A machine vision-based automatic packaging detection method for spiced chicken

Abstract

The application discloses a kind of based on machine vision's automatic packaging detection method of chicken, belong to food processing automation and quality detection field.It includes obtaining packaging environment parameter to generate compensation parameter;Optimized imaging data is generated based on compensation parameter adjustment imaging unit;Optimized imaging data is executed multispectral feature extraction to generate defect atlas;Product morphology and foreign matter are generated based on atlas and preset database analysis form and confirmation mark;Fusion determination result and confirmation mark generate comprehensive diagnostic instruction.The application adopts environmental parameter compensation imaging, multispectral feature analysis, biomechanical form matching and thermophysical foreign matter confirmation etc., can comprehensively detect packaging seal, content form and micro-foreign matter etc.Various quality problems, significantly improve the accuracy, reliability and anti-interference ability of automatic packaging detection under complex production environment.

Description

A Machine Vision-Based Automated Packaging Inspection Method for Braised Chicken Technical Field

[0001] This invention relates to the field of food processing automation and quality inspection, and in particular to an automated packaging inspection method for braised chicken based on machine vision. Background Technology

[0002] Pre-packaged cooked food products such as braised chicken have strict quality control requirements on automated production lines, including packaging integrity, contents appearance, and the presence of foreign objects. The quality inspection at the end of the production line is crucial. Currently, this process relies heavily on manual sampling or routine checks using basic machine vision systems to identify obvious defects such as severe damage to packaging bags and whether the seals are properly closed.

[0003] However, existing detection methods generally suffer from low accuracy and efficiency, especially on high-speed production lines, where the real-time performance and stability of detection face even greater challenges. Manual inspection is not only slow and costly, but also susceptible to the subjective judgment and fatigue of inspectors, making it difficult to guarantee consistency and achieve full inspection. Basic vision technology has limited ability to distinguish between subtle seal defects, tiny foreign objects hidden in product folds, and permissible morphological changes within the packaging itself, and genuine defects. Furthermore, detection systems are easily affected by common factors such as packaging surface reflections, oil stains, and moisture condensation. Changes in lighting in the production environment or minor vibrations of the conveyor belt can further reduce the reliability of detection, leading to missed detections of potentially defective products or misjudgments of qualified products. This not only affects production efficiency and product pass rates but may also pose risks to food safety. Summary of the Invention

[0004] To address the aforementioned problems, this invention provides an automated packaging inspection method for braised chicken based on machine vision. This method combines environmental parameter compensation-optimized imaging, multispectral feature extraction, and targeted morphology and foreign object analysis. This effectively suppresses environmental interference, enhances the comprehensive detection capabilities for packaging seals, contents morphology, and minute foreign objects, and significantly improves the accuracy and reliability of automated packaging inspection.

[0005] The above objectives can be achieved through the following approach:

[0006] A machine vision-based automated packaging inspection method for braised chicken includes: acquiring real-time vibration parameters and temperature gradient parameters of the braised chicken packaging on a conveyor belt to generate real-time environmental compensation parameters; dynamically adjusting the exposure time and supplementary lighting mode of the camera unit based on the real-time environmental compensation parameters to generate optimized imaging data; performing multispectral feature extraction on the optimized imaging data to generate a sealing defect map and a foreign object distribution map; matching the deformation parameters in the sealing defect map based on a preset biomechanical feature parameter library to generate a chicken body morphology compliance judgment result; generating a dynamic foreign object confirmation mark based on the coupling relationship between the thermal expansion coefficient inversion value of the foreign object distribution map and the temperature gradient parameters; and fusing the chicken body morphology compliance judgment result and the dynamic foreign object confirmation mark to generate a comprehensive packaging integrity diagnostic command.

[0007] Optionally, generating optimized imaging data includes: calculating a baseline exposure time based on the vibration frequency component in the real-time environmental compensation parameters, and superimposing the influence factor of the temperature gradient parameters to generate an exposure adjustment coefficient; triggering an alternating illumination mode of the near-infrared light channel and the visible light channel based on the exposure adjustment coefficient to generate optimized imaging data.

[0008] Optionally, the multispectral feature extraction of the optimized imaging data further includes: extracting the infrared absorption difference spectrum of the packaging film in the near-infrared light channel to generate a thermally induced deformation vector field; constructing surface texture depth convolution features in the visible light channel to generate a set of foreign object contour coordinates after condensate interference suppression; and combining the spatiotemporal alignment features of the thermally induced deformation vector field and the foreign object contour coordinate set to generate a vapor artifact isolation index.

[0009] Optionally, the generation of chicken body morphology compliance determination results includes: segmenting biological texture feature curves from the sealing defect atlas, extracting leg joint bending angle sequences and sternal curvature fluctuation parameters; calling a preset elasticity compensation database to perform meat elasticity correction on the sternal curvature fluctuation parameters to generate a standardized morphological feature vector; and performing distance metric matching between the standardized morphological feature vector and the standard anatomical template in the biomechanical feature parameter library to generate chicken body morphology compliance determination results.

[0010] Optionally, the generation of chicken body shape compliance determination results further includes: when the chicken body shape compliance determination result indicates that the joint angle deviates, activating the pressure tactile sensor to collect the meat rebound displacement; using the meat rebound displacement to invert the muscle stress relaxation curve and generate a dynamic shape correction factor; feeding the dynamic shape correction factor back to the standardized shape feature vector generation stage to update the preset elastic compensation database.

[0011] Optionally, generating the dynamic foreign object confirmation marker further includes: analyzing the spatial proximity between the abnormal morphological region determined by the chicken body morphology compliance judgment result and the foreign object location determined by the high-probability foreign object coordinates of the marker; and adjusting the confidence level and risk level of the dynamic foreign object confirmation marker based on the spatial proximity result.

[0012] Optionally, generating the dynamic foreign object confirmation marker includes: injecting the vapor artifact isolation index into a preset thermo-mechanical coupling criterion model to calculate the foreign object attachment confidence threshold; when the confidence of the foreign object contour coordinate set exceeds the foreign object attachment confidence threshold, generating the dynamic foreign object confirmation marker by marking high-probability foreign object coordinates, and outputting a packaging interruption signal.

[0013] Optionally, the generated packaging integrity comprehensive diagnostic command further includes: calculating the packaging stress concentration area based on the spatial distribution density of the dynamic foreign object confirmation mark; controlling the backlight unit to project a color temperature gradient light spot of a preset wavelength onto the stress concentration area to generate a stress visualization warning spectrum; and coupling the stress visualization warning spectrum with the conveyor belt speed control command in a time sequence to trigger an adaptive deceleration mechanism.

