] The absorbance data matrix X to be detected new To perform standardization:

[0370]

[0371] Where h new Represented as row index, j new Represented as a column index, It is represented as the hth matrix of the absorbance data to be detected after normalization new OK, J new Column data, get the standardized absorbance data matrix X * new ;

[0372] The calculation dimension is N new The projection matrix of the new sample ×k in the direction of the extracted k principal components:

[0373] M new =X * new W′

[0374] Where M new Represented as the projection matrix of the new sample in the direction of the extracted k principal components;

[0375] Among them, the projection of the new sample in the direction of the principal component after standardization is:

[0376]

[0377] It is represented as the projection of the new sample in the direction of the principal component after standardization, l new Represents the index of the principal component, where l new =1,2,…,k, Indicated as X * new Middle h new Row j new Elements of the column index, Denote the jth new Row 1 new Elements of column index;

[0378] Z new =M new (C hg )^

[0379] Where Z new Indicated as N new ×3 matrix, (C hg )^ represents the estimated value of the regression coefficient matrix with dimension k×3;

[0380] Calculate the hth new The content-normalized intermediate prediction value of a new sample at τ:

[0381]

[0382] Where, Expressed as h new The standardized intermediate predicted value of the τth content of the samples to be tested, c lτ Expressed as (C hg )^ in the first new The element in row and column τ;

[0383] Calculate the hth new The formula for the standardized predicted value of the τth content is obtained after principal component projection and regression coefficient calculation of the samples to be tested:

[0384]

[0385] Where Y hτ ′ represents the hth new The standardized predicted value of the content of the τth component in the samples to be tested, Expressed as h new The residual error of the sample to be tested on the prediction of the content of the τth component is 0;

[0386] Substitute into the prediction model, then the h new The prediction formula for the content of the τth component of a new sample is:

[0387]

[0388] Where, Expressed as h new The predicted value of the content of the τth component of a new sample, Y sτ Expressed as the standard deviation of the τth column in the dependent variable matrix Y, Expressed as the mean value of the τth column in the dependent variable matrix Y;

[0389] The protein content of rice product raw materials is:

[0390]

[0391] Where dω1 represents the protein content of rice product raw materials;

[0392] The moisture content of rice product raw materials is:

[0393]

[0394] Where dω2 represents the moisture content of the rice product raw material;

[0395] The amylose content of rice product raw materials is:

[0396]

[0397] Where dω2 represents the amylose content of rice product raw materials,

[0398] For example, the method to calculate the protein content of a sample is:

[0399] For N new The samples to be tested were divided into groups, and the wavelengths to be extracted were 1000, 1500, 1600, 1700, and 3200 cm -1 The absorbance is constructed with dimension N new ×5 absorbance data matrix X to be detected new :

[0400] The absorbance data matrix X to be detected new To perform standardization:

[0401]

[0402] The calculation dimension is N new The projection matrix of the new sample ×k in the direction of the extracted k principal components:

[0403] M new =X * new W′

[0404] Among them, the projection of the new sample in the direction of the principal component after standardization is:

[0405]

[0406] Z new =M new (C hg )^

[0407] Calculate the hth new Normalized intermediate predicted value of protein content of a new sample:

[0408]

[0409] Where, Expressed as h new 1 Normalized intermediate prediction value of protein content of a new sample, c l1 Expressed as (C hg )^ in the first new The element in row and column 1;

[0410] Calculate the hth newThe formula for the standardized predicted value of protein content obtained by principal component projection and regression coefficient calculation for each sample to be tested is:

[0411]

[0412] Where Y h1 ′ represents the hth new The protein content normalized prediction value is obtained after principal component projection and regression coefficient calculation of the samples to be tested. Expressed as h new The residual error of the protein content prediction of the sample to be tested is set to 0;

[0413] Substitute into the protein prediction model, then the h new The protein content prediction formula for a new sample is:

[0414]

