Method and device for rapidly detecting adulteration of cod surimi based on hyperspectral imaging technology

Abstract

The invention provides a rapid detection method and device for adulteration of cod surimi based on a hyperspectral imaging technology, and the method comprises the steps: preparing a to-be-detected cod surimi sample containing different component adulteration concentration proportions, collecting hyperspectral image information of the to-be-detected cod surimi sample, and carrying out black and white correction processing; extracting an ROI (Region of Interest) of the hyperspectral image after black and white correction; dividing all the obtained spectral data into a training set and a prediction set, and constructing a BP-ANN classification model; preprocessing the spectral data set, constructing a PLS regression model, and determining an optimal preprocessing mode; two variable screening algorithms, namely a self-adaptive reweighting algorithm CARS and an interval combinatorial optimization algorithm ICO, are adopted to simplify the model, the robustness of the model is improved, and the optimal model is selected. By adopting the hyperspectral imaging technology and combining the traditional chemometrics and deep learning, the method has the advantages of no damage, no pollution, automation, rapidness and high efficiency in the detection of the adulteration of the cod surimi.

Description

Technical Field

[0001] The invention belongs to the technical field of food quality detection, and in particular relates to a method and device for rapid detection of cod surimi adulteration based on hyperspectral imaging technology. Background Art

[0002] Cod is one of the original raw materials for surimi. It is a cold-water deep-sea fish that lives at the bottom of the deep sea and has a low degree of pollution. As a marine fish with high economic and nutritional value, its global annual production is about 6.22 million tons, accounting for nearly 8% of the total marine catch. However, overfishing, climate change and ocean acidification are threatening the sustainable development of cod resources, and the cod market is facing severe challenges. In addition, with the increasing demand for cod, the main producing areas of cod have begun to use methods such as restricting fishing and improving fishery development and breeding to respond. Therefore, the probability of adulteration of cod surimi has increased significantly, and problems such as adulteration, inferior products, and mislabeling are common. Therefore, further exploring methods for efficiently identifying the authenticity of cod surimi is crucial to its quality and safety issues and improving public health.

[0003] At present, most of the methods used in meat identification technology are loop-mediated isothermal amplification, DNA barcoding, liquid chromatography-mass spectrometry, enzyme-linked immunosorbent assay, CRISPR-Cas system, spectroscopy and electronic nose. Among them, the most widely used DNA-based fish species identification technology, such as real-time PCR, multiplex PCR, nuclear DNA barcoding and novel loop-mediated isothermal amplification. However, these DNA-based technologies are destructive, time-consuming and professional, and are not suitable for rapid and non-destructive differentiation of different fish.

[0004] Therefore, rapid and nondestructive detection technology for cod surimi adulteration will certainly have very important practical significance. Summary of the invention

[0005] The present invention provides a method and device for rapid detection of cod surimi adulteration based on hyperspectral imaging technology. Hyperspectral imaging technology is adopted. Hyperspectral images have the advantage of unified graphs and spectra. It combines traditional chemometrics and deep learning, and has the advantages of non-destructive, non-polluting, automated, rapid and efficient detection of cod surimi adulteration to solve the above-mentioned technical defects.

[0006] In a first aspect, the present invention provides a method for rapid detection of adulteration of cod surimi based on hyperspectral imaging technology, the method comprising the following steps:

[0007] S1. Prepare cod surimi samples to be tested containing different adulteration concentration ratios, collect hyperspectral image information of the cod surimi samples to be tested, and perform black and white correction processing;

[0008] S2, extracting the region of interest ROI of the hyperspectral image after black-and-white correction, and calculating the average spectral reflectance of all pixels in the region of interest ROI domain as the average spectral data of a cod surimi sample to be detected;

[0009] S3, dividing all the obtained spectral data into a training set and a prediction set, and constructing a BP-ANN classification model;

[0010] S4, preprocessing the spectral data set, building a PLS regression model, and determining the best preprocessing method;

[0011] S5. Adopt two variable screening algorithms, namely, the adaptive reweighting algorithm CARS and the interval combination optimization algorithm ICO, to simplify the model, improve the model robustness, and select the best model; and

[0012] S6. Determine the best model and visualize the adulteration concentration model to quickly detect adulteration.

