A wine box security anti-counterfeiting system based on an internet of things perception layer

Abstract

The present application relates to the field of anti-counterfeiting technology of Internet of Things, in particular to a wine box security anti-counterfeiting system based on Internet of Things sensing layer, comprising: a flexible sensing network is constructed on the inner surface of the wine box to capture the original sensing sequence of the physical state; the sequence is time-domain segmented and feature reconstructed to form a sealed integrity, environmental disturbance and identification optical feature set; the physical features and optical features are used respectively to independently deduce the packaging credibility and anti-counterfeiting credibility; the double credibility is input into a cross-validation model to output a suspicious score of the wine box state; the score is used to query a mapping table to obtain a preliminary warning level; finally, the warning information of adjacent wine boxes is aggregated for collaborative correction to generate a final security decision instruction. The present application realizes the monitoring of the physical integrity and circulation history from the inside of the package, and improves the accuracy and robustness of the anti-counterfeiting decision through multi-evidence cross-validation, which is suitable for the security management of high-end commodity packaging.

Description

Technical Field

[0001] This invention relates to the field of Internet of Things (IoT) anti-counterfeiting technology, and in particular to a wine box security anti-counterfeiting system based on the IoT sensing layer. Background Technology

[0002] Traditional anti-counterfeiting technologies for alcoholic beverages primarily rely on external labels, such as QR codes or RFID tags. The core of these solutions is to authenticate the label itself, verifying information based on pre-set static data. This approach has inherent limitations: the label is physically separated from the product packaging, making it easily detached and transferred to counterfeit products. External labels only provide one-dimensional information about authenticity and cannot detect whether the packaging has been illegally opened or subjected to abnormal physical environmental changes, resulting in limited anti-counterfeiting capabilities.

[0003] Current anti-counterfeiting verification processes mostly rely on a single-path decision-making approach, where the system directly outputs a authenticity conclusion after reading the label features. This method lacks analysis of the logical consistency between the packaging's physical state and the label information. When counterfeiters use recycled genuine packaging, the system's reliance solely on label authenticity ignores the risk of content substitution, making it ill-equipped to counter sophisticated counterfeiting techniques. Therefore, a solution is needed that can monitor the physical state from within the packaging and integrate multi-dimensional evidence for intelligent adjudication.

[0004] The problem this invention aims to solve is to achieve continuous monitoring of the physical integrity and circulation history of packaging from its source, and to improve the accuracy and reliability of the anti-counterfeiting system by integrating multi-dimensional features and cross-verification mechanisms. This aims to overcome the shortcomings of easily transferable external markings, single anti-counterfeiting dimensions, and single-path verification logic. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by proposing a wine box security and anti-counterfeiting system based on the Internet of Things (IoT) sensing layer.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a wine box security and anti-counterfeiting system based on the Internet of Things sensing layer, comprising:

[0007] The sensing and capturing module constructs a flexible sensing network attached to the inner surface of the wine box, and the flexible sensing network captures the original sensing sequence characterizing the physical state of the wine box.

[0008] The feature construction module performs temporal domain segmentation and feature reconstruction on the original sensing sequence to form a feature set including sealing integrity features, environmental disturbance features, and identification optical features.

[0009] The credibility derivation module performs multi-level fusion verification of the feature set, derives the packaging credibility using the sealing integrity feature and environmental disturbance feature, and independently derives the anti-counterfeiting credibility using the identification optical feature.

[0010] The cross-validation module inputs the encapsulation credibility and anti-counterfeiting credibility into a preset cross-validation model, which outputs a comprehensive suspicious score for the wine box status.

[0011] The early warning mapping module queries and obtains the initial early warning mark level in a preset suspicion level mapping table based on the suspicion score of the wine box status.

[0012] The collaborative decision-making module aggregates the warning label levels of the flexible sensing network carried by adjacent wine boxes, and performs collaborative correction on the preliminary warning label levels based on the aggregation results to generate the final safety decision instructions.

[0013] Preferably, the construction of the flexible sensing network attached to the inner surface of the wine box specifically includes:

[0014] The arrangement topology of the sensing units is planned according to the three-dimensional structure of the wine box to ensure that each major stress surface and key packaging seam of the wine box is covered by a sensing unit.

[0015] Configure the sensing unit to simultaneously acquire micro-deformation signals, temperature and humidity gradient signals, and reflection signals in a specific spectral band, which specifically refers to the 780nm-1100nm range in the near-infrared band.

[0016] The wake-up and sampling cycle of the sensing unit is set so that it enters a low-power listening mode when the wine box is stationary, and switches to an active sampling mode when an external stimulus is detected or the timing cycle is reached.

[0017] All sensing units deployed on a single wine box form the flexible sensing network through self-organization. The flexible sensing network integrates the collected micro-deformation signals, temperature and humidity gradient signals, and reflection signals in specific spectral bands into the original sensing sequence.

[0018] Preferably, the process of performing temporal domain segmentation and feature reconstruction on the original sensing sequence specifically includes:

[0019] The original sensing sequence is divided into continuous sensing data blocks according to a preset time window length;

[0020] For each sensing data block, frequency domain energy distribution and time domain statistics are extracted. The frequency domain energy distribution is used to characterize the vibration response mode of the sealed structure, and the time domain statistics are used to characterize the fluctuation mode of environmental parameters.

[0021] Image preprocessing and feature point matching are performed on the reflection signal of the specific spectral band to calculate the marker sharpness coefficient and pattern matching coefficient.

[0022] The vibration response pattern, environmental parameter fluctuation pattern, marking clarity coefficient, and pattern matching coefficient extracted within the same time window are packaged to form an entry of the feature set.

[0023] Preferably, the derivation of packaging reliability using seal integrity characteristics and environmental disturbance characteristics specifically includes:

[0024] Separate the sealing integrity feature and the environmental disturbance feature from the feature set;

[0025] The sealing integrity features are input into the trained packaging defect identification model, and the packaging defect identification model outputs a sealing score.

[0026] The environmental disturbance characteristics are compared with the standard transportation environment spectrum to calculate an environmental anomaly deviation degree.

[0027] A dynamic weighted algorithm is used to fuse the sealing performance score and the environmental anomaly deviation. The weights of the dynamic weighted algorithm are adaptively adjusted according to the current logistics stage of the wine box. The result of the fusion calculation is the packaging reliability.

[0028] Preferably, the step of independently deriving the anti-counterfeiting credibility using the optical features of the identifier specifically includes:

[0029] Extract the identification optical features from the feature set;

[0030] Run the anti-counterfeiting label authenticity identification process, which includes calculating the similarity between the label optical features and the pre-stored genuine product optical feature template, and checking whether there are trace patterns left by copying attacks in the optical features.

[0031] If the similarity is higher than the verification threshold and no trace pattern is detected, it is determined to be true and a basic truth value is assigned.

