Online real-time image recognition anti-counterfeiting verification method and system based on gold transaction

Abstract

The present application relates to the technical field of image recognition, and particularly relates to an online real-time image recognition anti-fake verification method and system based on gold transaction. The method comprises the following steps: collecting a microscopic texture image of a gold product; generating a unique digital fingerprint based on a physical unclonable feature of the microscopic texture image, and encrypting the unique digital fingerprint in combination with a UTC time stamp to generate a dynamic encryption label; extracting a feature descriptor of the microscopic texture image and performing texture comparison with a pre-stored database to obtain a texture matching degree; decrypting the dynamic encryption label to generate a dynamic code time-effect state; and performing anti-fake discrimination on the microscopic texture image of the gold product according to the texture matching degree and the dynamic code time-effect state to obtain a genuine product discrimination result and a fake product discrimination result. The present application improves the accuracy, security and efficiency of gold anti-fake by combining the microscopic texture image, near-infrared spectrum data, dynamic encryption label and blockchain technology.

Description

Technical Field

[0001] The present invention relates to the technical field of image recognition, and in particular to an online real-time image recognition anti-counterfeiting verification method and system based on gold transactions. Background Art

[0002] Anti-counterfeiting measures in gold trading primarily rely on physical identifiers, such as inscriptions, laser codes, and QR codes. These physical identifiers can help verify the authenticity of gold, but in large-scale online transactions, these methods are costly to implement and difficult to implement in real time. With the development of artificial intelligence and deep learning technologies, anti-counterfeiting technologies based on image recognition have emerged. By analyzing multi-dimensional information such as gold's surface features, luster, and texture, image recognition can accurately identify and verify gold. With the introduction of technologies such as deep convolutional neural networks (CNNs), the accuracy and speed of image recognition technology have significantly improved. Especially with the continuous advancement of big data processing and real-time processing capabilities, online real-time image recognition anti-counterfeiting verification methods are becoming a standard feature of gold trading platforms. However, existing technologies typically rely on single physical identifiers (such as laser inscriptions and QR codes), which are easily forged or tampered with, and cannot guarantee high security. Furthermore, these technologies often rely on a single verification step, resulting in a high risk of counterfeit goods entering the market, which in turn reduces the security and credibility of gold trading. Summary of the Invention

[0003] Based on this, it is necessary to provide an online real-time image recognition anti-counterfeiting verification method and system based on gold transactions to solve at least one of the above technical problems.

[0004] To achieve the above-mentioned purpose, an online real-time image recognition anti-counterfeiting verification method based on gold transactions is provided, the method comprising the following steps:

[0005] Step S1: Collect a micro-texture image of the gold product; generate a unique digital fingerprint based on the physical unclonable characteristics of the micro-texture image, and encrypt it with the UTC timestamp to generate a dynamic encrypted tag;

[0006] Step S2: extract the feature descriptor of the micro texture image and compare it with the pre-stored database to obtain the texture matching degree; decrypt the dynamic encryption tag and generate the dynamic code time status;

[0007] Step S3: Perform anti-counterfeiting identification on the micro-texture image of the gold product based on the texture matching degree and the aging status of the dynamic code to obtain the authenticity identification results and the counterfeit identification results; store the counterfeit identification results in the blockchain for evidence recording, and generate a genuine gold verification storage record;

[0008] Step S4: Based on the authenticity identification result, a multi-level review mechanism is initiated, and the micro-texture image of the gold product is again subjected to material intrinsic property anti-counterfeiting identification through the multi-level review mechanism. A secondary anti-counterfeiting identification result is obtained and synchronously stored in the blockchain, generating a secondary gold anti-counterfeiting identification storage record.

[0009] This invention uses a dynamic encrypted tag generated by encrypting a digital fingerprint generated from a microtexture image with a UTC timestamp to ensure that each gold product has a unique and unforgeable identity. This encrypted tag dynamically changes over time, making each anti-counterfeiting tag unique and authentic, enhancing the security of gold products. By extracting feature descriptors from the microtexture image and comparing them with a database, the texture characteristics of the gold product can be accurately identified. Combined with the detection of anomalous absorption peaks in near-infrared spectral data, spectral deviations can be effectively identified, further improving the accuracy of anti-counterfeiting identification. The results of authentic gold identification are stored in the blockchain for evidence, ensuring data immutability and providing consumers with a traceable verification record, enhancing transparency and market trust in gold products. A multi-level verification mechanism is activated to verify authenticity, reducing the possibility of counterfeiting by re-identifying the intrinsic properties of the material. This mechanism ensures multiple safeguards in the identification process and improves the reliability of the anti-counterfeiting technology. By combining texture features, spectral data, and the dynamic encrypted tag, this multi-layered anti-counterfeiting approach ensures that gold products are highly resistant to counterfeiting in all aspects and is adaptable to various counterfeiting methods. Therefore, the present invention improves the accuracy, security and efficiency of gold anti-counterfeiting by combining micro-texture images, near-infrared spectral data, dynamic encryption tags and blockchain technology.

[0010] Preferably, step S1 includes the following steps:

[0011] Step S11: collecting near-infrared spectrum data of the gold product using a near-infrared spectrometer; collecting an original microscopic image of the surface of the gold product using an industrial microscope, and performing denoising and contrast enhancement on the image to generate a microscopic enhanced image;

[0012] Step S12: performing lattice structure segmentation on the microscopic enhanced image to generate a lattice distribution binary mask;

[0013] Step S13: extracting the pore area in the mask, and counting the number, spacing and shape factor of the pores in the pore area to generate a pore feature vector;

[0014] Step S14: fusing the lattice distribution binary mask and the pore feature vector and marking them as physical unclonable feature data; generating a digital fingerprint of the gold product based on the physical unclonable feature data to obtain a unique digital fingerprint;

[0015] Step S15: Perform AES-256 encryption on the unique digital fingerprint in combination with the UTC timestamp to generate a dynamic encryption tag.

[0016] This invention combines a near-infrared spectrometer with an industrial microscope to capture the physical and surface characteristics of gold products from various angles. Near-infrared spectral data provides information on the chemical composition of gold, while microscopic images reveal surface texture details. This combination of multi-source data provides a more comprehensive basis for subsequent anti-counterfeiting assessments. De-noising and contrast enhancement improve the quality of microscopic images, ensuring the accuracy of subsequent lattice structure segmentation and pore feature extraction. Image enhancement enhances the visibility of minute details, facilitating the precise extraction of surface texture and pore information. Lattice structure segmentation and feature extraction of pore regions quantify the physical properties of the gold surface. The number, spacing, and shape factor of pores are unique characteristics of gold materials. These features serve as a unique digital fingerprint, adding a layer of depth to the anti-counterfeiting of gold products. By integrating the lattice distribution binary mask with the pore feature vector, the generated physical unclonable feature data is ensured to be highly unique and unclonable. These features serve as a unique identifier for gold products, ensuring their authenticity and uniqueness. Based on the generated physical unclonable feature data, a unique digital fingerprint is generated and, combined with the UTC timestamp, AES-256 encrypted to form a dynamic encrypted tag. AES-256 encryption ensures the tag's security and tamper-proofing, while the dynamic encrypted tag's real-time changes enhance anti-counterfeiting efforts, ensuring it cannot be copied or tampered with. The encryption and feature fusion technology employed throughout the entire process ensures that the gold product's anti-counterfeiting mechanism cannot be easily cracked. The combination of digital fingerprints and encrypted tags not only enhances the security of anti-counterfeiting technology but also provides a traceable digital record of gold products, enabling consumers to verify the authenticity of gold at any time.

[0017] Preferably, step S14 includes the following steps:

[0018] Step S141: fusing the lattice distribution binary mask with the pore feature vector and marking them as physical unclonable feature data;

[0019] Step S142: extracting the eigenvectors of the physical unclonable feature data, and performing high-dimensional mapping conversion on the eigenvectors to generate high-dimensional feature matrix data;

[0020] Step S143: performing hash coding calculation on the high-dimensional feature matrix data to generate unique hash fingerprint data;

[0021] Step S144: performing error correction processing on the unique hash fingerprint data to generate a unique digital fingerprint.

[0022] This invention fuses a lattice distribution binary mask with a pore feature vector to generate highly unique physical unclonable feature data. The physical features of each gold product are quantified based on its surface lattice structure and pore characteristics, making them impossible to copy or forge, providing a solid foundation for anti-counterfeiting. High-dimensional feature matrix data is generated by performing a high-dimensional mapping transformation on the feature vectors of the physical unclonable feature data. This high-dimensional mapping provides a more complex and in-depth representation of the original features, allowing the feature vectors to more comprehensively express the subtle differences between gold products, further enhancing the accuracy of anti-counterfeiting detection. The high-dimensional feature matrix data is calculated using hash coding to generate unique hash fingerprint data. The hash algorithm compresses the high-dimensional feature data into a unique hash value of a fixed length. This hash value is not only unique but also fast to retrieve and compare, facilitating subsequent verification and comparison. The generation of hash fingerprints effectively mitigates the risks of data leakage and forgery. Error correction of the unique hash fingerprint data further improves the accuracy and reliability of the fingerprint data. Error correction reduces the impact of minor errors during acquisition, processing, or storage, ensuring the stability and consistency of the unique digital fingerprint. The resulting unique digital fingerprint offers high security and accuracy, ensuring that each gold product possesses an unreplicable and unalterable identity. This enables effective traceability of gold products across various stages of development, while also eliminating the risk of counterfeiting and substitution, providing consumers with a more reliable verification method. This process combines physical unclonable features, high-dimensional feature mapping, hashing algorithms, and error correction to provide multiple safeguards against counterfeiting of gold products. Whether in practical applications or at any level of verification, it ensures data uniqueness and security, reducing the likelihood of forgery.

