Counterfeit commodity supervision and management method and system

Abstract

The invention relates to the field of intelligent management, and particularly discloses a counterfeit commodity supervision and management method and system, and the method comprises the steps: obtaining an image of a commodity and a picture of a certified commodity in a database through employing an artificial intelligence technology based on a deep neural network model, and carrying out the feature extraction of the image, and comparing and analyzing the features of the certified commodity image database to determine whether the commodity is a counterfeit commodity or not. The method can assist in identifying counterfeit commodities, so that the commodity supervision efficiency and accuracy are improved, and consumers and supervision departments are helped to better deal with counterfeit commodity problems.

Description

Technical Field

[0001] This application relates to the field of intelligent management, and more specifically, to a method and system for supervising and managing counterfeit goods. Background Technology

[0002] Counterfeit goods refer to products manufactured or sold using deceptive means that impersonate another person's brand or counterfeit another person's well-known goods. These products are usually similar to genuine products in appearance, packaging, trademarks, and logos, with the aim of misleading consumers into believing that they have purchased genuine products, thereby gaining profit.

[0003] Counterfeit goods may present quality problems, safety hazards, or involve fraud, posing risks to consumers' health and property. They also create unfair competition for legitimate businesses, which invest significant resources and effort in product development, brand promotion, and quality control. The presence of counterfeit goods can lead to decreased sales and harm the interests of legitimate businesses. Furthermore, counterfeit goods can cause price chaos, a crisis of consumer trust, and damage to brand image. Therefore, effective supervision and management can help consumers avoid purchasing low-quality or harmful counterfeit goods, protect consumer rights, combat counterfeiting, safeguard the rights of legitimate businesses, reduce the circulation of counterfeit goods, and maintain market order and stability.

[0004] However, traditional methods typically rely on manual supervision and detection of counterfeit goods. This requires significant human resources and time, increasing costs and workload, and hindering efficient large-scale monitoring. Secondly, counterfeiters often employ various methods to circumvent traditional monitoring. They may use disguises, repackage, or alter labels to conceal the true nature of counterfeit goods, making them difficult to detect. Traditional monitoring methods have limited ability to identify and counter these evasion tactics.

[0005] Therefore, we look forward to an optimized scheme for supervising and managing counterfeit goods. Summary of the Invention

[0006] To address the aforementioned technical problems, this application is proposed. Embodiments of this application provide a method and system for supervising and managing counterfeit goods. This method employs artificial intelligence technology based on a deep neural network model to acquire images of the product and images of genuine products in a database. Feature extraction is performed on the images, and the features from the genuine product image database are compared and analyzed to determine whether the product is counterfeit. This method can assist in identifying counterfeit goods, thereby improving the efficiency and accuracy of product supervision and helping consumers and regulatory authorities better address the problem of counterfeit goods.

[0007] According to one aspect of this application, a method for supervising and managing counterfeit goods is provided, comprising: Retrieve product images and images of genuine products from the database; Image feature analysis is performed on the image of the product and the image of genuine products in the database to obtain the product image feature map and the genuine product feature map, respectively. The difference between the product image feature map and the genuine product feature map is calculated to obtain a product contrast difference feature map, which is then segmented into multiple product local contrast difference feature maps. The multiple local contrast difference feature maps of the products are subjected to multi-scale fusion feature processing to obtain a global depth feature map of product contrast difference. Based on the product's comparative differential global depth feature map, it is determined whether the product is counterfeit.

[0008] According to another aspect of this application, a counterfeit goods supervision and management system is provided, comprising: The product image acquisition module is used to acquire images of products and pictures of genuine products in the database; The product image feature analysis module is used to perform image feature analysis on the image of the product and the image of genuine products in the database to obtain product image feature map and genuine product feature map respectively; The product feature difference calculation module is used to calculate the difference between the product image feature map and the genuine product feature map to obtain the product comparison difference feature map and segment it into multiple product local comparison difference feature maps; The multi-scale fusion feature processing module for commodities is used to perform multi-scale fusion feature processing on the multiple local contrast difference feature maps of commodities to obtain a global depth feature map of commodity contrast difference. The counterfeit product identification module is used to determine whether a product is counterfeit based on the product's comparative differential global depth feature map.

