Anti-counterfeiting and anti-theft marketing system and method based on QR code and AI visual recognition

Abstract

This invention relates to the field of anti-counterfeiting technology, specifically to an anti-counterfeiting and anti-theft system and method based on QR codes and AI visual recognition. The system includes the following steps: parsing QR code information and synchronously acquiring images; querying expected SKU information associated with the QR code; identifying the SKU of the product from the physical image; the backend server inputs the product image into a pre-trained AI visual model, which extracts the product's appearance features and determines its SKU; two-factor SKU matching verification; and processing the verification results. This invention employs a two-factor SKU matching verification mechanism combining QR code verification and AI visual recognition. Based on verifying the QR code verification status and validity period, it completes physical verification through product appearance feature extraction and carrier material analysis. This can identify non-physical theft in scenarios where QR code packages are leaked, reducing instances where genuine consumers are prevented from redeeming prizes and mitigating the adverse impact of theft on enterprise marketing security and brand reputation.

Description

Technical Field

[0001] This invention relates to the field of anti-counterfeiting technology, and in this context, to an anti-counterfeiting scanning marketing anti-theft system and method based on QR codes and AI visual recognition. Background Technology

[0002] Currently, QR code marketing systems are widely used in the consumer goods industry. The common practice is to print QR codes on product packaging. Some products also have an additional layer on top of the QR code, requiring the user to scratch it off and scan it with WeChat to claim a prize. Bottled products typically have the QR code inside the cap, which is then scanned upon opening the cap to redeem the prize. While redeeming the prize, the system records relevant information about the QR code, such as the scanning time and the WeChat account used. When the code is scanned again, it will display previous scanning information, helping consumers verify the product's authenticity.

[0003] However, a problem exists across the industry: when consumer goods companies distribute QR code data packages to their packaging suppliers (such as bottle cap, label, and bag suppliers) to produce coded packaging materials, inadequate employee management by these suppliers leads to code leaks. As a result, before consumers purchase products or scan the code, the QR code may have already been scanned by others, potentially leading to the theft of promotional offers. Furthermore, when consumers scan the code after purchasing goods, they are notified that the product has already been scanned, which has a significant negative impact on the consumer goods manufacturing industry and severely damages brand reputation. Therefore, we propose an anti-counterfeiting and anti-theft system and method based on QR code and AI visual recognition for QR code scanning marketing. Summary of the Invention

[0004] The purpose of this invention is to provide an anti-counterfeiting and anti-theft system and method for QR code marketing based on QR code and AI visual recognition, in order to solve the problem mentioned in the background art that the existing QR code marketing relies solely on the QR code verification status for prevention and control, lacks a physical verification process, and is easily stolen by non-physical items when the code package is leaked, which affects consumer rights and brand reputation.

[0005] To address the aforementioned technical problems, one objective of this invention is to provide a method for preventing counterfeiting and fraudulent use of QR codes in marketing based on QR code and AI visual recognition, comprising the following steps:

[0006] S1. QR code information parsing and image synchronous acquisition: When a consumer scans the QR code on a product using WeChat, the WeChat client simultaneously captures the current view as a product image during the scanning process, simultaneously parses the encoding information contained in the QR code, and sends the encoding information and the product image synchronously acquired during the scan to the backend server.

[0007] S2. Querying Expected SKU Information Associated with QR Code: The backend server queries the database for expected SKU information associated with the encoding information based on the encoding information, and synchronously verifies the QR code redemption status and validity period.

[0008] S3. Product SKU Recognition Based on Real-World Images: The backend server inputs real-world images of products into a pre-trained AI vision model, which then extracts the product's appearance features and determines its SKU.

[0009] S4. Two-factor SKU matching verification: The backend server compares and verifies the SKUs identified by the AI ​​visual model with the expected SKUs to determine whether they are consistent.

[0010] S5. Verification Result Processing: The backend server performs differentiated processing based on the comparison verification results. When the verification is successful, marketing reward authorization is executed and the scanning data is recorded to complete the single QR code verification and locking. When the verification fails, multi-dimensional visual features are used to analyze the material of the carrier on which the QR code is located, distinguishing between non-physical scanning (suspected fraud) and physical scanning (non-genuine products) scenarios and executing corresponding risk control operations.

[0011] As a further improvement to this technical solution, in step S1, the parsing of the QR code information and the synchronous acquisition of the image include the following steps:

[0012] S1.1 When consumers use WeChat's scan function to scan the QR code on the product, the camera viewfinder is brought up on the WeChat app, and the QR code encoding and parsing process and the viewfinder frame acquisition process are started simultaneously.

[0013] S1.2 The WeChat client performs real-time parsing of the QR code in the viewfinder, extracts the encoded information contained in the QR code, and if the parsing of the encoded information fails, it returns an "invalid QR code" prompt to the user and ends the process.

[0014] S1.3 At the same moment the QR code encoding and parsing is successful, WeChat automatically captures the complete image of the current camera viewfinder frame as a picture of the product. The image includes the QR code area and the surrounding product appearance area, without requiring the user to perform any additional manual shooting operation.

[0015] S1.4 The WeChat client integrates the parsed encoded information with the synchronously captured product images into a verification data packet, which is then encrypted and sent to the backend server.

[0016] As a further improvement to this technical solution, in step S2, the query for expected SKU information associated with the QR code includes the following steps:

[0017] S2.1 The backend server receives the encoded information transmitted from WeChat, verifies the integrity and legality of the encoded information, and retrieves the verification status, single-use identifier and preset validity period of the QR code bound to the encoded information.

[0018] S2.2 If the QR code is in a verified state or has exceeded its validity period, return the corresponding prompt to the user directly and terminate the verification process;

[0019] S2.3 The backend server uses the verified and valid unused encoded information as a search index to match the pre-established SKU database;

[0020] S2.4 The backend server performs a precise search in the SKU database based on the unique binding relationship between the coding information and the product SKU, and retrieves the expected SKU information that uniquely corresponds to the coding information.

[0021] As a further improvement to this technical solution, in step S3, the preprocessing and model input for physical image product SKU recognition includes the following steps:

[0022] S3.1 The backend server performs preprocessing on the product images in sequence, including size normalization, image denoising, and adaptive brightness adjustment.

[0023] S3.2 The backend server inputs the pre-processed product images into a dedicated AI visual model trained separately for each product SKU.

