A vision-based cigarette packet print traceability system

Abstract

The application discloses a kind of based on vision's cigarette shell package printing traceability system, system includes: micro multi-view stroboscopic acquisition module, through orthogonal polarized light path and multi-path stroboscopic exposure, obtain the gray image set of the specific printing area of the measured cigarette shell package;Surface micro normal vector reconstruction module, based on Lambertian diffuse reflection model Calculation of the normalized surface normal vector of pixel, generate three-dimensional micro normal vector field map;Ink fiber topology hash coding module, extract normal depth component and carry out local binary pattern feature dimension reduction, generate fixed-length physical hash sequence;High concurrency anti-fake traceability verification module is used for cloud binding storage in factory stage, and in terminal verification stage, calculate the hamming distance of the measured hash sequence and the registered fixed-length physical hash sequence, combined with tolerance threshold output determination result, the application eliminates specular reflection interference, converts three-dimensional micro topological form into discrete sequence, improves anti-fake security and verification efficiency.

Description

Technical Field

[0001] This invention relates to the field of machine vision and anti-counterfeiting traceability technology, specifically a vision-based cigarette pack printing traceability system. Background Technology

[0002] Cigarette packaging typically requires anti-counterfeiting and traceability verification during manufacturing and distribution. Currently, anti-counterfeiting technology based on visual feature extraction has seen initial application in this field due to its non-contact and non-destructive characteristics. However, in the actual image acquisition stage, the surface of cigarette packaging usually has a transparent reflective layer. Conventional optical acquisition equipment easily produces specular reflections when illuminating the surface, causing highlight masking of the microscopic morphology of the underlying paper fibers and printed ink, making it impossible to extract pure optical signals. Furthermore, during on-site verification in the distribution process, operators typically use unconstrained handheld terminals for image acquisition. Handheld operation inevitably leads to random variations in the relative working distance and spatial pitch angle between the acquisition device and the cigarette packaging surface, causing a significant deviation between the actual illumination angle and the production line's factory calibration conditions, thus reducing the reliability of the basic image input data.

[0003] Most existing visual anti-counterfeiting and traceability systems rely on two-dimensional planar image features for authenticity comparison. However, these features are highly susceptible to interference from fluctuations in ambient light intensity during extraction. More seriously, because two-dimensional features lack the physical depth of three-dimensional structural measurements, counterfeiters can easily acquire and replicate the two-dimensional texture patterns on packaging surfaces using high-precision planar scanning and copying techniques. Conventional two-dimensional visual verification systems cannot effectively distinguish between genuine microscopic undulations and counterfeit planar replicas, making the entire anti-counterfeiting and traceability system vulnerable to malicious attacks and resulting in low security.

[0004] Furthermore, for physical anti-counterfeiting solutions at the microscopic level, directly storing and comparing features of continuous and complex three-dimensional microscopic underlying data would significantly increase the storage space requirements of the cloud database and the computational load during the retrieval stage. This massive data processing method is difficult to meet the high-concurrency verification needs of industrial production lines and consumer markets. Moreover, existing rigid feature comparison algorithms lack adaptability to actual circulation environments. Cigarette packaging inevitably experiences legitimate surface paper fiber wear and microscopic deformation during warehousing, transportation, and sales. Traditional precise comparison modes are easily interfered with by such legitimate microscopic deformation noise, leading to misjudgments in traceability and compromising the robustness of the entire traceability system. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a vision-based cigarette packaging printing traceability system. This system solves the problems of unreliable underlying data caused by interference from surface reflection and handheld posture deviation in image acquisition of existing anti-counterfeiting and traceability systems, the lack of physical depth in conventional two-dimensional anti-counterfeiting features making them easy to be counterfeited by high-precision scanning and copying technologies, the excessive computational load of three-dimensional feature comparison, and the low robustness of judgment due to the incompatibility of rigid comparison algorithms with legitimate micro-wear during packaging circulation.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a vision-based traceability system for cigarette pack printing, comprising: The microscopic multi-field stroboscopic acquisition module is set at the inspection station of the cigarette packing production line. It is used to acquire grayscale image sets of a specific printed area on the surface of the cigarette pack under different lighting directions through multi-channel stroboscopic and synchronous exposure. The surface micro normal vector reconstruction module is connected to the micro multi-field strobe acquisition module. It is used to receive grayscale image sets, calculate the normalized surface normal vector of each pixel in the target printing area based on the Lambertian diffuse reflection physical model, and generate a three-dimensional micro normal vector field map. The ink fiber topological hash encoding module is connected to the surface micro normal vector reconstruction module. It is used to extract the normal depth component in the three-dimensional micro normal vector field map, perform local binary pattern feature extraction and dimensionality reduction on the normal depth component, and generate a fixed-length physical hash sequence. The high-concurrency anti-counterfeiting and traceability verification module communicates with the ink fiber topology hash encoding module. It is used to bind the fixed-length physical hash sequence with the production plaintext information and store it in the cloud database during the product manufacturing stage, and to obtain the test hash sequence on site during the terminal verification stage. The product traceability judgment result is output by measuring the Hamming distance between the test hash sequence and the fixed-length physical hash sequence.

[0007] Preferably, the microscopic multi-field stroboscopic acquisition module includes an array light source, four polarizers, an analyzer, an industrial camera, and a mechanical phase synchronization unit. The array light source consists of four light-emitting diodes (LEDs) evenly distributed in a ring around the optical axis of the industrial camera lens. The four polarizers are fixedly installed at the light-emitting ends of the four LEDs, and the analyzer is installed at the lens input end of the industrial camera. The polarization direction of the analyzer is at a 90-degree spatial angle with the polarization directions of the four polarizers to form an orthogonal polarization optical path. The mechanical phase synchronization unit is electrically connected to the spindle rotary encoder of the cigarette packaging production line. When the real-time mechanical angle pulse signal output by the spindle rotary encoder reaches the target phase threshold, it generates a synchronous trigger level to trigger the array light source to perform stroboscopic sequentially and trigger the industrial camera to perform synchronous exposure.

[0008] Preferably, the surface microscopic normal vector reconstruction module includes a Lambertian model establishment unit, an equation system construction unit, and a least squares solution unit. The Lambertian model establishment unit is used to extract the observed gray values ​​of specific pixels in the grayscale image set under different illumination directions to construct the observed grayscale column vector, and establish the photometric solid fundamental equation in combination with the pre-calibrated illumination direction matrix. The equation system construction unit is used to define the unnormalized surface normal vector as the scalar product of the surface albedo and the surface unit normal vector, and to construct an overdetermined linear equation system by performing algebraic substitution on the photometric solid fundamental equation system. The least squares solution unit is used to simultaneously multiply the transpose of the illumination direction matrix on both sides of the equal sign of the overdetermined linear equation system to convert it into a normal equation. By extracting the inverse matrix of the illumination direction matrix product, the analytical solution formula is executed to calculate the unnormalized surface normal vector corresponding to the specific pixel.

[0009] Preferably, the surface micro normal vector reconstruction module further includes an albedo separation unit, a normal vector normalization unit, and a normal vector field generation unit. The albedo separation unit is used to perform vector modulo operation on the unnormalized surface normal vector to calculate the surface albedo corresponding to a specific pixel. The normal vector normalization unit is used to divide the unnormalized surface normal vector by the surface albedo to remove the reflectivity attribute and calculate the normalized surface normal vector through vector division. The normal vector field generation unit is used to traverse the coordinates of all pixels within the target printing area and arrange the summarized normalized surface normal vectors into a three-dimensional micro normal vector field map, wherein the three-dimensional micro normal vector field map is constructed as a multi-dimensional matrix storing normal depth data.

