A method for online monitoring of gel strength change in surimi thermal processing based on hyperspectral imaging technology

Abstract

The present application relates to a kind of methods for monitoring the change of gel strength in surimi thermal processing process based on hyperspectral imaging technology online, comprising: collecting the original hyperspectral data of surimi sample in heating process, and determining the gel strength of the surimi sample;The original hyperspectral data is pretreated;Based on the original hyperspectral data and gel strength after pretreatment, a data set is constructed;The data set is used to train and verify PLS model, and the gel strength prediction model is obtained;The gel strength prediction model is used to monitor the change of gel strength in surimi thermal processing process.The present application can realize the online, fast and non-destructive monitoring of gel strength in industrial surimi two-stage heating process.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of food quality detection, in particular to a method for online monitoring of gel strength change in surimi thermal processing based on hyperspectral imaging technology. BACKGROUND

[0002] In recent years, with the rapid development of society and the guarantee of abundant fishery resources, surimi and its products are deeply loved by consumers due to their good gelatinization characteristics, high nutritional value, crisp taste, instant nature and compliance with the current dietary concept of the public. At present, surimi raw materials are mainly used to produce surimi products such as fish balls, fish cakes, fish rolls, fish sausages, fish tofu and aquatic simulation foods (such as artificial crab meat, shrimp meat, shark fin and shell column), which have high economic value. However, due to the influence of various factors on surimi during processing, the yield and quality of surimi are seriously affected, so it is crucial to develop a method for online monitoring of gel strength rapid and non-destructive detection in surimi thermal processing.

[0003] In recent years, other non-destructive technologies for monitoring quality changes during food processing include infrared imaging, near-infrared spectroscopy, machine vision and electronic nose, etc. These technologies collect relatively single sample information, while hyperspectral imaging technology can obtain both spectral data and image information of the sample, greatly enriching the data set and realizing visual quality analysis. Therefore, the method for online monitoring of gel strength change in surimi thermal processing has important practical significance. SUMMARY

[0004] The purpose of the present application is to provide a method for online monitoring of gel strength change in surimi thermal processing based on hyperspectral imaging technology, to solve the problems existing in the prior art and realize online, rapid and non-destructive monitoring of gel strength in industrial surimi two-stage heating process.

[0005] To achieve the above purpose, the present application provides the following scheme:

[0006] A method for online monitoring of gel strength change in surimi thermal processing based on hyperspectral imaging technology, comprising:

[0007] Collecting original hyperspectral data of surimi samples during heating and measuring the gel strength of the surimi samples;

[0008] Pretreating the original hyperspectral data;

[0009] Based on the pretreated original hyperspectral data and the gel strength, constructing a data set;

[0010] Using the data set to train and verify the PLS model to obtain a gel strength prediction model;

[0011] Using the gel strength prediction model, monitoring the change of gel strength in the surimi thermal processing process.

[0012] Optionally, the preprocessing of the original hyperspectral data comprises:

[0013] Black and white correction is performed on the original hyperspectral data; wherein the original hyperspectral data comprises VNIR and NIR dual-band hyperspectral data.

[0014] An area of interest of the corrected hyperspectral data is extracted.

[0015] The mean value of the spectral reflectance of all pixels in the area of interest is obtained as the average spectral data of the surimi sample.

[0016] The average spectral data is processed by standard normal variable transformation, multivariate scatter correction, first derivative and second derivative.

[0017] Optionally, the method for black and white correction of the original hyperspectral data is:

[0018]

[0019] Wherein, I represents the corrected reflectance hyperspectral image data, I0 represents the dark image, Rb represents the original hyperspectral image, and Rw represents the white reference image.

[0020] Optionally, the expression of the PLS model is:

[0021] Y=XA+B=XW*C+B=DC+B

[0022] Z*=Z(P'Z) -1

[0023] Wherein, A is the PLS coefficient, B is the residual matrix of Y, D is the score matrix of X, Z is the PLS weight, P and C are the loadings of X and Y respectively, Z* is the regression coefficient matrix, X is the independent variable, Y is the dependent variable, and P' is the transpose of the loading matrix of the independent variable matrix X.