[0024] As a further improvement to this technical solution, in step S3, the feature extraction and standardization for physical image product SKU recognition specifically includes:

[0025] The AI ​​vision model performs a full-area visual scan of the product image.

[0026] AI visual models extract multi-dimensional key appearance features from real product images, including product packaging texture, brand logo, shape, and relative size ratio of the packaging.

[0027] The AI ​​vision model performs feature vector standardization on the extracted multi-dimensional key appearance features.

[0028] As a further improvement to this technical solution, in step S3, the feature comparison and SKU determination for physical image product SKU recognition specifically include:

[0029] The AI ​​visual model performs targeted matching and comparison between the standardized appearance feature vector and the pre-stored corresponding unique standard SKU feature vector;

[0030] AI visual models calculate feature similarity based on matching and comparison results;

[0031] The AI ​​visual model determines whether the feature similarity meets the preset judgment conditions. If it does, the standard SKU is identified as the SKU to which the product image belongs. If the feature similarity of the feature vector of the unique standard SKU does not meet the preset judgment conditions, the user is returned a "product cannot be identified" prompt and the process ends.

[0032] As a further improvement to this technical solution, in step S4, the two-factor SKU matching verification includes the following steps:

[0033] S4.1 The backend server obtains the SKU results of the products identified by the AI ​​visual model;

[0034] S4.2 The backend server retrieves the expected SKU information corresponding to the previously queried encoding information;

[0035] S4.3 The backend server will perform a complete consistency comparison between the identified SKU and the expected SKU;

[0036] S4.4 The backend server determines whether the two-factor SKU matching verification passes based on the comparison results.

[0037] As a further improvement to this technical solution, in step S5, the processing of the verified result specifically includes:

[0038] The backend server determined that the two-factor SKU matching verification result was consistent and passed.

[0039] The backend server executes the authorization operation for marketing prize redemption;

[0040] The backend server immediately marks the current QR code as verified, locks the single-use permission, and prohibits repeated scanning and verification.

[0041] The backend server records relevant information about this legitimate scan, and calculates the scan time, scan location, corresponding SKU information, and incorporates it into a unified scan database;

[0042] The backend server periodically analyzes the common characteristics and behavioral patterns of legitimate and abnormal QR code scanning data in the QR code scanning database, and iteratively optimizes the feature library and risk control early warning rules of the AI ​​vision model.

[0043] As a further improvement to this technical solution, in step S5, the processing of verification failure results specifically includes:

[0044] The backend server determined that the two-factor SKU matching verification result was inconsistent and failed.

[0045] The backend server intercepts the distribution of marketing resources;

[0046] The backend server relies on the full-image visual information combined with multi-dimensional features such as reflectivity, texture, and image clarity to perform material analysis on the carrier of the QR code and identify the material type of the object being scanned.

[0047] If the identification result is paper or electronic screen material, it is judged as suspected fraudulent use and a deep risk control mechanism is triggered: the scanning application guides the person scanning the code to authorize the acquisition of mobile phone number, location information and environmental photos, the above information is associated with the scanning time and QR code encoding information and marked as "suspected fraudulent use" and the evidence chain is completely stored, and the user is returned with a "system review in progress" prompt.

[0048] If the identification result indicates that the product is a physical material, it is determined to be a counterfeit product, and a "counterfeit product" message is returned to the user. No personal information is requested from the person who scanned the code.

[0049] The second objective of this invention is to provide an anti-counterfeiting and anti-theft marketing system based on QR codes and AI visual recognition. The computer program of this system, when running, executes the steps of the aforementioned anti-counterfeiting and anti-theft marketing method based on QR codes and AI visual recognition, including:

[0050] The QR code scanning and data collection interaction unit receives the consumer's QR code scanning operation, simultaneously collects the camera view as a product image during the QR code recognition process, parses the encoding information contained in the QR code, and uploads the encoding information and the product image together to the backend server.

[0051] The data verification and retrieval unit verifies the integrity and legality of the received coded information, checks the redemption status and validity period of the QR code, and retrieves the corresponding expected SKU information based on the unique binding relationship between the coded information and the product SKU.

[0052] The AI ​​visual intelligent recognition unit performs preprocessing on the product image, including size normalization, image denoising, and adaptive brightness adjustment. The preprocessed image is then input into a dedicated AI visual model to complete the extraction of product appearance features, feature standardization processing, and automatic identification of product SKU.

[0053] The SKU comparison and verification unit compares the consistency of the SKU of the product output by the AI ​​visual recognition with the expected SKU retrieved by the data verification and retrieval unit to determine whether the two-factor SKU matching verification is passed.

[0054] The result risk control processing unit performs differentiated processing based on the comparison and verification results. When the verification is successful, it executes marketing reward authorization and locks the QR code single-time verification permission. When the verification fails, it completes the QR code carrier material analysis based on multi-dimensional visual features such as reflectivity, texture shape, and imaging clarity, and distinguishes between suspected fraudulent use and non-genuine product scenarios to perform corresponding risk control operations.

[0055] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0056] This invention employs a two-factor SKU matching verification mechanism that combines QR code verification with AI visual recognition. Based on verifying the QR code verification status and validity period, it completes physical verification through product appearance feature extraction and carrier material analysis. This can identify non-physical fraudulent activities in scenarios where QR code packages are leaked, reduce the situation where genuine consumers are unable to redeem prizes, and reduce the adverse impact of fraudulent activities on enterprise marketing security and brand reputation. Attached Figure Description

[0057] Figure 1 This is a schematic diagram of the overall method steps of the anti-counterfeiting QR code marketing anti-theft method in this invention;

[0058] Figure 2 This is a schematic diagram of the overall framework of the anti-counterfeiting QR code marketing anti-theft system in this invention;

[0059] The meanings of the labels in the diagram are as follows:

[0060] 1. QR code scanning and data collection interaction unit;

[0061] 2. Data verification and retrieval unit;

[0062] 3. AI visual intelligent recognition unit;

[0063] 4. SKU comparison and verification unit;

[0064] 5. Result Risk Control and Handling Unit. Detailed Implementation

[0065] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0066] like Figure 1 As shown, this embodiment provides a method for preventing counterfeiting and fraudulent use of QR codes in marketing based on QR code and AI visual recognition, including:

[0067] S1. QR Code Information Parsing and Image Synchronization: When a consumer scans a QR code on a product using WeChat, WeChat simultaneously captures the current view as a product image during the scanning process, parses the encoded information contained in the QR code, and sends the encoded information along with the product image acquired synchronously during the scan to the backend server. Specifically, this includes the following steps:

[0068] S1.1 When consumers use WeChat's scan function to scan the QR code on the product, the camera viewfinder is brought up on the WeChat app, and the QR code encoding and parsing process and the viewfinder frame acquisition process are started simultaneously.

