Method and system for rapid detection of reheating degree of meat dishes based on electrochemical sensing

CN122282892BActive Publication Date: 2026-08-28INST OF AGRO FOOD SCI & TECH CHINESE ACADEMY OF AGRI SCI +1
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202610677468.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2026-05-18
Publication Date
2026-08-28
Estimated Expiration
2046-05-18

AI Technical Summary

Technical Problem

[0011]本发明还有一个目的是提供一种基于电化学传感的肉类菜肴复热程度快速检测方法,其旨在解决现有检测方法主观性强、耗时长、破坏样品、设备昂贵,以及单一电化学传感器选择性不足且难以适应不同食材属性和复热方式变化的技术问题

Benefits of technology

1、本发明提供的基于电化学传感的肉类菜肴复热程度快速检测方法,通过融合电化学传感器完整时间响应曲线的动态特征参数(如响应上升时间、峰值、积分面积等)与食材属性参数(肉类种类、脂肪含量、烹饪方式)及复热方式参数,并利用深度学习模型进行综合判别,有效解决了现有技术中单一传感器选择性差、易受干扰以及模型难以适应食材和复热方式变化的难题。该方法能够快速、准确、无损地输出复热程度等级,为餐饮业、食品加工及家庭烹饪提供了一种客观、通用且高效的复热品质评价工具,显著提升了检测的智能化水平和结果的可靠性。

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122282892B_ABST
    Figure CN122282892B_ABST
Patent Text Reader

Abstract

The application discloses a meat dish reheating degree rapid detection method and system based on electrochemical sensing, and belongs to the technical field of food detection. In view of the problem that the reheating degree of a meat dish cannot be quickly and accurately evaluated in the prior art, the application first places a meat dish to be detected in a space to generate headspace gas at a preset equilibrium temperature and time; uses an electrochemical sensor unit to collect a complete time response curve of the headspace gas acting on a sensitive layer, extracts dynamic characteristic parameters such as a response rise time, a peak response value, a time to reach the peak, a curve integral area and a half-life period in a falling stage; meanwhile, obtains food material attribute parameters and reheating mode parameters; and inputs the above parameters into a pre-trained deep learning prediction model, and outputs a reheating degree grade from the model and presents the reheating degree grade on a display unit. The method can be used for quickly detecting the reheating degree of a meat dish in the catering industry, food processing and home cooking, and guarantees food quality and edible safety.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of food testing technology. More specifically, this invention relates to a rapid detection method and system for the reheating degree of meat dishes based on electrochemical sensing. Background Technology

[0002] Meat dishes often require refrigeration or freezing after cooking and reheating before consumption. The degree of reheating directly affects the sensory quality, textural characteristics, and food safety of the dish. Insufficient reheating may result in the core temperature not reaching the required sterilization level, posing a risk to microbial safety; excessive reheating will cause the meat to become tough, lose juices, and undergo excessive fat oxidation, producing unpleasant flavors. Therefore, accurately determining the degree of reheating of meat dishes is of great importance for the catering industry, food processing, and home cooking.

[0003] Currently, the methods for evaluating the reheating degree of meat dishes mainly rely on sensory evaluation and routine physicochemical analysis. Sensory evaluation typically involves trained evaluators scoring the color, odor, taste, and texture of the reheated samples. This method is affected by individual sensory differences, psychological state, and fatigue levels of the evaluators, often resulting in significant dispersion between different evaluators and even at different times for the same evaluator, leading to poor repeatability and reproducibility. Furthermore, sensory evaluation requires a relatively long time to organize the evaluation team, making rapid, online, or on-site testing impossible. For applications requiring large-scale testing or production line quality control, the efficiency of sensory evaluation is clearly insufficient.

[0004] Physicochemical testing methods include determining lipid oxidation products (such as thiobarbituric acid value), protein degradation products (such as volatile basic nitrogen), and free fatty acid content. These methods typically require pretreatment steps such as crushing, homogenizing, extraction, and derivatization of the sample, making the process cumbersome and time-consuming, ranging from tens of minutes to several hours. Furthermore, the testing process requires the use of various chemical reagents and organic solvents, increasing costs and generating chemical waste. More importantly, physicochemical testing is destructive; samples tested cannot be used or sold afterward. For scenarios requiring rapid information on the degree of reheating to guide subsequent processing or consumer decisions, these methods are insufficient.

[0005] Instrumental analytical methods, such as gas chromatography and gas chromatography-mass spectrometry, can separate and identify volatile organic compounds generated during reheating, providing relatively detailed component information. However, these instruments are expensive, bulky, require specialized personnel to operate, have long analysis cycles, and typically necessitate sending samples to a laboratory for testing, making rapid on-site detection impossible. Even with headspace sampling to simplify pretreatment, the entire analytical process still takes a considerable amount of time, failing to meet the needs of real-time detection.

[0006] In the practical application of reheat degree detection, there is a technical challenge that has not yet been effectively solved: the degree of reheating depends not only on the reheating temperature and time, but also significantly on the properties of the food itself and the reheating method. Different types of meat (such as livestock and poultry, red and white meat) have differences in muscle fiber structure, fat content, moisture content and distribution, resulting in different patterns of volatile release during reheating. Furthermore, the cooking methods used for the initial cooking of meat dishes (frying, stir-frying, steaming, braising, baking, etc.) alter their surface structure, porosity, and internal component distribution, thus affecting the mass and heat transfer processes during subsequent reheating. Reheating methods (microwave reheating, water bath reheating, direct heat reheating) are completely different in their heating principles and heat transfer paths. Microwaves rely on the rotation of polar molecules to generate heat, water baths rely on heat conduction, and direct heat combines conduction and radiation. These differences lead to significant differences in the volatile characteristics produced when the same dish achieves the same degree of reheating under different methods. Existing testing methods typically only establish evaluation models for a single reheating condition or a single food variety. Once the type of food, cooking method, or reheating method changes, the applicability of the original model decreases significantly, requiring recalibration and verification. This greatly limits the versatility and promotional value of the testing methods.

[0007] Researchers have attempted to evaluate the degree of reheating of meat using a single-gas sensor-based detection device, indirectly reflecting the reheating state by measuring changes in the concentration of specific volatiles (such as ammonia and hydrogen sulfide). However, the main difficulties encountered in practical detection are: the headspace gas composition generated during meat reheating is complex, with multiple volatile organic compounds coexisting; the cross-sensitivity of a single sensor to different gases leads to insufficient selectivity, making it difficult for the response signal to uniquely correspond to the degree of reheating. Furthermore, the sensor is sensitive to changes in ambient temperature and humidity, and the reheating of meat dishes is often accompanied by a large release of water vapor, which interferes with the sensor's response signal and causes misjudgments. To overcome these difficulties, researchers have tried combining signal filtering and temperature compensation methods, but due to the lack of dynamic analysis of the complete response process, relying solely on steady-state response values ​​or single feature points still makes it difficult to effectively distinguish the subtle differences in volatile composition at different degrees of reheating.

[0008] On the other hand, the degree of reheating itself is a continuously changing process, but in practical applications, it often needs to be divided into several levels (such as insufficient reheating, moderate reheating, and excessive reheating). The boundaries between different levels are not distinct, but rather there are gradual intervals. Traditional threshold-based methods are prone to classification errors when dealing with such ambiguous boundaries. To establish a robust level discrimination model, it is necessary to extract information from multiple dimensions that can characterize the dynamic process of volatile release and integrate prior knowledge such as food properties and reheating methods. However, existing technologies lack an effective multi-source information fusion framework, resulting in insufficient model generalization ability.

[0009] In summary, existing technologies for detecting the degree of reheating in meat dishes suffer from the following main problems: sensory evaluation is highly subjective and has poor repeatability; physicochemical testing is time-consuming, cumbersome, and damages samples; instrumental analysis equipment is expensive and cannot be used on-site; single sensors have poor selectivity and are susceptible to interference; and existing models are difficult to adapt to changes in food properties and reheating methods. These problems limit the development of rapid, accurate, and versatile methods for detecting the degree of reheating. Therefore, a novel detection method that can overcome these shortcomings is needed. Summary of the Invention

[0010] One object of the present invention is to solve at least the above-mentioned problems and to provide at least the advantages that will be described later.

[0011] Another objective of this invention is to provide a rapid detection method for the reheating degree of meat dishes based on electrochemical sensing, which aims to solve the technical problems of existing detection methods, such as high subjectivity, long time consumption, sample damage, expensive equipment, insufficient selectivity of single electrochemical sensors, and difficulty in adapting to different food properties and reheating methods.

[0012] To achieve these objectives and other advantages of the present invention, a rapid detection method for the reheating degree of meat dishes based on electrochemical sensing is provided, comprising the following steps: Step 1: Place the meat dish to be tested in a closed or semi-closed space and maintain it at a preset equilibrium temperature for a preset equilibrium time to generate headspace gas. Step 2: Use the electrochemical sensor unit to collect the electrochemical response signal generated by the interaction between the headspace gas and the sensor sensitive layer, and record the complete time response curve of the electrochemical response signal; Step 3: Extract dynamic characteristic parameters from the time response curve, which include at least one or more of the following: response rise time, peak response value, time to reach peak value, area under the integral of the response curve, and half-life of the fall phase. Step 4: Obtain the ingredient attribute parameters and reheating method parameters of the meat dish to be tested. The ingredient attribute parameters include at least one of the following: meat type, fat content, water content, and cooking method parameters. The meat type is quantified by the ratio of muscle fiber types or red / white meat classification. The cooking method parameters include at least one of the following: heating temperature, heating time, temperature change pattern, and endpoint temperature corresponding to different cooking methods. Cooking methods include at least one of the following: frying, stir-frying, steaming, braising, and baking. The reheating method parameters include at least one of the following: reheating start temperature, reheating power, reheating time, and reheating endpoint temperature. Reheating methods include at least one of the following: microwave reheating, water bath reheating, and direct flame reheating. Step 5: Input the dynamic feature parameters extracted in Step 3 and the food attribute parameters and reheating method parameters obtained in Step 4 into a pre-trained deep learning prediction model. The deep learning prediction model is a model built on a deep neural network or convolutional neural network, which is used to map the reheating level according to the input features. Step 6: The deep learning prediction model outputs the degree of reheating of the reheated meat dish to be detected and presents it on the display unit.

[0013] Preferably, the specific method for extracting dynamic feature parameters in step three of the rapid detection method for the reheating degree of meat dishes based on electrochemical sensing includes: S31. Perform model fitting for the rising and falling phases of the time response curve respectively; S32. Fit the rising phase curve to an exponential growth model: ; Where R(t) is the response value at time t, R0 is the initial baseline value, A is the adsorption amplitude, and τ is the adsorption time constant; A and τ are extracted as feature parameters. S33. Fit the descent phase curve to an exponential decay model: ; Among them, R ∞ To recover the stable response value, B is the desorption amplitude and τ3 is the desorption time constant; B and τ3 are extracted as characteristic parameters. S34. Compare the fitted feature parameters A, τ, B, τ3 with the peak response value R. max Together they constitute the dynamic characteristic parameters.

[0014] Preferably, the training process of the deep learning prediction model in the rapid detection method for the reheating degree of meat dishes based on electrochemical sensing includes: T1. Obtain multiple samples of meat dishes with different known reheating methods and different reheating times, and collect their electrochemical response signals and extract dynamic characteristic parameters according to steps one to three. At the same time, record the food attribute parameters and reheating method parameters of each sample. T2. The dynamic feature parameters, food attribute parameters and reheating method parameters are combined into an input feature vector, and the corresponding reheating degree level is used as the output label. T3. Construct a prediction model based on a deep neural network or a convolutional neural network, and use supervised learning to train the deep learning prediction model with the goal of minimizing the loss function between the predicted value and the label. T4. During training, cross-validation is used to optimize the model hyperparameters, and an early stopping mechanism is introduced to prevent overfitting.

[0015] Preferably, in the rapid detection method for the reheating degree of meat dishes based on electrochemical sensing, step six divides the reheating degree into four levels: no reheating, slight reheating, moderate reheating, and excessive reheating, or uses a reheating degree score expressed as a continuous numerical value.

