Risk prediction method of heavy metals in crayfish under different processing modes

Abstract

The application discloses a method for predicting heavy metal risk of crayfish under different processing modes, measures the heavy metal content of crayfish tails of crayfish under various processing modes, calculates the heavy metal risk HI calculation value of children and adults, measures the water content, weight, crude fat content, soluble protein content, whiteness value, texture characteristics and odor characteristics of crayfish tails under various processing modes, establishes a prediction model, constructs a training sample set, a verification sample set and a prediction sample set, and trains and optimizes the prediction model through the training sample set and the verification sample set. The prediction model is simple to establish, the input layer selects simple physicochemical indexes, the output layer is a relatively complex heavy metal risk, and simple work is used to replace complicated work.

Description

Technical Field

[0001] This invention belongs to the field of hazardous substance risk prediction technology, specifically relating to a method for predicting the risk of ingesting heavy metals from crayfish processed under different conditions. It is applicable to predicting the risk level of heavy metal hazards to human health from processed crayfish products. Background Technology

[0002] Because crayfish have a high resistance and adaptability to environmental metal pollution, some metals may accumulate in their bodies, posing a potential threat to human health. Even at very low doses, toxic metals such as mercury (Hg), cadmium (Cd), chromium (Cr), and lead (Pb) can induce a range of cancers, neurotoxicity, nephrotoxicity, and organ failure. Furthermore, excessive amounts of essential trace elements such as copper (Cu), zinc (Zn), nickel (Ni), and manganese (Mn) can also harm the human body. Therefore, as one of my country's most popular aquatic products, the heavy metal pollution problem of crayfish has recently received increasing attention. Most research on the risks to human health posed by heavy metals in crayfish abdominal muscles has focused on raw crayfish. However, food processing involves complex changes that can lead to physicochemical alterations, such as changes in the content and form of heavy metals. Risk assessments based on raw samples may cause consumers to overestimate or underestimate the risks associated with processing.

[0003] Furthermore, heavy metal changes during food processing are an uncontrollable phenomenon. Therefore, some linear models, such as multiple linear regression and partial least squares regression, are not suitable for modeling and intelligently monitoring complex food processing processes. BP-ANN, a neural network system based on biomimetic principles, possesses rapid parallel information processing capabilities and excellent nonlinear mapping abilities, making it an ideal method for building nonlinear models. Currently, BP neural networks have been applied in food preservation, food processing, enzyme engineering, and food hazard monitoring. However, there is currently no method for predicting heavy metal hazards to humans caused by crayfish tail processing products. Therefore, establishing a rapid, simple, and accurate monitoring method is particularly urgent. Summary of the Invention

[0004] The purpose of this invention is to address the current limitations and technical complexity of monitoring heavy metal hazards in processed crayfish products by providing a method for predicting the risk of heavy metal ingestion from crayfish under different processing methods. The method is easy to implement, simple, and quick to operate, meeting the needs of people monitoring and predicting the risk level of heavy metal hazards from crayfish ingestion.

[0005] The above-mentioned objectives of the present invention are achieved through the following technical means:

[0006] A method for predicting the risk of heavy metal intake from crayfish under different processing methods includes the following steps:

[0007] Step 1: Detection of heavy metal content in samples obtained from crayfish processed using various methods. This includes the following steps:

[0008] Step 1.1: Process crayfish using multiple methods (boiling, frying, steaming, and seasoning).

[0009] Step 1.2: After processing, use plastic tweezers to remove the edible shrimp tail from the crayfish shell as samples for each processing method. After drying and homogenizing the shrimp tail at 65℃, weigh about 0.5g of the sample into a polytetrafluoroethylene digestion vessel, add hydrogen peroxide and nitric acid to digest the sample.

[0010] Step 1.3: The contents of nine heavy metals, As, Ba, Cd, Cr, Cu, Hg, Mn, Ni and Pb, in shrimp tails obtained by digestion of various processing methods were detected by inductively coupled plasma optical emission spectrometry (ICP-OES).

[0011] Step 2: Calculation of HI values ​​for heavy metal risk in children and adults.