[0415] Then, based on the protein, moisture, and amylose content of rice product raw materials and the proportion of cracked grains, yellowed grains, and insect-damaged grains, a quality evaluation formula for rice product raw materials was constructed:

[0416]

[0417] Where Q represents the quality evaluation formula for rice product raw materials, ε1, ε2, and ε3 represent weight coefficients, which need to be set according to different actual conditions, and ε1+ε2+ε3=1, and β represents the adjustment parameter, with a default range of 1-10;

[0418] The weight coefficient reflects the relative importance of various factors to the quality of rice product raw materials and can be adjusted according to one's own needs and product characteristics. Taking the production of high-end sushi rice as an example, a high amylose content will make the taste drier and the rice harder, while a low amylose content will make the taste softer and more sticky. The moisture content will affect the water absorption and expansion taste during the steaming process. The recommended weights are set to ε1=0.2, ε2=0.3, and ε3=0.5.

[0419] The quality thresholds are set as Q1, Q2 and Q3, and Q1>Q2>Q3>0. Set Q1=0.18, Q2=0.14, Q3=0.10. When Q>Q1, the quality of the rice product raw material is evaluated as excellent. When Q1≥Q>Q2, the quality of the rice product raw material is evaluated as good. When Q2≥Q>Q3, the quality of the rice product raw material is evaluated as qualified. When Q3≥Q, the quality of the rice product raw material is evaluated as poor.

[0420] For example, the protein content dω1 = 0.18, the moisture content dω2 = 0.12, the amylose content dω3 = 0.24, the proportion of cracked grains, yellowed grains and insect-damaged grains Ptotal =0.005, adjustment parameter β=5, weight coefficient ε1=0.4, ε2=0.3, ε3=0.3,

[0421]

[0422] Q1≥Q>Q2

[0423] The quality assessment of rice product raw materials is qualified.

[0424] See also Figure 2 The present invention further provides a device for predicting the quality of rice product raw materials based on deep feature fusion. The system is used to execute the above-mentioned method for predicting the quality of rice product raw materials based on deep feature fusion, comprising:

[0425] A spectral data acquisition module is used to select samples from rice product raw materials using a stratified sampling method, collect spectral data using infrared spectral imaging technology, and record characteristic wavelengths and absorbances related to protein, moisture, and amylose;

[0426] The content determination module is used to determine the protein content of the sample using the Kjeldahl method, the moisture content using the oven drying method, and the amylose content using the iodine colorimetric method. The characteristic wavelengths and absorbances associated with protein, moisture, and amylose are used to construct a table of related contents.

[0427] The model building module is used to extract the characteristic wavelength and absorbance of the sample from the relevant content table and establish a prediction model for protein, moisture and amylose content based on the partial least squares method;

[0428] The quality prediction module is used to use the stratified sampling estimation method to count the yellowed grains and insect-damaged grains in rice product raw materials, and use the prediction model to calculate the protein, moisture and amylose content, and construct the rice product raw material quality evaluation formula to predict the quality.

[0429] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters in the formulas are set by technicians in this field according to actual conditions.

[0430] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed by hardware or software depends on the specific application and design constraints of the technical solution.

[0431] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, and may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment as needed.

[0432] The above is only a specific implementation method of the present application, but the scope of protection of the present application is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed in this application, which should be covered by the scope of protection of the present application.

Claims

1. A method for predicting the quality of rice product raw materials based on deep feature fusion, characterized in that: The specific steps include: Step 1: Samples were selected from the rice product raw materials using a stratified sampling method. Spectral data were collected using infrared spectroscopy imaging technology to record the characteristic wavelengths and absorbances related to protein, moisture, and amylose. Step 2: Determine the protein content of the sample using the Kjeldahl method, the moisture content using the oven drying method, and the amylose content using the iodine colorimetric method. Construct a table of the relevant contents using the characteristic wavelengths and absorbances associated with protein, moisture, and amylose. Step 3: Extract the characteristic wavelength and absorbance of the sample from the relevant content table, and establish a prediction model for protein, moisture, and amylose content based on the partial least squares method; Step 4: Use the stratified sampling estimation method to count the yellowed and insect-damaged grains in the rice product raw materials, and use the prediction model to calculate the protein, moisture and amylose content, and construct the rice product raw material quality evaluation formula to predict the quality.

