Commodity anti-counterfeiting system and method based on public key infrastructure and artificial intelligence fusion detection

Abstract

The invention discloses a commodity anti-counterfeiting system and method based on public key infrastructure (PKI) and artificial intelligence (AI) fusion detection. The system comprises a merchant side, a verification center and a user side. And the merchant side manages the private key and uploads training data containing commodity codes and fine physical characteristics. The verification center stores the public key, the data and an AI model trained according to the public key and the data and aiming at fine physical characteristics. The verification center executes two steps of cooperative verification: the first step is PKI verification to confirm the authenticity and integrity of the anti-counterfeiting code digital identity; and only after passing, executing AI physical identification at the server side in the second step, processing the user image frame, and obtaining a physical entity fine feature result by using an AI model to judge the authenticity. And the user side scans the code, shoots, and transmits and receives data and results. According to the invention, through cooperative gating of PKI and server-side high-precision AI identification, dual verification of a digital identity and a physical entity is realized, the anti-counterfeiting level is improved, and high-imitation products are effectively identified.

Description

Technical Field

[0001] The present invention relates to the technical field of product anti-counterfeiting, and particularly to a product anti-counterfeiting system and method that combines public key infrastructure for security verification and artificial intelligence for physical object recognition and fusion detection. Background Art

[0002] With the development of the cultural and creative industries and the tourism industry, the problem of counterfeiting of various cultural and creative products and souvenirs (such as fridge magnets, badges, models, etc.) has become increasingly prominent. It not only infringes on the intellectual property rights and economic interests of designers and sellers, but also affects consumers' purchasing experiences and brand reputations. Traditional anti-counterfeiting measures, such as ordinary printed labels and packaging anti-counterfeiting codes, are easily copied or transferred, and it is difficult to effectively prevent the situation of "false goods with genuine codes". Especially for cultural and creative products with high appearance similarity but different details, it is more challenging to verify authenticity.

[0003] Some existing digital anti-counterfeiting technologies, such as simple QR code traceability, mainly verify the validity of digital information and cannot check the physical object itself. Some solutions that attempt to combine image recognition either rely on the limited computing power of the user's mobile device and are difficult to run complex and high-precision recognition models, or the recognition technology itself is sensitive to changes in light and angle and has insufficient robustness. For example, it is difficult to capture the possible subtle three-dimensional features, material textures, or specific process defects of cultural and creative products only by comparing the 2D photos taken by users and pre-stored images.

[0004] Therefore, when the existing technologies are applied to the anti-counterfeiting of products such as cultural and creative products, there are generally problems in simultaneously, automatically, and reliably verifying the authenticity of the digital identity of products, the authority of the source, and the consistency of the physical entity itself features. In particular, there is a lack of a comprehensive anti-counterfeiting solution that can effectively combine a strong security mechanism (such as public key infrastructure) with high-precision and anti-interference physical feature recognition (including the processing of complex details or multi-modal information) executed using the powerful computing power of the cloud. Summary of the Invention

[0005] The main purpose of the present invention is to overcome the deficiencies of the existing technologies and provide a product anti-counterfeiting system and method based on the fusion detection of public key infrastructure and artificial intelligence.

[0006] To achieve the above purpose, the present invention provides a product anti-counterfeiting system based on the fusion detection of public key infrastructure and artificial intelligence, and the system includes: A merchant terminal, configured to enable a merchant to securely store its private key and upload physical product data including a unique product code and reflecting the subtle physical features of the individual or batch of the product for artificial intelligence recognition model training to a verification center; The verification center, configured as a cloud server cluster, is used to store merchant public keys, product information, the physical product data, and an artificial intelligence recognition model for the subtle physical features trained based on the physical product data; the verification center is further configured to: Receive a verification request from the user terminal, which includes anti-counterfeiting code data and real-time captured product image frame data; The first verification step: Perform public key infrastructure verification, which includes: decrypting the anti-counterfeiting code data using the verification center private key to obtain the merchant identifier and digital signature, and verifying the validity of the digital signature using the stored corresponding merchant public key. This signature is generated by the merchant using its private key for the information containing the unique product code to confirm the authenticity of the digital identity and data integrity of the anti-counterfeiting code; The second verification step, which is executed on the premise that the first verification step passes: If the public key infrastructure verification passes, then perform artificial intelligence physical object recognition on the server side of the verification center. The recognition includes: loading the artificial intelligence recognition model associated with the product, processing the received product image frame data, and using the model for inference to obtain a physical object recognition result related to the subtle features of the product physical entity, which is used to judge the authenticity of the physical object; Based on the public key infrastructure verification result and the artificial intelligence physical object recognition result, comprehensively determine the authenticity of the product; The user terminal, configured as a mobile terminal application, is used to scan the anti-counterfeiting code on the product, capture the product image frame data in real time, send the anti-counterfeiting code data and image frame data to the verification center, and receive and display the product authenticity determination result from the verification center.