[0010] Preferably, the ink fiber topological hashing encoding module includes a depth component extraction unit, a spatial gradient calculation unit, a neighborhood sampling unit, and a binary pattern calculation unit. The depth component extraction unit is used to extract the normal depth component perpendicular to the physical surface of the target printing area from the normalized surface normal vector. The spatial gradient calculation unit uses the discrete pixel difference operator to calculate the partial derivatives of the normal depth component in the horizontal and vertical directions, and calculates the two-dimensional spatial gradient magnitude at a specific pixel by performing the sum of squares and square root operations of the partial derivatives, thereby generating a two-dimensional spatial gradient magnitude matrix. The spatial gradient calculation unit filters high-frequency response data based on the gradient magnitude judgment threshold to locate the physical topological boundary. The neighborhood sampling unit is used to set a specific pixel in the two-dimensional spatial gradient magnitude matrix as the center pixel, construct a circular sampling neighborhood on the two-dimensional spatial gradient magnitude matrix with the center pixel, and extract the neighborhood gradient magnitude on the sampling circumference. The binary pattern calculation unit is used to use the signed step function to binarize the difference between the neighborhood gradient magnitude and the center gradient magnitude, and generate a local binary pattern response value by multiplying and adding with binary weights, thereby generating a local binary pattern response matrix.

[0011] Preferably, the ink fiber topology hash encoding module further includes a spatial pooling unit, a binarization unit, and a sequence concatenation unit. The spatial pooling unit is used to divide the local binary pattern response matrix into non-overlapping local grid regions according to the target length value of the hash sequence, and calculate the feature representative value corresponding to each local grid region to generate a one-dimensional feature array. The binarization unit is used to extract the median of the feature representative values ​​in the one-dimensional feature array as the hash quantization threshold, and generate a binary value set corresponding to the one-dimensional feature array through numerical comparison operation. The sequence concatenation unit is used to concatenate the binary value set according to the initial two-dimensional spatial coordinate order of the local grid regions to generate a fixed-length physical hash sequence.

[0012] Preferably, the high-concurrency anti-counterfeiting and traceability verification module includes a data packaging unit, a cloud communication unit, and a cloud database. The data packaging unit is used to obtain the production timestamp, production batch number, and production line identification code of the cigarette pack to be tested as production plaintext information. It combines and encapsulates the fixed-length physical hash sequence with the production plaintext information to generate a product registration data packet. The cloud communication unit is used to perform transport layer encryption on the product registration data packet and send it to the cloud database. The cloud database is used to decrypt the product registration data packet, set the fixed-length physical hash sequence as the primary key index in the relational data mapping table, and synchronously write the production plaintext information configured as the associated field of the primary key index.

[0013] Preferably, the high-concurrency anti-counterfeiting and traceability verification module includes an inspection and verification terminal, which is configured as a portable handheld device. The hardware front end of the inspection and verification terminal integrates a portable microscopic multi-field-of-view stroboscopic acquisition mechanism. The portable microscopic multi-field-of-view stroboscopic acquisition mechanism contains a handheld macro camera. The front end of the portable microscopic multi-field-of-view stroboscopic acquisition mechanism is fixedly installed with a fitting positioning light shield. The fitting positioning light shield is used to tightly press against the surface of a specific printed area of ​​the cigarette pack to be tested, so as to physically lock the relative working distance and spatial pitch angle of the handheld macro camera. The end face of the fitting positioning light shield is equipped with a pressure contact switch. The pressure contact switch is used to generate a level trigger signal when a preset pressure fitting state is reached, so as to control the inspection and verification terminal to acquire the diffuse reflection grayscale image set to be tested on site.

[0014] Preferably, the high-concurrency anti-counterfeiting and traceability verification module also includes a field hash generation unit built into the processor of the inspection and verification terminal. The field hash generation unit is used to receive the diffuse reflection grayscale image set to be tested on site, perform least squares solution and normalization processing to generate a field micro three-dimensional normal vector field map; and through spatial block pooling and dynamic median threshold binarization operations, the local binary mode response values ​​are spliced ​​into a test hash sequence with the same data length as the fixed-length physical hash sequence.

[0015] Preferably, the high-concurrency anti-counterfeiting and traceability verification module includes a sequence retrieval unit, a distance measurement unit, and a traceability discrimination unit. The sequence retrieval unit is used to send a traceability retrieval request containing plaintext association fields to the cloud database and extract the corresponding fixed-length physical hash sequence. The distance measurement unit is used to extract the binary values ​​of the fixed-length physical hash sequence and the hash sequence to be tested at the same index position, perform a logical XOR operation, and perform an accumulation operation on the logical XOR operation result to calculate the Hamming distance that represents the degree of sequence difference. The traceability discrimination unit is configured with a tolerance threshold. The tolerance threshold is obtained by pre-calibrating the Hamming distance of the genuine worn sample. It is used to determine that the cigarette pack to be tested is genuine when the Hamming distance is less than or equal to the tolerance threshold, and to output a warning interception signal to block traceability when the Hamming distance is greater than the tolerance threshold.

[0016] This invention provides a vision-based traceability system for cigarette pack packaging printing, which has the following advantages: 1. This invention, by setting up a microscopic multi-field stroboscopic acquisition module containing orthogonal polarization optical paths, combined with the spatial orthogonal arrangement of the polarizer and analyzer, effectively blocks the specular reflection light generated by the transparent dielectric layer on the surface of the cigarette packaging, and extracts the pure diffuse reflection light signal from the underlying paper fibers and printed ink. Simultaneously, a fitted positioning light shield with a pressure contact switch is configured in the inspection and verification terminal to physically lock the relative working distance and spatial pitch angle during handheld acquisition, replacing mechanical synchronization signal triggering exposure. This ensures a high degree of consistency between on-site verification acquisition conditions and production line factory calibration conditions, improving the reliability of the underlying image input data.

[0017] 2. This invention utilizes a surface microscopic normal vector reconstruction module to establish a fundamental photometric equation based on the Lambertian diffuse reflection physical model. It then solves the overdetermined linear equations using the least squares method, transforming a two-dimensional grayscale image into three-dimensional physical topological features. The system further removes surface reflectivity attributes through albedo separation and normalization, directly extracting the normal depth component that characterizes the microscopic morphology of the ink-paper fiber interface. This computational logic eliminates the interference of external light intensity fluctuations on feature extraction. Furthermore, through three-dimensional depth structural measurement, the system effectively defends against counterfeiting techniques employing high-precision planar scanning and copying, enhancing the security of anti-counterfeiting and traceability.

[0018] 3. This invention utilizes an ink-fiber topological hash encoding module to calculate the two-dimensional spatial gradient magnitude using discrete pixel difference operators. Combined with a local binary pattern algorithm and spatial pooling mechanism, it reduces the dimensionality of continuous and complex microscopic normal depth data into discrete, fixed-length physical hash sequences. During the cloud-based comparison phase, the high-concurrency anti-counterfeiting and traceability verification module calculates the Hamming distance between the registered sequence and the sequence to be tested, and performs authenticity judgment based on a pre-defined tolerance threshold. This feature reduction and binarization comparison mechanism significantly reduces the storage space requirements and retrieval computational load of the cloud database, meeting the high-concurrency verification needs of industrial production lines and the market. Furthermore, the tolerance judgment logic eliminates the interference of legitimate microscopic deformations during the physical circulation of packaging on the verification results, ensuring the robustness of the traceability system. Attached Figure Description

[0019] Figure 1 This is a diagram of the architecture of the present invention. Detailed Implementation

[0020] The technical solutions in 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, and 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.

