A method for evaluating the freshness of scallops based on a multi-layer perceptron
By constructing a fully connected neural network model using an electronic nose system based on a multilayer perceptron, the problem of poor universality in predicting TVB-N for scallops under various preservation treatments in existing technologies is solved. This enables rapid, non-destructive, and accurate evaluation of scallop freshness and is applicable to TVB-N prediction for scallops under various preservation treatments.
Patent Information
- Application Number
- CN202610702502.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-05-21
- Publication Date
- 2026-07-10
AI Technical Summary
Existing models have poor versatility when dealing with scallops that undergo various preservation treatments, and cannot accurately predict TVB-N values simultaneously, resulting in inaccurate evaluation of scallop freshness.
An electronic nose system based on multilayer perceptron (MLP) was adopted to detect the volatile odor signal in the headspace of scallops, construct training and test sets, perform data standardization, design a feedforward fully connected neural network model, and optimize the network weights using the backpropagation algorithm to achieve rapid and lossless prediction of TVB-N for scallops with different preservation treatments.
It enables rapid, non-destructive, and accurate prediction of TVB-N content in scallops with different preservation treatments. It has a wide range of applications, meets the needs of rapid on-site testing in the cold chain circulation process, and improves the efficiency of aquatic product quality control.
Smart Images

Figure CN122361737A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of rapid detection technology of aquatic product freshness, specifically to a method for evaluating the freshness of scallops based on a multilayer sensor. Background Technology
[0002] Shellfish are a crucial pillar of my country's aquaculture industry, holding a significant position due to their rich nutritional and economic value. According to data from the "China Fisheries Statistical Yearbook 2024," my country's total shellfish production has reached 17.14 million tons, with scallops accounting for 1.85 million tons annually, ranking third among shellfish producers. Scallop meat is tender and rich in high-quality protein, essential amino acids, various vitamins, and trace minerals such as zinc and selenium. It also contains functional components such as taurine and omega-3 polyunsaturated fatty acids, making it popular among consumers (Grienke, U., Silke, J., & Tasdemir, D. (2014). Bioactive compounds from marine mussels and their effects on human health. FoodChemistry, 142, 48-60; Lee, S.-Y., Kim, H.-J., & Han, J.-S. (2013). Anti-inflammatory Effect of Oyster Shell Extract in LPS-stimulated Raw 264.7 Cells. Preventive nutrition and food science, 18(1), 23-29). However, shellfish such as scallops are characterized by their loose structure, high water activity, and rich content of protein and unsaturated fatty acids. During the post-harvest storage, transportation, and sales process, they are highly susceptible to microbial growth, activation of endogenous enzymes, and external physicochemical factors, leading to a series of spoilage phenomena. Specifically, this manifests as decreased meat elasticity, water loss, browning, and accumulation of off-flavors. This not only seriously affects the sensory quality and nutritional value of the product but also poses a risk of foodborne illness due to the accumulation of biogenic amines (Yu, D., Wu, L., Regenstein, JM, Jiang, Q., Yang, F., Xu, Y., & Xia, W. (2019). Recent advances in quality retention of non-frozen fish and fishery products: A review. Critical Reviews in Food Science and Nutrition, 60(10), 1747-1759).Furthermore, the large-scale disposal of spoiled aquatic products not only causes serious waste of resources and economic losses, but also increases carbon emissions and exacerbates the environmental burden (Rathod, NB, Ranveer, RC, Bhagwat, PK, Ozogul, F., Benjakul, S., Pillai, S., & Annapure, US (2021). Cold plasma for the preservation of aquatic food products: An overview. Comprehensive Reviews in Food Science and Food Safety, 20(5), 4407-4425). Therefore, rapid and accurate assessment of scallop freshness is of great significance for improving food safety, reducing carbon emissions, and promoting environmental protection.
[0003] Furthermore, most of the novel TVB-N prediction methods for aquatic products reported in current research only build models for single-treated or untreated aquatic products. When applied to scallops with multiple preservation treatments (such as biological preservatives and chemical preservatives), the prediction accuracy of the models drops significantly due to the large impact of different treatment methods on the volatile components of scallops, resulting in poor versatility. To address these shortcomings, this invention proposes an electronic nose-TVB-N rapid prediction method for scallops with various preservation treatments based on MLP. Summary of the Invention
[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide an electronic nose-TVB-N rapid prediction method for scallops that is applicable to various preservation treatments. This method solves the problem that existing models have poor versatility and cannot be applied to TVB-N prediction of scallops with different preservation treatments at the same time, and achieves non-destructive, rapid and accurate evaluation of scallop freshness during refrigeration.