[0024] Optionally, before training the PLS model using the data set, it comprises:

[0025] The PLS model is simplified using VCPA, VCPA-IRIV, VCPA-GA and IRIV algorithms.

[0026] Optionally, the simplification of the PLS model using VCPA, VCPA-IRIV, VCPA-GA and IRIV algorithms comprises:

[0027] Based on the VCPA algorithm, an exponential decreasing function is used to gradually reduce the number of variables; and a binary matrix sampling is used to create a subset;

[0028] Based on the IRIV algorithm, the usefulness of the variables is evaluated by observing the change of RMSECV after adding or deleting variables;

[0029] Based on the VCPA-GA algorithm, the variable space is narrowed, and the process of natural selection, crossover and mutation is simulated to further optimize the selected variables to find the optimal variable combination;

[0030] Based on the VCPA-IRIV algorithm, the variable space is first reduced by VCPA, and then classified and iterated by IRIV to optimize the selection.

[0031] Optionally, collecting the original hyperspectral data of the surimi sample in the heating process comprises:

[0032] Placing the container containing the surimi on the moving platform to align the spectral acquisition device;

[0033] When the platform moves, the spectral acquisition device acquires complete spectral information of the surimi.

[0034] The present application has the following advantages:

[0035] The present application pretreats the surimi, then performs water bath two-stage heating, collects hyperspectral data in the VNIR (400-1000nm) and NIR (900-1700nm) bands during the thermal processing process, and measures the gel strength. The obtained hyperspectral data is black and white corrected, ENVI is used to collect the spectra of all pixels in the region of interest, and the average is taken to represent a sample. The spectral data is pretreated, the optimal pretreatment method is determined, the spectral data and the gel strength are used to construct a partial least squares regression model, the variable screening algorithm is used to obtain the best model effect in the VNIR and NIR bands during the thermal processing process, and the change of the gel strength of the surimi during the thermal processing process is visualized, so that the change of the gel strength during the thermal processing process can be monitored online in practical application, and the visualization of the gel strength is realized, so that non-professionals can clearly understand the specific situation during the thermal processing process. BRIEF DESCRIPTION OF DRAWINGS

[0036] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0037] Figure 1A process flow diagram of a method for online monitoring of gel strength change in a surimi thermal processing process based on hyperspectral imaging technology according to an embodiment of the present application;

[0038] Figure 2 A feature variable result graph extracted by four variable screening algorithms according to an embodiment of the present application;

[0039] Figure 3 An optimal model parameter diagram according to an embodiment of the present application; wherein (a) is a principal component score diagram of the best PLS model cross-validation, and (b) is a result regression diagram of the best model;

[0040] Figure 4 A visual analysis diagram of the optimal model according to an embodiment of the present application. DETAILED DESCRIPTION

[0041] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0042] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0043] As shown in the drawings, Figure 1 the present embodiment proposes a method for online monitoring of gel strength change in a surimi thermal processing process based on hyperspectral imaging technology, comprising:

[0044] Collecting original hyperspectral data of surimi samples in a heating process, and measuring the gel strength of the surimi samples;

[0045] Pretreating the original hyperspectral data;

[0046] Based on the pretreated original hyperspectral data and the gel strength, constructing a data set;

[0047] Training and verifying a PLS model using the data set, and obtaining a gel strength prediction model;

[0048] Monitoring the gel strength change in the surimi thermal processing process using the gel strength prediction model.

[0049] Specifically, in the embodiment, a method for online monitoring of gel strength change in surimi thermal processing based on hyperspectral imaging technology, comprising: obtaining live dry kg frozen silver carp surimi (AA grade) from Xiamen Anjing, thawing the surimi and putting it into a meat grinder to break it up, adding 2.5% surimi and water to adjust the moisture content to 80% and loading it into an aluminum box; heating at 40°C for 30 minutes, heating at 90°C for 20 minutes, taking 30 samples every 5 minutes, a total of 300 samples; collecting hyperspectral data of the surimi samples and measuring the gel strength of the surimi samples; extracting the spectral data of the sample hyperspectral region of interest, constructing a PLS quantitative model of the spectral data and the surimi gel strength; simplifying the model and improving the robustness of the model using VCPA, VCPA-IRIV, VCPA-GA, IRIV variable screening algorithms; the embodiment can realize online, rapid and non-destructive monitoring of the gel strength of the surimi in the two-stage heating process in industry.

