Commodity identification method, program product, electronic device and storage medium
Patent Information
- Application Number
- PCT/CN2025/130669
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2025-03-18
- Filing Date
- 2025-10-28
- Publication Date
- 2026-09-24
Smart Images
Figure CN2025130669_24092026_PF_FP_ABST
Abstract
Description
A product identification method, a program product, an electronic device, and a storage medium
[0001] Cross-references to related applications
[0002] This application claims priority to Chinese patent application CN202510321729.7, filed on March 18, 2025, entitled “A method for identifying a commodity, a program product, an electronic device and a storage medium”, the entire contents of which are incorporated herein by reference. Technical Field
[0003] This application relates to the field of commodity identification, and more specifically, to a commodity identification method, a program product, an electronic device, and a storage medium. Background Technology
[0004] Traditional product identification methods rely on manual operation, where individuals inspect each identification point individually to obtain the identification result. Due to the large number of inspection items, lengthy process, and variations in personnel's adherence to standards, discrepancies in the results can occur. Summary of the Invention
[0005] The purpose of this application is to provide a product identification method, program product, electronic device, and storage medium to improve the above-mentioned problems.
[0006] In a first aspect, embodiments of this application provide a product identification method, comprising: acquiring a product image and a product label image of the product to be identified; extracting image features from the product image to obtain image features; obtaining a genuine product score based on the image features; the genuine product score representing the probability that the product is genuine; extracting product attribute information based on the product label image; and identifying the product to be identified based on the genuine product score and the product attribute information to obtain an identification result.
[0007] In the above implementation process, by combining the image features of the product image and the product attribute information of the product label image, the product is analyzed and judged from the dimensions of image features and product attribute information, which can provide a more comprehensive understanding of the product; and the authenticity score provides a quantitative basis for the identification result; the product attribute information refines the characteristics of the product. The identification result obtained based on the authenticity score and product attribute information is more accurate and interpretable.
[0008] Optionally, in this embodiment of the application, the method further includes: obtaining additional product information of the product to be identified; wherein, the additional product information is information about the product to be identified other than product attribute information; extracting text features from the additional product information to obtain additional product features; and identifying the product to be identified based on the authenticity score and product attribute information to obtain an identification result, including: identifying the product to be identified based on the authenticity score, product attribute information, and additional product features to obtain an identification result.
[0009] In the above implementation process, the additional product information includes content not found in the product image and product label image, complementing the authenticity score and product attribute information, thus helping to more accurately determine the authenticity of the product to be identified. Furthermore, the source of the additional product information is not limited to the product image and product label image, enabling identification based on multiple information sources, improving the accuracy of identification, and providing a more definitive basis for the identification results.
[0010] Optionally, in this embodiment of the application, the product image includes a product foreground image; obtaining the product image of the product to be identified includes: obtaining multiple product foreground images corresponding to multiple shooting angles of the product to be identified.
[0011] In the above implementation process, by acquiring multiple foreground images of the product from multiple shooting angles, the interference of background factors in the product images on the identification is reduced. Furthermore, during the identification process, the foreground images of the product from multiple angles can be analyzed to improve the accuracy of identification.
[0012] Optionally, in this embodiment of the application, obtaining the genuine product score based on image features includes: calculating the distance between the image features and the genuine product classification center and the counterfeit product classification center corresponding to the identification part of the product image, respectively, to obtain distance data; the genuine product classification center is the genuine product feature of at least one product unit in the identification part; the counterfeit product classification center is the counterfeit product feature of at least one product unit in the identification part; wherein, a product unit represents a specification corresponding to the product; and the genuine product score of the product image is determined based on the distance data.
[0013] In the above implementation process, the genuine product classification center and the counterfeit product classification center of the identification area are associated with the product unit. Different classification centers can represent the characteristics of different product units. By calculating the distance between image features and at least one classification center, the refinement and accuracy of identification are improved. Furthermore, the distance data reflects the similarity between the product image features and the genuine product classification center and the counterfeit product classification center. The distance data is converted into a genuine product score for the product image, providing a quantitative basis for identification results.
[0014] Optionally, in this embodiment of the application, the method further includes: determining the target classification center that is closest to the image features based on the minimum value in the distance data; obtaining product unit information of the target classification center; the product unit information is used to characterize the specification attribute information corresponding to the product to be identified; performing consistency verification between the product unit information and the product attribute information to obtain a consistency verification result; and identifying the product to be identified based on the genuine product score and the product attribute information to obtain an identification result, including: identifying the product to be identified based on the genuine product score, the product attribute information, and the consistency verification result to obtain an identification result.
[0015] In the above implementation process, image features can reflect intuitive information such as the appearance and texture of the product, and product attribute information provides a basic description of the product in the product label image. By verifying the consistency between product unit information and product attribute information, the reliability of product attribute information is verified. The consistency verification result is considered in the identification result, which improves the accuracy of identification.