[0032] By combining the historical reading stability data of the aforementioned optical features, a reliability enhancement calculation is performed on the basic authenticity value, and the output value after the enhancement calculation is the anti-counterfeiting credibility.

[0033] Preferably, the workflow of the cross-validation model specifically includes:

[0034] Receive the encapsulation credibility and the anti-counterfeiting credibility as input;

[0035] Within the cross-validation model, the encapsulation credibility is mapped to the physical security dimension, and the anti-counterfeiting credibility is mapped to the information authenticity dimension.

[0036] Establish a logical constraint relationship between the physical security dimension and the information authenticity dimension. The logical constraint relationship defines the matching conditions that the values ​​of the two dimensions should satisfy under normal conditions.

[0037] Based on the aforementioned logical constraints, calculate the inconsistency measure between the encapsulation credibility and the anti-counterfeiting credibility;

[0038] Based on the magnitude of the inconsistency metric, the suspicious score of the wine box status is calculated using the scoring function embedded in the cross-validation model.

[0039] Preferably, the process of using the suspicion level mapping table specifically includes:

[0040] The suspicion level mapping table is a predefined lookup table that divides the continuous suspicious score range of the wine box status into multiple discrete intervals.

[0041] Each interval is associated with a preset text description label and a corresponding numerical level, which serves as the initial warning level.

[0042] The calculated suspicious score of the wine box is compared with each interval in the suspicious level mapping table to determine the interval to which it belongs;

[0043] Output the numerical level and textual description label corresponding to the interval to which the suspicious score of the wine box belongs.

[0044] Preferably, the process of collaboratively correcting the preliminary warning marker level based on the aggregation results to generate the final security decision instruction specifically includes:

[0045] Centered on the target wine box, determine the set of adjacent wine boxes associated with it based on logistics information or spatial proximity;

[0046] Obtain the current preliminary warning level for each wine box in the set of adjacent wine boxes;

[0047] The distribution of each warning marker level in the adjacent wine box set is statistically analyzed, and the group warning consensus degree is calculated.

[0048] Based on the degree of consensus among the groups regarding early warning, a preset correction strategy is selected;

[0049] Based on the selected correction strategy, the initial warning mark level of the target wine box is numerically adjusted, and the adjusted result is the level information carried in the final safety decision instruction.

[0050] Preferably, the selection and execution rules of the correction strategy specifically include:

[0051] Set high and low thresholds for the level of consensus in group early warning;

[0052] If the consensus of the group's early warning is higher than the high threshold, it is determined that the group's opinions are highly consistent, and a strengthening correction strategy is adopted to adjust the initial early warning mark level of the target wine box towards the mode level in the set of adjacent wine boxes;

[0053] If the consensus of the group's early warning is lower than the low threshold, it is determined that the group's opinions are scattered, and a conservative correction strategy is adopted, making only minor adjustments to or maintaining the initial early warning mark level of the target wine box.

[0054] If the consensus level of the group's early warning is between the low threshold and the high threshold, a weighted correction strategy is adopted. Weights are assigned according to the spatial distance or logistics correlation strength between each adjacent wine box and the target wine box, and a weighted average calculation is performed to adjust the initial early warning label level of the target wine box.

[0055] Preferably, the construction steps of the cross-validation model include:

[0056] Collect samples of the sealing reliability and anti-counterfeiting reliability of normal and abnormal wine boxes from historical data to form a training sample set;

[0057] The encapsulation credibility and anti-counterfeiting credibility in the training sample set are normalized so that the values ​​of different dimensions are mapped to the same scale.

[0058] Clustering algorithms are used to divide the normalized training sample set into multiple clusters, each cluster representing a typical correlation pattern between the physical security dimension and the information authenticity dimension.

[0059] For each cluster, calculate the joint probability distribution of the encapsulation credibility and anti-counterfeiting credibility of its internal samples, and define the logical constraint relationship of the cluster based on the joint probability distribution;

[0060] By integrating the logical constraints of all clusters, the cross-validation model is constructed.

[0061] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0062] By constructing a flexible sensing network on the inner surface of the wine box, the original sensor sequence reflecting the physical state of the packaging is directly captured. This technical solution realizes the transformation of anti-counterfeiting information collection from externally attached labels to the internal structural state of the packaging itself. The captured seal integrity characteristics and environmental disturbance characteristics provide a direct and inseparable data source for determining whether the packaging has experienced illegal opening or abnormal physical environment. In this way, the basis for anti-counterfeiting verification is no longer a static label that can be peeled off and transferred, but a dynamic historical record of the packaging's own integrity, improving the non-replicability and reliability of anti-counterfeiting elements.

[0063] The system employs a technical approach that independently derives dual credibility from physical and optical features, then fuses these results through a cross-validation model. Sealing integrity and environmental disturbance characteristics jointly derive the packaging credibility reflecting the packaging's physical state, while the identification optical features independently derive the anti-counterfeiting credibility. The conclusions of both technical approaches serve as input, and a pre-defined model analyzes their inherent logical consistency, outputting a comprehensive suspicion score. This evolves anti-counterfeiting verification from a single-path binary judgment to a multi-evidence chain correlation analysis and contradiction detection. The system can effectively identify inconsistencies between physical state evidence and identification optical evidence, thus, even if the identification itself is genuine, it can still determine the risk of the packaging being illegally used, enhancing its ability to identify complex fraudulent methods.

[0064] After generating the initial warning level, a collaborative decision-making mechanism based on warning information from adjacent wine boxes is introduced. This step utilizes the spatial clustering properties of products in the circulation environment to correct individual warning conclusions based on group consistency. This can mitigate misjudgments caused by random errors of individual sensing units, and at the same time, by identifying spatially correlated anomaly patterns, improve the overall monitoring accuracy and decision-making stability of the system in batch safety incidents. Attached Figure Description

[0065] Figure 1 This is a timing diagram of the wine box security and anti-counterfeiting system based on the Internet of Things sensing layer described in this invention;

[0066] Figure 2 Flowchart for constructing a flexible sensing network;

[0067] Figure 3 Flowchart for cross-validation model operation;

[0068] Figure 4 A heatmap showing the percentage of suspicious score intervals at each stage of wine box logistics;

[0069] Figure 5 A radar chart for multi-dimensional evaluation of suspicious status scores for wine boxes. Detailed Implementation

[0070] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0071] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0072] See Figure 1 A flexible sensing network is constructed attached to the inner surface of the wine box. This network captures and outputs the original sensing sequence characterizing the physical state of the wine box. The system performs temporal segmentation and feature reconstruction on the original sensing sequence, forming a feature set including sealing integrity features, environmental disturbance features, and identification optical features. The system performs multi-level fusion verification of the feature set. On the one hand, it uses the sealing integrity features and environmental disturbance features to derive the packaging credibility; on the other hand, it independently derives the anti-counterfeiting credibility using the identification optical features. The system inputs the packaging credibility and anti-counterfeiting credibility into a preset cross-validation model, which outputs a comprehensive wine box state suspicion score. Based on this suspicion score, the system queries a preset suspicion level mapping table to obtain a preliminary warning label level. The system aggregates the warning label levels of the flexible sensing networks carried by adjacent wine boxes and performs collaborative correction on the aforementioned preliminary warning label levels based on the aggregation result, thereby generating the final security decision instruction.

