A trademark infringement detection method and system

By combining super-resolution image reconstruction technology and CNN detection model, feature sub-bases are extracted and similarity measurement calculations are performed, the problems of insufficient feature information and large amount of calculations in traditional trademark infringement detection are solved, and the accuracy of trademark infringement judgment is improved. and consistency.

CN118691857BActive Publication Date: 2024-11-05CHINA JILIANG UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411170737.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Priority Date
2023-12-29
Filing Date
2024-08-26
Publication Date
2024-11-05
Estimated Expiration
2044-08-26

AI Technical Summary

Technical Problem

In traditional trademark infringement detection methods, the image feature information is insufficient, the feature retrieval matching calculation is large, and the similarity measurement technology is single, resulting in low accuracy of prediction results, and low efficiency in obtaining evidence from offline manual methods and in a timely manner.

Method used

Combining super-resolution image reconstruction technology and CNN-based detection model, m feature sub-base collections are extracted, and similarity measurement calculations are performed to reduce the system calculation volume and improve the accuracy and objectivity of trademark infringement judgments.

Benefits of technology

It improves the accuracy and consistency of trademark infringement judgments, reduces the system's computing volume, and achieves faster and more accurate trademark similarity judgments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118691857B_ABST
    Figure CN118691857B_ABST
Patent Text Reader

Abstract

The present invention discloses a trademark infringement detection method and system, which combines super-resolution image reconstruction technology with a CNN-based detection model. The super-resolution image reconstruction technology improves the resolution of the trademark image to be detected, so as to fully extract the feature information in the subsequent CNN detection stage, and first extracts m feature sub-library sets during the CNN detection process, and then performs similarity measurement calculations, thereby reducing the amount of system operations and improving the accuracy of trademark infringement judgments. The present invention applies artificial intelligence technology to trademark infringement judgments, further improving the objectivity and consistency of trademark similarity judgments, and can solve the current dilemma in trademark infringement judicial practice. In addition, based on blockchain technology, the collection and storage of infringement evidence is carried out online, further improving the efficiency and security of evidence collection and storage.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application belongs to the field of artificial intelligence technology, and in particular, relates to a trademark infringement detection method and system. Background Art

[0002] Trademark infringement refers to the use of a trademark that is identical or similar to a registered trademark on identical or similar goods without the permission of the trademark owner, or other acts that interfere with or hinder the trademark owner from using its registered trademark and damage the legitimate rights and interests of the trademark owner. With the development of e-commerce, there are more and more trademark infringements on products sold online, and products are sold at low prices by using other people's brands, which brings relatively great pressure to the original product companies and brings some confusion to the market.

[0003] However, traditional trademark infringement detection methods have problems such as insufficient extraction of image feature information, large computational complexity of feature retrieval and matching, and single similarity measurement technology, resulting in low accuracy of prediction results. At the same time, trademark protection in the field of e-commerce is still mostly carried out in the traditional offline mode, that is, the trademark right holder obtains infringement evidence through offline manual methods and then conducts corresponding protection based on the evidence. However, the offline manual method of obtaining evidence has problems such as low efficiency and delay. Summary of the invention

[0004] In view of the defects in the prior art, the present invention provides a trademark infringement detection method and system, which combines super-resolution image reconstruction technology with a CNN-based detection model, and first extracts m feature sub-library sets during the CNN detection process, and then performs similarity measurement calculations, thereby reducing the amount of system operations, improving the accuracy of trademark infringement determination, and further improving the objectivity and consistency of trademark similarity determination. To achieve the above purpose, the present invention provides the following technical solutions:

[0005] A trademark infringement detection method comprises the following steps:

[0006] (1) Establishing a graphic trademark protection database; the graphic trademark protection database is used to collect and store officially announced graphic trademark information according to a preset collection strategy. The graphic trademark information includes trademark names, trademark images, and trademark holders;

[0007] (2) Collecting the graphic trademark information of the products to be tested on the e-commerce platform, wherein the graphic trademark information of the products to be tested includes the trademark name, trademark image, trademark holder or trademark licensor; wherein the collection of the graphic trademark information of the products to be tested is carried out by using a web crawler according to a predetermined strategy;

