A method and system for meat quality evaluation based on digital image processing

Abstract

The embodiment of the application provides a meat product quality evaluation method and system based on digital image processing, and belongs to the technical field of image processing. The method comprises the following steps: acquiring a multi-angle / multi-channel image sequence of a meat product sample to obtain a calibration sequence for reflection component separation; performing joint demixing and inverse rendering correction of specular and diffuse reflection to obtain a net texture diffuse reflection image and a polarization map; extracting fascia rupture features based on the net texture diffuse reflection image and the polarization map, and correcting texture consistency based on the fascia rupture features to determine a tissue degradation score to form a freeze-thaw damage sensitive index; performing gated weighted fusion based on the freeze-thaw damage sensitive index to generate a freeze-thaw damage index vector, and inputting the freeze-thaw damage index vector into a texture evaluation model to output a texture evaluation result. The scheme of the application realizes adaptive fusion and quantitative evaluation of multi-source freeze-thaw damage features in the structure, moisture and optical layers, thereby significantly improving the accuracy and reliability of meat product quality evaluation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, in particular to a meat product quality evaluation method based on digital image processing and a meat product quality evaluation system based on digital image processing. BACKGROUND

[0002] At present, the quality evaluation of meat products mainly relies on two ways of physical detection and sensory judgment. Physical detection such as shear force determination and texture analyzer analysis can quantify tenderness and elasticity indicators, but it needs to destroy the sample structure and cannot realize rapid detection on the production line. Sensory judgment is intuitive, but it is greatly influenced by individual experience and subjective factors, and has poor repeatability and comparability. With the expansion of cold chain circulation and deep processing industry, it has become a common phenomenon that meat products undergo multiple freeze-thaw cycles during storage and transportation, and freeze-thaw damage often leads to muscle fiber structure fracture, water loss and tissue density decrease, directly affecting the taste and water retention of meat products. However, existing detection methods are difficult to non-contact, rapid and objective evaluate the microstructure degradation caused by freeze-thaw, and still remain at the macro physical parameter level, lacking the ability of fine perception of internal tissue continuity and optical characteristic changes.

[0003] In recent years, digital image processing has been preliminarily applied in the field of food quality detection, mainly focusing on meat color, fat ratio or surface defect recognition, but the research on the microstructure damage characteristics caused by freeze-thaw is still very limited. Existing image analysis methods are mostly based on color or gray scale statistical features, which are difficult to reflect the deep structure information such as muscle texture orientation, fascia continuity and micro-optical behavior; at the same time, different lighting conditions and the existence of surface water film will introduce strong reflection interference, which makes the image features deviate from the real tissue state, limiting the reliability and universality of image methods in texture evaluation scenarios.

[0004] Therefore, it is urgent to propose an image analysis method that can accurately identify the influence of freeze-thaw damage on the continuity of meat product tissue structure and optical properties under non-destructive conditions, so that the image features can truly reflect the degree of tissue degradation, and realize the objective quantitative evaluation of meat product quality. The solution of this problem has important significance for improving the quality monitoring ability of cold chain products and realizing online quality inspection of meat products. SUMMARY

[0005] The purpose of the embodiments of the present application is to provide a meat product quality evaluation method and system based on digital image processing, to at least solve the problems that the freeze-thaw damage of meat products is difficult to accurately identify by non-destructive methods, and the image features cannot truly reflect the degree of tissue structure degradation in the prior art.

[0006] To achieve the above object, the application provides a meat product quality evaluation method based on digital image processing, which comprises the following steps: acquiring a multi-angle / multi-channel image sequence of a meat product sample, and completing image calibration based on a gray card and / or a polarized plate reference image to obtain a calibration sequence for reflection component separation; performing joint specular and diffuse reflection unmixing and inverse rendering correction on the calibration sequence based on the prior knowledge of meat refractive index and water film thickness to obtain a net texture diffuse reflection image and a polarization map; extracting fascia rupture features of tissue structure continuity based on the net texture diffuse reflection image and the polarization map, and correcting texture consistency based on the fascia rupture features to determine a tissue degradation score to form a freeze-thaw damage sensitive index; performing gated weighted fusion based on the freeze-thaw damage sensitive index to generate a freeze-thaw damage index vector, and inputting the freeze-thaw damage index vector into a quality evaluation model to output a quality evaluation result.

[0007] Optionally, the multi-angle / multi-channel image sequence of the meat product sample is acquired by synchronously collecting reflection images of the meat product sample at multiple incident angles and / or multiple polarization angles in a preset shooting device, and capturing texture information of different wave bands through multi-spectral channels; and the collected image sequence is spatially registered and time-synchronized according to angle and channel indexes to generate multi-angle / multi-channel image sequence data for representing optical response characteristics of the sample surface.

[0008] Optionally, the image calibration based on the gray card and / or the polarized plate reference image is completed to obtain a calibration sequence for reflection component separation, which comprises the following steps: determining the brightness response curve and the color mapping matrix of each channel based on the gray card reference image, and sequentially performing brightness linearization and color uniformity correction on the multi-channel image sequence; extracting the transmission direction and the extinction ratio parameters corresponding to each polarization angle based on the polarized plate reference image, establishing a polarization direction correction matrix to compensate for the polarization angle error and the polarization intensity attenuation; and after the linearization, color uniformity correction and / or compensation are completed, performing spatial geometric registration and view consistency optimization on the multi-angle / multi-channel image to generate the calibration sequence for reflection component separation.

[0009] Optionally, the specular and diffuse reflection joint unmixing and inverse rendering correction are performed on the calibration sequence based on the meat refractive index and water film thickness priori to obtain a net texture diffuse reflection image and a polarization map, including: establishing a priori function in accordance with the Fresnel reflection model according to the meat refractive index, and combining the water film thickness parameter to construct a specular reflection energy constraint term to limit the energy distribution range of the specular and diffuse reflection components at the pixel level; multi-angle consistency optimization is performed on the calibration sequence, the specular component initial value is solved by using the polarization angle response difference, and the preliminary separated diffuse reflection image is obtained by using the sparse regularization and smooth constraint joint optimization; after obtaining the preliminary diffuse reflection image, the inverse rendering correction is performed according to the incident light direction, the material priori and the ambient light intensity parameter to restore the surface reflection distribution consistent with the illumination, and the net texture diffuse reflection image is generated; at the same time, the polarization degree and polarization direction distribution are determined based on the corresponding relationship between the reflectivity and transmissivity at each polarization angle in the separation process, and the corresponding polarization map is generated.

