A Double-Layer Anti-Counterfeiting Label System and Method Based on Artificial Intelligence

Abstract

The present invention relates to the field of anti-counterfeiting technology, specifically a double-layer anti-counterfeiting label system and method based on artificial intelligence, including: an image acquisition module, a color block analysis module, a particle distribution module, and an anti-counterfeiting identification module; collecting an initial anti-counterfeiting image containing an anti-counterfeiting label according to user requirements, splitting the initial anti-counterfeiting image into a color block border image, a particle distribution image, and a two-dimensional code image; analyzing the color block border image to obtain the edge features of the color block border image, and performing feature matching between the edge features of the color block border image and the two-dimensional code image to obtain edge matching features; performing matching analysis on the metal particle distribution in the particle distribution image to obtain the particle distribution state of the metal particles; performing feature matching between the particle distribution state and the two-dimensional code image to obtain a particle state vector; verifying the initial anti-counterfeiting image based on the edge matching features and the particle state vector; improving the recognition accuracy and reliability of the anti-counterfeiting label.

Description

Technical Field

[0001] The present invention relates to the field of anti-counterfeiting technology, and specifically to a dual-layer anti-counterfeiting label system and method based on artificial intelligence. Background Art

[0002] In the field of anti-counterfeiting technology, traditional anti-counterfeiting methods often rely on single anti-counterfeiting means, such as single QR code verification, special ink printing, etc. Although these methods have improved the anti-counterfeiting ability of products to a certain extent, there is still a risk of being counterfeited or cracked. With the development of artificial intelligence technology, anti-counterfeiting technology has also begun to develop towards intelligence and multi-layer. However, most of the existing intelligent anti-counterfeiting systems only focus on single-level anti-counterfeiting information, such as only verifying QR codes or only analyzing special textures, lacking comprehensive anti-counterfeiting verification means.

[0003] For example, Chinese Patent Publication No. CN114580597A discloses a method for producing and supplementing NFC anti-counterfeiting labels. The method includes: receiving a first login instruction from a first production equipment group; performing a first equipment login inspection; if the inspection is successful, performing production task allocation to generate a first production task list and sending it down; receiving a first defective label statistical list when the production task is completed; if the first defective label statistical list is not empty, receiving a second login instruction from a first supplementary labeling equipment group; performing a second equipment login inspection; if the inspection is successful, performing supplementary labeling task allocation to generate a first supplementary labeling task list and sending it down; receiving a second defective label statistical list when the supplementary labeling task is completed; if the first defective label statistical list is empty, forming a first production task backup record from the first production task list and saving it; if not empty, forming a first production task backup record from the first production task list and the second defective label statistical list and saving it.

[0004] For example, Chinese Patent Publication No. CN116227524A discloses a label-based anti-counterfeiting system. The method includes: the generating party of the anti-counterfeiting label obtains an initial anti-counterfeiting image containing a graphic code for anti-counterfeiting, and splits the initial anti-counterfeiting image into multiple sub-images; the multiple sub-images are printed on a label substrate in multiple times to generate a printed anti-counterfeiting image corresponding to the graphic code, and there are random offsets between the printing positions of the multiple sub-images on the label substrate; the verifying party of the anti-counterfeiting label obtains a to-be-verified anti-counterfeiting image collected by a user through a terminal; obtaining a printed anti-counterfeiting image corresponding to the graphic code contained in the to-be-verified anti-counterfeiting image; verifying the authenticity of the to-be-verified anti-counterfeiting image according to the image similarity between the to-be-verified anti-counterfeiting image and the printed anti-counterfeiting image.

[0005] Description of the prior art: When dealing with anti-counterfeiting labels, in addition to setting anti-counterfeiting labels through NFC, it is also described that the offset between multiple sub-images can be used for anti-counterfeiting. However, these anti-counterfeiting measures have the problem of a single setting method, and it is easy to be counterfeited and cracked. By setting multi-level comprehensive verification on the content of the anti-counterfeiting label, the recognition accuracy and reliability of the anti-counterfeiting label can be improved. Summary of the Invention

[0006] To solve the above technical problems, the technical solution adopted by the present invention is: An artificial intelligence-based double-layer anti-counterfeiting label system, including: An image acquisition module, configured to collect an initial anti-counterfeiting image containing an anti-counterfeiting label according to user requirements, and split the initial anti-counterfeiting image into a color block border image, a particle distribution image, and a two-dimensional code image.

[0007] A color block analysis module, configured to analyze the color block border image, determine the edge features of each sub-image in the color block border image, combine the edge features of each sub-image to obtain the edge features of the color block border image, and perform feature matching between the edge features of the color block border image and the two-dimensional code image to obtain the edge matching features of the color block border image.

[0008] A particle distribution module, configured to perform matching analysis on the metal particle distribution in the particle distribution image to obtain the particle distribution state of the metal particles; perform feature matching between the particle distribution state and the two-dimensional code image to obtain a particle state vector.