[0013] Preferably, step S3 divides all the obtained spectral data into a training set and a prediction set, including:

[0014] The full wavelength is used as the input parameter of the BP-ANN network model, and the sample category is used as the model output parameter. The input quantity in the model is 476, the output quantity is the sample adulteration concentration gradient 11, and the BP-ANN model training error is set to 10 -6 , the network training times is 1000.

[0015] Further preferably, the constructing of the BP-ANN classification model includes:

[0016] The network parameters of the model are: the hidden layer activation function is tansig, the output layer activation function is purelin, the training function is tansig, the number of nodes in the first hidden layer is 8, and the number of nodes in the second hidden layer is 88; the calculation formulas of tansig and purelin functions are as follows:

[0017]

[0018] purelin(x)=x

[0019] Among them, x is the weighted input of the neuron, the tansig function converts x to between -1 and 1; purelin(x) is the output of the neuron.

[0020] Preferably, step S4 of preprocessing the spectral data set includes:

[0021] The obtained average spectrum is preprocessed using standard normal variable transformation, multivariate scattering correction, first-order derivative, and second-order derivative methods to obtain hyperspectral data with smooth spectral information after noise removal.

[0022] Preferably, the constructing of the PLS regression model and determining the optimal preprocessing method comprises:

[0023] PLS projects the independent variable X and the dependent variable Y into the latent variable space to establish a linear regression model. PLS decomposes X and Y into several X scores D and constructs the PLS model using the following equation:

[0024] Y=XA+B=XW*C+B=DC+B

[0025] Z*=Z(P'Z) -1

[0026] Where A is the PLS coefficient, B is the Y residual matrix, D is the score matrix of X, Z is the PLS weight, P and C are the loadings of X and Y respectively, and Z* is the regression coefficient matrix.

[0027] Preferably, step S5 uses two variable screening algorithms, namely, an adaptive reweighting algorithm CARS and an interval combination optimization algorithm ICO, to simplify the model, specifically including:

[0028] CARS adopts the adaptive weighted sampling ARS strategy, which stores the sample data of each sampling in the PLS model, and screens them according to the absolute value weight of the regression coefficient, selects samples with high weights as new subsets, and eliminates samples with low weights;

[0029] The number of Monte Carlo cross-validation was set to 1000, the proportion of calibration samples collected each time was 0.8, the root mean square error value RMSECV of each interval model was calculated by the 5-fold interactive validation method, and the number of iterations was set to 50; the PLS model was established, and multiple iterative calculations were performed, and the bands in the smallest subset of the root mean square error RMSECV of the PLS model interactive validation were selected as feature bands;

[0030] The weight calculation formula for adaptive weighted sampling is as follows:

[0031]

[0032] Among them, b i is the regression coefficient of the ith variable, m is the number of variables in the current variable set, and this formula is used to normalize the absolute value of the regression coefficient of each variable so that the sum of their weights is 1.

[0033] Further preferably, the interval combinatorial optimization algorithm ICO is a chemical modeling algorithm based on the model cluster analysis strategy MPA framework, specifically including:

[0034] Divide the variables into several parts, each of which is an interval; then use the weighted bootstrap sampling method to generate subsets of different variable intervals, with the initial weight of each variable being 1; establish PLS models for these intervals respectively, and calculate the RMSECV value of each interval model through the 5-fold interactive validation method. The optimal model set is when the RMSECV value is the minimum.

[0035] Calculate the frequency of each variable interval in the optimal model, which is the weight of the next sampling, and calculate the RMSECV mean of the optimal model for this sampling; repeat multiple iterations until the RMSECV mean is higher than the previous round; at this time, the group of variable intervals with the smallest RMSECV mean is regarded as the final selected variable interval; the formula for calculating the sampling weight of each interval is as follows:

[0036]

[0037] Among them, f a is the frequency of occurrence in the optimal interval combination of b intervals, k b is the number of optimal interval combinations extracted.

[0038] Preferably, the step S1 of preparing the codfish surimi sample to be tested containing different adulteration concentration ratios comprises:

[0039] The adulterated samples were made by adulterating pure cod surimi, oil fish and cod in a ratio of 10%-100%, including 11 categories in total, 20 samples in each category, and a total of 220 samples, and each category was given a different identification number.

