Method and device for detecting hydrogen peroxide concentration of food packaging production line

Abstract

The invention provides a hydrogen peroxide concentration detection method and device for a food packaging production line, and the method comprises the following steps: preparing hydrogen peroxide solution samples with different concentrations, measuring the near infrared spectrum of the hydrogen peroxide solution samples to obtain near infrared spectrum data, forming an original spectrum matrix by the near infrared spectrum data, and calculating the concentration of hydrogen peroxide in the original spectrum matrix; meanwhile, forming an original concentration matrix by corresponding preparation concentration data; removing optical path errors and wave band screening from the original spectrum matrix, and then forming an input matrix with the original concentration matrix; inputting the input matrix into a 1D-CNN regression model for training to obtain a detection data feedback model; the near infrared spectrum data of the hydrogen peroxide solution with the hydrogen peroxide concentration to be detected are obtained, and after optical path errors and wave band screening are removed, the data are input into the detection data feedback model to obtain the hydrogen peroxide concentration value of the hydrogen peroxide solution with the hydrogen peroxide concentration to be detected. The concentration of hydrogen peroxide in the packaging production line is accurately detected.

Description

Technical Field

[0001] The present invention relates to the technical field of hydrogen peroxide detection, and in particular to a method and a device for detecting hydrogen peroxide concentration in a food packaging production line. Background Art

[0002] Hydrogen peroxide is unstable in nature and its aqueous solution concentration is easy to change, which directly affects the product quality and use effect of hydrogen peroxide solution. Especially in the fields of food, medicine, cosmetics, etc., the content of hydrogen peroxide must be strictly controlled, otherwise it may cause serious damage to human health. Therefore, developing a device that can quickly, efficiently and in real time monitor the content of hydrogen peroxide in food packaging production lines is crucial to ensure food safety and reduce the detection and maintenance costs in the production process. This is also a technical problem that needs to be solved urgently.

[0003] At present, there are many methods for detecting hydrogen peroxide concentration. Among them, chemical titration analysis has high accuracy, but has hysteresis, complex operation, and is difficult to meet automation requirements; spectrophotometry is limited by low sensitivity and poor reproducibility of results; fluorescence photometry has low fluorescence conversion efficiency and stability needs to be improved. In order to improve the stability and safety of food packaging production lines, a fast and high-precision concentration monitoring method is urgently needed. Summary of the invention

[0004] The purpose of the present invention is to provide a method and device for detecting the concentration of hydrogen peroxide in a food packaging production line, which can detect the concentration of hydrogen peroxide non-destructively, quickly and accurately.

[0005] To solve the above problems, the first aspect of the present invention provides a method for detecting hydrogen peroxide concentration in a food packaging production line, comprising the following steps:

[0006] Step 1000: preparing hydrogen peroxide solution samples of different concentrations, and measuring the near infrared spectrum of the hydrogen peroxide solution samples to obtain near infrared spectrum data, forming an original spectrum matrix with the near infrared spectrum data, and forming an original concentration matrix with the corresponding prepared concentration data;

[0007] Step 2000: removing the optical path error and the wavelength screening from the original spectrum matrix and combining it with the original concentration matrix to form an input matrix;

[0008] Step 3000: inputting the input matrix into a 1D-CNN regression model for training to obtain a detection data feedback model;

[0009] Step 4000: Acquire the near infrared spectrum data of the hydrogen peroxide solution to be tested, remove the optical path error and the band screening, and input the data into the detection data feedback model to obtain the hydrogen peroxide concentration value of the hydrogen peroxide solution to be tested.

[0010] Furthermore, step 2000 in the above-mentioned method for detecting hydrogen peroxide concentration in a food packaging production line includes:

[0011] Step 2100: preprocessing the original spectrum matrix by using multivariate scattering to remove optical path errors, thereby obtaining a preprocessed spectrum matrix;

[0012] Step 2200: performing band screening processing on the pre-processed spectrum matrix by competitive adaptive reweighted sampling method to obtain a screened spectrum matrix;

[0013] Step 2300: Combining the screened spectrum matrix and the original concentration matrix into the input matrix.