[0009] An anti-counterfeiting recognition module, configured to verify the initial anti-counterfeiting image based on the edge matching features and the particle state vector, determine the recognition method of the initial anti-counterfeiting image under different user requirements, and complete the anti-counterfeiting image recognition.

[0010] An artificial intelligence-based double-layer anti-counterfeiting label method, including: S1, collecting an initial anti-counterfeiting image containing an anti-counterfeiting label through the image acquisition module, and splitting it into a color block border image, a particle distribution image, and a two-dimensional code image.

[0011] S2, analyzing the color block border image, determining the edge features of each sub-image, and performing feature matching between the edge features of each sub-image and the two-dimensional code image to obtain edge matching features.

[0012] S3, performing matching analysis on the metal particle distribution in the particle distribution image to obtain the particle distribution state, and performing feature matching between the particle distribution state and the two-dimensional code image to obtain a particle state vector.

[0013] S4, verifying the initial anti-counterfeiting image based on the edge matching features and the particle state vector, and determining the recognition method according to user requirements to complete the anti-counterfeiting image recognition.

[0014] The beneficial effects of the present invention are as follows: First, the system of the present invention uses an image acquisition module to collect an initial anti-counterfeiting image containing an anti-counterfeiting label and splits it into a color block border image, a particle distribution image, and a QR code image. By analyzing the color block borders and metal particles in the initial anti-counterfeiting image respectively, the correlation between the corresponding data and the data bound in the QR code image can be obtained, providing a corresponding data basis for subsequent specific authentication of the anti-counterfeiting label. At the same time, the method of setting anti-counterfeiting marks for the current anti-counterfeiting label through metal particles and color block borders can reduce the problem of single setting means in anti-counterfeiting means, making it difficult for anti-counterfeiting labels to be imitated and cracked.

[0015] Second, the present invention analyzes the color block border image to determine the edge features of each sub-image and matches these edge features with the QR code image to obtain edge matching features. This step enhances the reliability of anti-counterfeiting verification by comparing the correlation between the color block border image and the QR code image. At the same time, the color block border is generated in an irregular manner, which can improve the setting effect of the anti-counterfeiting label and further reduce the risk of being imitated or cracked.

[0016] Third, the present invention performs matching analysis on the metal particle distribution in the particle distribution image to obtain the particle distribution state and matches this state with the QR code image to obtain a particle state vector. This step utilizes the randomness and non-replicability of the metal particle distribution to further improve the accuracy of anti-counterfeiting verification. At the same time, the vector after matching the metal particles is combined with the matching content of the color block border to complete the identification of the anti-counterfeiting image. This step realizes the comprehensive verification of the anti-counterfeiting label by considering anti-counterfeiting information at multiple levels. And it can determine the identification method according to user needs, providing different verification modes such as quick verification, detailed analysis, and batch processing to meet the needs of different users. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The present invention will be further described below with reference to the drawings and embodiments.

[0018] Figure 1 is a schematic diagram of a label of a double-layer anti-counterfeiting label system based on artificial intelligence.

[0019] Figure 2 is a schematic diagram of a system of a double-layer anti-counterfeiting label system based on artificial intelligence.

[0020] Figure 3 is a schematic flow diagram of a color block analysis module of a double-layer anti-counterfeiting label system based on artificial intelligence.

[0021] Figure 4 is a schematic flow diagram of a particle distribution module of a double-layer anti-counterfeiting label system based on artificial intelligence.

[0022] Figure 5 It is a schematic flow chart of a double-layer anti-counterfeiting label method based on artificial intelligence. Specific implementation manners

[0023] The embodiments of the present invention will be described in detail below. The embodiments described below are exemplary and are only used to explain the present invention, and should not be construed as a limitation to the present invention. For those not specified in the embodiments regarding specific technologies or conditions, they shall be carried out according to the technologies or conditions described in the literature in the art or according to the product specifications.

[0024] Refer to Figure 1 、 Figure 2 , a double-layer anti-counterfeiting label system based on artificial intelligence, comprising: an image acquisition module, a color block analysis module, a particle distribution module, and an anti-counterfeiting identification module; wherein, the output end of the image acquisition module is connected to the color block analysis module, the output end of the color block analysis module is connected to the particle distribution module, and the output end of the particle distribution module is connected to the anti-counterfeiting identification module.

[0025] The image acquisition module is used to collect an initial anti-counterfeiting image containing an anti-counterfeiting label according to user requirements, and split the initial anti-counterfeiting image into a color block border image, a particle distribution image, and a two-dimensional code image.

[0026] The color block analysis module is used to analyze the color block border image, determine the edge features of each sub-image in the color block border image, combine the edge features of each sub-image to obtain the edge features of the color block border image, and perform feature matching between the edge features of the color block border image and the two-dimensional code image to obtain the edge matching features of the color block border image.

[0027] The particle distribution module is used to perform matching analysis on the metal particle distribution in the particle distribution image to obtain the particle distribution state of the metal particles; perform feature matching between the particle distribution state and the two-dimensional code image to obtain a particle state vector.