[0008] (3) Determine the resolution of the trademark image of the product to be detected; perform corresponding infringement detection based on the judgment result of the resolution; the specific implementation is as follows: set a threshold value according to the resolution size of the graphic trademark data set in the graphic trademark protection library, and directly input the trademark image of the product to be detected with a resolution greater than or equal to the threshold into the second neural network model for detection; input the trademark image of the product to be detected with a resolution less than the threshold into the first neural network model for super-resolution image reconstruction processing, and input the generated super-resolution reconstructed trademark image into the second neural network model for detection; the second neural network model outputs recognition result information, wherein the recognition result information is used to indicate whether the graphic trademark of the product to be detected is infringing;

[0009] (4) In response to the identification result information indicating that the trademark to be detected is infringing, the graphic trademark of the product to be detected is determined to be an infringing trademark, and the product is determined to be an infringing product; capturing infringement evidence, and uploading the infringement evidence to a blockchain-based electronic evidence platform, wherein the infringement evidence includes images of the infringing product trademark, links to the infringing product, screenshots of the infringing product webpage content, and information about the infringing owner.

[0010] Furthermore, the first neural network model is a generative adversarial network GAN, which is composed of a generator built by a deep network and another discriminator built by a deep network. During the working process, the generator is responsible for mapping the input noise image into a false image that obeys the distribution of the real image, and the discriminator is responsible for learning the characteristics of the real image and distinguishing whether the image input into the network is a real image or a false image generated by the generator. The two images input into the discriminator are calculated through the loss function, and the weight of the GAN is reversely updated according to the calculated loss value. Finally, when the GAN converges, the super-resolution reconstructed trademark image is output.

[0011] Furthermore, the loss function is a perceptual loss function composed of content loss and adversarial loss.

[0012] Furthermore, the generator is specifically composed of multiple residual blocks, each residual block combines a convolution layer, a batch normalization layer and a ReLU activation layer, and finally uses a sub-pixel convolution layer to achieve image size enlargement; the discriminator uses multiple convolution blocks for feature extraction and uses LeakyReLU as the activation function layer.

[0013] Furthermore, the second neural network model is a convolutional neural network (CNN); the CNN components include an input layer, a hidden layer and an output layer, wherein the hidden layer of the CNN is composed of a convolutional layer, a pooling layer, and a fully connected layer is added; the output layer is a Softmax classifier, the input of which is the score of the predicted image in a certain category, and the output is the probability of the predicted image being in that category.

[0014] Furthermore, the working process of the second neural network model CNN includes the following steps:

[0015] (3-1) In the training phase, the trademark images of different categories in the graphic trademark protection database are input into the CNN, the output of the fully connected layer is extracted as the CNN features of the trademark image, and the CNN features of each trademark image are combined to construct a trademark image feature database. ; Use the output of the output layer in the CNN model as the classification category probability of the trademark image to establish a classification category probability information database of the trademark image; trademark image feature database Defined as ,in Indicates The trademark image feature sub-library of categories, A database of trademark image features representing all categories;

[0016] (3-2) Input the image F to be detected into CNN to obtain the CNN features of the image F to be detected , and the set of candidate categories ranked top m by classification probability , m is a predefined natural number, the image F to be detected is a trademark image of the product to be detected with a resolution greater than or equal to a threshold or a super-resolution reconstructed trademark image; , Indicates that the image F to be tested is classified by CNN to obtain the classification category probability ranking The categories are composed of candidate categories; according to the classification results, from the trademark image feature database Extract , ,in express Classification probability ranking The feature sub-library of It is the feature sub-library set corresponding to the set of categories ranked top m in classification probability;

[0017] (3-3) Similarity metric calculation: Calculate the CNN features of the image F to be detected respectively With feature sub-library collection Similarity between And the classification probability of the image F to be detected and the set of candidate categories with the top m classification probability rankings Similarity between , using cosine distance as the measure of similarity, the two similarities are weighted and fused to form the total similarity ; , represents weighted weight;

[0018] (3-4) Set an infringement determination threshold to determine whether there is a total similarity greater than the infringement determination threshold; if so, select the trademark image Q with the largest total similarity, and compare whether the trademark holder or trademark licensor in the trademark information of the product graphic to be detected is consistent with the trademark holder corresponding to the trademark image Q; if they are inconsistent, determine that the product graphic trademark to be detected is infringing, and output the recognition result information.

[0019] Another object of the present invention is to provide a trademark infringement detection system, which is used to execute the above-mentioned trademark infringement detection method.

[0020] Another object of the present invention is to provide a computer-readable storage medium storing one or more programs, wherein the one or more programs enable a computer to execute the above-mentioned trademark infringement detection method.