[0014] Optionally, the triggering adaptive deceleration mechanism includes: calculating the predicted value of the expansion trend of the stress concentration area and generating a pressure risk index by associating it with the weight data of the same batch of packaging; when the pressure risk index exceeds a preset composite threshold, adjusting the conveyor belt running speed according to the exponential decay curve, and synchronously updating the anatomical template weight coefficient in the biomechanical characteristic parameter library.

[0015] Optionally, the method further includes: monitoring the feature space continuity of the optimized imaging data; recording fault phase parameters when a sudden signal distortion is detected; acquiring historical operating data and constructing an association mapping table between the fault phase parameters and the real-time environmental compensation parameters based on the historical operating data to generate an anti-interference plan set; and selecting a combination of compensation parameters from the anti-interference plan set according to the temperature gradient parameters to autonomously reconstruct the imaging data generation link.

[0016] Compared with the prior art, the present invention has the following advantages:

[0017] 1. Improved detection accuracy and anti-interference capability in complex environments. This invention optimizes the imaging process through real-time environmental parameter compensation. Combined with multispectral information analysis, it effectively suppresses interference from production line vibration and packaging reflections and moisture. This significantly improves the accuracy of identifying minute foreign objects and complex sealing defects, reducing false positives and missed detections.

[0018] 2. This invention enables more comprehensive inspection of packaging condition. It not only performs foreign object and seal checks, but also introduces a method for determining compliance with internal product form based on biomechanical characteristics. The inspection dimensions and coverage are significantly expanded compared to traditional technologies, enabling the discovery of more potential quality issues.

[0019] 3. Enhanced dynamic adaptability and intelligence of detection. This invention includes a feedback correction mechanism for product shape determination. It features adaptive speed control of the conveyor belt based on real-time risk assessment. Its fault self-diagnosis and autonomous reconstruction capabilities of the imaging link further ensure long-term stable operation under different working conditions.

[0020] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 is a flowchart illustrating an automated packaging and inspection method for braised chicken based on machine vision according to an embodiment of the present invention.

[0023] Figure 2 is a timing diagram of alternating NIR / VIS irradiation according to an embodiment of the present invention.

[0024] Figure 3 is a comparison diagram of the sternal curvature of an embodiment of the present invention and that of a standard template.

[0025] Figure 4 is a density diagram of the packaging stress concentration area according to an embodiment of the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] Referring to Figure 1, one embodiment of the present invention proposes an automated packaging inspection method for braised chicken based on machine vision. It adopts a combination of environmental parameter compensation optimization imaging, multispectral feature extraction, and targeted morphology and foreign object analysis, which can effectively suppress environmental interference, improve the comprehensive detection capability of packaging seal, contents morphology and small foreign objects, and significantly improve the accuracy and reliability of automated packaging inspection.

[0028] The method described in this embodiment specifically includes:

[0029] Obtain real-time vibration and temperature gradient parameters of the braised chicken packaging on the conveyor belt, and generate real-time environmental compensation parameters;

[0030] Specifically, the aim is to perceive the real-time physical environment during packaging inspection. Vibration parameters can be vibration frequency and amplitude data monitored in real time by accelerometers or similar devices, used to quantify the physical instability at the moment of imaging. Temperature gradient parameters can be obtained by monitoring the temperature of the packaging or its surrounding environment and calculating its rate of change over time, used to assess the possibility of water vapor condensation or material thermal response. Based on these real-time perceived parameters, one or a set of real-time environmental compensation parameters are calculated and generated. These parameters can be indicators that comprehensively reflect the current level of disturbance, or they can be coefficient values ​​directly used for subsequent camera unit adjustments, providing a real-time basis for dynamically optimizing imaging.

[0031] Based on the real-time environmental compensation parameters, the exposure time and fill light mode of the camera unit are dynamically adjusted to generate optimized imaging data;

[0032] Specifically, the acquired environmental compensation parameters are used to proactively optimize imaging conditions to obtain higher-quality images. The camera unit, such as an industrial camera, dynamically adjusts its exposure time based on these compensation parameters. When high vibrations are detected, the exposure time can be shortened to reduce motion blur. The supplementary lighting mode involves intelligent control of the light source, adjusting its illumination intensity, timing, or spectral characteristics according to the compensation parameters to suppress interference from factors such as packaging reflections or surface moisture. Through this environmentally aware dynamic adjustment, optimized imaging data with higher clarity and less interference can be generated.

[0033] Multispectral feature extraction is performed on the optimized imaging data to generate sealing defect maps and foreign object distribution maps;

[0034] Specifically, key information related to packaging quality is extracted from the optimized image data. Multispectral feature extraction utilizes the different information that different spectral channels in the image data may carry. Through image processing and analysis algorithms, including edge detection, texture analysis, and absorption or reflectance analysis, features indicating potential problems are identified and separated. Two types of key maps are generated after processing: first, a sealing defect map, which highlights or marks the possible abnormal locations and morphological features in the packaging sealing area. These anomalies may include microchannels, uneven compression, or contamination, among other things. Second, a foreign matter distribution map, which indicates the location and basic outline of potential foreign objects inside or on the surface of the packaging.

[0035] Based on a preset biomechanical feature parameter library, the deformation parameters in the sealing defect map are matched to generate a chicken body morphology compliance judgment result.

[0036] Specifically, the assessment evaluates whether the shape of the braised chicken product within the packaging conforms to the standards. A pre-set biomechanical characteristic parameter library stores key morphological and physical characteristic parameters of standard qualified braised chicken. This information can cover the size range of major parts, the normal bending angle range of joints, and the allowable elastic deformation range, collectively forming a series of standard templates. During testing, characteristic parameters reflecting the actual shape of the braised chicken under test are extracted from optimized imaging data; these are referred to here as deformation parameters, such as actual joint angles or contour curvature obtained through image segmentation. These extracted actual parameters are compared with the standard templates in the parameter library using a specific matching algorithm. Based on the comparison results, it is determined whether the product shape is within the acceptable range, ultimately generating a compliance judgment result for the chicken body shape. This result can be a simple pass or fail status, or it can include more specific diagnostic information, such as indicating joint angle deviations.