[0009] Compared with existing technologies, this application provides a method and system for supervising and managing counterfeit goods. It employs artificial intelligence technology based on a deep neural network model to acquire images of the product and images of genuine products in a database, extracts features from the images, and compares these features with those from the database of genuine product images to determine whether the product is counterfeit. This method can assist in identifying counterfeit goods, thereby improving the efficiency and accuracy of product supervision and helping consumers and regulatory authorities better address the problem of counterfeit goods. Attached Figure Description

[0010] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0011] Figure 1 This is a flowchart of a method for supervising and managing counterfeit goods according to an embodiment of this application.

[0012] Figure 2 This is a schematic diagram of the architecture of a counterfeit goods supervision and management method according to an embodiment of this application.

[0013] Figure 3 This is a flowchart illustrating the process of performing image feature analysis on the image of the product and the image of the genuine product in the database to obtain the product image feature map and the genuine product feature map, respectively, in the counterfeit product supervision and management method according to the embodiments of this application.

[0014] Figure 4 This is a flowchart illustrating the process of extracting image features from an image of a product to obtain a product image feature map in a method for supervising and managing counterfeit goods according to an embodiment of this application.

[0015] Figure 5 This is a block diagram of a counterfeit goods supervision and management system according to an embodiment of this application. Detailed Implementation

[0016] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0017] It should be understood that the steps described in the method embodiments of this disclosure may be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of this disclosure is not limited in this respect.

[0018] In the description of embodiments of this disclosure, the term "comprising" and similar terms should be understood as open-ended inclusion, i.e., "including but not limited to". The term "based on" should be understood as "at least partially based on". The term "one embodiment" or "the embodiment" should be understood as "at least one embodiment". The terms "first", "second", etc., may refer to different or the same objects. Other explicit and implicit definitions may also be included below.

[0019] It should be noted that the use of the terms "a" or "a plurality of" in this disclosure is illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0020] Counterfeit goods refer to products manufactured and sold by forging or impersonating other people's brands, trademarks, patents, or well-known products. Specifically, counterfeit goods often misuse well-known brands or trademarks, are manufactured with low-cost, inferior materials and processes, and are made to resemble or be identical to genuine products through imitation of appearance, labeling, and packaging to deceive consumers. Compared to genuine products, they are of inferior quality, prone to quality problems, and may pose a threat to consumers' health and safety. Secondly, counterfeit goods are usually sold at prices lower than the market price to attract consumers. This is because counterfeiters often do not bear the costs of legitimate channels and brand owners, allowing them to obtain high profits at lower prices. Furthermore, counterfeit goods often use forged packaging, labels, certificates, etc., making them difficult to distinguish from genuine products in appearance. Counterfeit goods cause significant harm to consumers, brand owners, and market order. Consumers who purchase counterfeit goods may suffer economic losses and safety risks, brand owners' reputations and market share may be damaged, and market order may be disrupted. Therefore, combating counterfeit goods through supervision and management is an important task for protecting consumer rights, maintaining market order, and promoting economic development.

[0021] Current technologies rely heavily on manual supervision and inspection, making the monitoring and identification of counterfeit goods a labor-intensive and time-consuming process. This hinders efficient large-scale monitoring, leading to increased costs and workload. Furthermore, counterfeiters frequently employ various strategies to evade traditional methods. They may disguise products, alter packaging, or tamper with labels to conceal the true nature of counterfeit goods, making them difficult to detect. Traditional monitoring methods have limited ability to identify and overcome these evasion techniques. Therefore, an optimized method for supervising and managing counterfeit goods is desired. This method involves acquiring images of the product and comparing them with images of genuine products in a database, performing image feature analysis and comparative analysis to further determine whether the product is counterfeit. This approach can improve the efficiency and accuracy of product supervision, helping consumers and regulatory authorities better address the counterfeit goods problem.

[0022] Figure 1This is a flowchart of a method for supervising and managing counterfeit goods according to an embodiment of this application. Figure 2 This is a schematic diagram of the architecture of a method for supervising and managing counterfeit goods according to an embodiment of this application. Figure 1 and Figure 2 As shown, the counterfeit goods supervision and management method according to an embodiment of this application includes: S110, acquiring an image of a product and an image of a genuine product in a database; S120, performing image feature analysis on the image of the product and the image of the genuine product in the database to obtain a product image feature map and a genuine product feature map, respectively; S130, calculating the difference between the product image feature map and the genuine product feature map to obtain a product comparison difference feature map and segmenting it into multiple product local comparison difference feature maps; S140, performing multi-scale fusion feature processing on the multiple product local comparison difference feature maps to obtain a product comparison difference global depth feature map; and S150, determining whether the product is a counterfeit product based on the product comparison difference global depth feature map.