[0007] Optionally, when the verification center performs artificial intelligence physical object recognition, it is further configured to: Receive at least two different modalities of product image frame data captured by the user terminal, where the modalities include visible light images and infrared images; Perform preprocessing and artificial intelligence model inference on the image frame data of different modalities respectively; Adopt a fusion strategy to integrate the inference results of different modalities.

[0008] Optionally, the fusion strategy is a late fusion strategy, including weighted averaging of the similarity or confidence of the inference outputs of each modality.

[0009] Optionally, the processing performed by the verification center on the product image frame data includes at least one of the following: deblurring, noise suppression, image normalization, adaptive histogram equalization, and illumination invariance processing.

[0010] Optionally, the anti-counterfeiting code is a two-dimensional code, and its data structure includes data blocks encrypted by the public key of the verification center, and the data blocks contain merchant identifiers and the digital signature.

[0011] Optionally, the artificial intelligence recognition model is a classification model or a metric learning model, and / or the model training utilizes physical commodity data including manually marked geometric features and depth visual features automatically extracted through deep learning.

[0012] Optionally, the communication between the merchant side and the verification center is encrypted using the Secure Sockets Layer protocol or the Transport Layer Security protocol based on the Transmission Control Protocol; the communication between the user side and the verification center uses the Transmission Control Protocol or the Hypertext Transfer Security Protocol.

[0013] Optionally, the cryptographic algorithms used for public key infrastructure verification and anti-counterfeiting code encryption include the RSA algorithm for key generation and signature, and the AES algorithm for symmetric encryption.

[0014] The present invention also provides a commodity anti-counterfeiting method based on the integration detection of public key infrastructure and artificial intelligence. This method is implemented using the above system and includes the following steps: The merchant manages the private key through the merchant side and uploads the unique commodity code and physical commodity data reflecting the subtle physical characteristics of the commodity individual or batch for artificial intelligence recognition model training to the verification center; The verification center trains and generates an artificial intelligence recognition model for the subtle physical characteristics based on the physical commodity data and stores it; The merchant generates an anti-counterfeiting code for the commodity, and the generation process includes: digitally signing the information containing the unique commodity code using the merchant's private key, and encrypting the data block containing the merchant identifier and the digital signature using the public key of the verification center; The user scans the anti-counterfeiting code on the commodity through the user side, and real-time collects the commodity image frame data, and sends the anti-counterfeiting code data and the image frame data to the verification center; The verification center receives the request and performs verification, including: The first verification step: perform public key infrastructure verification: use the private key of the verification center to decrypt the anti-counterfeiting code data to obtain the merchant identifier and the digital signature, and use the corresponding merchant public key to verify the validity of the signature to confirm the authenticity of the anti-counterfeiting code digital identity and the data integrity; The second verification step, which is executed on the premise that the first verification step passes: perform artificial intelligence physical object recognition on the server side: load the artificial intelligence recognition model, process the received image frame data, and use model inference to obtain the physical object recognition result related to the subtle characteristics of the commodity physical entity for judging the authenticity of the physical object; The verification center comprehensively determines the authenticity of the product based on the results of the first verification step and the results of the second verification step, and sends the determination result to the user terminal; The user terminal receives and displays the determination result.

[0015] Optionally, the user terminal collects and sends at least two different modalities of product image frame data, including visible light images and infrared images; the steps for the verification center to perform artificial intelligence physical object recognition include: Processing and performing artificial intelligence model inference on the image frame data of different modalities respectively; Adopting a fusion strategy to integrate the inference results of different modalities.