[0037] Based on the coupling relationship between the inverted value of the thermal expansion coefficient of the foreign object distribution map and the temperature gradient parameter, a dynamic foreign object confirmation marker is generated;

[0038] Specifically, to distinguish between genuine physical foreign objects and interfering substances, such as oil stains, water droplets, or the product's own tissue, this step utilizes the different thermophysical properties that different substances may possess. By analyzing heat-related information in the imaging data and combining it with the temperature gradient parameters of the current environment, the thermal response behavior of suspected areas on the foreign object distribution map can be evaluated. This behavior is compared with the expected thermal responses of known common foreign objects or interfering substances; for example, the thermal responses of metal fragments and water droplets are typically different. By analyzing the coupling relationship between this thermal response behavior and the temperature gradient, the confirmation accuracy of genuine foreign objects can be improved, and dynamic foreign object confirmation markers with high confidence can be generated.

[0039] By combining the chicken body morphology compliance determination result with the dynamic foreign object confirmation mark, a comprehensive diagnostic instruction for packaging integrity is generated.

[0040] Specifically, the obtained morphological compliance judgment results are fused and evaluated with the foreign object confirmation mark. This fusion process can be based on preset logical rules; a simple example is that any confirmed foreign object detected is deemed non-compliant. More complex decision models can also be used for fusion. The ultimate goal is to output a clear and unambiguous comprehensive diagnostic instruction for packaging integrity. This instruction summarizes the conclusion of this inspection, such as compliant or non-compliant, and specifies the reason as the presence of foreign objects or morphological abnormalities. This instruction can be used to trigger subsequent automated processing, such as controlling the rejection device to remove non-compliant products, or recording defect information in the quality management system.

[0041] By real-time environmental perception, optimized imaging, multispectral feature extraction, and combined with targeted morphological conformity determination and physical property-based foreign object confirmation, a comprehensive diagnostic command is finally fused together, forming the basic process of this invention, which can effectively improve the accuracy and reliability of automated packaging inspection of braised chicken.

[0042] Optionally, generating optimized imaging data includes:

[0043] The baseline exposure time is calculated based on the vibration frequency component in the real-time environmental compensation parameters, and the exposure adjustment coefficient is generated by superimposing the influence factor of the temperature gradient parameters.

[0044] Specifically, an adjustment coefficient that incorporates environmental influences is used to guide imaging parameters. First, frequency components representing the main vibration characteristics of the current device or environment are extracted from the acquired real-time environmental compensation parameters. Simultaneously, the impact factor of quantifying the potential interference of temperature changes on imaging is calculated based on the magnitude and trend of the real-time temperature gradient parameters. Based on a preset or product-type-adjusted baseline exposure time, the exposure adjustment coefficient is calculated by combining the vibration frequency components and the temperature gradient impact factor. This calculation process can be implemented according to specific strategies; an exemplary calculation formula is as follows:

[0045] K exp =T base *(1-α*tanh(κ*f vib ))*(1+β*F temp ),

[0046] In the formula, K exp T is the exposure adjustment factor. base As the baseline exposure time, f vib For the detected main vibration frequency components, F temp α is the temperature gradient influence factor, κ and α are adjustment parameters related to vibration suppression, and β is the weighting coefficient of the temperature gradient influence.

[0047] Based on the exposure adjustment coefficient, an alternating illumination mode between the near-infrared light channel and the visible light channel is triggered to generate optimized imaging data.

[0048] Specifically, illumination control and image acquisition are performed based on the calculated exposure adjustment factor. This exposure adjustment factor value is used to trigger and regulate a special illumination mode involving alternating or coordinated illumination of the near-infrared (NIR) and visible light (VIS) channels, as shown in Figure 2. For example, the value of this exposure adjustment factor can determine the duration, relative intensity, or illumination sequence of the near-infrared and visible light sources within a complete imaging acquisition cycle. By precisely controlling this alternating illumination mode and the possible accompanying synchronous adjustment of camera exposure time, the final acquired image data is optimized, effectively suppressing motion blur and moisture interference.

[0049] For example, assuming a baseline exposure time of 10ms, and relevant adjustment parameters set to α = 0.3, κ = 0.02, and β = 0.2. If a vibration frequency component of 60Hz is detected at a certain moment, and the calculated temperature gradient influence factor is 0.5, the calculated exposure adjustment coefficient is approximately 8.25ms. Imaging is then performed based on this exposure adjustment coefficient. For example, within a total time of approximately 8.25ms, the near-infrared light source and the visible light source are turned on sequentially according to specific logic, and images are acquired. Finally, a frame of optimized imaging data containing dual-channel information is output, effectively suppressing blur and moisture interference.

[0050] Optionally, performing multispectral feature extraction on the optimized imaging data further includes:

[0051] The infrared absorption difference spectrum of the packaging film is extracted under the near-infrared light channel to generate a thermally induced deformation vector field;

[0052] Specifically, this step utilizes near-infrared image information from optimized imaging data. By comparing the actual absorption intensity of each pixel in the packaging film area of ​​the near-infrared image with the pre-stored standard near-infrared absorption characteristics of the packaging material, the difference value is calculated, generating an infrared absorption difference spectrum. Furthermore, by analyzing the changes in this difference spectrum over a short time series, especially the pixel intensity change patterns related to temperature field changes, the distribution of minute thermally induced deformations on the packaging film surface can be estimated and represented as a thermally induced deformation vector field.

[0053] Surface texture depth convolution features are constructed in the visible light channel to generate a set of foreign object contour coordinates after condensation water interference suppression.

[0054] Specifically, this step utilizes visible light image information from optimized imaging data, focusing on surface feature analysis. Deep learning techniques, specifically a Convolutional Neural Network (CNN) model, are employed to analyze and extract texture features from the packaging surface. This CNN model is trained to distinguish the textures of packaging materials, the product itself, and potential foreign objects. The model is designed and trained to suppress common visual disturbances such as condensation droplets or water stains. After processing the visible light image, the CNN model outputs a set of interference-suppressed coordinates of the suspected foreign object, labeled with its location and contour information. Each coordinate or contour in this set may also be accompanied by a confidence score output by the CNN model.

[0055] By combining the spatiotemporal alignment features of the thermally induced deformation vector field and the foreign object contour coordinate set, a vapor artifact isolation index is generated.