[0023] In step S110, an image of the product and images of genuine products in the database are acquired. It should be understood that by acquiring the product image and comparing it with images of genuine products in the database, comparison and identification can be performed to detect and distinguish between genuine and counterfeit products. This image comparison method can assist traditional manual supervision and inspection work, improving efficiency and accuracy.

[0024] In step S120, image feature analysis is performed on the image of the product and the image of a genuine product in the database to obtain a product image feature map and a genuine product feature map, respectively. Accordingly, considering that both the product image and the genuine product image contain key features of the product, such as color, texture, and shape, and these key features are very useful for subsequent comparative analysis of the products, the technical solution of this application, by performing image feature analysis on the image of the product and the image of a genuine product in the database to obtain a product image feature map and a genuine product feature map, can extract key features related to the product from the product image and the genuine product image, providing a basis and support for the subsequent identification of counterfeit products.

[0025] Figure 3 This is a flowchart illustrating the process of performing image feature analysis on an image of a product and an image of a genuine product in a counterfeit goods supervision and management method according to an embodiment of this application, to obtain a product image feature map and a genuine product feature map, respectively. Specifically, in an embodiment of this application, as... Figure 3As shown, image feature analysis is performed on the image of the product and the image of genuine products in the database to obtain the product image feature map and the genuine product feature map, respectively. The process includes: S210, extracting image features from the image of the product to obtain the product image feature map; and S220, inputting the image of genuine products in the database into a convolutional neural network model as a filter to obtain the genuine product feature map.

[0026] Figure 4 This is a flowchart illustrating the process of extracting image features from an image of a product to obtain a product image feature map in a counterfeit goods supervision and management method according to an embodiment of this application. Specifically, in an embodiment of this application, as... Figure 4 As shown, the process of extracting image features from the image of the product to obtain the product image feature map includes: S310, performing image denoising on the image of the product to obtain a denoised product image; and S320, applying spatial attention enhancement to the denoised product image to obtain the product image feature map.

[0027] Specifically, in step S310, image denoising is performed on the product image to obtain a denoised product image. Accordingly, considering that product images may be affected by various types of noise, such as image sensor noise, compression noise, and noise caused by changes in illumination, noise and interference may affect the observation and analysis of the product image, leading to misjudgment or misunderstanding. Therefore, in the technical solution of this application, image denoising is further performed on the product image to obtain a denoised product image. This effectively removes these noises, improves the visualization effect of the product image, and makes details clearer. This is crucial for observing the details, textures, markings, and other features of the product, helping to more accurately analyze and identify the product, thereby improving the accuracy and reliability of product identification, authenticity judgment, and user experience.

[0028] More specifically, in embodiments of this application, image denoising of the product image to obtain a denoised product image includes: inputting the product image into a product image denoising module based on an automatic codec to obtain the denoised product image.

[0029] It's worth noting that the product image denoising module based on an autoencoder is a method for denoising product images using an autoencoder algorithm. Specifically, an autoencoder is an unsupervised learning algorithm consisting of an encoder and a decoder. The encoder maps input data (e.g., a product image) to a latent space representation, while the decoder maps the latent space representation back to the reconstructed input data. By training the autoencoder, it can learn the low-dimensional representation and reconstruction capabilities of the data. In the product image denoising module, the autoencoder is used to learn the low-dimensional representation and reconstruction capabilities of product images. The training process includes the following steps: First, a training set is prepared, containing the original product image and its corresponding noisy image. Then, the original product image is input to the encoder, which maps it to a latent space representation. The encoder parameters are trained by minimizing the reconstruction error, i.e., the difference between the original image and the image reconstructed by the decoder. Next, the encoder's output is used as the decoder's input, which maps it back to the reconstructed product image. The decoder parameters are trained by minimizing the reconstruction error. Finally, using the trained encoder and decoder, the noisy product image is input to the encoder, and the decoder generates the denoised image. The key idea behind automatic codecs is to extract clean features from noisy images by learning the low-dimensional representation and reconstruction capabilities of data. Through proper training and tuning, automatic codecs can effectively reduce noise in product images, providing clearer and more realistic images. This product image denoising module based on an automatic codec can be used to preprocess product image data, improving the accuracy and reliability of subsequent product recognition, authenticity verification, and image analysis tasks.