[0016] According to the technical content disclosed in the present invention, the following beneficial effects are achieved: By deeply integrating and collaborating the public key infrastructure (PKI) that provides trusted digital identity verification with the artificial intelligence (AI) technology that is executed on the server side of the verification center and specifically targets the high-precision recognition of the subtle physical characteristics of products, the present invention constructs a closed-loop product anti-counterfeiting system and method, bringing significant beneficial effects. The system first uses PKI to compulsorily verify the authenticity of the digital identity and data integrity of the anti-counterfeiting code, effectively intercepting invalid or tampered digital identifiers and avoiding unnecessary consumption of subsequent AI computing resources; only after the PKI verification passes, the system will trigger the high-precision AI model on the server side to conduct a detailed physical feature analysis of the product image taken by the user in real time, so as to judge the authenticity of the physical object itself, making up for the inherent defect that PKI cannot confirm the physical entity. This "gated" collaborative dual verification mechanism that first performs digital identity verification and then physical feature recognition, and the latter is based on the success of the former, greatly improves the anti-counterfeiting threshold, forcing counterfeiters to simultaneously break through the standard password system and accurately copy the subtle physical characteristics of products that are difficult to imitate. In addition, placing the complex AI recognition task on the server side allows the use of more advanced models (such as deep metric learning or fine classification models) and combines technologies such as multi-modal data fusion and image preprocessing, which can accurately identify highly similar high-quality fakes and improve the verification robustness in complex shooting environments. In summary, the present invention provides an anti-counterfeiting solution with high security, accurate recognition, and user-friendliness, which is particularly suitable for products with rich physical details and easy to be highly imitated.

[0017] Through the following detailed description of the exemplary embodiments of the present invention with reference to the accompanying drawings, other features and advantages of the present invention will become clear. Description of the Drawings

[0018] The accompanying drawings incorporated in the specification and constituting a part of the specification illustrate the embodiments of the present invention, and together with the description are used to explain the principles of the present invention.

[0019] Figure 1 It is a schematic diagram of the overall architecture and interaction process of the commodity anti-counterfeiting system provided by the embodiment of the present invention based on the integration of public key infrastructure and artificial intelligence detection. Detailed implementation manners

[0020] Now, various exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings. It should be noted that: unless otherwise specifically stated, the relative arrangements of components and steps, numerical expressions and values set forth in these embodiments do not limit the scope of the present invention.

[0021] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way a limitation on the present invention or its application or use.

[0022] Technologies, methods and devices known to those of ordinary skill in the relevant art may not be discussed in detail, but where appropriate, the technologies, methods and devices should be regarded as part of the specification.

[0023] In all the examples shown and discussed herein, any specific values should be construed as merely exemplary and not as a limitation. Thus, other examples of the exemplary embodiments may have different values.

[0024] It should be noted that: similar reference numerals and letters denote similar items in the following drawings, and thus, once an item is defined in one drawing, it need not be further discussed in subsequent drawings.

[0025] See Figure 1 , and the present invention will be further described below in conjunction with a more detailed embodiment.

[0026] Embodiment: Anti-counterfeiting of a cultural and creative product (such as a fridge magnet) Initialization and data preparation phase (merchant side & verification center): Merchant registration and certificate acquisition: The designer or seller of cultural and creative products (merchants) access the secure web page background provided by the verification center through their authorized computers. Submit registration information, cooperation agreements, etc. After the verification center reviews, it provides the merchant with a digital certificate containing its public key (for example, in PEM format, based on RSA-2048), and emphasizes that the merchant needs to properly store the private key using a hardware security module or a security USB key, etc. The verification center stores the public key of the merchant.

[0027] Commodity Information and Training Data Upload: For each cultural and creative product (taking refrigerator magnets as an example), the merchant sets a unique batch number or product code for it. At the standardized acquisition station, using a fixed camera position and stable light source, record high-definition MP4 videos of multiple samples of the refrigerator magnet from multiple angles such as the front and side that can reflect its key design features (such as three-dimensional shape, surface printed pattern details, special material areas), or take a series of high-definition image sequences. The merchant logs in to the background, creates an entry for the refrigerator magnet, inputs the name, code, material description, etc., and uploads the corresponding training video / image data.

