Rapid and non-destructive beef carcass rating method by using electronic nose

Abstract

The invention relates to the technical field of detection rating, in particular to an electronic nose-based fast and lossless beef carcass rating method, which comprises the following steps of: implementing directional adsorption sampling on a target carcass through a multi-directional adsorption probe of an electronic nose, and synchronously recording the distribution tensor of volatile substances in a time-space dimension; inputting the distribution tensor into competitive adsorption compensation, carrying out molecular adsorption decoupling calculation, and outputting a correction concentration matrix of each independent volatile substance; coupling the correction concentration matrix with an ATP degradation kinetic model in real time, generating a metabolism characteristic spectrum through a time-varying concentration compensation algorithm, and inputting the metabolism characteristic spectrum into a grading neural network subjected to muscle texture optimization to output a final grading result; according to the method, through time-varying compensation and ATP degradation dynamic integral correction, the metabolic dynamics after slaughter are tracked in real time, the limitation that a traditional static rating model ignores the metabolic hysteresis effect is broken through, and the final grading result still has high consistency and reliability under the conditions of environment change, slaughter batch difference and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of detection and rating, and in particular to a method for rapidly and non-destructively rating beef carcasses using an electronic nose. Background Art

[0002] The grading of beef cattle carcasses is a crucial quality control step in the slaughtering, processing and circulation links, and is directly related to the market pricing of meat products and consumer acceptance. Traditional carcass grading methods mainly rely on manual visual inspection or texture analysis based on surface images. Evaluation indicators usually include appearance characteristics such as muscle color, fat distribution and intramuscular fat content. However, such methods are highly subjective, have poor repeatability, and are insufficiently sensitive to changes in the metabolic state of deep tissues. They are difficult to meet the modern beef cattle industry's demand for fast, non-destructive and highly consistent quality grading.

[0003] In recent years, electronic nose technology, as a detection method that simulates the biological olfactory perception system, has been initially introduced into the field of meat quality evaluation. By detecting volatile organic compounds (VOCs) released from the surface of meat, the electronic nose can reflect some metabolic process information. However, the existing carcass grading technology based on electronic nose still faces many technical bottlenecks: First, the sampling method lacks adaptability to the muscle anatomical structure, resulting in insufficient spatial representativeness of volatile substance collection; second, the competitive adsorption effect of multi-component gases on the sensor surface has not been effectively modeled, and the detection results are easily affected by component interference and have limited accuracy; third, the continuous metabolic dynamic changes of the carcass after slaughter are not taken into account, which makes the grading model show greater volatility and misjudgment risk at different detection time points. Summary of the Invention

[0004] The present invention provides a method for rapid and non-destructive grading of beef cattle carcasses using an electronic nose. This method combines the structural characteristics of muscle tissue, decouples multi-component adsorption interference, and dynamically compensates for the effects of metabolic evolution after slaughter, thereby improving the accuracy, stability, and standardization of the beef cattle carcass grading process.

[0005] A method for rapid and non-destructive beef cattle carcass grading using an electronic nose comprises the following steps:

[0006] S1: The multi-directional adsorption probe of the electronic nose is used to carry out directional adsorption sampling on the target carcass, and the distribution tensor of volatile substances in the time-space dimension is simultaneously recorded;

[0007] S2: Inputting the distribution tensor into competitive adsorption compensation, performing molecular adsorption decoupling calculation based on the Langmuir-Freundlich composite isotherm model, and outputting a correction concentration matrix of each independent volatile substance;

[0008] S3: The correction concentration matrix is coupled with the ATP degradation kinetic model in real time, and a metabolic characteristic spectrum is generated through a time-varying concentration compensation algorithm. The spectrum is input into a hierarchical neural network optimized by muscle texture to output the final rating result.

[0009] Optionally, the S1 specifically includes:

[0010] S11, scanning the direction of muscle fibers on the carcass surface with a laser texture scanner to generate a muscle texture vector map;

[0011] S12, controlling the adsorption probe microchannel array and the target area muscle fibers to form an optimal adsorption angle θ;

[0012] S13, start multi-stage pulse adsorption;

[0013] S14, the acquisition circuit records the spatial coordinates, timestamp, and three-dimensional gas concentration value of each adsorption point, and constructs a distribution tensor.

[0014] Optionally, the multi-stage pulse adsorption includes:

[0015] The first stage: 0.5kPa negative pressure adsorption of surface free molecules;

[0016] Phase 2: 1.2 kPa pulse pressure penetrates the sarcolemma to absorb deep substances;

[0017] The third stage: 0.8kPa steady-state adsorption maintains gas channel.