[0021] Please see the appendix Figure 1 This invention provides a vision-based cigarette pack printing traceability system, comprising: a microscopic multi-field strobe acquisition module, a surface microscopic normal vector reconstruction module, an ink fiber topology hash encoding module, and a high-concurrency anti-counterfeiting and traceability verification module, wherein the surface microscopic normal vector reconstruction module and the ink fiber topology hash encoding module are both deployed in an industrial control computer.

[0022] The microscopic multi-field stroboscopic acquisition module is installed at the inspection station of the cigarette packing production line. The microscopic multi-field stroboscopic acquisition module includes an array light source, a polarizer, an analyzer, an industrial camera, and a mechanical phase synchronization unit. The array light source consists of four light-emitting diodes. The central emission wavelength of the four light-emitting diodes matches the absorption spectrum of the target ink. The four light-emitting diodes are evenly distributed in a ring around the optical axis of the industrial camera lens, and the spatial angle between adjacent light-emitting diodes is 90 degrees.

[0023] Polarizers are respectively installed at the light emission ends of the four light-emitting diodes; analyzers are installed at the lens entrance end of the industrial camera, and the polarization directions of the analyzers and polarizers are orthogonal to each other. The mechanical phase synchronization unit establishes an electrical connection with the main spindle rotary encoder of the packaging production line. It is used to intercept the mechanical motion stationary phase signal output by the main spindle rotary encoder and use the stationary phase signal as the trigger clock for the array light source strobe and the industrial camera exposure.

[0024] During system operation, the microscopic multi-field stroboscopic acquisition module triggers the array light source to stroboscopically based on the received static phase signal. The industrial camera synchronously performs continuous exposure operation to acquire grayscale image sets of a specific printed area on the surface of the cigarette shell under four different illumination directions. Due to the orthogonal polarization setting of the analyzer and the polarizer, the analyzer blocks the specular reflection light generated by the coating on the surface of the cigarette shell. The light entering the target surface of the industrial camera is diffuse reflection light scattered by the paper fiber and ink interface.

[0025] The surface microscopic normal vector reconstruction module and the microscopic multi-field stroboscopic acquisition module are connected via an industrial bus to receive grayscale image sets. Based on the Lambertian diffuse reflection physical model, the surface microscopic normal vector reconstruction module combines the illumination direction matrix and the pixel grayscale equations, solves the pixel grayscale equations using the least squares method, calculates the surface albedo and normalized surface normal vector of each pixel in the target area, and generates a three-dimensional microscopic normal vector field map of the smoke shell to be tested.

[0026] The ink fiber topology hashing module is connected to the surface micro normal vector reconstruction module to extract the normal depth component in the three-dimensional micro normal vector field map and calculate the two-dimensional spatial gradient magnitude matrix. Furthermore, the ink fiber topology hashing module uses the local binary mode algorithm to extract the topological features of the two-dimensional spatial gradient magnitude matrix to locate the boundary formed by the ink along the gaps in the paper fibers and generate a fixed-length physical hash sequence.

[0027] The high-concurrency anti-counterfeiting and traceability verification module communicates with the ink fiber topology hash encoding module. During the product manufacturing stage, the high-concurrency anti-counterfeiting and traceability verification module stores the generated fixed-length physical hash sequence into a cloud database or local storage medium. During the terminal verification stage, the high-concurrency anti-counterfeiting and traceability verification module receives the hash sequence to be tested collected on-site and calculates the Hamming distance between the hash sequence to be tested and the physical hash sequence in the database. By comparing the Hamming distance with the preset tolerance threshold, the product traceability judgment result is output.

[0028] The microscopic multi-field stroboscopic acquisition module specifically includes an array light source, a polarizer, an analyzer, and an industrial camera; The array light source consists of a first light-emitting diode, a second light-emitting diode, a third light-emitting diode, and a fourth light-emitting diode. The first light-emitting diode, the second light-emitting diode, the third light-emitting diode, and the fourth light-emitting diode are evenly distributed in a ring around the optical axis of the industrial camera lens. The spatial angle between adjacent light-emitting diodes is set to 90 degrees. The central emission wavelength of the array light source is matched with the absorption spectrum of the target printed ink on the surface of the cigarette pack to be tested. Specifically, the central emission wavelength of the array light source coincides with the spectral absorption peak of the target printed ink. The wavelength matching setting makes the edge area of ​​the printed ink mark present high optical contrast during the image acquisition process, highlighting the physical morphological differences at the junction of the ink mark and paper fibers. There are four polarizers, which are fixedly installed at the light emission ends of the first light-emitting diode, the second light-emitting diode, the third light-emitting diode and the fourth light-emitting diode respectively. The polarizers modulate the natural light emitted by the array light source into linearly polarized light. The analyzer is mounted on the lens entrance of an industrial camera via a rotatable filter ring structure. By rotating the rotatable filter ring structure, the polarization direction of the analyzer can be adjusted so that the spatial angle between the polarization direction of the analyzer and the polarizer is kept at 90 degrees, thereby forming an orthogonal polarization optical path in the hardware space. The surface of cigarette packs is usually coated with a transparent reflective medium layer. When linearly polarized light emitted from the polarizer shines on the surface of the cigarette pack, the surface of the transparent reflective medium layer produces specular reflection light, while the paper fibers and ink at the bottom produce diffuse reflection light. The specular reflection light maintains the original polarization state of the incident light, while the diffuse reflection light undergoes multiple physical scatterings in the random fiber network inside the paper base material. The polarization state of the diffuse reflection light produces a depolarization effect and is transformed into unpolarized light. Since the analyzer and polarizer are in orthogonal polarization, the analyzer blocks the specular reflection light that maintains its original polarization state, preventing the specular reflection light from penetrating into the photosensitive element. The diffuse reflection light that undergoes depolarization passes through the analyzer and enters the photosensitive target surface of the industrial camera. The industrial camera then obtains a microscopic topography image with filtered surface reflection interference through the orthogonal polarization optical path, providing diffuse reflection physical image input for subsequent 3D topology calculations.