[0010] Optionally, the fascia rupture feature of the tissue structure continuity is extracted based on the net texture diffuse reflection image and the polarization map, including: constructing a local direction structure tensor in the net texture diffuse reflection image to obtain a texture main direction field, and using phase consistency detection to identify a structure discontinuity region as a fascia rupture candidate region; in combination with the polarization degree distribution reflecting the optical anisotropy of the tissue in the polarization map, the polarization consistency screening is performed on the candidate region to eliminate the false ruptures caused by the illumination or water film reflection; the main connected path of the fascia is reconstructed based on the skeletonization and minimum spanning tree constraint in the screened region, the rupture proportion, rupture length and direction distribution parameters are calculated, and the fascia rupture feature for representing the tissue structure continuity is obtained.

[0011] Optionally, the texture consistency is corrected based on the fascia rupture feature to determine the tissue degradation score, and a freeze-thaw damage sensitive index is formed, including: mapping the fascia rupture feature to the corresponding texture main direction field, performing direction correction and continuity compensation on the local texture consistency index, and generating a corrected texture consistency map; the texture energy attenuation rate and anisotropy change rate are calculated in the corrected texture consistency map to reflect the structure degradation degree caused by the rupture, and the tissue degradation score is obtained; the tissue degradation score is normalized in combination with the polarization degree feature and the diffuse reflection intensity feature in the polarization map representing the water change to form a freeze-thaw damage sensitive index for representing the freeze-thaw damage degree.

[0012] Optionally, the texture energy attenuation rate and the anisotropy change rate are calculated in the corrected texture consistency map to reflect the structural degradation degree caused by the fracture, and a tissue degradation score is obtained, including: performing multi-scale Gabor filtering on the corrected texture consistency map to extract the texture energy distribution at different spatial frequencies, and calculating the decline rate of the energy mean value at each scale as the texture energy attenuation rate; the anisotropy index is obtained based on the structure tensor, and the anisotropy difference before and after the fracture is calculated in the local neighborhood as the anisotropy change rate; the texture energy attenuation rate and the anisotropy change rate are weighted and fused according to the regional weight to obtain a tissue degradation score reflecting the comprehensive effect of the texture direction stability and the texture energy loss.

[0013] Optionally, the freeze-thaw damage sensitive index is used to perform gated weighted fusion to generate a freeze-thaw damage index vector, and the texture evaluation model is input to output a texture evaluation result, including: dividing the freeze-thaw damage sensitive index into texture feature components, polarization feature components and reflection intensity components according to the source, and inputting them into a gated weighted module to calculate channel weight coefficients; based on the weight coefficients, each component is normalized and weighted to generate a freeze-thaw damage index vector reflecting the comprehensive effect of structural degradation, water migration and optical change; the freeze-thaw damage index vector is input into the texture evaluation model, and the tenderness value and the elasticity value are calculated through the regression parameters obtained by training, and the texture comprehensive score is determined by the weighted combination of the two; at the same time of outputting the texture comprehensive score, the variance estimation and the confidence interval inference in the texture evaluation model are used to generate the corresponding uncertainty result, forming a complete texture evaluation result.

[0014] Optionally, the texture evaluation model is constructed based on supervised learning, and the construction rule is: taking the tenderness value and the elasticity value labeled by artificial as supervised labels, taking the freeze-thaw damage index vector as input features, and establishing the mapping relationship between the features and the texture labels through a multi-layer fully connected neural network to obtain the texture evaluation model; wherein, in the training stage, a weighted least squares loss function is used to balance the regression error of tenderness and elasticity, and a regularization constraint is introduced; in the inference stage, the texture comprehensive score is determined based on the regression mean value output by the model, and the uncertainty interval is calculated based on the variance or Dropout sampling result output by the model.

[0015] In a second aspect, the application provides a meat product quality evaluation system based on digital image processing, comprising: an acquisition unit configured to acquire a multi-angle / multi-channel image sequence of a meat product sample, and complete image calibration based on a gray card and / or a polarizing plate reference image to obtain a calibrated sequence for reflection component separation; a correction unit configured to perform specular and diffuse reflection joint unmixing and inverse rendering correction on the calibrated sequence based on prior knowledge of meat refractive index and water film thickness to obtain a net texture diffuse reflection image and a polarization map; a feature extraction unit configured to extract fascia rupture features based on the net texture diffuse reflection image and the polarization map, and correct texture consistency based on the fascia rupture features to determine a tissue degradation score to form a freeze-thaw damage sensitive index; and an output unit configured to perform gated weighted fusion based on the freeze-thaw damage sensitive index to generate a freeze-thaw damage index vector, and input the freeze-thaw damage index vector into a quality evaluation model to output a quality evaluation result.

[0016] Through the above technical solution, the application accurately restores the optical reflection characteristics of the surface of the meat product sample by multi-angle and multi-channel imaging combined with illumination and polarization calibration, effectively eliminating the interference caused by illumination conditions and water film reflection; further, the net texture diffuse reflection image and the polarization map that truly reflect the texture structure of the tissue are obtained by joint unmixing and inverse rendering correction, so that the image features are consistent with the actual tissue state. By extracting fascia rupture features and correcting texture consistency, the degree of degradation of the continuity of the tissue structure is quantitatively expressed, thereby constructing a feature index that is highly sensitive to freeze-thaw damage changes; finally, the tenderness and elasticity results are output by gated weighted fusion and a quality evaluation model, achieving non-destructive, quantitative and repeatable meat product quality evaluation.

[0017] Other features and advantages of the application will be described in detail in the following detailed description. BRIEF DESCRIPTION OF DRAWINGS

[0018] The accompanying drawings are included to provide a further understanding of the application and constitute a part of the specification, which together with the detailed description, serve to explain the application. In the drawings:

[0019] Figure 1 is a flowchart of a meat product quality evaluation method based on digital image processing provided by an embodiment of the application;

[0020] Figure 2 is a system structure diagram of a meat product quality evaluation system based on digital image processing provided by an embodiment of the application. DETAILED DESCRIPTION

[0021] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only for illustration and explanation of the embodiments of this application and are not intended to limit the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0022] It should be noted that if the embodiments of this application involve directional indicators (such as up, down, left, right, front, back, etc.), the directional indicators are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indicators will also change accordingly.