[0016] Preferably, in the aforementioned rapid detection method for the reheating degree of meat dishes based on electrochemical sensing, step six, the determination of the reheating degree level adopts a graded threshold determination method, which includes: Establish a grading threshold database: A mapping and association database is pre-established through systematic experiments for different combinations of food attribute parameters, cooking method parameters, and reheating method parameters. The database construction process includes: measuring the electrochemical response signal characteristic values ​​corresponding to the concentrations of aldehydes produced at different reheating levels, which are either peak response values ​​or response integral areas; simultaneously, combining the human sensory evaluation results of the corresponding samples, determining the aldehyde concentration thresholds corresponding to each reheating level, including no reheating, mild reheating, moderate reheating, and excessive reheating; finally, storing the data as grading threshold ranges for combinations of food attribute parameters, cooking method parameters, and reheating method parameters; wherein the reheating method includes at least one of microwave reheating, water bath reheating, and direct flame reheating. Threshold matching and correction during detection: Based on the ingredient attribute parameters, cooking method parameters, and reheating method parameters of the meat dish to be detected obtained in step four, the corresponding initial grading threshold T0 is matched and retrieved from the grading threshold library; the initial grading threshold T0 is a set of threshold vectors, including the boundary threshold T from no reheating to slight reheating. 01 The boundary threshold T from mild reheating to moderate reheating 02 And the boundary threshold T from moderate reheating to excessive reheating 03 Each boundary threshold represents the concentration value of the key aldehyde substance that divides adjacent reheating levels under the combined conditions, in micrograms per gram or a value corresponding to the response value of the electrochemical sensor. Adaptive correction: Obtain the baseline correction factor B for the current testing batch, where B = X. blank -X ref ;X blank X represents the characteristic value of the electrochemical response signal of the unreheated sample in the current batch. ref The standard response value is the value of the sample that was not reheated during library construction; the initial grading threshold is adaptively corrected using the formula T. cal,i =T 0i +B, where i=1,2,3; Level determination: Compare the corrected grading threshold T cal,i The numerical relationship between the peak response value or the response integral area X obtained from the current detection: If X ≤ T cal,1 If Tcal,1 <X≤T cal,2 If T cal,2 <X≤T cal,3 If X > T, it is determined to be moderate reheating; cal,3 If so, it is determined to be excessive reheating.

[0017] Preferably, the rapid detection method for the reheating degree of meat dishes based on electrochemical sensing has an initial grading threshold T. 01 T 02 T 03 The methods for determining this include: Establish multiple mapping relationships: First mapping relationship: Establish the correspondence between sensor response values ​​and the content of aldehydes in the headspace; The second mapping relationship is to establish the correspondence between the content of aldehydes and the reference index, which is the reactant value of thiobarbituric acid or the peroxide value. The third mapping relationship: Based on the results of human sensory evaluation, establish the correspondence between the reference index and the degree of reheating; Determine the grading threshold: Based on the transfer of the first to third mapping relationships mentioned above, calculate the sensor response value corresponding to the boundary of each reheating degree level, and use it as the initial grading threshold T. 01 T 02 T 03 ; Consistency verification and storage: The consistency of candidate grading thresholds for the same level boundary calculated through different paths is verified; when the difference exceeds the preset range, the threshold determined by sensory evaluation combined with reference indicators shall be used, and the final determined grading threshold shall be stored in the grading threshold library.

[0018] Preferably, in the rapid detection method for the reheating degree of meat dishes based on electrochemical sensing, in step two, the electrochemical sensor unit includes at least one gas-sensitive sensing electrode, which is a gas-sensitive sensing electrode with a ternary composite material formed of MXene, ZIF-8, and Co3O4 as the sensitive layer and an interdigitated electrode as the transducer element; the preparation method of the gas-sensitive sensing electrode includes the following steps: Ti3C2T x Nitrogen-doped porous carbon-confined Co3O4 nanocrystalline composite materials derived from MXene and ZIF-8 were mixed at a mass ratio of 1:1 to 1:3, dispersed in a solvent, and then subjected to ultrasonic treatment to obtain a ternary composite sensitive material. The obtained ternary composite sensitive material is made into a slurry, coated onto the interdigital electrode, and dried to obtain the final product.

[0019] Preferably, in the rapid detection method for the reheating degree of meat dishes based on electrochemical sensing, step five involves performing an interference mode decoupling step before inputting the dynamic feature parameters and ingredient attribute parameters into the deep learning prediction model. W1. Pre-establish an interference characteristic spectral library: Expose the electrochemical sensor unit to at least one interference standard gas, which includes at least one of water vapor, ethanol vapor, and acetic acid vapor. Extract the dynamic characteristic parameters of the interferences using the same method as in steps two and three to construct the interference characteristic matrix M. int M int Each column corresponds to a dynamic feature parameter vector of a type of interference. W2, regarding the interference feature matrix M int Perform singular value decomposition to obtain the orthogonal basis vectors U of the disturbance subspace. int U int Each column represents the main direction of change in the feature space of the interfering object; W3, convert the currently detected original feature vector F cur Projected onto U int On the orthogonal complement space, we obtain the projection vector after eliminating the influence of interference: ; Among them, F cur The feature vector is composed of the dynamic feature parameters extracted in step three and the ingredient attribute parameters, cooking method parameters, and reheating method parameters obtained in step four. The cooking method parameters include at least one of heating temperature, heating time, temperature change pattern, and endpoint temperature; the reheating method parameters include at least one of reheating start temperature, reheating power, reheating time, and reheating endpoint temperature. int U represents the orthogonal basis vectors of the interference subspace; int T F cur U represents the projection coefficients of the original feature vector onto the interference subspace. int (U int T F cur F represents the component of the original feature vector that belongs to the interference subspace; proj The projection vector after eliminating interference components is the projection of the original eigenvector onto the orthogonal complement space of the interference subspace. W4, with F proj As input to the deep learning prediction model, the feature vector is used.

[0020] The present invention also provides a detection system for implementing the above-described rapid detection method for the reheating degree of meat dishes based on electrochemical sensing, comprising: A portable detection device is provided with an electrochemical sensor unit, which includes at least one gas-sensitive electrode for collecting headspace gas from the reheated meat dish to be tested and generating an electrochemical response signal. The signal acquisition and processing module is used to record the complete time response curve of the electrochemical response signal and extract dynamic characteristic parameters from it. The input module is used to input the ingredient attribute parameters, cooking method parameters, and reheating method parameters of the reheated meat dish to be tested. The ingredient attribute parameters include at least one of the following: meat type, fat content, and water content. The cooking method parameters include at least one of the following: heating temperature, heating time, temperature change pattern, and endpoint temperature. The reheating method parameters include at least one of the following: reheating start temperature, reheating power, reheating time, and reheating endpoint temperature. A storage module that stores pre-trained deep learning prediction models; The prediction module is used to input dynamic feature parameters, food attribute parameters, and reheating method parameters into the deep learning prediction model to obtain the reheating degree level. The display module is used to output and display the degree of reheating.

[0021] Preferably, the portable detection device in the detection system is configured to be attached to the inside or outside of the meat dish packaging, or directly to the surface of the meat dish, and is powered and triggered by a mobile terminal via near-field communication.

[0022] The present invention has at least the following beneficial effects: 1. The present invention provides a rapid detection method for the reheating degree of meat dishes based on electrochemical sensing. This method integrates dynamic characteristic parameters (such as response rise time, peak value, and integral area) of the complete time response curve of the electrochemical sensor with ingredient attribute parameters (meat type, fat content, cooking method) and reheating method parameters, and utilizes a deep learning model for comprehensive discrimination. This effectively solves the problems of poor selectivity of single sensors, susceptibility to interference, and difficulty in adapting models to changes in ingredients and reheating methods in existing technologies. This method can quickly, accurately, and non-destructively output the reheating degree level, providing an objective, universal, and efficient tool for evaluating reheating quality in the catering industry, food processing, and home cooking, significantly improving the intelligence level of detection and the reliability of results.

[0023] 2. This invention extracts deep kinetic characteristic parameters such as adsorption amplitude (A), adsorption time constant (τ), desorption amplitude (B), and desorption time constant (τ3) by finely fitting the rising and falling phases of the sensor's time response curve using exponential growth and exponential decay models, respectively. Compared to using only single static indicators such as peak or steady-state values, these dynamic parameters can more comprehensively and sensitively characterize the complete process of interaction between headspace volatiles and the sensitive layer of meat dishes under different reheating degrees, significantly improving the richness and specificity of feature information, providing higher-quality input for deep learning models, and thus enhancing the model's ability to distinguish subtle reheating differences.

[0024] 3. This invention innovatively introduces a graded threshold determination and adaptive correction mechanism. By pre-establishing a graded threshold library covering different combinations of food properties, cooking methods, and reheating methods, and combining it with the baseline correction factor of the current detection batch, the threshold is dynamically and adaptively corrected. This effectively eliminates systematic errors caused by differences between sensor batches, environmental fluctuations, and changes in sample background. This method not only provides a reliable auxiliary or independent determination path for deep learning prediction models, but also solves the key technical problem of poor applicability of traditional fixed threshold methods under different detection conditions, ensuring the stability and reproducibility of detection results across different times and batches. 4. This invention performs an interference mode decoupling step before the feature input model. By pre-constructing a feature spectral library of interfering substances (water vapor, ethanol, acetic acid, etc.) and using singular value decomposition to obtain an orthogonal basis for the interference subspace, the original feature vector is projected onto this orthogonal complementary space, thereby effectively filtering out the contribution components of common environmental interfering substances to the sensor response signal. This technical solution significantly enhances the anti-interference ability and robustness of the method in real and complex detection environments (such as reheating processes accompanied by the release of large amounts of water vapor), ensuring that the features extracted from the composite response can more realistically reflect the changes in volatiles caused by the degree of reheating itself, and greatly reducing the risk of misjudgment.

[0025] Other advantages, objectives and features of the present invention will be apparent in part from the following description, and in part from the understanding of those skilled in the art through study and practice of the invention. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of the process for rapid detection of the reheating degree of meat dishes based on electrochemical sensing, as described in this invention. Detailed Implementation

[0027] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments, so that those skilled in the art can implement it based on the description.

[0028] It should be understood that terms such as “having,” “comprising,” and “including” as used herein do not exclude the presence or addition of one or more other elements or combinations thereof.

[0029] It should be noted that, unless otherwise specified, the experimental methods described in the following implementation plan are all conventional methods, and the reagents and materials described are all commercially available unless otherwise specified.