[0012] The target hazard factor (THQ) is evaluated based on the ratio of the amount absorbed by the human body to the reference amount. The formula for calculating the residual hazard factor THQ(i) of the i-th heavy metal is as follows:

[0013]

[0014] In the formula, C i Let be the concentration (mg / kg, wet weight) of the i-th heavy metal in crayfish tails, where heavy metals ∈ {As, Ba, Cd, Cr, Cu, Hg, Mn, Ni, Pb}; 'a' is the conversion factor from dry weight to wet weight of the crayfish tails; 'FIR' is the estimated daily dietary intake rate of crayfish tails, divided into the estimated daily dietary intake rate of crayfish tails for children and the estimated daily dietary intake rate of crayfish tails for adults (children 0.072 kg / person / day; adults 0.168 kg / person / day); 'EF' is the population exposure frequency (365 days / year). (years); ED is the exposure time (mean lifespan 70 years); RfD(i) is the oral reference dose of the i-th heavy metal (μg / kg / day; As, 0.3; Ba, 200; Cd, 1; Cr, 1500; Cu, 40; Hg, 0.5; Mn, 140; Ni, 20; Pb, 4); AT is the mean exposure time (365 days / year × exposure time); BW is the mean weight, divided into the mean weight of children and the mean weight of adults (70 kg for adults; 16 kg for children).

[0015] When THQ(i) < 1, it can be considered that there is no health risk; when THQ(i) ≥ 1, it can be considered that the possible health risk cannot be ruled out.

[0016] FIR selects the estimated daily dietary intake rate of children from crayfish tails, BW selects the average weight of children, and the heavy metal risk HI value for children is the sum of the residual hazard coefficients of each heavy metal.

[0017] FIR selects the estimated daily dietary intake rate of crayfish tails for adults, BW selects the average weight of adults, and the calculated value of heavy metal risk HI for adults is the sum of the residual hazard coefficients of each heavy metal.

[0018] The formula for calculating the sum of the residual hazard coefficients of each heavy metal is as follows:

[0019] HI=THQ(As)+THQ(Ba)+THQ(Cd)+THQ(Cr)+THQ(Cu)+THQ(Hg)

[0020] +THQ(Mn)+THQ(Ni)+THQ(Pb) (2)

[0021] Similarly, an HI value ≥ 1 indicates that heavy metals pose a potential danger to the human body; otherwise, the risk is negligible.

[0022] Step 3: Determination of physicochemical indices of the model input layer

[0023] The input parameters of the prediction model's input layer include seven physicochemical indicators: shrimp tail moisture content, shrimp tail weight, shrimp tail crude fat content, shrimp tail soluble protein content, shrimp tail whiteness value, textural properties, and odor characteristics. Textural properties were measured using a physical property testing instrument, including hardness, elasticity, adhesiveness, and resilience. Odor characteristics were measured using an electronic nose FOX4000, which includes 18 different types of sensors. The detection methods for the physicochemical indicators are as follows:

[0024] Step 3.1: The moisture content of shrimp tails processed by different methods was determined according to the 105°C constant weight method in GB 5009.3-2016. The moisture content is shown in Table 1.

[0025] Table 1 shows the moisture content and processing loss of crayfish tails under different processing conditions.

[0026]

[0027] Shrimp tail weight and moisture content are expressed as 8 mean ± standard deviation. Different lowercase letters in the same column indicate significant differences between the means (p < 0.05).

[0028] Step 3.2: The crude fat content of shrimp tails processed by different methods was determined according to GB 5009.6-2016. 0.5g of dried shrimp tails was accurately weighed and determined by acid extraction. The result was expressed as g / 100g and the average value of 8 repeated measurements was taken.