[0018] Optionally, the optimal adsorption angle θ satisfies:

[0019]

[0020] Among them, μ is the muscle density coefficient, ρ is the fat ratio, D parallel is the adsorption efficiency parallel to the muscle fiber direction, D perpendicular is the adsorption efficiency perpendicular to the muscle fiber direction.

[0021] Optionally, the S2 specifically includes:

[0022] S21, extracting the original concentration vector in the time-space dimension from the distribution tensor, where each element includes concentration components in the parallel, vertical and depth directions;

[0023] S22, establish the multi-substance competitive adsorption kinetic equation θ on the sensor surface i ,θ i is the surface coverage of the i-th substance;

[0024] S23, performing iterative decoupling calculations and outputting a correction concentration matrix, where the matrix dimensions are aligned with the distribution tensor space coordinates.

[0025] Optionally, the iterative decoupling calculation in S23 includes:

[0026] S231, initializing the concentration of each substance, initially setting it to the original extraction concentration;

[0027] S232, calculating the k-th iteration coverage;

[0028] S233, updating the substance concentration by the inverse compensation equation;

[0029] S234, after each iteration is completed, it is determined whether the current concentration change is less than the set convergence threshold. If the convergence condition is met, the iteration process is terminated, otherwise the iteration is continued.

[0030] Optionally, the S2 also includes a dynamic update mechanism, which includes recalibrating the adsorption affinity coefficient and the heterogeneity index through a micro mass spectrometry verification unit at preset time intervals, and automatically triggering a probe replacement reminder when the calibration error exceeds an error threshold.

[0031] Optionally, the S3 specifically includes:

[0032] S31, establishing a time-varying compensation coefficient. The time-varying compensation coefficient is described by an exponential function and is calculated based on the time difference between the slaughter recording time and the current detection time. A decay rate factor is introduced into the compensation coefficient. The decay factor is associated with the pH value detected in real time on the carcass surface and is dynamically corrected according to the change in pH value.

[0033] S32, dynamically correcting the calibration concentration matrix;

[0034] S33, construct metabolic profile:

[0035] In the spatial dimension, the corrected concentration matrix is divided into fixed grids, and the local concentration gradient is calculated in each grid;

[0036] In the time dimension, the concentration decay rate of the carcass is extracted at multiple time points after slaughter, and the concentration gradient, concentration decay rate, pH value, and decay rate factor are combined into a unified high-dimensional feature vector, namely the metabolic profile, which is used as the input of the hierarchical neural network.

[0037] S34, inputs the muscle texture optimized hierarchical neural network and calculates and outputs the final rating.

[0038] Optionally, the dynamic correction in S32 includes applying the time-varying compensation coefficient to the original concentration value at the current spatial position, and at the same time, calculating the integral term from the slaughter time to the current time period, wherein the integral term includes the changing process of ATP degradation reaction and metabolite generation reaction, and the integral term adopts real-time numerical solution to ensure compensation of concentration changes during the time evolution process.

[0039] Optionally, the input of the hierarchical neural network includes a metabolic profile and a muscle texture vector map, and the hierarchical neural network includes:

[0040] The metabolic profile and muscle texture vector map are jointly extracted through a three-dimensional convolutional layer. The convolution operation covers a predetermined spatial region and continuous time period to capture spatiotemporal correlation characteristics.

[0041] Entering the fat compensation unit, based on the intramuscular fat content obtained by near-infrared detection, the weights of the output features of the convolutional layer are adjusted to increase the importance of features of carcass samples with high fat content;

[0042] Different weights are assigned to each feature channel through the temporal attention gating unit. The gating mechanism outputs the attention coefficient of each channel based on the combined information of each channel feature and the detection time, after neural network weight mapping and activation function processing. The outputs of all feature channels are weighted and summarized to finally generate the rating score of the beef carcass.

[0043] Beneficial effects of the present invention:

[0044] The present invention scans the direction of muscle fibers in beef carcasses and dynamically adjusts the sampling direction of the adsorption probe in combination with an angle optimization formula, allowing the microfluidic array to adapt to the anisotropic characteristics of muscle texture during the adsorption process, effectively improving the difference in adsorption efficiency parallel to and perpendicular to the muscle fiber directions. At the same time, a staged pressure gradient adsorption strategy is introduced to achieve multi-level collection of surface and deep metabolites, enhancing the spatial representativeness and detection sensitivity of metabolic characteristic signals, and providing high-quality input data for subsequent classification models.