[0030] In the description of this invention, the terms "lateral", "longitudinal", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", and "outer" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0031] like Figure 1 As shown, this invention provides a rapid detection method for the reheating degree of meat dishes based on electrochemical sensing, which includes the following steps: Step 1: Place the meat dish to be tested in a closed or semi-closed space and maintain it at a preset equilibrium temperature for a preset equilibrium time to generate headspace gas. A closed or semi-closed space refers to a container that can limit the escape of headspace gas but is not necessarily completely sealed. For example, a glass bottle with a sealing cap, a small environment formed by heat sealing a food-grade plastic bag, or a limited space formed by attaching a portable detection device to the surface of the dish packaging can be used. The preset equilibrium temperature is usually set to a fixed value between 40 degrees Celsius and 60 degrees Celsius, such as 50 degrees Celsius. This temperature can promote the release of volatile substances from the dish matrix into the headspace without causing further thermal changes in the dish. The preset equilibrium time is generally set to 5 to 15 minutes to allow the concentration of volatile substances in the headspace to reach a relatively stable state. Headspace gas refers to the gas phase portion located in a closed or semi-closed space that is in contact with the surface of a solid or liquid sample but not in direct contact with the sample matrix. It contains various organic and inorganic compounds that have volatilized from the sample. Step 2: Acquire the electrochemical response signal generated by the interaction between headspace gas and the sensor's sensitive layer using an electrochemical sensor unit, and record the complete time response curve of the electrochemical response signal. The electrochemical sensor unit refers to a sensing element that can convert the adsorption or reaction of gas molecules into electrical signals (such as current, resistance, and potential). In this method, a chemiluminescence resistive sensor based on the change in the resistance of the sensitive layer is preferred. The complete time response curve refers to the continuous curve of the electrical signal changing with time from the moment the sensor contacts the headspace gas, through the signal rise to the peak value, and then back to the baseline. Step 3: Extract dynamic characteristic parameters from the time response curve, which include at least one or more of the following: response rise time, peak response value, time to reach peak value, integral area of ​​response curve, and half-life of the fall phase. Dynamic characteristic parameters refer to quantitative indicators extracted from the curve that can characterize the dynamic behavior of the interaction between the gas and the sensitive layer, such as the time taken from the start of contact to the peak value, the height of the peak value, and the area under the curve. Step 4: Obtain the ingredient attribute parameters and reheating method parameters of the meat dish to be tested. The ingredient attribute parameters include at least one of the following: meat type, fat content, water content, and cooking method parameters. The meat type is quantified by the ratio of muscle fiber types or red meat / white meat classification. The cooking method parameters include at least one of the following: heating temperature, heating time, temperature change pattern, and endpoint temperature corresponding to different cooking methods. Cooking methods include at least one of the following: frying, stir-frying, steaming, braising, and baking. The reheating method parameters include at least one of the following: reheating start temperature, reheating power, reheating time, and reheating endpoint temperature. Reheating methods include at least one of the following: microwave reheating, water bath reheating, and direct heat reheating. The "ingredient attribute parameters" include the meat type (which can be quantified by the ratio of red muscle fibers to white muscle fibers, for example, the proportion of red muscle fibers in red meat is higher than that in white meat), fat content (expressed in grams per 100 grams), water content (expressed as a percentage), and cooking method parameters (recording the process conditions used during the initial cooking). The “Reheating Method Parameters” include the reheating start temperature (the center temperature of the dish before reheating), reheating power (for microwave reheating, it refers to the output power; for direct flame reheating, it refers to the heating level), reheating time (the total time from the start to the end of heating), and reheating end temperature (the center temperature of the dish when reheating stops). Step 5: Input the dynamic feature parameters extracted in Step 3 and the food attribute parameters and reheating method parameters obtained in Step 4 into a pre-trained deep learning prediction model. The deep learning prediction model is a model built on deep neural networks or convolutional neural networks, used to map the reheating level based on the input features. The deep learning prediction model refers to a multi-layer nonlinear mapping model built on deep neural networks or convolutional neural networks. After being trained on a large number of samples, this model can output the reheating level based on the input feature vector. Step Six: The deep learning prediction model outputs the degree of reheating of the reheated meat dish to be detected and presents it on the display unit. The display unit can be an LCD screen, a light-emitting diode array, or an interface for communicating with a mobile terminal.

[0032] In the above technical solution, step one is implemented as follows: Take a portion of the meat dish to be tested, such as a portion of refrigerated braised pork, and place it in a 200 ml glass sample bottle. Tighten the cap with a silicone sealing gasket to form a sealed space. Place the sample bottle in a constant temperature chamber, set the equilibrium temperature to 50 degrees Celsius, and the equilibrium time to 10 minutes. During this period, volatile substances in the dish (such as aldehydes, ketones, alcohols, etc.) are generated from the degradation of fats and proteins and diffuse into the headspace. Step two is implemented as follows: Connect the electrochemical sensor unit (e.g., a sensor chip composed of interdigitated electrodes and a sensitive layer) to an electrochemical workstation or a portable potentiostat via wires. After the equilibrium time is completed, use a syringe to extract 5 ml of gas from the headspace of the sample bottle and inject it into the sensor test chamber, or insert the sensor probe directly into the headspace region of the sample bottle. Set the electrochemical workstation to constant voltage mode (e.g., apply a DC voltage of 0.1 volts), with a sampling frequency of 10 points per second, and continuously record the changes in the sensor's current or resistance value. Timing begins from the moment of gas injection, typically recording a complete response curve for 300 to 600 seconds, including the baseline phase (a stable signal before gas contact), the rise phase (rapid signal change due to gas interaction with the sensitive layer), the peak plateau (adsorption and desorption reaching dynamic equilibrium), and the fall phase (signal gradually recovering after gas supply stops). Step three involves importing the recorded time response data into data processing software (such as Python's NumPy library or MATLAB). First, the curve is smoothed, for example, using a moving average filter to remove high-frequency noise. Then, differentiation is used to find the response start point (the moment the signal first exceeds the baseline noise threshold), the peak point (the moment corresponding to the maximum signal value), and the recovery point (the moment the signal drops to within 10% of the difference between the peak and the baseline). Based on this, the response rise time (the time difference from the start point to the peak point), the peak response value (the maximum signal value minus the baseline average), the time to reach the peak (the absolute time of the peak point), the area under the integral of the response curve (integrating the signal values ​​of the rise and fall phases over time, typically using the trapezoidal rule), and the half-life of the fall phase (the time taken for the signal to fall from the peak point to half the difference between the peak and the baseline). These parameters are stored numerically. Step four is implemented as follows: Ingredient attribute parameters are input via the user interface or retrieved from a pre-stored database. For example, for braised pork belly, the meat type is selected as "pork," and the red / white meat attribute is further quantified by myoglobin content, setting the red meat label to 1. Fat content is input after consulting a food composition table or measuring it using a rapid fat analyzer, for example, 25 grams per 100 grams. Moisture content is input after measurement using a drying method, for example, 55%.The cooking method parameters are recorded as "Braised," with corresponding heating temperatures of 180 degrees Celsius (caramelizing stage) and 100 degrees Celsius (stewing stage), a heating time of 60 minutes, and a three-stage temperature change pattern (first high-temperature stir-frying, then medium-heat stewing, and finally reducing the sauce), with a final temperature of 100 degrees Celsius (the center temperature at the end of stewing). Reheating parameters: The user selects "Microwave Reheating," inputs a starting reheating temperature of 4 degrees Celsius (when removed from the refrigerator), a reheating power of 700 watts, a reheating time of 2 minutes, and a final reheating temperature of 65 degrees Celsius measured by an infrared thermometer. Step 5 is implemented as follows: The dynamic feature parameters obtained in Step 3 (e.g., rise time 15 seconds, peak response 120 nanoamps, peak time 45 seconds, integral area 3500 nanoamps·second, half-life 30 seconds) and the food attribute parameters and reheating method parameters obtained in Step 4 are combined into a one-dimensional feature vector, for example, [15, 120, 45, 3500, 30, 1, 25, 55, 180, 60, 100, 4, 700, 120, 65]. This vector is input into a pre-trained deep learning prediction model. The model can be a three-layer fully connected deep neural network with 15 nodes in the input layer, 32 and 16 nodes in the two hidden layers respectively, and 4 nodes in the output layer (corresponding to four reheating levels) or 1 node (outputting a continuous score). The model parameters (weights and biases) have been determined through training with a large number of samples. After receiving the feature vector, the model performs forward propagation calculations, executing linear transformations and nonlinear activations (such as the ReLU function) layer by layer, ultimately outputting a four-dimensional probability vector. The index corresponding to the maximum value is the predicted reheating level. Step six is ​​implemented by sending the model's output level (e.g., "moderate reheating") via serial communication to a 128×64 resolution organic light-emitting diode (OLED) display screen for text display. Different colored LED indicators can be used (green for moderate reheating, yellow for mild reheating, and red for excessive reheating). Users can then determine whether the reheating level of the meat dish is appropriate based on the displayed result.

[0033] The above-described technical solution enables rapid and non-destructive detection of the reheating degree of meat dishes. Compared to methods relying solely on the steady-state value of a single sensor, this implementation significantly improves the accuracy and robustness of detection by integrating multiple dynamic features from the complete time response curve, as well as prior information such as ingredient properties and reheating methods. This method avoids the subjectivity of sensory evaluation and the destructive nature of physicochemical analysis. Furthermore, due to the use of a portable electrochemical sensor, the equipment cost is far lower than that of a gas chromatography-mass spectrometry (GC-MS) system, making it suitable for on-site use in restaurant kitchens, food processing production lines, or home kitchens. In addition, by inputting different ingredients and reheating parameters, the same model can adapt to various dish types without retraining, demonstrating good versatility.

[0034] In another technical solution, the specific method for extracting dynamic feature parameters in step three of the rapid detection method for the reheating degree of meat dishes based on electrochemical sensing includes: S31. Perform model fitting for the rising and falling phases of the time response curve respectively; S32. Fit the rising phase curve to an exponential growth model: ; Where R(t) is the response value at time t, R0 is the initial baseline value, A is the adsorption amplitude, and τ is the adsorption time constant; A and τ are extracted as characteristic parameters; the adsorption amplitude refers to the maximum increment that the signal can rise in this exponential growth model, reflecting the total adsorption capacity of the sensitive layer for volatiles. The adsorption time constant is a parameter characterizing the speed of the adsorption process, and its value is equal to the time required for the signal to rise from the beginning to reach approximately 63.2% of the final adsorption amplitude. The smaller the time constant, the faster the adsorption. Among them, the exponential growth model is a mathematical function used to describe the monotonically increasing signal over time with a gradually decreasing rate of increase. Its basic form is that the response value is equal to the initial value plus an increment that is exponentially related to time. This model is used to fit the process of the signal rising from the baseline to the peak after the sensor comes into contact with headspace gas. S33. Fit the descent phase curve to an exponential decay model: ; Among them, R ∞ To recover the stable response value, B is the desorption amplitude, and τ3 is the desorption time constant; B and τ3 are extracted as feature parameters; the exponential decay model is used to describe the process of the signal decreasing from the peak to the stable value. Its basic form is that the response value is equal to the final stable value plus a component that decreases exponentially with time; the desorption amplitude refers to the total change from the peak to the final stable value during the signal decline phase, reflecting the degree of complete desorption of volatiles adsorbed on the sensitive layer; the desorption time constant characterizes the speed of the desorption process and is equal to the time required for the signal to decrease from the peak to approximately 63.2% of the desorption amplitude. The smaller the time constant, the faster the desorption. S34. Compare the fitted feature parameters A, τ, B, τ3 with the peak response value R. max Together, these parameters constitute the dynamic characteristic parameters. The peak response value refers to the net response after subtracting the baseline value from the maximum signal value on the full-time response curve. These parameters together constitute a richer set of dynamic characteristics than a single static indicator.

[0035] In the above technical solution, firstly, in step S31, the acquired time response curve is segmented. The data processing program automatically identifies the rising phase of the curve: the data segment from the signal start point (when the signal value first exceeds the baseline average plus three times the baseline noise standard deviation) to the peak point (the maximum signal value). The falling phase: the data segment from the peak point to the signal recovering to the baseline value (when the signal value drops to within one time the baseline average plus one noise standard deviation). If the curve does not fully recover within the measurement time, it is truncated to the end of the recording. In step S32, the data of the rising phase is fitted with an exponential growth model. Specifically, a nonlinear least squares method, such as the Levenberg-Marquardt algorithm, is used. The model expression is described in words as follows: the response value at any time t is equal to the initial baseline value plus the adsorption amplitude multiplied by the exponential function with the natural constant as the base, where the exponent of the exponential function is negative at time t divided by the adsorption time constant. The initial baseline value is the average signal value of the 10 seconds before the start of the curve. The fitting algorithm iteratively adjusts the adsorption amplitude and adsorption time constant to minimize the sum of squared residuals between the model's predicted values ​​and the actual measured values. After fitting, the adsorption amplitude and adsorption time constant are extracted. For example, for a serving of overheated fried chicken nuggets, the adsorption amplitude obtained from the fitting during the rising phase might be as high as 180 nanoamps, while the adsorption time constant is 8 seconds, indicating that a large amount of volatiles are generated and adsorbed rapidly. In step S33, the data during the falling phase is fitted using an exponential decay model. The model expression is: the response value at any time t is equal to the response value after recovery to stability plus the desorption amplitude multiplied by an exponential function with the natural constant as the base, where the exponent of the exponential function is negative at time t minus the peak time, divided by the desorption time constant. The response value after recovery to stability is the average signal value of the last 10 seconds of the curve. The fitting also uses the nonlinear least squares method to extract the desorption amplitude and desorption time constant. For example, for the same serving of overheated fried chicken nuggets, the desorption amplitude might be 160 nanoamps, and the desorption time constant might be 25 seconds, indicating that the volatiles desorb slowly. In step S34, the adsorption amplitude, adsorption time constant, desorption amplitude, desorption time constant obtained from the above fitting, along with the peak response value read directly from the curve, are combined to form a five-dimensional dynamic feature parameter vector. These parameters are saved in floating-point format for use in subsequent steps. To ensure the reliability of the fitting, the number of data points in both the rising and falling phases must be no less than a certain number (e.g., 10). If the data quality is poor (e.g., excessive noise causing the fitting to fail to converge), the program will automatically use an alternative method: directly calculating the average slope of the rising phase as a substitute indicator of the adsorption rate, and the average decay rate in the first period of the falling phase (e.g., within 10 seconds) as a substitute indicator of the desorption rate.