[0029] Step 3.3: The soluble protein content of shrimp tails processed by different methods was determined using the BCA method. 0.5g of dried shrimp tail was weighed, and an appropriate amount of n-hexane was added. The mixture was vortexed to remove oil and fat, and the n-hexane was discarded. 10ml of phosphate buffer solution (pH 7.2, 0.6mol / L NaCl) was added and mixed thoroughly. The mixture was vortexed for 30s, allowed to stand at 4℃ for 3 hours, and then centrifuged at 10000r / min for 10 minutes to obtain the supernatant. The protein concentration was measured using the BCA method. Eight parallel measurements were performed for each type of shrimp tail processed by different methods.

[0030] Step 3.4: The measurement of whiteness values ​​of shrimp tails processed using different methods was conducted using the following steps: The crayfish were shelled, and the 2nd-3rd abdominal segments of the tail were taken. Color was measured using an automatic colorimeter, and the brightness value (L*), red-green value (a*), and yellow-blue value (b*) were recorded. Each sample was measured 8 times in parallel. The formula for calculating whiteness (W) is as follows:

[0031]

[0032] The measurement results of the whiteness value of shrimp tails are shown in Table 2:

[0033] Table 2 shows the effects of different processing methods on the color of crayfish abdominal muscles.

[0034]

[0035] The shrimp tail color parameters are expressed as 8 mean ± standard deviation. Different lowercase letters in the same column indicate significant differences between the means (p < 0.05).

[0036] Step 3.5: The textural properties of shrimp tails processed using different methods were measured using the following steps: After processing, the crayfish were shelled, and the middle part of the tail was taken and cut into uniform cubes of 1cm × 1cm × 8mm. The texture was measured using a TA-XTplus material property analyzer. Parameters were set as follows: trigger type Auto, pre-test rate 2mm / s, mid-test rate 1mm / s, post-test rate 1mm / s, compression strength 50%, and residence time between two compressions 5s. The compression probe was a stainless steel P / 36R cylindrical probe. Eight samples of shrimp tails from each processing method were collected, and the average value was taken. The textural property test results are as follows: Figure 5 As shown, textural properties include hardness, elasticity, adhesiveness, and resilience.

[0037] Step 3.6: The odor characteristics of shrimp tails were measured using the following steps: Odor characteristics were collected using an electronic nose. 2g of processed shrimp tails were placed in a 10ml headspace vial, sealed, and allowed to stand for 30min. The electronic nose measurement parameters were as follows: carrier gas flow rate 150ml / min; injection volume 2.5ml / s; injection speed 2.5ml / s; acquisition delay time 300s; acquisition time interval 1s; test acquisition time 120s; incubation temperature 50℃; incubation time 120s. Eight measurements were performed on shrimp tails processed using each method. Principal component analysis was performed on the measurement data to obtain the odor characteristics of the shrimp tails.

[0038] The output layer of the BP neural network prediction model outputs the HI predicted values ​​of heavy metal risk for children and adults.

[0039] Step 4: Establishment of the prediction model based on BP neural network (BP-ANN model)

[0040] Step 4.1: The prediction model adopts the BP-ANN model. The BP-ANN model is established using the net = newff(P,T,S,TF,BTF) function in MATLAB. Here, P is the input data matrix (the elements of the input data matrix include the shrimp tail moisture content, shrimp tail weight, shrimp tail crude fat content, shrimp tail soluble protein content, shrimp tail whiteness value, texture characteristics and odor characteristics corresponding to the processing methods in Step 3, and also include the calculated heavy metal risk HI values ​​for children and adults corresponding to the processing methods calculated in Step 2), T is the output layer data matrix (the elements of the output layer data matrix include the calculated predicted heavy metal risk HI values ​​for children and adults corresponding to the processing methods), S is the number of hidden layer neurons, TF is the node transfer function, and BTF is the training function.

[0041] Step 4.2: The input of the prediction model's input layer consists of the shrimp tail moisture content, shrimp tail weight, shrimp tail crude fat content, shrimp tail soluble protein content, shrimp tail whiteness value, texture characteristics, and odor characteristics corresponding to the processing method measured in Step 3. It also includes the calculated HI values ​​for heavy metal risk in children and adults corresponding to the processing method calculated in Step 2. The output of the prediction model's output layer consists of the predicted HI values ​​for heavy metal risk in children and adults.