[0021] In combination with all the above technical solutions, the present invention has the following advantages compared with the prior art:

[0022] The super-resolution image reconstruction technology is combined with the CNN-based detection model. The resolution of the trademark image to be detected is improved through the super-resolution image reconstruction technology, so as to fully extract the feature information in the subsequent CNN detection stage. In the CNN detection process, m feature sub-library sets are first extracted, and then the similarity measurement calculation is performed, which reduces the system operation amount and improves the accuracy of trademark infringement judgment. The present invention applies artificial intelligence technology to trademark infringement judgment, further improving the objectivity and consistency of trademark similarity judgment. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 It is a schematic diagram of the execution of the trademark infringement detection method of the present invention. DETAILED DESCRIPTION

[0024] The following is a sample of the embodiment and the attached Figure 1 Further description is given. Obviously, the described embodiments are partial embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work belong to the protection scope of the present invention.

[0025] The present invention provides a trademark infringement detection method, referring to Figure 1 The trademark infringement detection method shown in the figure is a schematic diagram for executing the trademark infringement detection method, and the trademark infringement detection method includes the following steps:

[0026] (1) Establishing a graphic trademark protection database; the graphic trademark protection database is used to collect and store officially announced graphic trademark information according to a preset collection strategy. The graphic trademark information includes trademark names, trademark images, and trademark holders;

[0027] (2) Collecting the graphic trademark information of the products to be tested on the e-commerce platform, which includes the trademark name, trademark image, trademark holder or trademark licensor;

[0028] The work of collecting the graphic trademark information of the goods to be tested is carried out through a web crawler according to a predetermined strategy;

[0029] (3) Determine the resolution of the trademark image of the product to be detected; perform corresponding infringement detection based on the judgment result of the resolution; the specific implementation is as follows: set a threshold according to the resolution size of the graphic trademark data set in the graphic trademark protection library, and directly input the trademark image of the product to be detected with a resolution greater than or equal to the threshold into the second neural network model for detection; input the trademark image of the product to be detected with a resolution less than the threshold into the first neural network model for super-resolution image reconstruction processing, and input the generated super-resolution reconstructed trademark image into the second neural network model for detection; the second neural network model outputs recognition result information, wherein the recognition result information is used to indicate whether the graphic trademark of the product to be detected is infringing;

[0030] (4) In response to the identification result information indicating that the trademark to be detected is infringing, the graphic trademark of the product to be detected is determined to be an infringing trademark, and the product is determined to be an infringing product; capturing infringement evidence, and uploading the infringement evidence to a blockchain-based electronic evidence platform, wherein the infringement evidence includes images of the infringing product trademark, links to the infringing product, screenshots of the infringing product webpage content, and information about the infringing owner.

[0031] Preferably, the first neural network model is a generative adversarial network GAN, which is composed of a generator built by a deep network and another discriminator built by a deep network. During the working process, the generator is responsible for mapping the input noise picture into a false picture that obeys the distribution of the real picture, and the discriminator is responsible for learning the characteristics of the real picture and distinguishing whether the picture input into the network is a real picture or a false picture generated by the generator. The two pictures input into the discriminator are calculated through the loss function, and the weight of the GAN is reversely updated according to the calculated loss value, so as to achieve the effect of making the generator learn the distribution of real data. Finally, when the GAN converges, the super-resolution reconstructed trademark picture is output.

[0032] Preferably, the loss function is a perceptual loss function composed of content loss and adversarial loss. Different from the mean square error loss function used in traditional super-resolution algorithms, the perceptual loss function can effectively improve the quality of the reconstructed image.

[0033] Preferably, the generator is specifically composed of multiple residual blocks, each residual block combines a convolution layer, a batch normalization layer (Batch Normalization layer, referred to as BN layer) and a ReLU activation layer, and finally uses a sub-pixel convolution layer to achieve image size enlargement; the discriminator uses multiple convolution blocks to extract features, and uses LeakyReLU as the activation function layer. After the input image is extracted by the convolution layer, the number of channels of the feature map continues to increase and the size continues to decrease, and finally the probability of the authenticity of the image is obtained through two fully connected layers.

[0034] We found that the resolution of the images of the trademarks to be detected is relatively low, and the low-resolution trademark images contain less feature information than the high-resolution trademark images. The present invention uses super-resolution reconstruction technology based on generative adversarial networks as image preprocessing for trademark infringement detection, so that low-resolution trademark images are constructed into high-resolution images and then input into the trademark infringement detection model. Due to the improvement of the resolution of trademark images, it will be easier to extract useful information from trademark images, thereby improving the accuracy of image feature extraction and matching in trademark image infringement judgment.