[0016] Optionally, in this embodiment of the application, the identification of the product to be identified based on the authenticity score and product attribute information, and the identification result obtained, includes: inputting the authenticity score and product attribute information into a large language model to obtain the identification result.
[0017] In the above implementation process, the authenticity score provides the probability that a product is authentic, while the product attribute information provides a basic description of the product. The large language model can comprehensively consider multiple pieces of information, enabling identification analysis from multiple perspectives, reducing the possibility of misjudgment and improving the accuracy of identification. Furthermore, the large language model has powerful reasoning capabilities, producing more natural output results. It can automatically process the input authenticity score and product attribute information, reducing the workload of manual intervention and processing, and improving the efficiency of identification.
[0018] Optionally, in this embodiment of the application, the identification result includes the authenticity of the product to be identified and the reason for determining authenticity. Before inputting the genuine product score and product attribute information into the large language model to obtain the identification result, the method further includes: obtaining a dataset, the dataset including sample genuine product scores, sample product attribute information, and labeled data corresponding to the sample genuine product scores and sample product attribute information, the labeled data including genuine / fake labels and the reason for determining authenticity; and using a pre-trained large language model to learn knowledge of identifying the authenticity of products and attribution based on the dataset.
[0019] In the above implementation process, a pre-trained large language model is used to learn knowledge of distinguishing genuine and counterfeit goods and attribution based on the dataset. The processed large language model can better adapt to the product identification task and improve the accuracy of product identification. Furthermore, since the pre-trained large language model has learned knowledge of distinguishing genuine and counterfeit goods and attribution based on the dataset, the output identification results include the genuine / counterfeit category of the product and the reason for counterfeiting, helping users to better understand the identification results.
[0020] Secondly, embodiments of this application also provide a product identification device, comprising: an image acquisition module for acquiring a product image and a product label image of the product to be identified; a feature extraction module for extracting image features from the product image to obtain image features; a genuine product score module for obtaining a genuine product score based on the image features; the genuine product score represents the probability that the product is genuine; an attribute information module for extracting product attribute information based on the product label image; and an identification module for identifying the product to be identified based on the genuine product score and the product attribute information to obtain an identification result.
[0021] Thirdly, embodiments of this application also provide a computer program product, including computer program instructions, which are executed by a processor to perform the method provided in the first aspect or any implementation thereof.
[0022] Fourthly, embodiments of this application also provide an electronic device, including: a processor and a memory, the memory storing computer program instructions, which are executed by the processor to perform the method provided in the first aspect or any implementation thereof.
[0023] Fifthly, embodiments of this application also provide a computer-readable storage medium storing computer program instructions, which, when executed by a processor, perform the method provided in the first aspect or any implementation thereof.
[0024] By employing the product identification method, program product, electronic device, and storage medium provided in this application, and combining the image features of the product image with the product attribute information of the product label image, the product can be analyzed and judged from both the image feature dimension and the product attribute information dimension, thus providing a more comprehensive understanding of the product. Furthermore, the authenticity score provides a quantitative basis for the identification result, while the product attribute information refines the characteristics of the product. The identification result obtained based on the authenticity score and product attribute information has higher accuracy and interpretability. Attached Figure Description
[0025] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.
[0026] Figure 1 is a flowchart illustrating a product identification method provided in an embodiment of this application;
[0027] Figure 2 is a schematic diagram of the product identification device provided in an embodiment of this application;
[0028] Figure 3 is a schematic diagram of the structure of the electronic device provided in the embodiment of this application. Detailed Implementation
[0029] The embodiments of the technical solution of this application will now be described in detail with reference to the accompanying drawings. These embodiments are only used to more clearly illustrate the technical solution of this application and are therefore merely examples, and should not be used to limit the scope of protection of this application.
[0030] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs; the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit this application.
[0031] In the description of the embodiments of this application, technical terms such as "first" and "second" are used only to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly defined.
[0032] Please refer to Figure 1, which shows a flowchart illustrating a product identification method provided in an embodiment of this application. The product identification method provided in this application can be applied to electronic devices, which may include physical devices such as servers, PCs, tablets, or smartphones, or virtual devices such as virtual machines or containers. The electronic device can be a single device, a combination of multiple devices, or a cluster of a large number of devices. The product identification method may include:
[0033] Step S110: Obtain the product image and product label image of the product to be identified.
[0034] Step S120: Extract image features from the product image to obtain image features.
[0035] Step S130: Obtain the authenticity score based on image features; the authenticity score represents the probability that the product is authentic.
[0036] Step S140: Extract product attribute information based on the product label image.
[0037] Step S150: Based on the genuine product score and product attribute information, identify the product to be identified and obtain the identification result.