[0069] Specifically, consumers open the WeChat client, use the built-in scan function, and align the viewfinder with the QR code in a preset location on the product packaging (including under the scratch-off coating, inside the packaging, inside the sealed label, and other common coding locations). After the camera viewfinder is activated on WeChat, two background processes are started simultaneously and in parallel: one is the real-time parsing process of the QR code encoding information, which continuously decodes and recognizes the graphics in the viewfinder; the other is the real-time caching and acquisition process of the viewfinder frame, which continuously captures and temporarily caches the current viewfinder frame. The two processes run synchronously and do not block each other, achieving a seamless connection between the scanning action and the image acquisition. No additional shooting operation is required from the user, simplifying the user's operation path to the greatest extent.

[0070] S1.2 The WeChat client performs real-time parsing of the QR code in the viewfinder, extracts the encoded information contained in the QR code, and if the parsing of the encoded information fails, it returns an "invalid QR code" prompt to the user and ends the process.

[0071] Specifically, WeChat uses its built-in standardized QR code decoding algorithm to decode and recognize the captured QR code image frame by frame in real time. When a compliant QR code image is recognized, it automatically extracts core encoding information such as the unique product code, marketing activity identifier, and system verification code. If the QR code cannot be parsed due to printing defects, physical damage, obstruction, forgery, alteration, or being an unauthorized code, WeChat displays a clear "Invalid QR Code" message on the front-end interface and automatically terminates the scanning verification process. This prevents invalid data from being transmitted to the back-end server, improving system efficiency and allowing users to quickly understand the reason for the scanning failure. Successful code parsing is a necessary prerequisite for subsequent physical verification and marketing prize redemption; a screen capture operation is triggered simultaneously upon successful parsing.

[0072] S1.3 At the same moment the QR code encoding and parsing is successful, WeChat automatically captures the complete image of the current camera viewfinder frame as a picture of the product. The image includes the QR code area and the surrounding product appearance area, without requiring the user to perform any additional manual shooting operation.

[0073] Specifically, at the same moment that the WeChat decoding algorithm confirms that the QR code encoding information has been successfully parsed and verified, the system automatically retrieves the complete image of the latest frame from the real-time cached viewfinder frame as the product image for this verification. The entire process does not require the user to click the shooting button or manually confirm and submit; the user only needs to complete a single scan to complete the image acquisition.

[0074] The captured image fully covers the area where the QR code is located and the main area of ​​the product packaging around it. It includes the complete visual features of the QR code carrier, which can support the determination of reflectivity, texture and clarity in the subsequent carrier material analysis stage. It also includes the core visual information such as the appearance texture, brand logo and shape outline of the overall product packaging, which can meet the feature extraction needs of subsequent AI visual model SKU recognition.

[0075] In addition, the synchronous capture mechanism ensures that the QR code and the actual product are in the same imaging environment and the same shooting angle, effectively avoiding the problems of inconsistent images and misaligned features caused by step-by-step shooting. At the same time, it greatly reduces the user's operation time and improves the smoothness and convenience of scanning interaction.

[0076] S1.4 The WeChat client integrates the parsed encoded information with the synchronously captured product images into a verification data packet, which is then encrypted and sent to the backend server.

[0077] Specifically, after completing the encoding parsing and image capture, WeChat automatically encapsulates the complete QR code encoding information and the synchronously collected product images into a standardized verification data packet. The data packet contains a built-in data integrity verification identifier to ensure that the encoded information and the product images correspond one-to-one, without misalignment or loss. Subsequently, WeChat reliably sends the verification data packet to the marketing system's backend server through a secure and encrypted transmission channel.

[0078] The entire transmission process uses an industry-standard encryption protocol to effectively prevent data interception and tampering. This provides complete, authentic, and one-to-one corresponding input data for subsequent core processes such as QR code status verification, AI visual recognition, SKU matching verification, and carrier material analysis, ensuring the accuracy and reliability of the subsequent verification process.

[0079] S2. Querying Expected SKU Information Associated with QR Code: The backend server queries the database for expected SKU information associated with the encoding information, and synchronously verifies the QR code redemption status and validity period. Specifically, this includes the following steps:

[0080] S2.1 The backend server receives the encoded information transmitted from WeChat, verifies the integrity and legality of the encoded information, and retrieves the verification status, single-use identifier and preset validity period of the QR code bound to the encoded information.

[0081] Specifically, after receiving the QR code encoding information transmitted from WeChat through a secure encrypted interface, the backend server immediately initiates dual verification of integrity and legality:

[0082] Integrity verification: Check whether the field length, data format, and system check bits of the encoded information conform to the preset specifications, and investigate problems such as data transmission packet loss, format disorder, and missing fields;

[0083] Legality verification: Compare whether the code is an authorized and valid code generated by the enterprise's marketing system, and exclude forged codes, garbled characters, and external unauthorized codes.

[0084] After the verification is completed synchronously, the server uses the code as the unique retrieval index to synchronously retrieve three key status information from the core database of the marketing system: the redemption status of the QR code (redeemed / unredeemed), the single-use identifier (marked as one-time redemption, not allowed to be reused), and the preset effective usage period of the marketing activity (activity start time, end time).

[0085] This step filters out invalid and illegal codes at the data source, ensuring the effectiveness and security of subsequent verification processes.

[0086] S2.2 If the QR code is in a verified state or has exceeded its validity period, return the corresponding prompt to the user directly and terminate the verification process;

[0087] Specifically, the backend server performs real-time checks on the verification status and validity period of the retrieved QR codes, and quickly blocks QR codes that do not meet the usage conditions:

[0088] If the QR code is marked as verified, the server will return a message to the user's front end via the WeChat interface: "This QR code has been used. Please do not scan it again." This will automatically terminate the verification process.

[0089] If the current system time exceeds the activity's validity period, the server will return a "This QR code has expired and cannot participate in the activity" message to the user's front end, and the verification process will be terminated simultaneously.

[0090] This exception interception mechanism can quickly respond to invalid requests, reduce unnecessary background operations, improve system response efficiency, and clearly inform users of the reason for termination, avoiding user confusion.