[0028] The anti-counterfeiting identification module is used to verify the initial anti-counterfeiting image based on the edge matching features and the particle state vector, determine the identification method of the initial anti-counterfeiting image under different user requirements, and complete the anti-counterfeiting image identification.

[0029] For the initial anti-counterfeiting image, it can be collected through image acquisition and recognition by a camera, etc. At the same time, according to the distribution of each part in the initial anti-counterfeiting image, the color block border image and the particle distribution image for identifying the authenticity of the current anti-counterfeiting label are extracted; the color block border image and the particle distribution image will be distributed on the left side of the initial anti-counterfeiting image, and the right side of the initial anti-counterfeiting image will contain a QR code and a verification code, which is convenient for quick identification when authenticating the anti-counterfeiting label. For the color border image, a border image randomly generated by multiple color blocks is used to encrypt the content corresponding to the current anti-counterfeiting label. Then, the distribution form of metal particles is used as the particle distribution image, and the encrypted information is represented according to the data form of the particle distribution. When any content in the color block border image and the particle distribution image does not correspond, the current anti-counterfeiting label is considered fake.

[0030] When normally recognizing the QR code image, the features existing in the QR code image will also be matched with the color block border and the metal particle distribution to further verify whether the randomly distributed metal particles and color blocks conform to the pre-set rules, so as to complete the identification setting of the anti-counterfeiting label.

[0031] When using the anti-counterfeiting label, on one side of it, a border is composed of unique and irregular color blocks. The edges and corresponding areas of each color block need to be recognized on this border to verify whether the color block is the content bound by the QR code. When binding, a hash value will be set for the color block border image. Then, when the QR code is recognized, it will check whether the image is consistent with the image corresponding to the pre-set QR code to find out whether the border of the current initial anti-counterfeiting image is a genuine product.

[0032] In an embodiment of the present invention, as Figure 3 shown, the implementation method of the edge features of the color block border image further includes: preprocessing the color block border image and converting the color block border image into a grayscale histogram. The grayscale histogram converts the color image into a grayscale image and simultaneously performs filtering processing on the grayscale image to reduce the influence of noise on subsequent steps. Commonly used filtering methods include Gaussian filtering, median filtering, etc.

[0033] According to the distribution positions of the color blocks in the color block border image, the color block border area is sequentially divided into sub-images corresponding to each color block; the edge features of each sub-image are recognized.

[0034] The implementation method of recognizing the edge features of each sub-image includes: performing contour detection on the color blocks in each sub-image, and setting the contour pixel points corresponding to each color block according to the minimum circumscribed rectangle of the contour detection.

[0035] Combine the contour pixel points into a sequence of edge coordinates, extract the shape descriptor and Fourier descriptor of the contour pixel points, set the mapping relationship for the edge coordinate sequence, shape descriptor, and Fourier descriptor, and output the edge features of each sub-image.

[0036] The shape descriptor refers to the Hu moment (7-dimensional invariant moment). The Hu moment is a shape descriptor used for image recognition and shape analysis, which is based on the invariant moment theory. The invariant moment is a description of the shape features of an image and can maintain invariance under image translation, rotation, and scale change. The Hu moment includes seven normalized invariant moments, which are all functions of geometric moments and have translational, rotational, and scaling invariance.

[0037] The Hu moment calculates the geometric moment of the image based on the contour information, that is, calculates according to the distance between the contour pixel points, then normalizes the geometric moment, and calculates seven invariant moments to obtain the shape descriptor.

[0038] The Fourier descriptor uses the Fourier transform to apply the discrete Fourier transform to the contour point set to obtain the representation of each point in the frequency domain, and extracts the amplitude and phase of the first 20 low-frequency components from the frequency domain data as the Fourier descriptor. This is used to describe the contour of each color block in the corresponding area. It should be noted that here the color blocks represent sub-images composed of multiple irregular shapes of the same color, and the information of each color block is recognized in this sub-image.

[0039] According to the edge features of each sub-image, calculate the position and area corresponding to each color block in each sub-image; and construct the color block adjacency relationship between each color block according to the position and area corresponding to each color block. The color block adjacency relationship describes the adjacent relationship between each color block. At this time, the position can use the color block border image that has not been divided into sub-images as the coordinate system. For example, taking the lower left corner of the color block border image as the coordinate origin and the two adjacent sides as the coordinate axes to quantify the positions of all color blocks in the sub-images, and then identify its area according to the pixel points included in each color block.

[0040] According to the color block adjacency relationship between each color block, splice the edge features of each sub-image and output the edge features of the color block border image.

[0041] When performing feature splicing, use the raster scan and marking method to connect the edge features of the sub-images through the color block adjacency relationship of multiple color blocks to obtain the topological structure between each color block, and splice and combine the adjacent parts in the edge features according to the topological structure of each color block to obtain the edge features describing the current color block border image.