[0038] In step S110, the product to be identified is the product whose authenticity needs to be verified. The product image refers to the actual image of the product to be identified; the product label image is an image containing product information such as labels, hang tags, and nameplates on the product to be identified. Both the product image and the product label image can be obtained by capturing the product to be identified using a camera device, or they can be retrieved from a database or other storage system.
[0039] In one scenario, the product image includes the product label, and the product image and the product label image can be considered as one entity; that is, the image can serve as both a product image and a product label image.
[0040] In step S120, the image features are feature vectors extracted from the product image, which may include information such as color, texture, shape, or edges. Image feature extraction methods include: using an encoder to convert the product image into lower-dimensional image features; or using a pre-trained feature extraction model to process the product image to obtain image features.
[0041] In step S130, the authenticity score can be a numerical value representing the probability that the product is genuine. For example, the larger the value, the greater the probability that the product to be identified is genuine. Alternatively, the authenticity score can be a combination of a label and a numerical value, such as "Genuine 0.9" or "Fake 0.8," where "Genuine 0.9" represents a 90% probability that the product to be identified is genuine, and "Fake 0.8" represents an 80% probability that the product to be identified is fake (i.e., the probability of the product being fake is higher, and the probability of it being genuine is lower). This method requires combining the label and the numerical value to determine the probability that the product is genuine, rather than only considering the magnitude of the numerical value.
[0042] One way to obtain a product authenticity score is to use a deep learning model (such as a convolutional neural network, CNN) to classify image features and output the probability that the product is authentic. For example, a large number of images of authentic and counterfeit products can be collected, and image features can be extracted. The extracted sample image features are labeled with authentic product scores. The deep learning model is then trained using these sample image features to obtain a trained authentic product score model. This trained model is then used to predict the authentic product score based on the image features. This method can automatically learn the complex relationship between image features and the probability of authenticity, offering high efficiency and convenience.
[0043] It can also calculate the distance between the features of the product image to be identified and the classification centers of genuine and counterfeit products, and convert the distance into a genuine product score. For example, multiple features can be extracted from multiple known genuine product images, and a center vector can be calculated using these features as the genuine product classification center; similarly, multiple features can be extracted from multiple counterfeit product images, and a center vector can be calculated using these features as the counterfeit product classification center. Then, the distances between the feature images and the genuine and counterfeit product classification centers are calculated, and the distance values are converted into genuine product scores. This method is simple to calculate and saves computational resources.
[0044] Of course, one can also use probability distribution calculation models to obtain the authenticity score corresponding to image features. For example, the image features of the product to be identified can be input into an estimated probability density model, and the probability that it belongs to the authenticity distribution can be calculated as the authenticity score. This method can capture the distribution characteristics of features and is applicable even for cases with complex feature distributions, with a high accuracy rate for authenticity scores.
[0045] In step S140, product attribute information includes information such as the product's origin, item number, size, color, material, brand, and / or model. This product attribute information can be extracted from the product label image using Optical Character Recognition (OCR) technology, or a deep learning-based text detection model can be used to detect and recognize text regions in the product label image to obtain the product attribute information.
[0046] In step S150, the identification result is used to characterize the authenticity category label, the probability of authenticity (counterfeit), and / or the reasons for authenticity (for genuine products) or counterfeit (for counterfeit products) of the product to be identified. The authenticity score and product attribute information can be input into a large language model. For example, the large language model can identify the product to be identified and obtain the identification result. The large language model can pre-learn knowledge about identifying the authenticity of products and attribution.
[0047] Other machine learning models, such as logistic regression, support vector machines, and random forests, can be pre-trained and used to predict new products to be identified. By using the authenticity score and product attribute information as input features, the model will output the probability or category label that the product is authentic.
[0048] Of course, authentication results can also be based on preset rules. For example, if a shoe was released a long time ago and is showing significant wear, the required score for authenticity can be appropriately lowered. The rules can be determined based on the actual situation.
[0049] In the implementation of the above embodiments: by combining the image features of the product image and the product attribute information of the product label image, the product is analyzed and judged from the dimensions of image features and product attribute information, which can provide a more comprehensive understanding of the product; and the authenticity score provides a quantitative basis for the identification result; the product attribute information refines the characteristics of the product, and the identification result obtained based on the authenticity score and product attribute information has higher accuracy and interpretability.
[0050] Optionally, in this embodiment of the application, the product to be identified is determined based on the authenticity score and product attribute information to obtain the identification result, including:
[0051] Obtain additional information about the product to be identified; this additional information includes details about the product other than its attributes. This additional information can be derived from order information or records of the product in the inventory system; it includes details such as order date, weight, color, country of origin, and size.
[0052] It should be noted that product supplementary information and product attribute information can contain the same content, but their sources are different. For example, product attribute information includes the product's color, which is obtained through label image recognition; while product supplementary information also includes the product's color, which is obtained through system records (such as inventory records or user reviews). Of course, product supplementary information can also include content not included in product attribute information.