[0036] The above-described technical solution, compared to using only coarse-grained features such as peak value or integral area, can more precisely characterize the dynamics of the interaction between the sensor and headspace gas. For samples with similar reheating degrees but different ingredients, the peak response values ​​may be very similar, but the adsorption and desorption time constants may differ significantly. This difference stems from the different compositions of thermal degradation products of proteins and fats in different meats, which in turn affect the binding rate with the sensitive layer. By incorporating adsorption and desorption kinetic parameters into the feature set, deep learning models can utilize these subtle differences to improve classification accuracy. Furthermore, these kinetic parameters also have certain physical significance: the adsorption time constant is related to the size and polarity of volatile molecules, and the desorption time constant is related to the binding strength of active sites on the surface of the sensitive layer, thus providing a basis for subsequent mechanistic analysis.

[0037] In another technical solution, the training process of the deep learning prediction model for the rapid detection of the reheating degree of meat dishes based on electrochemical sensing includes: T1. Obtain multiple samples of meat dishes with different known reheating methods and different reheating times, and collect their electrochemical response signals and extract dynamic characteristic parameters according to steps one to three. At the same time, record the food attribute parameters and reheating method parameters of each sample. T2. The dynamic feature parameters, food attribute parameters and reheating method parameters are combined into an input feature vector, and the corresponding reheating degree level is used as the output label. T3. Construct a prediction model based on a deep neural network or convolutional neural network, using supervised learning to minimize the loss function between the predicted value and the label, and train the deep learning prediction model. Supervised learning is a machine learning paradigm in which each input sample is assigned a known output label (i.e., the true reheat level) during the training phase. The model minimizes the prediction error by learning the mapping relationship between the input and the output. The loss function is a quantitative indicator that measures the difference between the model's predicted value and the true label. Common classification loss functions are cross-entropy loss and regression loss functions are mean squared error. T4. During training, cross-validation is used to optimize model hyperparameters, and an early stopping mechanism is introduced to prevent overfitting. Cross-validation is a model evaluation and hyperparameter optimization method that divides the training dataset into K similarly sized subsets. Each time, the model is trained using K-1 subsets, and its performance is validated on the remaining subset. After K iterations, the average performance is used as an estimate of the generalization ability under that hyperparameter configuration. The early stopping mechanism is a technique to prevent overfitting. During training, the loss value on the validation set is monitored simultaneously. When the validation loss no longer decreases but increases for several consecutive training epochs, training is terminated early, and the model parameters with the minimum validation loss are saved. Hyperparameters are parameters that need to be manually set before model training, such as the number of layers in the neural network, the number of neurons per layer, the learning rate, the batch size, and the regularization coefficient. They are different from the weights and biases that are automatically learned through training.

[0038] In the above technical solution, in step T1, a sample dataset is prepared. A sufficient number of samples of different types of meat dishes are collected, covering common meats (pork, beef, chicken, etc.); each type of meat includes different cooking methods (frying, stir-frying, steaming, braising, roasting); dishes under each cooking method undergo different reheating methods (microwave, water bath, direct heat) and different numbers of reheating cycles (from no reheating to multiple reheatings). For each sample, headspace balancing and sensor response measurements are performed according to steps one to three, dynamic feature parameters are extracted, and food attribute parameters and reheating method parameters are recorded. Finally, a feature matrix and a corresponding label vector are formed. In step T2, the features and labels are combined into a training set. For classification tasks, labels are typically one-hot encoded. The entire dataset is randomly divided into a training set, a validation set, and a test set. In step T3, a prediction model is constructed. Taking a deep neural network as an example, a sequential model is defined using a deep learning framework, where the input layer dimension equals the feature dimension. Add fully connected layers, setting an appropriate number of neurons and ReLU activation function. Optionally, add batch normalization and dropout layers to prevent overfitting. Finally, add an output layer: for classification tasks, the number of neurons in the output layer equals the number of classes, and the activation function is Softmax; for regression tasks, the number of neurons in the output layer is 1, and the activation function is linear. Select Adam as the optimizer and set the initial learning rate. Loss function: use classification cross-entropy for classification tasks and mean squared error for regression tasks. During training, input the training set into the model, iterate one batch at a time, calculate the loss function value, and then backpropagate to update the weights. In step T4, use cross-validation to optimize hyperparameters. Search for hyperparameters such as learning rate and dropout rate within a certain range. For each set of hyperparameters, perform K-fold cross-validation and record the average validation accuracy. Select the hyperparameter combination with the highest average accuracy as the final model's hyperparameters. Then retrain the model using the entire training set with these hyperparameters. An early stopping mechanism is also introduced: set the monitoring metric as validation loss, set a patience value, and stop the model if the validation loss does not decrease after several consecutive rounds, restoring the optimal model parameters. After training, evaluate the model performance on a test set that was never used in the training, calculating accuracy, precision, recall, and F1 score. Save the trained model parameters and a description of the model structure as a file for deployment into the detection system.

[0039] The deep learning prediction model obtained by the above technical solution exhibits good generalization ability and robustness. Cross-validation effectively avoids performance fluctuations caused by the randomness of training set partitioning, ensuring the model remains stable under different combinations of ingredients and reheating conditions. The early stopping mechanism prevents the model from over-memorizing noise in the training set, stopping training at an appropriate time. The final model file is of moderate size, allowing for easy embedding into the microcontroller of a portable detection device or storage in a mobile application for offline inference. This method ensures that the model can provide reliable and consistent prediction results when facing unknown samples in practical applications.

[0040] In another technical solution, the rapid detection method for the reheating degree of meat dishes based on electrochemical sensing, in step six, classifies the reheating degree into four levels: no reheating, mild reheating, moderate reheating, and excessive reheating, or uses a reheating degree score expressed as a continuous numerical value. "No reheating" refers to meat dishes that have not undergone any reheating after their initial cooking, usually in a refrigerated or frozen state, with a low core temperature, retaining the original texture and flavor of freshly cooked meat, and exhibiting the lowest degree of fat oxidation. "Mild reheating" refers to dishes that have undergone short-term or low-temperature reheating, reaching a certain core temperature range, with a warm surface but possibly still cool interior. At this point, some volatile substances begin to be released but not fully, and the amount of fat oxidation products generated is relatively small, so the dish may feel undercooked when eaten. Moderate reheating refers to reheating conditions that bring the core temperature of the dish to a suitable range for consumption. This is the reheating endpoint temperature recommended by most restaurants. At this point, the dish is thoroughly heated, the flavor is fully released, the degree of fat oxidation is moderate, and the taste is close to the initial cooking state. Overheating refers to reheating for too long or at too high a temperature, resulting in excessively high temperatures in the center of the dish, even causing localized charring. This leads to severe fat oxidation producing unpleasant flavors, and excessive protein denaturation causing the meat to become tough and resulting in significant juice loss. In addition to the four discrete levels mentioned above, a "continuous numerical rating of reheating degree" is also allowed, such as a continuous range from 0 to 100 points, where 0 points correspond to no reheating at all, and 100 points correspond to extreme overheating. Users can judge for themselves based on the rating.

[0041] For quality control in catering businesses and quick assessment by home users, discrete ratings are more intuitive. In practice, the output layer of a deep learning prediction model can be set to multiple neurons, using the Softmax activation function to output a probability vector representing the probability of a sample belonging to each rating level. The category with the highest probability is then taken as the final rating. The rating can be displayed as text on the display unit, along with a simple progress bar to represent the probability distribution. For research or quality control scenarios requiring more fine-grained differentiation, continuous ratings can be used. In this case, the output layer of the prediction model is a single neuron, using a linear activation function to output a real number. To limit the output to a certain range, a limiting operation can be added after the output. Continuous ratings can provide more information than discrete ratings. For example, two samples may both be rated as "moderately reheated," but one rating is close to the boundary of mild reheating, while the other is close to the boundary of overheating. The chef can fine-tune the reheating time based on the ratings. When training a continuous rating model, continuous real ratings are needed as labels. These real ratings can be obtained by averaging scores from multiple trained sensory evaluators or by linearly mapping physicochemical indicators. The loss function uses mean squared error. In actual testing, users can select the output mode (discrete or continuous) through the input module of the testing system. If continuous scoring is selected, in addition to displaying the numerical score, the display unit can also present the results in the form of a circular gauge or bar chart, making the results more intuitive.

[0042] The grading method described above caters to the needs of different user groups. For ordinary household users, the four discrete grades are simple, clear, and easy to understand, avoiding the confusion caused by complex numbers. For R&D departments or quality control personnel in food processing enterprises, continuous scoring provides higher resolution, enabling the tracking of subtle changes in reheating degree and optimization of reheating process parameters. The two modes can be switched as needed, increasing the flexibility of the method. Regardless of the mode used, this method can output results quickly, and the detection time is not affected by the grading method.

[0043] In another technical solution, the rapid detection method for the reheating degree of meat dishes based on electrochemical sensing, in step six, the determination of the reheating degree level adopts a graded threshold determination method, which includes: Establish a grading threshold database: A mapping and association database is pre-established through systematic experiments for different combinations of food attribute parameters, cooking method parameters, and reheating method parameters. The database construction process includes: measuring the electrochemical response signal characteristic values ​​corresponding to the concentrations of aldehydes produced at different reheating levels, which are either peak response values ​​or response integral areas; simultaneously, combining the human sensory evaluation results of the corresponding samples, determining the aldehyde concentration thresholds corresponding to each reheating level, including no reheating, mild reheating, moderate reheating, and excessive reheating; finally, storing the data as grading threshold ranges for combinations of food attribute parameters, cooking method parameters, and reheating method parameters; wherein the reheating method includes at least one of microwave reheating, water bath reheating, and direct flame reheating. Threshold matching and correction during detection: Based on the ingredient attribute parameters, cooking method parameters, and reheating method parameters of the meat dish to be detected obtained in step four, the corresponding initial grading threshold T0 is matched and retrieved from the grading threshold library; the initial grading threshold T0 is a set of threshold vectors, including the boundary threshold T from no reheating to slight reheating. 01 The boundary threshold T from mild reheating to moderate reheating 02 And the boundary threshold T from moderate reheating to excessive reheating 03 Each boundary threshold represents the concentration value of the key aldehyde substance that divides adjacent reheating levels under the combined conditions, in micrograms per gram or a value corresponding to the response value of the electrochemical sensor. Adaptive correction: Obtain the baseline correction factor B for the current testing batch, where B = X. blank -X ref ;X blank X represents the characteristic value of the electrochemical response signal of the unreheated sample in the current batch. ref The standard response value is the value of the sample that was not reheated during library construction; the initial grading threshold is adaptively corrected using the formula T. cal,i =T 0i +B, where i=1,2,3; Level determination: Compare the corrected grading threshold T cal,i The numerical relationship between the peak response value or the response integral area X obtained from the current detection: If X ≤ T cal,1 If T cal,1 <X≤T cal,2 If T cal,2 <X≤T cal,3 If X > T, it is determined to be moderate reheating; cal,3 If so, it is determined to be excessive reheating.

