A method for intelligent identification and analysis of anti-counterfeiting marks

Abstract

The present invention relates to the field of commodity anti-counterfeiting technology, specifically to an intelligent identification and analysis method for anti-counterfeiting marks. First, environmental parameters are collected in real time, and a holographic image of the anti-counterfeiting mark is obtained by multi-spectral imaging technology. The image is corrected using environmental data to generate a standardized hologram; then, the microscopic anti-counterfeiting features in the dynamic dot matrix hologram are extracted to construct a holographic feature vector. Encrypted barcode data is collected synchronously, and the error correction parameters and decoding thresholds of the barcode decoding algorithm are dynamically adjusted through an environmental compensation model, and a first optimized barcode decoding algorithm is obtained for decryption. An asymmetric encryption algorithm is used to verify the timeliness matching of the hologram timestamp and the barcode dynamic key, and the matching degree of the real-time holographic feature vector and the pre-stored standard vector is calculated through a feature similarity algorithm. When the matching degree is greater than the preset threshold and the dynamic key verification is passed, it is determined to be authentic and the dynamic key is updated.

Description

Technical Field

[0001] The present invention relates to the technical field of commodity anti-counterfeiting, and in particular to an intelligent recognition and analysis method for anti-counterfeiting marks. Background Art

[0002] With the rapid development of social economy and the continuous innovation of information technology, product anti-counterfeiting has become an important link in safeguarding consumer rights, protecting corporate brand reputation and maintaining market order. At present, the market mainly uses traditional technologies such as QR code, barcode, optical recognition to verify anti-counterfeiting labels, but these methods often have problems such as low recognition accuracy and slow response when dealing with complex environments.

[0003] Existing technologies mainly rely on recognition algorithms based on image processing and feature matching. Their algorithm models are not stable enough in complex and changeable practical application scenarios and are easily affected by environmental interference and insufficient data. At the same time, traditional anti-counterfeiting systems generally lack the ability to adaptively learn and dynamically update, and it is difficult to meet the needs of e-commerce, financial bills, supply chain management, and administrative supervision for real-time, high-precision recognition and large-scale data analysis. Especially in application fields such as administration, commerce, finance, management or supervision, the requirements for real-time data processing, risk warning and regulatory decision-making are becoming increasingly high, and traditional methods have obvious limitations in information integration, dynamic response and intelligent judgment. In addition, with the continuous upgrading of counterfeiting technology, traditional static verification methods are difficult to respond to new counterfeiting methods in a timely manner, making it difficult to achieve an anti-counterfeiting closed loop.

[0004] Therefore, an intelligent recognition and analysis method for anti-counterfeiting marks is proposed. Summary of the invention

[0005] The purpose of the present invention is to provide an intelligent identification and analysis method for anti-counterfeiting marks. First, environmental parameters are collected in real time, and a holographic image of the anti-counterfeiting mark is obtained by multispectral imaging technology. The image is corrected using environmental data to generate a standardized hologram; then, the microscopic anti-counterfeiting features in the dynamic dot matrix hologram are extracted to construct a holographic feature vector. Encrypted barcode data is collected synchronously, and the error correction parameters and decoding thresholds of the barcode decoding algorithm are dynamically adjusted through the environmental compensation model to obtain a first optimized barcode decoding algorithm for decryption. An asymmetric encryption algorithm is used to verify the timeliness matching of the hologram timestamp and the barcode dynamic key, and the matching degree of the real-time holographic feature vector and the pre-stored standard vector is calculated by a feature similarity algorithm. When the matching degree is greater than the preset threshold and the dynamic key verification is passed, it is determined to be authentic and the dynamic key is updated.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] An intelligent identification and analysis method for anti-counterfeiting marks, comprising:

[0008] Real-time monitoring and collection of environmental data; collecting the holographic image of the anti-counterfeiting mark by multi-spectral imaging technology; adjusting the holographic image by environmental data to generate a standardized holographic image; extracting the microscopic anti-counterfeiting features of the dynamic dot matrix hologram from the standardized holographic image to generate a holographic feature vector;

[0009] Collect the encrypted barcode data of the anti-counterfeiting mark; construct an environmental compensation model to dynamically adjust the barcode decoding algorithm parameters, obtain a first optimized barcode decoding algorithm, decode the encrypted barcode data, and generate decoded barcode information;

[0010] The holographic feature vector and the barcode information are judged, and the specific process includes:

[0011] Compare the timeliness and consistency of the holographic image timestamp and the barcode dynamic key through an asymmetric encryption algorithm;

[0012] Calculate the similarity between the holographic feature vector and the pre-stored standard feature vector to generate the holographic feature matching degree;

[0013] If the holographic feature matching degree is not less than the preset threshold and the dynamic key verification is passed, it is judged to be authentic and a new dynamic key is generated and synchronized to the anti-counterfeiting mark; otherwise, the multi-level early warning mechanism is activated, the abnormal log is recorded and manual review is requested.

[0014] Preferably, the environmental data includes lighting conditions, shooting angle deviation, shooting distance deviation, ambient temperature, ambient humidity, mechanical vibration, mechanical interference, and physical state of the tag; and the holographic image includes visible light, infrared and ultraviolet band images.