[0005] This invention provides a rapid prediction method for TVB-N using an electronic nose based on MLP, applicable to scallops undergoing various preservation treatments, comprising the following steps: (1) The headspace volatile odor of scallop samples treated with preservation soaking solution was detected by electronic nose, the response signals of each sensor were collected, and the TVB-N value of scallop samples treated with preservation soaking solution was determined by automatic Kjeldahl nitrogen analyzer. (2) Dataset construction and partitioning The dataset was constructed using the electronic nose sensor response signal as the input feature and the measured TVB-N value as the output variable. Five parallel replicates were set for each refrigeration time point. All samples were randomly divided into training set and test set in a 4:1 ratio. K-fold cross-validation (K=10) was used to ensure the consistency of data distribution between training set and test set. (3) Data standardization processing To eliminate the dimensional differences in the response signals of different electronic nose sensors and improve the model's convergence speed and stability, the data is standardized. The process is as follows: calculate the mean (μ) and standard deviation (s) of each feature based on the training set data, and perform standardization transformation on both the training and test sets. z
[0006] Where x is the original sensor response value and z is the standardized value, this method avoids the leakage of test set data information and ensures the fairness of model evaluation; (4) MLP model structure design and training The MLP model constructed in this invention is a feedforward fully connected neural network, with the following structure: Input layer: The number of nodes is the same as the number of electronic nose sensors (10 nodes), and the input is the standardized electronic nose feature signal; Hidden layer: A single hidden layer is set (the neuron is optimized according to the data features), and the ReLU activation function is used to introduce nonlinear fitting ability, which solves the problem that traditional linear models cannot capture the complex nonlinear relationship between multi-processed scallop volatile signals and TVB-N; Output layer: 1 node, outputting the predicted TVB-N value; The model training uses the backpropagation algorithm with mean squared error (MSE) as the loss function. The network weights and biases are iteratively optimized to minimize the error between the predicted and measured values. (5) Model performance evaluation Mean squared error (MSE), mean absolute error (MAE), and mean absolute percentage error (MAPE) are used to evaluate model performance.
[0007] In one embodiment of the present invention, in step (1), the preservation soaking solution is either an ectoine preservation soaking solution or a compound phosphate preservation soaking solution; wherein the ectoine preservation soaking solution uses ectoine as the sole active ingredient, distilled water or deionized water as the solvent, and is stirred to ensure that the ectoine is completely dissolved and the soaking solution is clear and transparent, thereby preparing a preservation soaking solution with an ectoine concentration of 0.05wt%-0.2wt% (mass fraction); the compound phosphate preservation soaking solution is prepared from sodium tripolyphosphate (72.5wt%), sodium hexametaphosphate (15wt%), and sodium pyrophosphate (12.5wt%), thereby preparing a preservation soaking solution with a concentration of 0.1wt%.
[0008] In one embodiment of the present invention, in step (1), the pretreated scallop adductor muscle is completely immersed in the preservation soaking solution, preferably scallop adductor muscle: soaking solution = 1:3, the soaking temperature is controlled at 4 ℃, and the soaking time is 2 h; during the soaking process, the soaking solution is kept covering the scallop meat, and it is gently stirred once every 30 minutes to ensure uniform contact.
[0009] In one embodiment of the present invention, the scallops are pre-treated as follows: Fresh scallops are selected, and after shelling by hand or machine, the viscera and skirt impurities are removed. The scallops are rinsed with water 3-5 times to remove surface mucus and residual impurities. The rinsed scallop meat is placed on absorbent paper and gently pressed to remove surface free water, ensuring that there are no obvious water droplets on the surface of the meat to avoid diluting the concentration of the soaking solution.
[0010] Compared with existing technologies, the electronic nose-TVB-N rapid prediction method for scallops applicable to various preservation treatments of the present invention achieves the following beneficial effects: This invention employs an electronic nose to collect headspace volatile signals from scallops. By independently standardizing the training set, it eliminates dimensional differences in sensor signals, eliminating the need for complex sample pretreatment. This enables rapid, non-destructive detection of TVB-N (Total Volatility Bypass-N), with short batch testing times, meeting the rapid on-site testing needs of cold chain logistics. It achieves efficient and non-destructive freshness detection, providing a highly effective technical solution for quality control in the aquatic product distribution process.