[0014] Furthermore, step 2100 in the above-mentioned method for detecting hydrogen peroxide concentration in a food packaging production line includes the following steps:

[0015] Step 2110: Calculate the average spectrum of the hydrogen peroxide solution sample to obtain an average spectrum matrix;

[0016] Step 2120: Perform regression fitting on the original spectral matrix and the average spectral matrix, solve to obtain a first regression coefficient and construct a first regression function, and obtain the corrected preprocessed spectral matrix through the first regression function.

[0017] Furthermore, step 2200 in the above-mentioned method for detecting hydrogen peroxide concentration in a food packaging production line includes the following steps:

[0018] Step 2210: using the Monte Carlo sampling method to randomly sample part of the sample data from the corrected preprocessed spectrum matrix into the modeling set, and the remaining sample data into the prediction set, and establish a first least squares model, and repeat the above sampling process several times, and record the absolute value weight of the second regression coefficient in the first least squares model of each sampling process;

[0019] Step 2220: Use an exponential decay function to remove the relatively small wavelength values ​​in the absolute value weight of the second regression coefficient, and calculate the ratio R of the retained wavelength points. i ;

[0020] Step 2230: At each sampling, the number of variables selected by adaptive weighted sampling is (R i ×n) wavelength variables, and perform PLS modeling to calculate the cross-validation mean square error;

[0021] Step 2240: After the sampling is completed, the characteristic wavelength subset and the corresponding cross-validation mean square error value at each sampling are obtained, and the characteristic wavelength subset corresponding to the minimum cross-validation mean square error value is selected as the characteristic wavelength.

[0022] Furthermore, step 3000 in the above-mentioned method for detecting hydrogen peroxide concentration in a food packaging production line specifically comprises the following steps:

[0023] Step 3100: randomly divide the input matrix into two parts, respectively serving as a training matrix and a test matrix;

[0024] Step 3200: Perform an average pooling operation on the global image to obtain global information, and perform a one-dimensional convolution operation to calculate the one-dimensional convolution kernel size k. c , the calculation formula is:

[0025]

[0026] Where C is the number of channels, |x| odd is the odd number closest to x, b = 1, γ = 2;

[0027] Step 3300: Obtain the inter-channel relationship weight through convolution operation, and map the convolution output to [0, 1] through Sigmoid function to obtain the attention weight;

[0028] Step 3400: Apply the attention weight to the output of the convolutional layer to obtain an output weighted result, thereby obtaining a detection data feedback model.

[0029] Furthermore, in the above-mentioned method for detecting hydrogen peroxide concentration in a food packaging production line, step 3500: using the detection data feedback model to test the test set to obtain a predicted value, and comparing the predicted value with the actual value to calculate the correlation coefficient r between the two. 2 and root mean square error RMSE.

[0030] According to another aspect of the present invention, a hydrogen peroxide concentration detection device for a food packaging production line is provided, comprising the following modules:

[0031] Module 1000: used to prepare hydrogen peroxide solution samples of different concentrations, and measure the near infrared spectrum of the hydrogen peroxide solution samples to obtain near infrared spectrum data, and form the near infrared spectrum data into an original spectrum matrix, and at the same time form the corresponding prepared concentration data into an original concentration matrix;

[0032] Module 2000: used to remove the optical path error and the wavelength screening from the original spectrum matrix and then form an input matrix with the original concentration matrix;

[0033] Module 3000: used to input the input matrix into the 1D-CNN regression model for training to obtain a detection data feedback model;

[0034] Module 4000: used to obtain the near-infrared spectrum data of the hydrogen peroxide solution to be tested, and after removing the optical path error and band screening, input it into the detection data feedback model to obtain the hydrogen peroxide concentration value of the hydrogen peroxide solution to be tested.