[0015] Preferably, the step of adjusting the holographic image by using the environmental data to generate a standardized holographic image specifically includes:

[0016] Based on the light intensity data, an adaptive histogram equalization algorithm is used to adjust the image contrast, and a multi-scale Retinex algorithm is used to eliminate the color cast caused by uneven lighting.

[0017] According to the shooting angle deviation data, the affine transformation model is used to compensate for the image geometric distortion, and the original viewing angle of the hologram is restored in combination with the calibration parameters;

[0018] By using the camera distance deviation data, the bicubic interpolation algorithm is used to adjust the image resolution to the preset pixel density to ensure the clarity of the dynamic dot matrix pixels;

[0019] Based on the ambient temperature and humidity data, the image is nonlinearly stretched to compensate for the microstructural deformation caused by thermal expansion and contraction of the label material.

[0020] For image blur caused by mechanical vibration, a blind deconvolution algorithm based on motion estimation is used to restore high-frequency details;

[0021] According to the physical state of the label, a repair model is constructed using a generative adversarial network to identify and repair surface defects, wrinkles, and damaged areas in the holographic image and generate a standardized holographic image.

[0022] Preferably, extracting the microscopic anti-counterfeiting features of the dynamic dot matrix hologram from the standardized holographic image to generate a holographic feature vector specifically includes:

[0023] A multi-scale convolutional neural network is constructed. The input layer receives the standardized holographic image, and the hidden layer extracts multi-granular features through 3×3 convolution kernels, 5×5 convolution kernels and 7×7 convolution kernels. The feature maps of different scales are fused through the output layer to generate a multi-scale feature map. The entropy value of the pixel distribution in each channel, the diffraction grating phase difference and the gradient change caused by the viewing angle are calculated through the channel attention mechanism, and the channel weights of the multi-scale feature map are dynamically allocated to generate a multi-scale feature enhancement map. The local binary pattern and the directional gradient histogram are jointly encoded to quantize the microscopic arrangement pattern of the dynamic dot matrix to generate a comprehensive feature map. The comprehensive feature map is fused through the fully connected layer to generate a holographic feature vector.

[0024] Preferably, the environmental compensation model includes an environmental data preprocessing layer, a parameter mapping layer, an adaptive optimization layer and a result output layer;

[0025] The environmental data preprocessing layer processes the collected environmental data through a normalization algorithm and a feature extraction algorithm to generate an environmental feature vector;

[0026] The parameter mapping layer maps the environmental feature vector to a preliminary parameter adjustment value of the barcode decoding algorithm through a convolutional neural network;

[0027] The adaptive optimization layer generates dynamically optimized barcode decoding parameters by making real-time corrections to the preliminary parameter adjustment values ​​and comprehensively considering real-time environmental changes and historical verification data;

[0028] The result output layer generates a first optimized barcode decoding algorithm by applying the optimized barcode decoding parameters to the barcode decoding algorithm.

[0029] Preferably, the specific process of decoding the encrypted barcode data using the first optimized barcode decoding algorithm to generate the decoded barcode information is as follows:

[0030] The YOLOv5 model is used to locate the encrypted barcode area, and the barcode and the background are separated through morphological operations. The barcode image is denoised and binarized according to the parameters in the first optimized barcode decoding algorithm to obtain standard barcode data. The standard barcode data is decrypted using an asymmetric encryption algorithm to generate barcode information. The barcode information includes a dynamic key plaintext and a timestamp.

[0031] Preferably, the specific process of obtaining the holographic feature matching degree is as follows:

[0032] The holographic feature vector and the pre-stored standard feature vector are input into the twin neural network; the twin neural network has a dual-branch structure, specifically: branch one: input the real-time holographic feature vector, map it to the 128-dimensional embedding space through three fully connected layers, and generate a first embedding vector; branch two: input the pre-stored standard holographic feature vector, use the same network structure for mapping, and generate a second embedding vector;

[0033] The first embedding vector and the second embedding vector are measured by cosine similarity, Mahalanobis distance and dynamic weight allocation to generate a multidimensional similarity matrix; the multidimensional similarity matrix is ​​weightedly fused to obtain the holographic feature matching degree.

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] 1. The present invention significantly improves the stability and accuracy of anti-counterfeiting mark recognition by introducing multi-spectral imaging technology and environmental data adaptive adjustment method. First, the system monitors multiple environmental parameters such as light, angle, distance, temperature and humidity, and mechanical vibration in real time, and uses adaptive histogram equalization, multi-scale Retinex algorithm and affine transformation model to standardize the holographic image, effectively eliminating the influence of external interference on image color, contrast and geometry. Secondly, high-frequency details are restored through bicubic interpolation and motion estimation blind deconvolution technology to ensure the clear presentation of dynamic lattice microstructures and give full play to the advantages of holographic image details. This method improves the problem of easy distortion of traditional fixed parameter processing mode in complex environments, provides an efficient and accurate image preprocessing solution for anti-counterfeiting marks, and improves the overall anti-counterfeiting recognition level.

[0036] 2. The present invention adopts multi-scale convolutional neural network, channel attention mechanism and local binary pattern joint encoding to deeply mine the dynamic lattice features in the holographic image and realize the accurate extraction of micro anti-counterfeiting features. The twin neural network is used to calculate the multi-dimensional similarity between the real-time acquisition features and the standard features, and the cosine similarity, Mahalanobis distance and dynamic weight allocation are used for comprehensive evaluation, which significantly improves the accuracy of feature matching. At the same time, the environmental compensation model is innovatively constructed, which organically combines normalization preprocessing, convolution mapping and adaptive optimization, effectively corrects the image distortion caused by changes in illumination, angle, distance and temperature and humidity, and ensures the real-time dynamic adjustment of barcode decoding parameters.