[0053] In an optional embodiment, consistency checks can be performed on common information in product attribute information and product additional information, and the check results can be used for product identification.
[0054] Textual feature extraction is performed on product information to obtain additional product features. For example, natural language processing techniques can be used to segment text in product order information or inventory system records and extract keywords, phrases, etc., as additional product features. Alternatively, the text can be input into a pre-trained machine learning model, which will automatically learn and extract additional product features.
[0055] Based on the authenticity score, product attribute information, and additional product features, the product to be identified is determined to obtain an identification result. Similar to step S150, a large language model or other pre-trained models can be used to predict new products to be identified to obtain an identification result. Alternatively, a weighted scoring method can be used to obtain the identification result.
[0056] In the implementation of the above embodiments: the product supplementary information includes content not included in the product image and product label image, complementing the authenticity score and product attribute information, and helping to more accurately determine the authenticity of the product to be identified. Furthermore, the source of the product supplementary information is not the product image and product label image, enabling identification based on multiple information sources, improving the accuracy of identification, and providing a more definitive basis for the identification results.
[0057] Optionally, in this embodiment of the application, the product image includes a product foreground image; obtaining the product image of the product to be identified includes: obtaining multiple product foreground images corresponding to multiple shooting angles of the product to be identified.
[0058] For example, multiple cameras, either fixed or non-fixed, can be used to photograph the product to be identified from different angles. Each camera is responsible for capturing images of the product from different angles, such as front, back, left, right, top, and bottom, thus obtaining multiple product images. Then, an image segmentation algorithm is used to process the product images, separating the product from the background to obtain the corresponding foreground image.
[0059] Optionally, the shooting light source can adopt a symmetrical design, with multiple light sources installed on the upper and lower parts of the product respectively, so that the internal light source provides sufficiently uniform brightness. Diffuse light can also be used to supplement the internal lighting of the device, reducing interference from external ambient light and improving the stability of imaging lighting.
[0060] In one implementation, the product label image includes a product label foreground image; acquiring the product label image of the product to be identified includes: acquiring multiple product label foreground images corresponding to multiple shooting angles of the product to be identified. The acquisition method can be found in the method for acquiring product foreground images.
[0061] After acquiring multiple product foreground images, the product to be identified can be determined based on the authenticity score and product attribute information corresponding to each of the multiple product foreground images, thus obtaining the result of the product to be identified.
[0062] In the implementation of the above embodiments: by acquiring multiple foreground images of the product from multiple shooting angles, the interference of background factors in the product images on the identification is reduced, and the product foreground images from multiple angles can be combined for analysis during the identification process to improve the accuracy of identification.
[0063] Optionally, in this embodiment of the application, obtaining the authenticity score based on image features includes:
[0064] Based on the identification area corresponding to the product image, the distance between the image features and the genuine product classification center and the counterfeit product classification center corresponding to the identification area is calculated to obtain distance data.
[0065] Each product image includes a corresponding identification area, which is related to the angle of the product image. For example, if the product image is taken from behind the product, the identification area is the area behind the product. It can be understood that the number of identification areas corresponds to the number of product images taken from different angles.
[0066] Each identification area includes a corresponding authentic product category center and a counterfeit product category center. Each identification area may have one or more of these categories. An authentic product category center represents at least one product unit with authentic characteristics at the identification area; a counterfeit product category center represents at least one product unit with counterfeit characteristics at the identification area. Product units are used to identify the specifications of the product, including size, color, or material, etc. For example, the same bag may come in small, medium, and large sizes; the same shoes or clothing may come in different sizes and colors; and the same shoes may have a glossy or matte finish, etc.
[0067] When there are multiple authentic product classification centers for the identification area, then multiple product units of the product have corresponding authentic product classification centers. For example, the product is a backpack, and the identification area is the back of the bag. The backpack comes in three sizes: small, medium, and large. The authentic product features of the back area of the small-sized bag are the authentic product classification center corresponding to the small size (denoted as SKU1); the authentic product features of the back area of the medium-sized bag are the authentic product classification center corresponding to the medium size (denoted as SKU2); and the authentic product features of the back area of the large-sized bag are the authentic product classification center corresponding to the large size (denoted as SKU3). For example, the authentic product features of the back area of the small-sized bag can be obtained as follows: multiple sample authentic product images of the back area of the small-sized bag are pre-collected; image features are extracted from each of the above multiple sample authentic product images of the small-sized bag to obtain multiple authentic product sample image features; the authentic product features are obtained from the multiple authentic product sample image features, and the obtained authentic product features are used as the authentic product classification center (SKU1) corresponding to the small size.
[0068] Similarly, the counterfeit product category center includes three categories corresponding to different sizes. It's understandable that these categories can include their corresponding product unit information; for example, the authentic product category center SKU1 for the smaller size has the product unit information for the smaller size.