[0042] Step 4.3: Construct multiple sets of data samples. Each set includes shrimp tail moisture content, shrimp tail weight, shrimp tail crude fat content, shrimp tail soluble protein content, shrimp tail whiteness value, textural characteristics, and odor characteristics corresponding to a specific processing method. It also includes the corresponding calculated heavy metal risk HI values ​​for children and adults. Randomly divide the multiple sets of data samples (up to 40 sets) into a training set, a validation set, and a prediction set: 70% for the training set, 15% for the validation set, and 15% for the prediction set. The training set is used to train the parameters of the prediction model. The validation set is used to continuously adjust the parameters of the prediction model during training to ensure the prediction error is less than the set prediction error. The prediction set is used to test the prediction error of the prediction model after training is complete.

[0043] Step 4.4: Determine the training functions, node transfer functions between the input and hidden layers, and node transfer functions between the hidden and output layers of the prediction model. The training functions for the prediction model mainly include the gradient descent function `traingd`, the gradient descent function `traingdm` for momentum backpropagation, the gradient descent function `traingda` for dynamic adaptive learning rate, the gradient descent function `traingdx` for momentum backpropagation and dynamic adaptive learning rate, and the training function `trainlm` for Levenberg-Marquardt (nonlinear least squares method). The main function of the node transfer functions in the prediction model is to connect the input, hidden, and output layers, so determining the transfer functions between each layer is crucial. Commonly used node transfer functions include the linear transfer function `purelin`, the tangent sigmoid transfer function `tansig`, and the logarithmic sigmoid transfer function `logsig`.

[0044] Step 4.5: The choice of the number of hidden layer neurons is very important, as it directly affects the performance of the neural network. The empirical formula for the number of hidden layer neurons is as follows:

[0045]

[0046] In the formula, N is the number of hidden layer neurons; l is the number of input layer neurons; m is the number of output layer neurons; and A is a constant between 0 and 10.

[0047] Step 4.6: Train the prediction model using the training sample set. After training, obtain the optimal parameters of the prediction model. Adjust the node transfer function and the number of hidden layer neurons of the trained prediction model using the validation sample set, so that the prediction error of the heavy metal risk HI prediction value for children and adults is less than the set prediction error.

[0048] Compared with the prior art, the present invention has the following advantages:

[0049] 1. By using formulas to calculate, the abstract heavy metal concentration in crayfish tails is transformed into a directly expressible measure of the degree of harm of heavy metals to the human body.

[0050] 2. The prediction model is simple to build; the input layer uses simple physicochemical indicators, while the output layer deals with more complex heavy metal risks. It replaces tedious work with simple tasks.

[0051] 3. This method can also be applied to the assessment and prediction of edible hazards of other aquatic organisms, such as freshwater fish products. Attached Figure Description

[0052] Figure 1 Experimental procedure for detecting metal content in the abdominal muscles of crayfish.

[0053] Figure 2 This is the BP-ANN network topology, namely, input layer - hidden layer - output layer.

[0054] Figure 3 The effect of different processing methods on the fat content in the abdominal muscle of crayfish is one of the input layer parameters (crude fat content in the crayfish tail).

[0055] Figure 4 The effect of different processing methods on the soluble protein content in the abdominal muscle of crayfish is one of the input layer parameters (soluble protein content in the crayfish tail).

[0056] Figure 5 The effect of different processing methods on the textural properties of crayfish abdominal muscle is one of the input layer parameters (textural properties).

[0057] Figure 6 The effect of different processing methods on the flavor characteristics of crayfish abdominal muscle is one of the input layer parameters (odor characteristics).

[0058] Figure 7 The calculated heavy metal risk HI values ​​for children and adults in crayfish abdominal muscles under different processing methods are shown. (A) represents the calculated heavy metal risk HI value for adults, and (B) represents the calculated heavy metal risk HI value for children.

[0059] Figure 8 This is the interface for running the model program in MATLAB when the function `net = newff(P,T,S,TF,BTF)` is executed. It mainly includes basic information such as the number of iterations, error performance, and runtime.