[0091] S2.3 The backend server uses the verified and valid unused encoded information as a search index to match the pre-established SKU database;

[0092] Specifically, once it is determined that the encoding information has passed verification and the QR code is in a valid state that has not been verified and is within its validity period, the backend server will use the valid encoding as the unique retrieval index and automatically connect to the standardized SKU database pre-built by the system.

[0093] The SKU database in this embodiment is a lightweight database dedicated to the merchant, which only includes data on specific products that the merchant is currently participating in this marketing campaign and does not store irrelevant product information. At the same time, the AI ​​vision model used for subsequent recognition is a single SKU-specific training mode, which is trained only on the specific product appearance features of the merchant. There is no need to collect and store massive amounts of general product data, which greatly reduces the server storage pressure and model training cost. The training cycle is short, the deployment efficiency is high, and it can quickly adapt to the marketing needs of different merchants.

[0094] The SKU database has a built-in strong association mapping rule of "unique QR code code - product SKU", which ensures that each valid code can be accurately matched to the corresponding product SKU, without any association confusion or data redundancy issues.

[0095] S2.4 The backend server performs a precise search in the SKU database based on the unique binding relationship between the coding information and the product SKU, and retrieves the expected SKU information that uniquely corresponds to the coding information.

[0096] Specifically, based on the "one code corresponds to a unique SKU" binding relationship, the backend server performs precise search calculations in the SKU database to quickly locate and retrieve the expected SKU information that uniquely corresponds to the current code. This information includes core data such as the product standard SKU number, category attributes, packaging form, and appearance feature benchmark identifiers.

[0097] After the retrieval is completed, the server will temporarily store the expected SKU information in the system cache as benchmark reference data for subsequent AI visual recognition and two-factor SKU matching verification, ensuring the uniqueness and accuracy of subsequent verification processes.

[0098] S3. Product SKU Recognition from Physical Images: The backend server inputs physical images of products into a pre-trained AI vision model, which extracts the product's appearance features and determines its SKU; where:

[0099] In this step, the preprocessing and model input for physical product image SKU recognition includes the following steps:

[0100] S3.1 The backend server performs preprocessing on the product images in sequence, including size normalization, image denoising, and adaptive brightness adjustment.

[0101] Specifically, after receiving the product images uploaded by the user, the backend server performs three layers of standardized preprocessing on the images in sequence—size normalization, image denoising, and adaptive brightness adjustment—to eliminate image interference caused by differences in shooting environment and equipment, and to improve the stability and accuracy of AI visual recognition.

[0102] Size normalization: The server scales / crops real-world images of different resolutions and aspect ratios to the standard input size preset by the AI ​​vision model, ensuring that the image format fully matches the model's input requirements and avoiding recognition anomalies due to size incompatibility.

[0103] Image denoising: The filtering algorithm removes noise and blur interference caused by shooting shake, light scattering, and lens smudges from the image, while preserving the core texture, outline, and pattern features of the product's appearance.

[0104] Brightness adaptive adjustment: For poor shooting scenarios such as excessive darkness, overexposure, and backlighting, it automatically balances the brightness and contrast of the image to restore the true appearance, color and details of the product, ensuring that the model can accurately extract effective features.

[0105] After the above preprocessing, the physical images are optimized into interference-free, standardized, and high-definition recognition materials, providing a reliable foundation for accurate recognition by subsequent AI models.

[0106] S3.2 The backend server inputs the pre-processed product images into a dedicated AI visual model trained separately for each product SKU.

[0107] Specifically, after preprocessing, the backend server inputs the standardized images into a pre-trained single-SKU dedicated AI vision model.

[0108] The AI ​​vision model in this embodiment adopts a lightweight design that is merchant-specific and trained independently for each SKU. It is trained only for the specific products that the merchant is currently participating in the marketing campaign. It does not require the collection and storage of massive amounts of general product data, nor does it rely on large-scale general vision models, which greatly reduces model training costs, server storage pressure and inference time.

[0109] Meanwhile, single-SKU dedicated training allows AI vision models to deeply focus on the unique appearance features of the corresponding product, resulting in higher recognition accuracy and a lower misjudgment rate.

[0110] The AI ​​visual model corresponds one-to-one with the SKU database in step S2. Each dedicated AI visual model matches only one product SKU, ensuring that images can be quickly and accurately identified without redundant calculations. This fully meets the lightweight and rapid deployment needs of the consumer goods industry's marketing scenarios.

[0111] In this step, the feature extraction and standardization for physical product image SKU identification specifically includes:

[0112] The AI ​​vision model performs a full-area visual scan of the product image.

[0113] AI vision models extract multi-dimensional key appearance features from real product images, including product packaging texture, brand logo, shape, and relative size proportions. The relative size proportions of the packaging include the length-to-width ratio and the relative dimensions of each part of the product packaging.

[0114] The AI ​​vision model performs feature vector standardization on the extracted multi-dimensional key appearance features.

[0115] Specifically, after preprocessing the product images and inputting the standardized images into a pre-trained single-SKU-specific AI vision model, the AI ​​vision model immediately performs a full-domain visual region scan on the input standardized images. This full-domain scan covers the entire effective imaging pixel range of the image, without missing any local areas. During the scan, the model automatically identifies and distinguishes the main product area from the background area, accurately locking the core visual feature areas of the product packaging. At the same time, it filters out non-product interference information such as the shooting environment, background clutter, and irrelevant decorations, retaining only the effective visual data of the main product. This provides a clean and complete image foundation for subsequent feature extraction, avoiding problems such as incomplete product feature extraction and recognition deviations caused by background interference or missed areas.

[0116] Furthermore, the AI ​​vision model, based on a product feature recognition benchmark formed through training specifically for each SKU, extracts multi-dimensional key appearance features from the locked main visual region of the product. These extracted features cover the core recognition dimensions of the product, specifically including four categories: product packaging texture, brand logo, appearance shape, and packaging relative size ratio. Specifically, product packaging texture features include differentiated features such as the texture of the packaging material, surface printing patterns, and the feel of the coating; brand logo features include the outline of the product's exclusive brand identity, font style, color scheme, and fixed placement; appearance shape features include geometric features such as the overall outline of the product, packaging structure style, corner curvature, and component layout; and packaging relative size ratio features include the length-to-width ratio of the product packaging and the relative dimensions of each part. Because this model is a dedicated model trained for a single product SKU, the feature extraction process only captures the unique identifying features of the corresponding product, avoiding the extraction of generalized and non-discriminatory redundant features, further ensuring the specificity and accuracy of feature extraction.