[0028] Artificial Intelligence Model Training Stage (Verification Center): Feature Extraction and Dataset Construction: The model training unit in the verification center (running on a cloud GPU server) receives the training data. The system processes and extracts features aiming to comprehensively capture the subtle physical differences that can distinguish genuine from fake, including: (a) Automatic extraction: Use a convolutional neural network (CNN) to automatically learn and extract the deep visual features of the pattern texture, color distribution, and three-dimensional contour on the surface of the refrigerator magnet, as well as possible minor printing defects or material features. (b) Optional geometric feature annotation: For refrigerator magnets whose designs particularly rely on specific shapes or contours, key points or edge lines can be assisted in annotation.

[0029] Model Training: Target Detector Training (for Small Object Scenarios): For relatively small objects like refrigerator magnets, to reduce the interference caused by complex backgrounds when users take pictures, a lightweight target detection model can be trained first using a part of the training images with the positions of the refrigerator magnets marked. For example, the YOLOv5s architecture can be selected, pre-trained and fine-tuned for the refrigerator magnet dataset to obtain a detector that can efficiently and accurately locate the region of interest (ROI) of the refrigerator magnet in the user-uploaded image, and locate the commodity area where the refrigerator magnet is located. This step aims to improve the accuracy and robustness of subsequent feature recognition.

[0030] Feature Recognition Model Training: Based on the extracted feature dataset (if a target detector is used, mainly based on the features within the ROI area), train a high-precision feature recognition model, such as a deep metric learning model or a fine classification model, whose goal is to be able to distinguish the subtle feature differences between genuine refrigerator magnets and counterfeits or other objects. After training, associate and store the target detector (if used) and the feature recognition model for subtle physical features with the code of the refrigerator magnet. Model training can be carried out using physical commodity data containing manually marked geometric features and deep visual features automatically extracted through deep learning.

[0031] Anti-counterfeiting Code Generation and Coding Stage (Merchant Side): Signature and Encryption: During the production and packaging process of fridge magnets, the system obtains the unique code and production timestamp of the current batch or individual fridge magnets. The merchant uses their securely stored private key (RSA-2048) to digitally sign the "code + timestamp" information.

[0032] Data Encapsulation and Encryption: Combine the merchant identifier and the digital signature. Encrypt this data block using the verification center's public key through a hybrid encryption scheme (such as RSA+AES).

[0033] Encoding and Coding: Encode the encrypted data block into a QR code. Print the QR code on the independent small packaging bag of the fridge magnet, or directly print / laser mark it on the flat area on the back of the fridge magnet in a wear-resistant manner.

[0034] User Verification Phase (User Side & Verification Center): Scanning and Shooting: After purchasing the fridge magnet, the consumer opens the corresponding mobile application (user side, with camera and network permissions authorized). The App guides the user to first scan the QR code on the packaging or the fridge magnet itself. The App decodes to obtain the encrypted data.

[0035] Image Acquisition: Subsequently, the App prompts the user to aim the mobile terminal camera at the fridge magnet, trying to keep the fridge magnet centered in the viewfinder and stable, and take a real-time image. The App acquires consecutive visible light image frames, and if the mobile terminal supports and the function is enabled, it also acquires infrared image frames.

[0036] Data Upload: The App packs the parsed QR code encrypted data and the acquired several key image frames (visible light and optional infrared) into JSON or a similar format and sends it to the verification center via HTTPS.

[0037] Verification Center Processing Phase (Verification Center): Receiving and Parsing: The verification center API receives the request. The verification processing unit parses the data.

[0038] First Verification Step: Public Key Infrastructure Verification: Use the verification center's private key to decrypt the QR code data to obtain the merchant identifier and signature. Look up the merchant's public key and verify the validity of the signature. If it fails, it is determined as an "invalid code", and the process ends without subsequent AI recognition.

[0039] Second Verification Step: Artificial Intelligence Recognition (Server Side, Only Executed After the First Step Succeeds): If the PKI verification is successful, load the corresponding AI model (which may include an object detector and a feature recognizer) according to the recovered product code.