[0030] Specifically, in another feasible embodiment of this application, image denoising of the product image to obtain a denoised product image can be carried out through the following steps: 1. Collect a set of original product images as a training set. Generate a noisy version for each original image. Different noise models, such as Gaussian noise, salt-and-pepper noise, etc., can be used to add noise to the original images. 2. Use a noise estimation algorithm to estimate the noise in the noisy product image. The noise estimation algorithm can be based on statistical methods, model fitting, or deep learning methods. 3. Based on the noise estimation result, select an appropriate filter to reduce noise. Common filters include mean filters, median filters, Gaussian filters, etc. 4. Input the noisy product image into the selected filter to filter the image. The filter will adjust the pixel values ​​of the image according to the characteristics of its neighboring pixels to reduce the influence of noise. 5. Optimize and adjust the filter parameters according to the filtering effect and denoising quality. Evaluation metrics (such as mean square error, structural similarity index, etc.) can be used to evaluate the denoising effect, and the parameters are adjusted according to the evaluation results. 6. Image enhancement algorithms can be applied to improve the quality and visual effect of denoised product images. These algorithms may include contrast enhancement, sharpening, and color correction. In this embodiment, noise reduction of the product image is performed through steps such as noise estimation, filter selection, and parameter optimization. By selecting appropriate filters and optimizing parameters, noise in the image can be reduced, and image quality can be improved. When applied to actual product images, image enhancement can be performed as needed to further improve image quality and visualization.

[0031] Specifically, in step S320, the denoised product image is enhanced with spatial attention to obtain the product image feature map. It should be understood that a product image may contain multiple important regions, such as product logos, trademarks, and feature details, and the appearance and details of the product are crucial for subsequent identification of counterfeit goods. Therefore, to highlight important features in the product image, the model can focus more on the important parts of the product image and improve the accuracy of understanding and recognizing product features, as well as the attention paid to these features. In the technical solution of this application, the denoised product image is further enhanced with spatial attention to obtain the product image feature map, so as to understand the areas and criteria that the model focuses on during the decision-making process, thereby improving the understanding and analysis capabilities of the product.

[0032] More specifically, in embodiments of this application, obtaining the product image feature map by spatial attention enhancement of the denoised product image includes: obtaining the product image feature map by passing the denoised product image through a spatial attention mechanism based on a convolutional neural network.

[0033] In particular, in another possible implementation of this application, the denoised product image can be enhanced with spatial attention to obtain the product image feature map. STN is a mechanism for enhancing a model's attention to spatial transformations of an image. It can learn how to perform spatial transformations such as translation, rotation, and scaling on the input image, thereby improving the model's attention to and adaptability to the image. The following are the steps for implementing spatial attention enhancement using STN. First, input the denoised product image. Then, construct a convolutional neural network (CNN) model containing STN. This model can be divided into two parts: a feature extractor and a spatial transformer. Next, input the denoised product image into the feature extractor, extracting the image's feature representation through operations such as convolutional layers and pooling layers. Then, add a spatial transformer module to the output of the feature extractor. This module consists of the following: 1. Localization Network: A small neural network used to learn the parameters of spatial transformations. It can predict the parameters of transformations such as translation, rotation, and scaling. 2. Grid Generator: Generates a sampling grid based on the parameters output by the localization network. The grid defines the sampling position of each pixel in the input image. 3. Bilinear Sampler: Based on the generated sampling grid, the input image is sampled to obtain the transformed image. Then, the outputs of the feature extractor and the spatial transformer are fused, which can be achieved through simple element-wise addition or other fusion strategies. Finally, a product image feature map is obtained, highlighting key regions and features for subsequent product analysis, recognition, or other tasks. The key idea of ​​STN is to learn the parameters of the spatial transformer, enabling the model to adaptively adjust the spatial transformation of the input image, thereby enhancing its focus on key regions. This attention mechanism improves the model's robustness and generalization ability, making it suitable for various product image processing tasks.