[0050] Wherein, simplifying the PLS model is to reduce the variables for constructing the PLS model, and the time for model calculation is smaller. That is, these algorithms screen important variables by the contribution rate and importance of variables to the model.

[0051] Further, collecting the original hyperspectral data of the surimi samples during the heating process comprises:

[0052] Placing the container containing the surimi on the moving platform to align the spectral acquisition device;

[0053] When the platform moves, the spectral acquisition device acquires complete spectral information of the surimi.

[0054] Specifically, in the embodiment, first, the surimi is pretreated and placed in an aluminum box for two-stage heating;

[0055] The frozen surimi used in the embodiment is AA grade, comprising: thawing the surimi, cutting it into small pieces, putting it into a meat grinder and stirring for 3 minutes, adding 2% Nacl, adjusting the moisture content to 80%, continuing to stir for 3 minutes, and placing it in a cylindrical aluminum box with a diameter of 2 cm and a height of 2 cm for heating, heating at 40°C for 30 minutes, and heating at 90°C for 20 minutes.

[0056] Secondly, collect the hyperspectral data of the surimi during the heating process, and measure the gel strength of each surimi, a total of 300 data;

[0057] In the embodiment, the VNIR and NIR dual-band hyperspectral data are collected every 5 minutes during the heating process, and the gel strength of the surimi is measured immediately after the collection is completed, the test speed is 1 mm / s, the compression distance is 12 mm, and the trigger force is 15N.

[0058] Gel strength = breaking force (g) x breaking distance (cm)

[0059] In the high spectral image acquisition operation, the hyperspectral imaging system used in the embodiment is a hyperspectral imaging instrument, a platform control system, a computer, etc., wherein the NIR-HSI (900-1700 nm) system is composed of a spectral resolution of 4 nm imaging spectrometer (N17E, Specim, Finland), a resolution of 640*512 camera, a camera lens (640 miniRaptor, Northern Ireland), two tungsten lamps (LS-150, Wuliu optics, Taiwan), a moving platform driven by a stepper motor (HSIM-800, Wuliu optics, Taiwan) and a computer with software.

[0060] The specific method of hyperspectral image acquisition is: place the aluminum box containing surimi on the moving platform and align the camera. When the platform moves, the hyperspectral imaging instrument obtains the spectral information of a line of surimi spatial position in the entire spectral region, then the platform drives the aluminum box to move to obtain the spectral information of surimi at other positions, until the complete spectral information of the sample is obtained. The moving speed of the sample stage in the VNIR spectral imaging system is 8.3 mm / s, the exposure time is set to 90 ms, and the image resolution is 1604*605 pixel; because the spectral scanning range is 678 effective wavelengths in the 400-1000 nm band, the finally obtained three-dimensional data module size is 1604*605*678. The spectral resolution of the NIR spectral imaging instrument is 4 nm, the resolution of the 640*512 camera and the camera lens, and the image resolution is 640*459 pixel, because the spectral scanning range is 678 effective wavelengths in the 400-1000 nm band, so the finally obtained three-dimensional data module size is 640*459*512. The entire acquisition process is carried out in a dark box, and whiteboard and blackboard information is collected for subsequent correction to prevent light in the environment from affecting the collected hyperspectral images.

[0061] Further, the preprocessing of the original hyperspectral data includes:

[0062] The original hyperspectral data includes: VNIR and NIR dual-band hyperspectral data;

[0063] Extracting the region of interest of the corrected hyperspectral data;

[0064] Obtaining the mean value of the spectral reflectance of all pixels in the region of interest as the average spectral data of the surimi sample;

[0065] The average spectral data is processed by standard normal variable transformation, multivariate scatter correction, first derivative and second derivative.