[0023] Preferably, extracting the feature descriptor of the micro texture image and performing texture comparison with a pre-stored database in step S2 includes:

[0024] Extract the surface microstructure trend features of the microtexture image, analyze the microtexture image for rotation invariance, and obtain local rotation features;

[0025] Locate texture key points of micro texture images based on surface micro structure orientation features and local rotation features;

[0026] Calculate local feature descriptors for each texture keypoint;

[0027] Import the feature descriptors into the pre-stored database to perform feature descriptor comparison and obtain descriptor matching pairs;

[0028] The similarity of the descriptor matching pairs is calculated, and the descriptor matching pairs with the highest similarity and the descriptor matching pairs with the lowest similarity are screened out for weighted average processing to generate the texture matching degree.

[0029] By extracting surface microstructure orientation features from microtexture images and analyzing their rotational invariance, this method ensures that even when the texture image of a gold product is rotated, the features remain consistent. This process makes texture comparison unaffected by directional variations, increasing the stability and accuracy of the anti-counterfeiting technology at different angles. Based on the surface microstructure orientation features and local rotational characteristics, texture keypoints in the texture image can be precisely located. Accurate keypoint location provides the basis for subsequent feature descriptor calculation, ensuring the accuracy and reliability of the subsequent comparison process. Local feature descriptors are calculated for each texture keypoint, enabling detailed characterization of the texture features of each gold product. Local feature descriptors capture local details in the image and provide precise information during comparison, effectively distinguishing different gold products. By importing the extracted feature descriptors into a pre-stored database for comparison, texture features similar to those of gold products in the database can be quickly and efficiently matched. Descriptor comparison provides the basis for subsequent match degree calculation, enhancing the accuracy and speed of the verification process. Similarity is calculated for matching descriptor pairs, and the matching pairs with the highest and lowest similarity are selected through weighted mean processing. This process ensures that the final texture matching score comprehensively considers the most relevant matching pairs, while reducing the possibility of false matches and making the matching results more stable and reliable. Through the above feature extraction, comparison, and weighted average processing, the generated texture matching score accurately reflects the similarity between the gold product and the gold in the pre-stored database. This precise texture comparison mechanism improves the accuracy of gold anti-counterfeiting detection, making it difficult for counterfeit products to be easily verified using traditional comparison methods.

[0030] Preferably, step S2 further includes:

[0031] Perform spectral peak positioning on near-infrared spectral data to generate spectral peak position data;

[0032] Calculate the peak height, full width at half maximum and peak area of ​​the spectral peak position data respectively to obtain the spectral peak position characteristic data set;

[0033] Perform peak shape analysis on the spectral peak position data based on the spectral peak position characteristic data set to generate absorption peak morphology data;

[0034] Utilize the spectral peak position characteristic data set and absorption peak morphology data to detect and screen abnormal absorption peaks in near-infrared spectral data, generating abnormal absorption peak detection data and normal absorption peak screening data;

[0035] Perform spectral deviation analysis on abnormal absorption peak detection data and normal absorption peak screening data to generate spectral deviation values.

[0036] By performing peak location analysis on near-infrared spectral data, the present invention can accurately identify peak positions within the spectrum and generate spectral peak position data. This provides a foundation for subsequent spectral analysis, ensuring that each absorption peak is accurately captured and analyzed in detail. By calculating the peak height, full-width at half-maximum (FWHM), and peak area of ​​the spectral peak position data, a spectral peak position feature dataset is generated. This step provides a detailed description of the morphology and characteristics of each absorption peak, increasing the depth of spectral data analysis and facilitating the identification of anomalous absorption peaks and effective differentiation of normal absorption peaks. Peak shape analysis is performed based on the spectral peak position feature dataset to generate absorption peak morphology data. This process analyzes the properties of absorption peaks from a morphological perspective, providing additional dimensions for detecting anomalous absorption peaks and enhancing the accuracy of anomalous peak identification. Combining the spectral peak position feature dataset and absorption peak morphology data allows for the detection and screening of anomalous absorption peaks in near-infrared spectral data. This process effectively identifies anomalous absorption peaks that are significantly different from normal absorption peaks, avoiding errors caused by anomalous absorption peaks and thus improving the accuracy of spectral data analysis. Spectral deviation analysis is performed on the abnormal absorption peak detection data and the normal absorption peak screening data to generate spectral deviation values. This analysis step can identify and quantify deviations in the spectral data, helping to further correct errors in the data, ensuring the accuracy of the spectral data and providing precise spectral information for subsequent anti-counterfeiting identification. Abnormal absorption peak detection and screening can effectively eliminate abnormal signals in the spectral data, ensuring the cleanliness and reliability of the data. This reduces interference caused by noise or erroneous signals, improving the stability of the overall system and the reliability of anti-counterfeiting identification.

[0037] Preferably, decrypting the dynamic encryption tag in step S2 includes:

[0038] parsing the dynamic parameters of the dynamic encryption tag to perform hierarchical decoding of the tag structure of the dynamic encryption tag, thereby obtaining a hierarchical encrypted data unit;

[0039] Derivation of time-varying keys based on layered encrypted data units to obtain a dynamic decryption key sequence;

[0040] Reverse mapping the dynamic encryption tag according to the dynamic decryption key sequence to obtain the dynamic decryption intermediate state data;

[0041] Perform timeliness state synchronization and noise elimination on the dynamic decryption intermediate state data to obtain dynamic code timeliness state data;

[0042] By integrating the space-time dimension of the dynamic code aging status data, the dynamic code aging status is finally output.

[0043] By parsing the dynamic parameters of a dynamic encryption tag and performing layered decoding, the present invention effectively breaks down the complex tag structure into multiple encrypted data units. This layered process provides a clear structure for subsequent decryption operations, ensuring that each data unit can be processed independently, reducing the error rate and complexity of the decryption process. Based on the layered encrypted data units, time-varying key derivation is performed to generate a dynamic decryption key sequence. Time-varying key derivation enhances the security of the tag. The key is dynamically generated based on time during the decryption process, effectively avoiding the risk of static key cracking and significantly improving the tag's anti-attack capabilities. By performing inverse mapping on the dynamic decryption key sequence, the intermediate state data in the dynamic encryption tag can be effectively recovered. The polymorphic algorithm provides flexible decryption strategies in complex encryption environments, ensuring efficient data recovery in various situations and improving the adaptability and security of the decryption process. Time-sensitive state synchronization and noise removal of the dynamically decrypted intermediate state data help eliminate noise interference during the decryption process, ensuring more stable and accurate temporal data. Noise removal and time-sensitive state synchronization effectively improve the reliability and real-time performance of the final dynamic code, ensuring high accuracy of the decryption results under various circumstances. By integrating the spatial and temporal dimensions of dynamic code time-stamp data, the data can be effectively combined in both spatial and temporal dimensions. This fusion technology enhances the timeliness and spatial accuracy of tag decryption, improves the dynamic decryption process's adaptability to environmental changes, and enhances the tag's decryption stability in complex scenarios.

[0044] Preferably, the anti-counterfeiting determination of the micro texture image of the gold product according to the texture matching degree and the aging state of the dynamic code in step S3 includes:

[0045] The gold product's micro-texture image is subjected to anti-counterfeiting judgment based on texture matching, spectral deviation value and dynamic code aging status. When the following conditions are present at the same time, it is judged as authentic: texture matching is greater than or equal to 0.90 and less than or equal to 1.00, spectral deviation is less than or equal to 0.05, and dynamic code aging is greater than or equal to 90% and less than or equal to 100%;

[0046] When the following conditions occur at the same time, it is judged as a counterfeit identification result: the texture matching degree is less than 0.90 and greater than or equal to 0.80, the spectral deviation is greater than 0.05 and less than or equal to 0.15, and the dynamic code timeliness is less than 90% and greater than or equal to 70%.

[0047] This invention effectively improves the accuracy of anti-counterfeiting detection by combining three core metrics: texture matching (which measures the degree of match between the microtexture image of a gold product and a database archive), spectral deviation (which reflects whether the material's near-infrared spectral characteristics are abnormal), and dynamic code aging (which determines the timeliness of the digital fingerprint). The introduction of dynamic code aging makes the anti-counterfeiting system time-sensitive. Even if an attacker replicates the microtexture or spectral characteristics of the gold surface, they cannot forge the timeliness of the dynamic code, thereby enhancing the system's anti-counterfeiting capabilities. The spectral deviation threshold ensures that even if the surface texture is faked, the material's optical properties differ and can still be identified, preventing the counterfeiting of alloy coatings. The established counterfeit discrimination ranges (0.80 ≤ texture matching < 0.90, 0.05 < spectral deviation ≤ 0.15, and 70% ≤ dynamic code aging < 90%) prevent false rejections due to overly strict authenticity criteria and improve the system's adaptability in practical applications. This flexible anti-counterfeiting identification strategy can accurately identify high-quality counterfeits while reducing misjudgments of legitimate products, making the system more stable and reliable in actual industrial testing. This identification method, based on numerical calculations and independent of subjective human judgment, is suitable for automated anti-counterfeiting detection systems, significantly improving detection speed and efficiency.