[0034] Specifically, in step S220, images of genuine products from the database are input as filters into a convolutional neural network model to obtain the feature map of the genuine products. Furthermore, considering that convolutional neural networks (CNNs) are a widely used deep learning model in image processing tasks, and that genuine product images contain key information such as brand trademarks and logos, this application's technical solution uses images of genuine products from the database as filters into a convolutional neural network model to obtain the feature map of the genuine products. By using a CNN model, its multi-layer convolution and pooling operations can be utilized to extract image features, thereby capturing information such as local structure, texture, and shape in the image. This helps reduce differences caused by factors such as different angles and lighting, and helps distinguish differences between different products, thus generating a discriminative feature map of genuine products.

[0035] In step S130, the difference between the product image feature map and the genuine product feature map is calculated to obtain a product comparison difference feature map, which is then segmented into multiple product local comparison difference feature maps. Accordingly, in the context of product image processing, calculating the difference between the product image feature map and the genuine product feature map helps us obtain information about the differences between products. These differences may include variations in shape, color, texture, etc., which helps us more easily discover and understand the differences between products. Specifically, in the technical solution of this application, the difference between the product image feature map and the genuine product feature map is further calculated to obtain a product comparison difference feature map, which is then segmented into multiple product local comparison difference feature maps. This allows us to focus on the key features in the product image that differ from the genuine product. These features may be information related to the product brand, logo, packaging design, etc. By highlighting these features, we can more accurately compare and identify products. Furthermore, segmenting the product comparison difference feature map into multiple product local comparison difference feature maps can further extract local difference information about the product. This approach allows for more granular comparisons of differences between products, such as comparing distinctive features, patterns, or text, and provides more useful information for subsequent product analysis, identification, and authentication tasks, thereby improving the effectiveness and accuracy of product processing.

[0036] Specifically, in this embodiment, calculating the difference between the product image feature map and the genuine product feature map to obtain a product comparison difference feature map and segmenting it into multiple product local comparison difference feature maps includes: calculating the product comparison difference feature map between the product image feature map and the genuine product feature map using the following difference formula; wherein, the difference formula is: in, This represents the feature map of the product image. This indicates the difference based on position. This represents the feature image of the genuine product, and This represents the product comparison difference feature map.

[0037] In step S140, the multiple local contrast difference feature maps of the products are subjected to multi-scale fusion feature processing to obtain a global depth feature map of product contrast difference. It should be understood that product images typically contain features at different scales, such as overall appearance and detailed texture. Different scales contain different feature information. Therefore, in the technical solution of this application, by performing multi-scale fusion feature processing on the multiple local contrast difference feature maps of the products to obtain a global depth feature map of product contrast difference, we can comprehensively consider information at different scales, enabling the contrast difference feature map to more comprehensively express the characteristics of the product. Feature maps at different scales can capture different levels of detail and structural information, thereby providing more feature representation options, increasing the richness and expressive power of the feature map, and more accurately distinguishing the differences between products, making subsequent product analysis, recognition, and identification tasks more accurate and reliable.

[0038] Specifically, in this embodiment of the application, the multi-scale fusion feature processing of the multiple product local contrast difference feature maps to obtain a product contrast difference global depth feature map includes: passing the multiple product local contrast difference feature maps through a deep-shallow fusion convolutional network model to obtain multiple product local contrast difference depth feature maps; and concatenating the multiple product local contrast difference depth feature maps to obtain the product contrast difference global depth feature map.

[0039] Specifically, in this embodiment, the multiple product local contrast difference feature maps are processed through a shallow-deep fusion convolutional network model to obtain multiple product local contrast difference depth feature maps. Accordingly, considering that the shallow-deep fusion module can extract features from different network layers, including shallow and deep features, shallow features typically contain more low-level visual features, such as edges and textures, focusing more on detailed information. Deep features, on the other hand, are more abstract and semantic, providing higher-level semantic information. In the product contrast difference task, it is necessary to capture both the subtle differences between products and understand their semantic meaning. Based on this, in the technical solution of this application, the multiple local contrast difference feature maps of the products are processed through a deep-shallow fusion convolutional network model to obtain multiple local contrast difference depth feature maps of the products. By fusing the features of the shallow and deep layers, the trade-off between semantic information and detailed information can be balanced during the feature extraction process, capturing abstract and detailed information at different levels. This results in feature maps that have both semantic understanding capabilities and can retain detailed differences. By comprehensively utilizing their advantages, a more comprehensive and richer feature expression can be provided, which helps to better distinguish the differences between products and improve the accuracy of subsequent tasks (such as product recognition, classification, etc.).