[0037] 3. The present invention integrates the holographic image with the encrypted barcode information, realizes the dual verification of the timestamp and the dynamic key through an asymmetric encryption algorithm, and forms a multi-factor anti-counterfeiting verification system. YOLOv5 is used to locate the barcode area, and the dynamic key plaintext and time information are decoded by combining morphological processing and denoising binarization technology. The twin neural network is used to realize the multi-dimensional similarity measurement between the holographic feature and the standard feature vector to ensure that the matching degree reaches the preset threshold. The dual verification mechanism effectively prevents the risk of counterfeiting and tampering. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 A schematic diagram of a process flow of an intelligent identification and analysis method for anti-counterfeiting marks provided by the present invention;

[0039] Figure 2 A schematic diagram of the structure of an environmental compensation model provided by an embodiment of the present invention;

[0040] Figure 3 A schematic diagram of the intelligent identification structure of an anti-counterfeiting mark provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0041] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0042] Embodiment 1

[0043] See also Figure 1 The present invention provides an intelligent identification and analysis method for anti-counterfeiting marks, and the technical solution is as follows:

[0044] Real-time monitoring and collection of environmental data; collecting the holographic image of the anti-counterfeiting mark by multi-spectral imaging technology; adjusting the holographic image by environmental data to generate a standardized holographic image;

[0045] Furthermore, the environmental data includes lighting conditions, shooting angle deviation, shooting distance deviation, ambient temperature, ambient humidity, mechanical vibration, mechanical interference, and physical state of the tag; and the holographic image includes visible light, infrared, and ultraviolet band images.

[0046] In this embodiment, environmental data and multi-spectral holographic imaging are collected in real time to achieve image standardization, effectively correct lighting, angle, temperature and humidity interference, and improve anti-counterfeiting recognition accuracy.

[0047] Furthermore, the step of adjusting the holographic image by using the environmental data to generate a standardized holographic image specifically includes:

[0048] Based on the light intensity data, an adaptive histogram equalization algorithm is used to adjust the image contrast, and a multi-scale Retinex algorithm is used to eliminate the color cast caused by uneven lighting.

[0049] According to the shooting angle deviation data, the affine transformation model is used to compensate for the image geometric distortion, and the original viewing angle of the hologram is restored in combination with the calibration parameters;

[0050] By using the camera distance deviation data, the bicubic interpolation algorithm is used to adjust the image resolution to the preset pixel density to ensure the clarity of the dynamic dot matrix pixels;

[0051] Based on the ambient temperature and humidity data, the image is nonlinearly stretched to compensate for the microstructural deformation caused by thermal expansion and contraction of the label material.

[0052] For image blur caused by mechanical vibration, a blind deconvolution algorithm based on motion estimation is used to restore high-frequency details;

[0053] According to the physical state of the label, a repair model is constructed using a generative adversarial network to identify and repair surface defects, wrinkles, and damaged areas in the holographic image and generate a standardized holographic image.

[0054] In this embodiment, the present invention acquires holographic images by real-time collection of environmental data and multispectral imaging, uses histogram equalization and multi-scale Retinex algorithm to eliminate uneven illumination, and then uses affine transformation to correct angle deviation and bicubic interpolation to adjust resolution, combines nonlinear stretching, blind deconvolution and generative adversarial network to repair image defects, and each algorithm synergistically compensates for environmental interference to achieve hologram standardization, significantly improving anti-counterfeiting recognition accuracy. Effectively and significantly improve anti-counterfeiting security performance, so the multi-level compensation of image data proposed by the present invention is compared with the traditional single-factor correction, and the effectiveness is shown in Table 1.

[0055] Table 1 Comparison of environmental data correction effects

[0056] Environmental interference factors Error after single factor correction Error after multi-level compensation correction Improvement Uneven lighting Image contrast deviation ≥ 25% Image contrast deviation ≤5% +80% Shooting angle deviation Moire fringe distortion rate ≥ 30% Distortion elimination rate 98% +68% Temperature and humidity deformation Pixel displacement error ≥ 0.2mm Pixel displacement error ≤ 0.05mm -75% Mechanical vibration blur Peak signal-to-noise ratio ≤28dB Peak signal-to-noise ratio ≥42dB +50%

[0057] Extracting microscopic anti-counterfeiting features of the dynamic dot matrix hologram from the standardized holographic image to generate a holographic feature vector;

[0058] Furthermore, the microscopic anti-counterfeiting features of the dynamic dot matrix hologram are extracted from the standardized holographic image to generate a holographic feature vector, which specifically includes:

[0059] A multi-scale convolutional neural network is constructed. The input layer receives the standardized holographic image, and the hidden layer extracts multi-granular features through 3×3 convolution kernels, 5×5 convolution kernels and 7×7 convolution kernels. The feature maps of different scales are fused through the output layer to generate a multi-scale feature map. The entropy value of the pixel distribution in each channel, the diffraction grating phase difference and the gradient change caused by the viewing angle are calculated through the channel attention mechanism, and the channel weights of the multi-scale feature map are dynamically allocated to generate a multi-scale feature enhancement map. The local binary pattern and the directional gradient histogram are jointly encoded to quantize the microscopic arrangement pattern of the dynamic dot matrix to generate a comprehensive feature map. The comprehensive feature map is fused through the fully connected layer to generate a holographic feature vector.