[0035] Preferably, the second neural network model is a convolutional neural network (CNN); the CNN components include an input layer, a hidden layer and an output layer, wherein the hidden layer of the CNN consists of a convolutional layer, a pooling layer, and a fully connected layer is added.

[0036] The fully connected layer in CNN converts the input feature map into a feature vector, which contains crucial information about the input image. Because the convolutional layer before the fully connected layer saves information about the local features of the input image (such as color, texture, and shape, etc.), the fully connected layer saves the combined and aggregated information of all convolutional layers. The output layer is a Softmax classifier, which takes as input the score of the predicted image in a certain category and outputs the probability of the predicted image being in that category.

[0037] Preferably, the working process of the second neural network model CNN includes the following steps:

[0038] (3-1) In the training phase, the trademark images of different categories in the graphic trademark protection database are input into the CNN, the output of the fully connected layer is extracted as the CNN features of the trademark image, and the CNN features of each trademark image are combined to construct a trademark image feature database. The output (prob) of the output layer in the CNN model is used as the classification category probability of the trademark image, and a classification category probability information database of the trademark image is established. This category probability is a multidimensional vector after passing through the Softmax classifier, representing the probability that the sample belongs to a given class. Trademark Image Feature Database Defined as ,in Indicates The trademark image feature sub-library of categories, A database of trademark image features representing all categories.

[0039] (3-2) Input the image F to be detected into CNN to obtain the CNN features of the image F to be detected , and the set of candidate categories ranked top m by classification probability , m is a predefined natural number, and the image F to be detected is a trademark image of the product to be detected with a resolution greater than or equal to a threshold or a super-resolution reconstructed trademark image.

[0040] in , Indicates that the image F to be tested is classified by CNN to obtain the classification category probability ranking The categories are composed of candidate categories, and according to the classification results, the trademark image feature database is selected. Extract , ,in express Classification probability ranking The feature sub-library of It is the feature sub-library set corresponding to the set of categories ranked in the top m in terms of classification probability.

[0041] (3-3) Similarity metric calculation: Calculate the CNN features of the image F to be detected respectively With feature sub-library collection Similarity between And the classification probability of the image F to be detected and the set of candidate categories with the top m classification probability rankings Similarity between , using cosine distance as the measure of similarity, the two similarities are weighted and fused to form the total similarity ; , Represents weighted weight.

[0042] (3-4) Set an infringement determination threshold to determine whether there is a total similarity greater than the infringement determination threshold; if so, select the trademark image Q with the largest total similarity, and compare whether the trademark holder or trademark licensor in the trademark information of the product graphic to be detected is consistent with the trademark holder corresponding to the trademark image Q; if they are inconsistent, determine that the product graphic trademark to be detected is infringing, and output the recognition result information.

[0043] In addition, the present invention provides a trademark infringement detection system, which is used to execute the above-mentioned trademark infringement detection method.

[0044] It should be noted that those skilled in the art should understand that the implementation functions of the system in the implementation of the trademark infringement detection system can be understood by referring to the relevant description of the trademark infringement detection. The functions of the system shown in the implementation of the trademark infringement detection system can be implemented by a program (executable instruction) running on a processor, or by a specific logic circuit.

[0045] Accordingly, the present invention also provides a computer-readable storage medium, in which computer executable instructions are stored, and when the computer executable instructions are executed by a processor, the various method implementations of the present application are implemented. Computer-readable storage media include permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. The information can be a computer-readable instruction, a data structure, a module of a program, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, read-only compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device.

[0046] In addition, it should be understood that the above description is only a preferred embodiment of this specification and is not intended to limit the scope of protection of this specification. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of one or more embodiments of this specification should be included in the scope of protection of one or more embodiments of this specification.