[0011] This invention addresses the problem of poor versatility in existing models. It incorporates scallop samples from different preservation treatments (ecodyne treatment, phosphate treatment, and a blank control) into a training set, constructing and optimizing an MLP model that can simultaneously adapt to multiple treatment systems. This overcomes the limitation of traditional electronic nose prediction models, which are only applicable to samples with a single treatment, achieving universal and accurate prediction of TVB-N content in scallops with different treatments, thus broadening its applicability. It also solves the problem of large differences in volatile signals and decreased model prediction accuracy caused by different preservation treatments. Attached Figure Description
[0012] Figure 1 Changes in TVB-N for refrigerated scallops; Figure 2 Changes in the electronic nose of refrigerated scallops; Figure 3 Overall structure of the MLP model; Figure 4 Relationship between scallop hyperparameters and model performance metrics; (a) hidden neurons; (b) learning rate; (c) training epoch. Detailed Implementation
[0013] The technical solution of the present invention will be further described in more detail below with reference to specific embodiments and accompanying drawings. It should be understood that the specific embodiments described herein are only for the purpose of helping to understand the present invention and do not constitute a limitation on the present invention.
[0014] Example 1 1. Sample preparation: Select fresh scallops, remove the shells and viscera, take the scallop adductor muscles, rinse with clean water, and gently press with absorbent paper to absorb water; 2. All scallop adductor muscles were divided into a control group and an experimental group.
[0015] Control group (CK): Distilled water, containing no active ingredients.
[0016] Experimental group (Ectoine group): Group E1: 0.05% ectoine solution; Group E2: 0.1% ectoine solution; Group E3: 0.2% ectoine solution.
[0017] Phosphate group: 0.1% phosphate solution (sodium tripolyphosphate 72.5%, sodium hexametaphosphate 15%, sodium pyrophosphate 12.5%).
[0018] The pretreated scallop meat was completely immersed in the above-mentioned preservation soaking solution at a material-to-liquid ratio of 1:3 (g / mL). The soaking temperature was controlled at 4℃ and the soaking time was 2 hours. During the soaking process, the scallop meat was gently stirred with a glass rod once every 30 minutes (10 seconds each time) to ensure that the scallop meat was in full contact with the soaking solution.
[0019] 3. Remove the scallop adductor muscle, absorb any residual liquid on the surface, and seal. Store in a 4 ℃ refrigerator for 6 days, taking samples daily to measure various indicators.
[0020] Example 2 Changes in volatile basic nitrogen (TVB-N) in scallops during refrigeration The determination was performed using the automatic Kjeldahl method as per GB5009.2016, "National Food Safety Standard - Determination of Volatile Basic Nitrogen in Food".
[0021] Figure 1 This image shows the TVB-N values of refrigerated scallops after different treatments. TVB-N value is often used as an important indicator of the freshness of aquatic products, primarily representing the content of alkaline nitrogenous substances resulting from the decomposition of protein and non-protein nitrogenous substances by spoilage microorganisms and enzymes. The Chinese National Food Safety Standard "Fresh and Frozen Aquatic Animal Products" (GB 2733-2015) stipulates that the TVB-N value of scallops should be less than 15 mg / 100 g. Figure 1In the study, TVB-N levels in all groups increased with prolonged storage time, indicating that protein spoilage gradually intensified. This increase was attributed to the degradation of proteins by cathepsins and exogenous enzymes during storage, leading to an increased accumulation of protein hydrolysates and nitrogenous components. The control group showed the largest increase in TVB-N, reaching 16.89 mg / 100 g on day 5 of storage, exceeding 15 mg / 100 g. The phosphate group had significantly lower TVB-N values than the control group, which is attributed to phosphate's ability to effectively reduce mechanical damage to tissue structures, thereby decreasing the release of endogenous enzymes and the activity of psychrophilic bacteria. At the same addition level of 0.1%, the ectoine group had significantly lower TVB-N values than the phosphate group, indicating that ectoine has great potential as a phosphate substitute. The 0.2% ectoine group showed the smallest change in TVB-N levels over time, possibly because ectoine inhibited microbial proteolytic activity, thus delaying the increase in TVB-N levels.