2. The method for predicting the quality of rice product raw materials based on deep feature fusion according to claim 1, wherein: In step 1, selecting samples from the rice product raw materials using a stratified sampling method includes the following steps: The rice product raw materials were divided into 20 equal portions according to mass. For each portion of rice product raw materials, 10 samples were selected by simple random sampling and divided into 20 groups.

3. The method for predicting the quality of rice product raw materials based on deep feature fusion according to claim 2, wherein: The following steps are involved in collecting spectral data using infrared spectral imaging technology to record the characteristic wavelengths and absorbances associated with protein, moisture, and amylose: Using infrared spectral imaging technology, 20 groups of rice product raw material samples were focused on 900 to 3000 cm -1 Band, recording band 1000, 1500, 1600, 1700, 3200cm -1 The absorbance I λ , where λ = 1000, 1500, 1600, 1700 and 3200, and the characteristic wavelengths of proteins are set to 1500 and 1700 cm -1 The characteristic wavelengths of water are 1600 and 3200 cm -1 The characteristic wavelength of amylose is 1000 cm -1 .

4. The method for predicting the quality of rice product raw materials based on deep feature fusion according to claim 2, wherein: In step 2, the protein content of the sample is determined by the Kjeldahl method, the moisture content is determined by the oven drying method, and the amylose content is determined by the iodine colorimetric method, which includes the following steps: Each group of rice product raw material samples was ground and divided equally into three parts of mass A. The protein content of the samples was determined by Kjeldahl nitrogen determination, the moisture content was determined by oven drying method, and the amylose content was determined by iodine colorimetry. The method for determining the protein content of samples by Kjeldahl method is: Place a rice product raw material sample with a mass of A (in grams) into a dry Kjeldahl flask, add a mixture of 3g potassium sulfate and 1g copper sulfate and 20ml concentrated sulfuric acid, and continue heating until the liquid becomes blue-green, clear and transparent. Continue heating for 0.5 hour. After the Kjeldahl flask cools down, connect it to a Kjeldahl nitrogen determination distillation apparatus, add 20g / L boric acid solution and bromocresol green-methyl red mixed indicator to another conical flask, place the conical flask under the condenser of the distillation apparatus, and insert the lower end of the condenser below the liquid level of the boric acid solution. Add 400g / L sodium hydroxide solution to the Kjeldahl flask until the volume of the distillate, that is, the boric acid solution, reaches 100ml. Titrate the boric acid solution with 0.1mol / L hydrochloric acid standard solution. Stop when the solution changes from green to gray-red. At the same time, Take another Kjeldahl flask, add a mixture of 3g potassium sulfate, 1g copper sulfate and 20ml concentrated sulfuric acid and continue heating until the liquid is blue-green, clear and transparent, continue heating for 0.5 hour, after the Kjeldahl flask is cooled, connect it to the Kjeldahl nitrogen determination distillation apparatus, take another Erlenmeyer flask and add 20g / L boric acid solution and bromocresol green-methyl red mixed indicator, place the Erlenmeyer flask under the condenser of the distillation apparatus, insert the lower end of the condenser below the boric acid solution liquid level, add 400g / L sodium hydroxide solution to the Kjeldahl flask until the distillate, i.e., the volume of the boric acid solution reaches 100ml, titrate the boric acid solution with 0.1mol / L hydrochloric acid standard solution, stop when the solution becomes gray-red, and record the volume of the hydrochloric acid standard solution consumed, i.e., the volume of the hydrochloric acid standard solution consumed in the blank test; Calculation of nitrogen content: Where X represents the nitrogen content in the sample, c represents the concentration of the hydrochloric acid standard solution, V1 represents the volume of the hydrochloric acid standard solution consumed when