[0117] Finally, after extracting key appearance features from multiple dimensions, the AI ​​vision model performs L2 normalization combined with Min-Max feature vector standardization on the extracted discrete features. The AI ​​vision model first unifies the feature vector magnitude through L2 normalization for visual features of different dimensions and units, such as packaging texture, brand logo, appearance shape, and relative size ratio of packaging, and then maps them to the standard numerical range of [0,1] through Min-Max standardization. This uniformly converts them into feature vectors with fixed dimensions and standard numerical ranges, eliminating the comparison bias caused by differences in numerical magnitude and units between feature dimensions, and ensuring that the converted feature data fully matches the standard feature format of the pre-stored SKU feature library.

[0118] The standardized feature vectors possess unified quantitative standards and comparability, and can be directly used for subsequent comparison calculations with standard features in the SKU feature library, providing stable and reliable data support for the accurate determination of product SKUs.

[0119] In this step, the feature comparison and SKU determination for physical product image SKU identification specifically includes:

[0120] The AI ​​visual model performs targeted matching and comparison between the standardized appearance feature vector and the pre-stored corresponding unique standard SKU feature vector;

[0121] AI visual models calculate feature similarity based on matching and comparison results;

[0122] The AI ​​visual model determines whether the feature similarity meets the preset judgment conditions. If it does, the standard SKU is identified as the SKU to which the product image belongs. If the feature similarity of the feature vector of the unique standard SKU does not meet the preset judgment conditions, the user is returned a "product cannot be identified" prompt and the process ends.

[0123] Specifically, after completing the vector standardization processing of the product's appearance features, the AI ​​vision model will perform targeted matching and comparison between the standardized appearance feature vector of the product obtained from this scan and the feature vector of the unique standard SKU corresponding to the code pre-stored in the system. The comparison process is based on the vector space similarity algorithm and only performs calculations on the target feature vectors in the merchant's exclusive lightweight SKU feature library. No redundant data participates in the calculation, ensuring matching efficiency and recognition accuracy.

[0124] During the targeted matching and comparison process, the AI ​​visual model calculates the feature similarity between the physical object feature vector and the standard SKU feature vector using a weighted cosine similarity algorithm. The specific calculation formula is as follows:

[0125] ;

[0126] This indicates the similarity between the physical appearance features of a product and the features of a standard SKU.

[0127] Indicates the first The preset weight coefficients for the dimensional features range from 0 to 1, and are preset according to the importance of core anti-counterfeiting features such as brand logo and appearance.

[0128] This indicates the product image after standardization processing. 1D feature vector values;

[0129] This indicates the first standard SKU in the pre-stored SKU feature library. 1D feature vector values;

[0130] This represents the total dimension of the standardized feature vector;

[0131] The dimension index of the feature vector, with a value of ;

[0132] Feature similarity The value range is [0,1]. The closer the value is to 1, the higher the degree of matching between the product appearance features of the real product image and the standard SKU features; the closer the value is to 0, the lower the degree of matching between the two.

[0133] Furthermore, the AI ​​vision model will calculate the feature similarity. The system compares the results with preset judgment conditions. In this embodiment, the preset judgment conditions are set according to the anti-counterfeiting identification accuracy requirements. The similarity can be adjusted between 0.90 and 0.98 depending on the complexity of the product's appearance. When a unique standard SKU feature vector exists, the feature similarity is... When the preset conditions are met, the AI ​​vision model directly identifies the standard SKU as the SKU to which the actual product image belongs, and transmits the judgment result to the backend server simultaneously, providing a core basis for subsequent two-factor SKU matching verification.

[0134] Finally, if the feature similarity corresponding to the feature vector of this unique standard SKU is obtained after this targeted comparison, If none of the uploaded product images meet the system's preset criteria, it means that the uploaded images cannot be matched with a valid product SKU within the system. The AI ​​vision model immediately sends a signal to the backend server indicating SKU recognition failure. Upon receiving this signal, the backend server sends a clear "Product cannot be recognized" message to the user through the WeChat front-end interface and automatically terminates the entire QR code verification process to avoid unnecessary waste of system resources caused by invalid data transfer.

[0135] S4. Two-Factor SKU Matching Verification: The backend server compares and verifies the SKUs identified by the AI ​​visual model with the expected SKUs to determine if they match. Specifically, this includes the following steps:

[0136] S4.1 The backend server obtains the SKU results of the products identified by the AI ​​visual model;

[0137] Specifically, the backend server retrieves and obtains the SKU result of the product after feature comparison and similarity determination from the output of the AI ​​visual intelligent recognition unit through the system's internal data transmission interface. This SKU result is a uniquely determined standard product SKU code, containing unique identification information such as the product's category, specifications, and packaging form. The server temporarily stores the identified SKU data in the system's real-time computing cache to ensure the immediacy and stability of data retrieval, providing actual SKU data to be verified for subsequent comparison and verification.

[0138] S4.2 The backend server retrieves the expected SKU information corresponding to the previously queried encoding information;

[0139] Specifically, the backend server retrieves the expected SKU information obtained from the precise retrieval of QR code encoding information in step S2 from the system cache. This expected SKU information is legal benchmark data that the enterprise pre-binds the QR code to the product during the product production coding stage. It represents that the QR code can only correspond to the legitimate product SKU for verification. After the server retrieves it, it locks the benchmark data to prevent the data from being tampered with or replaced, ensuring the uniqueness and authority of the verification reference standard.

[0140] S4.3 The backend server will perform a complete consistency comparison between the identified SKU and the expected SKU;

[0141] Specifically, the backend server activates its built-in data accuracy comparison module to perform a full-field consistency comparison between the actual SKU data obtained from AI visual recognition and the expected SKU data. The comparison covers all character segments of the SKU code, product attribute identifiers, specification codes, packaging types, and all related fields, ensuring no field omissions or partial simplifications. Only when every field and every character of the two sets of SKU data completely matches is it considered to be completely consistent at the character level, eliminating comparison errors caused by missing fields.

[0142] S4.4 The backend server determines whether the two-factor SKU matching verification passes based on the comparison results.

[0143] Specifically, the backend server makes the final judgment on the two-factor SKU matching verification based on the complete consistency comparison results: if the identified SKU is completely consistent with the expected SKU in all fields, the two-factor SKU matching verification is deemed to have passed; if there is any mismatch in any field or inconsistent characters between the two sets of SKU data, the two-factor SKU matching verification is deemed to have failed.