[0060] In this embodiment, a multi-scale convolutional neural network and a channel attention mechanism are used to extract microscopic anti-counterfeiting features from a standard hologram, and a binary pattern and a gradient histogram are used to jointly encode and generate a feature vector. This method accurately captures the subtle structure of the image, significantly improves the accuracy and robustness of anti-counterfeiting recognition, and enhances anti-counterfeiting security.

[0061] Collect the encrypted barcode data of the anti-counterfeiting mark; construct an environmental compensation model to dynamically adjust the barcode decoding algorithm parameters, obtain a first optimized barcode decoding algorithm, decode the encrypted barcode data, and generate decoded barcode information;

[0062] The environmental compensation model includes an environmental data preprocessing layer, a parameter mapping layer, an adaptive optimization layer and a result output layer. Figure 2 ;

[0063] The environmental data preprocessing layer processes the collected environmental data through a normalization algorithm and a feature extraction algorithm to generate an environmental feature vector;

[0064] The parameter mapping layer maps the environmental feature vector to a preliminary parameter adjustment value of the barcode decoding algorithm through a convolutional neural network;

[0065] The adaptive optimization layer generates dynamically optimized barcode decoding parameters by making real-time corrections to the preliminary parameter adjustment values ​​and comprehensively considering real-time environmental changes and historical verification data;

[0066] The result output layer generates a first optimized barcode decoding algorithm by applying the optimized barcode decoding parameters to the barcode decoding algorithm.

[0067] In this embodiment, by constructing an environmental compensation model and dynamically adjusting the parameters of the barcode decoding algorithm, the decoding accuracy and robustness are improved, ensuring that the anti-counterfeiting barcode can still be stably and efficiently recognized in complex environments. This not only improves the accuracy of barcode recognition in various environments, but also strengthens the security and reliability of the barcode, and enhances the practicality and reliability of the anti-counterfeiting function. In addition, this dynamic optimization strategy can adapt to different environmental changes and improve decoding efficiency.

[0068] The specific process of decoding the encrypted barcode data using the first optimized barcode decoding algorithm to generate the decoded barcode information is as follows:

[0069] The YOLOv5 model is used to locate the encrypted barcode area, and the barcode and the background are separated through morphological operations. The barcode image is denoised and binarized according to the parameters in the first optimized barcode decoding algorithm to obtain standard barcode data. The standard barcode data is decrypted using an asymmetric encryption algorithm to generate barcode information. The barcode information includes a dynamic key plaintext and a timestamp.

[0070] In this embodiment, YOLOv5 is used to accurately locate the encrypted barcode area, and the barcode and background are separated through morphological operations. Standard barcode data is generated after denoising and binarization, and decrypted using an asymmetric encryption algorithm. The barcode information contains the dynamic key plaintext and timestamp. Such operations improve the accuracy of decoding and increase the security and reliability of anti-counterfeiting identification.

[0071] The holographic feature vector and the barcode information are judged, and the specific process includes:

[0072] Compare the timeliness and consistency of the holographic image timestamp and the barcode dynamic key through an asymmetric encryption algorithm;

[0073] Calculate the similarity between the holographic feature vector and the pre-stored standard feature vector to generate the holographic feature matching degree;

[0074] If the holographic feature matching degree is not less than the preset threshold and the dynamic key verification is passed, it is judged to be authentic and a new dynamic key is generated and synchronized to the anti-counterfeiting mark; otherwise, the multi-level early warning mechanism is activated, the abnormal log is recorded and manual review is requested.

[0075] The specific process of obtaining the holographic feature matching degree is as follows:

[0076] The holographic feature vector and the pre-stored standard feature vector are input into the twin neural network; the twin neural network has a dual-branch structure, specifically: branch one: input the real-time holographic feature vector, map it to the 128-dimensional embedding space through three fully connected layers, and generate a first embedding vector; branch two: input the pre-stored standard holographic feature vector, use the same network structure for mapping, and generate a second embedding vector;

[0077] The first embedding vector and the second embedding vector are measured by cosine similarity, Mahalanobis distance and dynamic weight allocation to generate a multidimensional similarity matrix; the multidimensional similarity matrix is ​​weightedly fused to obtain the holographic feature matching degree.

[0078] (1) Cosine similarity calculation:

[0079] ;

[0080] in, is the first embedding vector, is the second embedding vector, Used to characterize the directional consistency of feature vectors;

[0081] (2) Mahalanobis distance calculation:

[0082] ;

[0083] in, is the covariance matrix of the standard eigenvector, which is calculated by the 128-dimensional embedding vector of the offline training set. Reflects the statistical distance of feature distribution space;

[0084] (3) Dynamic weight allocation:

[0085] ;

[0086] in, and are the query matrix and key matrix of the attention mechanism, is the embedding dimension, Indicates Dynamic weights of dimensional features;

[0087] (4) Multi-dimensional similarity fusion:

[0088] ;

[0089] Among them, the weight coefficient satisfies , is the normalization factor of the maximum Mahalanobis distance of the training set, and the final similarity is the matching degree of the holographic features.