[0042] Meanwhile, the edge combination features will also perform feature matching with the relevant data on the QR code. For example, according to the extended data contained in the QR code, the size and position of each irregular color block in the edge combination features can be verified to complete the further matching of the QR code, ultimately improving the overall recognition speed and anti-counterfeiting effect.

[0043] Therefore, the implementation method of the edge matching features includes: based on the edge features of the color block border image, determining the relative distances and distribution sequences of each sub-image in the color block border image, obtaining the preset edge features using the QR code image, matching the relative distances and distribution sequences of each sub-image with the preset edge features, and determining the relative relationships and distribution quantities of multiple sub-images after matching.

[0044] The relative distance represents the relative distances between each sub-image and the relative distances between each color block in the sub-image. The distribution composition of each color block indicates the type, quantity, and arrangement method of the color blocks existing in the sub-image, representing the sequence of the corresponding order of the color blocks. Since each color block represents a graphic corresponding to a color, and this graphic is randomly generated in an irregular manner, then at this time, the type will be described according to the sides or shapes represented by this graphic. Then, according to whether there are multiple of the same type and how the graphics are sorted on the border, the distribution sequences of each sub-image are generated.

[0045] At this time, when outputting the relevant content of the color block border image, a border composed of irregular color blocks will be set with an identification code corresponding to the QR code image, and a preset edge feature with an initial setting will be generated. Then, the collected color block border image will be compared with this preset edge feature to understand whether the difference between the collected image and the initially set image meets the normal specifications, and to understand the type, quantity, and arrangement method of each color block at this time to describe the color block border image. Generally, when generating a border composed of irregular color blocks, a fixed-length hash value will be formed according to the distribution ratio of these data and the form of adjacent color blocks. After collecting the color block border image, the corresponding content on this image will be compared with this described hash value to determine the gap between the current image and the standard image.

[0046] After that, according to the relative distances and distribution sequences of each sub-image, it will be described whether some color blocks appear adjacent and symmetric at their corresponding positions under the image corresponding to the preset edge feature of this sub-image. And for those with adjacent and symmetric relationships, the quantities under these corresponding relationships will be identified to obtain the distribution quantity of the sub-image. This distribution quantity indicates what kind of matching relationship exists between the current color block border image and the preset edge feature, and records the existing relationship and the corresponding quantity.

[0047] The implementation methods for determining the relative relationships and distribution quantities of multiple sub-images after matching include: using a feature matching algorithm to calculate the similarity between the relative distances and distribution sequences of each sub-image during matching and the preset edge features. The calculated similarity uses Euclidean distance and Hamming distance to describe the similarity between these features. If the similarity of the relative distance and distribution sequence is described at this time, the cosine similarity is used to calculate the similarity under the relative distance, and the consistency of the sequence is used to describe the similarity under the type, quantity, and arrangement of color blocks, that is, the value of the consistent and similar characters in the entire sequence length is used to describe the similarity corresponding to the distribution sequence. It is also possible to calculate the sum of squared differences between the positions of this part of the corresponding data on the sub-image and the preset edge features as the matching result described at this time. Obtain the matching results of each sub-image and the preset edge features. According to the matching results of each sub-image and the preset edge features, use breadth-first search to construct the relative relationships of each sub-image, such as adjacent, symmetric, etc.; breadth-first search is to continuously combine and connect the results of whether there are adjacent and symmetric relationships between the corresponding color blocks in each sub-image and the preset edge features to obtain a relationship graph that can describe the features on all color blocks. Measure the data with different relative relationships on the relationship graph to obtain the distribution quantity of the sub-images. Then, according to these relative relationships and distribution quantities, the data in the matching process can be classified to identify whether there are corresponding errors.

[0048] Use the relative relationships and distribution quantities of the sub-images to perform consistency verification, identify the relevant granularity differences and area ratio differences of the sub-images, and obtain the difference degree of the sub-images.

[0049] The relevant granularity difference of the sub-image represents the average value of the pixel differences between the color blocks corresponding to the distribution quantity under the corresponding relative relationship of the sub-image; the area ratio difference is also the average value of the difference ratio of the areas contained in its color blocks. The corresponding values of the relevant granularity difference and the area ratio difference are used as the difference degree of the sub-image.

[0050] Judge whether the sub-image with the largest difference degree is a normal change. If it is a normal change, set an anti-counterfeiting mark for the color block border image, and combine the distribution sequences of each sub-image and output them as the edge matching features of the color block border image.

[0051] The finally output edge matching features will indicate under what circumstances the current color block border image has differences from the preset edge features and what the specific values of the differences are, so as to increase the difficulty of counterfeiting the current color block border image. As for judging whether the sub-image with the largest difference degree is a normal change, when the error in the difference degree exceeds 0.1%, it indicates that the currently recognized color block border image has an abnormal situation and is not a normal change; when the error value is less than 0.1%, it indicates that the currently recognized color block border image is an image that is normally matched with the two-dimensional code and is a product produced normally.