[0073] In one embodiment of the present invention, see [reference] Figure 2The process of constructing a flexible sensing network attached to the inner surface of the wine box involves planning the topology of the sensing units according to the three-dimensional structure of the wine box, ensuring that each major stress surface and key packaging seam of the wine box is covered by a sensing unit, and configuring the sensing unit to simultaneously collect micro-deformation signals, temperature and humidity gradient signals, and reflection signals in a specific spectral band. The specific spectral band refers to the 780nm-1100nm range in the near-infrared band. This band is selected based on the characteristics of the special fluorescent material of the anti-counterfeiting label built into the wine box. This fluorescent material will produce a characteristic reflection peak of 950nm-1050nm under 780nm excitation light, and this band avoids the background reflection interference of common paper packaging and ink, and can accurately capture the exclusive optical signal of the anti-counterfeiting label. The miniature near-infrared spectral sensor in the sensing unit collects continuous spectral reflectance data within this range at a fixed wavelength interval of 10nm to ensure the complete waveform covering the characteristic reflectance peak. The wake-up and sampling cycle of the sensing unit is set so that it enters a low-power listening mode when the wine box is stationary and switches to an active sampling mode when an external stimulus is detected or the timing cycle is reached. All sensing units deployed on a single wine box form a flexible sensing network through self-organization. This network integrates the various signals collected into the original sensing sequence. The process of temporal segmentation and feature reconstruction of the original sensing sequence involves cutting the original sensing sequence into continuous sensing data blocks according to a preset time window length. For each sensing data block, frequency domain energy distribution and temporal statistics are extracted. The frequency domain energy distribution is used to characterize the vibration response mode of the sealed structure, and the temporal statistics are used to characterize the fluctuation mode of environmental parameters. The reflection signal of the specific spectral band is continuous spectral data in the near-infrared range of 780nm-1100nm. The preprocessing process is designed for the signal characteristics of this band: first, noise signals generated by environmental infrared radiation in the band above 1000nm are removed, and the effective range data of 780nm-1000nm is retained; then, the diffuse reflection background interference of the inner surface of the paper wine box is eliminated through the baseline correction algorithm, and the characteristic reflection segment data of 950nm-1050nm corresponding to the fluorescent material of the anti-counterfeiting label is extracted; finally, the spectral curve of the characteristic reflection segment is converted into a two-dimensional grayscale image for subsequent feature point matching. Simultaneously, image preprocessing and feature point matching are performed on the reflection signals of specific spectral bands to calculate the marker sharpness coefficient and pattern matching coefficient. Finally, the vibration response mode, environmental parameter fluctuation mode, marker sharpness coefficient and pattern matching coefficient extracted within the same time window are packaged to form an entry in the feature set.

[0074] In practical implementation, the process of constructing a flexible sensing network attached to the inner surface of a wine box can be illustrated using a specific wine box form. Taking a cuboid paper wine box as an example, the topology of the sensing units is planned according to the three-dimensional structure of the wine box, ensuring that sensing units cover all six main outer surfaces of the wine box and the sealing seams of the top and bottom lids. At least three sensing units are deployed on each main stress-bearing surface, such as the bottom and sides, while sensing units are densely deployed along a line at critical sealing seams. The function of configuring the sensing units involves integrating multiple micro-sensors. A piezoelectric thin film sensor that synchronously collects micro-deformation signals, a digital temperature and humidity sensor that collects temperature and humidity gradient signals, and a micro near-infrared spectral sensor that collects reflection signals in specific spectral bands are packaged together within a single sensing unit. During the wake-up and sampling cycle of the sensing unit, the micro-deformation sensor remains in a low-power listening mode with a microampere current. When vibration or deformation excitation exceeding the threshold is detected, the entire sensing unit is woken up and enters the active sampling mode. At the same time, all sensors perform synchronous data acquisition with a period of 100 milliseconds. When the wine box is stationary without excitation and the preset ten-minute timer period is reached, the sensing unit will also briefly enter the active sampling mode to perform an environmental benchmark measurement. All sensing units deployed on a single wine box form a flexible sensing network through a self-organizing wireless mesh network protocol. The flexible sensing network integrates the micro-deformation signal, temperature and humidity gradient signal, and reflection signal of specific spectral bands collected by each sensing unit according to the timestamp and spatial coordinates to form a time-synchronized multimodal original sensing sequence.

[0075] In some embodiments, the process of temporal segmentation and feature reconstruction of the original sensing sequence has clearly defined parameters. The original sensing sequence is segmented according to a preset one-second time window length to obtain a series of continuous sensing data blocks. For a micro-deformation signal sensing data block within a time window, its frequency domain energy distribution is extracted by fast Fourier transform. The formula for calculating the frequency domain energy distribution is:

[0076]

[0077] in: This represents the characteristic frequency band of the preset sealing structure. Frequency domain energy distribution value within, It is the first time that the micro-deformation signal has undergone Fourier transform. One frequency component, This represents the energy of that frequency component. The numerical values ​​are used to characterize the vibration response pattern of the wine box sealing structure within the time period. For the temperature and humidity gradient signal sensing data block within the same time window, its maximum value, minimum value, and standard deviation are calculated as time-domain statistics. These time-domain statistics are used to characterize the fluctuation pattern of environmental parameters.

[0078] It is understandable that the processing of reflection signals in specific spectral bands is independent of vibration and environmental signal analysis. The system performs image preprocessing on the reflection signals in specific spectral bands collected within each time window. The preprocessing steps include grayscale conversion, binarization, and noise reduction. Subsequently, the processed image is matched with a pre-stored reference anti-counterfeiting label image for feature points. By statistically analyzing the proportion of successfully matched feature point pairs to the total number of feature points, a label sharpness coefficient and a pattern matching coefficient are calculated. The label sharpness coefficient reflects the imaging quality of the label image, and the pattern matching coefficient reflects the geometric similarity between the current label and the reference label. Finally, the system packages the vibration response mode, environmental parameter fluctuation mode, label sharpness coefficient, and pattern matching coefficient extracted within the same time window to form a complete entry in the feature set.