[0048] Preferably, step S4 includes the following steps:

[0049] Step S41: activating a multi-level review mechanism based on the authenticity identification result, wherein the multi-level review mechanism includes manual review and weight verification;

[0050] Step S42: performing microstructural stress response analysis on the gold product through a multi-level review mechanism to generate microstructural stress response data;

[0051] Step S43: performing lattice resonance frequency detection on the gold product according to the microstructure stress response data to obtain resonance frequency characteristic data;

[0052] Step S44: Use the resonance frequency characteristic data and microstructure stress response data to perform material intrinsic property anti-counterfeiting judgment on the microtexture image of the gold product, obtain a secondary anti-counterfeiting judgment result and synchronously store it in the blockchain, generating a secondary gold anti-counterfeiting judgment storage record.

[0053] This invention utilizes a multi-level verification mechanism (manual re-inspection + weight verification) to ensure more in-depth verification of suspicious samples, preventing misjudgments caused by single-detection errors. Incorporating microstructural stress response analysis, this method avoids relying solely on surface texture, spectral data, or encrypted tags while neglecting the structural properties of the material itself, thereby enhancing the reliability of the gold anti-counterfeiting system. Traditional anti-counterfeiting methods primarily rely on surface feature detection, while microstructural stress response analysis can detect changes in stress distribution in gold at the microscopic level, preventing the creation of high-quality counterfeits through surface imitation. Lattice resonance frequency detection identifies counterfeits based on the material's physical properties. Even if counterfeits are highly imitated in parameters such as appearance and density, their lattice resonance characteristics are difficult to replicate, significantly enhancing anti-counterfeiting effectiveness. The results of this secondary anti-counterfeiting verification are synchronously stored on the blockchain, ensuring data integrity and traceability, and preventing tampering or falsification of authentication results. The decentralized storage of blockchain data allows for the verification of every gold product inspection record, enhancing the credibility of the anti-counterfeiting system and preventing repeated counterfeiting or data falsification.

[0054] In this specification, an online real-time image recognition and anti-counterfeiting verification system based on gold transactions is provided, which is used to execute the above-mentioned online real-time image recognition and anti-counterfeiting verification method based on gold transactions. The online real-time image recognition and anti-counterfeiting verification system based on gold transactions includes:

[0055] The tag module is used to collect micro-texture images of gold products; based on the physical unclonable characteristics of the micro-texture image, a unique digital fingerprint is generated, which is encrypted with the UTC timestamp to generate a dynamic encrypted tag;

[0056] Feature analysis module is used to extract feature descriptors of micro texture images and compare textures with pre-stored databases to obtain texture matching degree; decrypt dynamic encryption tags and generate dynamic code aging status;

[0057] The primary anti-counterfeiting identification module is used to perform anti-counterfeiting identification on the micro-texture image of gold products based on texture matching and the aging status of the dynamic code to obtain the results of authenticity identification and counterfeit identification. The counterfeit identification results are stored in the blockchain for evidence recording, generating a primary verification storage record of authentic gold.

[0058] The secondary anti-counterfeiting identification module is used to initiate a multi-level review mechanism based on the authenticity identification results, and to conduct anti-counterfeiting identification of the material intrinsic properties of the micro-texture image of the gold product again through the multi-level review mechanism. The secondary anti-counterfeiting identification results are obtained and stored synchronously in the blockchain to generate a secondary gold anti-counterfeiting identification storage record.

[0059] The beneficial effect of this invention lies in generating a unique digital fingerprint based on the physically unclonable characteristics of microtexture images, ensuring that each gold product has a unique identification mark and preventing counterfeiting and tampering. Dynamic encryption, combined with a UTC timestamp, ensures that the digital fingerprint is unique at each point in time. Even if an attacker obtains current data, it cannot be used for future counterfeiting. By extracting microtexture features and comparing them with a pre-stored database, texture matching is calculated, effectively preventing surface counterfeiting methods such as gold plating and laser engraving. Detecting gold's specific absorption peaks and analyzing spectral deviations prevents tampering with adulteration or composition. Decrypting the tag reveals the dynamic code's aging status, ensuring the integrity and traceability of the gold's identity information. Combining texture matching, spectral deviation, and dynamic code aging status, the first round of anti-counterfeiting verification is performed to quickly identify authentic gold. Authentic gold information is stored on the blockchain, ensuring that its data cannot be tampered with and forming a record of the gold's authenticity. If the authenticity verification result is positive, a multi-level review mechanism is initiated, including manual re-inspection and weight verification, to further improve accuracy. Through microstructural stress response analysis and resonant frequency signature detection, the intrinsic properties of gold are ensured to be consistent with its inherent physical characteristics, eliminating adulteration or counterfeiting. Therefore, this invention improves the accuracy, security, and efficiency of gold anti-counterfeiting by combining microtexture images, near-infrared spectral data, dynamic encryption tags, and blockchain technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 A flowchart of the steps of an online real-time image recognition anti-counterfeiting verification method based on gold transactions;

[0061] Figure 2 for Figure 1 Detailed implementation steps of step S1 in FIG.

[0062] Figure 3 for Figure 1 Detailed implementation steps of step S4 in FIG.

[0063] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0064] The following is a clear and complete description of the technical method of the present invention in conjunction with the accompanying drawings. It is obvious that the embodiments described are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative efforts are within the scope of protection of the present invention.

[0065] In addition, the accompanying drawings are merely schematic illustrations of the present invention and are not necessarily drawn to scale. Identical reference numerals in the figures denote identical or similar parts, and thus repetitive descriptions thereof will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities that do not necessarily correspond to physically or logically separate entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor and / or microcontroller approaches.

[0066] It should be understood that although the terms "first," "second," and the like may be used herein to describe various elements, these elements should not be limited by these terms. These terms are used solely to distinguish one element from another. For example, a first element may be referred to as a second element, and similarly, a second element may be referred to as a first element, without departing from the scope of the exemplary embodiments. The term "and / or" as used herein includes any and all combinations of one or more of the listed associated items.

[0067] To achieve this, please refer to Figures 1 to 3 , an online real-time image recognition anti-counterfeiting verification method based on gold transactions, the method comprising the following steps:

[0068] Step S1: Collect a micro-texture image of the gold product; generate a unique digital fingerprint based on the physical unclonable characteristics of the micro-texture image, and encrypt it with the UTC timestamp to generate a dynamic encrypted tag;

[0069] Step S2: extract the feature descriptor of the micro texture image and compare it with the pre-stored database to obtain the texture matching degree; decrypt the dynamic encryption tag and generate the dynamic code time status;

[0070] Step S3: Perform anti-counterfeiting identification on the micro-texture image of the gold product based on the texture matching degree and the aging status of the dynamic code to obtain the authenticity identification results and the counterfeit identification results; store the counterfeit identification results in the blockchain for evidence recording, and generate a genuine gold verification storage record;

[0071] Step S4: Based on the authenticity identification result, a multi-level review mechanism is initiated, and the micro-texture image of the gold product is again subjected to material intrinsic property anti-counterfeiting identification through the multi-level review mechanism. A secondary anti-counterfeiting identification result is obtained and synchronously stored in the blockchain, generating a secondary gold anti-counterfeiting identification storage record.

[0072] This invention uses a dynamic encrypted tag generated by encrypting a digital fingerprint generated from a microtexture image with a UTC timestamp to ensure that each gold product has a unique and unforgeable identity. This encrypted tag dynamically changes over time, making each anti-counterfeiting tag unique and authentic, enhancing the security of gold products. By extracting feature descriptors from the microtexture image and comparing them with a database, the texture characteristics of the gold product can be accurately identified. Combined with the detection of anomalous absorption peaks in near-infrared spectral data, spectral deviations can be effectively identified, further improving the accuracy of anti-counterfeiting identification. The results of authentic gold identification are stored in the blockchain for evidence, ensuring data immutability and providing consumers with a traceable verification record, enhancing transparency and market trust in gold products. A multi-level verification mechanism is activated to verify authenticity, reducing the possibility of counterfeiting by re-identifying the intrinsic properties of the material. This mechanism ensures multiple safeguards in the identification process and improves the reliability of the anti-counterfeiting technology. By combining texture features, spectral data, and the dynamic encrypted tag, this multi-layered anti-counterfeiting approach ensures that gold products are highly resistant to counterfeiting in all aspects and is adaptable to various counterfeiting methods. Therefore, the present invention improves the accuracy, security and efficiency of gold anti-counterfeiting by combining micro-texture images, near-infrared spectral data, dynamic encryption tags and blockchain technology.