[0090] In this embodiment, the asymmetric encryption algorithm and the twin neural network dual judgment mechanism are used to compare the holographic image timestamp and the barcode dynamic key in real time to ensure the consistency verification of information. The embedded vector is generated by double-branch full-connection mapping, combined with cosine similarity, Mahalanobis distance, and dynamic weight weighted fusion, and multi-dimensional calculation is used to obtain the holographic feature matching degree. This method not only quickly determines the authenticity, but also triggers multi-level warnings, ensuring the accuracy and efficiency of anti-counterfeiting identification, and improving the security and intelligence level of the overall system.

[0091] The present invention acquires holographic images by real-time collection of environmental data and combining multispectral imaging technology, and uses a variety of image processing algorithms (such as adaptive histogram equalization, multi-scale Retinex, affine transformation, bicubic interpolation, blind deconvolution and generative adversarial network) to standardize holographic images, significantly reducing the influence of illumination, angle, temperature and humidity and mechanical interference, and improving the stability and accuracy of holographic images. At the same time, a multi-scale convolutional neural network and a channel attention mechanism are used to extract microscopic anti-counterfeiting features, and the binary mode and gradient histogram are combined for encoding to ensure the accurate capture and efficient matching of anti-counterfeiting features. For encrypted barcode data, an environmental compensation model is constructed to dynamically optimize the decoding parameters, and YOLOv5 is used to accurately locate the barcode area, combined with morphological processing, denoising, binarization and asymmetric encryption and decoding technology to improve the robustness and security of barcode recognition. In the anti-counterfeiting judgment link, the timestamp consistency is verified by an asymmetric encryption algorithm, and the holographic feature matching degree is calculated by using a twin neural network combined with cosine similarity, Mahalanobis distance and dynamic weight weighting to ensure the efficiency and accuracy of anti-counterfeiting judgment. In addition, the system is equipped with dynamic key synchronization and multi-level early warning mechanisms, which can monitor abnormal situations in real time and trigger security responses, thereby significantly improving the accuracy and robustness of anti-counterfeiting identification and enhancing the security effectiveness of the anti-counterfeiting system.

[0092] Embodiment 2

[0093] This embodiment uses multi-spectral imaging, environmental data compensation, dynamic dot matrix holographic feature extraction, and encrypted barcode decoding technologies to achieve real-time monitoring, image standardization, microscopic feature extraction, and verification of anti-counterfeiting labels on cigarette sealing paper. This method not only monitors the acquisition environment (including lighting, shooting angle, distance, temperature and humidity, vibration, mechanical interference, and physical state of the label, etc.) in real time, but also uses a multi-level algorithm to correct image deviations caused by environmental changes, ensuring that the holographic image and barcode information embedded on the cigarette sealing paper have a high degree of consistency and anti-counterfeiting capabilities.

[0094] Real-time monitoring and collection of environmental data; collecting the holographic image of the anti-counterfeiting mark by multi-spectral imaging technology; adjusting the holographic image by environmental data to generate a standardized holographic image;

[0095] Furthermore, the environmental data includes lighting conditions, shooting angle deviation, shooting distance deviation, ambient temperature, ambient humidity, mechanical vibration, mechanical interference, and physical state of the tag; and the holographic image includes visible light, infrared, and ultraviolet band images.

[0096] The environmental data of the sealing paper is collected in real time through the multi-spectral imaging device integrated in the tobacco sorting line, including the light intensity range of 200 to 1000 lux, the shooting angle of plus or minus 15 degrees, the object distance of 10 to 30 cm, the environmental humidity of 30 to 85 percent, and the vibration frequency of 5 to 50 Hz. The three bands of visible light wavelength of 400 to 700 nanometers, near infrared of 850 nanometers and ultraviolet 365 nanometers are used to synchronously collect the holographic image of the sealing paper, among which the ultraviolet band is used to capture the characteristics of fluorescent ink. The environmental data also includes: the curvature of the surface of the sealing paper, the reflectivity coefficient of the ink, the thermal expansion coefficient of the substrate and the vibration spectrum of the conveyor belt; the curvature of the surface of the sealing paper is obtained by laser triangulation, ranging from 0-15°; the reflectivity coefficient of the ink ranges from 0.2 to 0.8; the thermal expansion coefficient of the substrate is 1.5×10-5 / ℃; the vibration spectrum of the conveyor belt is 5-200Hz.