[0023] Furthermore, if the embodiments of this application involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, features defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the technical solutions of various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. If the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed in this application.

[0024] like Figure 1 As shown, this invention provides a method for evaluating the texture of meat products based on digital image processing, the method comprising:

[0025] Step S1: Obtain multi-angle / multi-channel image sequences of meat product samples, and perform image calibration based on gray card and / or polarizer reference images to obtain a calibration sequence for reflection component separation.

[0026] Specifically, acquiring multi-angle / multi-channel image sequences of meat product samples includes: simultaneously acquiring reflection images of meat product samples at multiple incident angles and / or multiple polarization angles in a preset imaging device, capturing texture information of different bands through multispectral channels; spatially registering and temporally synchronizing the acquired image sequences according to angle and channel indexes to generate multi-angle / multi-channel image sequence data for characterizing the optical response characteristics of the sample surface.

[0027] Further, based on the gray card and / or polarizing plate reference image, image calibration is completed to obtain a calibration sequence for reflection component separation, including: determining the brightness response curve and color mapping matrix of each channel based on the gray card reference image, and sequentially performing brightness linearization and color uniformity correction on the multi-channel image sequence; based on the polarizing plate reference image, the transmission direction and extinction ratio parameters corresponding to each polarization angle are extracted, and a polarization direction correction matrix is established to compensate for polarization angle errors and polarization intensity attenuation; after linearization, color uniformity correction and / or compensation are completed, spatial geometric registration and view angle consistency optimization are performed on the multi-angle / multi-channel image to generate a calibration sequence for reflection component separation.

[0028] In the embodiment of the application, in order to accurately capture the optical characteristics of the meat product surface, a multi-angle and multi-channel collaborative imaging data acquisition process needs to be established first. Specifically, during the shooting stage, by configuring an imaging device with adjustable incident angle and polarization angle, the meat product sample is uniformly irradiated under multiple incident light conditions, and is synchronously collected through multiple spectral channels, thereby obtaining a multi-angle / multi-channel image sequence containing different waveband information such as visible light, near-infrared and even short-wave infrared. Each frame of image not only records the brightness and color distribution information of the sample surface, but also implicitly records the optical response characteristics such as muscle fiber direction, surface water film state and local refraction difference. Those skilled in the art can understand that the change of illumination angle and polarization direction will directly affect the proportional distribution of specular reflection and diffuse reflection, so the joint collection of multi-angle and multi-polarization is a necessary basis for subsequent reflection component separation.

[0029] In order to ensure the space-time consistency between the acquisition sequences, after the image acquisition is completed, the image data of different angles and channels needs to be spatially registered and time-synchronized. Spatial registration can use a registration algorithm based on feature point matching or phase correlation to map images under different viewing angles to a unified coordinate space to eliminate geometric distortion caused by viewing angle offset in the imaging process; time synchronization is realized through a trigger signal or a time sequence controller to keep the exposure time of different channels consistent, avoiding brightness errors caused by light source flickering or sample reflection changes. The multi-angle / multi-channel image sequence data obtained after processing can completely express the optical response characteristics of the sample surface in a unified coordinate system, providing data support for subsequent illumination modeling and reflection component separation.

[0030] After obtaining the image sequence, image calibration needs to be completed through gray card and / or polarizing plate reference image to ensure the comparability of data under different shooting conditions. The gray card reference image is used to establish the brightness response curve and the color mapping matrix. By regression fitting the gray response of the gray card area, the linearization coefficient of the imaging device can be obtained, thereby realizing the linearization correction of the image brightness. At the same time, the color space conversion matrix is calculated based on the multi-point mapping results of the standard color card or the gray card, so that the color difference between different waveband channels is unified, thereby eliminating the color deviation caused by the light source color temperature and the sensor response difference.

[0031] In the polarization calibration process, the polarizing plate reference image is used to extract the transmission direction and the extinction ratio parameters corresponding to each polarization angle. Since the transmittance and extinction characteristics under different polarization angles will affect the distribution of the reflected component, it is necessary to establish a polarization direction correction matrix to compensate for the polarization angle deviation and polarization intensity attenuation. This process is usually completed by measuring the reflection response of a standard sample with a known polarization direction, and the obtained parameters are used to correct the polarization deviation in the actual sample imaging.

[0032] After completing the brightness linearization, color uniformity correction and polarization compensation, spatial geometric registration and view angle consistency optimization of the multi-angle images are still needed, so that each angle image corresponds to the same surface area in space, thereby ensuring the physical comparability of the subsequent pixel-level reflected component separation. The calibration sequence generated through the above calibration process can effectively suppress the interference of uneven illumination, sensor drift and polarization deviation, and provide standardized input for the subsequent specular and diffuse reflection joint unmixing. In this way, the accuracy of the reflected component separation and optical feature extraction of meat product surface can be significantly improved, and the imaging results can truly reflect the micro-texture structure and organization state of the sample, thereby laying a reliable image foundation for the quantitative identification of freeze-thaw damage.

[0033] Step S2: performing specular and diffuse reflection joint unmixing and inverse rendering correction on the calibration sequence based on the meat refractive index and water film thickness priori to obtain a net texture diffuse reflection image and a polarization map.

[0034] Specifically, a prior function conforming to the Fresnel reflection model is established according to the meat refractive index, and a specular reflection energy constraint term is constructed by combining the water film thickness parameter to limit the energy distribution range of the specular and diffuse reflection components at the pixel level; multi-angle consistency optimization is performed on the calibration sequence, the specular component initial value is solved by using the polarization angle response difference, and sparse regularization and smooth constraint joint optimization are used to obtain a preliminary separated diffuse reflection image; after obtaining the preliminary diffuse reflection image, inverse rendering correction is performed according to the incident light direction, material priori and ambient light intensity parameter to restore the surface reflection distribution consistent with the illumination, thereby generating a net texture diffuse reflection image; at the same time, the polarization degree and polarization direction distribution are determined based on the corresponding relationship between the reflectivity and transmissivity of each polarization angle in the separation process, thereby generating a corresponding polarization map.

