A digital watermark-based data processing method for certificate anti-counterfeiting

Abstract

The present invention relates to the field of machine learning technology, and in particular to a method for processing certificate anti-counterfeiting data based on digital watermarks. The method comprises the following steps: obtaining certificate image data, performing complex feature extraction and encoding processing and hybrid and digital watermark embedding protection processing on the certificate image data, generating mixed domain watermark embedded image data, performing multi-view feature fusion extraction processing and complex pattern and anomaly detection processing on the mixed domain watermark embedded image data, generating certificate pattern recognition result data, performing distributed blockchain anti-counterfeiting verification and consensus processing and anti-fraud and data analysis processing on the pattern recognition result data, and generating intelligent anti-fraud model data. The present invention can embed the designed watermark pattern and rules into the certificate image. The certificate digital watermark data generated in this way can be used to verify the authenticity of the certificate, ensure that the certificate has not been tampered with or forged, thereby providing a more reliable anti-counterfeiting effect.

Description

Technical Field

[0001] The present invention relates to the technical field of machine learning, and in particular to a method for processing certificate anti-counterfeiting data based on digital watermark. Background Art

[0002] Applying machine learning technology to digital watermark-based certificate anti-counterfeiting data processing methods can provide more efficient, accurate and automated anti-counterfeiting functions. Traditional digital watermark technology is mainly used to verify and protect the authenticity and integrity of certificates, but it has certain limitations in identifying and preventing fraud. By introducing machine learning technology, large amounts of certificate data can be better processed and analyzed to identify and predict possible fraud. Machine learning models can learn and identify patterns, features and regularities in certificates, thereby helping to detect forgery, tampering or other fraud. There may be challenges in migrating machine learning models from one environment or data set to another. Certificates from different regions or different types may have different features and patterns. Summary of the invention

[0003] The present invention provides a digital watermark-based certificate anti-counterfeiting data processing method to solve at least one of the above technical problems.

[0004] To achieve the above object, the present invention provides a method for processing certificate anti-counterfeiting data based on digital watermark, the method comprising the following steps:

[0005] Step S1: Acquire the certificate image data, perform complex feature extraction and encoding processing on the certificate image data, and generate high-dimensional image feature data;

[0006] Step S2: Perform mixing and digital watermark embedding protection processing based on high-dimensional image feature data to generate mixed domain watermark embedded image data;

[0007] Step S3: Perform multi-view feature fusion extraction processing on the mixed domain watermark embedded image data to generate multi-view feature fusion data of the certificate;

[0008] Step S4: Perform complex pattern and anomaly detection processing based on the multi-view feature fusion data of the license and generate license pattern recognition result data;

[0009] Step S5: Perform distributed blockchain anti-counterfeiting verification and consensus processing on the certificate pattern recognition result data to generate certificate blockchain consensus anti-counterfeiting verification data;

[0010] Step S6: Perform anti-fraud and data analysis based on the consensus anti-counterfeiting verification data of the certificate blockchain to generate intelligent anti-fraud model data.