[0079] In some embodiments, the topology of the sensing units can be adjusted according to the material of the wine box. For wine boxes with high rigidity, such as wood or metal, the deployment density of sensing units at the seams can be higher than that on the main stress-bearing surfaces. In specific implementations, the length of the time window is not fixed; the system can dynamically adjust the time window length according to the logistics stage, using a longer time window during the storage and resting stage and a shorter time window during the transportation vibration stage to capture more refined dynamic changes. Optionally, the calculation of the frequency domain energy distribution focuses on the characteristic frequency bands. It was determined by training and analyzing the micro-deformation signals of a large number of normally packaged wine boxes and wine boxes with known packaging defects. This frequency band can effectively distinguish between signals generated by normal transportation vibration and abnormal vibration signals generated by packaging cracks or material fatigue.

[0080] In one embodiment of the present invention, the process of deriving the packaging reliability using sealing integrity features and environmental disturbance features is as follows: sealing integrity features and environmental disturbance features are separated from the feature set, the sealing integrity features are input into a trained packaging defect identification model, the model outputs a sealing score, the environmental disturbance features are compared with a standard transportation environment spectrum to calculate an environmental anomaly deviation, and a dynamic weighting algorithm is used to fuse the sealing score and the environmental anomaly deviation. The weights of the dynamic weighting algorithm are adaptively adjusted according to the current logistics stage of the wine box, and the result of the fusion calculation is the packaging reliability. Meanwhile, the process of independently deriving the anti-counterfeiting credibility using the optical features of the identifier involves extracting the optical features of the identifier from the feature set and running the authenticity identification process of the anti-counterfeiting label. This process includes calculating the similarity between the optical features of the identifier and the pre-stored genuine optical feature template, and checking whether there are trace patterns left by copying attacks in the optical features. If the similarity is higher than the verification threshold and no trace patterns are detected, it is determined to be genuine and assigned a basic authenticity value. Combining the historical reading stability data of the optical features of the identifier, the reliability enhancement calculation is performed on the basic authenticity value. The output value after the enhancement calculation is the anti-counterfeiting credibility.

[0081] In practical implementation, the process of deriving packaging reliability using sealing integrity features and environmental disturbance features can be illustrated through a scenario of monitoring the status of a wine box in the logistics chain. The system separates sealing integrity features and environmental disturbance features from the feature set. The sealing integrity feature is the frequency domain energy distribution value extracted in the embodiment to characterize the vibration response mode of the sealing structure. The environmental disturbance feature is the time domain statistics extracted in the embodiment to characterize the fluctuation mode of environmental parameters. The sealing integrity feature is input into the trained packaging defect identification model. The packaging defect identification model is a classification neural network trained based on historical data. It receives the frequency domain energy distribution value and outputs a sealing score between 0 and 1. The closer the sealing score is to 1, the higher the sealing integrity. The environmental disturbance feature is compared with the standard transportation environment spectrum. The standard transportation environment spectrum defines the baseline curves of the allowable temperature and humidity fluctuation range and vibration intensity under different stages of road transportation, air transportation, and warehouse storage. By calculating the Euclidean distance between the current environmental disturbance feature and the baseline curve corresponding to the current logistics stage, an environmental anomaly deviation is calculated. The larger the environmental anomaly deviation value, the more significant the environmental anomaly.

[0082] In some embodiments, a dynamic weighted algorithm is used to fuse the sealing performance score and the environmental anomaly deviation. The weights of the dynamic weighted algorithm are adaptively adjusted according to the current logistics stage of the wine box. For example, when the wine box is in the stage of road transportation with severe vibration, a higher weight is given to the sealing performance score; when the wine box is in the stage of temperature- and humidity-sensitive constant temperature warehouse storage, a higher weight is given to the environmental anomaly deviation. The packaging reliability is calculated using the following weighted fusion formula:

[0083]

[0084] in: Represents the reliability of the packaging. The sealing score represents the output of the packaging defect identification model. This represents the calculated degree of environmental anomaly deviation (normalized to the range of 0 to 1). This represents a dynamic weighting factor determined by the current logistics stage, with values ​​between 0 and 1. The result of the fusion calculation... This refers to the packaging reliability; the higher the value, the more reliable the physical packaging state.

[0085] It is understandable that the process of independently deriving the anti-counterfeiting credibility using the identification optical features is executed in parallel with the process of deriving the encapsulation credibility. The system extracts the identification optical features from the feature set. The identification optical features include the identification sharpness coefficient and pattern matching coefficient calculated in the embodiment. The system runs the anti-counterfeiting label authenticity identification process. The anti-counterfeiting label authenticity identification process includes calculating the similarity between the identification optical features and the pre-stored genuine optical feature template. The pre-stored genuine optical feature template includes a baseline sharpness threshold and a baseline pattern feature point set. The similarity calculation combines the ratio of the identification sharpness coefficient to the baseline sharpness threshold, as well as the overlap rate between the current feature point matching result and the baseline pattern feature point set. It also checks whether there are trace patterns left by copying attacks in the optical features. The trace patterns are manifested as moiré fringe noise or edge ghosting features at specific frequencies.

[0086] In some embodiments, if the similarity is higher than a preset verification threshold and no trace pattern is detected, it is determined to be true and a basic authenticity value is assigned. The basic authenticity value is a value between 0 and 1. Its initial value is obtained by linear mapping of the similarity calculation result. The reliability enhancement calculation is performed on the basic authenticity value by combining the historical reading stability data of the identification optical feature. The historical reading stability data refers to the variance between the similarity calculation result of the identification optical feature and the trace pattern detection result in the past multiple sampling periods. The smaller the variance, the more stable the reading. The enhancement calculation smooths the basic authenticity value through a stability factor, and the output value is the anti-counterfeiting credibility.

[0087] In one embodiment of the present invention, see [reference] Figure 3The workflow of the cross-validation model is as follows: it receives the packaging credibility and anti-counterfeiting credibility as inputs, maps the packaging credibility to the physical security dimension and the anti-counterfeiting credibility to the information authenticity dimension within the model, establishes a logical constraint relationship between the physical security dimension and the information authenticity dimension, and defines the matching conditions that the values ​​of the two dimensions should meet under normal conditions. Based on the logical constraint relationship, it calculates the inconsistency measure between the packaging credibility and the anti-counterfeiting credibility, and calculates the suspicious score of the wine box status through the scoring function embedded in the model according to the magnitude of the inconsistency measure. The construction steps of the cross-validation model include: collecting samples of the packaging credibility and anti-counterfeiting credibility of normal and abnormal wine boxes from historical data to form a training sample set; normalizing the packaging credibility and anti-counterfeiting credibility in the training sample set to map values ​​of different dimensions to the same scale; using a clustering algorithm to divide the normalized training sample set into multiple clusters, with each cluster representing a typical association pattern between a physical security dimension and an information authenticity dimension; calculating the joint probability distribution of packaging credibility and anti-counterfeiting credibility of samples within each cluster and defining the logical constraint relationship of the cluster based on the joint probability distribution; and integrating the logical constraint relationships of all clusters to construct the final cross-validation model.