[0056] Specifically, this step aims to fuse information from different dimensions of near-infrared and visible light channels to more reliably distinguish between real foreign objects and visual artifacts. The thermally induced deformation vector field and foreign object contour coordinate set generated in the first two steps are spatiotemporally aligned to ensure that the analysis focuses on feature information corresponding to the same spatial location and similar time points, resulting in spatiotemporally aligned features that can be used for combined analysis. Then, a vapor artifact isolation index is calculated based on these aligned features. This index quantifies the probability that a suspected region marked by the foreign object contour coordinate set is a stable physical entity or a transient artifact. An exemplary calculation formula is as follows:

[0057]

[0058] In the formula, I artifact f(C) is the vapor artifact isolation index. obj_confidence ) represents the contribution function of the confidence level of the foreign object coordinate set. A represents the magnitude of the divergence of the thermally induced deformation vector field within the foreign object coordinate region, reflecting the deformation activity. align Representing the spatiotemporal alignment feature, g and h are non-negative monotonically increasing functions, and γ is the weighting coefficient. The higher the exponent value, the lower the probability that the suspected foreign object is an artifact.

[0059] For example, suppose a visible light CNN detects a suspected foreign object contour with a confidence contribution function value of 0.9. Corresponding near-infrared analysis shows that the divergence contribution function of the thermally induced deformation vector field in this region is 0.1. Assume the spatiotemporal alignment feature contribution function is 1, and the weighting coefficient is 5. Using the aforementioned example formula, the vapor artifact isolation index is calculated to be 0.6. This relatively high index value increases the confidence that the detection result is a real foreign object rather than a vapor artifact, providing important evidence for subsequent foreign object confirmation steps.

[0060] Optionally, the generated chicken body morphology compliance determination result includes:

[0061] Biological texture feature curves were segmented from the sealing defect map, and leg joint bending angle sequences and sternal curvature fluctuation parameters were extracted.

[0062] Specifically, this step aims to accurately quantify the key morphological features of the braised chicken from the image. Image segmentation techniques, such as active contour models or deep learning-based segmentation networks, can be used to segment the main region of the braised chicken inside the packaging from the optimized imaging data. Within the segmented region, texture analysis algorithms, such as those based on Gabor filters or Local Binary Patterns (LBP), are applied to extract biological texture feature curves that characterize the direction or distribution of muscle texture. Simultaneously, key anatomical landmarks of the braised chicken are identified using image skeletonization algorithms or machine learning-based keypoint detection models, such as the corresponding points of the hip, knee, and ankle joints in the legs, and the contour points of the sternum. Based on these landmarks, a series of quantified morphological parameters are calculated. These parameters may include a leg joint flexion angle sequence, which is a set of values ​​describing the angles of one or more key leg joints, and a sternal curvature fluctuation parameter. This parameter can be obtained by performing polynomial fitting or other curve analysis methods on the extracted sternal contour curve to calculate statistics such as the root mean square error (RMSE) or variance of curvature change compared to the standard template curve, in order to quantify the regularity of the sternal shape.

[0063] The preset elastic compensation database is invoked to perform meat elasticity correction on the sternal curvature fluctuation parameters, generating a standardized morphological feature vector;

[0064] Specifically, considering that different production batches, processing techniques, or temperature and humidity environments may lead to variations in the elasticity of braised chicken, thus affecting the measured values ​​of morphological parameters, this step introduces elasticity compensation correction. An elasticity compensation database needs to be pre-built. This database can be obtained through extensive experimental measurements and stores a model showing the relationship between changes in meat elasticity and measured values ​​of key morphological parameters under different observable conditions. Observable conditions include product temperature, storage time, and extracted partial texture features, while key morphological parameters include sternal curvature fluctuation parameters. This can be a lookup table, a regression model, or a small neural network. During detection, relevant information for the current sample is input into this database or model to obtain an elasticity compensation factor or correction amount for the current sample. This factor or correction amount is then applied to correct the originally extracted sternal curvature fluctuation parameters to obtain corrected curvature parameters that better reflect the inherent skeletal morphology. An exemplary correction relationship is as follows:

[0065] C corrected =C raw *F elastic ,

[0066] In the formula, C corrected For the corrected sternal curvature fluctuation parameter, C raw F represents the original extracted sternal curvature fluctuation parameter. elastic The elastic compensation factor is obtained by querying or calculating from the elastic compensation database based on current relevant conditions. Then, the extracted leg joint flexion angle sequence, the corrected sternal curvature fluctuation parameter, and other potentially extracted and corrected morphological feature parameters are combined. To eliminate the influence of different feature parameter dimensions and numerical ranges, this combined vector needs to be standardized. For example, Z-score standardization can be used to make its mean 0 and standard deviation 1, or Min-Max method can be used to scale it to a standard interval of [0,1] or [-1,1], ultimately forming a standardized, multi-dimensional morphological feature vector for subsequent matching and discrimination.

[0067] The standardized morphological feature vector is matched with the standard anatomical template in the biomechanical feature parameter library by distance measurement to generate a chicken body morphology compliance judgment result.

[0068] Specifically, the standardized morphological feature vector representing the current morphology of the tested braised chicken is compared with a standard anatomical template stored in a pre-set biomechanical feature parameter library. This parameter library is built on a large amount of qualified sample data and contains one or more feature vector templates representing the ideal or standard morphology of braised chicken. The comparison process is carried out in the standardized feature space, using a suitable distance metric to quantify the degree of difference between the tested sample vector and the standard template vector, as shown in Figure 3. Commonly used distance metrics can be Euclidean distance, which is simple and intuitive to calculate, or Mahalanobis distance, which takes into account the correlation between features. The choice of distance metric may depend on the specific features of each dimension in the feature vector. The purpose of calculating the distance is to obtain a quantitative morphological deviation value. Taking Euclidean distance as an example, its calculation formula is:

[0069]

[0070] In the formula, D is the calculated Euclidean distance, and V sample V is the standardized morphological feature vector of the tested sample. template Let 'i' represent the feature vector of the standard anatomical template, and 'n' represent the total dimension of the feature vector. Finally, the calculated distance value is compared with a pre-set morphological conformity threshold based on product quality requirements. If the calculated distance value is less than or equal to the threshold, it indicates that the tested sample's morphology is sufficiently close to the standard template, and the roasted chicken is deemed compliant, generating a "qualified" result. Conversely, if the calculated distance value is greater than the threshold, it is deemed non-compliant, and a more specific judgment result can be generated based on the contribution of each dimension's features during distance calculation, such as explicitly indicating "joint angle deviation" or "abnormal sternal curvature."