[0040] In some embodiments, it also includes positioning the area where the product is located: If the target detector is loaded, first input the received image frame into the detector (e.g., the pre-trained refrigerator magnet YOLOv5s model). The detector outputs the bounding box (ROI) of the refrigerator magnet in the image. Subsequent processing will focus on the area where the product is located.

[0041] Preprocessing: Perform image enhancement processing on the original image frame (or the detected area where the product is located): deblurring, noise suppression, image normalization, adaptive histogram equalization, illumination invariance processing, etc.

[0042] Feature extraction and comparison / recognition: Input the preprocessed image (or the area where the product is located) into the feature recognition model. The model extracts its subtle physical features and compares them with the reference features of the genuine refrigerator magnet of this model stored, and outputs a similarity score (or classification confidence) as the physical entity recognition result related to the subtle physical features.

[0043] In some embodiments, it also includes multimodal fusion: If there are visible light and infrared image frames at the same time, perform the above preprocessing and recognition steps respectively to obtain their respective similarity scores. Then perform late fusion. For example, perform weighted averaging on the similarities or confidences output by each modality to integrate the inference results of different modalities.

[0044] Result determination: Compare the final similarity with a preset threshold (e.g., 0.98). If it is greater than or equal to the threshold, it is determined that the physical entity is consistent, otherwise it is "the physical entity does not match".

[0045] Result feedback stage (verification center & user side): Generate response: Combine the results of the first verification step (PKI verification) and the second verification step (AI recognition) to generate a final conclusion: (a) Success: Prompt "Genuine product certification: [product name]", and may be accompanied by a product story or relevant link. (b) Invalid code: Prompt "The anti-counterfeiting code is invalid. Please confirm the source." © Physical entity does not match: Prompt "The anti-counterfeiting code is valid, but the physical features do not match. Beware of counterfeits."

[0046] Send response: Package the conclusion and send it back to the user-side App via HTTPS (or other secure transmission protocols such as the secure socket layer / transport layer protocol based on TCP).

[0047] User display: The App receives and clearly displays the verification result to the user. The entire interaction process strives to be completed within a few seconds.

[0048] In some embodiments, it also includes considerations regarding robustness and false positives / negatives: In order to improve the robustness of the system in actual applications and reduce the probabilities of false positives (genuine products being judged as fakes) and false negatives (fake products being judged as genuine products), the following measures can be taken: Image preprocessing enhancement: Apply more complex image preprocessing algorithms before AI recognition, such as motion blur removal, light adaptive correction, noise filtering, etc., to reduce the impact of environmental factors on image quality.

[0049] Multi-modal data fusion: As mentioned above, combine different modal information such as visible light and infrared, and utilize their respective advantages (such as infrared being more sensitive to certain material features) to improve the comprehensiveness and anti-interference ability of recognition.

[0050] Data augmentation training: During the model training stage, perform data augmentation operations on training samples such as random rotation, scaling, cropping, brightness / contrast adjustment, adding noise, etc., so that the model learns the robustness to these changes.

[0051] Threshold dynamic adjustment and optimization: According to the feedback and test data in actual applications, continuously optimize the threshold for similarity determination to balance the false positive rate and false negative rate. Different thresholds can be set for different product lines or batches.

[0052] Model selection and continuous iteration: Select or design a deep learning model architecture that is sensitive to subtle features and has strong generalization ability. Collect failure cases in actual verification and continuously retrain and iteratively optimize the model.

[0053] Confidence evaluation and fuzzy determination: For samples with similarity scores near the threshold (for example, between 0.95 - 0.98), fuzzy prompts such as "Suspected genuine product, please check carefully" or guiding the user to take more detailed photos can be output instead of directly giving an absolute determination to reduce the negative impact caused by extreme misjudgments.

[0054] (For small objects) Location of the commodity area: As mentioned in the embodiment, by performing object detection to locate the commodity area first, the interference of complex backgrounds can be effectively excluded, significantly improving the accuracy and robustness of small object recognition.

[0055] Communication security and encryption algorithms: To ensure data transmission security, the communication between the merchant side and the verification center can be encrypted using the Secure Sockets Layer (SSL) protocol or Transport Layer Security (TLS) protocol based on the Transmission Control Protocol (TCP). The communication between the user side and the verification center can use the Transmission Control Protocol (TCP) or the Hypertext Transfer Protocol Secure (HTTPS).