[0052] At this time, it indicates that 0.1% means that the color block border image currently recognized should have a very small error with the content bound to the QR code. This 0.1% error may be caused by the image processing method after the image is taken. In the scenario where the error is small enough to be ignored, the color block border can be considered normal, thus completing the anti-counterfeiting recognition.

[0053] When the difference is normal after matching, the sequence of the specific distribution of color blocks in the color block border image will be selected for output to assist the metal particles to complete the double-layer authentication of the anti-counterfeiting image.

[0054] In an embodiment of the present invention, the particle distribution module is mainly used to identify the distribution state of metal particles in the particle distribution image. After obtaining a coordinate sequence, the random auxiliary index and the combined index of each position are determined to describe the anti-counterfeiting information represented by the metal particles.

[0055] Such as Figure 4 As shown, the implementation method for matching and analyzing the distribution of metal particles in the particle distribution image includes: identifying the coordinates corresponding to each metal particle according to the positions of the metal particles in the particle distribution image, and generating a particle coordinate sequence; the particle distribution image represents an image of the irregular distribution of metal particles in a specific area, and the distribution state of the metal particles is used as an anti-counterfeiting method for the anti-counterfeiting label. And the metal particles are of the same size. Starting from the lower left corner in the specific area, the area where the metal particles are located is equally divided into multiple rectangles of the same size, and then the coordinates corresponding to each metal particle are identified and input into the coordinate sequence to obtain the coordinates corresponding to the metal particles.

[0056] Record the distances and angles between the metal particles, convert the particle coordinate sequence into a topological structure matrix, and output the connection of the metal particles in the topological structure matrix as the particle distribution state. The topological structure matrix is mainly used to combine the data of the positions, distances, angles, and adjacency relationships of the metal particles to describe the specific distribution of the current metal particles, which is convenient for subsequent analysis of the actual distribution of the metal particles and the corresponding data in the QR code image to determine whether the distribution of the metal particles meets the preset rules.

[0057] The implementation method for feature matching between the particle distribution state and the QR code image includes: based on the distances between the metal particles in the particle distribution state, associating the particle distribution state with the corresponding data in the QR code image to generate a relationship data set of the particle distribution state; the relationship data set is obtained by associating the topological structure matrix corresponding to the particle distribution state with the corresponding data set in the QR code image to determine whether the distribution of the current metal particles is correct. The corresponding data in the QR code image is the image of the distribution of the metal particles, which is used to check the distribution of the metal particles in the current particle distribution image.

[0058] Obtain the position index of the particle distribution state from the relational dataset, and combine the position indexes to obtain each position combination index. The position index represents the position of the metal particles in the current particle distribution state and the position of the corresponding data in the QR code image. Each position index represents a numerical pair that correlates the coordinates of these two positions; each position combination index is a statistic of this data. For example, each position index includes the physical coordinates of the metal particles and the logical coordinates of the corresponding data block in the QR code. Then, for each position index, calculate the deviation distance and deviation angle of the two coordinates corresponding to the position index as the deviation distance and deviation angle of each position index, and combine these data into each position combination index.

[0059] Use each position combination index to classify the particle distribution state, and calculate the average deviation distance and average deviation angle of the position indexes under each classification.

[0060] According to the average deviation distance and average deviation angle of the position indexes under each classification, set the position encoding for each classification; combine the position encodings into a particle distribution vector according to the dimension corresponding to each classification.

[0061] When performing classification, the implementation method of using each position combination index to classify the particle distribution state includes: performing multi-dimensional hierarchical clustering on the particle distribution state, and sequentially performing clustering on spatial position, distribution pattern, and deviation angle association.

[0062] Cut the area of the current metal particle distribution into multiple grid areas of the same size. There are metal particles in each grid area. Calculate the average deviation distance and average deviation angle of the position indexes corresponding to the metal particles in each grid area; as the clustering result of the spatial position.

[0063] The grid area is converted into a density histogram according to the density of the metal particles. Use the density histogram for KL divergence clustering. Calculate the KL divergence corresponding to the probability distributions of the values of the metal particles in the grid area and the area of the metal particle distribution, and cluster the metal particles in each grid area according to the value of the KL divergence to obtain multiple clustering clusters. Output the average deviation distance and average deviation angle of the position indexes corresponding to the metal particles in each clustering cluster to obtain the clustering result of the distribution pattern.