[0097] Furthermore, in the scenario of tobacco packaging production line, in view of the special properties of the sealing paper PET substrate and the environmental characteristics of the production line, the specific process of adjusting the holographic image through environmental data to generate a standardized holographic image is optimized as follows:

[0098] After geometric correction (affine transformation, thermal deformation compensation, etc.), The physical location of the corresponding sealing paper;

[0099] Light compensation module:

[0100] Based on the light intensity of the production line and the reflective characteristics of tobacco ink, the polarized light adaptive histogram equalization algorithm is used to suppress the mirror reflection interference of the gold wire while improving the image contrast; the multi-scale Retinex algorithm with weighted tobacco spectral characteristics is used to eliminate the color deviation of fluorescent ink in the ultraviolet band. , the compensation formula is:

[0101] ;

[0102] in, For the The weight coefficient of each spectral band, For the The reflected component of the segment image characterizes the optical properties of the material;

[0103] Geometry Correction Module:

[0104] According to the positioning data of the robot arm and the speed of the conveyor belt of the production line, a six-degree-of-freedom affine transformation model is used to compensate for the moiré fringe distortion caused by the tilt of the sealing paper. Combined with the actual object distance obtained by the laser rangefinder, an improved bicubic interpolation algorithm is used to stabilize the image resolution to 5080dpi, ensuring that the edge sharpness of the micro-text is ≥90%;

[0105] Thermal deformation compensation module:

[0106] Based on the data from the warehouse temperature and humidity sensors (temperature 20-45°C, humidity 30%-70%RH), pixel-level compensation is performed using the PET material deformation formula:

[0107] ;

[0108] in, is the deformation of PET material, is the thermal expansion coefficient of the PET substrate, with a typical value of 2.3×10 -5 / ℃, is the base size, is the real-time ambient temperature, Standard ambient temperature, is the coefficient of moisture expansion, and the common value for cigarette sealing paper is 1.7×10 -6 / %RH, is the real-time ambient humidity, is the standard ambient humidity;

[0109] Based on the environmental data, a dynamic compensation matrix is ​​constructed to perform geometric correction and optical compensation on the holographic image to generate a standardized holographic image with a compensation error of plus or minus 0.05 mm. The micro-text array in the standardized holographic image is extracted through a deep convolutional network with a minimum line width of 0.1 mm, and the light-variable diffraction features are identified to generate a 128-dimensional holographic feature vector.

[0110] The matrix encryption barcode on the surface of the sealing paper is collected synchronously, which complies with the ECC200 standard and has a version of 32 by 32. The environmental compensation model is applied to dynamically adjust the contrast threshold range of the decoding algorithm to 0.3 to 0.7, and adjust the error correction level to L1 to L4.

[0111] Vibration blur removal module:

[0112] According to the vibration spectrum of the conveyor belt collected by the acceleration sensor (main frequency 5-50Hz, amplitude 0.1-2g), the motion blur point diffusion range is constructed function:

[0113] ;

[0114] in, is a sine function, is the horizontal coordinate component of the fuzzy displacement caused by vibration, is the ordinate component of the fuzzy displacement caused by vibration, is the frequency domain horizontal coordinate, corresponding to the image spatial frequency, is the frequency domain ordinate, is the circumference of a circle, is an imaginary number;

[0115] Physical damage repair module:

[0116] For typical defects of sealing paper, such as curling > 3mm / m, glue contamination area > 5%, and edge wear depth > 0.05mm, a special generative adversarial network is trained:

[0117] The generator uses the U-Net architecture, inputs the damaged hologram (512×512×3), and outputs the repair mask; the discriminator uses the PatchGAN structure, focusing on detecting the continuity of the cigarette spot color;

[0118] The visible light, near-infrared and ultraviolet three-band images are registered through the multispectral fusion module to generate a fusion weight map and obtain a standardized holographic image.

[0119] Extract the microscopic anti-counterfeiting features of the dynamic dot matrix hologram from the standardized holographic image of the cigarette sealing paper and generate the holographic feature vector, which includes:

[0120] Multi-scale feature extraction network construction:

[0121] The input layer receives a 640×480 pixel three-channel standardized holographic image, and the input range is normalized to [0,1];

[0122] The hidden layer adopts the improved Inception-v4 architecture, deploying three groups of convolution kernels of 3×3, 5×5, and 7×7 in parallel to obtain the edge features of micro-text, the distribution pattern of fluorescent ink, and the fiber grid structure of PET substrate respectively;

[0123] The output layer fuses multi-scale feature maps through a 1×1 convolution kernel.

[0124] Channel Attention Mechanism

[0125] When calculating the weight of each channel, the characteristic parameters of cigarette sealing paper are introduced: the entropy weight is obtained based on the discreteness of the fluorescent ink in the ultraviolet band, the diffraction angle deviation of the holographic grating is quantified by the phase difference weight, and the texture blur gradient caused by the vibration of the conveyor belt is analyzed by the gradient weight;

[0126] In the feature encoding stage, the topological structure of the fiber intersections is quantified by the rotation-invariant LBP operator, and the consistency of the direction of the gold wire is verified by the vibration-compensated HOG algorithm. After PCA dimensionality reduction, it is fused with the CNN features to generate a 128-dimensional comprehensive vector.

[0127] The fully connected layer uses triplet loss optimization, and TensorRT acceleration is used to reduce the single-frame processing time to ≤15ms. An online incremental learning module is simultaneously embedded, and network parameters are updated every 10,000 images to adapt to the vibration interference of the production line.

[0128] A multi-spectral competition mechanism (dynamic competition mechanism of visible light and ultraviolet band weights) is introduced to achieve light adaptation, and quantitative standards such as microtext SSIM ≥ 0.95 and fiber Hellinger distance ≤ 0.05 are set. Compared with the general method, the false detection rate is reduced to below 0.01%, meeting the robust detection requirements for typical defects such as curling and glue marks in high-speed sorting scenarios; visible light and ultraviolet band weights The dynamic competition mechanism formula is as follows:

[0129] ;

[0130] in, is the edge energy value of the visible light band, is the edge energy value of the ultraviolet band, is an infinitesimal value.