titrating the sample, V0 represents the volume of the hydrochloric acid standard solution consumed in the blank experiment, m represents the mass of the sample, and 0.014 represents the molar mass of nitrogen; Converted protein content: C p =X×6.25 Where C p Expressed as protein content; The method for determining moisture content by oven drying is: Select a weighing bottle with a mass of m1, place a rice product raw material sample with a mass of A into the weighing bottle, and record the total mass as m2. Place the weighing bottle containing the rice product raw material sample in an oven at 105°C for 3 hours. After drying, remove the weighing bottle and record the mass of the rice product raw material sample and the weighing bottle as m3. Calculate the moisture content: Where C w Expressed as moisture content; The method for determining the content of amylose by iodine colorimetry is as follows: Weigh 0.1000g of amylose standard, add 1ml of anhydrous ethanol, then add 9.0mL of 1mol / L sodium hydroxide solution, heat in a boiling water bath for 10 minutes, cool and transfer to a 100mL volumetric flask, dilute to the mark with 1.0mg / mL distilled water, and then pipette a set of volume data solutions starting from 0 and increasing in increments of 1.0mL to 10.0mL into a 100mL volumetric flask, dilute to the mark with distilled water, and dilute to a concentration starting from 0 and increasing in increments of 10.0μg / mL. Prepare a series of standard solutions with concentration data of 100 μg / mL to 100.0 μg / mL. Weigh the amylopectin standard and prepare a solution with a concentration of 1.0 mg / mL according to the preparation method of the above amylose standard solution. Then dilute it to a series of standard solutions with different concentrations according to the same gradient. Weigh 2.0 g of potassium iodide and dissolve it in a small amount of distilled water. Add 0.2 g of iodine and, after complete dissolution, dilute to 100 mL with distilled water. Store in a brown bottle. Weigh 40 g of 1 mol / L sodium hydroxide solution and dissolve it in distilled water. After cooling, dilute to 1000 mL. A rice product raw material sample of mass A was placed in a 100 mL conical flask, 10 mL of anhydrous ethanol was added and stirred evenly, the mixture was soaked for 1 hour, filtered, and cooled to room temperature. The solution was then transferred to a 100 mL volumetric flask, diluted to the mark with distilled water, and shaken to obtain a rice product raw material sample solution. 2.0 mL of amylose standard solution of different concentrations was respectively drawn into a 25 mL colorimetric tube, 1.0 mL of iodine reagent was added to each solution, and the solution was diluted to the mark with distilled water. The solution was shaken and allowed to stand for 15 minutes. Distilled water was used as a blank control, and the absorbance of each standard solution was measured at a wavelength of 620 nm using a spectrophotometer. A standard curve was plotted with amylose concentration as the abscissa and absorbance as the ordinate to obtain the amylose regression equation. A 2.0 nm rice product raw material sample solution was drawn into a 25 nm colorimetric tube, and iodine reagent was added and the volume was adjusted according to the steps for drawing the standard curve. The solution was shaken and allowed to stand for 15 minutes to obtain a test rice product raw material sample, and the absorbance of the sample solution was measured at a wavelength of 620 nm. Then, a blank test was performed without adding the sample according to the sample measurement steps, and the blank absorbance at a wavelength of 620 nm was measured; At a wavelength of 620 nm, the difference between the absorbance of the rice product raw material sample solution and the blank absorbance was calculated, and the amylose concentration was obtained by comparing it with the amylose regression equation, and the dilution factor was calculated: Where n represents the dilution multiple; The calculation formula for amylose content is: Where C Al Expressed as amylose content, c al It is expressed as the amylose concentration obtained from the amylose regression equation in ug / ml, and V is expressed as the solution volume of the test rice product raw material sample in ml.