[0064] Generate the deviation angle correlation matrix for each grid area according to the deviation angle of the position index. Perform principal component analysis on the deviation angle correlation matrices of adjacent grid areas, and calculate the average deviation distance and average deviation angle corresponding to the position index of the principal component analysis output data as the clustering result of the deviation angle correlation. When generating the deviation angle correlation matrix, generate the corresponding deviation angle correlation matrix for the metal particles existing in the current grid area according to the deviation angle existing within the position index. Then, perform principal component analysis on the deviation angle correlation matrices in adjacent grid areas, and use the several data with the top-ranked contribution degrees in the output of the principal component analysis as the clustering results corresponding to the adjacent grid areas. Calculate the average value of the deviation angles and deviation distances at the corresponding position indexes to obtain the clustering result under the deviation angle correlation. At the same time, when performing principal component analysis and outputting the corresponding data, it will be output according to the cumulative variance contribution rate of the output features, that is, the sum of the features of the output data accounts for the proportion of the sum of all eigenvalues. At this time, 70% is set as the threshold of the cumulative variance contribution rate to select the situation of the deviation angle correlation matrix of the adjacent grid area when it becomes the corresponding eigenvalues using principal component analysis. Finally, obtain multiple data combinations similar to clustering, and calculate the average values of the corresponding deviation distances and deviation angles of these data combinations respectively to obtain the analysis result of the corresponding data at this time.

[0065] Set position encodings for the average deviation distance and average deviation angle corresponding to the position index of the data after clustering in sequence, and combine the position encodings according to the clustering dimension to obtain the particle distribution vector. The particle distribution vector represents the combination of the corresponding data after clustering in multiple dimensions. Finally, these values can be converted into a specific hash value to represent the result of metal particle recognition.

[0066] At the same time, the implementation method of the particle distribution module also includes: identifying the texture of each metal particle in the particle distribution image, obtaining the etching characteristics of each metal particle in the particle distribution state. The etching characteristics represent the texture after laser etching on the surface of the metal particle. Then, perform hierarchical analysis according to the etching characteristics and the metal particle distribution, and perform distribution layer verification and etching layer analysis respectively. Based on the analysis results of the two layers, complete the analysis of the particle distribution state. Since the textures corresponding to laser etching are mostly delicate contents, an optical microscope or an electron microscope can be used to magnify and observe the metal particles at this time, and the traces of laser etching can be clearly seen.

[0067] This processing method is generally used when similar suppliers, etc. verify product anti-counterfeiting labels, and is not applicable to the scenarios used by the cameras described above. The content described above can be the scenario where users use mobile phones or other devices to scan the QR code, color block border image, and particle distribution image after purchasing the product to complete the anti-counterfeiting label verification.

[0068] Perform hierarchical analysis on the particle distribution state according to the etching characteristics of each metal particle, and sequentially determine the distribution layer analysis result and the etching layer analysis result.

[0069] For the distribution layer analysis result, it represents the relevant numerical values of the particle distribution vector calculated using the position index. The etching layer analysis result is to identify the holographic image existing on the metal particle and determine whether the holographic image is consistent with the data characteristics bound to the two-dimensional code, so as to complete the recognition result of the etching layer.

[0070] For example, the etching layer analysis result is expressed as follows: perform feature matching between the etching characteristics and the image characteristics in the bound data of the two-dimensional code image, calculate the similarity of the etching characteristics, and when the similarity of the etching characteristics is greater than 95%, it is determined that the current etching layer analysis result is normal. The similarity can be calculated by representing the etching characteristics using a grayscale histogram and comparing the pixel point values of the etching characteristics with the corresponding pixel point values in the database, and using the calculation method of cosine similarity to complete the feature matching.

[0071] Construct the particle state vector according to the distribution layer analysis result and the etching layer analysis result.

[0072] At this time, when constructing the particle distribution state vector using two levels, since the distribution layer analysis result represents the process of constructing the particle state vector using the position index, at this time, the etching layer analysis result is used to perform data identification on the content in the distribution layer analysis result. For example, when there is an inconsistency between the etching layer analysis result and the data bound to the two-dimensional code, identify the metal particles corresponding to the inconsistency, record the metal particles at the corresponding positions, and upload the corresponding image data to facilitate subsequent analysis of the reasons for the inconsistency during the etching analysis of the current anti-counterfeiting label; if they are consistent, it is considered that the current metal particle analysis result is normal, and the content of the distribution layer analysis result is used to construct the particle state vector.

[0073] In an embodiment of the present invention, after completing the identification of the color block border and metal particles on the anti-counterfeiting label, it is possible to directly obtain whether the current anti-counterfeiting label is valid based on these two identification results. At the same time, since different users have different authentication contents for the anti-counterfeiting label, the verification method of the anti-counterfeiting label can be adjusted.

[0074] For example, ordinary users can directly scan the two-dimensional code and scan the corresponding color block border for verification, or in addition to verifying the color block border, they also need to verify the metal particle distribution. At this time, they can directly use a normal mobile device through a specific software for verification; if it is verified by manufacturers or distributors, etc., they will not only directly verify the color block border and metal particles, but also verify the texture existing on the metal particles, and judge whether the anti-counterfeiting label is valid based on the results of the three verifications.

[0075] Therefore, the implementation methods of the anti-counterfeiting identification module include: obtaining the identification requirements of the user, such as quick verification, detailed analysis, batch processing, etc., and determining the identification method of the current initial anti-counterfeiting image according to the user's identification requirements; for the quick verification requirement, the matching steps and output results can be simplified; for the detailed analysis requirement, a detailed matching report and particle status information can be provided; for the batch processing requirement, the processing flow can be optimized to improve efficiency.