[0044] The grading threshold library is a pre-built database stored within the detection system, containing multiple thresholds for different combinations of food properties, cooking methods, and reheating methods. Aldehydes are a class of major volatile compounds produced during the oxidation of meat fats, and their concentration is highly correlated with the degree of reheating. Thiobarbituric acid reactant value is a commonly used physicochemical indicator for evaluating the degree of fat oxidation, calculated by measuring the absorbance of the pink substance generated by the reaction of the sample with thiobarbituric acid reagent at a specific wavelength. Peroxide value indicates the content of hydroperoxides in oils and fats, and is also an early indicator of fat oxidation. The baseline correction factor is used to correct for signal drift caused by batch-to-batch differences in sensors and changes in environmental conditions. The standard response value of the unreheated sample is the characteristic value of the electrochemical response signal of the unreheated reference sample measured under standard sensors and conditions when the threshold library was established.

[0045] The grading threshold determination method of the above technical solution includes three main parts: establishing a grading threshold library, threshold matching and correction during detection, and grade determination. The first part involves establishing the grading threshold library. Combinations of different meats, different fat contents, different cooking methods, and different reheating methods are selected. For each combination, a series of samples with different reheating degrees are prepared. For each sample, three measurements are performed simultaneously: electrochemical sensor response measurement (recording peak response values ​​or response integral areas according to steps one to three), aldehyde concentration determination, and sensory evaluation. Through regression analysis, a mapping relationship between sensor response values ​​and aldehyde concentrations is established. Simultaneously, the thiobarbituric acid reactant value and peroxide value of each sample are measured to establish a correspondence between aldehyde concentrations and these reference indicators. Finally, based on the sensory evaluation results, the thiobarbituric acid reactant value or peroxide value threshold corresponding to the boundary of each reheating degree level is determined. Then, the corresponding aldehyde concentration is calculated in reverse through the mapping relationship and mapped back to the sensor response value to obtain the initial grading threshold T. 01 T 02 T 03 These thresholds are stored in a database, with each record containing the ingredient attributes, cooking method, reheating method, and corresponding threshold. The second part involves threshold matching and correction during testing. The user inputs the ingredient attributes, cooking method, and reheating method of the sample to be tested via the input module. The system searches for matching records in the graded threshold database. If a perfect match is found, the threshold is retrieved directly; otherwise, interpolation is used to calculate the threshold based on attribute similarity weights. Then, the baseline correction factor B for the current batch is obtained. Specific operation: Within the same testing batch, a known unreheated sample of the same type of meat is taken, and its electrochemical response signal characteristic value X is measured following the same steps. blank Calculate B=X blank -X ref , where X refThis refers to the unreheated standard response value stored during database creation. Then, the initial threshold is corrected: T cal1 =T 01 +B,T cal2 =T 02 +B,T cal3 =T 03 +B. Part Three, Grade Determination. Measure the characteristic value X of the electrochemical response signal of the current sample. Compare X with the corrected threshold: if X is less than or equal to T... cal1 If X is greater than T, it is determined that the heat has not been reheated; cal1 And less than or equal to T cal2 If X is greater than T, it is determined to be mild reheat; cal2 And less than or equal to T cal3 If X is greater than T, it is determined to be moderate reheating; cal3 The result was determined to be excessive reheating. The result was output through the display unit.

[0046] The graded threshold determination method described above provides a computationally efficient and highly interpretable approach to determining the degree of reheating. Compared to deep learning models, this method does not require complex matrix operations or large amounts of storage space, and can run in real time on resource-constrained microcontrollers. The introduction of the baseline correction factor B effectively eliminates systematic errors caused by sensor aging, changes in environmental temperature and humidity, and batch-to-batch differences, improving the stability and consistency of the detection results. The establishment of a graded threshold library enables this method to adapt to different ingredients and reheating conditions, avoiding misjudgments caused by a single fixed threshold. This method can serve as a backup or cross-validation tool for deep learning models, further improving the reliability of the detection system.

[0047] In another technical solution, the rapid detection method for the reheating degree of meat dishes based on electrochemical sensing has an initial grading threshold T. 01 T 02 T 03 The methods for determining this include: Establish multiple mapping relationships: First mapping relationship: Establish the correspondence between sensor response values ​​and the content of aldehydes in the headspace; The second mapping relationship is to establish the correspondence between the content of aldehydes and the reference index, which is the reactant value of thiobarbituric acid or the peroxide value. The third mapping relationship: Based on the results of human sensory evaluation, establish the correspondence between the reference index and the degree of reheating; Determine the grading threshold: Based on the transfer of the first to third mapping relationships mentioned above, calculate the sensor response value corresponding to the boundary of each reheating degree level, and use it as the initial grading threshold T. 01 T 02 T 03 ; Consistency verification and storage: The consistency of candidate grading thresholds for the same level boundary calculated through different paths is verified; when the difference exceeds the preset range, the threshold determined by sensory evaluation combined with reference indicators shall be used, and the final determined grading threshold shall be stored in the grading threshold library.

[0048] "Multiple mapping relationships" refer to a chain that links sensor response values ​​to sensory evaluation results through multiple independent intermediate variables. The "first mapping relationship" is the quantitative relationship between the sensor response value and the aldehyde content in the headspace, typically established by preparing aldehyde standard gases of different concentration gradients and measuring the sensor response to create a calibration curve. The "second mapping relationship" is the relationship between aldehyde content and a reference index (thiobarbituric acid reactant value or peroxide value), established through parallel measurements of the same batch of samples. The "third mapping relationship" is the relationship between the reference index and the degree of reheating, determined based on human sensory evaluation results. "Conformity verification" refers to comparing whether the thresholds of the same level boundary calculated through different paths are consistent to verify the reliability of the mapping relationship.

[0049] The above technical solution is as follows: First, establish multiple mapping relationships. The first mapping relationship is established by purchasing aldehyde standards and diluting them with nitrogen to prepare at least several standard gases of different concentrations. The peak response value at each concentration is measured using an electrochemical sensor unit identical to the detection method, with each concentration repeated multiple times. Regression analysis is performed using concentration as the independent variable and response value as the dependent variable to obtain the calibration curve equation. This equation is stored as the first mapping relationship. The second mapping relationship is established by selecting several meat dish samples with different reheating degrees. Three measurements are performed simultaneously on each sample: headspace aldehyde concentration, thiobarbituric acid reactant value, and peroxide value. Correlation analysis is conducted between aldehyde concentration and thiobarbituric acid reactant value to establish a regression equation. Similarly, the relationship between aldehyde concentration and peroxide value is established. The third mapping relationship is established by organizing trained sensory evaluators to score the reheating degree of the above samples. The grade boundaries are determined by analyzing the relationship between sensory scores and thiobarbituric acid reactant values. Then, these boundary values ​​are converted into aldehyde concentration boundaries using the second mapping relationship, and then converted into sensor response value boundaries using the first mapping relationship, thus obtaining the initial classification threshold T. 01 T 02 T 03 To ensure reliability, consistency verification is performed. Alternatively, another approach can be taken: directly establishing a mapping between sensor response values ​​and sensory scores, and calculating another set of thresholds T. 01' T 02' T 03'The two sets of thresholds are compared. If the relative difference is within an acceptable range, the consistency is considered good, and the average of the two sets of thresholds is used as the final threshold. If the difference exceeds a preset range, the threshold determined by sensory evaluation combined with a reference index is used, as the reference index is an internationally recognized standard method for lipid oxidation and is more reliable than direct sensor mapping. Finally, the thresholds that have passed consistency verification are stored in a hierarchical threshold library. Each record also includes the sensor model, environmental conditions, and the unreheated standard response value X at the time of library construction. ref This is for future revisions and use.

[0050] The multi-mapping and consistency verification method employing the above-described technical solution ensures that the initial grading thresholds have a solid physicochemical basis and sensory relevance. Compared to empirical thresholds that solely rely on sensor response values, this method uses dual anchoring of thiobarbituric acid reactant values ​​and sensory evaluations, giving the thresholds clear biological and chemical significance. The consistency verification step effectively identifies conflicts between different mapping paths, ensuring that the final thresholds are more reliable. The threshold library established by this method is traceable and updatable. When sensor batches are replaced or detection conditions change, corrections can be made using the baseline correction factor B without requiring a complete recalibration of the thresholds, thus reducing maintenance costs.

[0051] In another technical solution, the rapid detection method for the reheating degree of meat dishes based on electrochemical sensing, in step two, the electrochemical sensor unit includes at least one gas-sensitive sensing electrode, which is a gas-sensitive sensing electrode with a ternary composite material formed of MXene, ZIF-8, and Co3O4 as the sensitive layer and interdigitated electrodes as the transducer elements; the preparation method of the gas-sensitive electrode includes the following steps: Ti3C2T x Nitrogen-doped porous carbon-confined Co3O4 nanocrystalline composite materials derived from MXene and ZIF-8 were mixed at a mass ratio of 1:1 to 1:3, dispersed in a solvent, and then subjected to ultrasonic treatment to obtain a ternary composite sensitive material. The obtained ternary composite sensitive material is made into a slurry, coated onto the interdigital electrode, and dried to obtain the final product.

[0052] The aforementioned technical solution specifically relates to the material composition and preparation method of the gas-sensitive electrode in an electrochemical sensor unit. "MXene" is a type of two-dimensional transition metal carbide or nitride, specifically referring to Ti3C2T. x T xThis indicates surface terminal groups (such as hydroxyl, oxygen, or fluorine). MXenes possess a layered structure, high electrical conductivity, and a large specific surface area, making them suitable as conductive substrates for gas sensing. "ZIF-8" is a type of zeolite imidazole ester framework material, formed by the coordination of zinc ions and 2-methylimidazolium, exhibiting a regular microporous structure (pore size approximately 1.16 nm) and good thermal stability. "ZIF-8-derived nitrogen-doped porous carbon-confined Co3O4 nanocrystal composite material" refers to a process using ZIF-8 as a precursor. Through pyrolysis, the organic framework of ZIF-8 is carbonized to form nitrogen-doped porous carbon. Simultaneously, cobalt ions loaded in the ZIF-8 channels are converted in situ into cobalt tetroxide (Co3O4) nanocrystals. These nanocrystals are confined within the channels or carbon matrix of the porous carbon, exhibiting small size and uniform distribution. "Ternary composite material" refers to a composite system composed of MXene, the aforementioned ZIF-8 derived carbon-confined Co3O4 nanocrystalline composite material, and Co3O4 itself. MXene provides the conductive network, ZIF-8 derived carbon provides gas adsorption and enrichment, and Co3O4 provides catalytic activity. "Interdigitated electrode" is a transducer with an alternating arrangement of finger-shaped electrodes, typically formed from gold or platinum on a ceramic or silicon substrate via photolithography or screen printing. When a sensitive layer is applied to the interdigitated electrode, the resistance change of the sensitive layer can be measured through the interdigitated electrode. The "sensitive layer" is a thin film of gas-sensitive material coated on the interdigitated electrode, typically with a thickness between a few micrometers and tens of micrometers.