[0056] In the public key infrastructure verification and anti-counterfeiting code encryption processes involved in the present invention, mature cryptographic algorithms can be adopted. For example, the RSA algorithm (such as RSA-2048) can be used for asymmetric key generation and digital signature, and the AES algorithm (such as AES-128 or AES-256) can be used for symmetric encryption (such as encrypting the data block itself in a hybrid encryption scheme). In summary, the technical solution of the present invention constructs a collaborative system integrating public key infrastructure and server-side artificial intelligence recognition, bringing significant beneficial effects: Strong security: By using asymmetric encryption and digital signature technologies, it ensures the credibility and non-forgeability of the source of anti-counterfeiting labels, and the integrity and anti-tampering of data during transmission. It effectively intercepts counterfeited or tampered anti-counterfeiting codes at the digital level.

[0057] High-precision physical recognition: Placing complex artificial intelligence recognition tasks on the verification center server with powerful computing capabilities enables the operation of more advanced and precise models (such as deep learning models, multi-modal fusion models) to accurately capture and compare the subtle physical characteristics (texture, three-dimensional shape, material, printing details, etc.) of physical goods, effectively distinguishing genuine and fake physical objects and overcoming the limitation of insufficient computing power of user-end devices.

[0058] Dual verification mechanism: Combining the dual barriers of digital identity authentication (PKI) and physical entity recognition (AI) requires counterfeiters to break the password system and precisely replicate the physical characteristics of goods simultaneously, greatly increasing the difficulty and cost of counterfeiting.

[0059] High robustness: The image processing and AI models on the server side can use more complex algorithms and data augmentation techniques to effectively resist adverse factors in practical applications such as light changes, shooting angle deviations, and background interference, improving the stability and reliability of verification. Especially for scenarios such as cultural and creative products that may be small in size and have complex backgrounds, the optional ROI positioning technology further enhances the robustness of recognition.

[0060] Automation and user-friendliness: Users only need to perform simple operations of scanning codes and taking pictures to quickly obtain authoritative verification results. The verification process has a high degree of automation and a good user experience.

[0061] Scalability and flexibility: The system architecture is clear, supports multiple cryptographic algorithms and AI models, and can be configured and upgraded according to different product characteristics and security requirements. The server-side models can be continuously optimized and iterated, and the recognition ability can be improved without users updating the App.

[0062] Therefore, the present invention provides an efficient, secure, reliable and user-friendly anti-counterfeiting solution for goods, which is particularly suitable for fields such as cultural and creative products, luxury goods, and important certificates with high anti-counterfeiting requirements.

[0063] Although some specific embodiments of the present invention have been described in detail by way of examples, those skilled in the art should understand that the above examples are only for illustration purposes and not for limiting the scope of the present invention. Those skilled in the art should understand that the above embodiments can be modified without departing from the scope and spirit of the present invention. The scope of the present invention is defined by the appended claims.

Claims

1. A commodity anti-counterfeiting system based on public key infrastructure and artificial intelligence fusion detection, characterized in that: include: The merchant side is configured to allow the merchant to safely keep its private key and upload physical product data containing the unique product code and the subtle physical characteristics of individual or batch products for AI recognition model training to the verification center; The verification center is configured as a cloud server cluster, and is used to store the merchant public key, product information, the physical product data, and an artificial intelligence recognition model for the subtle physical features generated and trained based on the physical product data; the verification center is also configured as: Receiving a verification request from a user end including anti-counterfeiting code data and real-time collected commodity image frame data; First verification step: performing public key infrastructure verification, the verification includes: using the verification center private key to decrypt the anti-counterfeiting code data to obtain the merchant identification and digital signature, and using the stored corresponding merchant public key to verify the validity of the digital signature, which is generated by the merchant using its private key to the information containing the unique code of the product to confirm the digital identity authenticity and data integrity of the anti-counterfeiting code; The second verification step is performed on the premise that the first verification step is passed: if the public key infrastructure verification is passed, then artificial intelligence physical object recognition is performed on the server side of the verification center, and the recognition includes: loading an artificial intelligence recognition model associated with the commodity, processing the received commodity image frame data, and using the model to perform reasoning to obtain a physical object recognition result related to the subtle features of the physical entity of the commodity, which is used to determine the authenticity of the physical object; Comprehensively determine the authenticity of the product based on the public key infrastructure verification result and the artificial intelligence physical object recognition result; The user end is configured as a mobile terminal application, which is used to scan the anti-counterfeiting code on the product, collect the product image frame data in real time, send the anti-counterfeiting code data and image frame data to the verification center, and receive and display the product authenticity determination results from the verification center.