5. The method for predicting the quality of rice product raw materials based on deep feature fusion according to claim 4, characterized in that: In the method for determining the amylose content by iodine colorimetry, the specific steps for obtaining the amylose regression equation include: Construct a univariate linear regression equation: y=a+bx Where y represents absorbance, x represents amylose concentration, a represents intercept, and b represents slope; By experimentally measuring the absorbance of different concentrations of amylose standard solutions, a series of corresponding data points (x1, y1), (x2, y2), (x 11 ,y 11 ), recorded as (x i ,y i ); Calculate the mean value of amylose concentration: Where, It is expressed as the mean value of amylose concentration; Calculate the average absorbance of different concentrations of amylose standard solutions: Where, It is expressed as the average absorbance of standard solutions of amylose at different concentrations; Calculate the slope of a linear regression equation: Compute the intercept of a univariate linear regression equation:

6. The method for predicting the quality of rice product raw materials based on deep feature fusion according to claim 5, characterized in that: In step 3, the method for establishing the prediction model of protein, moisture and amylose content based on the partial least squares method is as follows: Based on the 20 sets of data extracted, each set of data contains the absorbance of 5 characteristic wavelengths, and the dimension of the absorbance data matrix is set to 20×5: Where X represents the absorbance data matrix; Set the dependent variable matrix to 20×3: Where Y represents the dependent variable matrix; Calculate the mean of columns 1, 2, and 3 of the dependent variable matrix Y: Where, and They are respectively represented as the average values of the 1st, 2nd and 3rd columns of the dependent variable matrix Y, and h is represented as the row index of the matrix; Calculate the standard deviation of columns 1, 2, and 3 of the dependent variable matrix Y: Where s1, s2 and s3 represent the standard deviations of the first, second and third columns of the dependent variable matrix Y. Standardize the absorbance data matrix X and the dependent variable matrix Y: Where g hj Represented as original data, where j represents the column index of the matrix, j∈(j dy ,j dr ), j dy Expressed as the column index of the absorbance data matrix X, j dr Represented as the column index of the dependent variable matrix Y, Expressed as the mean value of the jth column, s j Expressed as the standard deviation of the jth column, the standardized results are substituted back into the original absorbance data matrix X and dependent variable matrix Y; The mean value of column j is: The standard deviation of the jth column is Calculate the covariance matrix of the absorbance data matrix X and the dependent variable matrix Y: Where Cov(X, Y) represents the covariance matrix of the absorbance data matrix X and the dependent variable matrix Y; Assume that the process matrix with dimension q×q is: B=YY T Where B represents the process matrix; The process 2 matrix with dimension p×p is: C=X T YY T X Where C represents the process two matrix; Compute eigenvalues and eigenvectors: Since CW=λ′W, we have: (C-λ′I)W=0 Where λ′ is the eigenvalue, W is the eigenvector, and I is the identity matrix with dimension p×p; Construct the characteristic equation of the process two matrix: det(C-λ′I)=0 Where det(C-λ′I) is the characteristic equation of the process 2 matrix, and p eigenvalues λ′1, λ′2, …, λ′ are