[0045] The present invention accurately models the influence of gas diffusion in different directions on adsorption behavior by establishing a multi-substance competitive adsorption equation based on the Langmuir-Freundlich composite model and combining it with a directional weighting mechanism. On this basis, an iterative decoupling calculation process based on reverse compensation is designed to effectively eliminate the nonlinear suppression effect of sensor surface coverage and improve the detection of low-concentration metabolites.

[0046] The present invention introduces a three-dimensional convolutional neural network with metabolic characteristic spectrum and laser muscle texture map as dual input in beef cattle carcass grading. By designing a convolution kernel with spatial scale matching, an adaptive compensation mechanism for fat content and a temporal attention gating unit, the comprehensive characteristic changes of beef cattle carcasses in space, time and tissue composition are accurately captured. In particular, through time-varying compensation and ATP degradation kinetic integral correction, the metabolic dynamics after slaughter are tracked in real time, breaking through the limitation of traditional static rating models that ignore metabolic lag effects, so that the final grading results are still highly consistent and reliable under conditions such as environmental changes and differences in slaughter batches. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only for the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0048] Figure 1 This is a flow chart of a rating method according to an embodiment of the present invention;

[0049] Figure 2 Schematic diagram of a hierarchical neural network according to an embodiment of the present invention. DETAILED DESCRIPTION

[0050] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments. It is also noted that, to provide a more detailed description, the following embodiments are best and preferred embodiments, and those skilled in the art may employ alternative methods for implementing certain known technologies. Furthermore, the accompanying drawings are intended only to provide a more detailed description of the embodiments and are not intended to limit the present invention.

[0051] It should be noted that references in the specification to "one embodiment," "an embodiment," "an exemplary embodiment," "some embodiments," etc. indicate that the described embodiments may include specific features, structures, or characteristics, but not every embodiment necessarily includes such specific features, structures, or characteristics. In addition, when specific features, structures, or characteristics are described in conjunction with an embodiment, it is within the knowledge of persons skilled in the relevant art to implement such features, structures, or characteristics in conjunction with other embodiments (whether or not explicitly described).

[0052] In general, terms can be understood, at least in part, from their use in context. For example, depending at least in part on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in the singular sense, or can be used to describe a combination of features, structures, or characteristics in the plural sense. Additionally, the term "based on" can be understood as not necessarily intended to convey an exclusive set of factors, but can instead, depending at least in part on the context, allow for the presence of other factors that are not necessarily explicitly described.

[0053] like Figure 1-Figure 2 As shown, a method for rapid and non-destructive beef cattle carcass grading using an electronic nose comprises the following steps:

[0054] S1: The multi-directional adsorption probe of the electronic nose is used to carry out directional adsorption sampling on the target carcass, and the distribution tensor of volatile substances in the time-space dimension is simultaneously recorded;

[0055] S2: Input the distribution tensor into the competitive adsorption compensation, perform molecular adsorption decoupling calculation based on the Langmuir-Freundlich composite isotherm model, and output the correction concentration matrix of each independent volatile substance;

[0056] S3: The correction concentration matrix is coupled with the ATP degradation kinetic model in real time, and a metabolic signature profile is generated through a time-varying concentration compensation algorithm. The profile is then input into a hierarchical neural network optimized for muscle texture to output the final rating result.

[0057] S1 specifically includes:

[0058] S11, using a laser texture scanner to scan the direction of muscle fibers on the carcass surface and generate a muscle texture vector map.

[0059] S12, controlling the piezoelectric drive direction-changing mechanism (106) so that the probe microchannel array and the muscle fibers in the target area are at an optimal adsorption angle θ, where the angle θ satisfies:

[0060]

[0061] Among them, μ is the muscle density coefficient, ρ is the fat ratio, D parallel is the adsorption efficiency parallel to the muscle fiber direction, D perpendicular is the adsorption efficiency perpendicular to the muscle fiber direction.

[0062] Adsorption efficiency parallel to the muscle fiber direction D parallel The adsorption test results of the probe in the direction of muscle fibers are used to calculate the mass of target volatile substances adsorbed per unit time, which can be calibrated with the standard gas response value in the initial sampling stage.

[0063] Adsorption efficiency perpendicular to the muscle fiber direction D perpendicular Using the same method as above, repeat the adsorption operation in the vertical direction and calculate the corresponding collection efficiency. A common practice is to set up a control adsorption path and compare the difference in adsorption amount.