[0060] Figure 9 The fitting results of the prediction model for the heavy metal risk (HI) prediction values ​​of adults are shown.

[0061] Figure 10 The fitting results of the prediction model for the heavy metal risk (HI) prediction values ​​of children. Detailed Implementation

[0062] To facilitate understanding and implementation of the present invention by those skilled in the art, the present invention will be further described in detail below with reference to examples. The implementation examples described herein are only for illustration and explanation and are not intended to limit the present invention.

[0063] A method for predicting heavy metal hazards caused by human ingestion of processed crayfish tail products, comprising the following steps:

[0064] Step 1: Detection of heavy metal content in samples obtained from crayfish processed using various methods.

[0065] In this embodiment, each processing method includes thawing enough crayfish with micro-flow water, randomly selecting 120 crayfish and processing them using four different methods, and setting up a raw sample blank control. There are 8 parallel samples under each processing method to collect sufficient data. The main processing methods for crayfish are as follows: (1) Boiling: Boiling in a stainless steel container for 15 minutes; (2) Frying: Heating in rapeseed oil at 180℃ for 5 minutes; (3) Steaming: Steaming in a stainless steel container for 15 minutes; (4) Spicy crayfish: After frying the crayfish for 5 minutes, add 330ml of beer, 180g of broad bean paste, 20g of dried chili peppers, 5g of Sichuan peppercorns, and some scallions, ginger, and garlic, and boil for 20 minutes.

[0066] After processing, the edible shrimp tails were removed from the crayfish shells using plastic tweezers as samples for each processing method. After the shrimp tails were dried and homogenized at 65°C, about 0.5g of the sample was weighed into a polytetrafluoroethylene digestion vessel, and hydrogen peroxide and nitric acid were added to digest the sample.

[0067] Nine heavy metals in crayfish tails were detected by ICP-OES (inductively coupled plasma atomic emission spectrometry) under four different processing conditions: boiling, frying, steaming, and seasoning (spicy). Table 1 shows the average heavy metal content (mg / kg, dry weight) in crayfish tails under different processing methods.

[0068] Table 3 shows the average heavy metal content (mg / kg, dry weight) in crayfish tails under different processing methods.

[0069]

[0070] Step 2: Calculation of heavy metal hazard assessment indicators

[0071] The target hazard factor (THQ) is evaluated based on the ratio of the amount absorbed by the human body to the reference amount. The formula for calculating the residual hazard factor THQ(i) of the i-th heavy metal is as follows:

[0072]

[0073] In the formula, C i Let be the concentration (mg / kg, wet weight) of the i-th heavy metal in crayfish tails, where heavy metals ∈ {As, Ba, Cd, Cr, Cu, Hg, Mn, Ni, Pb}; 'a' is the conversion factor from dry weight to wet weight of crayfish tails; 'FIR' is the estimated daily dietary intake rate of crayfish tails (0.072 kg / person / day for children; 0.168 kg / person / day for adults); 'EF' is the population exposure frequency (365 days / year); 'ED' is the exposure time (mean lifespan 70 years); 'RfD(i)' is the oral reference dose of the i-th heavy metal (μg / kg / day; As, 0.3; Ba, 200; Cd, 1; Cr, 1500; Cu, 40; Hg, 0.5; Mn, 140; Ni, 20; Pb, 4); 'AT' is the mean exposure time (365 days / year × exposure time); and 'BW' is the mean body weight (70 kg for adults; 16 kg for children).

[0074] When THQ(i) < 1, it can be considered that there is no health risk; when THQ(i) ≥ 1, it can be considered that the possible health risk cannot be ruled out.

[0075] FIR selects the estimated daily dietary intake rate of children from crayfish tails, BW selects the average weight of children, and the heavy metal risk HI value for children is the sum of the residual hazard coefficients of each heavy metal.

[0076] FIR selects the estimated daily dietary intake rate of crayfish tails for adults, BW selects the average weight of adults, and the calculated value of heavy metal risk HI for adults is the sum of the residual hazard coefficients of each heavy metal.