calculated. p ; For each eigenvalue λ i′ , solve the homogeneous linear equations (C-λ′ i′ I) W = 0, get the corresponding eigenvector W i′ ; Compare p eigenvalues λ′1, λ′2, …, λ′ p The size of the eigenvalue is found, which is recorded as λ′ max , and its corresponding eigenvector is recorded as W1; After that, calculate the number of principal components to be extracted: Assuming the number of principal components is k, the variance contribution rate of the i′th principal component is: Where, v i′ Expressed as the variance contribution rate of the i′th principal component; The calculation formula for the cumulative variance contribution rate of the first k principal components is: Starting from the first principal component, calculate the cumulative variance contribution rate v1, v2, ..., v p , when it reaches or exceeds 0.85 for the first time, k is the number of principal components determined; Compute the first component of the absorbance data matrix X: t1=X·W1 Where X represents the first component of the absorbance data matrix X; Calculate the regression coefficient of the dependent variable Y on t1: Where C1 represents the regression coefficient of the dependent variable Y on t1; Calculate the first component of the dependent variable matrix Y: u1=t1C1 Where u1 represents the first component of the dependent variable matrix Y; Calculate the residual matrix for the absorbance data matrix X: E1=X-t1P1 Where E1 represents the residual matrix of the absorbance data matrix X, P1 represents the first load vector, and the calculation formula is: Calculate the residual matrix of the dependent variable matrix Y: F1=Y-t1C1 T Where F1 is the residual matrix of the dependent variable matrix Y; Repeat the above steps for the residual matrix E1 of the absorbance data matrix X and the residual matrix F1 of the dependent variable matrix Y until the kth (K>1) pair of components t k and u k : Calculate E k-1 , F k-1 The covariance matrix of : Where Cov(E k-1 , F k-1 ) is expressed as E k-1 , F k-1 The covariance matrix of Calculate the residual matrix E at the kth iteration k-1 and F k-1 The cross-covariance characteristic matrix of : D=(E k-1 ) T F k-1 (F k-1 ) T E k-1 Where D is the cross-covariance feature matrix; Perform eigendecomposition on it and select the eigenvector corresponding to the maximum eigenvalue as W k , calculate E k-1 The kth component of: t k =E k-1 W k Calculate F k-1 t k The regression coefficient of : Where C k Indicated as F k-1 t k The regression coefficient of Calculate F k-1 The kth component of: u k =t k C k Calculate the residual matrix of the k-th extracted component: E k =E k-1 -t k P k T Where, E k It is expressed as the residual matrix of the k-th extracted component, P k is represented as the kth load vector, where Calculate the residual matrix 2 of the k-th extracted component: F k =F k-1 -t k C k T Where, F k It is represented as the residual matrix 2 of the k-th extracted component; Construct the composition matrix: T cf =[t1,t2,…,t k ] Where, T cf Represented as a component matrix; Construct the regression coefficient matrix: C hg =[C1,C2,…,C k ] T Where C hg Expressed as a regression coefficient matrix; Construct a standardized regression model: Due to t i′ =XW i′ , set W′=[W1,W2,…,W k ] The standardized prediction model is: The normalized prediction model Y′ is denormalized, and the prediction model of protein is: Where y h1 Represented as a prediction model for proteins; The prediction model for moisture is: Where y h2 It is expressed as a prediction model of moisture; Where y h3 Expressed as a prediction model for amylose.