[0040] Preferably, the step of collecting the hyperspectral image information of the cod surimi sample to be tested and performing black and white correction processing comprises:

[0041] The acquisition process of hyperspectral image information is carried out in a dark box. The parameters before acquiring the hyperspectral image are set as follows: exposure time 43ms, platform movement time 4.5cm / s, and the angle between the two 150W tungsten lamps and the platform is 50 degrees; the black and white correction processing uses the following black and white correction formula:

[0042]

[0043] Among them, the corrected reflectance hyperspectral image I is expressed in units of relative reflectance %; Rb represents the original hyperspectral image; I0 is a dark image with 0% reflectance; Rw is a white reference image with 100% reflectance.

[0044] In a second aspect, an embodiment of the present invention provides a rapid detection device for adulteration of cod surimi based on hyperspectral imaging technology, comprising:

[0045] A sample preparation module configured to prepare cod surimi samples to be tested containing different adulteration concentration ratios;

[0046] The spectrum acquisition and processing module is configured to acquire the hyperspectral image information of the cod surimi sample to be detected, perform black-white correction processing, extract the region of interest ROI of the hyperspectral image after black-white correction, and calculate the average value of the spectral reflectance of all pixels in the region of interest ROI domain as the average spectrum data of a cod surimi sample to be detected;

[0047] A first model building module is configured to divide all obtained spectral data into a training set and a prediction set, and build a BP-ANN classification model;

[0048] a second model building module, configured to preprocess the spectral data set, build a PLS regression model, and determine an optimal preprocessing method;

[0049] Model optimization module, configured to use two variable screening algorithms, namely, adaptive reweighting algorithm CARS and interval combination optimization algorithm ICO, to simplify the model, improve the model robustness, and select the best model;

[0050] The detection module is configured to determine the best model and visualize the adulteration concentration model to quickly detect whether it is adulterated.

[0051] Compared with the prior art, the beneficial results of the present invention are:

[0052] The present invention proposes a method and device for detecting adulteration of cod surimi based on hyperspectral imaging technology. The hyperspectral imaging technology is used. The hyperspectral image has the advantage of graph-spectrum integration. It combines traditional chemometrics and deep learning, and has the advantages of no damage, no pollution, automation, fast and high efficiency. The CARS and ICO methods are used to obtain the characteristic wavelength, and a partial least squares regression model is established to improve the accuracy of detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] The accompanying drawings are included to provide a further understanding of the embodiments and are incorporated into and constitute a part of this specification. The accompanying drawings illustrate the embodiments and are used together with the description to explain the principles of the present invention. It will be easy to recognize other embodiments and many expected advantages of the embodiments because they become better understood by reference to the following detailed description. The elements of the drawings are not necessarily to scale with each other. The same reference numerals refer to corresponding similar parts.

[0054] Figure 1 is a schematic diagram of a process for rapid detection of adulteration of cod surimi based on hyperspectral imaging technology according to an embodiment of the present invention;

[0055] Figure 2is a schematic diagram of a hyperspectral imaging system according to an embodiment of the present invention;

[0056] Figure 3 It is a confusion matrix diagram of the BP-ANN qualitative model of an embodiment of the present invention;

[0057] Figure 4 This is a diagram of the best PLS model result after screening variables in an embodiment of the present invention;

[0058] Figure 5 A schematic diagram of the visualized distribution of adulteration concentration of cod surimi according to an embodiment of the present invention;

[0059] Figure 6 Schematic diagram of a rapid detection device for adulteration of codfish surimi based on hyperspectral imaging technology according to an embodiment of the present invention. DETAILED DESCRIPTION

[0060] The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the relevant invention, rather than to limit the invention. It is also necessary to explain that, for ease of description, only the parts related to the relevant invention are shown in the accompanying drawings.

[0061] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0062] In a first aspect, an embodiment of the present invention discloses a method for rapid detection of adulteration of cod surimi based on hyperspectral imaging technology, such as Figure 1 As shown, the method comprises the following steps:

[0063] S1. Prepare cod surimi samples to be tested containing different adulteration concentration ratios, collect hyperspectral image information of the cod surimi samples to be tested, and perform black and white correction processing;

[0064] Specifically, the adulterated codfish surimi samples to be tested with different adulteration ratios used in this embodiment include: adulterated samples prepared by adding pure codfish surimi, oil fish and cod in a ratio of 10%-100%, with a total of 11 categories, 20 samples in each category, a total of 220 samples, and each category is provided with a different identification number.