[0066] Specifically, in the present embodiment, a region of interest (ROI) of the corrected hyperspectral image is extracted, and the mean value of the spectral reflectance of all pixels in the ROI is taken as the average spectrum of the surimi sample.

[0067] The black and white correction of the hyperspectral data of the present embodiment uses the following formula:

[0068]

[0069] wherein the corrected reflectance hyperspectral image I is expressed in relative reflectance (%); Rb represents the original hyperspectral image; I0 is a dark image (0% reflectance), and Rw is a white reference image (100% reflectance).

[0070] In the preferred embodiment, a circular region with a radius of 60 pixels is taken as the region of interest. The range of the region of interest is determined according to the circular aluminum box containing the surimi. In order to ensure that the region of interest is a complete circular region in the case of direct online monitoring of the surimi container, the present embodiment sets the radius to 60 pixels to realize online monitoring of the change in gel strength during the thermal processing of surimi. A total of 600 spectral data of surimi are obtained under the dual-band.

[0071] The obtained average spectral data is preprocessed, and the preprocessed average spectral data and the gel strength are divided into a training set and a prediction set in a ratio of 7:3 for subsequent modeling.

[0072] The present embodiment uses the standard normal variable transformation, multivariate scatter correction, first derivative, second derivative and other preprocessing methods on the average spectrum to obtain hyperspectral data with smooth spectral information after removing noise.

[0073] Due to the characteristics of light scattering on the surface of the surimi sample and the noise and interference caused by the instrument, the above four preprocessing methods are used to obtain the preprocessed spectral data, and the K-S grouping method is used to divide the data set into a training set and a prediction set in a ratio of 7:3.

[0074] Specifically, in the present embodiment, the data set is further used to train the PLS model before the PLS model is trained.

[0075] The PLS model is constructed using the gel strength and the spectral data, the PLS model is simplified using the VCPA, VCPA-IRIV (variable combination population analysis-iterative reserved information variable), VCPA-GA (variable combination population analysis-genetic algorithm) and IRIV (iterative reserved information variable) algorithms, and the optimal model is used for visual analysis.

[0076] The inspiration of VCPA comes from the "survival of the fittest" principle observed in natural evolution, which uses an exponentially decreasing function (EDF) to gradually reduce the number of variables. Binary matrix sampling is then employed to ensure equal selection opportunities for each variable, creating various subsets. Model overall analysis is used to identify the subset with the lowest cross-validated root mean square error (RMSECV).

[0077] IRIV is a new variable selection technique based on binary matrix transformation filtering method, which evaluates the usefulness of variables by observing the change in RMSECV after adding or deleting variables. After several iterations, IRIV retains strong and weak information variables while eliminating non-informative and interfering variables. The retained variables are then reversed to determine the optimal subset.

[0078] VCPA-GA further optimizes the selected variables by simulating the process of natural selection, crossover, and mutation after VCPA reduces the variable space and sets the final EDF residual variable to 100, to find the optimal variable combination.

[0079] The hybrid VCPA-IRIV strategy method combines VCPA and IRIV to handle high-dimensional spectral data. VCPA first reduces the variable space, and IRIV optimizes selection by categorizing and iterating variables to delete less informative variables. This hybrid method utilizes the systematic reduction of VCPA and the iterative filtering function of IRIV to optimize the variable selection process.

[0080] The task of simplifying the PLS model is to reduce the dimensionality of the X matrix by selecting important variables (features), making the PLS model more concise, accurate, and robust. When VCPA and its hybrid strategy algorithms perform stepwise regression, the model optimizes prediction performance by gradually adding or removing independent variables. Cross-validation techniques can help evaluate the impact of different feature subsets on model generalization, selecting the most suitable variable subset to simplify the PLS model. Ultimately, the goal of variable selection is to reduce the number of variables in the dataset, making the input dataset smaller and the model run faster.