[0035] In the embodiments of the present application, to accurately separate the reflection component formed by the joint action of the tissue structure and the surface water film on the surface of the meat sample, an optical constraint model is established based on the refractive index of meat and the thickness of the water film, the joint unmixing of specular and diffuse reflection is performed on the calibration sequence, and the surface reflection distribution consistent with the illumination is further recovered through inverse rendering correction, so as to obtain the net texture diffuse reflection image and the polarization map. Since the surface of the meat product is usually uneven wet tissue, the light experiences the refraction and scattering effect of the air-water film interface and the water film-muscle fiber interface during the incidence and reflection process, resulting in a complex energy distribution relationship of the specular reflection and diffuse reflection components at the pixel level. If only the conventional brightness correction or the empirical reflection model is used, the problem of overfitting in the highlight area or the loss of details in the dark area is easily caused, and the texture characteristics of the meat tissue cannot be truly reflected.

[0036] Therefore, first, the average refractive index of meat tissue (usually between 1.35 and 1.42) and the refractive index of the water film are combined with the incidence angle and the refraction angle to establish a reflectivity prior function in accordance with the Fresnel reflection theory. The incidence angle and the refraction angle are combined to establish a reflectivity prior function in accordance with the Fresnel reflection theory. The function is defined as:

[0037] ;

[0038] wherein, is determined by the refraction law:

[0039] ;

[0040] Based on the model, the reflectivity variation law of the incident light at different angles can be accurately described. Further, to represent the weakening effect of the surface water film thickness on the specular reflection energy, an energy constraint term is introduced:

[0041] ;

[0042] wherein, is the effective absorption coefficient of the water film, which is used to adjust the nonlinear attenuation relationship between the water film thickness and the reflection energy. The constraint term limits the upper bound of the specular reflection energy at the pixel level, so that the subsequent unmixing solution can avoid local reflection overestimation, thereby ensuring the true restoration of the texture details.

[0043] Then, multi-angle consistency optimization is performed on the calibration sequence. Since the observed brightness under different incident angles and polarization angles has physical correlation, the initial value of the specular component can be calculated using the difference in polarization angle response. Specifically, the reflection intensity of the same pixel under different polarization angles is fitted to obtain the preliminary estimate of the specular component . To improve the separation accuracy, a sparse regularization constraint term is introduced to maintain the local concentration characteristics of the specular reflection, and a smoothing constraint term is introduced to ensure the spatial continuity of the diffuse reflection image. The joint optimization objective function can be represented as:

[0044] ;

[0045] wherein, is the diffuse reflection component, and are the weight coefficients of the sparse and smooth constraints, respectively. By iteratively solving the equation, a preliminary diffuse reflection image with reasonable energy distribution and smooth boundaries can be obtained.

[0046] After obtaining the preliminary diffuse reflection image, inverse rendering correction is performed according to the incident light direction , the material albedo , and the ambient light intensity . This process uses a simplified Lambertian model to express the lighting consistency:

[0047] ;

[0048] By minimizing the difference between the predicted brightness and the observed brightness, self-consistent correction of the lighting is achieved, resulting in a net texture diffuse reflection image, which effectively eliminates the brightness difference under different angles and lighting conditions.

[0049] In addition, during the separation process, the degree of polarization P and the polarization direction are calculated based on the reflectivity and transmissivity under each polarization angle, wherein:

[0050] ;

[0051] wherein, . The calculated degree of polarization and polarization direction distribution are combined into a polarization map, which is used to represent the state of the surface water film and the optical anisotropy characteristics of the tissue.

[0052] Through the above optical modeling and optimization solution, the specular and diffuse reflection components can be effectively separated while maintaining image details, making the imaging results truly reflect the texture structure and optical properties of the surface tissue of meat products, and fundamentally improving the physical consistency of texture recovery and lighting compensation, providing a high reliability image basis for subsequent freeze-thaw damage sensitive feature extraction.

[0053] Step S3: extracting fascia rupture features of tissue structure continuity based on the net texture diffuse reflection image and the polarization map, and correcting the texture consistency based on the fascia rupture features to determine a tissue degradation score, to form a freeze-thaw damage sensitive index.

[0054] Specifically, the fascia rupture features of tissue structure continuity are extracted based on the net texture diffuse reflection image and the polarization map, including: constructing a local directional structure tensor in the net texture diffuse reflection image to obtain a texture main direction field, and using phase consistency detection to identify a structure discontinuity region as a fascia rupture candidate region; combining the polarization degree distribution reflecting the optical anisotropy of the tissue in the polarization map, performing polarization consistency screening on the candidate region to eliminate false ruptures caused by illumination or water film reflection; and based on the skeletonization and minimum spanning tree constraint, reconstructing the main connected path of the fascia in the screened region, calculating the rupture proportion, rupture length and direction distribution parameters, and obtaining the fascia rupture features for representing the tissue structure continuity.

[0055] Further, the fascia rupture features are corrected to determine the tissue degradation score based on the texture consistency, to form a freeze-thaw damage sensitive index, including: mapping the fascia rupture features to the corresponding texture main direction field, performing direction correction and continuity compensation on the local texture consistency index, and generating a corrected texture consistency map; calculating the texture energy attenuation rate and the anisotropy change rate in the corrected texture consistency map to reflect the structure degradation degree caused by the rupture, to obtain the tissue degradation score; and performing joint normalization processing on the tissue degradation score, the polarization degree features representing the water change in the polarization map, and the diffuse reflection intensity features, to form a freeze-thaw damage sensitive index for representing the freeze-thaw damage degree.

[0056] Specifically, the texture energy attenuation rate and the anisotropy change rate are calculated in the corrected texture consistency map to reflect the structure degradation degree caused by the rupture, to obtain the tissue degradation score, including: performing multi-scale Gabor filtering on the corrected texture consistency map to extract the texture energy distribution at different spatial frequencies, and calculating the decline rate of the energy mean value at each scale as the texture energy attenuation rate; calculating the anisotropy index based on the principal direction eigenvalue ratio of the structure tensor, and calculating the anisotropy difference before and after the rupture in the local neighborhood as the anisotropy change rate; and weighting and fusing the texture energy attenuation rate and the anisotropy change rate according to the regional weight, to obtain the tissue degradation score reflecting the comprehensive effect of the texture direction stability and the texture energy loss.