[0088] In its implementation, the cross-validation model's workflow is designed based on the correlation between the physical security dimension and the information authenticity dimension. The cross-validation model receives the encapsulation credibility and anti-counterfeiting credibility calculated from the implementation examples as direct inputs. Within the cross-validation model, the system maps the encapsulation credibility to a physical security dimension. The physical security dimension value directly reflects the integrity of the physical structure of the wine box's outer packaging. The system maps the anti-counterfeiting credibility to an information authenticity dimension. The information authenticity dimension value directly reflects the authenticity of the anti-counterfeiting label inside the wine box. A logical constraint relationship is established between the physical security dimension and the information authenticity dimension. This logical constraint relationship defines the matching conditions that the physical security dimension value and the information authenticity dimension value should meet under normal, tamper-free conditions. For example, a wine box with a high physical security dimension value should also have a high information authenticity dimension value, while a wine box with an extremely low physical security dimension value should theoretically not have a high information authenticity dimension value.

[0089] In some embodiments, an inconsistency measure between encapsulation credibility and anti-counterfeiting credibility is calculated based on logical constraints. This inconsistency measure quantifies the degree to which the currently observed combination of physical security and information authenticity deviates from the normal logical constraints. Based on the magnitude of the inconsistency measure, a suspiciousness score for the wine box is calculated using a scoring function embedded in the cross-validation model. This scoring function is designed as a monotonically increasing function of the inconsistency measure; a higher suspiciousness score indicates a more suspicious current wine box state. One implementation method uses the following formula:

[0090]

[0091] in: Represents a measure of inconsistency. Represents the reliability of the packaging. Represents the credibility of anti-counterfeiting measures. and These represent the center values ​​of the cluster to which the current sample belongs in the dimensions of physical security and information authenticity, respectively. The weight factor represents the cluster to which the current sample belongs. It is determined by the proportion of historical anomalies in samples within the cluster.

[0092] It is understandable that the construction steps of the cross-validation model begin with the collection and processing of historical data. Samples of the packaging credibility and anti-counterfeiting credibility of normal and abnormal wine boxes are collected from historical data to form a training sample set. Normal wine box samples come from products with complete and accurate logistics records, while abnormal wine box samples come from recalled products or experimental samples that are known to have been opened, replaced, or counterfeited. The packaging credibility and anti-counterfeiting credibility in the training sample set are normalized by using the min-max normalization method to map the values ​​of different dimensions to a unified [0,1] scale.

[0093] In some embodiments, a clustering algorithm is used to divide the normalized training sample set into multiple clusters. The clustering algorithm used is a Gaussian mixture model. Each cluster represents a typical association pattern between the physical security dimension and the information authenticity dimension. For example, one cluster may represent a normal pattern of "high physical security and high information authenticity", while another cluster may represent a potential substitution and forgery pattern of "low physical security but high information authenticity". For each cluster divided by the clustering algorithm, the joint probability distribution of the encapsulation credibility and anti-counterfeiting credibility of all samples within it is calculated. The joint probability distribution is fitted using a kernel density estimation method. Based on the fitted joint probability distribution, the logical constraint relationship of the current cluster is defined. The logical constraint relationship is specifically represented by a conditional probability function, which describes the reasonable distribution range of the information authenticity dimension value under a given physical security dimension value. The logical constraint relationships of all clusters are integrated to construct the final cross-validation model. The integration method is to assign a confidence weight to the logical constraint relationship of each cluster and construct a global decision function.

[0094] In one embodiment of the present invention, the process of using the suspicion level mapping table is as follows: the mapping table, as a predefined lookup table, divides the continuous suspicious score range of the wine box status into multiple discrete intervals. Each interval is associated with a preset text description label and a corresponding numerical level, with the numerical level serving as a preliminary warning marker level. The system compares the calculated suspicious score of the wine box status with each interval in the suspicion level mapping table to determine the interval to which it belongs, and finally outputs the numerical level and text description label corresponding to the interval to which the suspicious score of the wine box status belongs. In specific implementation, the process of using the suspicion level mapping table can be illustrated with a specific scoring mapping scenario. The suspicion level mapping table is a static query data structure predefined before system deployment. Its core function is to divide the continuous suspicious score range of the wine box status into multiple discrete intervals. For example, the theoretical output range of the suspicious score of the wine box status is 0 to 1. The system divides this range into five consecutive intervals, each interval being associated with a preset text description label and a corresponding numerical level. The numerical level serves as a preliminary warning marker level, and the text description label is used to provide an intuitive interpretation of the status.

[0095] In some embodiments, the division and mapping relationship of the numerical range of suspicious box status scores can be represented by a specific table. The table defines the boundaries of each interval, the corresponding numerical level, and the textual description label. The system compares the calculated suspicious box status score with each interval in the suspicious level mapping table to determine the interval to which it belongs. The comparison process is achieved by sequentially checking whether the suspicious box status score falls within the minimum and maximum value range of a certain interval. After determining the interval, the system outputs the numerical level and textual description label corresponding to the interval to which the suspicious box status score belongs. A method for determining interval indexes. The mathematical expression is as follows:

[0096]

[0097] in: This represents the calculated interval index number (corresponding to the numerical level). This represents the suspicious score of the wine box state that needs to be mapped. This represents the minimum score (e.g., 0) covered by the mapping table. This represents the constant width of each interval (e.g., when the total number of intervals is 5). ), Represents the floor function, when If the calculated result exceeds the maximum number of intervals, it is limited to the maximum number of intervals. See Table 1.

[0098] Table 1: Suspicion Level Mapping Table

[0099]

[0100] It is understandable that the use of Table 1 is a deterministic query operation. Assuming the system calculates a suspicious status score of 0.35 for a particular wine box, according to Table 1, this score falls within the interval [0.2, 0.4). The system will output a numerical level 2 and the textual description label "Slightly Suspicious Status." This numerical level 2 is the initial warning level for the wine box. Similarly, a wine box with a suspicious status score of 0.72 will be mapped to a numerical level 4 and the label "Highly Suspicious Status." In some embodiments, the interval division of the suspicious status mapping table may not use an equal-width method. For example, a denser interval division can be used in the range of higher suspicious status scores to achieve a more refined distinction of high-risk states. The interval boundary values ​​are determined based on the percentile distribution of historical statistical data. Optionally, the suspicious status mapping table can be configured to support dynamic updates. After the system has accumulated a large amount of new suspicious status score data over a long period of operation, the interval boundary values ​​can be recalculated and adjusted according to the new data distribution to adapt to changes in the score distribution of different batches of products or different logistics environments.