[0081] In this implementation, PLS modeling is performed using pretreated spectral data and gel strength of surimi. The optimal pretreatment method is determined, and the results are shown in Table 1. Then, based on the optimal pretreatment PLS model, VCPA, VCPA-IRIV, VCPA-GA, and IRIV algorithms are used to simplify the PLS model, and the results are shown in Table 2. The selected wavelengths are shown in Figure 2 , and the optimal model is shown in Figure 3 , where Figure 3 (a) is the principal component score of the best PLS model cross-validation, Figure 3 (b) is the result regression of the best model, and the visualization analysis is performed using the optimal model, as shown inFigure 4 PLS is shown.

[0082] PLS projects the independent variables X and dependent variables Y into a latent variable space to build a linear regression model. This process takes into account the covariance information between X and Y to maximize the correlation between X and Y. PLS decomposes X and Y into several X scores (D) to build the PLS model using the following equation:

[0083] Y = XA + B = XW*C + B = DC + B

[0084] Z* = Z(P'Z) -1

[0085] where A is the PLS coefficients, B is the Y residual matrix, D is the X scores matrix, Z is the PLS weights, P and C are the loadings of X and Y, respectively, and Z* is the regression coefficient matrix. The set of data projected from the spectral data is called the orthogonal factors of the “latent variables”. The optimal number of orthogonal factors depends on the prediction error, usually achieved by using the lowest value of the prediction residual error sum of squares (PRESS).

[0086] Table 1. PLS model results based on different pretreatment methods at two wavelengths

[0087]

[0088] Variable Combination Population Analysis (VCPA), this strategy includes two key processes. First, the Exponential Decline Function (EDF) is a simple and effective “survival of the fittest” principle in Darwinian natural evolution, which is used to determine the number of variables to be kept and constantly reduced in the variable space. Assuming the EDF is set to N, it means it will go through N runs to iteratively filter the variables. In other words, the variable space is reduced by N times. The formula for calculating the ratio of remaining variables in the i-th run of EDF is shown in (1). Second, in each EDF run, the Binary Matrix Sampling (BMS) strategy is used to provide the same selection opportunity for each variable and generate different variable combinations to generate the subset population to build the sub-model population. Then, Model Population Analysis (MPA) is adopted to find the variable subsets with lower cross-validation root mean square error (RMSECV). The frequency of each variable appearing in the best 10% of sub-models is calculated. The higher the frequency, the more important the variable.

[0089] r i = e -θi (1)

[0090] where θ is a constant parameter that controls the EDF curve. θ is related to the curvature of the EDF, and is positively correlated with the speed of the descending curve. It can be calculated under the following conditions: (1) at the beginning, i is equal to 0, and all p variables are used for modeling, resulting in r0=1; (2) at the rN In the next iteration, there are still ω variables left; then r N is equal to ω / p. ω is the number of variables left after running N times. Under the above condition, θ can be determined as: Under the above condition, θ can be determined as:

[0091]

[0092] Based on the hybrid strategy of VCPA, VCPA continuously narrows the variable space based on EDF, eventually making large variable space small and optimized. To solve the current limitation of GA and IRIV on a large number of variables, VCPA is modified and coupled with GA and IRIV, resulting in a hybrid strategy for variable selection. In addition, this hybrid strategy can help VCPA make up for its defect of tending to select too few variables. The hybrid strategy based on VCPA includes the following two steps: Step 1: Perform VCPA to narrow the variable space. For the modified VCPA, in this work, ω is set to 100 in the EDF step, which means there are still 100 variables that need to be further optimized by GA and IRIV. Step 2: Perform GA and IRIV to further optimize the remaining 100 variables. These 100 variables are gradually preserved by eliminating other variables that contribute little to N iterations of EDF. The remaining variables span a small and optimized space, making it easier and better for GA and IRIV to select the best subset of variables.

[0093] Table 2 PLS model results under dual-band based on four variable screening algorithms

[0094]

[0095] The spectral values corresponding to the pixel points in the prepared surimi hyperspectral image are extracted, and the selected spectral values are substituted into the optimal model to realize the prediction of the gel strength information of each pixel point in the surimi hyperspectral image. Finally, through information fusion of hyperspectral imaging technology, the distribution diagram of the information in the surimi heat processing process on the plane is obtained according to the coordinate information of the pixels and the corresponding adulteration concentration, as shown in FIG. 4. Figure 4 In the gel strength distribution diagram, the color is from blue to red. The redder the color, the higher the adulteration concentration, and the bluer the color, the lower the adulteration concentration.