[0057] In the embodiment of the present application, after the acquisition and correction of the net texture diffuse reflectance image and the polarization map are completed, the further key step is to quantitatively extract the continuity features of the meat tissue structure, so as to identify and describe the microscopic fracture, structural degradation and optical unevenness features caused by freeze-thaw damage from the image level. The fascia layer in the meat tissue is a key structure that connects muscle fibers, restricts water distribution and maintains tissue integrity, which is extremely susceptible to the combined effects of micro-crack propagation, water recrystallization and fiber stretching and contraction during the freeze-thaw cycle. When fascia rupture occurs, the optical anisotropy of the local tissue decreases significantly, the diffuse reflectance distribution appears direction disorder, and the polarization retention ability decreases. Therefore, by constructing a quantifiable fascia rupture feature based on the net texture diffuse reflectance image and the polarization map, not only can the change law of the continuity of the tissue structure be directly reflected, but also a key feature input for subsequent quantitative evaluation of tissue degradation is provided.

[0058] Specifically, first, the local directional structure tensor is calculated in the net texture diffuse reflectance image, to represent the texture orientation and direction stability of each pixel neighborhood. The structure tensor is defined as the second-order statistics of the local gray gradient, and its form is:

[0059] ;

[0060] wherein, and respectively represent the gradient components of the image gray in the x and y directions, is the Gaussian smoothing kernel under the scale σ. By eigen-decomposition of the structure tensor, the principal direction vector and the corresponding eigenvalue ratio can be obtained, so as to determine the main direction field of the texture and the local direction intensity. The principal direction angle reflects the average direction of the muscle fiber arrangement, and the eigenvalue ratio represents the direction consistency of the texture. For normal muscle tissue, the principal direction field presents a continuous and smooth change in the local region; and once the fascia rupture occurs, the direction field of the region mutates sharply, and the eigenvalue ratio sharply decreases, so the phase consistency detection can be used to identify the structure discontinuous regions.

[0061] The phase consistency detection is a structure feature detection method based on local phase information, which can maintain stable response under different brightness conditions. By decomposing the image through a multi-scale Log-Gabor filter bank, the consistency measure of the local phase under each scale is calculated. When the local phase mutation is significant and the direction gradient is high, it can be determined as a potential fracture region. The structure discontinuous region obtained by this method is defined as the fascia rupture candidate region, representing the positions where tissue delamination or micro-cracks may exist.

[0062] However, there is a certain thickness of water film on the surface of the meat sample, which will cause partial reflection of high light or polarization anomaly, which may lead to misjudgment. Therefore, it is necessary to combine the polarization degree distribution in polarization mapping to screen the polarization consistency of the above candidate regions. The polarization degree P reflects the consistency of the reflected light wave in the polarization direction, and its calculation formula is:

[0063] ;

[0064] wherein, and are the reflection intensities along the maximum and minimum polarization directions. For the real fascia rupture area, due to the change of micro-normal direction of the surface caused by the destruction of the tissue structure, the polarization degree usually shows a significant decrease; while the pseudo rupture area caused by the high light of the light or the interference of the water film has a higher polarization degree and consistent polarization direction. Therefore, by setting the polarization consistency threshold, non-structural rupture areas can be effectively removed, and rupture areas reflecting the destruction of the real tissue structure can be retained.

[0065] After obtaining the screened rupture area, in order to accurately depict the topology of the fascia layer, skeletonization and connected reconstruction operations need to be performed. Skeletonization is used to extract the center path of the rupture area, and through the morphological thinning algorithm, the edge pixels are gradually removed and the main structure is retained. Then, based on the minimum spanning tree (MST) constraint, the skeleton nodes are connected and optimized to ensure that the connection relationship between the rupture fragments meets the actual physical structure. Through this reconstruction process, the main connected path of the fascia can be recovered, and the rupture feature parameters can be further extracted, including the rupture proportion, the rupture length and the direction distribution parameter. Among them, the rupture proportion is defined as the proportion of the rupture area to the total area of the fascia, the rupture length is calculated by the cumulative length of the skeleton path, and the direction distribution parameter is calculated according to the variance of the main direction angle of each rupture fragment, which is used to reflect the dispersion degree of the rupture orientation. The final fascia rupture feature vector can quantitatively describe the integrity and directionality change of the tissue structure, which is an important input for the subsequent texture consistency correction.

[0066] After obtaining the fascicle rupture feature, the overall texture consistency needs to be corrected based on the feature to reflect the local direction discontinuity and energy attenuation effect caused by the rupture. Specifically, the fascicle rupture feature is mapped to the corresponding texture principal direction field, and the direction correction and continuity compensation are performed on the direction consistency index in each pixel neighborhood. The process of direction correction mainly aims at the direction deviation between the rupture area and the non-rupture area. By calculating the angle difference of the principal direction at the rupture edge and performing weighted smoothing, the transition of the texture principal direction field at the rupture boundary is more in line with the actual tissue stress distribution. Continuity compensation corrects the direction consistency index by interpolation or local reconstruction, so that the texture features of the surrounding area of the rupture can be fully expressed. After the above correction, the corrected texture consistency map reflecting the real texture state of the tissue can be obtained, providing basic data for subsequent degradation analysis.

[0067] In the corrected texture consistency map, in order to further quantify the degradation degree of the tissue structure, the texture energy attenuation rate and the anisotropy change rate need to be calculated. First, the image is convolved and decomposed by using a multi-scale Gabor filter to extract the texture energy distribution at different spatial frequencies. For each scale s, the filter response energy is defined as The square sum of the filter output corresponding to the scale is defined as:

[0068] ;

[0069] Wherein, is the Gabor kernel of scale s. The texture energy attenuation rate can be obtained by calculating the decline rate of the average energy of each scale, which describes the trend of texture energy loss as the spatial frequency increases. High energy attenuation rate usually means that there are more micro-fractures or detail losses in the tissue structure, indicating that the tissue compactness decreases.

[0070] On the other hand, the anisotropy change rate is used to evaluate the change degree of the texture direction stability before and after the rupture. Based on the structure tensor of the corrected texture consistency map, the principal direction eigenvalue ratio is calculated as And the difference value in the local neighborhood before and after the rupture is calculated as

[0071] ;

[0072] A larger anisotropy change rate represents that the local directionality is destroyed and the texture tends to be disordered. Finally, the texture energy attenuation rate and the anisotropy change rate are weighted and fused according to the regional weight to obtain the tissue degradation score reflecting the comprehensive effect of the texture direction stability and the texture energy loss.