[0011] The present invention provides a method for processing certificate anti-counterfeiting data based on digital watermark, which obtains certificate image data, performs complex feature extraction and encoding processing on the certificate image data, generates high-dimensional image feature data, adopts dynamic data acquisition processing, and obtains multiple different data acquisition results by adjusting image parameters, applying illumination transformation, simulating different environmental conditions, etc. This diversified data acquisition helps to capture the performance of certificate images under different conditions and increase the diversity and coverage of data. Based on the high-dimensional image feature data, hybrid and digital watermark embedding protection processing is performed to generate hybrid domain watermark embedded image data. Through hybrid domain mapping processing, features in different domains can be mapped and fused to improve the diversity and richness of features. Such processing helps to comprehensively utilize the information of multiple feature spaces, enhance the expressive power and discrimination of data, and perform digital watermark generation processing based on the certificate hybrid domain mapping data set, which can provide anti-counterfeiting and authenticity verification for the certificate. By designing specific watermark patterns and rules and converting them into digital watermark data, the authenticity of the certificate can be marked and unauthorized copying and tampering can be prevented. By performing watermark embedding processing on the certificate digital watermark data, the digital watermark can be embedded into the certificate image. Such processing can protect the concealment and imperceptibility of the watermark, ensure that the watermark and image content are integrated with each other, and prevent the watermark from being tampered with or deleted. The designed watermark pattern and rules are embedded in the certificate image. The certificate digital watermark data generated in this way can be used to verify the authenticity of the certificate, ensure that the certificate has not been tampered with or forged, and thus provide a more reliable anti-counterfeiting effect. The mixed domain watermark embedded image data is subjected to multi-view feature fusion extraction processing to generate certificate multi-view feature fusion data. By performing feature dimension reduction and projection space construction on the mixed domain watermark embedded image data, the dimension of the data can be reduced and key information can be retained, thereby reducing storage and computing costs and improving data processing and analysis efficiency. Complex pattern and anomaly detection processing is performed based on the certificate multi-view feature fusion data to generate certificate pattern recognition result data. Chaotic data distortion and compression processing is performed on the certificate hierarchical multimodal feature mapping data to generate certificate chaotic compression feature data. This processing process can reduce data redundancy and noise while retaining the main features of the data. The chaotic compression feature data is subjected to self-encoding driven deep anomaly detection processing to generate certificate deep anomaly labeling data. This processing can capture anomalies and important features in the data and further extract key information. Through information compression and extraction, the refinement and representation efficiency of data can be improved, and the complexity of data processing can be reduced. Distributed blockchain anti-counterfeiting verification and consensus processing are performed on the certificate pattern recognition result data to generate certificate blockchain consensus anti-counterfeiting verification data. Through distributed data allocation and distributed computing processing, computing resources can be fully utilized and parallel computing and processing can be achieved. This improves processing efficiency and speed, making the processing of large-scale certificate data more efficient and scalable.Based on the consensus anti-counterfeiting verification data of the license blockchain, anti-fraud and data analysis are carried out to generate intelligent anti-fraud model data. Anti-fraud decision generation based on model interpretation and analysis can generate personalized decision suggestions for different license data. According to the model's prediction results and interpretation information for the license, suggestions on the authenticity recognition and fraud risk of the license can be determined. Such targeted decision suggestions can help users more accurately assess the credibility and risk of the license and take corresponding measures. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 A schematic diagram of the steps of a method for processing certificate anti-counterfeiting data based on digital watermarking according to the present invention;

[0013] Figure 2 Detailed implementation flow chart of step S1;

[0014] Figure 3 Detailed implementation flow chart of step S2. DETAILED DESCRIPTION

[0015] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0016] The embodiment of the present application provides a method for processing certificate and anti-counterfeiting data based on digital watermark. The execution subject of the method for processing certificate and anti-counterfeiting data based on digital watermark includes but is not limited to: mechanical equipment, data processing platform, cloud server node, network transmission equipment, etc. equipped with the system can be regarded as the general computing node of the present application. The data processing platform includes but is not limited to: at least one of an audio management system, an image management system, and an information management system.

[0017] See also Figures 1 to 3 The present invention provides a method for processing certificate anti-counterfeiting data based on digital watermark, the method comprising the following steps:

[0018] Step S1: Acquire the certificate image data, perform complex feature extraction and encoding processing on the certificate image data, and generate high-dimensional image feature data;

[0019] Step S2: Perform mixing and digital watermark embedding protection processing based on high-dimensional image feature data to generate mixed domain watermark embedded image data;

[0020] Step S3: Perform multi-view feature fusion extraction processing on the mixed domain watermark embedded image data to generate multi-view feature fusion data of the certificate;

[0021] Step S4: Perform complex pattern and anomaly detection processing based on the multi-view feature fusion data of the license and generate license pattern recognition result data;

[0022] Step S5: Perform distributed blockchain anti-counterfeiting verification and consensus processing on the certificate pattern recognition result data to generate certificate blockchain consensus anti-counterfeiting verification data;

[0023] Step S6: Perform anti-fraud and data analysis based on the consensus anti-counterfeiting verification data of the certificate blockchain to generate intelligent anti-fraud model data.