[0071] For example, suppose that the leg joint flexion angle sequence and the original sternal curvature fluctuation parameters of a sample are obtained through image segmentation and feature extraction. After correction using an elasticity compensation database, a standardized morphological feature vector of [0.6, 0.85] is generated. The corresponding standard anatomical template vector of [0.5, 0.8] is found in the biomechanical feature parameter library. The difference between the two is calculated using Euclidean distance, and the distance value is approximately 0.112. If the preset morphological compliance threshold is 0.15, since the calculated distance is less than or equal to the threshold, the final chicken body morphological compliance judgment result is "qualified".

[0072] Optionally, the generated chicken body morphology compliance determination result also includes:

[0073] When the chicken body shape compliance determination result indicates that the joint angle deviates, the pressure tactile sensor is activated to collect the meat rebound displacement.

[0074] Specifically, the trigger is activated only when the chicken body morphology compliance assessment clearly indicates a specific non-compliance condition of "joint angle deviation." Upon receiving this trigger signal, an actuator equipped with a pressure tactile sensor, such as the end effector of a small robotic arm or a specialized probe, is precisely moved to the outer area of ​​the packaging where the joint with the determined angle deviation is located. Subsequently, the pressure tactile sensor applies a brief, standardized pressure to the packaging surface according to a preset program. This pressure can be a rapid impact or a brief constant pressure. During this process and for a short period after the pressure is released, the sensor continuously measures and records the displacement changes of the sensor probe or the packaging surface caused by the deformation and rebound of the underlying meat under pressure. Among these measurements, the key measurement data reflecting the elasticity and viscosity characteristics of the meat, namely the amount of meat rebound displacement, is extracted for subsequent analysis.

[0075] The muscle stress relaxation curve is inverted by the meat rebound displacement, and a dynamic morphology correction factor is generated.

[0076] Specifically, by utilizing the data sequence of meat rebound displacement over time, combined with known sensor force parameters and possible mechanical property models of packaging materials, this model is used to decouple the influence of the packaging itself. The stress relaxation characteristics of the muscle tissue near the joint can be inferred through biomechanical model inversion. Commonly used models include, but are not limited to, the Standard Linear Solid (SLS) model or other more complex viscoelastic constitutive models. By fitting the theoretical response curves of these models with the actual measured rebound displacement data, key parameters describing the viscoelastic behavior of the meat in a specific sample and location can be estimated, such as the effective elastic modulus, viscosity coefficient, or characteristic relaxation time. These parameters collectively define the stress relaxation curve of the muscle at that location. Based on these physical parameters inverted from actual measurements, a dynamic morphological correction factor is calculated. This factor aims to quantify the difference between the actual mechanical properties of the current tested sample at that specific location and the standard or average state, for subsequent compensation or database updates. One possible calculation method is to calculate this factor based on the inverted parameters:

[0077] F correct =ω1*E measured +ω2*τ measured ,

[0078] In the formula, F correct E is the dynamic morphological correction factor. measured τ represents the effective elastic modulus of this part of the current sample obtained through model inversion. measuredω1 and ω2 are the stress relaxation time constants of this part of the current sample obtained through model inversion, and are preset weighting coefficients determined based on experience or experiments.

[0079] The dynamic morphological correction factor is fed back to the standardized morphological feature vector generation stage to update the preset elastic compensation database.

[0080] Specifically, the calculated dynamic morphology correction factor is primarily used for feedback learning and optimization. A major application is updating the pre-defined elasticity compensation database. The dynamic morphology correction factor obtained from the current measurement, along with its associated contextual information such as the current batch number, ambient temperature, specific joint location measured, and original angular deviation, can be used as a new data instance or evidence. This new data is then used to update the relational model or lookup table stored within the elasticity compensation database. The update method can employ various online learning or adaptive algorithms. For example, a simple weighted average can be used to adjust the estimated compensation factor values ​​under corresponding conditions in the database. Alternatively, if the database is model-based, methods such as Kalman filtering or recursive least squares (RLS) can be used to iteratively update the model parameters. This continuous feedback update allows the elasticity compensation database to dynamically adapt to the real changes in meat elasticity under different batches and environmental conditions, thereby improving the accuracy of future elasticity compensation steps. Furthermore, in some implementations, the dynamic morphology correction factor may also be used to perform a final fine-tuning of the standardized morphological feature vector of the current sample to obtain a more accurate final judgment result, but this is usually not its primary feedback path.

[0081] For example, suppose the result of the braised chicken package number SN001 is "unqualified - left knee joint angle deviation". This optional process is triggered. The pressure tactile sensor moves to the packaging position corresponding to the left knee joint, applies pressure and measures the rebound, obtaining a set of displacement data. Through viscoelastic model fitting, the effective elastic modulus and relaxation time constant of the meat at that location are obtained. Based on these parameters and preset weights, a dynamic morphology correction factor value of 1.08 is calculated. Assuming the standard is 1.0, this indicates that the sample is slightly harder or has slightly better elasticity at that location. The current ambient temperature is recorded as 15℃ and the batch number is BATCH05. Then, using the factor value of 1.08, the temperature of 15℃, and the batch number BATCH05, the corresponding compensation model parameters or lookup table entries in the elasticity compensation database are updated, making the elasticity compensation more accurate when the BATCH05 batch is tested in an environment of around 15℃.

[0082] Optionally, the generation of the dynamic foreign object confirmation marker includes:

[0083] The vapor artifact isolation index is injected into a preset thermo-mechanical coupling criterion model to calculate the foreign object adhesion confidence threshold.

[0084] Specifically, this step aims to dynamically set a standard for confirming a real foreign object by evaluating an index to determine whether a suspected target is a vapor artifact. A pre-defined thermo-mechanical coupling criterion model is invoked. This model can be a machine learning model, such as a Support Vector Machine (SVM) or decision tree, or a complex rule system built based on expert knowledge. The model receives a generated vapor artifact isolation index as key input, reflecting the thermal stability or non-artifact characteristics of the suspected target. The model may also incorporate current ambient temperature parameters or other relevant contextual information. The "thermo-mechanical coupling" characteristic of the model is reflected in its comprehensive evaluation of the target's quasi-thermal stability and its relative probability of physical adhesion to the product or packaging, ultimately outputting a dynamically calculated foreign object adhesion confidence threshold. This threshold is generated in real-time for the currently evaluated target, representing the required self-detection confidence level under the current evidence to confirm it as a potentially "attached" or real foreign object of concern. An exemplary threshold calculation relationship might be as follows:

[0085] T adhesion =T base_adhesion *(1+δ*I artifact ),

[0086] In the formula, T adhesion T is the calculated foreign object adhesion confidence threshold. base_adhesion I is a basic threshold set according to product and testing requirements. artifact δ is the input vapor artifact isolation index, and δ is a preset adjustment coefficient used to adjust the intensity of the isolation index's influence on the final threshold.