[0073] In the embodiment of the present invention, reference Figure 1 FIG. 1 is a flow chart showing the steps of an online real-time image recognition and anti-counterfeiting verification method based on gold transactions according to the present invention. In this example, the online real-time image recognition and anti-counterfeiting verification method based on gold transactions includes the following steps:

[0074] Step S1: Collect a micro-texture image of the gold product; generate a unique digital fingerprint based on the physical unclonable characteristics of the micro-texture image, and encrypt it with the UTC timestamp to generate a dynamic encrypted tag;

[0075] In this embodiment of the present invention, a high-precision microscopic imaging device with a resolution of at least 5000 DPI (such as a confocal microscope or electron microscope) is used to acquire microtexture images of the gold surface. Spectral data is collected using a near-infrared (NIR) spectrometer with a wavelength range of 800-2500 nm to ensure the integrity of the texture and material property data. To reduce environmental interference, the temperature is controlled at 22±1°C and the humidity at 50%±5% during the acquisition process. The equipment is calibrated using a standard gold sample, and the average value is calculated from at least 10 repeated measurements to improve data consistency. Subsequently, feature extraction is performed on the microtexture images using CNN and ResNet deep learning models. 200-500 key points are detected using SIFT, ORB, or SURF algorithms. After deduplication and dimensionality reduction, a unique digital fingerprint is generated. This fingerprint data is converted into a 64-byte hash value using the SHA-256 hash algorithm and stored in binary format to improve storage and transmission efficiency. A UTC timestamp, accurate to 1 millisecond, is then obtained from a standard Network Time Protocol (NTP) server. This is then combined with the digital fingerprint and encrypted using the AES-256 symmetric encryption algorithm, with key management utilizing 256-bit ECC (Elliptic Curve Cryptography) or 2048-bit RSA. Finally, the generated dynamic encryption tag is stored as a 300DPI QR code, NFC tag, or RFID tag, attached to the gold artifact, and uploaded to a blockchain or cloud database for evidence storage. To verify system stability, microtexture data is repeatedly collected under varying lighting conditions (500-1500 lux), temperatures (10-40°C), and angles (0-90°, sampled every 10°). Hash collision tests are performed to ensure that fingerprint data is not duplicated due to similar samples. Database comparisons are also used to verify the decryptability of the encrypted tag and the fingerprint match, ensuring system stability and accuracy.

[0076] Step S2: extract the feature descriptor of the micro texture image and compare it with the pre-stored database to obtain the texture matching degree; decrypt the dynamic encryption tag and generate the dynamic code time status;

[0077] In this embodiment of the present invention, 200-500 key points are extracted from the collected microtexture image using the SIFT, ORB, and SURF algorithms. Feature descriptors are then calculated to generate a 128-dimensional local feature vector. Subsequently, the extracted feature descriptors are compared with standard texture fingerprint data in a pre-stored database using the FLANN (Fast Nearest Neighbor Search) or BF (Brute Force Matching) algorithm. The Euclidean or Hamming distance is calculated, and the degree of match is measured using cosine similarity. The match score range is set between 0 and 100%, with a threshold of 85%. When the match exceeds the threshold, the gold product is identified as identical. Subsequently, the near-infrared spectral data (800-2500nm band) of the gold product is tested for anomalous absorption peaks. The spectrum is smoothed using the Savitzky-Golay filter method, and the first and second derivatives are calculated to screen for anomalous absorption peaks in the spectral curve. Based on the K-Sigma method, an absorption peak anomaly detection threshold of μ±3σ (mean ± 3 times the standard deviation) is set to identify spectral deviations outside the normal range and calculate the spectral deviation value ΔA. For ΔA > 0.05 (absorbance units: AU), the cause of the deviation is further analyzed and compared with a standard spectral library to detect changes in material composition. During the dynamic encryption tag decryption process, the encrypted data from the QR code, NFC, or RFID tag is first read and decrypted using the AES-256 symmetric decryption algorithm combined with a pre-stored key. The resulting SHA-256 hash fingerprint and UTC timestamp are then parsed. The timestamp is verified to be within the ±5-minute tolerance range, and the decrypted hash value is compared with the original fingerprint stored in the database to ensure fingerprint consistency. Finally, based on the decryption result, the dynamic code validity status is generated and annotated with decryption success or failure status information. In the event of a decryption failure, the error type (including timestamp expiration, fingerprint mismatch, or data corruption) is recorded for further analysis.

[0078] Step S3: Perform anti-counterfeiting identification on the micro-texture image of the gold product based on the texture matching degree and the aging status of the dynamic code to obtain the authenticity identification results and the counterfeit identification results; store the counterfeit identification results in the blockchain for evidence recording, and generate a genuine gold verification storage record;

[0079] In this embodiment of the present invention, based on the texture matching score obtained in step S2, if the matching score exceeds 85%, the texture images are considered highly similar, further verifying authenticity. If the matching score is less than 85%, the product is identified as a potential counterfeit and marked as "suspected counterfeit." Based on the spectral deviation value ΔA in step S2, if ΔA is less than 0.05 and no abnormal absorption peaks are detected, the spectral characteristics are considered normal and the gold product is authentic. If ΔA exceeds 0.05 or significant abnormal spectral absorption peaks are detected, the gold product is considered to have compositional deviations and is a counterfeit, and a warning is issued. Based on the decrypted dynamic code validity status in step S2, if the dynamic code timestamp is within the range of ±5 minutes, the dynamic tag is considered valid and the verification is valid. If the dynamic code timestamp exceeds the allowable deviation range, a "Dynamic code expired" message is displayed, indicating that the gold product is a counterfeit. A weighted average method is used to combine the texture matching score, spectral deviation value, and dynamic code validity to make the final anti-counterfeiting determination. A comprehensive score is calculated, with a matching weight of 40%, a spectral deviation weight of 30%, and a dynamic code timeliness weight of 30%. A comprehensive score of 80% or higher is considered authentic, while a score below 80% is considered counterfeit. If the comprehensive score is 80% or higher, an authenticity determination result is generated and marked as "authentic." If the comprehensive score is less than 80%, a counterfeit determination result is generated and marked as "counterfeit." For gold products determined to be authentic, the system automatically stores the authenticity determination result on the blockchain for evidence recording. Once stored on the blockchain, a verification record of the authenticity of the gold is generated, ensuring that the data cannot be tampered with and providing a reliable basis for subsequent traceability.

[0080] Step S4: Based on the authenticity identification result, a multi-level review mechanism is initiated, and the micro-texture image of the gold product is again subjected to material intrinsic property anti-counterfeiting identification through the multi-level review mechanism. A secondary anti-counterfeiting identification result is obtained and synchronously stored in the blockchain, generating a secondary gold anti-counterfeiting identification storage record.

[0081] In an embodiment of the present invention, during the anti-counterfeiting verification process for gold products, once a preliminary determination of authenticity is made, the system automatically initiates a multi-level review mechanism for further secondary anti-counterfeiting verification. Once a preliminary determination of authenticity is made, the system automatically triggers the multi-level review mechanism. The review process is divided into three levels: First-level review (optional): The system automatically analyzes and compares data (including texture matching, spectral deviation, and dynamic code validity) for further verification. Second-level review: A manual auditor conducts physical sample analysis of the gold product, further collects microtexture images through equipment, reacquires near-infrared spectral data, and compares this with previous data for in-depth analysis. Third-level review: A comprehensive appraisal is conducted by a third-party certification agency, including physical and chemical property verification such as composition analysis and atomic spectroscopy, ensuring that all anti-counterfeiting clues are verified. During this multi-level review process, the system uses the gold product's microtexture images and near-infrared spectral data to conduct anti-counterfeiting verification based on the material's intrinsic properties. Microtexture images of the gold surface are reacquired using a high-precision microscope, and deep feature extraction is performed using the latest image recognition algorithms (such as convolutional neural networks). The system compares the gold product to a pre-stored database of standard gold samples and calculates similarity. If the similarity exceeds a set threshold (e.g., 90%), the gold product is determined to meet the intrinsic properties. Near-infrared spectral data is recollected and compared to the standard spectral data using statistical analysis methods such as PLS-DA (Partial Least Squares Discriminant Analysis) to determine whether it matches the spectral characteristics of gold. If the spectral data matches the standard data, the product is confirmed to be gold; otherwise, it is a counterfeit. Based on the analysis results of the multi-level verification mechanism, the system conducts a secondary anti-counterfeiting verification on the gold product. If the product is confirmed to be authentic at all verification levels, a "Secondary Anti-Counterfeiting Verification Result" is generated and marked as "Authentic." If the product is confirmed to be a counterfeit at any level of verification, a "Secondary Anti-Counterfeiting Verification Result" is generated and marked as "Counterfeit." For gold products that are secondarily determined to be authentic, the system stores the secondary anti-counterfeiting verification results on the blockchain for evidence. At the same time, a secondary gold anti-counterfeiting identification storage record is generated and uploaded to the blockchain database in an unalterable manner, providing permanent verifiable credentials for the authenticity of gold products.

[0082] As an example of the present invention, refer to Figure 2 As shown, in this example, step S1 includes:

[0083] Step S11: collecting near-infrared spectrum data of the gold product using a near-infrared spectrometer; collecting an original microscopic image of the surface of the gold product using an industrial microscope, and performing denoising and contrast enhancement on the image to generate a microscopic enhanced image;

[0084] Step S12: performing lattice structure segmentation on the microscopic enhanced image to generate a lattice distribution binary mask;

[0085] Step S13: extracting the pore area in the mask, and counting the number, spacing and shape factor of the pores in the pore area to generate a pore feature vector;

[0086] Step S14: fusing the lattice distribution binary mask and the pore feature vector and marking them as physical unclonable feature data; generating a digital fingerprint of the gold product based on the physical unclonable feature data to obtain a unique digital fingerprint;

[0087] Step S15: Perform AES-256 encryption on the unique digital fingerprint in combination with the UTC timestamp to generate a dynamic encryption tag.