2. The system according to claim 1, characterized in that When performing artificial intelligence physical object recognition, the verification center is also configured to: Receiving commodity image frame data in at least two different modalities collected by a user terminal, wherein the modalities include a visible light image and an infrared image; Preprocess image frame data of different modalities and perform artificial intelligence model inference respectively; A fusion strategy is used to integrate the reasoning results of different modalities.

3. The system according to claim 2, characterized in that The fusion strategy is a late fusion strategy, which includes weighted averaging of the similarity or confidence of each modal reasoning output.

4. The system according to claim 1, characterized in that The processing performed by the verification center on the commodity image frame data includes at least one of the following: deblurring, noise suppression, image standardization, adaptive histogram equalization, and illumination invariance processing.

5. The system according to claim 1, characterized in that The anti-counterfeiting code is a two-dimensional code, and its data structure includes a data block encrypted by the public key of the verification center, and the data block contains the merchant identification and the digital signature.

6. The system according to claim 1, characterized in that The artificial intelligence recognition model is a classification model or a metric learning model, and / or the model training utilizes physical commodity data containing manually marked geometric features and deep visual features automatically extracted through deep learning.

7. The system according to claim 1, characterized in that The communication between the merchant end and the verification center is encrypted using the Secure Sockets Layer protocol or the Secure Transport Layer protocol based on the Transmission Control Protocol; the communication between the user end and the verification center is encrypted using the Transmission Control Protocol or the Hypertext Transfer Protocol Security Protocol.

8. The system according to claim 1, characterized in that The cryptographic algorithms used for the public key infrastructure authentication and anti-counterfeiting code encryption include the RSA algorithm for key generation and signature, and the AES algorithm for symmetric encryption.

9. A commodity anti-counterfeiting method based on public key infrastructure and artificial intelligence fusion detection, characterized in that: The following steps are involved: Merchants manage private keys through the merchant end and upload the unique product code and physical product data that reflects the subtle physical characteristics of individual or batch products for AI recognition model training to the verification center; The verification center generates and stores an artificial intelligence recognition model for the subtle physical features based on the training of the physical commodity data; The merchant generates an anti-counterfeiting code for the product, and the generation process includes: using the merchant's private key to digitally sign the information containing the unique code of the product, and encrypting the data block containing the merchant's identification and the digital signature using the verification center's public key; The user scans the anti-counterfeiting code on the product through the user terminal, collects the product image frame data in real time, and sends the anti-counterfeiting code data and image frame data to the verification center; The verification center receives the request and performs verification, including: The first verification step: perform public key infrastructure verification: use the verification center private key to decrypt the anti-counterfeiting code data to obtain the merchant identification and digital signature, and use the corresponding merchant public key to verify the validity of the signature to confirm the authenticity of the anti-counterfeiting code digital identity and data integrity; The second verification step is performed on the premise that the first verification step is passed: performing artificial intelligence physical object recognition on the server side: loading the artificial intelligence recognition model, processing the received image frame data, and using the model reasoning to obtain the physical object recognition results related to the subtle features of the physical entity of the commodity, which are used to determine the authenticity of the physical object; The verification center comprehensively determines the authenticity of the product based on the result of the first verification step and the result of the second verification step, and sends the determination result to the user terminal; The client receives and displays the determination result.

10. The method according to claim 9, characterized in that The user terminal collects and sends commodity image frame data in at least two different modes, including visible light images and infrared images; the verification center performs artificial intelligence physical object recognition steps including: Process image frame data of different modalities and perform AI model inference separately; A fusion strategy is used to integrate the reasoning results of different modalities.

Images