[0087] When the confidence level of the foreign object contour coordinate set exceeds the foreign object attachment confidence threshold, the dynamic foreign object confirmation mark is generated by marking high-probability foreign object coordinates, and a packaging interruption signal is output.

[0088] Specifically, this step performs the final foreign object confirmation decision. A confidence score is generated along with the foreign object contour coordinate set. This score reflects the detection algorithm's confidence that the contour belongs to an independent target. This confidence score is compared with a dynamically calculated foreign object attachment confidence threshold for that target. If the confidence score is greater than the foreign object attachment confidence threshold, it indicates that the target is not only clearly detected, but also has a sufficiently high probability of being a real foreign object according to the thermo-mechanical coupling model. Therefore, the foreign object is determined to be a high-probability real foreign object. At this point, a confirmation action is performed, generating a dynamic foreign object confirmation mark by marking the high-probability foreign object coordinates. This typically means marking the state associated with the foreign object coordinates as "confirmed foreign object" in the internal data. This mark itself or its generation event is the "dynamic foreign object confirmation mark." Simultaneously, as a necessary linkage measure, a packaging interruption signal is immediately output. This signal is a digital or logical signal sent to the production line control system, such as a Programmable Logic Controller (PLC), or directly to the actuator of the rejection device to trigger the interception, rejection, or other predetermined processing procedures for the currently non-conforming packaging. If the confidence score does not exceed the calculated dynamic threshold, the suspected foreign object will not be finally confirmed, no confirmation mark will be generated, and no interruption signal will be output.

[0089] Optionally, the generation of the dynamic foreign object confirmation marker further includes:

[0090] Analyze the spatial proximity between the abnormal morphological regions determined by the chicken body morphology compliance judgment results and the foreign object locations determined by the marked high-probability foreign object coordinates;

[0091] Specifically, the aim is to correlate and analyze the spatial relationship between morphological issues of the product itself and detected foreign objects. This involves acquiring definite morphological anomaly information indicating a specific location or region, as well as the coordinates of preliminarily confirmed high-probability foreign objects, also carrying location information. Then, their spatial proximity is assessed by calculating the spatial distance between these two or determining whether they fall within each other's predefined neighboring areas.

[0092] Based on the spatial proximity results, adjust the confidence level and risk level of the dynamic foreign object confirmation marker.

[0093] Specifically, based on the spatial proximity obtained from the analysis, the confidence level or associated risk level of the dynamic foreign object identification marker is adjusted. If an identified foreign object is spatially adjacent to an area of ​​abnormal shape, the risk it poses to the overall product quality may be greater. Therefore, in such cases of high proximity, the final risk level score of the foreign object marker is increased or its confidence value is adjusted accordingly. This marker attribute, adjusted for proximity context information, will be used to generate more accurate comprehensive diagnostic instructions.

[0094] Optionally, the generated packaging integrity comprehensive diagnostic instruction further includes:

[0095] Calculate the stress concentration area of ​​the packaging based on the spatial distribution density of the dynamic foreign object confirmation marks;

[0096] Specifically, the location information of dynamic foreign object identification markers is used to assess the potential physical stress risk on the packaging. The spatial coordinates of all identification markers currently on the packaging are collected. Then, a spatial density estimation algorithm is used to calculate the spatial distribution density of these marker points; kernel density estimation is an optional technique for achieving this function. This algorithm generates a density map. On this density map, areas with density values ​​exceeding a preset density threshold are identified and determined; these high-density areas are defined as stress concentration areas in the packaging. The physical meaning of this definition is that the accumulation of multiple foreign objects may cause or indicate high local stress in the packaging material or a potential risk of breakage. An exemplary kernel density estimation calculation might involve the following formula:

[0097]

[0098] In the formula, Here, x represents the estimated spatial distribution density at location x, N is the total number of dynamic foreign object identification markers, h is the bandwidth parameter, and x is the distance between x and x. i Let be the spatial coordinates of the i-th marker, and K be the kernel function, a commonly used kernel function being the Gaussian kernel function. By setting an appropriate density threshold to discriminate the calculated spatial distribution density estimate, the packaging stress concentration area can be delineated, as shown in Figure 4.

[0099] The backlight unit is controlled to project a color temperature gradient light spot of a preset wavelength onto the stress concentration area to generate a stress visualization warning spectrum;

[0100] Specifically, a programmable backlight unit located below or to the side of the conveyor belt is controlled. This backlight unit can be an LED array light source capable of independently controlling pixels. It is instructed to precisely project a special light spot onto the identified stress concentration area of ​​the package as it passes through. The key characteristic of this light spot is its color temperature gradient, meaning that the color temperature of the light within the spot gradually changes in space. Color temperature is an indicator describing whether light is warm or cool. This gradient can be a smooth transition from warm light on one side to cool light on the other, and the overall light remains within a preset visible spectrum. When this light with a specific color temperature gradient penetrates or reflects onto the packaging material where stress concentration exists, especially for transparent or translucent plastic films, the birefringence effect that may occur due to stress will observably alter the polarization state or spectral composition of the transmitted or reflected light. This alteration will cause the observed color temperature gradient pattern to be distorted, color-shifted, or exhibit interference fringes compared to the original projected pattern. By capturing the image of the light spot after passing through the stress concentration area of ​​the package using a camera unit, a stress visualization warning map is formed. The camera unit can be a camera that performs the main detection task, or it can be a camera specifically designed for this stress visualization step. This image can visually reflect the non-uniformity of stress distribution.

[0101] The stress visualization warning map is coupled with the conveyor belt speed control command in a time sequence to trigger an adaptive deceleration mechanism.

[0102] Specifically, image analysis is performed on the stress visualization warning spectrum to quantify the stress magnitude or risk level reflected in the spectrum. This is achieved by calculating specific indicators such as the distortion degree of the color temperature gradient pattern, the distribution area of ​​specific colors, or the density of interference fringes. Then, the stress risk assessment results obtained from the analysis are tightly coupled in time with the conveyor belt speed control logic. This means that based on the real-time assessed stress risk level, corresponding conveyor belt speed control commands are generated in a very short time, ensuring that action can be taken before the current package or the immediately following package reaches the next critical processing station. If the analysis results indicate a high risk in the stress concentration area, such as when the calculated stress index exceeds a preset safety threshold, a deceleration or stop command is generated. This command is sent to the conveyor belt drive control system, thereby triggering the adaptive deceleration mechanism. Here, "triggered" refers to issuing a signal to initiate the mechanism.