[0069] During the distance calculation process, the image features of the product to be identified need to be calculated separately with the three genuine product classification centers and the three counterfeit product classification centers to obtain distance data. The distance data includes the distance value between the image feature and each genuine product classification center and counterfeit product classification center. The distance value is used to characterize the difference between the image feature and the genuine product classification center or the counterfeit product classification center. The smaller the distance value, the smaller the difference.
[0070] Multiple product units each have corresponding authentic product classification centers. This is advantageous because different sizes of goods may differ in details; for example, the logos on small-sized bags and large-sized bags may differ, making it difficult to find a single authentic product classification center that simultaneously represents the authentic characteristics of different product sizes. Therefore, this application's embodiments associate classification centers (a collective term for authentic and counterfeit product classification centers) with product units. A classification center represents the authentic (or counterfeit) characteristics of a single product unit at the identification point, rather than the authentic (or counterfeit) characteristics of all product sizes at the identification point, thus improving the precision of identification.
[0071] Furthermore, in some authentication scenarios, if the product to be authenticated is a small-sized item, but its logo is identified as being identical to that of a large-sized backpack, this product is clearly problematic (e.g., the logo has been swapped). If the authentication process does not consider size differences, and the product to be authenticated matches any size of genuine product, thus being identified as genuine, then authentication errors may occur in such cases. Therefore, in this embodiment, the differences between different product sizes are fully considered during the authentication process. Even if the image recognition shows a match with a genuine product (e.g., the distance value is less than a threshold), consistency verification can still be performed using the corresponding product unit information, thereby improving the accuracy of authentication. This solution will be explained later.
[0072] If there is a genuine product classification center at the identification point, the product may have a single product unit, meaning it doesn't differentiate by size. In this case, a genuine product classification center can be calculated using the genuine product characteristics at the identification point. The same logic applies to counterfeit product classification centers. Alternatively, if the product has multiple sizes, but the differences between sizes are negligible at the identification point, then multiple product units at that identification point can be used to calculate a single genuine product classification center. During distance calculation, only the image data needs to be compared with one genuine product classification center and one counterfeit product classification center. The number of classification centers can be determined based on the characteristics of the product.
[0073] After obtaining the distance data, the authenticity score of the product image is determined based on the distance data. The numerical and label components of the authenticity score can be determined separately based on the distance data, and the authenticity score is obtained by combining the numerical and label components.
[0074] For example, based on the minimum value in the distance data, a target classification center that is closest to the image features can be determined. This target classification center can be either a genuine product classification center or a counterfeit product classification center. The label for the genuine product classification center can be "genuine," and the label for the counterfeit product classification center can be "counterfeit." The minimum value is mapped to a score interval (e.g., [0,1]), and the mapping method can include thresholding, linear mapping, or non-linear mapping. For instance, suppose the target classification center closest to the minimum value is the counterfeit product classification center for small sizes, with a minimum value of 0.1. Using a thresholding method, the distance value is converted into a genuine product score. Assuming the threshold is 0.5, if the distance value is less than the threshold, the numerical part of the genuine product score is 1; otherwise, it is 0. Since the minimum value is less than the threshold, the numerical part of the genuine product score is 1. And because the target classification center closest to the minimum value is the counterfeit product classification center for small sizes, labeled "counterfeit," the label part of the genuine product score is "counterfeit." The final genuine product score can be "counterfeit 1," representing that the product to be identified has a high probability of being counterfeit.
[0075] Of course, you can also use a plus sign "+" or a minus sign "-" or other identifiers as the label part; the above are just examples.
[0076] In the implementation of the above embodiments: Authentic product classification centers and counterfeit product classification centers are associated with product units. Different classification centers can represent the characteristics of different product units. By calculating the distance between image features and classification centers, the precision and accuracy of identification are improved. Furthermore, the distance data reflects the similarity between the product image features and the authentic and counterfeit product classification centers. Converting the distance data into an authentic product score for the product image provides a clear basis for the identification results.
[0077] In an optional embodiment, before obtaining the distance data, the genuine product classification center and the counterfeit product classification center corresponding to the identification part can be identified first. The method further includes:
[0078] Step 1: Obtain multiple genuine product images and multiple counterfeit product images of at least one product unit at the identification area. For example, you can obtain multiple genuine product images and multiple counterfeit product images of the rear area of a small-sized backpack; and multiple genuine product images and multiple counterfeit product images of the rear areas of medium-sized and large-sized backpacks.
[0079] Step 2: Extract image features from multiple genuine product images and multiple counterfeit product images respectively, to obtain multiple features for genuine product images and multiple features for counterfeit product images. The feature extraction method is the same as described above.
[0080] Step 3: Process the features of multiple genuine product images for the same product unit to obtain the genuine product classification center for that product unit. Similarly, process the features of multiple counterfeit product images for the same product unit to obtain the counterfeit product classification center for that product unit. Processing methods include feature clustering, taking the mean, or taking the median, etc.