[0024] The present invention provides a method for processing certificate anti-counterfeiting data based on digital watermark, which obtains certificate image data, performs complex feature extraction and encoding processing on the certificate image data, generates high-dimensional image feature data, adopts dynamic data acquisition processing, and obtains multiple different data acquisition results by adjusting image parameters, applying illumination transformation, simulating different environmental conditions, etc. This diversified data acquisition helps to capture the performance of certificate images under different conditions and increase the diversity and coverage of data. Based on the high-dimensional image feature data, hybrid and digital watermark embedding protection processing is performed to generate hybrid domain watermark embedded image data. Through hybrid domain mapping processing, features in different domains can be mapped and fused to improve the diversity and richness of features. Such processing helps to comprehensively utilize the information of multiple feature spaces, enhance the expressive power and discrimination of data, and perform digital watermark generation processing based on the certificate hybrid domain mapping data set, which can provide anti-counterfeiting and authenticity verification for the certificate. By designing specific watermark patterns and rules and converting them into digital watermark data, the authenticity of the certificate can be marked and unauthorized copying and tampering can be prevented. By performing watermark embedding processing on the certificate digital watermark data, the digital watermark can be embedded into the certificate image. Such processing can protect the concealment and imperceptibility of the watermark, ensure that the watermark and image content are integrated with each other, and prevent the watermark from being tampered with or deleted. The designed watermark pattern and rules are embedded in the certificate image. The certificate digital watermark data generated in this way can be used to verify the authenticity of the certificate, ensure that the certificate has not been tampered with or forged, and thus provide a more reliable anti-counterfeiting effect. The mixed domain watermark embedded image data is subjected to multi-view feature fusion extraction processing to generate certificate multi-view feature fusion data. By performing feature dimension reduction and projection space construction on the mixed domain watermark embedded image data, the dimension of the data can be reduced and key information can be retained, thereby reducing storage and computing costs and improving data processing and analysis efficiency. Complex pattern and anomaly detection processing is performed based on the certificate multi-view feature fusion data to generate certificate pattern recognition result data. Chaotic data distortion and compression processing is performed on the certificate hierarchical multimodal feature mapping data to generate certificate chaotic compression feature data. This processing process can reduce data redundancy and noise while retaining the main features of the data. The chaotic compression feature data is subjected to self-encoding driven deep anomaly detection processing to generate certificate deep anomaly labeling data. This processing can capture anomalies and important features in the data and further extract key information. Through information compression and extraction, the refinement and representation efficiency of data can be improved, and the complexity of data processing can be reduced. Distributed blockchain anti-counterfeiting verification and consensus processing are performed on the certificate pattern recognition result data to generate certificate blockchain consensus anti-counterfeiting verification data. Through distributed data allocation and distributed computing processing, computing resources can be fully utilized and parallel computing and processing can be achieved. This improves processing efficiency and speed, making the processing of large-scale certificate data more efficient and scalable.Based on the consensus anti-counterfeiting verification data of the license blockchain, anti-fraud and data analysis are carried out to generate intelligent anti-fraud model data. Anti-fraud decision generation based on model interpretation and analysis can generate personalized decision suggestions for different license data. According to the model's prediction results and interpretation information for the license, suggestions on the authenticity recognition and fraud risk of the license can be determined. Such targeted decision suggestions can help users more accurately assess the credibility and risk of the license and take corresponding measures.

[0025] In the embodiment of the present invention, reference Figure 1 The above is a schematic flow chart of the steps of a method for processing certificate anti-counterfeiting data based on digital watermarks of the present invention. In this example, the method for processing certificate anti-counterfeiting data based on digital watermarks includes the following steps:

[0026] Step S1: Acquire the certificate image data, perform complex feature extraction and encoding processing on the certificate image data, and generate high-dimensional image feature data;

[0027] In an embodiment of the present invention, certificate image data is acquired, dynamic data acquisition processing is performed on the certificate image data to generate an original certificate image data set, unsupervised deep segmentation processing is performed based on the original certificate image data set to generate a segmented image element set, heterogeneous feature encoding processing is performed on the segmented image element set to generate certificate feature encoding data, high-dimensional feature mapping processing is performed on the certificate feature encoding data set to generate high-dimensional image feature data.

[0028] Step S2: Perform mixing and digital watermark embedding protection processing based on high-dimensional image feature data to generate mixed domain watermark embedded image data;

[0029] In an embodiment of the present invention, a weighted feature fusion regularization formula is used to calculate high-dimensional image feature data to generate a weighted distribution feature data set, and a mixed domain mapping process is performed on the weighted distribution feature data set to generate a mixed domain mapping data set. A digital watermark generation process is performed based on the mixed domain mapping data set for certificates to generate digital watermark data for certificates. The digital watermark data for certificates is watermark embedded to generate mixed domain watermark embedded image data.

[0030] Step S3: Perform multi-view feature fusion extraction processing on the mixed domain watermark embedded image data to generate multi-view feature fusion data of the certificate;

[0031] In an embodiment of the present invention, high-dimensional projection space construction processing is performed on the mixed-domain watermark embedded image data to generate a high-dimensional projection space data set, holographic visual mapping processing is performed based on the high-dimensional projection space data set to generate a holographic visual mapping data set for certificates and licenses, high-order feature analysis processing is performed on the holographic visual mapping data set for certificates and licenses to generate a high-order feature analysis data set for certificates and licenses, and Gaussian compression and multi-view fusion calculation formulas are used to calculate the high-order feature analysis data for certificates and licenses to generate multi-view feature fusion data for certificates and licenses.