[0103] Optionally, the triggering adaptive deceleration mechanism includes:

[0104] Calculate the predicted expansion trend of the stress concentration area and generate a pressure risk index by correlating it with the weight data of the same batch of packaging.

[0105] Specifically, analyzing stress concentration areas on packaging may involve continuously observing these areas over a short period, analyzing changes in their size, shape, or strength indicators, and using a time-series forecasting model to calculate a predicted trend for the expansion of the stress concentration area. This predicted value is used to determine whether the stress concentration is rapidly worsening. Simultaneously, it is necessary to obtain the weight data of packaging from the same batch as the currently inspected packaging. This data can come from weighing stations upstream of the production line or be retrieved from the Manufacturing Execution System (MES) based on batch number. Obtaining weight data is intended to be a factor influencing risk assessment, as heavier packaging may generate greater actual physical stress at stress concentration points. Finally, the calculated predicted expansion trend value is integrated with the associated packaging weight data, and a comprehensive stress risk index is generated using a pre-defined risk assessment model or weighted calculation formula.

[0106] When the pressure risk index exceeds the preset composite threshold, the conveyor belt speed is adjusted according to the exponential decay curve, and the anatomical template weight coefficient in the biomechanical characteristic parameter library is updated synchronously.

[0107] Specifically, the pressure risk index is compared with a preset composite threshold representing the upper limit of acceptable risk. If and only if the calculated pressure risk index exceeds this composite threshold, it indicates that the risk is too high and intervention is required. At this point, two parallel actions are performed: first, the conveyor belt speed is adjusted. The conveyor belt speed is reduced according to a preset exponential decay curve model. This means that the speed does not drop to zero instantaneously or to a fixed low speed, but rather decays smoothly and rapidly over time or distance to a lower safe speed or stops completely. An exemplary speed adjustment model is as follows:

[0108]

[0109] In the formula, t is the time calculated from the start of deceleration, v(t) is the instantaneous speed of the conveyor belt at time t, v0 is the speed of the conveyor belt before deceleration is triggered, and v min Here, τ represents the target minimum velocity, e is the base of the natural logarithm, and τ is the exponential decay time constant, which determines the rate of deceleration. Secondly, the biomechanical feature parameter library is updated synchronously. This update is based on currently detected high-pressure risk events, adaptively adjusting the weight coefficients of the standard anatomical templates stored in the library. This constitutes a learning mechanism, suggesting that the current high-pressure risk may be related to the vulnerability of certain morphological features. Therefore, by adjusting the weights of corresponding features in the templates, potential high-risk morphologies can be identified earlier when performing morphological inspections on products under similar batches or conditions.

[0110] For example, suppose the calculated pressure risk index is 0.92, while the preset composite threshold is 0.8. Since the index exceeds the threshold, a control action is triggered. It instructs the conveyor belt speed to decelerate from the current 1 meter per second, following an exponential decay model. Simultaneously, it accesses a biomechanical feature parameter library, finds a standard anatomical template associated with the current product batch, and fine-tunes the weighting coefficients of features related to packaging stress, for example, adjusting the original weight from 1.0 to 1.05, to be more sensitive to such risks in subsequent inspections.

[0111] Optionally, the method further includes:

[0112] Monitor the feature space continuity of the optimized imaging data, and record the fault phase parameters when a sudden signal distortion is detected;

[0113] Specifically, this step involves real-time health monitoring of the imaging process. The imaging data stream is continuously analyzed and optimized, extracting several key indicators within a predetermined feature space. These indicators may include statistical analysis of image brightness distribution, texture complexity, or frequency domain energy distribution. By tracking the changes in these indicator values ​​over time, it is monitored whether they remain within a expected stable range, i.e., monitoring the continuity of the feature space. When one or more indicators exhibit a sharp jump or persistent anomaly exceeding a preset threshold, i.e., a sudden signal distortion is detected, it is determined that a failure or strong interference in the imaging link may have occurred. At this time, fault phase parameters related to this event are recorded. These parameters may include information such as the time of the event, the identified distortion type code, and the abnormal feature indicator values.

[0114] Acquire historical operating data, and construct an association mapping table between the fault phase parameters and the real-time environmental compensation parameters based on the historical operating data to generate a set of anti-interference contingency plans;

[0115] Specifically, this step aims to leverage historical experience to guide fault response. It requires access to and processing of historical data accumulated over long-term operation. This data should include previously recorded fault phase parameters and the corresponding real-time environmental compensation parameters at the time of the fault. By analyzing this historical data, for example using statistical correlation analysis or machine learning methods, a correlation mapping table can be constructed. This table reveals potential historical patterns of correlation between specific fault types and specific environmental compensation parameter settings. Based on the patterns learned from this mapping table, one or more anti-interference plans can be pre-generated, forming an anti-interference plan set. Each plan typically provides a set of compensation parameter combinations considered relatively more stable and less prone to errors for a specific type of environmental parameter condition.

[0116] Based on the temperature gradient parameters, a combination of compensation parameters is selected from the anti-interference plan set to autonomously reconstruct the imaging data generation link.

[0117] Specifically, when the self-diagnostic recovery procedure needs to be initiated, the current real-time environmental parameters, especially the temperature gradient parameters, are obtained. Using the current temperature gradient parameters as a query condition, a search or match is performed within the previously generated set of anti-interference contingency plans to select a predefined combination of compensation parameters best suited to the current environmental conditions. Once selected, an autonomous reconstruction operation is executed. This means automatically adjusting the relevant control logic and parameter settings responsible for generating optimized imaging data, applying the selected combination of compensation parameters. This typically involves changing the specific strategies for calculating camera exposure time or controlling the illumination mode, thereby altering the characteristics of the imaging data generation chain. The aim is to switch to a more robust operating mode based on historical experience to overcome or avoid currently detected faults or potential interference.