[0073] After obtaining the tissue degradation score, to comprehensively characterize the multi-dimensional variation characteristics in the freeze-thaw process, the score is combined with the polarization degree feature reflecting the water migration characteristics and the diffuse reflectance intensity feature in the polarization mapping for joint normalization processing. The decrease in the polarization degree often corresponds to the redistribution of water in the tissue or the rupture of the intercellular water film, and the decrease in the diffuse reflectance intensity reflects the weakening of the tissue density and light scattering ability. By normalization and weighted fusion, these different source but physically related features are integrated into a single index vector, i.e., a freeze-thaw damage sensitive index. This index comprehensively considers three factors of structural fracture, energy loss and water state change, and is a core parameter for reflecting the freeze-thaw damage degree of meat products.

[0074] Step S4: performing gated weighted fusion based on the freeze-thaw damage sensitive index to generate a freeze-thaw damage index vector, and inputting a texture evaluation model to output a texture evaluation result.

[0075] Specifically, the freeze-thaw damage sensitive index is divided into texture feature components, polarization feature components and reflectance intensity components according to the source, and is respectively input into a gated weighting module to calculate channel weight coefficients; based on the weight coefficients, normalization and weighted fusion are performed on each component to generate a freeze-thaw damage index vector reflecting the comprehensive effects of structural degradation, water migration and optical changes; the freeze-thaw damage index vector is input into the texture evaluation model, the tenderness value and the elasticity value are calculated through the regression parameters obtained by training, and the texture comprehensive score is determined by the weighted combination of the two; at the same time of outputting the texture comprehensive score, the uncertainty result corresponding to the texture comprehensive score is generated by using the variance estimation and confidence interval inference inside the texture evaluation model, and a complete texture evaluation result is formed.

[0076] Further, the texture evaluation model is constructed based on supervised learning, and the construction rule is: taking the artificially labeled tenderness value and elasticity value as the supervised label, taking the freeze-thaw damage index vector as the input feature, establishing the mapping relationship between the features and the texture label through a multi-layer fully connected neural network to obtain the texture evaluation model; wherein, in the training stage, a weighted least squares loss function is used to balance the regression error of tenderness and elasticity, and a regularization constraint is introduced; in the inference stage, the texture comprehensive score is determined based on the regression mean value output by the model, and the uncertainty interval is calculated based on the variance or Dropout sampling result output by the model.

[0077] In the last stage of freeze-thaw damage identification, the multi-source features extracted above need to be fused into a unified index that can quantitatively represent the change of meat product texture, so as to comprehensively reflect the influence of freeze-thaw process on the tissue structure, optical properties and water content of meat products. For this purpose, a multi-component fusion strategy based on a gating weighting mechanism is designed to generate a freeze-thaw damage index vector and input it into a texture evaluation model, and finally output a texture evaluation result with physical consistency and statistical confidence. This process actually performs semantic compression of the image structure information and optical polarization information at the feature layer, so that the complex freeze-thaw damage mechanism is expressed in a low-dimensional and highly correlated form in the numerical space.

[0078] Specifically, first, the freeze-thaw damage sensitive indexes are divided into three types of feature components according to the source type: texture feature component, polarization feature component and reflection intensity component. The texture feature component is mainly derived from the corrected texture consistency map and the tissue degradation score, reflecting the continuity and direction stability of muscle fiber arrangement; the polarization feature component is derived from the polarization degree and polarization direction information of the polarization mapping, representing the optical anisotropy changes caused by water migration and intercellular interface changes during freeze-thaw process; the reflection intensity component is taken from the brightness response of the net texture diffuse reflection image, reflecting the tissue density and surface scattering characteristics. These three types of features respectively represent the degradation characteristics of meat products in structure, optics and scattering behavior, and are the basic information sources for constructing freeze-thaw damage indexes.

[0079] On this basis, in view of the differences in physical meaning and numerical scale of different feature components, a gating weighting module is introduced to adaptively calculate the weight coefficients of each channel. The gating weighting module performs statistical analysis on the input feature mean, variance and local gradient response based on a nonlinear mapping structure, and outputs a weight factor in the interval (0, 1) through a Sigmoid gating function. This process can be formalized as:

[0080] ;

[0081] wherein, is the statistical description vector of the i-th type of feature component, and are the gating mapping parameters, is the Sigmoid function. Through this weight mechanism, the features sensitive to texture change (such as polarization degree reduction and direction consistency destruction) can be automatically highlighted in the fusion stage, and the features affected by light or measurement noise can be suppressed.

[0082] Then, based on the calculated weight coefficients, each feature component is subjected to normalized weighted fusion. The fusion rule is defined as:

[0083] ;

[0084] The normalization processing ensures that different source features remain comparable in numerical scale while avoiding feature imbalance caused by excessive weight of a single channel. After fusion, the freeze-thaw damage index vector is obtained has a clear physical meaning, i.e., simultaneously encoding the structural degradation degree, water migration intensity, and optical response change in a vector space, thereby realizing cross-modal expression of freeze-thaw damage mechanism. This index vector not only has the comprehensiveness of high-dimensional features, but also retains the interpretability of original physical information.

[0085] Subsequently, the freeze-thaw damage index vector is input into the texture evaluation model to obtain the final texture evaluation result. The texture evaluation model is constructed based on supervised learning in this embodiment and is trained by a large number of artificially labeled meat product sample data. Specifically, the artificially measured tenderness value and elasticity value are used as supervised labels to construct the input-output correspondence relationship. The model structure adopts a multi-layer fully connected neural network form, including an input layer, multiple hidden layers, and an output layer, wherein the hidden layers use ReLU activation functions to enhance the non-linear fitting ability. The core goal of the model training process is to minimize the weighted mean square error between the predicted result and the artificially labeled result, and the loss function can be represented as:

[0086] ;

[0087] wherein, and are the tenderness and elasticity prediction values output by the model, is a weight coefficient for balancing the error contribution of the two types of indicators; is a regularization term to suppress overfitting; and are the artificially labeled tenderness true value and elasticity true value, respectively. In the training phase, the network parameters are updated iteratively through the backpropagation algorithm until the loss function converges.