[0040] It's worth noting that the deep-shallow fusion module is a module used in deep learning models to fuse feature representations from different levels, thereby improving feature expressiveness and generalization performance. It is commonly used in the design of Convolutional Neural Networks (CNNs) or other deep learning models. Specifically, the main function of the deep-shallow fusion module is to fuse features from different levels to comprehensively utilize their strengths and information. Generally, deeper features possess higher-level semantic information, while shallower features focus more on details and low-level visual features. The deep-shallow fusion module obtains more comprehensive and richer feature representations by appropriately combining and fusing these features.

[0041] There are several ways to implement a shallow-deep fusion module, one common approach being the use of skip connections or multi-scale convolution operations. For example, shallow-deep fusion can be achieved by element-wise addition, concatenation, or convolution of shallow and deep features. This balances semantic and detail information during feature extraction, thereby improving feature expressiveness and generalization performance. In summary, a shallow-deep fusion module is used in deep learning models to fuse feature representations from different levels. Through reasonable combination and fusion, it comprehensively utilizes the advantages of both deep and shallow features, enhancing feature richness and expressiveness, thus improving model performance across various tasks.

[0042] In step S150, based on the product comparison differential global depth feature map, it is determined whether the product is counterfeit. It should be understood that the product comparison differential global depth feature map is created by comparing the product to be detected with a genuine product to identify differences between them, capturing subtle variations and changes, including minor shape and texture differences. Therefore, in the technical solution of this application, the product comparison differential global depth feature map is used to determine whether the product is counterfeit. That is, the global depth feature map extracted by the deep learning model can represent the characteristics of the product holistically, including information on shape, texture, and color. These features reflect the uniqueness and authenticity of the product, thus helping to distinguish between genuine and counterfeit products. Secondly, through the deep learning model, rapid feature extraction and comparison differential analysis can be performed on a large number of products, thereby achieving efficient counterfeit product detection. Compared with traditional manual detection methods, this automated method is more efficient, accurate, and adaptable to large-scale product data.

[0043] Specifically, in this embodiment of the application, determining whether a product is counterfeit based on the product comparison differential global depth feature map includes: calculating the posterior latent feature attention of the aggregation node of the product comparison differential global depth feature map to obtain an optimized product comparison differential global depth feature map; Furthermore, the optimized product comparison differential global depth feature map is processed by a classifier to obtain a classification result, which is used to indicate whether the product is counterfeit.

[0044] Specifically, in the technical solution of this application, it is considered that if there are similarities between genuine product images in the database, the genuine product feature maps obtained by the convolutional neural network model may have a certain degree of duplicate information. This is because the convolutional neural network captures shared feature patterns such as texture, shape, and structure in the image when extracting features. If multiple genuine product images have similar textures or shapes, their corresponding feature representations in the feature maps may be similar or repetitive, leading to feature redundancy. At the same time, the product image denoising module based on the automatic encoder-decoder may introduce some redundant information during the denoising process. The goal of the denoising module is to restore the original product image, but in practice, the denoising module may retain some noise-related image details while removing noise. These details may affect the generation of the product contrast difference feature map, leading to feature redundancy. Furthermore, in the process of cascading multiple product local contrast difference depth feature maps, there may be a certain degree of redundancy between different local feature maps. The cascading operation merges the information of different local feature maps together, but there may be overlapping receptive fields between different local feature maps, that is, some features may appear repeatedly in different local feature maps. This repetition can lead to feature redundancy, as the same feature appears multiple times in the global feature map. In other words, the product comparison difference global depth feature map will have feature redundancy, which may increase the amount of redundant information in the features and reduce their discriminative and expressive power. To reduce feature redundancy, the technical solution of this application calculates the posterior latent feature attention of the aggregation nodes of the product comparison difference global depth feature map to obtain an optimized product comparison difference global depth feature map.

[0045] Specifically, in the embodiments of this application, calculating the posterior latent feature attention of the aggregation nodes of the product contrast difference global depth feature map to obtain an optimized product contrast difference global depth feature map includes: calculating the posterior latent feature attention of the aggregation nodes of the product contrast difference global depth feature map using the following optimization formula to obtain an optimized product contrast difference global depth feature map; wherein, the optimization formula is: in, This represents the product's contrast-difference global depth feature map. Location feature value This represents the value of the logarithmic function with base 2. This indicates dot product by position. This represents the optimized product contrast difference global depth feature map. Location feature value This indicates exponentiation.