[0065] The cod and oil fish were purchased from *Dong Supermarket or *Bao. The oil fish and cod were put into a meat grinder according to a preset ratio and blended for 3 minutes to make the meat sample into minced meat. In actual application, the meat sample can be ground into a particle size of 2 mm.

[0066] In this embodiment, when performing the hyperspectral image acquisition operation, the hyperspectral imaging system used is a hyperspectral imager, a platform control system, a computer, etc. Figure 2 As shown in the figure, the SWIR-HIS (900-1700nm) system consists of an imaging spectrometer with a spectral resolution of 4nm (N17E, Specim, Finland), a camera with a resolution of 640×512, a camera lens (640mini Raptor, Northern Ireland), two tungsten lamps (LS-150, Wuling Optics, Taiwan, China), a mobile platform driven by a stepper motor (HSIM-800, Wuling Optics, Taiwan, China) and a computer with software.

[0067] The entire acquisition process of the hyperspectral acquisition software is carried out in a dark box to prevent the light in the environment from affecting the acquired hyperspectral images. The parameters before acquiring the hyperspectral image are: exposure time 43ms, platform movement time 4.5cm / s, and the angle between the two 150W tungsten lamps and the platform is 50 degrees.

[0068] In a preferred embodiment, the black and white correction formula used in the black and white correction process is:

[0069]

[0070] Among them, the corrected reflectance hyperspectral image I is expressed in units of relative reflectance (%); Rb represents the original hyperspectral image; I0 is a dark image (0% reflectance), and Rw is a white reference image (100% reflectance).

[0071] S2, extracting the region of interest ROI of the hyperspectral image after black-and-white correction, and calculating the average spectral reflectance of all pixels in the region of interest ROI domain as the average spectral data of a cod surimi sample to be detected;

[0072] In this embodiment, ENVI 5.3 is used to select the region of interest of the hyperspectral image after black and white correction. ENVI5.3 is a powerful remote sensing image processing software with a wide range of applications in the field of remote sensing image processing and analysis, providing users with powerful tools and functions to help users extract valuable information from remote sensing data.

[0073] In this embodiment, a circular area with a radius of 60 pixels is used as the region of interest. The range of the region of interest is determined based on the circular aluminum box containing the surimi. In order to ensure that the region of interest can be a complete circular area when the surimi container is directly used for adulteration detection, the radius is set to 60 pixels in this embodiment to achieve non-destructive detection of cod surimi. A total of 220 spectral data of cod surimi to be tested are obtained.

[0074] S3, dividing all the obtained spectral data into a training set and a prediction set, and constructing a BP-ANN classification model;

[0075] The BP-ANN classification model is a model for preliminary identification of the concentration of adulterated cod surimi. In this embodiment, the spectral data obtained in step S2 is divided into a training set and a prediction set, the full wavelength is used as the input parameter of the BP-ANN network model, and the sample category is used as the model output parameter.

[0076] Figure 3 The confusion matrix of the BP-ANN qualitative model is shown. In the model, the input is 476; the output is the sample adulteration concentration gradient 11. The BP-ANN model training error is set to 10 -6 , the number of network training times is 1000. After multiple tests and adjustments to the network structure, the optimal model parameters are determined as follows: the hidden layer activation function is tansig, the output layer activation function is purelin, the training function is tansig, the number of nodes in the first hidden layer is 8, and the number of nodes in the second hidden layer is 88. Based on the above network parameters, a BP-ANN model is established. The calculation formulas for the tansig and purelin functions are as follows:

[0077]

[0078] purelin(x)=x

[0079] Among them, x is the weighted input of the neuron, the tansig function converts x to between -1 and 1; purelin(x) is the output of the neuron.

[0080] S4, preprocessing the spectral data set, building a PLS regression model, and determining the best preprocessing method;

[0081] In this embodiment, the average spectrum is subjected to preprocessing methods such as standard normal variable transformation, multivariate scattering correction, first-order derivative, second-order derivative, etc., to obtain hyperspectral data with smooth spectral information after noise removal.

[0082] Due to the light scattering characteristics of the cod surimi sample surface and the noise and interference caused by the instrument, the above four preprocessing methods were used to obtain the preprocessed spectral data to construct a PLS regression model and determine the best preprocessing method. The results are shown in Table 1 below. Figure 4 Graph of the best PLS model results after variable screening.