[0081] For example, clustering the features of multiple authentic images of the back area of a small-sized backpack yields an authentic product category center; similarly, clustering the features of multiple counterfeit images of the back area of a small-sized backpack yields a counterfeit product category center. The same principle applies to medium-sized and large-sized backpacks, allowing for the acquisition of corresponding authentic and counterfeit product category centers.
[0082] Thus, for each of the three backpack sizes, corresponding authentic and counterfeit classification centers are obtained in the rear area as an identification dimension. During identification, the image features of the image to be identified are compared with the authentic and counterfeit classification centers of the three sizes to calculate the distance data and obtain the corresponding authentic score.
[0083] Optionally, in this embodiment, the method further includes: determining the target classification center closest to the image feature based on the minimum value in the distance data. Following the above embodiment, the image feature is used to calculate the distance between the genuine product classification center and the counterfeit product classification center for the three sizes, respectively, to obtain six distance values. The minimum value is the distance to the genuine product classification center of the small-sized backpack, and the genuine product classification center of the small-sized backpack is then determined as the target classification center closest to the image feature.
[0084] Obtain the product unit information from the target category center; the product unit information is used to represent the specification attributes of the product to be identified. For example, if the target category center is the genuine product category center for small-sized backpacks, then the product unit information can be "small-sized backpacks".
[0085] The consistency verification between product unit information and product attribute information is performed to obtain the verification result. Product attribute information is extracted from the product label image, while product unit information is based on image features, i.e., it comes from the product image. By verifying the consistency between the specification information in the product attribute information and the product unit information, it is possible to identify whether the product unit information in the product image (i.e., the product itself) is consistent with the information in the product label image, thereby achieving the verification effect and improving the accuracy of identification.
[0086] Following the above embodiment, if the product unit information is a small backpack and the information in the product attribute information is also a small backpack, then the consistency check result is consistent; if the information in the product attribute information is a large backpack, then the consistency check result is inconsistent.
[0087] Based on the authenticity score and product attribute information, the product to be identified is processed to obtain the identification results, including:
[0088] Based on the authenticity score, product attribute information, and consistency verification results, the product to be authenticated is identified to obtain the authentication result. During the identification process, the consistency verification result also needs to be considered. For example, if the consistency verification result shows consistent information, the authenticity score and product attribute information can be used for identification according to the original calculation method to obtain the authentication result; if the consistency verification result shows inconsistent information, the authentication result can be directly determined as counterfeit.
[0089] For example, a large language model can be used to identify the product to be authenticated and obtain the authentication result. The large language model can then learn, based on a pre-set dataset, the knowledge to distinguish between genuine and counterfeit products according to the authenticity score, product attribute information, and consistency verification results. Afterwards, the authenticity score, product attribute information, and consistency verification results are input into the learned large language model to obtain the authentication result.
[0090] In the implementation of the above embodiments: image features can reflect intuitive information such as the appearance and texture of the product, and product attribute information provides a basic description of the product in the product label image. By verifying the consistency between product unit information and product attribute information, the reliability of product attribute information is verified, providing a reliable basis for product identification.
[0091] Optionally, in this embodiment of the application, the identification of the product to be identified based on the authenticity score and product attribute information, and the identification result obtained, includes: inputting the authenticity score and product attribute information into a large language model to obtain the identification result.
[0092] The authenticity score and product attribute information can be converted into a token sequence. This token sequence is then input into a large language model to obtain the identification result. For example, the authenticity score can be converted into a string format, the product attribute information can be segmented and encoded, and the processed authenticity score and product attribute information can be concatenated into a token sequence. This generated token sequence is then input into a large language model, such as GPT or the Tongyi Qianwen model. The large language model will generate an identification result based on the input token sequence.
[0093] Optionally, in this embodiment of the application, the identification of the product to be identified is based on the authenticity score, product attribute information and additional product features, and the identification result is obtained, including: inputting the authenticity score, product attribute information and additional product features into a large language model to obtain the identification result.
[0094] Similarly, the authenticity score, product attribute information, and additional product features can be converted into a token sequence, and the token sequence can be input into a large language model to obtain the identification result.
[0095] In the implementation of the above embodiments: the authenticity score provides the probability that a product is authentic, while the product attribute information provides a basic description of the product. The large language model can comprehensively consider multiple pieces of information, enabling identification analysis from multiple perspectives, reducing the possibility of misjudgment, and improving the accuracy of identification. Furthermore, the large language model can automatically process the input authenticity score and product attribute information, reducing the workload of manual intervention and processing, and improving the efficiency of identification.
[0096] Optionally, in this embodiment of the application, before inputting the authenticity score and product attribute information into the large language model to obtain the identification result, the method further includes:
[0097] Obtain the dataset, which includes the sample genuine product score, sample product attribute information, and the labeled data corresponding to the sample genuine product score and sample product attribute information. The labeled data includes genuine product labels and counterfeit product labels and reasons.