[0101] See Figure 4 This is a heatmap showing the percentage of suspicious score ranges at each stage of wine box logistics. Darker colors indicate a higher percentage for that range at the corresponding stage. The production and packaging stage has an extremely high percentage in the [0.0, 0.2) range (corresponding to "normal status"). This indicates strong initial stability of the wine boxes during production. The trunk transportation / last-mile delivery stage shows an increase in the percentage of medium-to-high suspicious score ranges such as [0.4, 0.6) and [0.6, 0.8), reflecting the significant impact of environmental disturbances during transportation on the wine box status. The retail shelf placement stage shows an increase in the percentage of the [0.2, 0.4) range ("slightly suspicious status"), possibly due to secondary disturbances during the final distribution process. This graph visually presents the changing patterns of suspicious wine box status across logistics stages, identifies high-risk logistics stages, optimizes the monitoring frequency for these stages, verifies the effectiveness of status control across the entire "production-distribution" chain, and assists in adjusting early warning thresholds.

[0102] In one embodiment of the present invention, the process of collaboratively correcting the preliminary warning label level based on the aggregation result and generating the final safety decision instruction is as follows: taking the target wine box as the center and determining its associated set of neighboring wine boxes based on logistics information or spatial proximity, obtaining the current preliminary warning label level of each wine box in the set of neighboring wine boxes, statistically analyzing the distribution of each warning label level in the set of neighboring wine boxes and calculating the group warning consensus, selecting a preset correction strategy based on the magnitude of the group warning consensus, and adjusting the preliminary warning label level of the target wine box numerically according to the selected correction strategy. The specific adjustment rule is that the mode of the warning levels in the set of neighboring wine boxes is taken as the target value. If the difference between the preliminary warning level of the target wine box and the mode level is Δ (Δ is a positive integer, with a level range of 1-5), then the adjustment range is 80% of Δ (rounded down to an integer multiple of 0.5), and the adjusted level must not exceed the effective range of 1-5. For example, if the initial warning level for the target wine box is 4, and the mode level of the adjacent wine box set is 1, then Δ=3, the adjustment range is 3×80%=2.4, rounded down to 2.0, ultimately adjusting the target wine box level to 4-2.0=2.0. If the initial warning level for the target wine box is 2, and the mode level of the adjacent wine boxes is 5, then Δ=3, the adjustment range is 2.0, ultimately adjusting the level to 2+2.0=4.0. If Δ=1, the adjustment range is 0.8, rounded down to 0.5, for example, adjusting the initial level 3 to 3.5 or 2.5 (depending on the mode direction). During adjustment, regardless of the size of Δ, the difference between the adjusted level and the mode level must not exceed 1.0, ensuring that the adjustment reflects group consensus while avoiding erroneous corrections due to extreme differences. The adjusted result is the level information carried in the final safety decision instruction. The selection and execution rules for the correction strategy are as follows: a high threshold and a low threshold are set for the group's early warning consensus. If the group's early warning consensus is higher than the high threshold, it is determined that the group's opinions are highly consistent and a strengthening correction strategy is adopted, which adjusts the initial early warning mark level of the target wine box to the mode level of the adjacent wine box set by a large margin. If the group's early warning consensus is lower than the low threshold, it is determined that the group's opinions are dispersed and a conservative correction strategy is adopted, which only makes minor adjustments to the initial early warning mark level of the target wine box or keeps it unchanged. If the group's early warning consensus is between the low threshold and the high threshold, a weighted correction strategy is adopted, which allocates weights according to the spatial distance or logistics correlation strength between each adjacent wine box and the target wine box and performs a weighted average calculation to adjust the initial early warning mark level of the target wine box.

[0103] In specific implementation, the process of collaboratively correcting the preliminary warning label level based on the aggregation results and generating the final safety decision instruction can be illustrated by the monitoring scenario of wine boxes on a logistics pallet. Taking the target wine box as the center, the set of adjacent wine boxes associated with it is determined based on logistics information or spatial proximity. For example, when the target wine box is located inside a standard shipping carton, the other wine boxes in the same carton are determined as the set of adjacent wine boxes. When the target wine box is located on a storage pallet, all wine boxes on the same layer of the entire pallet are determined as the set of associated adjacent wine boxes. The current preliminary warning label level of each wine box in the set of adjacent wine boxes is obtained. The preliminary warning label level is the numerical level output after querying the suspicious level mapping table in the embodiment.

[0104] In some embodiments, the distribution of each warning marker level in adjacent wine box sets is statistically analyzed, and the group warning consensus degree is calculated. The group warning consensus degree is an indicator that quantifies the consistency of warning opinions within adjacent wine box sets. One specific method uses the following formula:

[0105]

[0106] in: The degree of consensus among representative groups in early warning. This represents the total number of boxes in the adjacent set of boxes (including the target box). Represents the set of the first The initial warning level for each wine box, Representing all The arithmetic mean of the initial warning level markings on each wine box. The theoretical range representing the initial warning level (e.g., if the level range is 1 to 5, then...) ), Representing the The absolute deviation of each wine box grade from the average grade.

[0107] It is understandable that a preset correction strategy is selected based on the level of consensus on early warning. A high threshold and a low threshold are set for the consensus level, with the high threshold set at 0.8 and the low threshold at 0.3. If the consensus level is higher than 0.8, the group is considered to have highly consistent opinions, and a strengthening correction strategy is adopted. The initial early warning level of the target wine box is adjusted significantly towards the mode level of the adjacent wine box set, with an adjustment range of one complete level unit. If the consensus level is lower than 0.3, the group is considered to have dispersed opinions, and a conservative correction strategy is adopted. Only a minor adjustment or no change is made to the initial early warning level of the target wine box, with the minor adjustment range not exceeding 0.5 level units. If the consensus level is between 0.3 and 0.8, a weighted correction strategy is adopted. Weights are allocated based on the spatial distance or logistical correlation between each adjacent wine box and the target wine box, and a weighted average is calculated to adjust the initial early warning level of the target wine box. The initial early warning level of the target wine box is numerically adjusted according to the selected correction strategy, and the adjusted result is the level information carried in the final safety decision instruction.

[0108] In some embodiments, a specific example of collaborative correction is as follows: the initial warning level of the target wine box is 4 (highly suspicious), and its neighboring wine box set contains a total of 12 wine boxes. The initial warning level of the remaining 11 wine boxes is as follows: 9 wine boxes are at level 1 (normal), and 2 wine boxes are at level 2 (slightly suspicious). The group warning consensus degree is calculated. The initial warning level is set to 0.15. Since 0.15 is below the low threshold of 0.3, the system adopts a conservative correction strategy. Given that the vast majority of wine boxes in the adjacent set are in normal condition, the system slightly adjusts the initial warning level of the target wine box from 4 to 3.5. This 3.5 is the level information carried in the final safety decision instruction. Optionally, in the weighted correction strategy, the weights can be allocated based on signal strength. Adjacent wine boxes with higher communication signal strength to the target wine box are considered to be spatially closer or have a closer logistical connection, and therefore are given higher weights. Optionally, the final safety decision instruction may include not only the adjusted level information, but also the correction strategy type that triggered the instruction and the original initial warning level, to facilitate auditing and traceability analysis by the backend system.