[0083] PLS projects the independent variable X and the dependent variable Y into the 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 fractions D and constructs the PLS model using the following equation:

[0084] Y=XA+B=XW*C+B=DC+B

[0085] Z*=Z(P'Z) -1

[0086] Where A is the PLS coefficients, B is the Y residual matrix, D is the score matrix of X, 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 and is usually achieved by using the lowest value of the prediction residual error sum of squares (PRESS).

[0087] Table 1

[0088]

[0089]

[0090] S5. Adopt two variable screening algorithms, namely, the adaptive reweighting algorithm CARS and the interval combination optimization algorithm ICO, to simplify the model, improve the model robustness, and select the best model; and

[0091] Specifically, the competitive adaptive reweighting algorithm (CARS) and interval combinatorial optimization (ICO) variable screening algorithm are used to simplify the model, improve the model robustness, and select the best model. The results are shown in Table 2.

[0092] CARS adopts an adaptive weighted sampling (ARS) strategy, saves the sample data of each sampling in the PLS model, and screens according to the absolute value weight of the regression coefficient, selects samples with high weights as new subsets, and eliminates samples with low weights. The number of Monte Carlo cross-validation is set to 1000, the proportion of calibration samples collected each time is 0.8, and the RMSECV value of each interval model is calculated by the 5-fold interactive validation method, and the number of iterations is set to 50. On this basis, the PLS model is established, and multiple iterative calculations are performed to select the bands in the smallest subset of the root mean square error (RMSECV) of the PLS model interactive validation as the feature bands. The weight calculation formula for adaptive weighted sampling is as follows:

[0093]

[0094] Among them, b iis the regression coefficient of the ith variable, m is the number of variables in the current variable set, and this formula is used to normalize the absolute value of the regression coefficient of each variable so that the sum of their weights is 1.

[0095] ICO is a chemical modeling algorithm based on the framework of model cluster analysis strategy (MPA). Its implementation process is as follows: Divide the variables into several equal parts, each of which is an interval. Then use the weighted bootstrap sampling method to generate a subset of different variable intervals randomly composed of different variable intervals, and the initial weight of each variable is 1. Establish PLS models for these intervals respectively, and calculate the RMSECV value of each interval model by the 5-fold interactive validation method. When the RMSECV value is the minimum, it is the optimal model set. Calculate the frequency of each variable interval in the optimal model, which is the weight of the next sampling, and calculate the RMSECV mean of the optimal model for this sampling. Repeat multiple iterations until the RMSECV mean is higher than the previous round. At this time, the group of variable intervals with the smallest RMSECV mean is regarded as the final selected variable interval. The following is the formula for calculating the sampling weight of each interval:

[0096]

[0097] Among them, f a is the frequency of occurrence in the optimal interval combination of b intervals, k b is the number of optimal interval combinations extracted.

[0098] Table 2

[0099]

[0100] S6. Determine the best model and visualize the adulteration concentration model to quickly detect adulteration.

[0101] Specifically, the spectral values ​​corresponding to the pixels in the prepared hyperspectral image of the cod surimi sample are extracted, and the selected spectral values ​​are substituted into the optimal model to predict the adulteration concentration information at each pixel in the hyperspectral image of the cod surimi. Finally, the distribution map of the adulteration information of the cod surimi on the plane is reconstructed according to the coordinate information of the pixel and its corresponding adulteration concentration through information fusion of hyperspectral imaging technology, such as Figure 5 As shown. In the adulteration concentration distribution diagram, the colors range from blue to red. The redder the color, the higher the adulteration concentration, and the bluer the color, the lower the adulteration concentration.

[0102] In a second aspect, an embodiment of the present invention provides a rapid detection device for adulteration of cod surimi based on hyperspectral imaging technology, such as Figure 6 As shown, it includes: a sample preparation module 61, a spectrum acquisition and processing module 62, a first model building module 63, a second model building module 64, a model optimization module 65 and a detection module 66.