[0096] The embodiment proposes a method for online monitoring the change of gel strength in the surimi heat processing process. A PLS regression model is constructed between the spectral data of surimi in the heat processing process and the gel strength, and a feature screening method is used to simplify the model to determine the optimal model for visual analysis. In actual processing application, the surimi does not need to be sampled and detected, and no pollution is generated, so as to achieve the effect of online monitoring.

[0097] The above described embodiments are only to illustrate the preferred modes of the present application, and are not intended to limit the scope of the present application. Any modification and improvement made by those skilled in the art to the technical solutions of the present application without departing from the design spirit of the present application shall fall within the protection scope of the present application as defined by the claims.

Claims

1. A method for online monitoring the change of gel strength during thermal processing of surimi based on hyperspectral imaging technology, characterized in that, The application relates to a method for monitoring the gel strength of surimi during a heating process. The method comprises the following steps: collecting original hyperspectral data of surimi samples during a heating process and measuring the gel strength of the surimi samples; preprocessing the original hyperspectral data; constructing a data set based on the preprocessed original hyperspectral data and the gel strength; training and verifying a PLS model by using the data set to obtain a gel strength prediction model; monitoring the change of the gel strength of surimi during a heating process by using the gel strength prediction model; The preprocessing of the original hyperspectral data comprises the following steps: black and white correction of the original hyperspectral data; wherein the original hyperspectral data comprises VNIR and NIR dual-band hyperspectral data; extracting a region of interest from the corrected hyperspectral data; obtaining the mean value of the reflectivity of all pixel spectra in the region of interest as the average spectral data of the surimi sample; performing standard normal variable transformation, multivariate scatter correction, first derivative and second derivative processing on the average spectral data; Before training the PLS model by using the data set, the method further comprises the following steps:

2. The method for monitoring the change of gel strength in the thermal processing of surimi online based on hyperspectral imaging technology according to claim 1, characterized in that, simplifying the PLS model by using VCPA, VCPA-IRIV, VCPA-GA and IRIV algorithms. wherein, represents corrected reflectance hyperspectral image data, represents a dark image, represents an original hyperspectral image, represents a white reference image.

3. The method for monitoring the change of gel strength in the thermal processing of surimi online based on hyperspectral imaging technology according to claim 1, characterized in that, The method for black and white correction of the original hyperspectral data comprises the following steps: The expression of the PLS model is as follows: Z* = Z(P'Z) -1 Y=XA+B=XZ*C+B=DC+B 4. The method for monitoring the change of gel strength in the thermal processing of surimi online based on hyperspectral imaging technology according to claim 1, characterized in that, wherein A is a PLS coefficient, B is a residual matrix of Y, D is a score matrix of X, Z is a PLS weight, P and C are loads of X and Y respectively, Z* is a regression coefficient matrix, X is an independent variable, Y is a dependent variable, and P' is the transpose of the load matrix of the independent variable matrix X. The simplification of the PLS model by using VCPA, VCPA-IRIV, VCPA-GA and IRIV algorithms comprises the following steps: based on the VCPA algorithm, an exponential decreasing function is used to gradually reduce the number of variables; and a binary matrix sampling is used to create a subset; based on the IRIV algorithm, the usefulness of a variable is evaluated by observing the change of RMSECV after adding or deleting the variable; based on the VCPA-GA algorithm, the variable space is reduced, and the selected variables are further optimized by simulating the process of natural selection, crossover and mutation to find the optimal variable combination; 5. The method for online monitoring the change of gel strength during thermal processing of surimi based on hyperspectral imaging technology according to claim 1, characterized in that, based on the VCPA-IRIV algorithm, the variable space is first reduced by VCPA, and then the variables are classified and iterated by IRIV to optimize the selection. The collection of original hyperspectral data of surimi samples during a heating process comprises the following steps: placing a container containing surimi on a moving platform to align with a spectral acquisition device; when the platform moves, acquiring complete spectral information of the surimi by the spectral acquisition device.

Images