[0053] Following the above preparation method, the specific steps are as follows: First, prepare the required raw materials. MXene (Ti3C2T x The Ti3AlC2MAX phase powder can be prepared using a conventional fluoride etching method: Ti3AlC2MAX phase powder is mixed with lithium fluoride and hydrochloric acid, stirred at 35-40°C for 24 hours to etch away the aluminum layer, then washed by centrifugation to neutrality, and ultrasonically exfoliated to obtain a few-layer MXene dispersion, which is then freeze-dried for later use. The nitrogen-doped porous carbon-confined Co3O4 nanocrystalline composite material derived from ZIF-8 can be obtained as follows: First, ZIF-8 crystals are synthesized. ZIF-8 is then immersed in an alcohol solution containing cobalt ions, allowing the cobalt ions to diffuse into the ZIF-8 channels, followed by filtration and drying. The cobalt-loaded ZIF-8 is placed in a tube furnace and heated to approximately 400°C under an inert atmosphere (such as argon) and held for 1-2 hours to carbonize the ZIF-8 and simultaneously convert the cobalt ions into Co3O4 nanocrystals, yielding a black powder. The Co3O4 nanocrystals in this composite material are typically 2-5 nanometers in size and uniformly distributed within the nitrogen-doped porous carbon matrix. The above-prepared Ti3C2MAX phase powder can be further processed using this method. xMXene and ZIF-8-derived nitrogen-doped porous carbon-confined Co3O4 nanocrystalline composites were weighed in a mass ratio of 1:1 to 1:3. For example, 20 mg of MXene and 40 mg of the composite (mass ratio 1:2), or 20 mg of MXene and 20 mg of the composite (mass ratio 1:1), or 20 mg of MXene and 60 mg of the composite (mass ratio 1:3) could be used. The weighed powders were placed in an agate mortar, and an appropriate amount of solvent was added as a dispersion medium. The solvent could be deionized water, anhydrous ethanol, or N,N-dimethylformamide. The amount of solvent should be sufficient to form a flowable slurry, typically 0.5-1 mL of solvent per 10 mg of solids. The mixture was ground in the mortar for 5-10 minutes, then transferred to a glass bottle and ultrasonically treated for 30-60 minutes in an ultrasonic cleaner at a power of 100-150 watts and a frequency of 40 kHz. During the ultrasonic process, the layered structure of MXene is further exfoliated, while the porous carbon and nanocrystals in the composite material are uniformly dispersed between the MXene layers, forming a uniform slurry of the ternary composite sensitive material. After ultrasonic treatment, the slurry is directly used for coating. The interdigitated electrodes need to be cleaned before use: the alumina ceramic substrate with the interdigitated gold electrodes is sequentially immersed in acetone, ethanol, and deionized water, ultrasonically cleaned for 10 minutes in each, and then dried with nitrogen. The above slurry is then coated onto the sensitive area of ​​the interdigitated electrodes using either a drop-coating or spin-coating method. The drop-coating method is as follows: 20-50 microliters of slurry are drawn up using a micropipette and evenly dropped onto the surface of the interdigitated electrodes, allowing it to spread naturally and cover the entire interdigitated area. The coated electrodes are placed in a vacuum drying oven and dried at 60 degrees Celsius for 4-6 hours to allow the solvent to completely evaporate. If it is necessary to increase the thickness of the sensitive layer, the drop-coating and drying steps can be repeated 1-2 times. After drying, the sensitive layer is firmly attached to the interdigitated electrodes, forming a gas-sensitive sensing electrode. In use, the electrode is connected to the measurement circuit via leads. A constant DC voltage (typically 0.5-1 volt) is applied across the interdigitated electrodes to measure the current or resistance flowing through them. When the electrode is exposed to headspace gas containing aldehydes, the resistance of the sensing layer changes, and the change in resistance reflects the gas concentration.

[0054] The gas-sensitive electrode prepared using this technical solution has a sensitive layer composed of a ternary composite of MXene, ZIF-8-derived nitrogen-doped porous carbon, and Co3O4 nanocrystals. In this structure, MXene provides a continuous conductive channel, ensuring rapid transmission of electrical signals; the ZIF-8-derived porous carbon has a large specific surface area and abundant microporous structure, effectively adsorbing and enriching aldehyde molecules in the headspace; and the Co3O4 nanocrystals, as catalytic active centers, catalyze the oxidation reaction of aldehyde molecules at relatively low temperatures, releasing electrons and causing a change in the resistance of the sensitive layer. The three materials can be mixed and ultrasonically treated to form a homogeneous composite system without complex covalent modification or high-temperature annealing steps, making the preparation process simple and easy to scale up. This electrode exhibits high response sensitivity to aldehydes (such as hexanal, heptanal, and nonanal), while showing relatively low response to common interfering substances such as water vapor and alcohols due to the hydrophobic properties and molecular sieving effect of ZIF-8-derived carbon, demonstrating a certain degree of selectivity. This electrode can operate within a temperature range of room temperature to 100 degrees Celsius, meeting the requirements for headspace gas detection during the reheating of meat dishes. Integrating this electrode into a portable detection device enables rapid, on-site detection of aldehyde markers in reheated meat dishes.

[0055] In another technical solution, the rapid detection method for the reheating degree of meat dishes based on electrochemical sensing, in step five, before inputting the dynamic feature parameters and ingredient attribute parameters into the deep learning prediction model, also performs an interference mode decoupling step: W1. Pre-establish an interference characteristic spectral library: Expose the electrochemical sensor unit to at least one interference standard gas, which includes at least one of water vapor, ethanol vapor, and acetic acid vapor. Extract the dynamic characteristic parameters of the interferences using the same method as in steps two and three to construct the interference characteristic matrix M. int M int Each column corresponds to a dynamic feature parameter vector of a type of interference. W2, regarding the interference feature matrix M int Perform singular value decomposition to obtain the orthogonal basis vectors U of the disturbance subspace. int U int Each column represents the main direction of change in the feature space of the interfering object; W3, convert the currently detected original feature vector F cur Projected to U int On the orthogonal complement space, we obtain the projection vector after eliminating the influence of interference: ; Among them, F curThe feature vector is composed of the dynamic feature parameters extracted in step three and the ingredient attribute parameters, cooking method parameters, and reheating method parameters obtained in step four. The cooking method parameters include at least one of heating temperature, heating time, temperature change pattern, and endpoint temperature; the reheating method parameters include at least one of reheating start temperature, reheating power, reheating time, and reheating endpoint temperature. int U represents the orthogonal basis vectors of the interference subspace; int T F cur U represents the projection coefficients of the original feature vector onto the interference subspace. int (U int T F cur F represents the components of the original feature vector that belong to the interference subspace; proj The projection vector after eliminating interference components is the projection of the original eigenvector onto the orthogonal complement space of the interference subspace. W4, with F proj As input to the deep learning prediction model, the feature vector is used.

[0056] The "Interference Feature Spectrum Library" is a pre-built matrix database that records the dynamic feature parameter vectors of various common interfering gases (such as water vapor, ethanol vapor, and acetic acid vapor) for electrochemical sensor units. "Singular Value Decomposition" (SVD) is a matrix factorization technique that decomposes a matrix into a product of three matrices. The middle diagonal matrix contains singular values, and the two orthogonal matrices on the left and right contain left and right singular vectors, respectively. In this method, after performing SVD on the interference feature matrix, the resulting left singular vectors constitute a set of orthogonal bases for the interference subspace. The "interference subspace" refers to the low-dimensional linear space spanned by the feature vectors of the interfering substances; the signal features generated by any interfering substance can be represented as a linear combination of vectors in this subspace. The "orthogonal complement space" is the space perpendicular to the interference subspace, i.e., the set of vectors orthogonal to all vectors in the interference subspace. Projecting the original feature vectors onto the orthogonal complement space eliminates components belonging to the interference subspace, thus preserving information unrelated to the interference. The "projection coefficients" refer to the magnitudes of the components of the original feature vectors in each orthogonal base direction of the interference subspace.

[0057] The interference mode decoupling steps in the above technical solution are performed before the feature input model, specifically divided into W1 to W4. In step W1, an interference feature spectral library is pre-established. Under standard conditions, the electrochemical sensor unit is exposed to standard gases of different concentrations of interference, such as water vapor, ethanol vapor, and acetic acid vapor. For each concentration of each interference, dynamic feature parameters are extracted according to steps two and three, including rise time, peak response value, time to peak value, integral area, half-life, adsorption amplitude, adsorption time constant, desorption amplitude, and desorption time constant. All these feature vectors are arranged column-wise to form an interference feature matrix M. int In step W2, for M int Perform singular value decomposition. Calculate M using a numerical computation library. int =U int *Σ*V T U int Σ is an orthogonal matrix whose column vectors are left singular vectors; Σ is a diagonal matrix whose diagonal elements are singular values; V is an orthogonal matrix. The left singular vectors corresponding to the k largest singular values ​​are selected to form the orthogonal basis vector matrix U of the disturbance subspace. int The selection criterion for k is that the sum of squares of the first k singular values ​​accounts for more than a certain threshold, such as 95%, of the sum of squares of all singular values. These k orthogonal basis vectors represent the main directions of change in the features of the interference. In step W3, the original feature vector F obtained from the current detection is... cur Process it. F cur This vector is composed of the dynamic feature parameters extracted in step three and the ingredient attribute parameters, cooking method parameters, and reheating method parameters obtained in step four. Interference decoupling only targets the dynamic feature components directly related to the sensor response; other features remain unchanged. The projection vector F is calculated. proj The specific operation is as follows: First, calculate the projection coefficients of the original dynamic feature vector onto the interference subspace, i.e., U. int The transpose of the vector is multiplied by the dynamic eigenvector. Then, the components belonging to the disturbance subspace, i.e., U, are reconstructed. int Multiply by this projection coefficient. Finally, subtract the interference component from the original dynamic feature vector to obtain the projection vector after eliminating the interference effect. For non-dynamic feature parts (food properties, reheating method parameters), they are not affected by gas interference and therefore remain unchanged. The final F proj It is formed by concatenating the dynamic feature vector after eliminating interference and the remaining non-dynamic feature vector. In the W4 step, F proj The feature vectors are used as input to the deep learning prediction model. To ensure the stability of the singular value decomposition, the interference feature spectral library needs to be updated regularly, for example, by re-measuring the interference response at regular intervals or rebuilding it when the sensor is replaced.

[0058] The interference mode decoupling step employing the above technical solution effectively eliminates the influence of common interfering substances such as water vapor, ethanol, and acetic acid on the sensor response. During reheating, meat dishes release a large amount of water vapor, which itself causes changes in the sensor's resistance, but this change is independent of the degree of reheating. After processing with this method, humidity-induced response fluctuations are effectively suppressed. The orthogonal basis of the interference subspace determined by singular value decomposition has clear physical meaning, corresponding to the characteristic modes of different interfering substances. By eliminating components in these directions, the true signals related to lipid oxidation products are retained. This method does not require an additional humidity sensor or temperature compensation circuit; interference suppression can be achieved solely through software algorithms, reducing hardware costs. Furthermore, the computational complexity of the projection operation is low, allowing for real-time execution on a microcontroller without affecting detection speed.

[0059] The present invention also provides a detection system for implementing the above-described rapid detection method for the reheating degree of meat dishes based on electrochemical sensing, comprising: A portable detection device is provided with an electrochemical sensor unit, which includes at least one gas-sensitive electrode for collecting headspace gas from the reheated meat dish to be tested and generating an electrochemical response signal. The signal acquisition and processing module is used to record the complete time response curve of the electrochemical response signal and extract dynamic characteristic parameters from it. The input module is used to input the ingredient attribute parameters, cooking method parameters, and reheating method parameters of the reheated meat dish to be tested. The ingredient attribute parameters include at least one of the following: meat type, fat content, and water content. The cooking method parameters include at least one of the following: heating temperature, heating time, temperature change pattern, and endpoint temperature. The reheating method parameters include at least one of the following: reheating start temperature, reheating power, reheating time, and reheating endpoint temperature. A storage module that stores pre-trained deep learning prediction models; The prediction module is used to input dynamic feature parameters, food attribute parameters, and reheating method parameters into the deep learning prediction model to obtain the reheating degree level. The display module is used to output and display the degree of reheating.

[0060] The above technical solution includes a rapid detection system for the reheating degree of meat dishes based on electrochemical sensing. This system comprises multiple functional modules. The "portable detection device" refers to a small, lightweight device that can be held in the hand or attached to the surface of packaging, integrating an electrochemical sensor unit. The "signal acquisition and processing module" typically consists of a microcontroller (such as an STM32 series chip) and an analog-to-digital converter, responsible for controlling the sensor's excitation signal, acquiring response data, performing signal conditioning, and extracting features. The "input module" can be a touchscreen, buttons, or an interface for communication with a mobile terminal via Bluetooth, used by the user to input food attribute parameters and reheating method parameters. The "storage module" is a non-volatile memory (such as flash memory or an SD card) used to store the weights and biases of the trained deep learning prediction model, as well as a grading threshold library. The "prediction module" is the inference engine running in the microcontroller, loading the model from the storage module, performing forward propagation calculations on the feature vectors, and outputting the reheating degree level. The "display module" can be an LCD screen, an OLED screen, or an indicator light array used to present the results. In addition, the system may include a power management module (battery or near-field communication power supply circuit) and a communication module (Bluetooth, Wi-Fi, or near-field communication). It should be noted that the electrochemical sensor unit in this system must be able to contact the headspace gas of the meat dish being tested. Therefore, when installing the portable detection device, it is necessary to ensure that the gas-sensitive area of ​​the sensor is located inside the packaging or connected to the headspace through a breathable structure.