[0077] The formula for calculating the sum of the residual hazard coefficients of each heavy metal is as follows:

[0078] HI=THQ(As)+THQ(Ba)+THQ(Cd)+THQ(Cr)+THQ(Cu)+THQ(Hg)

[0079] +THQ(Mn)+THQ(Ni)+THQ(Pb) (2)

[0080] Similarly, an HI value ≥ 1 indicates that heavy metals pose a potential danger to the human body; otherwise, the risk is negligible.

[0081] Step 3: Measurement of input parameters of the input layer of the prediction model of the BP neural network.

[0082] The input parameters of the prediction model of the BP neural network include seven physicochemical indicators: shrimp tail moisture content, shrimp tail weight, shrimp tail crude fat content, shrimp tail soluble protein content, shrimp tail whiteness value, texture properties, and odor characteristics.

[0083] Step 4: Building the Prediction Model

[0084] Step 4.1: The prediction model adopts the BP-ANN model. The BP-ANN model is established using the net = newff(P,T,S,TF,BTF) function in MATLAB. Here, P is the input data matrix (the elements of the input data matrix include the shrimp tail moisture content, shrimp tail weight, shrimp tail crude fat content, shrimp tail soluble protein content, shrimp tail whiteness value, texture characteristics and odor characteristics corresponding to the processing methods in Step 3, and also include the calculated heavy metal risk HI values ​​for children and adults corresponding to the processing methods calculated in Step 2), T is the output layer data matrix (the elements of the output layer data matrix include the calculated predicted heavy metal risk HI values ​​for children and adults corresponding to the processing methods), S is the number of hidden layer neurons, TF is the node transfer function, and BTF is the training function.

[0085] Step 4.2: The input of the prediction model's input layer consists of the shrimp tail moisture content, shrimp tail weight, shrimp tail crude fat content, shrimp tail soluble protein content, shrimp tail whiteness value, texture characteristics, and odor characteristics corresponding to the processing method measured in Step 3. It also includes the calculated HI values ​​for heavy metal risk in children and adults corresponding to the processing method calculated in Step 2. The output of the prediction model's output layer consists of the predicted HI values ​​for heavy metal risk in children and adults.

[0086] Step 4.3: Construct multiple sets of data samples. Each set of data samples includes the shrimp tail moisture content, shrimp tail weight, shrimp tail crude fat content, shrimp tail soluble protein content, shrimp tail whiteness value, texture characteristics, and odor characteristics corresponding to a certain processing method. Each set of data samples also includes the corresponding heavy metal risk HI calculation values ​​for children and adults. Based on the multiple sets of data samples, generate a training sample set, a validation sample set, and a prediction sample set. In this embodiment, there are a total of 40 sets of data samples. 28 sets of data samples are randomly selected as the training sample set, 6 sets of data samples are used as the validation sample set, and 6 sets of data samples are used as the prediction sample set.

[0087] Step 4.4: Determine the training function and node transfer function of the prediction model.

[0088] As a type of backpropagation (BP) algorithm, Levenberg-Marquardt (nonlinear least squares) is considered the preferred method due to its fast learning ability, especially when using medium-sized neural networks. Therefore, the training function for the prediction model is chosen as `trainlm`. Given a fixed training function `trainlm`, the number of hidden layer neurons is set to 5 to obtain a better transfer function.

[0089] The prediction model input validation sample set included shrimp tail moisture content, shrimp tail weight, shrimp tail crude fat content, shrimp tail soluble protein content, shrimp tail whiteness value, texture characteristics, odor characteristics, and calculated heavy metal risk HI values ​​for children and adults. Under the same network structure, weights, and thresholds, different node transfer functions were selected between the input and output layers of the prediction model. The relationship between the mean squared error (MSE) and mean absolute error (MAE) between the predicted heavy metal risk HI values ​​for adults and their corresponding calculated values ​​is shown in Table 4. The minimum error was achieved when the node transfer function between the input and hidden layers was logsig, and the transfer function between the hidden and output layers was tansig, with a prediction set MSE of 0.0113 and an MAE of 0.0821. Therefore, the transfer function between the input and hidden layers is logsig, and the transfer function between the hidden and output layers is tansig.