[0035] The above technical solution of the present invention has the following beneficial technical effects:

[0036] By preprocessing the original spectrum, using the competitive adaptive reweighted sampling method to screen the preprocessed spectrum to obtain effective spectral feature points, as the screening spectrum matrix, the screening spectrum matrix and the original concentration matrix form an input matrix, and input it into the 1D-CNN regression model, so as to obtain the hydrogen peroxide concentration value of the hydrogen peroxide solution to be tested. The process of the present invention is concise and clear, and it is easy to operate in the actual production line. The concentration of hydrogen peroxide in the packaging production line can be accurately monitored to eliminate food safety risks. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 Schematic diagram of the steps of the detection method in an embodiment of the present invention;

[0038] Figure 2 It is a schematic diagram of a data processing flow in an embodiment of the present invention;

[0039] Figure 3 1D-CNN model and predicted value correlation diagram in the embodiment of the present invention;

[0040] Figure 4 Schematic diagram of the correlation between the measured value and the predicted value of the detection feedback model in the embodiment of the present invention

[0041] Figure 5 It is a structural block diagram of the detection device in an embodiment of the present invention. DETAILED DESCRIPTION

[0042] In order to make the purpose, technical scheme and advantages of the present invention clearer, the present invention is further described in detail below in conjunction with specific embodiments and with reference to the accompanying drawings. It should be understood that these descriptions are only exemplary and are not intended to limit the scope of the present invention. In addition, in the following description, the description of well-known structures and technologies is omitted to avoid unnecessary confusion of the concept of the present invention.

[0043] The embodiments of the present invention are described below in conjunction with the accompanying drawings. Figure 1 and Figure 2 A method for detecting hydrogen peroxide concentration shown in this embodiment includes the following steps:

[0044] Step 1000: prepare hydrogen peroxide solution samples of different concentrations, and measure the near infrared spectra of the hydrogen peroxide solution samples to obtain near infrared spectrum data, and form an original spectrum matrix with the near infrared spectrum data, and form an original concentration matrix with the corresponding prepared concentration data.

[0045] In this embodiment, a total of 160 hydrogen peroxide solution samples were prepared. To ensure the accuracy of the prepared concentration, the prepared concentration data of the hydrogen peroxide solution samples were confirmed by conventional methods. Then, the near-infrared spectral data of the 160 hydrogen peroxide solution samples and their corresponding prepared concentration data were combined into an original spectral matrix A. i (j×n) and the original concentration matrix Y(n×1), where j represents the number of wavelength variables and n is the number of samples.

[0046] Step 2000: After removing the optical path error and band screening from the original spectrum matrix, the matrix is ​​combined with the original concentration matrix to form an input matrix.

[0047] The specific steps of removing the optical path error from the original spectrum matrix are as follows:

[0048] Step 2100: preprocessing the original spectrum matrix by using multivariate scattering to remove the optical path error, and obtaining a preprocessed spectrum matrix, specifically including the following steps:

[0049] Step 2110: Calculate the average spectrum of the hydrogen peroxide solution sample to obtain an average spectrum matrix;

[0050] Step 2120: Perform regression fitting on the original spectral matrix and the average spectral matrix, solve to obtain the first regression coefficient and construct the first regression function, and obtain the corrected preprocessed spectral matrix through the first regression function.

[0051] The average spectrum matrix calculation formula is: The regression function obtained by regression fitting is: Where m i and b i is the regression coefficient, so the corrected preprocessed spectrum matrix is:

[0052]

[0053] Step 2200: Perform band screening processing on the pre-processed spectral matrix through competitive adaptive reweighted sampling method to obtain a screened spectral matrix.

[0054] Step 2210: Use the Monte Carlo sampling method to randomly sample part of the sample data from the corrected preprocessed spectral matrix into the modeling set, and the remaining sample data into the prediction set, and establish the first least squares model, and repeat the above sampling process several times, and record the absolute value weight of the second regression coefficient in the first least squares model of each sampling process.

[0055] In this embodiment, the Monte Carlo sampling method is used, and the sampling times N is 5 times. The absolute value weight calculation formula of the second regression coefficient is:

[0056]

[0057] Where m is the number of variables remaining in each sampling.