[0061] The specific implementation of the detection system described above is as follows: The portable detection device's casing is made of food-grade plastic material, measuring approximately 60 mm × 40 mm × 15 mm and weighing approximately 30 grams. An opening is located at the bottom of the casing, covered by a breathable and waterproof membrane (e.g., a polytetrafluoroethylene membrane with a pore size of 0.2-0.5 micrometers). This membrane allows gas molecules to pass through but prevents liquid water and solid particles from entering. The electrochemical sensor unit is mounted directly above the opening. Internally, a printed circuit board is installed, integrating the following components: an electrochemical sensor unit (gas-sensitive electrode, employing interdigitated electrodes and a ternary composite sensitive layer); an analog front-end chip (e.g., LMP91000) to provide bias voltage to the sensor and amplify the current signal; a microcontroller (e.g., STM32F103C8T6) integrating a 12-bit analog-to-digital converter and a digital signal processing unit; a Bluetooth module (e.g., HC-05); an organic light-emitting diode display (0.96-inch size, 128×64 resolution); three buttons and three LEDs (red, yellow, and green); a micro USB interface for charging and data transfer; and a battery management chip and a lithium polymer battery (200 mAh capacity). The signal acquisition and processing module operates as follows: After the microcontroller starts, it performs a system self-test to check the stability of the sensor baseline. Then, a 1-volt DC voltage is applied to the interdigitated electrodes via the analog front-end chip. Initially, the sensor is stabilized in clean air for 60 seconds, and the baseline current value is recorded. When the user triggers the detection via a button or mobile application, the microcontroller begins sampling the sensor's current value at a rate of 10 times per second for 600 seconds. After sampling, the microcontroller runs an embedded signal processing algorithm: first, it applies a moving average filter (5-point window width) to the raw data; then, it automatically identifies the response start point (the first time the signal exceeds the baseline average plus three times the noise standard deviation), the peak point (the maximum signal value), and the recovery point (the signal drops to within 10% of the difference between the peak and the baseline). Based on this, it calculates the rise time, peak response value, time to reach the peak, integral area, and half-life. Simultaneously, it performs exponential fitting on the rise and fall phases to calculate the adsorption amplitude, adsorption time constant, desorption amplitude, and desorption time constant. These calculations are performed on the microcontroller using fixed-point arithmetic. Input module: The user selects menu items and inputs parameters using three buttons: button 1 switches input items (meat type, fat content, water content, cooking method, reheating method, etc.), button 2 increases the value or switches options, and button 3 confirms. To simplify input, the system can also preset default parameters for common dishes (e.g., "braised pork" corresponds to pork, 25% fat, 55% water content, and stewing method). Users can also choose to connect their phones via Bluetooth for more convenient parameter input in a mobile application. Storage module: The device's built-in flash memory chip (16 megabytes capacity) stores pre-trained deep learning models.The model is stored in 8-bit integer quantization to save storage space and accelerate inference. The model structure is also encoded as metadata. Prediction module: The microcontroller loads the model, combining the extracted dynamic feature parameters (5 in total) with user-input food attribute parameters (such as meat type code, fat content, water content, cooking method code, etc., approximately 6-8 in total) and reheating method parameters (reheating start temperature, power, time, etc., approximately 3-5 in total) into a feature vector, with a total dimension of approximately 15. Then, neural network inference is performed: matrix multiplication and activation function operations are performed layer by layer. Inference time is less than 100 milliseconds. Display module: The prediction result (e.g., "moderate reheating") is displayed on an OLED screen, while a green LED illuminates. If the result is overheating, a red LED flashes and a beep sounds. Users can also choose to send the result to their mobile phone via Bluetooth for storage. In use, the portable detection device needs to be fixed to the meat product packaging, ensuring that the opening of the device's breathable waterproof membrane faces inwards, i.e., the sensor's sensitive layer can contact the headspace inside the packaging. Specific mounting methods include: using medical-grade double-sided adhesive on the back of the device to attach it to the transparent film of the packaging, but a small hole (approximately 5 mm in diameter) needs to be pre-drilled in the packaging film and aligned with the hole on the device; or using dedicated testing packaging with a built-in ventilation window. For unpackaged dishes, the device can be placed next to the container using a stand, and a silicone tube can be used to draw the gas from the headspace of the container to the air inlet of the device's sensor.

[0062] The detection system employing this technical solution achieves hardware integration and portability of the entire detection method. The system weighs approximately 30 grams, making it easy to hold or temporarily fix in place. Due to its low-power design (the microcontroller consumes less than 1 mA in standby mode and approximately 30 mA during measurement), a 200 mA battery can support approximately 50 complete detections (each detection includes preheating, sampling, calculation, and display). The system is easy to operate: users simply fix the device to the packaging (ensuring the sensor is connected to the headspace), select the dish type using the buttons, press the "Start" button, and the results can be read after approximately 10-15 minutes (including headspace equilibration time). The system's detection accuracy has been validated with a large number of samples, achieving a classification consistency rate of over 90% compared to laboratory gas chromatography-mass spectrometry (GC-MS) analysis results. Because the system integrates an interference decoupling module, the fluctuation of detection results is minimal under different environmental humidity conditions. Furthermore, the system supports firmware and model updates via a USB interface; users can download update packages and upload them to the device via computer to upgrade the model and expand the dish database. The system's cost is controlled within 200 yuan (excluding the outer shell mold), making it suitable for mass production and market promotion.

[0063] In another technical solution, the detection system, the portable detection device is configured to be attached to the inside or outside of the meat dish packaging, or directly to the surface of the meat dish, and is powered by a mobile terminal and triggered for detection via near-field communication.

[0064] The portable detection device is designed as a passive near-field communication patch that is entirely placed inside food packaging. Its structure includes: a flexible printed circuit board, approximately 0.2 mm thick and 30 mm x 30 mm in area. The circuit board integrates a near-field communication antenna coil (typically a 3- to 5-turn copper coil with an outer diameter of approximately 25 mm), a near-field communication chip (e.g., NT3H2211), which integrates energy harvesting circuitry, rectification and regulation circuitry, non-volatile memory, and a simple state machine. Additionally, the circuit board houses an electrochemical sensor unit (such as a gas-sensitive electrode, including interdigitated electrodes and a ternary composite sensitive layer), analog front-end circuitry (which can be built from discrete low-power operational amplifiers), and a miniature organic light-emitting diode display (0.96 inches in size, 1 mm thick) or simplified to three LEDs. The entire device is encapsulated in a flexible, food-grade encapsulation material (such as polyethylene terephthalate film), leaving only a vent for the sensor, which is covered with a waterproof and breathable membrane. Food-grade double-sided adhesive is applied to the back of the device. In use, for pre-packaged meat dishes (such as frozen ready-to-eat chicken breast, vacuum-packed braised pork, etc.), during the food packaging production process, the patch is adhered to the inside of the transparent film of the packaging using double-sided adhesive, ensuring that the sensor's vent faces the headspace area inside the packaging. The patch and packaging film are well-sealed, preventing leakage of contents. After the user purchases the food, the patch remains inside the packaging. When the user reheats the dish and wants to test the degree of reheating, they bring a near-field communication (NFC) enabled smartphone close to the corresponding patch on the outside of the packaging (at a distance of less than 2 cm). The phone transmits a 13.56 MHz carrier wave via NFC. The NFC antenna coil on the patch senses the electromagnetic field, and after rectification and voltage regulation, generates a 3.3V DC voltage to power the entire patch. After power-on, the NFC chip inside the patch establishes a communication link with the phone. A dedicated application runs on the phone, and the user inputs the ingredient attributes (if there is a QR code or barcode on the packaging, it can be scanned and automatically imported), cooking method, and reheating method parameters through the application interface. Then, the user clicks "Start Detection". The mobile phone sends commands to the patch via near-field communication (NFC). The state machine within the patch initiates the detection process: First, it waits for a preset headspace balancing time (e.g., 5 minutes, timed by the application; upon completion, the mobile phone wakes up the patch again via NFC). After the balancing time ends, the analog front-end inside the patch begins acquiring sensor response signals, continuously acquiring them for 300 seconds (or ending early once the signal stabilizes), and temporarily storing the data in the buffer of the NFC chip (approximately 4 kilobytes). After acquisition, the patch sends the raw time response curve data to the mobile phone in batches via NFC. Upon receiving the data, the mobile application performs signal processing, feature extraction, interference decoupling, and deep learning inference (the model is stored on the mobile phone) to obtain the reheating level, which is then displayed on the mobile phone screen.Meanwhile, the miniature organic light-emitting diodes on the patch can also display a simplified result (e.g., using different colors or numbers to represent grades), but since the patch is inside the packaging, users may need to observe it through the transparent film. After detection, the phone disconnects from near-field communication, the patch loses power, and enters a passive sleep state. For direct application to meat dishes (e.g., unpackaged dishes for restaurant dining or home cooking), the patch can be attached to the inside of the plate (close to the dish but not soaked in broth) using food-grade double-sided tape, or to the inside of the pot lid, ensuring the sensor vent faces the air above the dish. The detection process is the same as described above.

[0065] The above-described technical solution enables a battery-free, ultra-thin, and disposable detection device, making it suitable for use as part of food packaging. Since no built-in battery is required, the device thickness can be controlled to within 1 mm, and the material cost of a single patch is approximately 3-5 RMB, making it suitable for single-use. Near-field communication power supply eliminates the hassle of battery replacement for users and avoids environmental pollution from battery disposal. The mobile terminal serves as a computing and display platform, utilizing the phone's powerful processor to run complex deep learning models, while the patch only handles simple data acquisition and communication, reducing the hardware requirements and cost of the patch. Due to the high penetration rate of mobile phones, users do not need to purchase a dedicated reader, lowering the barrier to entry. The patch is completely placed inside the packaging, with the sensor directly contacting the headspace, eliminating the need for additional openings in the packaging or concerns about sealing, resulting in more accurate and reliable detection results. In practical applications, consumers can purchase pre-cooked meals with detection patches from supermarkets, reheat them at home, and quickly determine whether reheating has been successful by simply bringing their phone close to the packaging, without opening the packaging and avoiding secondary contamination. This configuration is particularly suitable for quality monitoring in smart packaging and cold chain logistics. For catering businesses, the patch can be attached to the inside of plates or insulated lids to quickly check the reheating quality before serving, ensuring food quality. This technology enables a convenient "stick-and-use, mobile phone-readable" detection mode and has promising market application prospects.

[0066] Experimental Example: Validation of the performance of the method of the present invention in detecting the reheating degree of meat dishes 1. Experimental Objective The accuracy of the method of the present invention in detecting the reheating degree of meat dishes was verified and compared with existing detection methods.

[0067] 2. Test Samples and Procedures 2.1 Sample Preparation Three representative meat dishes were selected: braised pork belly (high-fat, red meat), cumin-fried beef (medium-fat, red meat), and steamed chicken breast (low-fat, white meat). Three reheating methods were used: microwave, water bath, and direct heat. Each method was graded into four levels: no reheating, slight reheating, moderate reheating, and overheating (determined based on a combination of peroxide value and sensory evaluation). A total of 36 groups were selected, with 3 replicates per group, for a total of 108 samples.

[0068] 2.2 Detection Method The method of this invention is as follows: extract the dynamic feature parameters (rise time, peak value, integral area, adsorption / desorption time constant, etc.) of the complete time response curve of the sensor as described above, fuse the food attribute parameters (fat content, water content, cooking method) and reheating method parameters (power, time, final temperature), and input them into a three-layer deep neural network for classification.

[0069] Comparison Method A (Static Feature Method): Only uses peak response value + rise time, SVM classifier, without incorporating prior information.

[0070] Comparison Method B (Multi-feature SVM method): Uses all dynamic features, does not fuse prior information, and uses an SVM classifier.

[0071] Sensory evaluation: Double-blind evaluation by 10 evaluators.

[0072] Physicochemical analysis: Peroxide value was determined according to GB 5009.227 (as a reference).

[0073] 3. Test Results 3.1 Sensor Performance The sensor has a detection limit of 0.09~0.35 ppm for reheat-characteristic aldehydes (hexanal, pentanal, etc.), a response time of 32~58 seconds, and a recovery time of 45~72 seconds; after 30 days of continuous testing, the response value decayed by 7.7% and the baseline drift was 3.0%, showing good stability.