[0144] S5. Verification Result Processing: The backend server performs differentiated processing based on the verification results. If verification passes, marketing reward authorization is granted and scanning data is recorded, completing a single QR code verification lock. If verification fails, multi-dimensional visual features are used to analyze the material of the QR code carrier, distinguishing between non-physical scanning (suspected fraud) and physical scanning (non-genuine products) scenarios, and corresponding risk control operations are performed. Among these:

[0145] In this step, the processing of verified results specifically includes:

[0146] The backend server determined that the two-factor SKU matching verification result was consistent and passed.

[0147] The backend server executes the authorization operation for marketing prize redemption;

[0148] The backend server immediately marks the current QR code as verified, locks the single-use permission, and prohibits repeated scanning and verification.

[0149] The backend server records relevant information about this legitimate scan, and calculates the scan time, scan location, corresponding SKU information, and incorporates it into a unified scan database;

[0150] The backend server periodically analyzes the common characteristics and behavioral patterns of legitimate and abnormal QR code scanning data in the QR code scanning database, and iteratively optimizes the feature library and risk control early warning rules of the AI ​​vision model.

[0151] Specifically, after the backend server completes the two-factor SKU matching verification, it executes the corresponding processing logic based on the verification results. When the product SKU identified by step S4 is completely consistent with the expected SKU bound to the QR code and the two-factor SKU matching verification is passed, the backend server immediately starts the standardized processing procedure that has passed the verification, and sequentially completes the marketing prize redemption authorization, QR code verification and locking, scanning data recording and system iteration optimization operations to protect the prize redemption rights of legitimate consumers, while improving system data accumulation and function upgrades.

[0152] Specifically, the backend server first locks the verification conclusion that the two-factor SKUs match, marks the conclusion as a valid QR code verification result, and simultaneously clears the temporary cache data in the verification process to ensure that the verification result cannot be tampered with or modified retrospectively, providing a unique and legal basis for subsequent prize redemption and cancellation operations.

[0153] Once the verification is successful, the backend server immediately executes the marketing reward authorization operation. Based on the merchant's preset marketing activity rules, it sends a compliant reward redemption instruction to the WeChat client. After receiving the authorization instruction, the WeChat client issues corresponding marketing rewards to consumers according to the activity configuration, including cash red envelopes, membership points, and product coupons. The detailed data of reward issuance is synchronously transmitted back to the backend server in real time for storage, ensuring that the entire reward redemption process is traceable and dispute-free.

[0154] Furthermore, the backend server employs an atomic consistency processing mechanism for prize redemption and verification. Simultaneously with the issuance of prize redemption authorization, a verification lock operation is performed on the current QR code. The atomicity of the verification status update is guaranteed through database transactions. If there is a network error, reward issuance fails, or the user fails to redeem the prize, the unverified state of the QR code is automatically rolled back within a preset timeout period. In the system's core database, the verification status of the QR code is updated from "unverified" to "verified" in a single update, and a single-use lock flag is activated, permanently prohibiting the QR code from participating in marketing prize redemption and anti-counterfeiting verification again. This prevents duplicate scanning, malicious verification, and cross-product prize redemption from a data perspective, ensuring that enterprise marketing resources are not abused.

[0155] Meanwhile, after the prize redemption and verification are completed, the back-end server automatically collects and encrypts and archives all-dimensional related information of this legitimate scan, including the scan trigger time, real-time location information of the scanning device, corresponding product SKU code, unique QR code code, reward distribution type and denomination. All data is uniformly stored in the system's scan database to form a complete legitimate scan ledger, ensuring that the data is authentic, complete, long-term, and tamper-proof.

[0156] Finally, the backend server collects and analyzes the legitimate and abnormal scanning data in the scanning database according to a preset cycle, explores the AI ​​recognition patterns under different shooting scenarios and product forms, as well as the behavioral characteristics and operation modes of abnormal scanning, and supplements and calibrates the SKU feature library of the AI ​​visual model based on the analysis results, and iteratively optimizes the risk control warning threshold and judgment rules in parallel, continuously improving the model recognition accuracy and risk control interception precision, and realizing the continuous iterative upgrade of system functions.

[0157] In this step, the handling of verification failures specifically includes:

[0158] The backend server determined that the two-factor SKU matching verification result was inconsistent and failed.

[0159] The backend server intercepts the distribution of marketing resources;

[0160] The backend server relies on the full-image visual information combined with multi-dimensional features such as reflectivity, texture, and image clarity to perform material analysis on the carrier of the QR code and identify the material type of the object being scanned.

[0161] If the identification result is paper or electronic screen material, it is judged as suspected fraudulent use and a deep risk control mechanism is triggered: the scanning application guides the person scanning the code to authorize the acquisition of mobile phone number, location information and environmental photos, the above information is associated with the scanning time and QR code encoding information and marked as "suspected fraudulent use" and the evidence chain is completely stored, and the user is returned with a "system review in progress" prompt.

[0162] If the identification result indicates that the product is a physical material, it is determined to be a counterfeit product, and a "counterfeit product" message is returned to the user. No personal information is requested from the person who scanned the code.

[0163] Specifically, when the backend server verifies the product SKU output by the AI ​​visual model through S4 two-factor SKU matching and determines that there is a mismatch in fields, characters, or attributes between the product SKU identified by the AI ​​visual model and the expected SKU bound to the QR code, and confirms that the two-factor SKU matching verification result is inconsistent and fails, the standardized risk control process for verification failure is immediately initiated. The entire process includes marketing resource interception, carrier material analysis, and scenario-specific differentiated risk control operations, strictly adhering to user privacy protection standards while intercepting fraudulent transactions.

[0164] Specifically, the backend server immediately intercepts the distribution of marketing resources upon confirming the verification failure, terminating all instructions and data interactions related to marketing reward redemption. This completely blocks the distribution of marketing resources such as cash red envelopes, membership points, and coupons, preventing the illegal theft of corporate marketing resources from the source and ensuring the safety of marketing activity funds.