[0090] Table 4 shows the error table corresponding to different node transfer functions.

[0091]

[0092]

[0093] Step 4.5: The choice of the number of hidden layer neurons is very important, as it directly affects the performance of the neural network. The empirical formula for the number of hidden layer neurons is as follows:

[0094]

[0095] In the formula, N is the number of hidden layer neurons; l is the number of input layer neurons; m is the number of output layer neurons; and A is a constant between 0 and 10. Therefore, the number of hidden layers in this embodiment is between 3 and 13.

[0096] To determine the specific number of hidden layers, with the training function `trainlm`, the transfer function between the input layer and the hidden layer `logsig`, the transfer function between the hidden layer and the output layer `tansig`, a learning rate `lr` = 0.3, and 1000 iterations, under the same network structure, weights, and thresholds, the model input validation sample set included shrimp tail moisture content, shrimp tail weight, shrimp tail crude fat content, shrimp tail soluble protein content, shrimp tail whiteness value, texture characteristics, odor characteristics, and calculated heavy metal risk HI values ​​for children and adults. Different numbers of hidden layer neurons were selected. The relationship between the mean squared error (MSE) and mean absolute error (MAE) between the predicted heavy metal risk HI values ​​for adults and their corresponding calculated values ​​is shown in Table 5. When the number of hidden layer neurons was 6, the minimum MSE was 0.051, and the minimum MAE was 0.166. Therefore, the number of hidden layer neurons in the model was determined to be 6.

[0097] Table 5 shows the prediction errors of the prediction models with different numbers of hidden layer nodes.

[0098]

[0099] Step 4.6: Train the prediction model using the training sample set. After training, obtain the optimal parameters of the prediction model. Adjust the node transfer function and the number of hidden layer neurons of the trained prediction model using the validation sample set, so that the prediction error of the heavy metal risk HI prediction value for children and adults is less than the set prediction error.

[0100] The program execution process of the prediction model BP-ANN is shown in the figure:

[0101] Clicking the "Regression" option in the program's interface will display the linear regression results of the BP-ANN prediction model on the training, validation, and prediction sample sets. Figure 5 As shown:

[0102] A BP-ANN model was established using shrimp tail moisture content, shrimp tail weight, shrimp tail crude fat content, shrimp tail soluble protein content, shrimp tail whiteness value, texture characteristics (PC1), and odor characteristics (PC1, PC2) as X variables, and the calculated heavy metal risk HI values ​​for children and adults as Y variables. The training function was selected as trainlm, the transfer function between the input layer and the hidden layer was logsig, and the transfer function between the hidden layer and the output layer was tansig.

[0103] Figure 9The fitting results of the prediction model for the heavy metal risk (HI) prediction values ​​of adults (including the regression coefficients R for the training sample set, validation sample set, and prediction sample set) are given. The regression coefficient R for the training sample set is greater than the regression coefficient R for the prediction sample set, which is greater than the regression coefficient R for the validation sample set. The regression coefficient R for the training sample set is 0.968. Figure 9 (Top left figure); The regression coefficient R for the validation sample set is 0.917 ( Figure 9 (Top right figure); The regression coefficient R for the predicted sample set is 0.956 ( Figure 9 (Lower left figure) shows the fitting correlation coefficient R of the prediction model BP-ANN. 2 It is 0.915.

[0104] Figure 10 The fitting results of the prediction model for the heavy metal risk HI value of children (including the regression coefficients R of the training sample set, validation sample set, and prediction sample set) are given. The regression coefficient R of the validation sample set is greater than that of the training sample set, which is greater than that of the prediction sample set. The regression coefficient R of the training sample set is 0.972. Figure 10 (Top left figure); The regression coefficient R for the validation sample set is 0.973 ( Figure 10 (Top right figure); the regression coefficient R for the predicted sample set is 0.928 ( Figure 10 (Lower left figure) shows the fitting correlation coefficient R of the prediction model BP-ANN. 2 It is 0.929.