[0074] 3.2 Classification accuracy comparison, as shown in Table 1. Table 1. Comparison of Classification Accuracy The accuracy of the method of this invention is 21.8 percentage points higher than that of the static feature method and 9.3 percentage points higher than that of the multi-feature SVM method. It is significantly better than sensory evaluation, close to the physicochemical analysis benchmark, and is a non-destructive testing method.

[0075] 3.3 Performance under different reheating methods, the results are shown in Table 2: Table 2 Performance under different reheating methods The accuracy of water bath reheating is low, mainly due to interference from a large amount of water vapor, which can be further improved through interference decoupling steps.

[0076] 3.4 The results of different ingredients are shown in Table 3. Table 3 Performance under different ingredients Low-fat samples release less aldehydes and have weaker signals, which can be enhanced by extending the equilibration time or by interfering with decoupling.

[0077] 4. Conclusion The method of this invention achieved an overall accuracy of 90.6% on 108 test samples, significantly outperforming sensory evaluation (67.5%) and traditional static characteristic methods (68.8%), and approaching the physicochemical analysis benchmark (93.8%). Furthermore, it achieves rapid (approximately 15 minutes), non-destructive, and low-cost detection (device cost approximately 50-100 RMB). This method can adapt to variations in ingredients, cooking methods, and reheating techniques, demonstrating good practicality and potential for wider application.

[0078] The number of devices and processing scale described herein are for the purpose of simplifying the description of the invention. Applications, modifications, and variations of the invention will be readily apparent to those skilled in the art.

[0079] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. They can be applied to various fields suitable for the present invention. For those skilled in the art, other modifications can be easily made. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and illustrations shown and described herein.

Claims

1. A rapid detection method for the reheating degree of meat dishes based on electrochemical sensing, characterized in that, Includes the following steps: Step 1: Place the meat dish to be tested in a closed or semi-closed space and maintain it at a preset equilibrium temperature for a preset equilibrium time to generate headspace gas. Step 2: Use the electrochemical sensor unit to collect the electrochemical response signal generated by the interaction between the headspace gas and the sensor sensitive layer, and record the complete time response curve of the electrochemical response signal; Step 3: Extract dynamic characteristic parameters from the time response curve, which include at least one or more of the following: response rise time, peak response value, time to reach peak value, area under the integral of the response curve, and half-life of the fall phase. Step 4: Obtain the ingredient attribute parameters and reheating method parameters of the meat dish to be tested. The ingredient attribute parameters include at least one of the following: meat type, fat content, and water content. The meat type is quantified by the ratio of muscle fiber types or red meat / white meat classification. Step 5: Input the dynamic feature parameters extracted in Step 3 and the food attribute parameters and reheating method parameters obtained in Step 4 into a pre-trained deep learning prediction model. The deep learning prediction model is a model built on a deep neural network or convolutional neural network, which is used to map the reheating level according to the input features. Step 6: The deep learning prediction model outputs the degree of reheating of the reheated meat dish to be detected and presents it on the display unit; In step five, before inputting the dynamic feature parameters and food attribute parameters into the deep learning prediction model, an interference mode decoupling step is also performed: W1. Pre-establish an interference characteristic spectral library: Expose the electrochemical sensor unit to at least one interference standard gas, which includes at least one of water vapor, ethanol vapor, and acetic acid vapor. Extract the dynamic characteristic parameters of the interferences using the same method as in steps two and three to construct the interference characteristic matrix M. int M int Each column corresponds to a dynamic feature parameter vector of a type of interference. W2, regarding the interference feature matrix M int Perform singular value decomposition to obtain the orthogonal basis vectors U of the disturbance subspace. int U int Each column represents the main direction of change in the feature space of the interfering object; W3, convert the currently detected original feature vector F cur Projected to U int On the orthogonal complement space, we obtain the projection vector after eliminating the influence of interference: ; Among them, F cur The feature vector is composed of the dynamic feature parameters extracted in step three and the ingredient attribute parameters, cooking method parameters, and reheating method parameters obtained in step four. The cooking method parameters include at least one of heating temperature, heating time, temperature change pattern, and endpoint temperature; the reheating method parameters include at least one of reheating start temperature, reheating power, reheating time, and reheating endpoint temperature. int U represents the orthogonal basis vectors of the interference subspace; int T F cur U represents the projection coefficients of the original feature vector onto the interference subspace. int (U int T F cur F represents the components of the original feature vector that belong to the interference subspace; proj The projection vector after eliminating interference components is the projection of the original eigenvector onto the orthogonal complement space of the interference subspace. W4, with F proj As input to the deep learning prediction model, the feature vector is used.

2. The rapid detection method for the reheating degree of meat dishes based on electrochemical sensing as described in claim 1, characterized in that, The specific methods for extracting dynamic feature parameters in step three include: S31. Perform model fitting for the rising and falling phases of the time response curve respectively; S32. Fit the rising phase curve to an exponential growth model: ; Where R(t) is the response value at time t, R0 is the initial baseline value, A is the adsorption amplitude, and τ is the adsorption time constant; A and τ are extracted as feature parameters. S33. Fit the descent phase curve to an exponential decay model: ; Among them, R ∞ To recover the stable response value, B is the desorption amplitude and τ3 is the desorption time constant; B and τ3 are extracted as characteristic parameters. S34. Compare the fitted feature parameters A, τ, B, τ3 with the peak response value R. max Together they constitute the dynamic characteristic parameters.

3. The rapid detection method for the reheating degree of meat dishes based on electrochemical sensing as described in claim 1, characterized in that, The training process of a deep learning prediction model includes: T1. Obtain multiple samples of meat dishes with different known reheating methods and different reheating times, and collect their electrochemical response signals and extract dynamic characteristic parameters according to steps one to three. At the same time, record the food attribute parameters and reheating method parameters of each sample. T2. The dynamic feature parameters, food attribute parameters and reheating method parameters are combined into an input feature vector, and the corresponding reheating degree level is used as the output label. T3. Construct a prediction model based on a deep neural network or a convolutional neural network, and use supervised learning to train the deep learning prediction model with the goal of minimizing the loss function between the predicted value and the label. T4. During training, cross-validation is used to optimize the model hyperparameters, and an early stopping mechanism is introduced to prevent overfitting.

4. The rapid detection method for the reheating degree of meat dishes based on electrochemical sensing as described in claim 1, characterized in that, In step six, the degree of reheating is divided into four levels: no reheating, mild reheating, moderate reheating, and excessive reheating, or a reheating degree score expressed as a continuous numerical value is used.

5. The rapid detection method for the reheating degree of meat dishes based on electrochemical sensing as described in claim 4, characterized in that, In step six, the degree of reheating is determined using a grading threshold method, which includes: Establish a grading threshold database: A mapping and association database is pre-established through systematic experiments for different combinations of food attribute parameters, cooking method parameters, and reheating method parameters. The database construction process includes: measuring the electrochemical response signal characteristic values ​​corresponding to the concentrations of aldehydes produced at different reheating levels, which are either peak response values ​​or response integral areas; simultaneously, combining the human sensory evaluation results of the corresponding samples, determining the aldehyde concentration thresholds corresponding to each reheating level, including no reheating, mild reheating, moderate reheating, and excessive reheating; finally, storing the data as grading threshold ranges for combinations of food attribute parameters, cooking method parameters, and reheating method parameters; wherein the reheating method includes at least one of microwave reheating, water bath reheating, and direct flame reheating. Threshold matching and correction during detection: Based on the ingredient attribute parameters, cooking method parameters, and reheating method parameters of the meat dish to be detected obtained in step four, the corresponding initial grading threshold T0 is matched and retrieved from the grading threshold library; the initial grading threshold T0 is a set of threshold vectors, including the boundary threshold T from no reheating to slight reheating. 01 The boundary threshold T from mild reheating to moderate reheating 02 And the boundary threshold T from moderate reheating to excessive reheating 03 Each boundary threshold represents the concentration value of the key aldehyde substance that divides adjacent reheating levels under the combined conditions, in micrograms per gram or a value corresponding to the response value of the electrochemical sensor. Adaptive correction: Obtain the baseline correction factor B for the current testing batch, where B = X. blank -X ref ;X blank X represents the characteristic value of the electrochemical response signal of the unreheated sample in the current batch. ref The standard response value is the value of the sample that was not reheated during library construction; the initial grading threshold is adaptively corrected using the formula T. cal,i =T 0i +B, where i=1,2,3; Level determination: Compare the corrected grading threshold T cal,i The numerical relationship between the peak response value or the response integral area X obtained from the current detection: If X ≤ T cal,1 If T cal,1 <X≤T cal,2 If T cal,2 <X≤T cal,3 If X > T, it is determined to be moderate reheating; cal,3 If so, it is determined to be excessive reheating.

6. The rapid detection method for reheating degree based on electrochemical sensing as described in claim 5, characterized in that, Initial grading threshold T 01 T 02 T 03 The methods for determining this include: Establish multiple mapping relationships: First mapping relationship: Establish the correspondence between sensor response values ​​and the content of aldehydes in the headspace; The second mapping relationship is to establish the correspondence between the content of aldehydes and the reference index, which is the reactant value of thiobarbituric acid or the peroxide value. The third mapping relationship: Based on the results of human sensory evaluation, establish the correspondence between the reference index and the degree of reheating; Determine the grading threshold: Based on the transfer of the first to third mapping relationships mentioned above, calculate the sensor response value corresponding to the boundary of each reheating degree level, and use it as the initial grading threshold T. 01 T 02 T 03 ; Consistency verification and storage: The consistency of candidate grading thresholds for the same level boundary calculated through different paths is verified; when the difference exceeds the preset range, the threshold determined by sensory evaluation combined with reference indicators shall be used, and the final determined grading threshold shall be stored in the grading threshold library.

7. The rapid detection method for the reheating degree of meat dishes based on electrochemical sensing as described in claim 1, characterized in that, In step two, the electrochemical sensor unit includes at least one gas-sensitive electrode, which is a gas-sensitive electrode with a ternary composite material formed of MXene, ZIF-8, and Co3O4 as the sensitive layer and an interdigitated electrode as the transducer element; the preparation method of the gas-sensitive electrode includes the following steps: Ti3C2T x Nitrogen-doped porous carbon-confined Co3O4 nanocrystalline composite materials derived from MXene and ZIF-8 were mixed at a mass ratio of 1:1 to 1:3, dispersed in a solvent, and then subjected to ultrasonic treatment to obtain a ternary composite sensitive material. The obtained ternary composite sensitive material is made into a slurry, coated onto the interdigital electrode, and dried to obtain the final product.

8. A detection system for implementing the rapid detection method for the reheating degree of meat dishes based on electrochemical sensing as described in any one of claims 1 to 7, characterized in that, include: A portable detection device is provided with an electrochemical sensor unit, which includes at least one gas-sensitive electrode for collecting headspace gas from the reheated meat dish to be tested and generating an electrochemical response signal. The signal acquisition and processing module is used to record the complete time response curve of the electrochemical response signal and extract dynamic characteristic parameters from it. The input module is used to input the ingredient attribute parameters, cooking method parameters, and reheating method parameters of the reheated meat dish to be tested. The ingredient attribute parameters include at least one of meat type, fat content, and water content. The cooking method parameters include at least one of heating temperature, heating time, temperature change pattern, and endpoint temperature. The reheating method parameters include at least one of reheating start temperature, reheating power, reheating time, and reheating endpoint temperature. A storage module that stores pre-trained deep learning prediction models; The prediction module is used to input dynamic feature parameters, food attribute parameters, and reheating method parameters into the deep learning prediction model to obtain the reheating degree level. The display module is used to output and display the degree of reheating.

9. The detection system as described in claim 8, characterized in that, The portable detection device is configured to be attached to the inside or outside of meat dish packaging, or directly to the surface of the meat dish, and is powered and triggered by a mobile terminal via near-field communication.

Citation Information

Patent Citations

  • Biomarker analysis method for over-cooked taste of reheated pre-stewed beef

    CN117074153A

  • Multi-gas accurate detection method and system

    CN121027428A