[0088] In this embodiment of the present invention, a gold product was scanned using a near-infrared spectrometer (800-2500nm band) to collect near-infrared spectral data. To ensure data accuracy, the spectral resolution was maintained at 1nm, the data acquisition time was 5 seconds, and spectral intensity correction was performed using the relative reflectance method. This method allows the spectral data to reflect the intrinsic chemical composition and molecular structure of the gold material. The gold surface was scanned using an industrial microscope (1000x magnification) to obtain a raw microscopic image. To enhance image usability, the image was first denoised using a Gaussian filter to remove noise, and histogram equalization was used to enhance image contrast, resulting in a clear microscopically enhanced image. This image clearly reveals subtle differences in the gold surface structure. Lattice structure segmentation was then performed on the microscopically enhanced image, using an adaptive threshold segmentation algorithm to separate the lattice structure from the amorphous portion of the image. A binary mask of the lattice distribution was generated through calculation, with the lattice structure accounting for approximately 70% of the mask. This mask was used to further analyze the microscopic features of the gold surface with high accuracy and clarity. Based on the lattice distribution binary mask, the pore region within the mask is extracted, and morphological operations (such as dilation and erosion) are used to further highlight the pore features. The number, spacing, and shape factor of the pores in the pore region are calculated. For example, if the number of pores is 100 per square millimeter, the pore spacing is 0.3 mm, and the pores have a shape factor greater than 0.8, approximately 60% of the total pores. This pore feature data generates a pore feature vector, which reflects the microscopic porosity of the gold surface material. The lattice distribution binary mask and the pore feature vector are combined to generate a unique dataset containing physically unclonable feature data. This dataset not only contains information on the gold surface microtexture but also incorporates the porosity characteristics of the gold material, representing its unique physical fingerprint that cannot be replicated on other gold products. Based on this physically unclonable feature data, the data is digitally fingerprinted using the SHA-256 hash algorithm to generate a unique digital fingerprint. This digital fingerprint serves as a unique identifier for the gold product. The digital fingerprint is encrypted using the AES-256 symmetric encryption algorithm, combined with the current UTC timestamp (for example, 15:00:00, March 18, 2025). This encryption generates a dynamic encrypted tag that includes the timestamp information. Each decrypted tag is unique and unalterable. This tag plays a crucial role in the anti-counterfeiting and verification of gold products, ensuring a fresh encrypted identity each time it is verified, preventing counterfeiting and tampering.

[0089] It is particularly important that step S12 further includes the following steps:

[0090] Step S121: performing lattice edge feature enhancement on the micro-enhanced image to generate edge enhancement data; performing lattice boundary detection on the micro-enhanced image based on the edge enhancement data to obtain boundary feature data;

[0091] Step S122: performing skeleton extraction on the micro-enhanced image using the boundary feature data to obtain lattice skeleton data; performing lattice morphological analysis on the micro-enhanced image using the lattice skeleton data to generate morphological feature data;

[0092] Step S123: extracting lattice texture features to obtain lattice texture feature data; performing regional lattice growth segmentation on the lattice skeleton data according to the lattice texture feature data to generate initial segmentation data;

[0093] Step S124: performing lattice contour correction on the initial segmentation data to obtain contour optimization data, wherein the lattice contour correction includes curve fitting, contour smoothing, defect repair and boundary refinement; constructing a mask based on the contour optimization data to obtain a lattice distribution binary mask.

[0094] In an embodiment of the present invention, lattice edge features are enhanced in a microscopically enhanced image using an edge detection algorithm (such as the Sobel operator). The Sobel operator calculates the image's gradient to identify changing edge regions, thereby highlighting the edge information of the lattice structure. The edge-enhanced image more clearly displays the lattice's boundaries and details. Next, lattice boundary detection is performed on the microscopically enhanced image using the edge-enhanced data. Lattice boundary feature data is extracted through methods such as threshold segmentation and contour tracing, providing clear boundary information for subsequent image analysis. After obtaining the lattice boundary features, skeleton extraction is further performed on the image. Skeleton extraction uses a thinning algorithm (such as the Zhang-Suen thinning algorithm) to simplify the lattice region in the image into a single-pixel-width skeleton structure, extracting the lattice skeleton data. This skeleton data can more intuitively represent the main structural features of the lattice. Subsequently, morphological analysis is performed on the microscopically enhanced image using the lattice skeleton data to analyze the lattice's morphological features, such as shape, symmetry, and pore structure. This process uses morphological operations such as erosion, dilation, and opening and closing to generate morphological feature data that further describes the geometric characteristics of the lattice. Based on the skeleton data and morphological analysis, texture feature extraction algorithms (such as the gray-level co-occurrence matrix (GLCM) or local binary pattern (LBP)) are used to extract the lattice's texture feature data. These features, including texture roughness, contrast, and directionality, reflect the microstructural properties of the lattice region. Using these texture feature data, a region growing algorithm is used to segment the lattice skeleton data. This algorithm expands from a seed point and merges similar regions based on texture features to generate initial lattice segmentation data. After obtaining this initial segmentation data, curve fitting techniques are used to refine the contours, smoothing irregular boundaries for a more accurate contour. Contour smoothing, such as using B-spline curves, removes noise and artifacts. Furthermore, defect repair techniques are used to fill in missing or damaged areas in the image to ensure the integrity of the lattice contour. Boundary refinement uses an iterative algorithm to refine contour details for greater accuracy. After lattice profile optimization, binarization is used to convert the optimized profile data into a lattice distribution binary mask. This process clearly distinguishes crystalline and amorphous regions, setting pixel values ​​in crystalline regions to 1 and those in amorphous regions to 0. The resulting lattice distribution binary mask accurately reflects the microscopic lattice distribution characteristics of gold products, providing an accurate data foundation for subsequent anti-counterfeiting analysis.

[0095] Preferably, step S14 includes the following steps:

[0096] Step S141: fusing the lattice distribution binary mask with the pore feature vector and marking them as physical unclonable feature data;

[0097] Step S142: extracting the eigenvectors of the physical unclonable feature data, and performing high-dimensional mapping conversion on the eigenvectors to generate high-dimensional feature matrix data;

[0098] Step S143: performing hash coding calculation on the high-dimensional feature matrix data to generate unique hash fingerprint data;

[0099] Step S144: performing error correction processing on the unique hash fingerprint data to generate a unique digital fingerprint.

[0100] In this embodiment of the present invention, a previously generated lattice distribution mask and pore feature vectors are fused using a weighted averaging method. Each pixel in the mask is combined with the corresponding pore feature value (such as pore number, spacing, and shape factor) to generate a physically unclonable feature dataset. These features, including microtexture and pore characteristics, are unique and highly discriminative. This fused dataset is professionally labeled as "physically unclonable feature data," which serves as the core basis for anti-counterfeiting of gold products. A high-precision algorithm is used to extract key eigenvectors from the physically unclonable feature data. Each eigenvector includes texture information, pore information, and their spatial relationships. These eigenvectors are then transformed using high-dimensional mapping techniques such as principal component analysis (PCA) or t-SNE (t-distributed stochastic neighbor embedding) to generate a high-dimensional feature matrix. This matrix represents the high-dimensional data space of the gold product's microstructure, with each data point representing a unique characteristic of the gold. Assuming the high-dimensional matrix has dimensions of 1024×1024, with 1000 feature points per dimension, these feature matrices can accurately reflect the original material properties of the gold product. The high-dimensional feature matrix data is input into a hash algorithm (such as SHA-512) to perform a hash code calculation. The hash algorithm compresses the high-dimensional feature matrix into a fixed-length hash value. This hash value serves as the unique fingerprint of the gold item. Regardless of changes in the input data, the hash value remains unique and difficult to reverse engineer. Assuming a 512-bit hash value, each generated hash fingerprint is highly unique and cannot be copied or forged. After the hash code calculation, error correction is performed to ensure the accuracy and stability of the fingerprint data. The generated hash fingerprint is encoded using an error correction algorithm (such as Hamming coding or Reed-Solomon algorithm) to eliminate computational errors and external interference. After error correction, the resulting unique digital fingerprint data is highly fault-tolerant and resistant to interference, ensuring that each decrypted digital fingerprint is unique and accurate. Ultimately, this digital fingerprint serves as the identity of the gold item and is stored on the blockchain along with all historical transaction records and stored information, ensuring permanent preservation and tamper-proofing.

[0101] Preferably, extracting the feature descriptor of the micro texture image and performing texture comparison with a pre-stored database in step S2 includes:

[0102] Extract the surface microstructure trend features of the microtexture image, analyze the microtexture image for rotation invariance, and obtain local rotation features;

[0103] Locate texture key points of micro texture images based on surface micro structure orientation features and local rotation features;

[0104] Calculate local feature descriptors for each texture keypoint;

[0105] Import the feature descriptors into the pre-stored database to perform feature descriptor comparison and obtain descriptor matching pairs;

[0106] The similarity of the descriptor matching pairs is calculated, and the descriptor matching pairs with the highest similarity and the descriptor matching pairs with the lowest similarity are screened out for weighted average processing to generate the texture matching degree.