[0064] Fat percentage ρ uses laser to calculate the body surface fat ratio based on the difference in scattering / absorption characteristics between fat and muscle tissue.

[0065] The muscle density coefficient μ is estimated by ultrasound combined with prior data to evaluate the density of muscle tissue. It can also be estimated based on the moisture content and texture comparison characteristics of muscle in known breeds and slaughter levels.

[0066] S13, start multi-stage pulse adsorption, which is carried out in the following three stages:

[0067] The first stage (0-2s): applying 0.5kPa negative pressure adsorption, mainly removing free molecules on the surface;

[0068] The second stage (2-5s): applying 1.2kPa pulse pressure to penetrate the sarcolemma and absorb deep substances;

[0069] The third stage (5-8s): Apply 0.8kPa steady-state adsorption to maintain the stability of the gas diffusion channel.

[0070] S14, record the following data of each adsorption point through the spatiotemporal synchronous acquisition circuit:

[0071] spatial coordinates (x, y, z);

[0072] Timestamp t (accuracy ±10ms);

[0073] Three-dimensional gas concentration value (C para ,C perp ,C depth ), corresponding to the gas concentration in the parallel, vertical and depth directions, C para ,C perp ,C depth They represent the gas concentration component in the direction parallel to the muscle fibers, the gas concentration component in the direction perpendicular to the muscle fibers, and the gas concentration component in the depth direction (subcutaneous direction);

[0074] And construct the distribution tensor accordingly:

[0075] T ijk =∑[C dir (t)×δ(x i ,y j ,z k )], where,dir∈{para,perp,depth};

[0076] Among them, T ijk Represents the distribution tensor, recording the adsorption concentration in the i, j, and k-th spatial discrete units, C dir (t) represents the gas concentration value in the specified direction (parallel / vertical / depth), as a function of time t, δ(x i ,y j ,z k ) represents the spatial position (x i ,y j ,z k ) at , para means parallel to the muscle fiber direction, perp means perpendicular to the muscle fiber direction, and depth means the direction of penetration perpendicular to the surface to the subcutaneous tissue.

[0077] S2 specifically includes:

[0078] S21, from the distribution tensor T ijk Extract the original concentration vector C in the time-space dimension raw(t,x,y,z), where each element contains the concentration components in the parallel, vertical and depth directions: [C para ,C perp ,C depth ], where C raw (t,x,y,z) represents the time-space concentration vector of the original sampling, which contains the gas concentration in the parallel, vertical and depth directions of each point.

[0079] S22, establish the multi-substance competitive adsorption kinetic equation on the sensor surface, expressed as:

[0080]

[0081] Among them, θ i is the surface coverage of the i-th substance, b i is the adsorption affinity coefficient of the i-th substance (obtained by micro-mass spectrometry calibration), n i is the heterogeneity index of the i-th substance (obtained by micro mass spectrometry calibration), n j is the heterogeneity index of the jth substance, α para ,α perp ,α depth are directional weight factors, which are the weight factors of the parallel direction concentration in the overall equivalent concentration, the vertical direction concentration in the overall equivalent concentration, and the depth direction concentration in the overall equivalent concentration, satisfying: α para +α perp +α depth =1.

[0082] S23, perform iterative decoupling calculation, the steps are as follows:

[0083] S231, initialize the concentration of each substance

[0084] S232, calculate the coverage of the kth iteration: represents the surface coverage of the i-th substance after the k-th iteration, represents the concentration of the i-th substance after the k-1th iteration;

[0085] S233, updates the species concentration through the inverse compensation equation:

[0086] represents the concentration of the i-th substance after the k-th iteration;

[0087] S234, when the convergence conditions are met: When , the iteration is terminated.

[0088] S24, output correction concentration matrix M corrected : Among them, the matrix dimension is the same as the distribution tensor T ijk The spatial coordinates are aligned, Represents the final concentration value of the nth substance after iteration convergence.

[0089] S25, dynamic update mechanism: recalibration every 8 hours through micro mass spectrometry verification i 、n i If the calibration error exceeds 5%, a probe replacement reminder will be triggered.

[0090] S3 specifically includes:

[0091] S31, establish time-varying compensation coefficient: Wherein, u represents the exponential decay rate factor, u = 0.0217 × [1 + 0.05 × (pH - 5.6)], pH is the carcass surface pH value obtained by near-infrared detection, T0 is the slaughter recording time, and t is the current detection time.