[0058] Step 2220: Use an exponential decay function to remove the relatively small wavelength values ​​in the absolute value weight of the second regression coefficient, and calculate the proportion R of the retained wavelength values i .

[0059] In this embodiment, when the MC sampling is used to establish the first least squares model for the i-th time, the ratio R of the wavelength points retained after removing the relatively small wavelengths in the absolute value weight of the second regression coefficient using the exponential decay function is i for:

[0060]

[0061] Among them, u and k are constants, and the calculation formulas are:

[0062]

[0063] Where N is the number of sampling times.

[0064] Step 2230: At each sampling, use the adaptive weighted sampling method to select (R i ×n) wavelength variables, and PLS (partial least squares) modeling was performed to calculate the cross-validation mean square error value.

[0065] Step 2240: After the sampling is completed, the characteristic wavelength subset and the corresponding cross-validation mean square error value at each sampling are obtained, and the characteristic wavelength subset corresponding to the minimum cross-validation mean square error value is selected as the characteristic wavelength.

[0066] Step 2300: Combine the screening spectrum matrix and the original concentration matrix into an input matrix.

[0067] Step 3000: Input the input matrix into a 1D-CNN (one-dimensional convolutional neural network) regression model for training to obtain a detection data feedback model.

[0068] In this embodiment, the 1D-CNN regression model structure is as follows:

[0069] The first convolutional layer: the number of input channels is 1, the number of output channels is 16, the convolution kernel size is 13, and the stride is 2.

[0070] Batch Normalization Layer: Batch normalization is performed on the output of the previous layer, which helps to speed up the model training process and improve the generalization ability of the model.

[0071] ReLU (Rectified Linear Unit) activation function layer: The output of the previous layer is processed by the ReLU activation function to introduce nonlinearity and enhance the expressiveness of the model.

[0072] The second convolution layer: the number of input channels is 16, the number of output channels is 32, the convolution kernel size is 11, the step size is 2, and the convolution operation on the features continues.

[0073] Repeat batch normalization and ReLU activation function layers.

[0074] Max pooling layer: The pooling window size is 2 and the stride is 1, which is used to downsample the feature map, reduce the amount of computation and retain important features.

[0075] Subsequent convolution, batch normalization, ReLU activation function, and pooling layers are stacked in sequence to gradually extract higher-level features.

[0076] Flattening layer: Flattens the multi-dimensional feature map into a one-dimensional vector so that it can be input into the fully connected layer for processing.

[0077] The first fully connected layer has an input dimension of 768 and an output dimension of 256, which is used to map the feature vector to a higher-dimensional feature space.

[0078] Repeating the ReLU activation function and Dropout layer helps to enhance the nonlinear expression ability of the model and prevent overfitting.

[0079] The second fully connected layer: the input dimension is 256, the output dimension is 64, and the features are continued to be mapped.

[0080] Repeat the ReLU activation function and Dropout layer.

[0081] The third fully connected layer: the input dimension is 64, the output dimension is 32, and the features continue to be mapped.

[0082] The last fully connected layer: the input dimension is 32 and the output dimension is 1, which is used to produce the final prediction result of the model.

[0083] By also including an attention layer after the last convolutional layer of the above 1D-CNN model, the attention weight of each spectrum point is calculated by the attention layer.

[0084] Step 3000 specifically includes the following steps:

[0085] Step 3100: Randomly divide the input matrix into two parts, as a training set and a test set respectively.

[0086] In this embodiment, 80% of the data in the input matrix is ​​randomly selected as a training set, and the remaining data is used as a test set.

[0087] Step 3200: Perform an average pooling operation on the global image to obtain global information, and perform a one-dimensional convolution operation to calculate the one-dimensional convolution kernel size k. c , the calculation formula is:

[0088]

[0089] Where C is the number of channels, |x| odd is the odd number closest to x, b=1, γ=2.