[0029] The microscopic multi-field image acquisition module specifically includes a mechanical phase synchronization unit, which comprises: a spindle rotary encoder, a phase signal conversion circuit board, and a trigger output interface. The main shaft rotary encoder is coaxially mounted on the end of the mechanical main shaft of the cigarette packing production line. The mechanical main shaft and the lever mechanism on the packaging production line are synchronously driven by the mechanical cam. The main shaft rotary encoder rotates with the mechanical main shaft and continuously outputs real-time mechanical angle pulse signals. The input terminal of the phase signal conversion circuit board is electrically connected to the signal output terminal of the spindle rotary encoder to receive the real-time mechanical angle pulse signal sent by the spindle rotary encoder; During the process of conveying the cigarette packs to be tested in the cigarette packing production line, the lever mechanism has periodic mechanical motion stagnation points during the reciprocating reversing motion phase. The cigarette packs to be tested are physically stationary at the mechanical motion stagnation points. During the system calibration phase, the packaging production line is moved to the mechanical motion stagnation point by jogging operation. The mechanical spindle angle output by the spindle rotary encoder at this time is read, and the read mechanical spindle angle is written into the register inside the phase signal conversion circuit board as the target phase threshold. The phase signal conversion circuit board analyzes the real-time mechanical angle pulse signal in real time. The phase signal conversion circuit board is equipped with a digital comparator. When the real-time mechanical angle pulse signal is consistent with the target phase threshold, the digital comparator generates a synchronous trigger level. The input terminal of the trigger output interface is connected to the phase signal conversion circuit board, and the output terminal of the trigger output interface is connected to the timing controller of the array light source and the external trigger pin of the industrial camera respectively. The trigger output interface sends the synchronous trigger level to the array light source and the industrial camera in parallel. After receiving the synchronous trigger level, the array light source performs microsecond-level strobe, and after receiving the synchronous trigger level, the industrial camera synchronously performs the exposure operation. Through the above-mentioned mechanical phase electrical signal linkage, the movements of the array light source and the industrial camera are limited to the time window when the smoke shell under test is in a physically stationary state, so as to obtain image data without motion blur.

[0030] The microscopic multi-field-of-view stroboscopic acquisition module specifically includes an array light source, an industrial camera, and a timing controller. The microsecond-level multi-channel stroboscopic and image acquisition process includes the following steps: Step 101: The timing controller receives the trigger signal and starts the timing logic. The timing controller is electrically connected to the trigger output interface of the mechanical phase synchronization unit to receive the synchronous trigger level sent by the mechanical phase synchronization unit. The synchronous trigger level indicates that the cigarette pack under test is in the mechanical motion stagnation state. After receiving the synchronous trigger level, the timing controller starts the microsecond-level timing generation logic to generate four independent drive pulse signals. Step 201: The timing controller controls the array light source and the industrial camera to perform synchronous stroboscopic exposure. Four drive pulse signals are transmitted to the first, second, third and fourth light-emitting diodes in the array light source, respectively. The timing controller controls the first light-emitting diode to generate microsecond-level pulse flashes and simultaneously sends a synchronous exposure signal to the industrial camera. The exposure time window of the synchronous exposure signal set by the timing controller covers the microsecond-level pulse flash width of the light-emitting diode to ensure that the photosensitive target surface fully absorbs the flash energy. During the flash of the first light-emitting diode, the industrial camera performs the first global shutter exposure to acquire the first microscopic morphology image under the first illumination direction.

[0031] After the first microscopic morphology image is acquired, the timing controller controls the second, third, and fourth light-emitting diodes to generate microsecond-level pulse flashes according to the preset time interval. The industrial camera synchronously performs global shutter exposure during each light-emitting diode flash, thereby acquiring the second microscopic morphology image under the second illumination direction, the third microscopic morphology image under the third illumination direction, and the fourth microscopic morphology image under the fourth illumination direction.

[0032] The four strobes of the array light source and the four exposures of the industrial camera are completed continuously within the time window of the synchronous trigger level. Since the time window corresponds to the mechanical motion stationary point of the cigarette pack to be tested, the image acquisition process eliminates pixel-level dynamic blur caused by the conveyor belt movement. The first microscopic morphology image, the second microscopic morphology image, the third microscopic morphology image and the fourth microscopic morphology image together constitute a grayscale image set and are transmitted to the surface microscopic normal vector reconstruction module.

[0033] The surface micro-normal vector reconstruction module that performs the above-mentioned micro-three-dimensional feature calculation specifically includes a specular reflection filtering unit and a Lambertian body model establishment unit.

[0034] The surface microscopic normal vector reconstruction module is communicatively connected to the microscopic multi-field stroboscopic acquisition module. It is used to receive a grayscale image set containing four images with different illumination directions. The image data in the grayscale image set is generated by converting diffuse reflection light signals filtered by orthogonal polarization optical paths.

[0035] The reflective medium layer on the surface of the cigarette pack generates specular reflection and diffuse reflection components under illumination. The specular reflection component exists on the surface of the reflective medium layer, causing optical highlights and masking the underlying paper fiber texture. The diffuse reflection component is formed by multiple scattering of light as it penetrates the reflective medium layer and enters the internal network of paper fibers. Based on the optical orthogonal polarization filtering structure of the microscopic multi-field stroboscopic acquisition module, the pixel brightness values ​​in the grayscale image set obtained by the specular reflection filtering unit are represented by a single diffuse reflection component.

[0036] The diffuse reflection component follows the physical properties of Lambertian reflection in terms of spatial luminance distribution. The Lambertian model establishes the microscopic surface of the target printing area as an ideal Lambertian diffuse reflection surface. In the ideal Lambertian diffuse reflection surface model, the observed grayscale of a specific pixel under illumination from light sources in different directions has a linear mapping relationship with the incident direction of the light source and the normal vector of the microscopic surface where the specific pixel is located.

[0037] The Lambertian model establishment unit establishes a local camera three-dimensional coordinate system. In the local camera three-dimensional coordinate system, the coordinates of any pixel in the target area are set as (x, y). The Lambertian model establishment unit obtains the spatial direction vectors of the four light-emitting diodes in the local camera three-dimensional coordinate system through the spatial calibration of the system in the early stage, and combines the four spatial direction vectors into a 4×3 illumination direction matrix L.

[0038] For a pixel (x, y), let the unit normal vector of the real surface be n(x, y), and the surface albedo be... The Lambertian model establishes the unit to extract the observed gray values ​​of the pixel points (x,y) in the gray image set corresponding to four different lighting directions, and constructs an observed gray column vector I(x,y) with a dimension of 4×1.

[0039] The Lambert model establishes the unit based on the Lambert law of reflection, defining the observed gray-level column vector I(x,y) and the surface albedo. The linear equation between the illumination direction matrix L and the surface unit normal vector n(x,y) is used to construct the following fundamental equation for photometric solids:

[0040] By constructing the aforementioned photometric three-dimensional fundamental equations, the Lambertian model establishment unit transforms discrete two-dimensional grayscale image data into a mathematical operation structure for physical three-dimensional topological features, providing an underlying computational framework for subsequently solving the three-dimensional normal vectors of each pixel. The surface microscopic normal vector reconstruction module that performs the aforementioned microscopic three-dimensional feature calculations specifically includes an equation system construction unit and a least squares solution unit.

[0041] The communication bus inside the surface microscopic normal vector reconstruction module transmits the fundamental photometric solid equations generated by the Lambertian model establishment unit to the equation system construction unit. The fundamental photometric solid equations include a 4×1 observation grayscale column vector I(x,y), a 4×3 illumination direction matrix L, and surface albedo. And the surface unit normal vector n(x,y).

[0042] The equation system construction unit defines the unnormalized surface normal vector u(x,y) and defines the unnormalized surface normal vector u(x,y) as the surface albedo. The scalar product with the surface unit normal vector n(x,y), based on the above definition, allows the equation system building unit to perform algebraic substitutions on the fundamental equations of the photometric solid, constructing a linear equation system:

[0043] Since the microscopic multi-field stroboscopic acquisition module contains the above four light-emitting diodes, the illumination direction matrix L provides four independent observation dimensions, so that the above linear equation system contains four independent equations. The unnormalized surface normal vector u(x,y) contains three unknown spatial components. Given that the number of independent equations is greater than the number of unknown spatial components, the equation system construction unit thus establishes the above linear equation system as an overdetermined linear equation system.