[0098] In an optional embodiment, the dataset may also include additional features of the sample products, in which case the labeled data consists of the sample genuine product score, sample product attribute information, and labeled data corresponding to the additional features of the products.
[0099] For example, image features can be extracted from known genuine and counterfeit products, and the distances between these features and the classification centers of genuine and counterfeit products can be calculated to obtain the genuine product score of the sample (see the process for obtaining the genuine product score). Product attribute information can also be extracted from the label images of known genuine and counterfeit products. Data preprocessing can also be performed on the collected datasets, including data cleaning, data augmentation, or format conversion.
[0100] If the true value of the combination of the sample genuine product score and the sample product attribute information is genuine, then the labeling data can be a genuine product label. If the true value of the combination of the sample genuine product score and the sample product attribute information is counterfeit, then the labeling data includes not only the counterfeit product label, but also the reason why it is counterfeit. Reasons for counterfeit products include non-original accessories, counterfeit accessories, questionable place of origin, or incompatible products.
[0101] This study utilizes a pre-trained large language model to learn knowledge for identifying product authenticity and attribution based on a dataset. The pre-trained large language model, trained on massive text data, possesses excellent language understanding and generation capabilities. Building upon this foundation, the model can be fine-tuned using the dataset to optimize its parameters, enabling it to better adapt to the product identification task and learn knowledge for distinguishing product authenticity and attribution. During fine-tuning, the cross-entropy loss function can be used to optimize the model's performance.
[0102] In the implementation of the above embodiments: a pre-trained large language model is used to learn knowledge of identifying genuine and counterfeit goods and attribution based on the dataset. The processed large language model can better adapt to the product identification task and improve the accuracy of product identification. Furthermore, since the pre-trained large language model has learned knowledge of identifying genuine and counterfeit goods and attribution based on the dataset, the output identification results include the genuine / counterfeit category of the product and the reason for counterfeiting, providing clearer identification criteria and helping users better understand the identification results.
[0103] Please refer to Figure 2 for a schematic diagram of the product identification device provided in this application embodiment; this application embodiment provides a product identification device 200, including:
[0104] Image acquisition module 210 is used to acquire product images and product label images of the product to be identified;
[0105] Feature extraction module 220 is used to extract image features from product images to obtain image features;
[0106] The Authenticity Score Module 230 is used to obtain the Authenticity Score based on image features; the Authenticity Score represents the probability that a product is authentic.
[0107] Attribute information module 240 is used to extract product attribute information based on product label images;
[0108] The identification module 250 is used to identify the product to be identified based on the authenticity score and product attribute information, and obtain the identification result.
[0109] Optionally, in this embodiment of the application, the product identification device 200 and the identification module 250 are further used to obtain additional product information of the product to be identified; wherein, the additional product information is information about the product to be identified other than product attribute information; text feature extraction is performed on the additional product information to obtain additional product features; based on the genuine product score, product attribute information and additional product features, the product to be identified is identified to obtain the identification result.
[0110] Optionally, in this embodiment of the application, the product identification device 200 includes a product foreground image; the image acquisition module 210 is further configured to acquire multiple product foreground images corresponding to multiple shooting angles of the product to be identified.
[0111] Optionally, in this embodiment of the application, the product identification device 200 and the genuine product score module 230 are further configured to calculate the distance between the image features and the genuine product classification center and the counterfeit product classification center corresponding to the identification part of the product image, respectively, to obtain distance data; the genuine product classification center is the genuine product feature of at least one product unit in the identification part; the counterfeit product classification center is the counterfeit product feature of at least one product unit in the identification part; and the genuine product score of the product image is determined based on the distance data.
[0112] Optionally, in this embodiment, the product identification device 200 further includes: a consistency verification module, used to determine the target classification center closest to the image features based on the minimum value in the distance data; obtain product unit information of the target classification center; the product unit information is used to characterize the specification attribute information corresponding to the product to be identified; perform consistency verification between the product unit information and the product attribute information to obtain a consistency verification result; and identify the product to be identified based on the genuine product score and the product attribute information to obtain an identification result, including: identifying the product to be identified based on the genuine product score, the product attribute information, and the consistency verification result to obtain an identification result.
[0113] Optionally, in this embodiment of the application, the product identification device 200 and the identification module 250 are further used to input the genuine product score and product attribute information into the large language model to obtain the identification result.
[0114] Optionally, in this embodiment of the application, the product identification device 200 further includes: a large language model attribution knowledge learning module, used to acquire a dataset, the dataset including sample genuine product scores, sample product attribute information, and labeled data corresponding to the sample genuine product scores and sample product attribute information, the labeled data including genuine product labels and counterfeit product labels and reasons; and using a pre-trained large language model to learn knowledge of identifying genuine and counterfeit products and attribution based on the dataset.