[0165] Furthermore, after intercepting marketing resources, the backend server initiates a multi-dimensional visual material analysis mechanism for the QR code carrier. Based on the full-domain visual information of the complete scanned image uploaded by the user, it comprehensively judges the image based on three core quantitative features: grayscale histogram reflectance intensity, grayscale co-occurrence matrix texture features, and Laplacian gradient variance imaging clarity. The grayscale histogram is used to statistically analyze reflectance intensity to distinguish the light reflection characteristics of the carrier surface. Paper and electronic screens exhibit uniform or specular reflectance, while physical product packaging displays natural diffuse reflection. The grayscale co-occurrence matrix is ​​used to extract texture morphology to identify the printing patterns and material textures on the carrier surface, distinguishing between the fiber texture of paper, the pixel matrix texture of a screen, and the unique packaging texture of a physical product. The Laplacian gradient variance is used to calculate imaging clarity to determine the imaging quality of the carrier, eliminating image blurring and pixel distortion issues associated with non-physical carriers, ultimately accurately identifying the carrier material type of the scanned object.

[0166] Specifically, if the material analysis results determine that the scanning medium is paper or an electronic screen, it is directly identified as suspected fraudulent use, and a deep risk control evidence collection mechanism is immediately triggered: the WeChat scanning front-end interface initiates authorization guidance to the scanner, applies for and obtains the scanner's mobile phone number and real-time geographical location information in accordance with regulations, and prompts the scanner to take a photo of the current scanning environment. After obtaining the user's authorization, the above information is collected in its entirety; at the same time, the server associates and binds the collected mobile phone number, location data, and environmental photo with the scanning time and the unique QR code encoding information, uniformly marks them with the "suspected fraudulent use" label, encrypts and stores them in the system's risk control evidence database, forming a complete and traceable evidence chain, and at the same time returns the "system review in progress" prompt message to the user's front end, without revealing the risk control evidence collection intention, and avoiding arousing the vigilance of the fraudster.

[0167] Specifically, if the material analysis results determine that the scanning carrier is made of the actual material of a genuine product, the backend server will classify this scanning behavior as scanning a counterfeit product. It will only return a clear "counterfeit" message to the user through the WeChat frontend, without initiating any personal information authorization requests to the person scanning the code, and without collecting privacy data such as phone numbers, locations, or photos. While clearly informing the user of the product's authenticity, it strictly protects the personal information security of legitimate users, in accordance with relevant user privacy protection regulations.

[0168] like Figure 2 As shown, this embodiment also provides an anti-counterfeiting and anti-theft marketing system based on QR code and AI visual recognition. The computer program running the system executes the steps of the aforementioned anti-counterfeiting and anti-theft marketing method based on QR code and AI visual recognition, including:

[0169] The QR code scanning and data collection interaction unit 1 receives the consumer's QR code scanning operation, and simultaneously collects the camera view as a picture of the product during the QR code recognition process, parses the encoding information contained in the QR code, and uploads the encoding information and the picture of the product together to the backend server.

[0170] Data verification and retrieval unit 2 verifies the integrity and legality of the received coded information, checks the redemption status and validity period of the QR code, and retrieves the corresponding expected SKU information based on the unique binding relationship between the coded information and the product SKU.

[0171] AI visual intelligent recognition unit 3 performs size normalization, image denoising, and brightness adaptive adjustment preprocessing on the product image. The preprocessed image is then input into the dedicated AI visual model to complete the extraction of product appearance features, feature standardization processing, and automatic recognition of product SKU.

[0172] SKU comparison and verification unit 4 compares the SKU of the product output by AI visual recognition with the expected SKU retrieved by the data verification and retrieval unit to determine whether the two-factor SKU matching verification is passed.

[0173] As a result, the risk control processing unit 5 performs differentiated processing based on the comparison and verification results. When the verification is successful, the marketing prize redemption authorization is executed and the QR code single-time redemption permission is locked. When the verification fails, the QR code carrier material is analyzed based on multi-dimensional visual features such as reflectivity, texture shape, and imaging clarity, and the corresponding risk control operation is executed to distinguish between suspected fraudulent use and non-genuine product scenarios.

[0174] Those skilled in the art will understand that the process of implementing all or part of the steps of the above embodiments can be carried out by hardware or by a program instructing the relevant hardware.

[0175] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention.

Claims

1. A method for preventing counterfeiting and fraudulent use of QR codes in marketing based on QR code and AI visual recognition, characterized in that: include: S1. QR code information parsing and image synchronous acquisition: When a consumer scans the QR code on a product using WeChat, the WeChat client simultaneously captures the current view as a product image during the scanning process, simultaneously parses the encoding information contained in the QR code, and sends the encoding information and the product image synchronously acquired during the scan to the backend server. S2. Querying Expected SKU Information Associated with QR Code: The backend server queries the database for expected SKU information associated with the encoding information based on the encoding information, and synchronously verifies the QR code redemption status and validity period. S3. Product SKU Recognition Based on Real-World Images: The backend server inputs real-world images of products into a pre-trained AI vision model, which then extracts the product's appearance features and determines its SKU. S4. Two-factor SKU matching verification: The backend server compares and verifies the SKUs identified by the AI ​​visual model with the expected SKUs to determine whether they are consistent. S5. Verification Result Processing: The backend server performs differentiated processing based on the comparison verification results. When the verification is successful, marketing reward authorization is executed and the scanning data is recorded to complete the single QR code verification and locking. When the verification fails, multi-dimensional visual features are used to analyze the material of the carrier on which the QR code is located, distinguish between non-physical scanning and physical scanning scenarios, and execute the corresponding risk control operations.

2. The anti-counterfeiting and anti-theft method for QR code marketing based on QR code and AI visual recognition according to claim 1, characterized in that, In step S1, the parsing of the QR code information and the synchronous acquisition of the image include the following steps: S1.1 When consumers use WeChat's scan function to scan the QR code on the product, the camera viewfinder is brought up on the WeChat app, and the QR code encoding and parsing process and the viewfinder frame acquisition process are started simultaneously. S1.2 The WeChat client performs real-time parsing of the QR code in the viewfinder, extracts the encoded information contained in the QR code, and if the encoding information parsing fails, returns an "invalid QR code" prompt to the user and ends the process; S1.3 At the same moment the QR code encoding and parsing is successful, WeChat automatically captures the complete image of the current camera viewfinder frame as a picture of the product. The image includes the QR code area and the surrounding product appearance area, without requiring the user to perform any additional manual shooting operation. S1.4 The WeChat client integrates the parsed encoded information with the synchronously captured product images into a verification data packet, which is then encrypted and sent to the backend server.