[0044] The least squares solution unit receives the overdetermined linear equation system. Addressing the redundancy in multi-field physical observations, the least squares solution unit uses the least squares method to solve the overdetermined linear equation system. The least squares solution unit simultaneously left-multiplies both sides of the equality sign of the overdetermined linear equation system by the transpose of the illumination direction matrix L. T Transform the overdetermined linear equation system into normal equations:

[0045] In the front-end optical hardware design, the annular non-coplanar distribution of the four light-emitting diodes ensures the product matrix L T L is full rank, thus making the product matrix L T L possesses the invertibility property, and the least squares solution unit multiplies the product matrix L in the normal equation. T L performs inverse matrix operations, using symbolic... -1 Representing the inverse matrix operation, through equality transformations, the least squares solution unit extracts the analytical solution formula for the unnormalized surface normal vector u(x,y):

[0046] The least squares solution unit extracts the measured pixel grayscale data and the spatial orientation matrix data calibrated by the system from the grayscale image set. The extracted measured pixel grayscale data and spatial orientation matrix data are then input into the analytical solution formula to perform matrix multiplication and inversion numerical calculations, and the algebraic operation result of the unnormalized surface normal vector u(x,y) corresponding to a specific pixel (x,y) in the target printing area is obtained.

[0047] The surface micro-normal vector reconstruction module that performs the above-mentioned micro-three-dimensional feature calculations specifically includes an albedo separation unit, a normal vector normalization unit, and a normal vector field generation unit.

[0048] The albedo separation unit receives the unnormalized surface normal vector u(x,y) output by the least squares solution unit. This unnormalized surface normal vector u(x,y) physically includes surface reflectivity and three-dimensional direction attributes. The three spatial components of the unnormalized surface normal vector u(x,y) are defined as u... x u y u zThe albedo separation unit performs vector modulo operation on the unnormalized surface normal vector u(x,y) to calculate the surface albedo corresponding to a specific pixel (x,y) within the target printing area. The calculation formula is:

[0049] Surface albedo Characterize the intrinsic reflectivity of the paper fiber-printed ink interface and extract surface albedo. The technical purpose is to eliminate the light and dark interference caused by changes in the direction of illumination and the geometric irregularities of the surface.

[0050] The normalization unit and the albedo separation unit are communicatively connected. The normalization unit extracts the unnormalized surface normal vector u(x,y) and the surface albedo. The unnormalized surface normal vector u(x,y) is divided by the surface albedo. This process strips away the reflectivity attribute, and the normalized surface normal vector n(x,y) is calculated using the normalized unit. The calculation formula is as follows:

[0051] After vector division, the normalized surface normal vector n(x,y) has a constant vector magnitude of 1, retaining the spatial three-dimensional direction components representing the undulation of the physical surface.

[0052] The normal vector field generation unit and the normal vector normalization unit are connected. Since the image acquisition process is based on the two-dimensional pixel array of the industrial camera, the normal vector field generation unit sets up coordinate traversal loop logic to control the albedo separation unit and the normal vector normalization unit to traverse the coordinates of all pixels in the target printing area and summarize the normalized surface normal vector n(x,y) corresponding to all pixel coordinates.

[0053] The normal vector field generation unit arranges and combines the normalized surface normal vectors into a three-dimensional micro normal vector field map according to the two-dimensional topological position in pixel space. The pixel resolution of the target printing area is set to M×N. The three-dimensional micro normal vector field map is constructed as a multi-dimensional matrix with dimensions of M×N×3 in terms of data structure. The multi-dimensional matrix stores the normal depth data of the ink penetration morphology along the micro gaps in the paper fibers within the target printing area. The surface micro normal vector reconstruction module transmits the three-dimensional micro normal vector field map to the subsequent ink fiber topology hash encoding module, providing the underlying data input for subsequent feature dimensionality reduction operations.

[0054] The ink fiber topology hash encoding module that performs the above micro-topological feature dimensionality reduction and hash encoding specifically includes a depth component extraction unit and a spatial gradient calculation unit. The depth component extraction unit is communicatively connected to the surface micro normal vector reconstruction module and receives a three-dimensional micro normal vector field map. The three-dimensional micro normal vector field map contains the normalized surface normal vector n(x,y) of all pixels in the target printing area.

[0055] The normalized surface normal vector n(x,y) contains three orthogonal spatial coordinate axis components. The depth component extraction unit extracts the normalized surface normal vector n. z The normal depth component n in (x,y) perpendicular to the physical surface of the target printing area z (x,y) is used to characterize the physical unevenness formed by ink penetration into the microscopic gaps in paper fibers.

[0056] The spatial gradient calculation unit is communicatively connected to the depth component extraction unit, and receives the normal depth component n of each pixel within the target printing area. z (x,y), taking advantage of the discrete characteristics of the image matrix, the spatial gradient calculation unit introduces a discrete pixel difference operator for the normal depth component n. z (x,y) performs partial derivative operations along two orthogonal directions of the two-dimensional image plane.

[0057] The two orthogonal directions of the two-dimensional image plane are defined as the horizontal x-direction and the vertical y-direction, respectively. The spatial gradient calculation unit calculates the normal depth component n. z Partial derivative of (x,y) with respect to the horizontal direction x And the partial derivative in the vertical direction y .

[0058] Based on the partial derivatives calculated above, the spatial gradient calculation unit performs sum of squares and square root operations to calculate the magnitude of the two-dimensional spatial gradient G(x,y) at a specific pixel (x,y). The calculation formula is as follows:

[0059] The spatial gradient calculation unit traverses all pixels within the target printing area, summarizes the two-dimensional spatial gradient magnitude G(x,y) of each pixel, and generates a two-dimensional spatial gradient magnitude matrix. The two-dimensional spatial gradient magnitude matrix records the relative rate of change of micro-depth within the target printing area.

[0060] The rate of change of depth at the interface between ink and paper fibers is represented as high-frequency response data in the two-dimensional spatial gradient amplitude matrix. The spatial gradient calculation unit has a preset gradient amplitude judgment threshold T. g The spatial gradient calculation unit will determine the gradient magnitude threshold T in the two-dimensional spatial gradient magnitude matrix. gThe numerical points are selected as high-frequency response data, thereby locating the physical topological boundary formed by the random penetration of ink along the gaps in paper fibers in the spatial topological dimension. The spatial gradient calculation unit outputs a two-dimensional spatial gradient magnitude matrix containing the physical topological boundary, providing basic coordinate data for subsequent local feature encoding.

[0061] The ink fiber topology hash encoding module, which performs the above-mentioned microscopic topological feature dimensionality reduction and hash encoding, specifically includes a neighborhood sampling unit and a binary pattern calculation unit.

[0062] The neighborhood sampling unit receives a two-dimensional spatial gradient magnitude matrix containing the physical topological boundary, wherein the two-dimensional spatial gradient magnitude matrix contains the two-dimensional spatial gradient magnitude of each pixel within the target printing area.

[0063] For a specific pixel in the two-dimensional spatial gradient magnitude matrix, the neighborhood sampling unit sets the specific pixel as the center pixel, extracts the two-dimensional spatial gradient magnitude of the center pixel, and sets the two-dimensional spatial gradient magnitude of the center pixel as the center gradient magnitude g. c .