[0103] In a specific embodiment, the sample preparation module 61 is configured to prepare a codfish surimi sample to be tested containing different adulteration concentration ratios; the spectrum acquisition and processing module 62 is configured to acquire the hyperspectral image information of the codfish surimi sample to be tested, and perform black and white correction processing, extract the region of interest ROI of the hyperspectral image after black and white correction, and calculate the average value of the spectral reflectance of all pixels in the region of interest ROI domain as the average spectral data of a codfish surimi sample to be tested; the first model construction module 63 is configured to divide all the obtained spectral data into a training set and a prediction set, and construct a BP-ANN classification model;

[0104] The second model building module 64 is configured to preprocess the spectral data set, build a PLS regression model, and determine the best preprocessing method; the model optimization module 65 is configured to simplify the model by using two variable screening algorithms, namely, an adaptive reweighting algorithm CARS and an interval combination optimization algorithm ICO, to improve the model robustness and select the best model; the detection module 66 is configured to determine the best model and visualize the adulteration concentration model to quickly detect whether it is adulterated.

[0105] The present invention proposes a method and device for detecting adulteration of cod surimi based on hyperspectral imaging technology. The hyperspectral imaging technology is used. The hyperspectral image has the advantage of graph-spectrum integration. It combines traditional chemometrics and deep learning, and has the advantages of no damage, no pollution, automation, fast and high efficiency. The CARS and ICO methods are used to obtain the characteristic wavelength, and a partial least squares regression model is established to improve the accuracy of detection.

[0106] The above description is only a preferred embodiment of the present invention and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the present invention is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above inventive concept. For example, the above features are replaced with the technical features with similar functions disclosed in the present invention (but not limited to) to form a technical solution.

Claims

1. A rapid detection method for adulteration of cod surimi based on hyperspectral imaging technology, characterized in that: The method comprises the following steps: S1. Prepare cod surimi samples to be tested containing different adulteration concentration ratios, collect hyperspectral image information of the cod surimi samples to be tested, and perform black and white correction processing; S2, extracting the region of interest ROI of the hyperspectral image after black-and-white correction, and calculating the average spectral reflectance of all pixels in the region of interest ROI domain as the average spectral data of a cod surimi sample to be detected; S3, dividing all the obtained spectral data into a training set and a prediction set, and constructing a BP-ANN classification model; S4, preprocessing the spectral data set, building a PLS regression model, and determining the best preprocessing method; S5. Use two variable screening algorithms, namely, the adaptive reweighting algorithm CARS and the interval combination optimization algorithm ICO, to simplify the model, improve the model robustness, and select the best model; and S6. Determine the best model and visualize the adulteration concentration model to quickly detect adulteration.

2. The method for rapid detection of adulteration of cod surimi based on hyperspectral imaging technology according to claim 1, characterized in that: Step S3 divides all the obtained spectral data into a training set and a prediction set, including: The full wavelength is used as the input parameter of the BP-ANN network model, and the sample category is used as the model output parameter. The input quantity in the model is 476, the output quantity is the sample adulteration concentration gradient 11, and the BP-ANN model training error is set to 10 -6 , the network training times is 1000.

3. The method for rapid detection of adulteration of cod surimi based on hyperspectral imaging technology according to claim 2, characterized in that: The construction of the BP-ANN classification model includes: The network parameters of the model are: the hidden layer activation function is tansig, the output layer activation function is purelin, the training function is tansig, the number of nodes in the first hidden layer is 8, and the number of nodes in the second hidden layer is 88; the calculation formulas of tansig and purelin functions are as follows: purelin(x)=x Among them, x is the weighted input of the neuron, the tansig function converts x to between -1 and 1; purelin(x) is the output of the neuron.

4. The method for rapid detection of cod surimi adulteration based on hyperspectral imaging technology according to claim 1, characterized in that: Step S4 preprocesses the spectral data set, including: The obtained average spectrum is preprocessed using standard normal variable transformation, multivariate scattering correction, first-order derivative, and second-order derivative methods to obtain hyperspectral data with smooth spectral information after noise removal.

5. The method for rapid detection of adulteration of cod surimi based on hyperspectral imaging technology according to claim 1, characterized in that: The construction of the PLS regression model and determination of the optimal preprocessing method include: PLS projects the independent variable X and the dependent variable Y into the latent variable space to establish a linear regression model. PLS decomposes X and Y into several X scores D and constructs the PLS model using the following equation: Y=XA+B=XW*C+B=DC+B Z*=Z(P'Z) -1 Where A is the PLS coefficient, B is the Y residual matrix, D is the score matrix of X, Z is the PLS weight, P and C are the loadings of X and Y respectively, and Z* is the regression coefficient matrix.