[0090] Step 3300: Obtain the inter-channel relationship weight through the convolution operation, and map the convolution output to [0,1] through the Sigmoid function to obtain the attention weight.

[0091] Step 3400: Apply the attention weight to the output of the convolutional layer to obtain an output weighted result, thereby obtaining a detection data feedback model.

[0092] Step 3500: Use the test data feedback model to test the test set to obtain the predicted value, and compare the predicted value with the actual value to calculate the correlation coefficient r between the two. 2 , correlation coefficient r 2 The closer the value is to 1, the better the prediction effect is. 2 The calculation formula is:

[0093]

[0094] Among them, y i The input value x i The actual value of The input value x i The predicted value of is the average of the actual values.

[0095] At the same time, the root mean square error RMSE is calculated as follows:

[0096]

[0097] The smaller the root mean square error RMSE value is, the stronger the prediction ability of the detection data feedback model is. The correlation diagram of the measured value and the predicted value of the 1D-CNN model is shown in the figure. Figure 3 As shown in the figure, the correlation diagram between the measured value and the predicted value of the detection feedback model is as follows Figure 4 As shown in the figure, the correlation coefficient r between the 1D-CNN model and the detection data feedback model 2 The comparison table of root mean square error RMSE is shown in Table 1:

[0098] Table 1

[0099]

[0100] Step 4000: Acquire near infrared spectrum data of a hydrogen peroxide solution to be tested for hydrogen peroxide concentration, remove optical path error and band screening, and input the data into a detection data feedback model to obtain a hydrogen peroxide concentration value of the hydrogen peroxide solution to be tested for hydrogen peroxide concentration.

[0101] refer to Figure 5 The present invention also provides a hydrogen peroxide concentration detection device for a food packaging production line, comprising the following modules:

[0102] Module 1000: used to prepare hydrogen peroxide solution samples of different concentrations, and measure the near infrared spectrum of the hydrogen peroxide solution samples to obtain near infrared spectrum data, and form the near infrared spectrum data into an original spectrum matrix, and at the same time form the corresponding prepared concentration data into an original concentration matrix;

[0103] Module 2000: used to remove the optical path error and band screening from the original spectrum matrix and then form an input matrix with the original concentration matrix;

[0104] Module 3000: used to input the input matrix into the 1D-CNN regression model for training to obtain a detection data feedback model;

[0105] Module 4000: used to obtain near-infrared spectrum data of a hydrogen peroxide solution to be tested for hydrogen peroxide concentration, and after removing optical path error and band screening, input the data into a detection data feedback model to obtain a hydrogen peroxide concentration value of the hydrogen peroxide solution to be tested for hydrogen peroxide concentration.

[0106] It should be understood that the above specific embodiments of the present invention are only used to illustrate or explain the principles of the present invention, and do not constitute a limitation of the present invention. Therefore, any modifications, equivalent substitutions, improvements, etc. made without departing from the spirit and scope of the present invention should be included in the protection scope of the present invention. In addition, the appended claims of the present invention are intended to cover all changes and modifications that fall within the scope and boundaries of the appended claims, or the equivalent forms of such scope and boundaries.

Claims

1. A method for detecting hydrogen peroxide concentration in a food packaging production line, characterized in that: The steps include: Step 1000: preparing hydrogen peroxide solution samples of different concentrations, and measuring the near infrared spectrum of the hydrogen peroxide solution samples to obtain near infrared spectrum data, forming an original spectrum matrix with the near infrared spectrum data, and forming an original concentration matrix with the corresponding prepared concentration data; Step 2000: removing the optical path error and the wavelength screening from the original spectrum matrix and combining it with the original concentration matrix to form an input matrix; Step 3000: inputting the input matrix into a 1D-CNN regression model for training to obtain a detection data feedback model; Step 4000: Acquire the near infrared spectrum data of the hydrogen peroxide solution to be tested, remove the optical path error and the band screening, and input the data into the detection data feedback model to obtain the hydrogen peroxide concentration value of the hydrogen peroxide solution to be tested.