[0032] Step S4: Perform complex pattern and anomaly detection processing based on the multi-view feature fusion data of the license and generate license pattern recognition result data;

[0033] In an embodiment of the present invention, feature hierarchy and multimodal mapping are performed on the multi-view feature fusion data of the license to generate hierarchical multimodal feature mapping data of the license, chaotic data distortion and compression are performed based on the hierarchical multimodal feature mapping data of the license to generate chaotic compressed feature data of the license, self-encoding driven deep anomaly detection processing is performed on the chaotic compressed feature data of the license to generate deep anomaly marking data of the license, and pattern recognition processing based on graph neural network is performed on the deep anomaly marking data of the license to generate pattern recognition result data of the license.

[0034] Step S5: Perform distributed blockchain anti-counterfeiting verification and consensus processing on the certificate pattern recognition result data to generate certificate blockchain consensus anti-counterfeiting verification data.

[0035] In an embodiment of the present invention, pattern recognition result encoding processing is performed on multi-view feature fusion data of certificates and licenses to generate a certificate pattern recognition coding set, distributed data allocation processing is performed based on the certificate pattern recognition coding set to generate a distributed pattern recognition result data set, blockchain data encapsulation processing is performed on the distributed pattern recognition result data set to generate a certificate blockchain encapsulation data set, consensus algorithm anti-counterfeiting verification processing is performed based on the certificate blockchain encapsulation data set to generate a certificate blockchain consensus anti-counterfeiting verification data set.

[0036] Step S6: Perform anti-fraud and data analysis based on the consensus anti-counterfeiting verification data of the certificate blockchain to generate intelligent anti-fraud model data.

[0037] In an embodiment of the present invention, data standardization processing is performed on the certificate blockchain consensus anti-counterfeiting verification data set to generate a standardized blockchain verification data set, feature engineering transformation processing is performed based on the standardized blockchain verification data set to generate characterized blockchain verification data, anti-fraud model training processing is performed based on the characterized blockchain verification data to generate certificate anti-fraud model data.

[0038] Preferably, step S1 comprises the following steps:

[0039] Step S11: Acquire the certificate image data, perform dynamic data acquisition processing on the certificate image data, and generate an original certificate image data set;

[0040] Step S12: performing unsupervised deep segmentation processing based on the original certificate image data set to generate a segmented image element set;

[0041] Step S13: performing heterogeneous feature coding processing on the segmented image element set to generate a certificate feature coding data set;

[0042] Step S14: Perform high-dimensional feature mapping processing on the certificate feature encoding data set to generate high-dimensional image feature data.

[0043] The present invention obtains the certificate image data, performs dynamic data acquisition processing on the certificate image data, generates the original certificate image data set, and obtains multiple different data acquisition results through dynamic data acquisition processing, including different lighting conditions, image parameters and environmental simulation. The original certificate image data set generated in this way has richer diversity and can cover the changes of certificate images in different scenes. Based on the original certificate image data set, unsupervised deep segmentation processing is performed to generate a segmented image element set. The feature representation learned by the deep learning model in the segmentation process can better capture the local structure and global context in the image. The generated segmented image element set can provide more informative features, help distinguish different image elements, and provide more accurate feature representation for subsequent anti-counterfeiting verification and recognition tasks. By training and optimizing the unsupervised deep learning model, the segmentation effect and generalization ability of the model for the certificate image can be improved. The generated segmented image element set can be used to evaluate and improve the performance of the deep learning model, thereby improving the accuracy and reliability of the anti-counterfeiting method. The segmented image element set is subjected to heterogeneous feature coding processing to generate a certificate feature coding data set. The heterogeneous feature coding fuses different types of features together, so that the generated feature coding data set has higher information richness. The feature extraction of each element or area can capture subtle image changes and detail information, thereby providing a more accurate and detailed feature description, and providing a more comprehensive information basis for subsequent certificate anti-counterfeiting processing. The certificate feature encoding data set is processed by high-dimensional feature mapping to generate high-dimensional image feature data. Through high-dimensional feature mapping, low-dimensional features can be converted into higher-dimensional feature representations. In the high-dimensional feature space, the certificate image features are expressed more accurately and in detail, thereby improving the representativeness of the features. This helps to better capture the subtle differences and local structures of the image, making the features more distinguishable and expressive.