[0131] Collect the encrypted barcode data of the anti-counterfeiting mark; construct an environmental compensation model to dynamically adjust the barcode decoding algorithm parameters, obtain a first optimized barcode decoding algorithm, decode the encrypted barcode data, and generate decoded barcode information;

[0132] The lightweight MobileNetV3 network is used to input the holographic feature vector and output the preliminary parameters of barcode decoding: binarization threshold, morphological operation kernel size and denoising strength; the parameters are dynamically modified based on the historical decoding success rate; the ZebraTechnologies decoding engine is injected to adapt to cigarette barcodes;

[0133] Cigarette encrypted barcode decoding process:

[0134] Step 1: Locate the barcode area

[0135] The improved YOLOv5s model is used: the input image is 640×480 pixels, and the coordinates of the barcode area on the sealing paper surface are output; 100,000 cigarette barcode images with interference samples are used as the training data set, and an attention module is added to improve the ability to suppress reflective areas; the interference samples include gold pull lines, glue marks, and wrinkles;

[0136] Step 2: Morphological operations:

[0137] Use opening operation to eliminate small reflective noise and closing operation to fill barcode breaks; and use binarization to compensate for image blur caused by humidity;

[0138] Step 3: Using asymmetric encryption using the national secret SM2 algorithm, the private key is stored in the tobacco traceability platform HSM module.

[0139] The holographic feature vector and the barcode information are judged, and the specific process includes: comparing the timeliness and consistency of the holographic image timestamp and the barcode dynamic key through an asymmetric encryption algorithm; calculating the similarity between the holographic feature vector and the pre-stored standard feature vector to generate a holographic feature matching degree;

[0140] If the holographic feature matching degree is not less than the preset threshold and the dynamic key verification is passed, it is judged to be authentic and a new dynamic key is generated and synchronized to the anti-counterfeiting mark; otherwise, the multi-level early warning mechanism is activated, the abnormal log is recorded and manual review is requested.

[0141] The holographic feature vector is jointly verified with the dynamic key embedded in the barcode. When the similarity reaches or exceeds 95% and the timestamp deviation is less than 5 minutes, the laser engraving device is triggered to generate a dynamic verification code on the edge of the sealing paper. Otherwise, the system activates a three-level warning mechanism: when the similarity is between 80% and 95%, a rescan is triggered; when the similarity is between 60% and 80%, the sorting line is locked; when the similarity is less than 60%, it is uploaded to the tobacco traceability platform.

[0142] In this embodiment, the multi-level warning is a three-level warning; the three-level warning includes:

[0143] Level 1 warning triggering conditions: holographic feature similarity is 80%-95% and the dynamic key is valid, and local rescanning is initiated;

[0144] Secondary warning trigger conditions: similarity between 60%-80% or key timeout, set to more than 5 minutes, lock the sorting line and collect hyperspectral images;

[0145] Level 3 warning triggering conditions: similarity < 60% and key verification fails, physical identification is activated and abnormal data packets are uploaded (including environmental sensor logs, image metadata, and device operating parameters).

[0146] This embodiment organically combines multispectral imaging, environmental data compensation, deep feature extraction and encrypted barcode decoding to achieve efficient monitoring and verification of the entire process of anti-counterfeiting labels for cigarette sealing paper. The system collects environmental parameters such as light intensity, shooting inclination, object distance, ambient humidity, vibration frequency, sealing paper surface curvature, ink reflection coefficient, substrate thermal expansion coefficient and conveyor belt vibration spectrum in real time, and uses a dynamic compensation matrix to perform geometric correction, optical compensation and thermal deformation correction on the holographic image to control the image error within ±0.05 mm. Deep convolutional networks and multi-scale feature extraction technology are used to accurately capture microtext and light-variable diffraction features, generate holographic feature vectors, and jointly verify with encrypted barcode information that meets the ECC200 standard. The barcode area positioning and dynamic parameter tuning are achieved through the improved YOLOv5 and MobileNetV3 models, and the national secret SM2 algorithm is used to ensure data security. At the same time, a three-level early warning mechanism is set up to respond to abnormal situations in a timely manner. For details, please refer to Figure 3 The overall solution significantly improves the accuracy, robustness and real-time performance of anti-counterfeiting identification, and effectively prevents the risk of counterfeiting. In order to verify the environmental adaptability and decoding robustness of the present invention, a variety of extreme interference scenarios were simulated in the tobacco sorting production line, and compared with the traditional ECC200 decoding solution. The results are shown in Table 2.

[0147] Table 2 Comparison of barcode decoding success rates

[0148]

[0149] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent identification and analysis of anti-counterfeiting marks, characterized in that: include: Real-time monitoring and collection of environmental data; collecting the holographic image of the anti-counterfeiting mark by multi-spectral imaging technology; adjusting the holographic image by environmental data to generate a standardized holographic image; extracting the microscopic anti-counterfeiting features of the dynamic dot matrix hologram from the standardized holographic image to generate a holographic feature vector; Collect the encrypted barcode data of the anti-counterfeiting mark; construct an environmental compensation model to dynamically adjust the barcode decoding algorithm parameters, obtain a first optimized barcode decoding algorithm, decode the encrypted barcode data, and generate decoded barcode information; The holographic feature vector and the barcode information are judged, and the specific process includes: Compare the timeliness and consistency of the holographic image timestamp and the barcode dynamic key through an asymmetric encryption algorithm; Calculate the similarity between the holographic feature vector and the pre-stored standard feature vector to generate the holographic feature matching degree; If the holographic feature matching degree is not less than the preset threshold and the dynamic key verification is passed, it is judged to be authentic and a new dynamic key is generated and synchronized to the anti-counterfeiting mark; otherwise, the multi-level early warning mechanism is activated, the abnormal log is recorded and manual review is requested.