2. The method for detecting hydrogen peroxide concentration in a food packaging production line according to claim 1, characterized in that: The step 2000 includes: Step 2100: preprocessing the original spectrum matrix by using multivariate scattering to remove optical path errors, thereby obtaining a preprocessed spectrum matrix; Step 2200: performing band screening processing on the pre-processed spectrum matrix by competitive adaptive reweighted sampling method to obtain a screened spectrum matrix; Step 2300: Combining the screened spectrum matrix and the original concentration matrix into the input matrix.

3. The method for detecting hydrogen peroxide concentration in a food packaging production line according to claim 2, characterized in that: The step 2100 includes the following steps: Step 2110: Calculate the average spectrum of the hydrogen peroxide solution sample to obtain an average spectrum matrix; Step 2120: Perform regression fitting on the original spectral matrix and the average spectral matrix, solve to obtain a first regression coefficient and construct a first regression function, and obtain the corrected preprocessed spectral matrix through the first regression function.

4. The method for detecting hydrogen peroxide concentration in a food packaging production line according to claim 3, characterized in that: The step 2200 includes the following steps: Step 2210: using the Monte Carlo sampling method to randomly sample part of the sample data from the corrected preprocessed spectrum matrix into the modeling set, and the remaining sample data into the prediction set, and establish a first least squares model, and repeat the above sampling process several times, and record the absolute value weight of the second regression coefficient in the first least squares model of each sampling process; Step 2220: Use an exponential decay function to remove the relatively small wavelength values ​​in the absolute value weight of the second regression coefficient, and calculate the ratio R of the retained wavelength points. i ; Step 2230: At each sampling, the number of variables selected by adaptive weighted sampling is (R i ×n) wavelength variables, and perform PLS modeling to calculate the cross-validation mean square error; Step 2240: After the sampling is completed, the characteristic wavelength subset and the corresponding cross-validation mean square error value at each sampling are obtained, and the characteristic wavelength subset corresponding to the minimum cross-validation mean square error value is selected as the characteristic wavelength.

5. The method for detecting hydrogen peroxide concentration in a food packaging production line according to claim 1, characterized in that: The step 3000 specifically includes the following steps: Step 3100: randomly divide the input matrix into two parts, respectively serving as a training matrix and a test matrix; Step 3200: Perform an average pooling operation on the global image to obtain global information, and perform a one-dimensional convolution operation to calculate the one-dimensional convolution kernel size k. c , the calculation formula is: Where C is the number of channels, |x| odd is the odd number closest to x, b = 1, γ = 2; Step 3300: Obtain the inter-channel relationship weight through convolution operation, and map the convolution output to [0, 1] through Sigmoid function to obtain the attention weight; Step 3400: Apply the attention weight to the output of the convolutional layer to obtain an output weighted result, thereby obtaining a detection data feedback model.

6. The method for detecting hydrogen peroxide concentration in a food packaging production line according to claim 5, characterized in that: Step 3500: Use the test data feedback model to test the test set to obtain the predicted value, and compare the predicted value with the actual value to calculate the correlation coefficient r between the two. 2 and root mean square error RMSE.

7. A hydrogen peroxide concentration detection device for a food packaging production line, characterized in that: Includes the following modules: Module 1000: used to prepare hydrogen peroxide solution samples of different concentrations, and measure the near infrared spectrum of the hydrogen peroxide solution samples to obtain near infrared spectrum data, and form the near infrared spectrum data into an original spectrum matrix, and at the same time form the corresponding prepared concentration data into an original concentration matrix; Module 2000: used to remove the optical path error and the wavelength screening from the original spectrum matrix and then form an input matrix with the original concentration matrix; Module 3000: used to input the input matrix into the 1D-CNN regression model for training to obtain a detection data feedback model; Module 4000: used to obtain the near-infrared spectrum data of the hydrogen peroxide solution to be tested, and after removing the optical path error and band screening, input it into the detection data feedback model to obtain the hydrogen peroxide concentration value of the hydrogen peroxide solution to be tested.

Images