[0046] Because the product comparison difference global depth feature map has feature redundancy, meaning it contains some unnecessary or repetitive feature information, the technical solution of this application calculates the posterior latent feature attention of the aggregation nodes of the product comparison difference global depth feature map to obtain an optimized product comparison difference global depth feature map. Thus, by calculating the attention posterior representation of the feature information at each position in the product comparison difference global depth feature map relative to the global feature distribution of the product comparison difference global depth feature map, the physicality of the feature map's self-expression is simulated. This reduces the dimensionality and complexity of the feature map by using point-by-point regression within the feature map, improving the internal structure of the optimized feature map, compressing a more compact feature representation, and thereby improving classification efficiency.

[0047] Specifically, in the embodiments of this application, the optimized product contrast difference global depth feature map is processed by a classifier to obtain a classification result, which is used to indicate whether the product is counterfeit. That is, the purpose of classifying the product contrast difference global depth feature map by a classifier is to divide the product into two categories: genuine and counterfeit, and generate a classification result to indicate whether the product is counterfeit. In detail, the classifier has the ability to distinguish between different categories. By training the classifier, it can learn the differences and characteristics between genuine and counterfeit products, thereby accurately classifying the product. The classifier can determine whether a product is genuine or counterfeit by analyzing and judging the product contrast difference global depth feature map. This improves the efficiency and accuracy of product supervision, helping consumers and regulatory authorities better address the problem of counterfeit products.

[0048] More specifically, in the embodiments of this application, the optimized product comparison difference global depth feature map is processed by a classifier to obtain a classification result, the classification result being used to indicate whether the product is counterfeit. This includes: expanding the optimized product comparison difference global depth feature map into an optimized product comparison difference global depth classification feature vector; using the fully connected layer of the classifier to fully connect and encode the optimized product comparison difference global depth classification feature vector to obtain an optimized product comparison difference global depth encoded classification feature vector; and passing the encoded classification feature vector through the Softmax classification function of the classifier to obtain the classification result.

[0049] In summary, the method for supervising and managing counterfeit goods based on the embodiments of this application is explained. It employs artificial intelligence technology based on a deep neural network model to acquire images of the product and images of genuine products in a database, extracts features from the images, and compares and analyzes these features with those from the database of genuine product images to determine whether the product is counterfeit. This method can assist in identifying counterfeit goods, thereby improving the efficiency and accuracy of product supervision and helping consumers and regulatory authorities better address the problem of counterfeit goods.

[0050] Figure 5 This is a block diagram of a counterfeit goods supervision and management system according to an embodiment of this application. Figure 5 As shown, the counterfeit goods supervision and management system 100 according to an embodiment of this application includes: a goods image acquisition module 110, used to acquire images of goods and images of genuine goods in a database; a goods image feature analysis module 120, used to perform image feature analysis on the images of the goods and the images of genuine goods in the database to obtain a goods image feature map and a genuine goods feature map respectively; a goods feature difference calculation module 130, used to calculate the difference between the goods image feature map and the genuine goods feature map to obtain a goods comparison difference feature map and segment it into multiple goods local comparison difference feature maps; a goods multi-scale fusion feature processing module 140, used to perform multi-scale fusion feature processing on the multiple goods local comparison difference feature maps to obtain a goods comparison difference global depth feature map; and a counterfeit goods identification module 150, used to determine whether the goods are counterfeit goods based on the goods comparison difference global depth feature map.

[0051] Here, those skilled in the art will understand that the specific operations of each step in the aforementioned counterfeit goods supervision and management system have been referenced above. Figures 1 to 4 The description of the methods for supervising and managing counterfeit goods is detailed here, and therefore, its repeated description will be omitted.

[0052] The above description is merely an example of the principles of this disclosure, and various modifications can be made by those skilled in the art without departing from the scope of this disclosure. The above embodiments are presented for illustrative purposes and not for limitation. This disclosure can also take many forms besides those expressly described herein. Therefore, it is intended to emphasize that this disclosure is not limited to the methods, systems, and apparatus expressly disclosed, but is intended to include variations and modifications made within the spirit and scope of the appended claims.