[0107] In an embodiment of the present invention, surface microstructure orientation features are extracted from a microtexture image using an image processing algorithm (such as the gray-level co-occurrence matrix method (GLCM) or wavelet transform). These microstructure orientation features can describe the primary directionality of the surface texture and reflect the orientation of the lattice structure. Next, a rotation-invariance analysis is performed on the image using rotation-invariance techniques (such as the scale-invariant feature transform (SIFT) or the speeded-up robust features (SURF) algorithm) to obtain local rotation features. This step ensures that the texture features remain stable under rotation, ensuring consistency across different angles and making them suitable for subsequent feature matching. After obtaining the surface microstructure orientation features and local rotation features, a feature point localization algorithm (such as Harris corner detection or FAST corner detection) is used to determine texture keypoints in the microtexture image. By combining surface orientation and rotation features, these keypoints can accurately describe the local characteristics of the texture, particularly maintaining consistency across different rotation angles, thereby enhancing the reliability of texture recognition. Once the texture keypoints are located, a descriptor generation algorithm (such as SIFT, SURF, or ORB) is used to calculate a local feature descriptor for each keypoint. These descriptors are high-dimensional vectors generated from texture information in the surrounding area, accurately representing the texture features of a local region. Each descriptor contains image information related to its corresponding texture keypoint, effectively distinguishing different texture regions. The extracted feature descriptors are compared with texture descriptors stored in a database. This comparison utilizes a feature matching algorithm, such as brute force matching or matching based on the FLANN (Fast Approximate Nearest Neighbor) algorithm. By calculating the distance between feature descriptors (typically using Euclidean distance or Hamming distance), one-to-one descriptor matching pairs are obtained. Each pair of matching descriptors represents a region with similar textures in the image. For each matching pair, the similarity is calculated, typically using metrics such as cosine similarity or normalized mutual information. The higher the similarity, the more similar the two texture regions. The descriptor matching pairs with the highest and lowest similarity are then selected. The features of these two extreme matching pairs are important for assessing texture similarity. The similarity values ​​of these matching pairs are then combined using a weighted average to generate the final texture matching score. The weight of the weighted mean can be adjusted according to the range of similarity and the reliability of the texture area to ensure that the calculation result accurately reflects the texture similarity.

[0108] Preferably, step S2 further includes:

[0109] Perform spectral peak positioning on near-infrared spectral data to generate spectral peak position data;

[0110] Calculate the peak height, full width at half maximum and peak area of ​​the spectral peak position data respectively to obtain the spectral peak position characteristic data set;

[0111] Perform peak shape analysis on the spectral peak position data based on the spectral peak position characteristic data set to generate absorption peak morphology data;

[0112] Utilize the spectral peak position characteristic data set and absorption peak morphology data to detect and screen abnormal absorption peaks in near-infrared spectral data, generating abnormal absorption peak detection data and normal absorption peak screening data;

[0113] Perform spectral deviation analysis on abnormal absorption peak detection data and normal absorption peak screening data to generate spectral deviation values.

[0114] In embodiments of the present invention, near-infrared spectral data is peak located using a spectral peak detection algorithm (such as one based on the second-order derivative method, the maximum derivative method, or an adaptive threshold). These algorithms can accurately identify absorption peaks in the spectral data and determine their locations. The spectral peak position data generated in this step contains the wavelength or frequency information of each absorption peak, serving as the basis for subsequent analysis. Based on the spectral peak position data, standard spectral analysis methods are used to calculate the peak height, full-width at half-maximum (FWHM), and peak area of ​​each spectral peak. These characteristic data further describe the morphological characteristics of each absorption peak. The peak height represents the maximum value of the absorption peak and reflects the magnitude of the absorption intensity in the spectrum; the full-width at half-maximum (FWHM) represents the width of the absorption peak and reflects the distribution range of the absorption peak; and the peak area represents the area under the absorption peak and reflects the light absorption intensity within a specific wavelength band. Calculating these characteristics provides detailed data support for subsequent peak shape analysis. The morphological characteristics of each absorption peak can be further extracted by performing peak shape analysis on the spectral peak position feature dataset (e.g., using Gaussian fitting, Lorentz fitting, etc.). These morphological features include the shape, symmetry, and goodness of fit of the absorption peak. During the analysis process, peak shape denoising is also required to reduce the impact of noise on peak shape. The resulting absorption peak morphology data helps identify anomalous absorption peaks. Using the spectral peak position feature dataset and absorption peak morphology data, anomaly detection algorithms (such as those based on statistical distribution models or machine learning-based anomaly detection models) are used to detect anomalous absorption peaks in the spectral data. This step allows for the identification of anomalous absorption peaks by setting thresholds (e.g., thresholds for peak height, peak width, goodness of fit, and other parameters). Anomalous absorption peaks typically manifest as peaks whose waveforms do not conform to the normal absorption peak morphology. Finally, through screening, normal and anomalous absorption peaks are distinguished, generating anomalous absorption peak detection data and normal absorption peak screening data. Spectral deviation analysis is performed on the anomalous absorption peak detection data and normal absorption peak screening data. Spectral deviation values ​​are generated by calculating the deviations between normal and anomalous absorption peaks (e.g., peak shift, spectral intensity difference, frequency drift, etc.).

[0115] Preferably, decrypting the dynamic encryption tag in step S2 includes:

[0116] parsing the dynamic parameters of the dynamic encryption tag to perform hierarchical decoding of the tag structure of the dynamic encryption tag, thereby obtaining a hierarchical encrypted data unit;

[0117] Derivation of time-varying keys based on layered encrypted data units to obtain a dynamic decryption key sequence;

[0118] Reverse mapping the dynamic encryption tag according to the dynamic decryption key sequence to obtain the dynamic decryption intermediate state data;

[0119] Perform timeliness state synchronization and noise elimination on the dynamic decryption intermediate state data to obtain dynamic code timeliness state data;

[0120] By integrating the space-time dimension of the dynamic code aging status data, the dynamic code aging status is finally output.

[0121] In embodiments of the present invention, dynamic parameters of a dynamic encryption tag are parsed using a tag structure parsing algorithm (e.g., a data structure-based hierarchical decoding method). These parameters typically include a timestamp, encryption identifier, encryption algorithm type, and related encryption key information. The tag structure is organized as multi-layered data units, and information about each encrypted data unit can be obtained by decoding each layer. Hierarchical decoding of the tag structure extracts hierarchical encrypted data units related to time, space, and encryption type. A time-varying key derivation algorithm (e.g., an encryption algorithm based on time series data and a key update mechanism) is used to derive a key sequence. These time-varying keys typically contain key information for each moment in the time series, and based on the rules of the encryption algorithm, the keys may change at different time points. Specifically, assuming the timestamp data is T1 = 1632680400 and T2 = 1632680500, key derivation generates a new key sequence based on the changes in each time period, such as K1 = 0x4F7B9D2C and K2 = 0x3A8D0B5F, to ensure the timeliness of the decryption process. Using the derived dynamic decryption key sequence, the dynamic encrypted tag is reverse mapped (for example, using an XOR algorithm, DES reverse mapping, or AES reverse decryption techniques). This process combines the ciphertext data in the encrypted tag with the key sequence to gradually restore the tag's intermediate decrypted state. For example, decrypting encrypted data C using key sequences K1 and K2 yields intermediate data D_intermediate = 0x9F4C8A3E, representing an intermediate state of the tag decryption. After obtaining the dynamic decrypted intermediate state data, timeliness state synchronization is performed. This step compares timeliness factors (such as system time T_sync = 1632680700) with the decrypted data to correct for time deviations during the decryption process. Furthermore, noise reduction algorithms (such as Kalman filters and smoothing algorithms) are applied to remove noise from the data, improving data accuracy. Ultimately, the processed dynamic decrypted intermediate state data becomes the dynamic code timeliness state data, such as T_state = 0x01A2B3C4. Finally, the dynamic code aging state data is fused in the spatial and temporal dimensions (using an algorithm based on a spatial-temporal correlation model), combining temporal and spatial information to generate the final dynamic code aging state. For example, the time series information T_sync = 1632680700 can be combined with the device location data (X = 23.5, Y = 45.6) to obtain the final dynamic code aging state C_final = 0xB7C9D3F8 using a spatiotemporal data fusion algorithm (such as a Kalman filter or convolutional neural network).

[0122] Preferably, the anti-counterfeiting determination of the micro texture image of the gold product according to the texture matching degree and the aging state of the dynamic code in step S3 includes:

[0123] The gold product's micro-texture image is subjected to anti-counterfeiting judgment based on texture matching, spectral deviation value and dynamic code aging status. When the following conditions are present at the same time, it is judged as authentic: texture matching is greater than or equal to 0.90 and less than or equal to 1.00, spectral deviation is less than or equal to 0.05, and dynamic code aging is greater than or equal to 90% and less than or equal to 100%;

[0124] When the following conditions occur at the same time, it is judged as a counterfeit identification result: the texture matching degree is less than 0.90 and greater than or equal to 0.80, the spectral deviation is greater than 0.05 and less than or equal to 0.15, and the dynamic code timeliness is less than 90% and greater than or equal to 70%.