2. The method for intelligent identification and analysis of anti-counterfeiting marks according to claim 1, characterized in that: The environmental data includes lighting conditions, shooting angle deviation, shooting distance deviation, ambient temperature, ambient humidity, mechanical vibration, mechanical interference, and physical state of the tag; the holographic image includes visible light, infrared and ultraviolet band images.

3. The method for intelligent identification and analysis of anti-counterfeiting marks according to claim 1, characterized in that: The step of adjusting the holographic image by using the environmental data to generate a standardized holographic image specifically includes: Based on the light intensity data, an adaptive histogram equalization algorithm is used to adjust the image contrast, and a multi-scale Retinex algorithm is used to eliminate the color cast caused by uneven lighting. According to the shooting angle deviation data, the affine transformation model is used to compensate for the image geometric distortion, and the original viewing angle of the hologram is restored in combination with the calibration parameters; By using the camera distance deviation data, the bicubic interpolation algorithm is used to adjust the image resolution to the preset pixel density to ensure the clarity of the dynamic dot matrix pixels; Based on the ambient temperature and humidity data, the image is nonlinearly stretched to compensate for the microstructural deformation caused by thermal expansion and contraction of the label material. For image blur caused by mechanical vibration, a blind deconvolution algorithm based on motion estimation is used to restore high-frequency details; According to the physical state of the label, a repair model is constructed using a generative adversarial network to identify and repair surface defects, wrinkles, and damaged areas in the holographic image and generate a standardized holographic image.

4. The method for intelligent identification and analysis of anti-counterfeiting marks according to claim 1, characterized in that: Extract the microscopic anti-counterfeiting features of the dynamic dot matrix hologram from the standardized holographic image and generate the holographic feature vector, which specifically includes: A multi-scale convolutional neural network is constructed. The input layer receives the standardized holographic image, and the hidden layer extracts multi-granular features through 3×3 convolution kernels, 5×5 convolution kernels and 7×7 convolution kernels. The feature maps of different scales are fused through the output layer to generate a multi-scale feature map. The entropy value of the pixel distribution in each channel, the diffraction grating phase difference and the gradient change caused by the viewing angle are calculated through the channel attention mechanism, and the channel weights of the multi-scale feature map are dynamically allocated to generate a multi-scale feature enhancement map. The local binary pattern and the directional gradient histogram are jointly encoded to quantize the microscopic arrangement pattern of the dynamic dot matrix to generate a comprehensive feature map. The comprehensive feature map is fused through the fully connected layer to generate a holographic feature vector.

5. The method for intelligent identification and analysis of anti-counterfeiting marks according to claim 1, characterized in that: The environmental compensation model includes an environmental data preprocessing layer, a parameter mapping layer, an adaptive optimization layer and a result output layer; The environmental data preprocessing layer processes the collected environmental data through a normalization algorithm and a feature extraction algorithm to generate an environmental feature vector; The parameter mapping layer maps the environmental feature vector to a preliminary parameter adjustment value of the barcode decoding algorithm through a convolutional neural network; The adaptive optimization layer generates dynamically optimized barcode decoding parameters by making real-time corrections to the preliminary parameter adjustment values ​​and comprehensively considering real-time environmental changes and historical verification data; The result output layer generates a first optimized barcode decoding algorithm by applying the optimized barcode decoding parameters to the barcode decoding algorithm.

6. The method for intelligent identification and analysis of anti-counterfeiting marks according to claim 1, characterized in that: The specific process of decoding the encrypted barcode data using the first optimized barcode decoding algorithm to generate the decoded barcode information is as follows: The YOLOv5 model is used to locate the encrypted barcode area, and the barcode and the background are separated through morphological operations. The barcode image is denoised and binarized according to the parameters in the first optimized barcode decoding algorithm to obtain standard barcode data. The standard barcode data is decrypted using an asymmetric encryption algorithm to generate barcode information. The barcode information includes a dynamic key plaintext and a timestamp.

7. The method for intelligent identification and analysis of anti-counterfeiting marks according to claim 1, characterized in that: The specific process of obtaining the holographic feature matching degree is as follows: The holographic feature vector and the pre-stored standard feature vector are input into the twin neural network; the twin neural network has a dual-branch structure, specifically: branch one: input the real-time holographic feature vector, map it to the 128-dimensional embedding space through three fully connected layers, and generate a first embedding vector; branch two: input the pre-stored standard holographic feature vector, use the same network structure for mapping, and generate a second embedding vector; The first embedding vector and the second embedding vector are measured by cosine similarity, Mahalanobis distance and dynamic weight allocation to generate a multi-dimensional similarity matrix; The multi-dimensional similarity matrix is ​​weightedly fused to obtain the holographic feature matching degree.

Images