[0064] The neighborhood sampling unit constructs a circular sampling neighborhood on the two-dimensional spatial gradient magnitude matrix, with the center pixel as the center and a preset physical pixel distance as the sampling radius R. P sampling points are uniformly selected on the circumference of this circular neighborhood. Since the coordinates of the sampling points on the circumference of the circular neighborhood may contain non-integer pixel coordinates, the neighborhood sampling unit uses a bilinear interpolation algorithm to calculate the two-dimensional spatial gradient magnitude at these non-integer pixel coordinates. The neighborhood sampling unit extracts the two-dimensional spatial gradient magnitudes corresponding to these P sampling points and sets the extracted two-dimensional spatial gradient magnitudes as the neighborhood gradient magnitude g. p The range of values ​​for the subscript p is: .

[0065] The binary mode calculation unit is communicatively connected to the neighborhood sampling unit and receives the central gradient magnitude g. c and all neighborhood gradient magnitudes g p The binary mode calculation unit will calculate the center gradient magnitude g. c Set the comparison threshold, and calculate the gradient magnitude g of each neighborhood. p With the center gradient magnitude g c The difference between them is then binarized using a sign step function.

[0066] Set the sign step function as The independent variable is defined as follows: Signed step function The calculation formula is:

[0067] The binary mode computation unit converts the relative gradient changes between each sampling point in the circular sampling neighborhood and the center pixel into binary feature states by executing the symbolic step function operation.

[0068] The binary mode calculation unit assigns a corresponding binary weight to each sampling point on the circumference of the circular sampling neighborhood. p The binarized binary feature state is then multiplied and added with the corresponding binary weights to calculate the local binary mode response value corresponding to the center pixel. Local binary mode response value The calculation formula is:

[0069] The binary mode calculation unit establishes a traversal loop mechanism to control the neighborhood sampling unit and the binary mode calculation unit to traverse all pixels in the two-dimensional spatial gradient magnitude matrix and execute the above calculation process, based on the local binary mode response values ​​corresponding to all pixels. The binary mode computation unit generates the local binary mode response matrix.

[0070] The local binary mode response matrix records the spatial distribution characteristics of the physical topological boundary within the target printing area. The binary mode calculation unit outputs the local binary mode response matrix, providing structured basic data for the subsequent generation of physical hash sequences.

[0071] The ink fiber topology hash encoding module, which performs the above-mentioned microscopic topological feature dimensionality reduction and hash encoding, specifically includes a spatial pooling unit, a binarization unit, and a sequence splicing unit.

[0072] The spatial pooling unit is communicatively connected to the binary pattern calculation unit and receives the local binary pattern response matrix. The local binary pattern response matrix contains the feature distribution data of the physical topological boundary within the target printing area. The spatial pooling unit extracts the preset hash sequence target length value K. Based on the hash sequence target length value K, the spatial pooling unit divides the local binary pattern response matrix into K non-overlapping local grid regions in a two-dimensional spatial dimension by cutting the matrix in a row-column proportional manner.

[0073] For each local grid region in the local binary mode response matrix, the spatial pooling unit extracts the local binary mode response values ​​of all pixels in the local grid region, performs an arithmetic mean operation on the extracted local binary mode response values, calculates the feature representative value corresponding to each local grid region, and the spatial pooling unit summarizes the feature representative values ​​of all local grid regions to generate a one-dimensional feature array of length K.

[0074] The binarization unit communicates with the spatial pooling unit, receives a one-dimensional feature array, sorts all the feature representative values ​​in the one-dimensional feature array, extracts the median of the sorted result, and sets the extracted median as the hash quantization threshold T. h .

[0075] The binarization unit reads the feature representative values ​​from the one-dimensional feature array one by one, and compares the read feature representative values ​​with the hash quantization threshold T. h Perform a numerical comparison; when the feature representative value is greater than or equal to the hash quantization threshold T... h When the binary value is less than the hash quantization threshold T, the binarization unit outputs a binary value of 1; when the feature representative value is less than the hash quantization threshold T, the binary value is less than the hash quantization threshold T. h At that time, the binarization unit outputs the binary value 0. Through the above numerical comparison operation, the binarization unit generates a set of binary values ​​corresponding to the one-dimensional feature array.

[0076] The sequence concatenation unit is communicatively connected to the binarization unit, receiving the set of binary values ​​output by the binarization unit. Following the initial two-dimensional spatial coordinates of the local grid region in the local binary mode response matrix, the sequence concatenation unit sequentially concatenates the generated K binary values ​​to generate a fixed-length physical hash sequence H of length K bits. reg .

[0077] Fixed-length physical hash sequence H reg The mapping module determines the irreversible physical topology of the ink-paper fiber interface within the target printing area. The ink-fiber topology hash encoding module then encodes a fixed-length physical hash sequence H. reg The data is transmitted to the high-concurrency anti-counterfeiting and traceability verification module, providing a unique retrieval benchmark for subsequent factory registration and terminal verification.

[0078] The high-concurrency anti-counterfeiting and traceability verification module that performs the above physical fingerprint registration and verification comparison specifically includes a data packaging unit, a cloud communication unit, and a cloud database.

[0079] The data packaging unit communicates with the ink fiber topology hash encoding module and receives a fixed-length physical hash sequence H. reg The data packaging unit synchronously connects to the manufacturing execution system of the cigarette packing production line to obtain the production plaintext information of the cigarette packing to be tested. The production plaintext information includes the production timestamp, production batch number, and production line identification code. The data packaging unit then packages the fixed-length physical hash sequence H... reg Combine and encapsulate the production plaintext information to generate a product registration data packet.

[0080] The cloud communication unit communicates with the data packaging unit to extract the product registration data packet. The cloud communication unit is configured with an encrypted transmission protocol and establishes a wide area network communication link with the cloud database through this protocol. The cloud communication unit performs transport layer encryption on the product registration data packet and sends the encrypted packet to the cloud database. The cloud database receives and decrypts the data packet, parses the underlying message structure, and extracts a fixed-length physical hash sequence H. reg And the corresponding production plaintext information, the cloud database internally constructs a relational data mapping table, and the cloud database uses a fixed-length physical hash sequence H reg A primary key index is set, and the production plaintext information is configured as the associated field of the primary key index. It is synchronously written into the relational data mapping table. After the cloud database completes the data writing operation, it generates a registration completion status code. The cloud database sends the registration completion status code to the cloud communication unit at the production line's outgoing end. Based on the received registration completion status code, the high-concurrency anti-counterfeiting and traceability verification module confirms that the physical topological characteristics of the cigarette pack to be tested and the production plaintext information are bound and stored at the outgoing end, providing a structured retrieval benchmark for anti-counterfeiting verification in subsequent circulation links.

[0081] The high-concurrency anti-counterfeiting and traceability verification module that performs the above-mentioned physical fingerprint registration and verification comparison specifically includes an audit verification terminal and an on-site hash generation unit.

[0082] The inspection and verification terminal is configured as a portable handheld device for reading the microscopic topological features of the cigarette pack to be tested during the circulation process. The hardware front end of the inspection and verification terminal integrates a portable microscopic multi-field stroboscopic acquisition mechanism. The portable microscopic multi-field stroboscopic acquisition mechanism replicates the optical extraction logic of the production line at the factory end. Internally, it includes a ring-shaped array of micro light-emitting diodes, a handheld macro camera, and an orthogonal polarizing lens group.