[0105] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention. For those skilled in the art, other variations or modifications can be made based on the above description and ideas, and it is neither necessary nor possible to exhaustively describe all implementation methods here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A method for predicting the risk of heavy metal intake from crayfish under different processing methods, characterized in that... Includes the following steps: Step 1: Process crayfish using various processing methods and measure the content of various heavy metals in the tails of crayfish processed using different methods; Step 2: Calculate the HI value for heavy metal risk in children and adults; Step 3: Measure the moisture content, weight, crude fat content, soluble protein content, whiteness value, texture characteristics, and odor characteristics of shrimp tails processed by various methods. Step 4: Establish a prediction model based on the BP-ANN model. The prediction model includes an input layer, hidden layers, and an output layer. Determine the training function, the node transfer function between the input and hidden layers, and the node transfer function between the hidden and output layers. Determine the number of neurons in the hidden layers. Multiple sets of data samples were constructed. Each set of data samples included the following parameters for shrimp tails: moisture content, weight, crude fat content, soluble protein content, whiteness value, textural properties, and odor characteristics, corresponding to a specific processing method. Each set of data samples also included the corresponding calculated HI values ​​for heavy metal risk in children and adults. Multiple sets of data samples are divided into training sample set, validation sample set, and prediction sample set. The prediction model is trained using a training sample set. After training, the optimal parameters of the prediction model are obtained. The node transfer function and the number of hidden layer neurons of the trained prediction model are adjusted using a validation sample set so that the prediction error of the heavy metal risk HI values ​​for children and adults is less than the set value.

2. The method for predicting the risk of heavy metal intake from crayfish under different processing modes according to claim 1, characterized in that, The heavy metals mentioned in step 1 include As, Ba, Cd, Cr, Cu, Hg, Mn, Ni, and Pb. Measuring the content of each heavy metal in the tails of crayfish processed in various ways includes the following steps: Remove the shrimp tail from the crayfish shell using plastic tweezers, and after drying and homogenizing at 65℃, weigh 0.5g of shrimp tail into a polytetrafluoroethylene digestion vessel, add hydrogen peroxide and nitric acid to digest the shrimp tail. The contents of nine heavy metals, As, Ba, Cd, Cr, Cu, Hg, Mn, Ni, and Pb, in shrimp tails obtained by various processing methods were detected by inductively coupled plasma atomic emission spectrometry (ICP-OES).

3. The method for predicting the risk of heavy metal intake from crayfish under different processing modes according to claim 1, characterized in that, The heavy metal risk HI calculation values ​​for children and adults in step 2 are based on the following formula for the residual hazard coefficient of the i-th heavy metal: In the formula, THQ(i) is the residual hazard coefficient of the i-th heavy metal, and C i denoted as , where is the concentration of the i-th heavy metal in crayfish tails; 'a' is the conversion factor from dry weight to wet weight of crayfish tails; 'FIR' is the estimated daily dietary intake rate of crayfish tails, divided into the estimated daily dietary intake rate of crayfish tails for children and the estimated daily dietary intake rate of crayfish tails for adults; 'EF' is the population exposure frequency. ED stands for exposure time; RfD(i) is the oral reference dose for the i-th heavy metal; AT stands for average exposure time. BW stands for average body weight, which is divided into average weight for children and average weight for adults. FIR uses the estimated daily dietary intake rate of crayfish tails for children, BW uses the average weight of children, and the heavy metal risk HI value for children is calculated as the sum of the residual hazard coefficients of each heavy metal. FIR selects the estimated daily dietary intake rate of crayfish tails for adults, BW selects the average weight of adults, and the calculated value of heavy metal risk HI for adults is the sum of the residual hazard coefficients of each heavy metal.

4. The method for predicting the risk of heavy metal intake from crayfish under different processing modes according to claim 1, characterized in that, The number of hidden layer neurons in step 4 is based on the following formula: In the formula, N is the number of neurons in the hidden layer; l is the number of neurons in the input layer; m is the number of neurons in the output layer; A is a constant between 0 and 10.

Images