Claims

1. A method for supervising and managing counterfeit goods, characterized in that, include: Retrieve product images and images of genuine products from the database; Image feature analysis is performed on the image of the product and the image of genuine products in the database to obtain the product image feature map and the genuine product feature map, respectively. The difference between the product image feature map and the genuine product feature map is calculated to obtain a product contrast difference feature map, which is then segmented into multiple product local contrast difference feature maps. The multiple local contrast difference feature maps of the products are subjected to multi-scale fusion feature processing to obtain a global depth feature map of product contrast difference. Based on the product's comparative differential global depth feature map, it is determined whether the product is counterfeit.

2. The method for supervising and managing counterfeit goods according to claim 1, characterized in that, Image feature analysis is performed on the image of the product and the images of genuine products in the database to obtain product image feature maps and genuine product feature maps, respectively, including: Image feature extraction is performed on the image of the product to obtain the product image feature map; Images of genuine products from the database are input into a convolutional neural network model that acts as a filter to obtain feature maps of the genuine products.

3. The method for supervising and managing counterfeit goods according to claim 2, characterized in that, Image feature extraction is performed on the image of the product to obtain the product image feature map, including: The image of the product is subjected to image denoising to obtain a denoised image of the product; The image of the product after noise reduction is enhanced by spatial attention to obtain the feature map of the product image.

4. The method for supervising and managing counterfeit goods according to claim 3, characterized in that, To perform image denoising on the image of the product to obtain a denoised image of the product, the method includes: inputting the image of the product into a product image denoising module based on an automatic codec to obtain the denoised image of the product.

5. The method for supervising and managing counterfeit goods according to claim 4, characterized in that, The process of obtaining a product image feature map by enhancing the image of the denoised product through spatial attention includes: passing the image of the denoised product through a spatial attention mechanism based on a convolutional neural network to obtain the product image feature map.

6. The method for supervising and managing counterfeit goods according to claim 5, characterized in that, Calculating the difference between the product image feature map and the genuine product feature map to obtain a product comparison difference feature map and segmenting it into multiple product local comparison difference feature maps includes: calculating the product comparison difference feature map between the product image feature map and the genuine product feature map using the following difference formula; The difference formula is as follows: in, This represents the feature map of the product image. This indicates the difference based on position. This represents the feature image of the genuine product, and This represents the product comparison difference feature map.

7. The method for supervising and managing counterfeit goods according to claim 6, characterized in that, The multiple local contrast difference feature maps of the products are subjected to multi-scale fusion feature processing to obtain a global depth feature map of product contrast difference, including: The local contrast difference feature maps of the multiple products are processed through a deep-shallow fusion convolutional network model to obtain the local contrast difference depth feature maps of the multiple products. The multiple local contrast difference depth feature maps of the products are concatenated to obtain the global contrast difference depth feature map of the products.

8. The method for supervising and managing counterfeit goods according to claim 7, characterized in that, Based on the product's comparative differential global depth feature map, determine whether the product is counterfeit, including: Calculate the posterior latent feature attention of the aggregation node of the product contrast difference global deep feature map to obtain the optimized product contrast difference global deep feature map; The optimized product comparison differential global deep feature map is passed through a classifier to obtain a classification result, which is used to indicate whether the product is counterfeit.

9. The method for supervising and managing counterfeit goods according to claim 8, characterized in that, Calculating the posterior latent feature attention of the aggregation nodes of the product contrast difference global depth feature map to obtain an optimized product contrast difference global depth feature map includes: calculating the posterior latent feature attention of the aggregation nodes of the product contrast difference global depth feature map using the following optimization formula to obtain an optimized product contrast difference global depth feature map. The optimization formula is as follows: in, This represents the product's contrast-difference global depth feature map. Location feature value This represents the value of the logarithmic function with base 2. This indicates dot product by position. This represents the optimized product contrast difference global depth feature map. Location feature value This indicates exponentiation.

10. A counterfeit goods supervision and management system, characterized in that, include: The product image acquisition module is used to acquire images of products and pictures of genuine products in the database; The product image feature analysis module is used to perform image feature analysis on the image of the product and the image of genuine products in the database to obtain product image feature map and genuine product feature map respectively; The product feature difference calculation module is used to calculate the difference between the product image feature map and the genuine product feature map to obtain the product comparison difference feature map and segment it into multiple product local comparison difference feature maps; The multi-scale fusion feature processing module for commodities is used to perform multi-scale fusion feature processing on the multiple local contrast difference feature maps of commodities to obtain a global depth feature map of commodity contrast difference. The counterfeit product identification module is used to determine whether a product is counterfeit based on the product's comparative differential global depth feature map.

Images