[0083] To eliminate spatial attitude errors caused by handheld operation, a fitted positioning light shield is fixedly installed at the front end of the portable microscopic multi-field stroboscopic acquisition mechanism. When inspectors operate the inspection and verification terminal at the anti-counterfeiting verification site, they press the end face of the fitted positioning light shield tightly against the surface of a specific printed area on the cigarette pack to be tested. The fitted positioning light shield isolates stray light from the external environment and forcibly locks the relative working distance and spatial pitch angle between the handheld macro camera and the surface of the cigarette pack to be tested, thereby ensuring that the illumination direction matrix of the micro LED array is strictly consistent with the illumination direction matrix calibrated at the production line. The central emission wavelength of the miniature LED array is consistent with the central emission wavelength of the array light source configured at the production line's factory end. Orthogonal polarizing lenses are distributed at the emission end of the miniature LED array and the incident end of the handheld macro camera, maintaining orthogonal polarization directions to block specular reflection light from the surface of the cigarette packaging under test. The inspection and verification terminal controls the miniature LED array to perform multi-channel strobe and controls the handheld macro camera to simultaneously expose, acquiring a set of diffuse reflection grayscale images of the area under test. The on-site hash generation unit is built into the processor of the inspection and verification terminal. It receives the diffuse reflection grayscale image set to be tested on-site and calls the same surface microscopic normal vector reconstruction algorithm and local binary pattern feature dimensionality reduction algorithm as the production line's outgoing end.

[0084] The on-site hash generation unit, based on the Lambertian diffuse reflection physical model, performs least squares solution and normalization processing on the on-site diffuse reflection grayscale image set to be measured, generating an on-site microscopic three-dimensional normal vector field map. Furthermore, the on-site hash generation unit calculates the spatial gradient magnitude of the on-site microscopic three-dimensional normal vector field map and extracts the local binary mode response values ​​corresponding to the high-frequency response data.

[0085] After spatial block pooling and threshold binarization, the field hash generation unit concatenates the local binary pattern response values ​​into a fixed-length field physical topology feature sequence. The field hash generation unit defines the generated field physical topology feature sequence as the hash sequence to be tested, H. query The hash sequence to be tested H query Data length and fixed-length physical hash sequence H reg The data lengths are completely consistent, and the on-site hash generation unit will generate the hash sequence H to be tested. query The data is cached in the memory of the inspection and verification terminal to provide on-site measured data input for subsequent cloud-based comparison and measurement.

[0086] The high-concurrency anti-counterfeiting and traceability verification module that performs the above physical fingerprint registration and verification comparison specifically includes a sequence retrieval unit, a distance measurement unit, and a traceability discrimination unit.

[0087] The sequence retrieval unit establishes a wide area network communication link with the cloud database. Based on plaintext association fields such as the production batch number attached to the surface of the cigarette pack to be tested, the sequence retrieval unit sends a traceability retrieval request to the cloud database. The cloud database performs primary key matching in the relational data mapping table and extracts the corresponding fixed-length physical hash sequence H. reg The sequence retrieval unit receives a fixed-length physical hash sequence H from the cloud database. reg .

[0088] The distance metric unit communicates with both the sequence retrieval unit and the on-site hash generation unit. The distance metric unit extracts the fixed-length physical hash sequence H sent from the cloud.reg Simultaneously extract the hash sequence H to be tested from the memory of the audit and verification terminal. query Fixed-length physical hash sequence H reg With the hash sequence to be tested H query The length of each binary sequence is K.

[0089] Distance metric unit extracts fixed-length physical hash sequence H bit by bit reg With the hash sequence to be tested H query For the binary values ​​at the same index k, the distance metric unit performs a logical XOR operation on the two extracted binary values. After traversing all K index bits, the distance metric unit performs an accumulation operation on all the logical XOR operation results to calculate the Hamming distance D, which represents the sequence dissimilarity. H The calculation formula is:

[0090] Among them, symbols This represents a logical XOR operation, where the index k ranges from 1 to K, which are positive integers. reg [k] represents the k-th binary value in the fixed-length physical hash sequence, H query [k] represents the k-th binary value in the hash sequence to be tested.

[0091] Cigarette packaging undergoes surface paper fiber wear and microscopic deformation during physical distribution. To eliminate the interference of legitimate microscopic deformation caused by physical distribution on the verification results, the traceability judgment unit is equipped with a tolerance threshold T. th Tolerance threshold T th The Hamming distance of genuine wear samples is pre-calibrated through statistical testing and used as a quantitative reference benchmark for determining authenticity.

[0092] The source tracing and discrimination unit receives the calculated Hamming distance D. H And the distance between Hamming and D H With tolerance threshold T th Compare the size of the row values.

[0093] When Hamming distance D H Less than or equal to the tolerance threshold T th At that time, the traceability discrimination unit determines that the physical topological features extracted on site match the factory registration data, confirms that the cigarette pack to be tested is genuine, and outputs a traceability success result containing production plaintext information to the inspection and verification terminal.

[0094] When Hamming distance D H Greater than the tolerance threshold T thWhen the source tracing and discrimination unit determines that the physical topological boundary extracted on site has undergone substantial change, the source tracing and discrimination unit confirms that the cigarette pack to be tested is a counterfeit or illegal product, and outputs a warning and interception signal to the inspection and verification terminal to block the source tracing.

[0095] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that modifications can be made to these embodiments without departing from the principles and spirit of the invention.

Claims

1. A vision-based traceability system for cigarette pack printing, characterized in that, include: The microscopic multi-field stroboscopic acquisition module is set at the inspection station of the cigarette packing production line. It is used to acquire grayscale image sets of a specific printed area on the surface of the cigarette pack under different lighting directions through multi-channel stroboscopic and synchronous exposure. The surface micro normal vector reconstruction module is communicatively connected to the micro multi-field strobe acquisition module. It is used to receive the grayscale image set, calculate the normalized surface normal vector of each pixel in the target printing area based on the Lambertian diffuse reflection physical model, and generate a three-dimensional micro normal vector field map. The ink fiber topological hash encoding module is communicatively connected to the surface micro normal vector reconstruction module. It is used to extract the normal depth component in the three-dimensional micro normal vector field map, perform local binary pattern feature extraction and dimensionality reduction on the normal depth component, and generate a fixed-length physical hash sequence. The high-concurrency anti-counterfeiting and traceability verification module is communicatively connected to the ink fiber topology hash encoding module. It is used to bind the fixed-length physical hash sequence with the production plaintext information and store it in the cloud database during the product manufacturing stage, and to obtain the test hash sequence on site during the terminal verification stage. It outputs the product traceability judgment result by measuring the Hamming distance between the test hash sequence and the fixed-length physical hash sequence.

2. The vision-based cigarette pack printing traceability system according to claim 1, characterized in that, The microscopic multi-field stroboscopic acquisition module includes an array light source, four polarizers, an analyzer, an industrial camera, and a mechanical phase synchronization unit. The array light source consists of four light-emitting diodes, which are evenly distributed in a ring around the optical axis of the industrial camera lens. The four polarizers are fixedly installed at the light-emitting ends of the four light-emitting diodes. The analyzer is installed at the lens entrance end of the industrial camera lens, and the polarization direction of the analyzer is at a 90-degree spatial angle with the polarization directions of the four polarizers to form an orthogonal polarization light path. The mechanical phase synchronization unit is electrically connected to the main shaft rotary encoder of the cigarette packing production line. When the real-time mechanical angle pulse signal output by the main shaft rotary encoder reaches the target phase threshold, it generates a synchronous trigger level to trigger the array light source to perform strobe sequentially and trigger the industrial camera to perform synchronous exposure.