[0022] Example 3 Changes in the electronic nose of scallops during refrigeration Flavor compounds were analyzed using an electronic nose. A 1 g scallop sample was placed in a 20 mL headspace vial and incubated at 60°C for 10 min at 500 rpm before injection. The instrument was calibrated before analysis. Gaseous compounds in the headspace were extracted using the sensor array at a flow rate of 300 mL / min. The entire measurement process lasted 6 minutes.
[0023] Figure 2 The effects of different treatments on the electronic nose of refrigerated scallops were investigated. Ten metal oxide sensors reflected changes in headspace volatile odors in the samples. Significant changes in the response values of the W5S and W1W sensors were observed during storage. During seafood storage, ammonia and nitrogen oxides produced by the decomposition of proteins and other substances are closely related to the freshness of seafood. The W5S sensor can reflect the freshness of seafood by detecting changes in nitrogen oxides. The W1W sensor was used to detect inorganic sulfides. During seafood storage, protein decomposition produces inorganic sulfides such as hydrogen sulfide, which can effectively reflect changes in volatile substances during the storage process, thus assessing freshness. Sensor results indicate that with prolonged storage, the oxidation of oils produces aldehydes, ketones, and other off-odor substances, or microbial growth and metabolism release volatile products, causing the scallop adductor muscles to emit unpleasant odors such as rancidity and putridity. In the later stages of storage, the response value of the W5S sensor in the ectoine treatment group was significantly reduced compared to the control group, indicating that ectoine may reduce nitrogen oxide production by inhibiting the production of protein decomposition products, which is consistent with the results of TVB-N.
[0024] Example 4 An electronic nose-TVB-N prediction model was constructed based on MLP.
[0025] (1) Dataset construction and partitioning The electronic nose sensor response signals of scallops during storage in all treatment groups (control group, 0.05% ectoine group, 0.1% ectoine group, 0.2% ectoine group, and 0.1% phosphate group) were selected as input features, and the measured TVB-N value was used as the output variable to construct the dataset. Five parallel replicates were set for each refrigeration time point, and all sample data were randomly divided into training and test sets in a 4:1 ratio. K-fold cross-validation (K=10) was used to ensure the consistency of data distribution between the training and test sets.
[0026] (2) Data standardization processing The mean (μ) and standard deviation (s) of each feature are calculated based on the training set data, and then the training set data and the test set data are standardized respectively.
[0027] z
[0028] Where x is the original sensor response value and z is the standardized value. This method avoids the leakage of test set data information and ensures the fairness of model evaluation.
[0029] (3) MLP model structure design and training The MLP model constructed in this invention is a feedforward fully connected neural network, with the structure as follows: Figure 3 As shown: Input layer: The number of nodes is the same as the number of electronic nose sensors (10 nodes), and the input is the standardized electronic nose feature signal; Hidden layer: A single hidden layer is set (the neuron can be optimized according to the data features). The ReLU activation function is used to introduce nonlinear fitting ability, which solves the problem that traditional linear models cannot capture the complex nonlinear relationship between multi-processed scallop volatile signals and TVB-N. Output layer: 1 node, output is the predicted TVB-N value.
[0030] The model training employs the backpropagation algorithm, using mean squared error (MSE) as the loss function. It iterative optimization of network weights and biases minimizes the error between predicted and measured values. The initial training configuration includes hidden neurons, a learning rate of 10, and training epochs of 0.01 and 300, respectively. Figure 4 This paper demonstrates the relationship between hyperparameters and model performance metrics in the TVB-N prediction model for scallops. The model performance fluctuates with increasing number of hidden neurons; at 14 neurons, the model RV... 2 The model reaches its peak performance, while MSE and MAE decrease to a lower level, indicating that the model's nonlinear fitting ability and generalization ability are optimally balanced at this point. Therefore, the optimal number of hidden neurons is determined to be 14. When the learning rate is 0.18, the model's R² is...2 Reaching its maximum value, MSE and MAE decrease to their minimum, maximizing model performance while ensuring training efficiency; therefore, the optimal learning rate was determined to be 0.18. As the number of training epochs increased from 100 to 500, the model's R² generally showed an upward trend, while MSE and MAE continuously decreased. This indicates that increasing the number of training epochs can improve the model's fitting ability and prediction accuracy. However, when the number of training epochs exceeds 350, the improvement in model performance tends to plateau, and further increases may lead to overfitting. Therefore, the optimal number of training epochs was determined to be 350.