Claims

1. A trademark infringement detection method, characterized in that: The following steps are involved: (1) Establishing a graphic trademark protection database; the graphic trademark protection database is used to collect and store officially announced graphic trademark information according to a preset collection strategy. The graphic trademark information includes trademark names, trademark images, and trademark holders; (2) Collecting the graphic trademark information of the goods to be tested, which includes the trademark name, trademark image, trademark holder or trademark licensor; (3) setting a threshold value according to the resolution size of the graphic trademark data set in the graphic trademark protection library, directly inputting the trademark image of the product to be detected with a resolution greater than or equal to the threshold value into the second neural network model for detection; inputting the trademark image of the product to be detected with a resolution less than the threshold value into the first neural network model for super-resolution image reconstruction processing, and inputting the generated super-resolution reconstructed trademark image into the second neural network model for detection; the second neural network model outputs recognition result information, wherein the recognition result information is used to indicate whether the graphic trademark of the product to be detected is infringing; (4) If it is judged to be an infringement, the graphic trademark of the product to be tested will be determined as an infringing trademark; The first neural network model in step (3) is a generative adversarial network (GAN), which is composed of a generator constructed by a deep network and a discriminator constructed by another deep network. During operation, the generator is responsible for mapping the input noise image into a false image that obeys the distribution of the real image, and the discriminator is responsible for learning the features of the real image and distinguishing whether the image input into the network is a real image or a false image generated by the generator. The two images input into the discriminator are calculated through the loss function, and the weight of the GAN is reversely updated according to the calculated loss value. Finally, when the GAN converges, the super-resolution reconstructed trademark image is output; The second neural network model in step (3) is a convolutional neural network (CNN); the CNN components include an input layer, a hidden layer and an output layer, wherein the hidden layer of the CNN is composed of a convolutional layer, a pooling layer, and a fully connected layer; the output layer is a Softmax classifier, the input of which is the score of the predicted image in a certain category, and the output is the probability of the predicted image being in this category; the working process of the second neural network model CNN includes the following steps: (3-1) In the training phase, the trademark images of different categories in the graphic trademark protection library are input into the CNN, the output of the fully connected layer is extracted as the CNN features of the trademark image, and the CNN features of each trademark image are combined to construct a trademark image feature database. ; Use the output of the output layer in the CNN model as the classification category probability of the trademark image to establish a classification category probability information database of the trademark image; trademark image feature database Defined as ,in Indicates The trademark image feature sub-library of categories, A database of trademark image features representing all categories; (3-2) Input the image F to be detected into CNN to obtain the CNN features of the image F to be detected , and the set of candidate categories ranked top m by classification probability , m is a predefined natural number, the image F to be detected is a trademark image of the product to be detected with a resolution greater than or equal to a threshold or a super-resolution reconstructed trademark image; , Indicates that the image F to be tested is classified by CNN to obtain the classification category probability ranking The categories are composed of candidate categories; according to the classification results, from the trademark image feature database Extract , ,in express Classification probability ranking The feature sub-library of It is the feature sub-library set corresponding to the set of categories ranked top m in classification probability; (3-3) Similarity metric calculation: Calculate the CNN features of the image F to be detected respectively With feature sub-library collection The similarity between And the classification probability of the image F to be detected and the set of candidate categories with the top m classification probability rankings The similarity between , using cosine distance as the measure of similarity, weighted fusion of these two similarities is performed to form the total similarity ; , represents weighted weight; (3-4) Set an infringement determination threshold to determine whether there is a total similarity greater than the infringement determination threshold; if so, select the trademark image Q with the largest total similarity, and compare whether the trademark holder or trademark licensor in the trademark information of the product graphic to be detected is consistent with the trademark holder corresponding to the trademark image Q; if they are inconsistent, determine that the product graphic trademark to be detected is infringing, and output the recognition result information.

2. A trademark infringement detection method as claimed in claim 1, characterized in that: The loss function is a perceptual loss function composed of content loss and adversarial loss.

3. A trademark infringement detection method as claimed in claim 1, characterized in that: The generator is specifically composed of multiple residual blocks, each residual block combines a convolution layer, a batch normalization layer and a ReLU activation layer, and finally uses a sub-pixel convolution layer to achieve image size enlargement; the discriminator uses multiple convolution blocks for feature extraction and uses LeakyReLU as the activation function layer.

4. A trademark infringement detection method as claimed in claim 1, characterized in that: The collecting of graphic trademark information of the goods to be detected is collecting graphic trademark information of the goods to be detected on the e-commerce platform.

5. A trademark infringement detection method as claimed in claim 1, characterized in that: If it is judged to be infringement, the product will be identified as an infringing product; infringement evidence will be captured and uploaded to the blockchain-based electronic evidence platform, where the infringement evidence includes trademark images of the infringing product, links to the infringing product, screenshots of the infringing product web page content, and information about the infringing owner.

6. A trademark infringement detection system, characterized in that: The trademark infringement detection system is used to execute the method according to any one of claims 1-5.

7. A computer-readable storage medium, characterized in that: One or more programs are stored, and the one or more programs enable a computer to execute the method according to any one of claims 1 to 5.