[0125] As an example of the present invention, refer to Figure 3 As shown, in this example, step S4 includes:

[0126] Step S41: activating a multi-level review mechanism based on the authenticity identification result, wherein the multi-level review mechanism includes manual review and weight verification;

[0127] Step S42: performing microstructural stress response analysis on the gold product through a multi-level review mechanism to generate microstructural stress response data;

[0128] Step S43: performing lattice resonance frequency detection on the gold product according to the microstructure stress response data to obtain resonance frequency characteristic data;

[0129] Step S44: Use the resonance frequency characteristic data and microstructure stress response data to perform material intrinsic property anti-counterfeiting judgment on the microtexture image of the gold product, obtain a secondary anti-counterfeiting judgment result and synchronously store it in the blockchain, generating a secondary gold anti-counterfeiting judgment storage record.

[0130] In this embodiment of the present invention, after a gold product undergoes preliminary authenticity verification, the system automatically initiates a multi-level verification mechanism. First, a manual review step involves human experts verifying the authenticity verification results generated by the system. This step includes a manual inspection of the gold product's appearance, weight, and other visible features to ensure the accuracy of the initial verification. Second, a weight verification step involves measuring the gold product's weight using a high-precision electronic balance (with an accuracy of up to 0.0001 gram) and comparing it to a standard weight for gold products to ensure compliance. If the product's weight is within the allowable error range, the process proceeds to the next step. If there is a significant error, the product is deemed counterfeit and further verification is discontinued. Following this multi-level verification mechanism, the gold product enters the microstructural stress response analysis phase. Piezoelectric stress sensors (with an accuracy of ±1%) are used to apply controllable pressure or stress to the gold product, simulating the mechanical response in a real-world environment. The sensors measure the gold product's response signals in real time after the stress is applied, capturing the stress distribution at the microscopic level. Using stress analysis algorithms (such as finite element analysis), microstructural stress response data is generated to characterize the gold product's microstructural response characteristics under varying stresses. Based on data obtained from microstructural stress response analysis, a laser resonance frequency tester (with an accuracy of 0.01Hz) was used to measure the lattice resonance frequency of gold products. The equipment measures the resonance response of gold products at different frequencies to extract the resonance frequency characteristics of the gold material. The microscopic lattice structure of gold products produces specific resonance frequencies, which reflect its intrinsic physical properties, such as the lattice's elastic modulus and hardness. Resonance frequency signature data is generated by comparing the resonance frequency range with that of standard gold products. After obtaining the resonance frequency signature data and microstructural stress response data, the system combines the microtexture image of the gold product to perform anti-counterfeiting verification based on its intrinsic material properties. First, a specialized anti-counterfeiting algorithm integrates the resonance frequency signature data, microstructural stress response data, and texture image features to generate a comprehensive anti-counterfeiting model. This model identifies the unique physical properties and microstructure of the gold product and compares it with authentic reference data in a database to generate a secondary anti-counterfeiting verification result. If the gold product meets the authenticity standards, it is marked as authentic; if not, it is classified as a counterfeit. Finally, the system stores the secondary anti-counterfeiting identification results and related data (such as resonance frequency, microstructure stress response data, etc.) in the blockchain to ensure the non-tamperability of the identification record, generate a valid gold anti-counterfeiting identification storage record, and provide data basis for subsequent traceability.

[0131] It is particularly important that step S43 further includes the following steps:

[0132] Step S43: performing lattice resonance frequency detection on the gold product according to the microstructure stress response data to obtain resonance frequency characteristic data;

[0133] Step S431: performing time-frequency decomposition on the microstructure stress response data to obtain time-frequency characteristic data; performing stress wave propagation characteristic analysis on the microstructure stress response data based on the time-frequency characteristic data to generate wave propagation data;

[0134] Step S432: extracting stress response modal features of the microstructure stress response data through the wave propagation data, and constructing a lattice vibration model to generate vibration model data; calculating the lattice elastic constants of the vibration model data, and analyzing the thermal vibration characteristics of the vibration model data through the lattice elastic constants to obtain thermal vibration data;

[0135] Step S433: performing a sweep frequency excitation test on the gold product based on the thermal vibration data to obtain frequency response data; performing resonance peak identification on the frequency response data to generate resonance frequency characteristic data.

[0136] In this embodiment of the present invention, time-frequency analysis is performed on the microstructural stress response data of a gold product. Assume that the microstructural stress response data has been collected using a high-precision stress sensor (such as a piezoelectric sensor) with a sampling rate of 10 kHz and a total data length of 5 seconds. Wavelet transform is used to analyze the time-frequency characteristics, selecting the Daubechies 4 (db4) wavelet as the base wavelet and selecting an appropriate number of decomposition layers (e.g., 4) to obtain time-frequency characteristic data for each decomposition layer. A time-frequency spectrum is extracted from the wavelet transform data for each layer, from which the frequency distribution within each time window can be obtained. This allows the time-frequency distribution of the microstructural stress response waves of the gold product to be observed. For example, the time-frequency analysis shows that the gold product has the strongest response in the frequency range of 2 kHz to 6 kHz, with significant amplitudes at 2.5 kHz and 4 kHz. Based on this time-frequency characteristic data, the propagation pattern of the stress wave in the gold product is analyzed. The finite difference method (FDM) or the transfer matrix method (TMM) is used to simulate stress wave propagation, generating wave propagation data for the gold product's microstructure. For example, a wave propagation velocity of 1500 m / s and a wave attenuation coefficient of 0.5 dB / cm can be determined. The main modal characteristics of the gold product's stress response are extracted from the wave propagation data. These modal characteristics typically manifest as significant responses at specific frequencies. For example, at 3 kHz and 5 kHz, the gold product exhibits strong resonances, which the wave propagation data indicates are resonant frequencies of the gold product's lattice structure. Finite element analysis (FEA) is used to model the gold product, assuming an FCC (face-centered cubic) lattice structure. In this model, the material elastic modulus is set to 120 GPa, the Poisson's ratio is set to 0.3, and the lattice mesh size is set to 0.01 mm. A vibration analysis is performed on the model using a solver such as ANSYS to calculate its natural frequencies. The calculations reveal significant modal responses at 3.2 kHz and 5.1 kHz. During this process, simulations were used to calculate the elastic constants of the gold lattice. Assuming a typical elastic constant of 50 GPa, the lattice stress response characteristics of gold were calculated, yielding the thermal vibration characteristics. This data provides the basis for subsequent thermal vibration analysis. Experimental testing of the gold was conducted using a swept-frequency excitation method. In the experiment, the excitation signal was swept from 1 kHz to 10 kHz, with a sampling frequency of 10 kHz and a sweep step of 1 Hz. A piezoelectric exciter was used to apply the excitation signal to the gold product, and the response at each frequency was recorded. For example, during the sweep process, the frequency was started at 1 kHz and increased by 1 Hz until it reached 10 kHz. At each frequency, the frequency response amplitude and phase were recorded. Under swept-frequency excitation, a set of frequency response data was obtained.Assuming the frequency response amplitude at 3.1kHz is -25dB, and the amplitude at 5.0kHz is -18dB, this indicates that the stress response of gold products is more pronounced at these frequencies. Based on the frequency response data, a peak detection algorithm (such as a peak search algorithm) is used to identify the resonant peaks in the frequency response. This resonant peak identification yields resonant frequency signature data. For example, 3.2kHz and 5.1kHz are identified as the two primary resonant frequencies of a gold product. These frequencies serve as unique physical characteristics of the gold product and are subsequently used for anti-counterfeiting assessments.

[0137] In this specification, an online real-time image recognition and anti-counterfeiting verification system based on gold transactions is provided, which is used to execute the above-mentioned online real-time image recognition and anti-counterfeiting verification method based on gold transactions. The online real-time image recognition and anti-counterfeiting verification system based on gold transactions includes:

[0138] The tag module is used to collect micro-texture images of gold products; based on the physical unclonable characteristics of the micro-texture image, a unique digital fingerprint is generated, which is encrypted with the UTC timestamp to generate a dynamic encrypted tag;

[0139] Feature analysis module is used to extract feature descriptors of micro texture images and compare textures with pre-stored databases to obtain texture matching degree; decrypt dynamic encryption tags and generate dynamic code aging status;

[0140] The primary anti-counterfeiting identification module is used to perform anti-counterfeiting identification on the micro-texture image of gold products based on texture matching and the aging status of the dynamic code to obtain the results of authenticity identification and counterfeit identification. The counterfeit identification results are stored in the blockchain for evidence recording, generating a primary verification storage record of authentic gold.

[0141] The secondary anti-counterfeiting identification module is used to initiate a multi-level review mechanism based on the authenticity identification results, and to conduct anti-counterfeiting identification of the material intrinsic properties of the micro-texture image of the gold product again through the multi-level review mechanism. The secondary anti-counterfeiting identification results are obtained and stored synchronously in the blockchain to generate a secondary gold anti-counterfeiting identification storage record.