[0031] (4) Model performance evaluation Mean squared error (MSE), mean absolute error (MAE), and mean absolute percentage error (MAPE) are used to evaluate model performance. The model built with optimal parameters has R... 2 =0.9502, MSE=0.4485, MAE=0.4928, MAPE=4.1598%, indicating that the model has excellent fitting ability and prediction accuracy. To further verify the actual prediction effect of the model, a portion of electronic nose data was randomly selected for TVB-N prediction, as shown in Table 1. The results show that the deviation between the predicted and measured values is small, and the model exhibits good prediction consistency on actual samples, providing a reliable method for rapid detection of TVB-N in scallops under different treatment groups during cold storage. This model not only achieves rapid, non-destructive, and accurate prediction of TVB-N content in scallops during cold storage, but also constructs a general prediction system adaptable to multiple treatment conditions, providing important theoretical basis and technical support for rapid screening of scallop freshness after different preservation methods in actual production.
[0032] Table 1 Comparison of predicted and actual values for TVB-N
[0033] The embodiments provided above are not intended to limit the scope of the invention, nor are the described steps intended to limit the order of execution. Any obvious modifications made to the invention by those skilled in the art based on existing common knowledge also fall within the scope of protection defined by the claims.
Claims
1. A method for evaluating the freshness of scallops based on a multilayer perceptron, characterized in that, The method includes: (1) The headspace volatile odor of scallop samples treated with preservation soaking solution was detected by electronic nose, the response signals of each sensor were collected, and the TVB-N value of scallop samples treated with preservation soaking solution was determined by automatic Kjeldahl nitrogen analyzer. (2) Dataset construction and partitioning The dataset was constructed using the electronic nose sensor response signal as the input feature and the measured TVB-N value as the output variable. Five parallel replicates were set for each refrigeration time point. All samples were randomly divided into training set and test set in a 4:1 ratio. K-fold cross-validation was used with K=10. (3) Data standardization processing The data is standardized as follows: The mean and standard deviation of each feature are calculated based on the training set data, and standardization transformations are performed on both the training and test sets: z Where x is the original sensor response value, z is the standardized value, μ is the mean, and s is the standard deviation; (4) MLP model structure design and training The MLP model is a feedforward fully connected neural network, with the following structure: Input layer: The number of nodes is the same as the number of electronic nose sensors, and the input is the standardized electronic nose feature signal; Hidden layer: A single hidden layer is set, and the ReLU activation function is used to introduce non-linear fitting capability; Output layer: 1 node, outputting the predicted TVB-N value; The model training uses the backpropagation algorithm with mean squared error (MSE) as the loss function. The network weights and biases are iteratively optimized to minimize the error between the predicted and measured values. (5) Model performance evaluation Mean squared error (MSE), mean absolute error (MAE), and mean absolute percentage error (MAPE) were used to evaluate model performance.
2. The method according to claim 1, characterized in that, In step (1), the preservation soaking solution is either ectoine preservation soaking solution or compound phosphate preservation soaking solution.
3. The method according to claim 2, characterized in that, The ectoine preservation soaking solution uses ectoine as the sole active ingredient, with distilled or deionized water as the solvent. The solution is stirred to ensure complete dissolution of ectoine, resulting in a clear and transparent soaking solution with an ectoine concentration of 0.05wt%-0.2wt%. The compound phosphate preservation soaking solution is prepared from sodium tripolyphosphate, sodium hexametaphosphate, and sodium pyrophosphate, resulting in a preservation soaking solution with a concentration of 0.1wt%.
4. The method according to claim 1, characterized in that, In step (1), the pre-treated scallop adductor muscles are completely immersed in the preservation soaking solution, the soaking temperature is controlled at 4 ℃, and the soaking time is 2 h; during the soaking process, the soaking solution is kept covering the scallop meat, and it is gently stirred once every 30 minutes to ensure uniform contact.
5. The method according to claim 4, characterized in that, The pre-processing method for scallops is as follows: Select fresh scallops, remove the shells by hand or machine, remove the viscera and skirt impurities, rinse with water to remove surface mucus and residual impurities; place the rinsed scallop meat on absorbent paper and gently press to absorb the surface free water, ensuring that there are no obvious water droplets on the surface of the meat.