7. The method for predicting the quality of rice product raw materials based on deep feature fusion according to claim 1, wherein: In step 4, using a stratified sampling estimation method to count the yellowed grains and insect-infested grains in the rice product raw materials includes the following steps: Divide the rice product raw materials into N new The same weight WT of rice raw materials were taken from each group as the test samples. The cracks, yellowing and insect-damaged grains in each sample of rice raw materials were identified using the Japanese Satake Grain Rating Instrument. The identified cracked grains, yellowing grains and insect-damaged grains were weighed separately and their respective weights wT were recorded. id , where id = 1, 2, ... N new ; Calculate the proportion of cracked kernels, yellowed kernels, and insect-damaged kernels in each sampling group: Where p id It is expressed as the percentage of cracked grains, yellowed grains and insect-damaged grains in group id; Calculate the proportion of cracked, yellowed, and insect-damaged rice grains in the raw materials for rice products: Where, P total It is expressed as the proportion of cracked grains, yellowed grains and insect-damaged grains in the raw materials of rice products.

8. The method for predicting the quality of rice product raw materials based on deep feature fusion according to claim 6, wherein: In step 4, the calculation of protein, moisture and amylose content using the prediction model includes the following steps: For N new The method for predicting the content of the τth species (τ=1 represents albumin, τ=2 represents protein, and τ=3 represents amylose) is as follows: The extracted bands are 1000, 1500, 1600, 1700, and 3200 cm -1 The absorbance is constructed with dimension N new ×5 absorbance data matrix X to be detected new : The absorbance data matrix X to be detected new To perform standardization: Where h new Represented as row index, j new Represented as a column index, It is represented as the hth matrix of the absorbance data to be detected after normalization new OK, J new Column data, get the standardized absorbance data matrix X * new ; The calculation dimension is N new The projection matrix of the new sample ×k in the direction of the extracted k principal components: M new =X * x ew W′ Where M new Represented as the projection matrix of the new sample in the direction of the extracted k principal components; Among them, the projection of the new sample in the direction of the principal component after standardization is: It is represented as the projection of the new sample in the direction of the principal component after standardization, l new Represents the index of the principal component, where l new =1,2,…,k, Expressed as X* new Middle h new Row j new Elements of the column index, Denote the jth new Row 1 new Elements of column index; Z new =M new (C hg )^ Where Z new Indicated as N new ×3 matrix, (C hg )^ represents the estimated value of the regression coefficient matrix with dimension k×3; Calculate the hth new The content-normalized intermediate prediction value of a new sample at τ: Where, Expressed as h new The standardized intermediate predicted value of the τth content of the samples to be tested, c lτ Expressed as (C hg )^ in the first new The element in row and column τ; Calculate the hth new The formula for the standardized predicted value of the τth content is obtained after principal component projection and regression coefficient calculation of the samples to be tested: Where Y hτ ′ represents the hth new The standardized predicted value of the content of the τth component in the samples to be tested, Expressed as h new The residual error of the sample to be tested on the prediction of the content of the τth component is 0; Substitute into the prediction model, then the h new The prediction formula for the content of the τth component of a new sample is: Where, Expressed as h new The predicted value of the content of the τth component of a new sample, Y sτ Expressed as the standard deviation of the τth column in the dependent variable matrix Y, Expressed as the mean value of the τth column in the dependent variable matrix Y; The protein content of rice product raw materials is: Where dω1 represents the protein content of rice product raw materials; The moisture content of rice product raw materials is: Where dω2 represents the moisture content of the rice product raw material; The amylose content of rice product raw materials is: Where dω2 represents the amylose content of the rice product raw material.

9. The method for predicting the quality of rice product raw materials based on deep feature fusion according to claim 8, characterized in that: In step 4, the method for constructing the quality evaluation formula of rice product raw materials to predict quality is: Based on the protein, moisture, and amylose content of rice product raw materials and the proportion of cracked grains, yellowed grains, and insect-damaged grains, a formula for evaluating the quality of rice product raw materials was constructed: Where Q is the quality evaluation formula for rice product raw materials, ε1, ε2, and ε3 are weight coefficients, and ε1+ε2+ε3=1, and β is the adjustment parameter; The quality thresholds are set to Q1, Q2 and Q3, and Q1>Q2>Q3>0. When Q>Q1, the quality of the rice product raw material is evaluated as excellent; when Q1≥Q>Q2, the quality of the rice product raw material is evaluated as good; when Q2≥Q>Q3, the quality of the rice product raw material is evaluated as qualified; when Q3≥Q, the quality of the rice product raw material is evaluated as poor.

10. A device for predicting the quality of rice product raw materials based on deep feature fusion, characterized by: The device is used to execute the method for predicting the quality of rice product raw materials based on deep feature fusion according to any one of claims 1 to 9, comprising: A spectral data acquisition module is used to select samples from rice product raw materials using a stratified sampling method, collect spectral data using infrared spectral imaging technology, and record characteristic wavelengths and absorbances related to protein, moisture, and amylose; The content determination module is used to determine the protein content of the sample using the Kjeldahl method, the moisture content using the oven drying method, and the amylose content using the iodine colorimetric method. The characteristic wavelengths and absorbances associated with protein, moisture, and amylose are used to construct a table of related contents. The model building module is used to extract the characteristic wavelength and absorbance of the sample from the relevant content table and establish a prediction model for protein, moisture and amylose content based on the partial least squares method; The quality prediction module is used to use the stratified sampling estimation method to count the yellowed grains and insect-damaged grains in rice product raw materials, and use the prediction model to calculate the protein, moisture and amylose content, and construct the rice product raw material quality evaluation formula to predict the quality.

Images