[0109] See Figure 5This is a radar chart for a multi-dimensional assessment of the suspiciousness score of wine boxes, used to compare the performance of normal and suspicious wine boxes in multiple dimensions. This radar chart, through a comprehensive comparison of multi-dimensional features, intuitively distinguishes the differences between normal and suspicious wine boxes, and can be used to assist in setting feature thresholds for anti-counterfeiting systems; it quickly locates the abnormal feature dimensions of suspicious wine boxes, providing direction for subsequent verification. By comparing multiple dimensions (environment, sealing, anti-counterfeiting, etc.) on the "radar surface," it intuitively quantifies the feature differences between normal and suspicious wine boxes, avoiding the one-sidedness of judging by a single indicator, and highlighting the shortcomings of abnormal features. It quickly locates the abnormal dimensions of suspicious wine boxes, reducing the blindness of verification.

[0110] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A wine box security and anti-counterfeiting system based on the Internet of Things (IoT) sensing layer, characterized in that, Includes the following modules: The sensing and capturing module constructs a flexible sensing network attached to the inner surface of the wine box, and the flexible sensing network captures the original sensing sequence characterizing the physical state of the wine box. The feature construction module performs time-domain segmentation and feature reconstruction on the original sensing sequence to form a feature set including sealing integrity features, environmental disturbance features and identification optical features. Among them, the sealing integrity feature is the frequency domain energy distribution value used to characterize the vibration response mode of the sealing structure, and the environmental disturbance feature is the time domain statistics used to characterize the fluctuation mode of environmental parameters. The credibility derivation module performs multi-level fusion verification of the feature set, derives the packaging credibility using the sealing integrity feature and environmental disturbance feature, and independently derives the anti-counterfeiting credibility using the identification optical feature. The derivation of packaging reliability using seal integrity characteristics and environmental disturbance characteristics specifically includes: Separate the sealing integrity feature and the environmental disturbance feature from the feature set; The sealing integrity feature is input into the trained packaging defect identification model, which is a classification neural network trained based on historical data. It receives the frequency domain energy distribution value and outputs a sealing score in the range of 0 to 1. The environmental disturbance characteristics are compared with the standard transportation environment spectrum, which defines the baseline curves of the allowable temperature and humidity fluctuation range and vibration intensity at different stages of logistics. By calculating the Euclidean distance between the current environmental disturbance characteristics and the baseline curve corresponding to the current logistics stage, an environmental anomaly deviation degree is calculated. A dynamic weighted algorithm is used to fuse the sealing score and the environmental anomaly deviation. The weights of the dynamic weighted algorithm are adaptively adjusted according to the current logistics stage of the wine box. The result of the fusion calculation is the packaging reliability. The method of simultaneously utilizing the optical features of the identifier to independently deduce the credibility of the anti-counterfeiting measures specifically includes: Extract the identification optical features from the feature set; Run the anti-counterfeiting label authenticity identification process, which includes calculating the similarity between the label optical features and the pre-stored genuine product optical feature template, and checking whether there are trace patterns left by copying attacks in the optical features. If the similarity is higher than the verification threshold and no trace pattern is detected, it is determined to be true and a basic truth value is assigned. Combining the historical reading stability data of the identified optical features, a reliability enhancement calculation is performed on the basic authenticity value. The output value after the enhancement calculation is the anti-counterfeiting credibility. The historical reading stability data refers to the variance between the similarity calculation result of the identified optical features and the trace pattern detection result over multiple past sampling periods. The smaller the variance, the more stable the reading. The reliability enhancement calculation uses a stability factor obtained based on the historical reading stability data to smoothly adjust the basic authenticity value. The cross-validation module inputs the encapsulation credibility and anti-counterfeiting credibility into a preset cross-validation model, which outputs a comprehensive suspicious score for the wine box status. The workflow of the cross-validation model specifically includes: Receive the encapsulation credibility and the anti-counterfeiting credibility as input; Within the cross-validation model, the encapsulation credibility is mapped to the physical security dimension, and the anti-counterfeiting credibility is mapped to the information authenticity dimension. Establish a logical constraint relationship between the physical security dimension and the information authenticity dimension. The logical constraint relationship defines the matching conditions that the values ​​of the two dimensions should satisfy under normal conditions. Based on the logical constraint relationship, the inconsistency measure between the packaging credibility and the anti-counterfeiting credibility is calculated. The inconsistency measure quantifies the degree to which the currently observed combination of physical security dimension and information authenticity dimension deviates from the normal logical constraint relationship. According to the magnitude of the inconsistency measure, the suspicious score of the wine box status is calculated through the scoring function embedded in the cross-validation model. The early warning mapping module queries and obtains the initial early warning mark level in a preset suspicion level mapping table based on the suspicion score of the wine box status. The collaborative decision-making module aggregates the warning label levels of the flexible sensing network carried by adjacent wine boxes, and performs collaborative correction on the preliminary warning label levels based on the aggregation results to generate the final safety decision instructions.

2. The anti-counterfeiting system for wine boxes based on the Internet of Things sensing layer according to claim 1, characterized in that, The construction of the flexible sensing network attached to the inner surface of the wine box specifically includes: The arrangement topology of the sensing units is planned according to the three-dimensional structure of the wine box to ensure that each major stress surface and key packaging seam of the wine box is covered by a sensing unit. Configure the sensing unit to simultaneously acquire micro-deformation signals, temperature and humidity gradient signals, and reflection signals in a specific spectral band, which specifically refers to the 780nm-1100nm range in the near-infrared band. The wake-up and sampling cycle of the sensing unit is set so that it enters a low-power listening mode when the wine box is stationary, and switches to an active sampling mode when an external stimulus is detected or the timing cycle is reached. All sensing units deployed on a single wine box form the flexible sensing network through self-organization. The flexible sensing network integrates the collected micro-deformation signals, temperature and humidity gradient signals, and reflection signals in specific spectral bands into the original sensing sequence.