[0118] For example, if abnormal brightness saturation is detected in the image during operation, the corresponding fault phase parameters are recorded. When self-diagnostic recovery is required, the current temperature gradient parameters are obtained and found to be in a high range. Based on this high temperature gradient, a set of anti-interference contingency plans previously constructed based on historical operational data is queried, and a set of optimized compensation parameters marked as suitable for high temperature gradient conditions is selected. Subsequently, the current imaging control parameters are automatically switched to this selected set of parameters, for example, by adjusting the exposure reference or supplementary lighting strategy, thus autonomously reconstructing the imaging data generation chain to eliminate the brightness saturation problem and achieve stable operation.

[0119] It should be noted that the formulas described above, through the principle of dimensional consistency and mathematical standardization methods (such as normalization, dimensionless parameter conversion, or unit system unification), can translate physical quantities with different properties into unitless standard values ​​or parameters that can be superimposed in the same dimension. This eliminates the interference of different dimensions on the computational logic, allowing the formulas to retain the original data distribution characteristics while possessing mathematical rationality and adaptability to objective laws. These are conventional technical methods and will not be elaborated further. The electrical connections between the various units described above do not necessarily represent direct or indirect connections; any indirect connection method is applicable to the embodiments of this invention as long as it achieves the purpose of this invention. The above descriptions are merely exemplary embodiments of this invention and should not be construed as limiting the scope of this invention.

[0120] All equivalent changes and modifications made in accordance with the teachings of this invention are still within the scope of this invention. Those skilled in the art will readily conceive of other embodiments of this invention upon considering the specification and the disclosure of practical truth. This application is intended to cover any variations, uses, or adaptations of this invention that follow the general principles of this invention and include common knowledge or conventional techniques in the art not described herein.

Claims

1. A machine vision-based automated packaging and inspection method for braised chicken, characterized in that, The method includes: acquiring real-time vibration parameters and temperature gradient parameters of the braised chicken packaging on the conveyor belt, and generating real-time environmental compensation parameters; dynamically adjusting the exposure time and supplementary lighting mode of the camera unit based on the real-time environmental compensation parameters to generate optimized imaging data; wherein, generating optimized imaging data includes: calculating a baseline exposure time based on the vibration frequency component in the real-time environmental compensation parameters, and superimposing the influence factor of the temperature gradient parameters to generate an exposure adjustment coefficient; triggering an alternating illumination mode of the near-infrared light channel and the visible light channel based on the exposure adjustment coefficient to generate optimized imaging data; and performing multispectral feature extraction on the optimized imaging data to generate a sealing defect map. The process of extracting multispectral features from the optimized imaging data includes: extracting the infrared absorption difference spectrum of the packaging film in the near-infrared light channel to generate a thermally induced deformation vector field; constructing surface texture depth convolution features in the visible light channel to generate a set of foreign object contour coordinates after condensate interference suppression; combining the spatiotemporal alignment features of the thermally induced deformation vector field and the foreign object contour coordinate set to generate a vapor artifact isolation index; and matching the deformation parameters in the sealing defect spectrum based on a preset biomechanical feature parameter library to generate a chicken body morphology compliance determination result. The generation of the chicken body morphology compliance determination result includes: from the sealing defect... Biological texture feature curves are segmented from the atlas, and leg joint bending angle sequences and sternal curvature fluctuation parameters are extracted. A preset elasticity compensation database is used to correct the sternal curvature fluctuation parameters for meat elasticity, generating a standardized morphological feature vector. This standardized morphological feature vector is then matched with a standard anatomical template in the biomechanical feature parameter library using a distance metric to generate a chicken body morphology compliance determination result. Based on the coupling relationship between the thermal expansion coefficient inversion value of the foreign object distribution atlas and the temperature gradient parameter, a dynamic foreign object confirmation marker is generated. The generation of the dynamic foreign object confirmation marker includes: injecting the vapor artifact isolation index into a preset thermo-mechanical coupling criterion model. The process involves calculating a foreign object attachment confidence threshold; when the confidence level of the foreign object contour coordinate set exceeds the foreign object attachment confidence threshold, generating a dynamic foreign object confirmation marker by marking high-probability foreign object coordinates, and outputting a packaging interruption signal; generating the dynamic foreign object confirmation marker further includes: analyzing the spatial proximity between the abnormal morphology area determined by the chicken body morphology compliance judgment result and the foreign object location determined by marking high-probability foreign object coordinates; adjusting the confidence level and risk level of the dynamic foreign object confirmation marker based on the spatial proximity result; and fusing the chicken body morphology compliance judgment result and the dynamic foreign object confirmation marker to generate a comprehensive packaging integrity diagnostic instruction.

2. The automated packaging and inspection method for braised chicken based on machine vision according to claim 1, characterized in that, The generation of chicken body shape compliance determination results also includes: when the chicken body shape compliance determination result indicates that the joint angle deviates, activating the pressure tactile sensor to collect the meat rebound displacement; using the meat rebound displacement to invert the muscle stress relaxation curve and generate a dynamic shape correction factor; feeding the dynamic shape correction factor back to the standardized shape feature vector generation stage to update the preset elastic compensation database.

3. The automated packaging and inspection method for braised chicken based on machine vision according to claim 1, characterized in that, The generated packaging integrity comprehensive diagnostic command further includes: calculating the packaging stress concentration area based on the spatial distribution density of the dynamic foreign object confirmation mark; controlling the backlight unit to project a color temperature gradient light spot of a preset wavelength onto the stress concentration area to generate a stress visualization warning spectrum; and coupling the stress visualization warning spectrum with the conveyor belt speed control command in a time sequence to trigger an adaptive deceleration mechanism.

4. The automated packaging and inspection method for braised chicken based on machine vision according to claim 3, characterized in that, The trigger adaptive deceleration mechanism includes: calculating the predicted value of the expansion trend of the stress concentration area and generating a pressure risk index by associating it with the weight data of the same batch of packaging; when the pressure risk index exceeds the preset composite threshold, adjusting the conveyor belt running speed according to the exponential decay curve, and synchronously updating the anatomical template weight coefficient in the biomechanical characteristic parameter library.

5. The automated packaging and inspection method for braised chicken based on machine vision according to claim 1, characterized in that, The method further includes: monitoring the feature space continuity of the optimized imaging data; recording fault phase parameters when a sudden signal distortion is detected; acquiring historical operating data and constructing an association mapping table between the fault phase parameters and the real-time environmental compensation parameters based on the historical operating data to generate an anti-interference plan set; selecting a combination of compensation parameters from the anti-interference plan set according to the temperature gradient parameters, and autonomously reconstructing the imaging data generation link.

Images