[0088] After training is completed, the texture evaluation model can realize texture prediction of freeze-thaw samples. In the inference phase, the tenderness value and elasticity value output by the model reflect the performance of meat products in terms of structural fracture and elastic recovery, respectively. In order to obtain a comprehensive texture score, the two are combined with a weight:

[0089] ;

[0090] wherein, is an empirically determined weight parameter, which is usually determined according to sensory evaluation results or experimental calibration. The combined score can intuitively reflect the overall taste and texture characteristics of meat products.

[0091] It is worth noting that the embodiment does not only output a single numerical value, but also provides an uncertainty estimation of the texture prediction result to reflect the confidence range of the model output. To this end, a variance estimation or Dropout sampling mechanism is introduced when training the model to approximate the uncertainty of the prediction distribution by the volatility of the network in multiple forward propagations. By calculating the variance or confidence interval of the output sample, the following uncertainty expression can be constructed:

[0092] ;

[0093] wherein, is the texture prediction value obtained by the k-th sampling, is the average value. The uncertainty result is particularly important in the texture evaluation process and can be used as a reference basis for determining sample abnormalities or measurement fluctuations in the quality control link.

[0094] Through the above steps, the obtained texture evaluation result not only contains a quantitative texture comprehensive score, but also includes a corresponding confidence interval and variance estimation, so that the entire evaluation process has statistical interpretability and decision reliability. Technically, this gating weighted fusion mechanism significantly improves the expression ability of freeze-thaw damage features, so that features of different physical sources can dynamically allocate weights in the fusion process according to the actual sample features, avoiding the feature bias problem caused by channel coupling in the traditional linear weighting method. At the same time, the texture evaluation model constructed based on supervised learning learns through end-to-end mapping, so that a nonlinear correspondence is established between the freeze-thaw damage indicators and the real texture attributes, ensuring that the evaluation result is highly consistent with the actual sensory quality of meat products.

[0095] As shown in Figure 2 , the embodiment of the present application provides a meat product texture evaluation system based on digital image processing, which comprises: an acquisition unit for acquiring a multi-angle / multi-channel image sequence of a meat product sample, and completing image calibration based on gray card and / or polarizing plate reference image to obtain a calibration sequence for reflection component separation; a correction unit for performing specular and diffuse reflection joint unmixing and inverse rendering correction on the calibration sequence based on meat refractive index and water film thickness priori to obtain net texture diffuse reflection image and polarization mapping; a feature extraction unit for extracting fascia rupture features based on net texture diffuse reflection image and polarization mapping, and correcting texture consistency based on fascia rupture features to determine tissue degradation score to form freeze-thaw damage sensitive indicators; an output unit for performing gating weighted fusion to generate freeze-thaw damage indicator vector based on the freeze-thaw damage sensitive indicators, and inputting the texture evaluation model to output texture evaluation results.

[0096] Those skilled in the art can understand that all or part of the steps of the method for implementing the above-mentioned embodiments can be completed by programs instructing relevant hardware, the programs are stored in a storage medium, and the programs include a plurality of instructions for enabling a single-chip microcomputer, a chip or a processor to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various storage media capable of storing program codes.

[0097] The optional embodiments of the present application are described in detail above in combination with the drawings, but the embodiments of the present application are not limited to the specific details in the above-described embodiments. Within the technical concept range of the embodiments of the present application, various simple modifications can be made to the technical solutions of the embodiments of the present application, and these simple modifications all belong to the protection scope of the embodiments of the present application. In addition, it should be noted that each specific technical feature described in the above-described specific embodiments can be combined in any appropriate manner without contradiction. In order to avoid unnecessary repetition, the embodiments of the present application will not be described again for various possible combinations.

[0098] In addition, various different embodiments of the present application can also be combined in any manner, as long as it does not deviate from the idea of the embodiments of the present application, and it should also be considered as disclosed by the embodiments of the present application.

Claims

1. A method for meat quality assessment based on digital image processing, characterized in that, The method comprises: Obtaining a multi-angle / multi-channel image sequence of a meat product sample, and completing image calibration based on a gray card and / or a polarizing plate reference image to obtain a calibrated sequence for reflection component separation; Performing specular and diffuse reflection joint unmixing and inverse rendering correction on the calibrated sequence based on meat refractive index and water film thickness priors to obtain a net texture diffuse reflection image and a polarization map; comprising: Establishing a prior function conforming to a Fresnel reflection model according to the meat refractive index, and combining the water film thickness parameter to construct a specular reflection energy constraint term to limit the energy distribution range of the specular and diffuse reflection components at the pixel level; wherein, The prior function is expressed as: ； wherein is the angle of incidence; is the angle of refraction; is the reflectance function value at the angle of incidence under the condition that the angle of incidence is The energy constraint term is expressed as: ； wherein is the effective absorption coefficient of the water film; is the surface water film thickness; denotes the specular reflection energy intensity at spatial position x; denotes the reflectivity function value at spatial position x, corresponding to the incidence angle under the condition Performing multi-angle consistency optimization on the calibrated sequence, using polarization angle response differences to solve the initial value of the specular component, and through joint optimization of the objective function to jointly optimize sparse regularization and smooth constraint, to obtain a preliminary separated diffuse reflection image; wherein, The joint optimization objective function is expressed as: ； wherein, is the diffuse reflection component; with are the weight coefficients for the sparsity and smoothness constraints, respectively; is the specular component; After obtaining the preliminary diffuse reflection image, performing inverse rendering correction according to the incident light direction, material prior and ambient light intensity parameter to restore the surface reflection distribution consistent with the illumination, to generate a net texture diffuse reflection image; At the same time, based on the corresponding relationship between the reflectivity and transmissivity of each polarization angle in the separation process, the polarization degree and polarization direction distribution are determined to generate the corresponding polarization map Based on the net texture diffuse reflection image and the polarization map, the fascia rupture feature of the tissue structure continuity is extracted, and the texture consistency is corrected based on the fascia rupture feature to determine the tissue degradation score, to form a freeze-thaw damage sensitive index; Based on the freeze-thaw damage sensitive index, a gate weighted fusion is performed to generate a freeze-thaw damage index vector, which is input into a texture evaluation model to output a texture evaluation result.