3. The anti-counterfeiting and anti-theft method for QR code marketing based on QR code and AI visual recognition according to claim 1, characterized in that, In step S2, the query for expected SKU information associated with the QR code includes the following steps: S2.1 The backend server receives the encoded information transmitted from WeChat, verifies the integrity and legality of the encoded information, and retrieves the verification status, single-use identifier and preset validity period of the QR code bound to the encoded information. S2.2 If the QR code is in a verified state or has exceeded its validity period, return the corresponding prompt to the user directly and terminate the verification process; S2.3 The backend server uses the verified and valid unused encoded information as a search index to match the pre-established SKU database; S2.4 The backend server performs a precise search in the SKU database based on the unique binding relationship between the coding information and the product SKU, and retrieves the expected SKU information that uniquely corresponds to the coding information.

4. The anti-counterfeiting and anti-theft method for QR code marketing based on QR code and AI visual recognition according to claim 1, characterized in that, In step S3, the preprocessing and model input for physical image product SKU recognition includes the following steps: S3.1 The backend server performs preprocessing on the product images in sequence, including size normalization, image denoising, and adaptive brightness adjustment. S3.2 The backend server inputs the pre-processed product images into a dedicated AI visual model trained separately for each product SKU.

5. The anti-counterfeiting and anti-theft method for QR code marketing based on QR code and AI visual recognition according to claim 1, characterized in that, In step S3, the feature extraction and standardization for physical product image SKU identification specifically includes: The AI ​​vision model performs a full-area visual scan of the product image. AI visual models extract multi-dimensional key appearance features from real product images, including product packaging texture, brand logo, shape, and relative size ratio of the packaging. The AI ​​vision model performs feature vector standardization on the extracted multi-dimensional key appearance features.

6. The anti-counterfeiting and anti-theft method for QR code marketing based on QR code and AI visual recognition according to claim 1, characterized in that, In step S3, the feature comparison and SKU determination for physical product image SKU recognition specifically include: The AI ​​visual model performs targeted matching and comparison between the standardized appearance feature vector and the pre-stored corresponding unique standard SKU feature vector; AI visual models calculate feature similarity based on matching and comparison results; The AI ​​visual model determines whether the feature similarity meets the preset judgment conditions. If it does, the standard SKU is identified as the SKU to which the product image belongs. If the feature similarity of the feature vector of the unique standard SKU does not meet the preset judgment conditions, the user is returned a "product cannot be identified" prompt and the process ends.

7. The anti-counterfeiting and anti-theft method for QR code marketing based on QR code and AI visual recognition according to claim 1, characterized in that, In S4, the two-factor SKU matching verification includes the following steps: S4.1 The backend server obtains the SKU results of the products identified by the AI ​​visual model; S4.2 The backend server retrieves the expected SKU information corresponding to the previously queried encoding information; S4.3 The backend server will perform a complete consistency comparison between the identified SKU and the expected SKU; S4.4 The backend server determines whether the two-factor SKU matching verification passes based on the comparison results.

8. The anti-counterfeiting and anti-theft method for QR code marketing based on QR code and AI visual recognition according to claim 1, characterized in that, In S5, the processing of the verified result specifically includes: The backend server determined that the two-factor SKU matching verification result was consistent and passed. The backend server executes the authorization operation for marketing prize redemption; The backend server immediately marks the current QR code as verified, locks the single-use permission, and prohibits repeated scanning and verification. The backend server records relevant information about this legitimate scan, and calculates the scan time, scan location, corresponding SKU information, and incorporates it into a unified scan database; The backend server periodically analyzes the common characteristics and behavioral patterns of legitimate and abnormal QR code scanning data in the QR code scanning database, and iteratively optimizes the feature library and risk control early warning rules of the AI ​​vision model.

9. The anti-counterfeiting and anti-theft method for QR code marketing based on QR code and AI visual recognition according to claim 1, characterized in that, In S5, the processing of verification failure results specifically includes: The backend server determined that the two-factor SKU matching verification result was inconsistent and failed. The backend server intercepts the distribution of marketing resources; The backend server uses full-image visual information combined with multi-dimensional features such as reflectivity, texture, and image clarity to perform material analysis on the carrier of the QR code and identify the material type of the object being scanned. If the identification result is paper or electronic screen material, it is judged as suspected fraudulent use and a deep risk control mechanism is triggered: the scanning application guides the person scanning the code to authorize the acquisition of mobile phone number, location information and environmental photos, the above information is associated with the scanning time and QR code encoding information and marked as "suspected fraudulent use" and the evidence chain is completely stored, and the user is returned with the "system review in progress" prompt. If the identification result indicates that the product is a physical material, it is determined to be a counterfeit product, and a "counterfeit product" message is returned to the user. No personal information is requested from the person who scanned the code.

10. A QR code and AI visual recognition-based anti-counterfeiting and anti-theft marketing system, characterized in that: When the computer program of the anti-counterfeiting and anti-theft marketing system based on QR code and AI visual recognition is executed, it performs the steps of the anti-counterfeiting and anti-theft marketing method based on QR code and AI visual recognition as described in any one of claims 1-9, including: The QR code scanning and data collection interaction unit (1) receives the consumer's QR code scanning operation, and simultaneously collects the camera view as a picture of the product during the QR code scanning and recognition process, parses the encoding information contained in the QR code, and uploads the encoding information and the picture of the product to the backend server together. The data verification and retrieval unit (2) verifies the integrity and legality of the received coded information, checks the redemption status and validity period of the QR code, and retrieves the corresponding expected SKU information based on the unique binding relationship between the coded information and the product SKU. AI visual intelligent recognition unit (3) performs size normalization, image noise reduction and brightness adaptive adjustment preprocessing on the real product image, inputs the preprocessed image into the exclusive AI visual model, and completes product appearance feature extraction, feature standardization processing and automatic product SKU recognition. SKU comparison verification unit (4), the SKU comparison verification unit (4) compares the consistency of the SKU of the product output by AI visual recognition with the expected SKU retrieved by the data verification and retrieval unit, and determines whether the two-factor SKU matching verification is passed; The result risk control processing unit (5) performs differentiated processing based on the comparison and verification results. When the verification is successful, it performs marketing prize redemption authorization and locks the QR code single redemption permission. When the verification fails, it performs QR code carrier material analysis based on multi-dimensional visual features such as reflectivity, texture shape, and imaging clarity, and distinguishes between suspected fraudulent use and non-genuine product scenarios to perform corresponding risk control operations.

Images