[0092] S32, the correction concentration matrix M corrected Perform dynamic correction, expressed as:

[0093]

[0094] The integral term is solved in real time by the Runge-Kutta method, k1 and k2 represent the ATP degradation rate parameters, m represents the empirical index of ATP degradation reaction, m = 1.2, g is the empirical index of metabolite generation reaction, g = 0.8, [ATP] represents the ATP concentration, and [M] is the metabolite concentration.

[0095] S33, construct metabolic profile P metabolic ,include:

[0096] Spatial dimension processing: Divide the modified matrix [M]′ into 5mm×5mm grids and calculate the concentration gradient within each grid:

[0097] Time dimension processing: Extract the concentration decay rate 15 minutes, 30 minutes, and 60 minutes after slaughter:

[0098] Δ[M]′ / Δt;

[0099] Forming the final metabolic profile P metabolic : The total number of features is 128.

[0100] S34, inputting the hierarchical neural network for muscle texture optimization, the specific steps are as follows:

[0101] S341, input data: input includes metabolic profile P metabolic (Concentration gradient, time decay rate, pH value, decay rate factor of each grid), muscle texture vector diagram;

[0102] S342, extract P through the 3D convolution layer metabolic Spatial-temporal correlation characteristics with laser texture images:

[0103] Use a 3D convolution kernel of size 5×5×3 (length×width×time) for convolution operation;

[0104] The step size is set to 2 × 2 × 1, i.e., stepping by 2 grids in space and continuously in time;

[0105] The activation function is ReLU activation.

[0106] S343, the fat compensation unit adjusts the feature weights, adjusts the feature channel weights according to the intramuscular fat content (IMF), improves the feature contribution of samples with high fat content, and performs weighted compensation on the convolution output feature weight W: W′=W×(1+0.3×IMF), where IMF represents the intramuscular fat content predicted by near-infrared, W represents the original feature weight matrix, and W′ is the compensated feature weight matrix;

[0107] S344, dynamically adjust the weight of each feature channel in the final rating based on the detection time:

[0108] Generate the attention coefficient α for each feature channel i :α i =σ(W α ·[h i ‖t]), where σ(·) is the Sigmoid activation function, W α is the temporal attention weight matrix

[0109] The eigenvector h i After concatenation with the detection time t, it passes through the fully connected layer W a And the Sigmoid function generates attention weights, and the final comprehensive score is: Score = ∑(α i FC i (P metabolic )), where h i is the feature representation vector, t is the detection time, and || represents the vector concatenation operation.

[0110] The present invention encompasses any alternatives, modifications, equivalents, and solutions that fall within the spirit and scope of the present invention. To provide a thorough understanding of the present invention, specific details are described in detail below in connection with the preferred embodiments of the present invention, but those skilled in the art will be able to fully understand the present invention without these detailed descriptions. Furthermore, to avoid unnecessary confusion regarding the essence of the present invention, well-known methods, processes, procedures, components, and circuits have not been described in detail.

[0111] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A method for rapid and non-destructive beef cattle carcass grading using an electronic nose, characterized in that: The following steps are involved: S1: The multi-directional adsorption probe of the electronic nose is used to carry out directional adsorption sampling on the target carcass, and the distribution tensor of volatile substances in the time-space dimension is simultaneously recorded; S2: Inputting the distribution tensor into competitive adsorption compensation, performing molecular adsorption decoupling calculation based on the Langmuir-Freundlich composite isotherm model, and outputting a correction concentration matrix of each independent volatile substance; S3: The correction concentration matrix is coupled with the ATP degradation kinetic model in real time, and a metabolic characteristic spectrum is generated through a time-varying concentration compensation algorithm. The spectrum is input into a hierarchical neural network optimized by muscle texture to output the final rating result.

2. The method for rapid and non-destructive beef cattle carcass grading using an electronic nose according to claim 1, wherein: Said S1 specifically includes: S11, scanning the direction of muscle fibers on the carcass surface with a laser texture scanner to generate a muscle texture vector map; S12, controlling the adsorption probe microchannel array and the target area muscle fibers to form an optimal adsorption angle θ; S13, start multi-stage pulse adsorption; S14, the acquisition circuit records the spatial coordinates, timestamp, and three-dimensional gas concentration value of each adsorption point, and constructs a distribution tensor.