[0115] It should be understood that this device corresponds to the above-described product identification method embodiment and is capable of performing the various steps involved in the above method embodiment. The specific functions of this device can be found in the description above, and detailed descriptions are omitted here to avoid repetition. The device includes at least one software functional module that can be stored in memory or embedded in the device's operating system (OS) in the form of software or firmware.
[0116] Please refer to Figure 3, which shows a schematic diagram of the structure of an electronic device provided in an embodiment of this application. An electronic device 300 provided in this application includes a processor 310 and a memory 320. The memory 320 stores machine-readable instructions executable by the processor 310. When the machine-readable instructions are executed by the processor 310, the method described above is performed.
[0117] The components shown in Figure 3 can be implemented using hardware, software, or a combination thereof. Electronic device 300 may be a physical device, such as a server or PC, or a virtual device, such as a virtual machine or virtualized container. Furthermore, electronic device 300 is not limited to a single device; it can also be a combination of multiple devices or a cluster of numerous devices.
[0118] This application also provides a storage medium storing a computer program, which is executed by a processor to perform the above-described method.
[0119] The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.
[0120] This application also provides a computer program product, including computer program instructions, which are executed by a processor to perform the method described above.
[0121] It should be understood that the disclosed apparatus and methods can also be implemented in other ways, as provided in the embodiments of this application. The apparatus embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, or they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.
[0122] In addition, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.
[0123] The above description is only an optional implementation of the embodiments of this application, but the protection scope of the embodiments of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the embodiments of this application should be covered within the protection scope of the embodiments of this application.
Claims
1. A method for identifying commodities, characterized in that, include: Obtain the product image and product label image of the product to be identified; Image features are extracted from the product image to obtain image features; Based on the image features, a genuine product score is obtained; the genuine product score represents the probability that the product is genuine. Extract product attribute information based on the product label image; Based on the authenticity score and the product attribute information, the product to be identified is identified, and the identification result is obtained.
2. The method according to claim 1, characterized in that, The method further includes: Obtain the additional product information of the product to be identified; wherein, the additional product information is information about the product to be identified other than the product attribute information; Text feature extraction is performed on the product's additional information to obtain product additional features; The process of identifying the product to be authenticated based on the authenticity score and the product attribute information, and obtaining the authentication result, includes: Based on the authenticity score, the product attribute information, and the product additional features, the product to be identified is identified, and the identification result is obtained.
3. The method according to claim 1, characterized in that, The product image includes a product foreground image; acquiring the product image of the product to be identified includes: Acquire multiple foreground images of the product from multiple shooting angles corresponding to the product to be identified.
4. The method according to claim 1, characterized in that, The step of obtaining the authenticity score based on the image features includes: Based on the identification area corresponding to the product image, the distance between the image features and the genuine product classification center and the counterfeit product classification center corresponding to the identification area is calculated to obtain distance data; the genuine product classification center is the genuine product feature of at least one product unit in the identification area; the counterfeit product classification center is the counterfeit product feature of at least one product unit in the identification area; wherein, a product unit represents a specification corresponding to the product; Based on the distance data, the authenticity score of the product image is determined.
5. The method according to claim 4, characterized in that, The method further includes: Based on the minimum value in the distance data, determine the target classification center that is closest to the image feature; Obtain the product unit information of the target classification center; the product unit information is used to characterize the specification attribute information corresponding to the product to be identified. The product unit information and the commodity attribute information are subjected to a consistency verification to obtain a consistency verification result. Based on the authenticity score and the product attribute information, the product to be identified is determined to obtain the identification result, including: Based on the authenticity score, the product attribute information, and the consistency verification result, the product to be identified is identified, and the identification result is obtained.
6. The method according to claim 1, characterized in that, The process of identifying the product to be authenticated based on the authenticity score and the product attribute information, and obtaining the authentication result, includes: The authenticity score and the product attribute information are input into the large language model to obtain the identification result.
7. The method according to claim 6, characterized in that, The identification result includes the authenticity of the product to be identified and the reasons for determining its authenticity. Before inputting the genuine product score and the product attribute information into the large language model to obtain the identification result, the method further includes: Obtain a dataset, which includes sample genuine product scores, sample product attribute information, and labeled data corresponding to the sample genuine product scores and sample product attribute information. The labeled data includes genuine and fake labels and the reasons for determining genuine and fake. The pre-trained large language model is used to learn knowledge to identify the authenticity of goods and to attribute them based on the dataset.
8. A computer program product, characterized in that, It includes computer program instructions that, when executed by a processor, perform the method as described in any one of claims 1 to 7.
9. An electronic device, characterized in that, include: A processor and a memory, the memory storing computer program instructions that, when executed by the processor, perform the method as described in any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer program instructions that, when executed by a processor, perform the method as described in any one of claims 1 to 7.