[0076] According to the identification method of the current initial anti-counterfeiting image, use the edge matching feature and the particle status vector to verify the initial anti-counterfeiting image in sequence.

[0077] For example, under the quick verification requirement, the verification can be completed only through the output content of the edge matching feature. When obtaining the edge matching feature, it will be judged whether the border of the color block is compared with the data bound to the two-dimensional code image, that is, the edge matching feature will include a flag indicating successful matching. Under the quick verification requirement, the verification can be completed only through this method.

[0078] For detailed analysis, both the edge matching feature and the particle status vector will be used to complete the verification of the preliminary anti-counterfeiting image. For the particle status vector, a set of position-encoded data will be output, or the position encoding converted into a hash value. This value can be directly compared with the data bound to the two-dimensional code. If they are consistent, the verification of the preliminary anti-counterfeiting image is completed.

[0079] If it is batch processing and identifying a batch of anti-counterfeiting labels, the identification flag regarding the etching feature will be searched in the particle status vector, and the edge matching feature of the color block border will be used to find the deficiencies in the production of anti-counterfeiting labels in the current batch to complete the comprehensive authentication of the anti-counterfeiting labels.

[0080] As Figure 5 shown, the present invention also provides an artificial intelligence-based double-layer anti-counterfeiting label method, including: S1, collecting an initial anti-counterfeiting image containing an anti-counterfeiting label through an image acquisition module, and splitting it into a color block border image, a particle distribution image, and a two-dimensional code image.

[0081] S2, analyzing the color block border image to determine the edge features of each sub-image, and performing feature matching between the edge features of each sub-image and the two-dimensional code image to obtain an edge matching feature.

[0082] S3, performing matching analysis on the metal particle distribution in the particle distribution image to obtain the particle distribution state, and performing feature matching between the particle distribution state and the two-dimensional code image to obtain a particle status vector.

[0083] S4, verifying the initial anti-counterfeiting image based on the edge matching feature and the particle status vector, and determining the identification method according to the user's requirements to complete the anti-counterfeiting image identification.

[0084] Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those of ordinary skill in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present invention, and still be covered by the protection scope of the present invention.

Claims

1. A double-layer anti-counterfeiting label system based on artificial intelligence, characterized in that: include: An image acquisition module is used to collect an initial anti-counterfeiting image including an anti-counterfeiting label according to user needs, and split the initial anti-counterfeiting image into a color block border image, a particle distribution image and a QR code image; The color block analysis module is used to analyze the color block border image, determine the edge features of each sub-image in the color block border image, combine the edge features of each sub-image to obtain the edge features of the color block border image, perform feature matching on the edge features of the color block border image and the two-dimensional code image, and obtain the edge matching features of the color block border image; The particle distribution module is used to match and analyze the metal particle distribution in the particle distribution image to obtain the particle distribution state of the metal particles; perform feature matching between the particle distribution state and the two-dimensional code image to obtain a particle state vector; The implementation methods of feature matching between particle distribution state and QR code image include: Based on the distance between each metal particle in the particle distribution state, the particle distribution state is associated with the data corresponding to the two-dimensional code image to generate a relational data set of the particle distribution state; Obtaining position indexes of particle distribution states from the relational data set, combining the position indexes to obtain position combination indexes; Use the combined index of each position to classify the particle distribution state, and calculate the average deviation distance and average deviation angle of the position index under each classification; According to the average deviation distance and average deviation angle of the position index under each category, a position code is set for each category; according to the dimension corresponding to each category, the position codes are combined into a particle distribution vector; The anti-counterfeiting recognition module is used to verify the initial anti-counterfeiting image based on edge matching features and particle state vectors, determine the recognition method of the initial anti-counterfeiting image under different user requirements, and complete the anti-counterfeiting image recognition.

2. The double-layer anti-counterfeiting label system based on artificial intelligence according to claim 1, characterized in that: The implementation methods of edge features of color block border images include: According to the distribution positions of the color blocks in the color block border image, the color block border area is divided into sub-images corresponding to the color blocks in sequence; and the edge features of each sub-image are identified; According to the edge features of each sub-image, the position and area of ​​each color block in each sub-image are calculated; and according to the position and area of ​​each color block, the color block adjacency relationship between the color blocks is constructed; According to the color block adjacency relationship between the color blocks, the edge features of each sub-image are spliced ​​and output as the edge features of the color block border image.

3. The double-layer anti-counterfeiting label system based on artificial intelligence according to claim 2 is characterized in that: The implementation methods of identifying the edge features of each sub-image include: Perform contour detection on the color blocks in each sub-image, and set the contour pixel points corresponding to each color block according to the minimum circumscribed rectangle of the contour detection; The contour pixels are combined into an edge coordinate sequence, and the shape descriptor and Fourier descriptor of the contour pixels are extracted. The edge coordinate sequence, shape descriptor and Fourier descriptor are mapped to each other and output as edge features of each sub-image.