3. The method for rapid and non-destructive beef cattle carcass grading using an electronic nose according to claim 2, wherein: The multi-stage pulse adsorption comprises: The first stage: 0.5kPa negative pressure adsorption of surface free molecules; Phase 2: 1.2 kPa pulse pressure penetrates the sarcolemma to absorb deep substances; The third stage: 0.8kPa steady-state adsorption maintains gas channel.

4. The method for rapid and non-destructive beef cattle carcass grading using an electronic nose according to claim 3, wherein: The optimal adsorption angle θ satisfies: Among them, μ is the muscle density coefficient, ρ is the fat ratio, D parallel is the adsorption efficiency parallel to the muscle fiber direction, D perpendicular is the adsorption efficiency perpendicular to the muscle fiber direction.

5. The method for rapid and non-destructive beef cattle carcass grading using an electronic nose according to claim 1, characterized in that: The S2 specifically includes: S21, extracting the original concentration vector in the time-space dimension from the distribution tensor, where each element includes concentration components in the parallel, vertical and depth directions; S22, establish the multi-substance competitive adsorption kinetic equation θ on the sensor surface i ,θ i is the surface coverage of the i-th substance; S23, performing iterative decoupling calculations and outputting a correction concentration matrix, where the matrix dimensions are aligned with the distribution tensor space coordinates.

6. The method for rapid and non-destructive beef cattle carcass grading using an electronic nose according to claim 5, characterized in that: The iterative decoupling calculation in S23 includes: S231, initializing the concentration of each substance, initially setting it to the original extraction concentration; S232, calculating the k-th iteration coverage; S233, updating the substance concentration by the inverse compensation equation; S234, after each iteration is completed, it is determined whether the current concentration change is less than the set convergence threshold. If the convergence condition is met, the iteration process is terminated, otherwise the iteration is continued.

7. The method for rapid and non-destructive beef cattle carcass grading using an electronic nose according to claim 6, characterized in that: The S2 also includes a dynamic update mechanism, which includes recalibrating the adsorption affinity coefficient and the heterogeneity index through a micro mass spectrometry verification unit at preset time intervals. When the calibration error exceeds the error threshold, a probe replacement reminder is automatically triggered.

8. The method for rapid and non-destructive beef cattle carcass grading using an electronic nose according to claim 1, wherein: The S3 specifically includes: S31, establishing a time-varying compensation coefficient. The time-varying compensation coefficient is described by an exponential function and is calculated based on the time difference between the slaughter recording time and the current detection time. A decay rate factor is introduced into the compensation coefficient. The decay factor is associated with the pH value detected in real time on the carcass surface and is dynamically corrected according to the change in pH value. S32, dynamically correcting the calibration concentration matrix; S33, construct metabolic profile: In the spatial dimension, the corrected concentration matrix is divided into fixed grids, and the local concentration gradient is calculated in each grid; In the time dimension, the concentration decay rate of the carcass is extracted at multiple time points after slaughter, and the concentration gradient, concentration decay rate, pH value, and decay rate factor are combined into a unified high-dimensional feature vector, namely the metabolic profile, which is used as the input of the hierarchical neural network. S34, inputs the muscle texture optimized hierarchical neural network and calculates and outputs the final rating.

9. The method for rapid and non-destructive beef cattle carcass grading using an electronic nose according to claim 8, characterized in that: The dynamic correction in S32 includes applying the time-varying compensation coefficient to the original concentration value at the current spatial position. At the same time, the integral term from the slaughter time to the current time period is calculated. The integral term includes the changing process of ATP degradation reaction and metabolite generation reaction. The integral term adopts real-time numerical solution to ensure compensation of concentration changes during the time evolution process.

10. The method for rapid and non-destructive beef cattle carcass grading using an electronic nose according to claim 9, characterized in that: The input of the hierarchical neural network includes a metabolic profile and a muscle texture vector map, and the hierarchical neural network includes: The metabolic profile and muscle texture vector map are jointly extracted through a three-dimensional convolutional layer. The convolution operation covers a predetermined spatial region and continuous time period to capture spatiotemporal correlation characteristics. Entering the fat compensation unit, based on the intramuscular fat content obtained by near-infrared detection, the weights of the output features of the convolutional layer are adjusted to increase the importance of features of carcass samples with high fat content; Different weights are assigned to each feature channel through the temporal attention gating unit. The gating mechanism outputs the attention coefficient of each channel based on the combined information of each channel feature and the detection time, after neural network weight mapping and activation function processing. The outputs of all feature channels are weighted and summarized to finally generate the rating score of the beef carcass.

Images