3. The anti-counterfeiting system for wine boxes based on the Internet of Things sensing layer according to claim 2, characterized in that, The process of performing temporal segmentation and feature reconstruction on the original sensing sequence specifically includes: The original sensing sequence is divided into continuous sensing data blocks according to a preset time window length; For each sensing data block, frequency domain energy distribution and time domain statistics are extracted. The frequency domain energy distribution is used to characterize the vibration response mode of the sealed structure, and the time domain statistics are used to characterize the fluctuation mode of environmental parameters. Specifically, for a micro-deformation signal sensing data block within a time window, its frequency domain energy distribution is extracted by fast Fourier transform. For a temperature and humidity gradient signal sensing data block within the same time window, its maximum value, minimum value, and standard deviation are calculated as time domain statistics. The formula for calculating the frequency domain energy distribution is: ； in: This represents the characteristic frequency band of the preset sealing structure. Frequency domain energy distribution value within, It is the first time that the micro-deformation signal has undergone Fourier transform. One frequency component, This represents the energy of that frequency component; Image preprocessing and feature point matching are performed on the reflection signal of a specific spectral band to calculate the marker sharpness coefficient and pattern matching coefficient. The specific spectral band refers specifically to the 780nm-1100nm range in the near-infrared band. The process of image preprocessing and feature point matching for the reflection signal of the specific spectral band includes: the reflection signal of the specific spectral band is continuous spectral data in the 780nm-1100nm near-infrared range; the preprocessing process is designed for the signal characteristics of this band: firstly, noise signals generated by ambient infrared radiation in the band above 1000nm are removed, retaining the effective range of 780nm-1000nm. The data is then processed by: first, eliminating diffuse background interference from the inner surface of the paper wine box using a baseline correction algorithm; second, extracting the 950nm-1050nm characteristic reflectance band data corresponding to the fluorescent material of the anti-counterfeiting label; finally, converting the spectral curve of this characteristic reflectance band into a two-dimensional grayscale image for subsequent feature point matching; and third, matching the processed image with a pre-stored benchmark anti-counterfeiting label image. By statistically analyzing the ratio of successfully matched feature point pairs to the total number of feature points, a label clarity coefficient and a pattern matching coefficient are calculated. The label clarity coefficient reflects the imaging quality of the label image, while the pattern matching coefficient reflects the geometric similarity between the current label and the benchmark label. The vibration response pattern, environmental parameter fluctuation pattern, marking clarity coefficient, and pattern matching coefficient extracted within the same time window are packaged to form an entry of the feature set. The vibration response mode is used to characterize the sealing structure and corresponds to the sealing integrity feature; the environmental parameter fluctuation mode is used to characterize the environmental parameters and corresponds to the environmental disturbance feature; the mark sharpness coefficient and the image matching coefficient together constitute the mark optical feature.

4. The anti-counterfeiting system for wine boxes based on the Internet of Things sensing layer according to claim 1, characterized in that, The encapsulation reliability is calculated using the following formula: ； in, Represents the reliability of the packaging. The hermeticity score represents the output of the packaging defect identification model. This represents the calculated environmental anomaly deviation, normalized to the range of 0 to 1. This represents a dynamic weighting factor determined by the current logistics stage, with a value between 0 and 1.

5. The anti-counterfeiting system for wine boxes based on the Internet of Things sensing layer according to claim 1, characterized in that, The inconsistency metric is calculated using the following formula: ； in, Represents a measure of inconsistency. Represents the reliability of the packaging. Represents the credibility of anti-counterfeiting measures. and These represent the center values ​​of the cluster to which the current sample belongs in the dimensions of physical security and information authenticity, respectively. The weight factor represents the cluster to which the current sample belongs. It is determined by the proportion of historical anomalies in samples within the cluster.

6. The anti-counterfeiting system for wine boxes based on the Internet of Things sensing layer according to claim 1, characterized in that, The process of using the suspicion level mapping table specifically includes: The suspicion level mapping table is a predefined lookup table that divides the continuous suspicious score range of the wine box status into multiple discrete intervals. Each interval is associated with a preset text description label and a corresponding numerical level, which serves as the initial warning level. The calculated suspicious score of the wine box is compared with each interval in the suspicious level mapping table to determine the interval to which it belongs; Output the numerical level and textual description label corresponding to the interval to which the suspicious score of the wine box belongs.

7. The anti-counterfeiting system for wine boxes based on the Internet of Things sensing layer according to claim 3, characterized in that, The anti-counterfeiting label authentication process includes calculating the similarity between the optical features of the label and a pre-stored genuine optical feature template. The pre-stored genuine optical feature template includes a baseline sharpness threshold and a baseline pattern feature point set. The similarity calculation combines the ratio of the label sharpness coefficient to the baseline sharpness threshold, as well as the overlap rate between the current feature point matching result and the baseline pattern feature point set.

8. The anti-counterfeiting system for wine boxes based on the Internet of Things sensing layer according to claim 1, characterized in that, The process of collaboratively correcting the preliminary warning marker level based on the aggregation results to generate the final security decision instruction specifically includes: Centered on the target wine box, determine the set of adjacent wine boxes associated with it based on logistics information or spatial proximity; Obtain the current preliminary warning level for each wine box in the set of adjacent wine boxes; The distribution of each warning marker level in the adjacent wine box set is statistically analyzed, and the group warning consensus degree is calculated. Calculate the consensus degree of the group's early warning Use the following formula: ； in, Represents the level of consensus in early warning among representative groups. This represents the total number of boxes in the adjacent set of boxes, including the target box. This represents the initial warning level for the i-th wine box in the set. Representing all The arithmetic mean of the initial warning level for each wine box. The theoretical range representing the initial warning level; Based on the degree of consensus among the groups regarding early warning, a preset correction strategy is selected; Based on the selected correction strategy, the initial warning mark level of the target wine box is numerically adjusted, and the adjusted result is the level information carried in the final safety decision instruction.

9. A wine box security and anti-counterfeiting system based on the Internet of Things sensing layer according to claim 8, characterized in that, The specific rules for selecting and implementing the correction strategy include: Set high and low thresholds for the level of consensus in group early warning; If the consensus of the group's early warning is higher than the high threshold, it is determined that the group's opinions are highly consistent, and a strengthening correction strategy is adopted to adjust the initial early warning mark level of the target wine box towards the mode level in the set of adjacent wine boxes; If the consensus of the group's early warning is lower than the low threshold, it is determined that the group's opinions are scattered, and a conservative correction strategy is adopted, making only minor adjustments to or maintaining the initial early warning mark level of the target wine box. If the consensus level of the group's early warning is between the low threshold and the high threshold, a weighted correction strategy is adopted. Weights are assigned according to the spatial distance or logistics correlation strength between each adjacent wine box and the target wine box, and a weighted average calculation is performed to adjust the initial early warning label level of the target wine box.

10. The anti-counterfeiting system for wine boxes based on the Internet of Things sensing layer as described in claim 1, characterized in that, The steps for constructing the cross-validation model include: Collect samples of the sealing reliability and anti-counterfeiting reliability of normal and abnormal wine boxes from historical data to form a training sample set; The encapsulation credibility and anti-counterfeiting credibility in the training sample set are normalized so that the values ​​of different dimensions are mapped to the same scale. Clustering algorithms are used to divide the normalized training sample set into multiple clusters, each cluster representing a typical correlation pattern between the physical security dimension and the information authenticity dimension. For each cluster, calculate the joint probability distribution of the encapsulation credibility and anti-counterfeiting credibility of its internal samples, and define the logical constraint relationship of the cluster based on the joint probability distribution; By integrating the logical constraints of all clusters, the cross-validation model is constructed.

Images