6. The method for rapid detection of adulteration of cod surimi based on hyperspectral imaging technology according to claim 1, characterized in that: Step S5 uses two variable screening algorithms, namely, the adaptive reweighting algorithm CARS and the interval combination optimization algorithm ICO, to simplify the model, specifically including: CARS adopts the adaptive weighted sampling ARS strategy, which stores the sample data of each sampling in the PLS model, and screens them according to the absolute value weight of the regression coefficient, selects samples with high weights as new subsets, and eliminates samples with low weights; The number of Monte Carlo cross-validation was set to 1000, the proportion of calibration samples collected each time was 0.8, the root mean square error value RMSECV of each interval model was calculated by the 5-fold interactive validation method, and the number of iterations was set to 50; the PLS model was established, and multiple iterative calculations were performed, and the bands in the smallest subset of the root mean square error RMSECV of the PLS model interactive validation were selected as feature bands; The weight calculation formula for adaptive weighted sampling is as follows: Among them, b i is the regression coefficient of the ith variable, m is the number of variables in the current variable set, and this formula is used to normalize the absolute value of the regression coefficient of each variable so that the sum of their weights is 1.

7. The method for rapid detection of adulteration of cod surimi based on hyperspectral imaging technology according to claim 6, characterized in that: The interval combinatorial optimization algorithm ICO is a chemical modeling algorithm based on the model cluster analysis strategy MPA framework, which includes: Divide the variables into several parts, each of which is an interval; then use the weighted bootstrap sampling method to generate subsets of different variable intervals, with the initial weight of each variable being 1; establish PLS models for these intervals respectively, and calculate the RMSECV value of each interval model through the 5-fold interactive validation method. The optimal model set is when the RMSECV value is the minimum. Calculate the frequency of each variable interval in the optimal model, which is the weight of the next sampling, and calculate the RMSECV mean of the optimal model for this sampling; repeat multiple iterations until the RMSECV mean is higher than the previous round; at this time, the group of variable intervals with the smallest RMSECV mean is regarded as the final selected variable interval; the formula for calculating the sampling weight of each interval is as follows: Among them, f a is the frequency of occurrence in the optimal interval combination of b intervals, k b is the number of optimal interval combinations extracted.

8. The method for rapid detection of adulteration of cod surimi based on hyperspectral imaging technology according to claim 1, characterized in that: The step S1 of preparing the codfish surimi samples to be tested containing different adulteration concentration ratios includes: The adulterated samples were made by adulterating pure cod surimi, oil fish and cod in a ratio of 10%-100%, including 11 categories in total, 20 samples in each category, and a total of 220 samples, and each category was given a different identification number.

9. The method for rapid detection of adulteration of cod surimi based on hyperspectral imaging technology according to claim 1, characterized in that: The method of collecting the hyperspectral image information of the cod surimi sample to be detected and performing black and white correction processing includes: The acquisition process of hyperspectral image information is carried out in a dark box. The parameters before acquiring the hyperspectral image are set as follows: exposure time 43ms, platform movement time 4.5cm / s, and the angle between the two 150W tungsten lamps and the platform is 50 degrees; the black and white correction processing uses the following black and white correction formula: Among them, the corrected reflectance hyperspectral image I is expressed in units of relative reflectance %; Rb represents the original hyperspectral image; I0 is a dark image with 0% reflectance; Rw is a white reference image with 100% reflectance.

10. A rapid detection device for adulteration of cod surimi based on hyperspectral imaging technology, characterized in that: include: A sample preparation module configured to prepare cod surimi samples to be tested containing different adulteration concentration ratios; The spectrum acquisition and processing module is configured to acquire the hyperspectral image information of the cod surimi sample to be detected, perform black-white correction processing, extract the region of interest ROI of the hyperspectral image after black-white correction, and calculate the average value of the spectral reflectance of all pixels in the region of interest ROI domain as the average spectrum data of a cod surimi sample to be detected; A first model building module is configured to divide all obtained spectral data into a training set and a prediction set, and build a BP-ANN classification model; a second model building module, configured to preprocess the spectral data set, build a PLS regression model, and determine an optimal preprocessing method; Model optimization module, configured to use two variable screening algorithms, namely, adaptive reweighting algorithm CARS and interval combination optimization algorithm ICO, to simplify the model, improve the model robustness, and select the best model; The detection module is configured to determine the best model and visualize the adulteration concentration model to quickly detect whether it is adulterated.

Images