3. The vision-based cigarette pack printing traceability system according to claim 1, characterized in that, The surface microscopic normal vector reconstruction module includes a Lambertian model establishment unit, an equation system construction unit, and a least squares solution unit; The Lambertian model establishment unit is used to extract the observed gray values ​​of specific pixels in the grayscale image set under different illumination directions to construct the observed grayscale column vector, and to establish the photometric solid fundamental equation in combination with the pre-calibrated illumination direction matrix. The equation system construction unit is used to define the unnormalized surface normal vector as a scalar product of the surface albedo and the surface unit normal vector, and to perform algebraic substitution on the photometric stereo fundamental equations to construct an overdetermined linear equation system. The least squares solving unit is used to simultaneously left-multiply the transpose of the illumination direction matrix on both sides of the equals sign of the overdetermined linear equation system to convert it into a normal equation. By extracting the inverse matrix of the product of the illumination direction matrices, the unnormalized surface normal vector corresponding to the specific pixel point is solved.

4. The vision-based cigarette pack printing traceability system according to claim 3, characterized in that, The surface micro normal vector reconstruction module also includes an albedo separation unit, a normal vector normalization unit, and a normal vector field generation unit. The albedo separation unit is used to perform vector modulo operation on the unnormalized surface normal vector to calculate the surface albedo corresponding to the specific pixel. The normal vector normalization unit is used to divide the unnormalized surface normal vector by the surface albedo, strip away the reflectivity attribute, and calculate the normalized surface normal vector. The normal vector field generation unit is used to traverse the coordinates of all pixels within the target printing area and arrange the summarized normalized surface normal vectors into the three-dimensional micro normal vector field map, wherein the three-dimensional micro normal vector field map is constructed as a multi-dimensional matrix storing normal depth data.

5. A vision-based traceability system for cigarette pack packaging printing according to claim 1, characterized in that, The ink fiber topological hash encoding module includes a depth component extraction unit, a spatial gradient calculation unit, a neighborhood sampling unit, and a binary pattern calculation unit; The depth component extraction unit is used to extract the normal depth component perpendicular to the physical surface of the target printing area from the normalized surface normal vector; The spatial gradient calculation unit uses the discrete pixel difference operator to calculate the partial derivatives of the normal depth component in the horizontal and vertical directions, generates a two-dimensional spatial gradient magnitude matrix by performing sum of squares and square root operations, and filters high-frequency response data based on the gradient magnitude judgment threshold to locate the physical topological boundary. The neighborhood sampling unit is used to set a specific pixel in the two-dimensional spatial gradient magnitude matrix as the center pixel, construct a circular sampling neighborhood on the two-dimensional spatial gradient magnitude matrix with the center pixel, and extract the neighborhood gradient magnitude on the sampling circumference. The binary mode calculation unit is used to binarize the difference between the neighborhood gradient magnitude and the central gradient magnitude using a symbolic step function, and generate a local binary mode response value by multiplying and adding it with binary weights, thereby generating a local binary mode response matrix.

6. A vision-based traceability system for cigarette pack packaging printing according to claim 5, characterized in that, The ink fiber topology hash encoding module also includes a spatial pooling unit, a binarization unit, and a sequence splicing unit; The spatial pooling unit is used to divide the local binary pattern response matrix into non-overlapping local grid regions according to the target length value of the hash sequence, and calculate the feature representative value corresponding to each local grid region to generate a one-dimensional feature array. The binarization unit is used to extract the median of the feature representative values ​​in the one-dimensional feature array as the hash quantization threshold, and to generate a set of binary values ​​corresponding to the one-dimensional feature array through numerical comparison operations. The sequence splicing unit is used to splice the binary value set according to the initial two-dimensional spatial coordinates of the local grid region to generate the fixed-length physical hash sequence.

7. A vision-based traceability system for cigarette pack packaging printing according to claim 1, characterized in that, The high-concurrency anti-counterfeiting and traceability verification module includes a data packaging unit, a cloud communication unit, and a cloud database; The data packaging unit is used to obtain the production timestamp, production batch number, and production line identification code of the cigarette pack to be tested as production plaintext information, and to combine and encapsulate the fixed-length physical hash sequence with the production plaintext information to generate a product registration data packet; The cloud communication unit is used to perform transport layer encryption on the product registration data packet and send it to the cloud database; The cloud database is used to decrypt the product registration data packet, set the fixed-length physical hash sequence as the primary key index in the relational data mapping table, and synchronously write the production plaintext information as the associated field of the primary key index.

8. A vision-based traceability system for cigarette pack packaging printing according to claim 1, characterized in that, The high-concurrency anti-counterfeiting and traceability verification module includes an inspection and verification terminal, which is configured as a portable handheld device; The hardware front end of the inspection and verification terminal integrates a portable microscopic multi-field stroboscopic acquisition mechanism, which contains a handheld macro camera. The front end of the portable microscopic multi-field stroboscopic acquisition mechanism is fixedly installed with a fitting positioning light shield. The fitted positioning light shield is used to press tightly against the surface of a specific printed area on the cigarette pack to be tested, so as to physically lock the relative working distance and spatial pitch angle of the handheld macro camera; the end face of the fitted positioning light shield is equipped with a pressure contact switch, which is used to generate a level trigger signal when a preset pressure fit is reached, so as to control the inspection and verification terminal to acquire the diffuse reflection grayscale image set to be tested on site.

9. A vision-based traceability system for cigarette pack packaging printing according to claim 8, characterized in that, The high-concurrency anti-counterfeiting and traceability verification module also includes a field hash generation unit built into the processor of the inspection and verification terminal; The on-site hash generation unit is used to receive the set of diffuse reflection grayscale images to be tested on-site, and perform least squares solution and normalization processing to generate a microscopic three-dimensional normal vector field map of the on-site. Then, through spatial block pooling and dynamic median threshold binarization, the local binary pattern response values ​​are concatenated and transformed into the test hash sequence with the same data length as the fixed-length physical hash sequence.

10. A vision-based traceability system for cigarette pack packaging printing according to claim 1, characterized in that, The high-concurrency anti-counterfeiting and traceability verification module includes a sequence retrieval unit, a distance measurement unit, and a traceability discrimination unit; The sequence retrieval unit is used to send a source retrieval request containing plaintext association fields to the cloud database to extract the corresponding fixed-length physical hash sequence; The distance metric unit is used to extract the binary values ​​of the fixed-length physical hash sequence and the hash sequence to be tested at the same index bit, perform a logical XOR operation, and perform an accumulation operation on the logical XOR operation result to calculate the Hamming distance that represents the difference between the sequences. The traceability discrimination unit is equipped with a tolerance threshold, which is obtained by pre-calibrating the Hamming distance of the genuine wear sample. When the Hamming distance is less than or equal to the tolerance threshold, the unit determines that the cigarette pack to be tested is genuine. When the Hamming distance is greater than the tolerance threshold, the unit outputs a warning interception signal to block the traceability.

Images