2. The digital image processing based meat quality evaluation method according to claim 1, characterized in that, Obtaining a multi-angle / multi-channel image sequence of a meat product sample, comprising: Synchronously collecting reflection images of the meat product sample under multiple incident angles and / or multiple polarization angles in a preset shooting device, and capturing texture information of different wave bands through multiple spectral channels; Spatially registering and time-synchronizing the collected image sequence according to the angle and channel index to generate multi-angle / multi-channel image sequence data for representing the optical response characteristics of the sample surface.

3. The digital image processing based meat quality evaluation method according to claim 1, characterized in that, Completing image calibration based on a gray card and / or a polarizing plate reference image to obtain a calibrated sequence for reflection component separation, comprising: Determining the brightness response curve and color mapping matrix of each channel based on the gray card reference image, and sequentially performing brightness linearization and color uniformity correction on the multi-channel image sequence; Extracting the transmission direction and extinction ratio parameters corresponding to each polarization angle based on the polarizing plate reference image, establishing a polarization direction correction matrix to compensate for the polarization angle error and polarization intensity attenuation; After completing the linearization, color uniformity correction and / or compensation, performing spatial geometric registration and view consistency optimization on the multi-angle / multi-channel image to generate a calibrated sequence for reflection component separation.

4. The digital image processing-based meat quality evaluation method according to claim 1, characterized in that, Extracting the fascia rupture feature of the tissue structure continuity based on the net texture diffuse reflection image and the polarization map, comprising: Constructing a local directional structure tensor in the net texture diffuse reflection image to obtain a texture main direction field, and using phase consistency detection to identify structure discontinuity regions as fascia rupture candidate regions; In combination with the polarization degree distribution reflecting the optical anisotropy of the tissue in the polarization mapping, polarization consistency screening is performed on the candidate region to eliminate false fractures caused by light or water film reflection; In the screened region, the main connected path of the fascia is reconstructed based on the skeletonization and minimum spanning tree constraints, the fracture proportion, fracture length and direction distribution parameters are calculated, and the fascia fracture features for characterizing the continuity of the tissue structure are obtained.

5. The digital image processing based meat quality assessment method according to claim 4, characterized in that, Based on the fascia fracture features, the texture consistency is corrected to determine the tissue degradation score, and a freeze-thaw damage sensitive index is formed, including: The fascia fracture features are mapped to the corresponding texture main direction field, the local texture consistency index is directionally corrected and continuity compensated, and a corrected texture consistency map is generated; In the corrected texture consistency map, the texture energy attenuation rate and the anisotropy change rate are calculated to reflect the degree of structural degradation caused by the fracture, and a tissue degradation score is obtained; The tissue degradation score is normalized with the polarization degree features and the diffuse reflectance intensity features in the polarization mapping to form a freeze-thaw damage sensitive index for characterizing the degree of freeze-thaw damage.

6. The digital image processing based meat quality assessment method according to claim 5, characterized in that, In the corrected texture consistency map, the texture energy attenuation rate and the anisotropy change rate are calculated to reflect the degree of structural degradation caused by the fracture, and a tissue degradation score is obtained, including: Multi-scale Gabor filtering is performed on the corrected texture consistency map to extract the texture energy distribution at different spatial frequencies, and the falling rate of the energy mean value at each scale is calculated as the texture energy attenuation rate; The principal direction eigenvalue ratio is calculated based on the structure tensor to obtain the anisotropy index, and the anisotropy difference before and after the fracture is calculated in the local neighborhood as the anisotropy change rate; The texture energy attenuation rate and the anisotropy change rate are weighted and fused according to the regional weight to obtain a tissue degradation score reflecting the comprehensive effect of texture direction stability and texture energy loss.

7. The digital image processing based meat quality evaluation method according to claim 1, characterized in that, Based on the freeze-thaw damage sensitive index, a freeze-thaw damage index vector is generated by gated weighted fusion, and a texture evaluation model is input to output a texture evaluation result, including: The freeze-thaw damage sensitive index is divided into texture feature components, polarization feature components and reflectance intensity components according to the source, and is input into the gated weighting module to calculate the channel weight coefficient; Based on the weight coefficient, each component is normalized and weighted to generate a freeze-thaw damage index vector reflecting the comprehensive effect of structural degradation, water migration and optical change; The freeze-thaw damage index vector is input into the texture evaluation model, the tenderness value and the elasticity value are calculated through the regression parameters obtained by training, and the texture comprehensive score is determined by the weighted combination of the two; At the same time of outputting the texture comprehensive score, the variance estimation and confidence interval inference inside the texture evaluation model are used to generate the corresponding uncertainty result, forming a complete texture evaluation result.

8. The digital image processing based meat quality assessment method according to claim 7, characterized in that, The texture evaluation model is constructed based on supervised learning, and the construction rule is: The tenderness value and the elasticity value labeled by artificial labeling are used as the supervised label, the freeze-thaw damage index vector is used as the input feature, and the mapping relationship between the feature and the texture label is established through a multi-layer fully connected neural network to obtain the texture evaluation model; wherein, In the training stage, a weighted least square loss function is adopted to balance the tenderness and elasticity regression error, and a regularization constraint is introduced; In the inference stage, the regression mean value based on the model output is used to determine the texture comprehensive score, and the variance or Dropout sampling result based on the model output is used to calculate the uncertainty interval.

9. A meat product quality evaluation system based on digital image processing, characterized by, The system is used to execute the meat texture evaluation method based on digital image processing in any one of claims 1-8, and the system comprises: An acquisition unit is configured to acquire a multi-angle / multi-channel image sequence of a meat sample, and complete image calibration based on a gray card and / or a polarizing plate reference image to obtain a calibrated sequence for reflection component separation; A correction unit is configured to perform specular and diffuse reflection joint unmixing and inverse rendering correction on the calibrated sequence based on meat refractive index and water film thickness priors to obtain a net texture diffuse reflection image and a polarization map; A feature extraction unit is configured to extract fascia rupture features based on the net texture diffuse reflection image and the polarization map, and correct texture consistency based on the fascia rupture features to determine a tissue degradation score, forming a freeze-thaw damage sensitive indicator; An output unit is configured to execute a gated weighted fusion based on the freeze-thaw damage sensitive indicator to generate a freeze-thaw damage indicator vector, and input the texture evaluation model to output a texture evaluation result.

Images