4. The double-layer anti-counterfeiting label system based on artificial intelligence according to claim 1, characterized in that: The implementation methods of edge matching features include: Based on the edge features of the color block border image, the relative distance and distribution sequence of each sub-image in the color block border image are determined, the preset edge features are obtained using the QR code image, the relative distance and distribution sequence of each sub-image are matched with the preset edge features, and the relative relationship and distribution quantity of multiple sub-images after matching are determined; Using the relative relationship and distribution number of sub-images, consistency verification is performed to identify the relevant particle size differences and area ratio differences of the sub-images, and obtain the degree of difference of the sub-images; It is determined whether the sub-image with the largest difference is a normal change. If it is a normal change, an anti-counterfeiting mark is set for the color block border image, and the distribution sequences of each sub-image are combined and output as the edge matching features of the color block border image.

5. The double-layer anti-counterfeiting label system based on artificial intelligence according to claim 1, characterized in that: The matching analysis of the metal particle distribution in the particle distribution image can be realized by: According to the position of the metal particles in the particle distribution image, the coordinates corresponding to each metal particle are identified to generate a particle coordinate sequence; The distance and angle between each metal particle are recorded, the particle coordinate sequence is converted into a topological structure matrix, and each metal particle in the topological structure matrix is ​​connected and output as a particle distribution state.

6. The double-layer anti-counterfeiting label system based on artificial intelligence according to claim 1, characterized in that: The implementation methods for classifying the particle distribution state include: Multi-dimensional hierarchical clustering of particle distribution states is performed, and spatial position, distribution pattern and deviation angle correlation clustering are performed in turn; The current metal particle distribution area is cut into multiple grid areas of the same size, each of which contains metal particles. The average deviation distance and average deviation angle of the corresponding position index of the metal particles in each grid area are calculated as the clustering result of the spatial position; The grid area is converted into a density histogram according to the density of the metal particles, and the density histogram is used for KL divergence clustering. The KL divergence corresponding to the probability distribution of the values ​​of the metal particles in the grid area and the area where the metal particles are distributed is calculated, and the metal particles in each grid area are clustered according to the value of the KL divergence to obtain multiple clusters. The average deviation distance and average deviation angle of the corresponding position index of the metal particles in each cluster are output to obtain the clustering result of the distribution pattern; According to the deviation angle of the position index, the deviation angle association matrix of each grid area is generated, and the principal component analysis is performed on the deviation angle association matrix of adjacent grid areas. The average deviation distance and average deviation angle of the position index corresponding to the principal component analysis output data are calculated as the clustering result of the deviation angle association.

7. The double-layer anti-counterfeiting label system based on artificial intelligence according to claim 1, characterized in that: The particle distribution module can also be implemented by: Identify the texture of each metal particle in the particle distribution image to obtain the etching characteristics of each metal particle under the particle distribution state; Performing a hierarchical analysis on the particle distribution state according to the etching characteristics of each metal particle, and determining the distribution layer analysis result and the etching layer analysis result in turn; According to the analysis results of the distribution layer and the etching layer, the particle state vector is constructed.

8. The double-layer anti-counterfeiting label system based on artificial intelligence according to claim 1, characterized in that: The implementation methods of the anti-counterfeiting identification module include: Obtaining the user's identification requirements, and determining the identification method of the current initial anti-counterfeiting image according to the user's identification requirements; According to the recognition method of the current initial anti-counterfeiting image, the initial anti-counterfeiting image is verified in turn using edge matching features and particle state vectors.

9. A double-layer anti-counterfeiting label method based on artificial intelligence, characterized in that: include: S1, collecting an initial anti-counterfeiting image including an anti-counterfeiting label through an image acquisition module, and splitting it into a color block border image, a particle distribution image and a QR code image; S2, analyzing the color block border image, determining the edge features of each sub-image, and matching the edge features of each sub-image with the two-dimensional code image to obtain edge matching features; S3, performing matching analysis on the metal particle distribution in the particle distribution image to obtain the particle distribution state, and performing feature matching between the particle distribution state and the two-dimensional code image to obtain a particle state vector; The implementation methods of feature matching between particle distribution state and QR code image include: Based on the distance between each metal particle in the particle distribution state, the particle distribution state is associated with the data corresponding to the two-dimensional code image to generate a relational data set of the particle distribution state; Obtaining position indexes of particle distribution states from the relational data set, combining the position indexes to obtain position combination indexes; Use the combined index of each position to classify the particle distribution state, and calculate the average deviation distance and average deviation angle of the position index under each classification; According to the average deviation distance and average deviation angle of the position index under each category, a position code is set for each category; according to the dimension corresponding to each category, the position codes are combined into a particle distribution vector; S4, based on edge matching features and particle state vectors, verifies the initial anti-counterfeiting image, determines the recognition method according to user needs, and completes the anti-counterfeiting image recognition.

Images