[0142] The beneficial effect of this invention lies in generating a unique digital fingerprint based on the physically unclonable characteristics of microtexture images, ensuring that each gold product has a unique identification mark and preventing counterfeiting and tampering. Dynamic encryption, combined with a UTC timestamp, ensures that the digital fingerprint is unique at each point in time. Even if an attacker obtains current data, it cannot be used for future counterfeiting. By extracting microtexture features and comparing them with a pre-stored database, texture matching is calculated, effectively preventing surface counterfeiting methods such as gold plating and laser engraving. Detecting gold's specific absorption peaks and analyzing spectral deviations prevents tampering with adulteration or composition. Decrypting the tag reveals the dynamic code's aging status, ensuring the integrity and traceability of the gold's identity information. Combining texture matching, spectral deviation, and dynamic code aging status, the first round of anti-counterfeiting verification is performed to quickly identify authentic gold. Authentic gold information is stored on the blockchain, ensuring that its data cannot be tampered with and forming a record of the gold's authenticity. If the authenticity verification result is positive, a multi-level review mechanism is initiated, including manual re-inspection and weight verification, to further improve accuracy. Through microstructural stress response analysis and resonant frequency signature detection, the intrinsic properties of gold are ensured to be consistent with its inherent physical characteristics, eliminating adulteration or counterfeiting. Therefore, this invention improves the accuracy, security, and efficiency of gold anti-counterfeiting by combining microtexture images, near-infrared spectral data, dynamic encryption tags, and blockchain technology.

[0143] The present invention is therefore intended to be illustrative and non-restrictive in all respects, with the scope of the invention being defined by the appended claims rather than the foregoing description, and all changes that come within the meaning and range of equivalents of the application documents are intended to be embraced therein.

[0144] The foregoing description is intended only to provide specific embodiments of the present invention, which will enable those skilled in the art to understand and implement the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not intended to be limited to the embodiments shown herein, but is to be construed in the widest possible manner consistent with the principles and novel features disclosed herein.

Claims

1. An online real-time image recognition anti-counterfeiting verification method based on gold trading, characterized in that: The following steps are involved: Step S1: collecting micro texture images of gold products; A unique digital fingerprint is generated based on the physical unclonable features of the microtexture image, and is encrypted in combination with the UTC timestamp to generate a dynamic encrypted tag; wherein step S1 includes the following steps: Step S11: collecting near-infrared spectrum data of the gold product using a near-infrared spectrometer; collecting an original microscopic image of the surface of the gold product using an industrial microscope, and performing denoising and contrast enhancement on the image to generate a microscopic enhanced image; Step S12: performing lattice structure segmentation on the microscopic enhanced image to generate a lattice distribution binary mask; Step S13: extracting the pore area in the mask, and counting the number, spacing and shape factor of the pores in the pore area to generate a pore feature vector; Step S14: fusing the lattice distribution binary mask and the pore feature vector and marking them as physical unclonable feature data; generating a digital fingerprint of the gold product based on the physical unclonable feature data to obtain a unique digital fingerprint; Step S15: Perform AES-256 encryption on the unique digital fingerprint in combination with the UTC timestamp to generate a dynamic encryption tag; wherein step S14 includes the following steps: Step S141: fusing the lattice distribution binary mask with the pore feature vector and marking them as physical unclonable feature data; Step S142: extracting the eigenvectors of the physical unclonable feature data, and performing high-dimensional mapping conversion on the eigenvectors to generate high-dimensional feature matrix data; Step S143: performing hash coding calculation on the high-dimensional feature matrix data to generate unique hash fingerprint data; Step S144: performing error correction processing on the unique hash fingerprint data to generate a unique digital fingerprint; Step S2: extract the feature descriptor of the micro texture image and compare it with the pre-stored database to obtain the texture matching degree; decrypt the dynamic encryption tag and generate the dynamic code time status; Step S3: Perform anti-counterfeiting identification on the micro-texture image of the gold product based on the texture matching degree and the aging status of the dynamic code to obtain the authenticity identification results and the counterfeit identification results; store the counterfeit identification results in the blockchain for evidence recording, and generate a genuine gold verification storage record; Step S4: Based on the authenticity identification result, a multi-level review mechanism is initiated, and the micro-texture image of the gold product is again subjected to material intrinsic property anti-counterfeiting identification through the multi-level review mechanism. A secondary anti-counterfeiting identification result is obtained and synchronously stored in the blockchain, generating a secondary gold anti-counterfeiting identification storage record.

2. The online real-time image recognition anti-counterfeiting verification method based on gold transaction according to claim 1 is characterized in that: Extracting feature descriptors of the micro texture image and performing texture comparison with a pre-stored database in step S2 includes: Extract the surface microstructure trend features of the microtexture image, analyze the microtexture image for rotation invariance, and obtain local rotation features; Locate texture key points of micro texture images based on surface micro structure orientation features and local rotation features; Calculate local feature descriptors for each texture keypoint; Import the feature descriptors into the pre-stored database to perform feature descriptor comparison and obtain descriptor matching pairs; The similarity of the descriptor matching pairs is calculated, and the descriptor matching pairs with the highest similarity and the descriptor matching pairs with the lowest similarity are screened out for weighted average processing to generate the texture matching degree.

3. The online real-time image recognition anti-counterfeiting verification method based on gold transaction according to claim 1, characterized in that: Step S2 further includes: Perform spectral peak positioning on near-infrared spectral data to generate spectral peak position data; Calculate the peak height, full width at half maximum and peak area of ​​the spectral peak position data respectively to obtain the spectral peak position characteristic data set; Perform peak shape analysis on the spectral peak position data based on the spectral peak position characteristic data set to generate absorption peak morphology data; Utilize the spectral peak position characteristic data set and absorption peak morphology data to detect and screen abnormal absorption peaks in near-infrared spectral data, generating abnormal absorption peak detection data and normal absorption peak screening data; Perform spectral deviation analysis on abnormal absorption peak detection data and normal absorption peak screening data to generate spectral deviation values.

4. The online real-time image recognition anti-counterfeiting verification method based on gold transaction according to claim 1, characterized in that: Decrypting the dynamic encryption tag in step S2 includes: parsing the dynamic parameters of the dynamic encryption tag to perform hierarchical decoding of the tag structure of the dynamic encryption tag, thereby obtaining a hierarchical encrypted data unit; Derivation of time-varying keys based on layered encrypted data units to obtain a dynamic decryption key sequence; Reverse mapping the dynamic encryption tag according to the dynamic decryption key sequence to obtain the dynamic decryption intermediate state data; Perform timeliness state synchronization and noise elimination on the dynamic decryption intermediate state data to obtain dynamic code timeliness state data; By integrating the space-time dimension of the dynamic code aging status data, the dynamic code aging status is finally output.

5. The online real-time image recognition anti-counterfeiting verification method based on gold transaction according to claim 1, characterized in that: The anti-counterfeiting determination of the micro texture image of the gold product according to the texture matching degree and the aging state of the dynamic code in step S3 includes: The gold product's micro-texture image is subjected to anti-counterfeiting judgment based on texture matching, spectral deviation value and dynamic code aging status. When the following conditions are present at the same time, it is judged as authentic: texture matching is greater than or equal to 0.90 and less than or equal to 1.00, spectral deviation is less than or equal to 0.05, and dynamic code aging is greater than or equal to 90% and less than or equal to 100%; When the following conditions occur at the same time, it is judged as a counterfeit identification result: the texture matching degree is less than 0.90 and greater than or equal to 0.80, the spectral deviation is greater than 0.05 and less than or equal to 0.15, and the dynamic code timeliness is less than 90% and greater than or equal to 70%.

6. The online real-time image recognition anti-counterfeiting verification method based on gold transaction according to claim 1, characterized in that: Step S4 includes the following steps: Step S41: activating a multi-level review mechanism based on the authenticity identification result, wherein the multi-level review mechanism includes manual review and weight verification; Step S42: performing microstructural stress response analysis on the gold product through a multi-level review mechanism to generate microstructural stress response data; Step S43: performing lattice resonance frequency detection on the gold product according to the microstructure stress response data to obtain resonance frequency characteristic data; Step S44: Use the resonance frequency characteristic data and microstructure stress response data to perform material intrinsic property anti-counterfeiting judgment on the microtexture image of the gold product, obtain a secondary anti-counterfeiting judgment result and synchronously store it in the blockchain, generating a secondary gold anti-counterfeiting judgment storage record.

7. An online real-time image recognition anti-counterfeiting verification system based on gold trading, characterized in that: For executing the online real-time image recognition anti-counterfeiting verification method based on gold transactions as claimed in claim 1, the online real-time image recognition anti-counterfeiting verification system based on gold transactions comprises: The tag module is used to collect micro-texture images of gold products; based on the physical unclonable characteristics of the micro-texture image, a unique digital fingerprint is generated, which is encrypted with the UTC timestamp to generate a dynamic encrypted tag; Feature analysis module is used to extract feature descriptors of micro texture images and compare textures with pre-stored databases to obtain texture matching degree; decrypt dynamic encryption tags and generate dynamic code aging status; The primary anti-counterfeiting identification module is used to perform anti-counterfeiting identification on the micro-texture image of gold products based on texture matching and the aging status of the dynamic code to obtain the results of authenticity identification and counterfeit identification. The counterfeit identification results are stored in the blockchain for evidence recording, generating a primary verification storage record of authentic gold. The secondary anti-counterfeiting identification module is used to initiate a multi-level review mechanism based on the authenticity identification results, and to conduct anti-counterfeiting identification of the material intrinsic properties of the micro-texture image of the gold product again through the multi-level review mechanism. The secondary anti-counterfeiting identification results are obtained and stored synchronously in the blockchain to generate a secondary gold anti-counterfeiting identification storage record.

Images