[0044] As an example of the present invention, refer to Figure 2 As shown, Figure 1 Detailed implementation steps of step S1 in the flowchart, in this example, step S1 includes:

[0045] Step S11: Acquire the certificate image data, perform dynamic data acquisition processing on the certificate image data, and generate an original certificate image data set;

[0046] In the examples of the present invention, image data of the target certificate is obtained, including digital images scanned, photographed or obtained from other sources, and the original certificate image data is preprocessed, including denoising, image enhancement, geometric correction and other operations to improve image quality and accuracy, and dynamic data acquisition processing is performed. By adjusting image parameters, applying illumination transformation, simulating different environmental conditions and other methods, multiple different data acquisition results are obtained, and all dynamic data acquisition results are combined to generate an original certificate image data set.

[0047] Step S12: performing unsupervised deep segmentation processing based on the original certificate image data set to generate a segmented image element set;

[0048] In the embodiment of the present invention, the original certificate image data set is input into the deep learning model, and the image segmentation is performed by using an unsupervised learning method, and deep learning technology, such as a convolutional neural network (CNN), is used to perform feature learning and representation learning on the original image to extract the target area in the image. During the training process, unsupervised strategies, such as self-organizing maps, clustering and other methods, are used to segment the image, divide the image pixels into regions with similar features, and generate a set of segmented image elements.

[0049] Step S13: performing heterogeneous feature coding processing on the segmented image element set to generate a certificate feature coding data set;

[0050] In an embodiment of the present invention, heterogeneous feature extraction is performed on the segmented image element set, and multiple feature extraction algorithms and methods are used, such as local binary pattern (LBP), histogram of oriented gradients (HOG), color histogram, etc., to extract multiple features from each element or region, and the extracted heterogeneous features are normalized, dimensionally reduced or compressed, and the processed heterogeneous features are encoded into a specific data format or vector representation to generate a certificate feature encoding data set.

[0051] Step S14: Perform high-dimensional feature mapping processing on the certificate feature encoding data set to generate high-dimensional image feature data.

[0052] In an embodiment of the present invention, a high-dimensional feature mapping method, such as a kernel technique or a deep neural network, is used to map a certificate feature encoding data set to a higher-dimensional feature space. The mathematical calculation formula or algorithm of the feature mapping method is used to convert low-dimensional features into high-dimensional features to capture more image features and relationships. In the mapped high-dimensional feature space, the features are further processed and analyzed, such as clustering, classification, and dimensionality reduction, to extract more discriminative and representative image features, thereby generating high-dimensional image feature data.

[0053] Preferably, step S2 comprises the following steps:

[0054] Step S21: Calculate the high-dimensional image feature data using a weighted feature fusion regularization formula to generate a weighted distribution feature data set;

[0055] Step S22: performing hybrid domain mapping processing on the weighted distribution feature data set to generate a hybrid domain mapping data set;

[0056] Step S23: Performing digital watermark generation processing based on the certificate mixed domain mapping data set to generate certificate digital watermark data;

[0057] Step S24: Perform watermark embedding processing on the digital watermark data of the certificate to generate mixed domain watermark embedded image data.

[0058] The present invention generates a hybrid domain mapping data set by performing hybrid domain mapping processing on a weight distribution feature data set. The hybrid domain mapping algorithm combines the mapping methods of multiple domains or models, and generates a hybrid domain feature data set by mapping and fusing the features in different domains. This feature fusion method can make full use of the advantages of different domains or models, improve the diversity and richness of feature expression, thereby enhancing the distinguishing and expressive ability of features, and increase the diversity and quantity of data by introducing more features of domains or models. This data enhancement method can provide more samples for training and testing models. The weight distribution feature data set is subjected to hybrid domain mapping processing to generate hybrid domain mapping data. The hybrid domain mapping algorithm combines the mapping methods of multiple domains or models, and can fuse and map the features in different domains, thereby providing richer and more comprehensive feature information. This feature fusion and enhancement can improve the representation ability and discrimination of the data set, so that the subsequent certificate anti-counterfeiting data processing method can be more accurately identified and verified. Digital watermark generation processing is performed based on the certificate hybrid domain mapping data set to generate certificate digital watermark data. Through the digital watermark generation processing based on the certificate hybrid domain mapping data set, the designed watermark pattern and rules can be embedded into the certificate image. The digital watermark data of the certificate generated in this way can be used to verify the authenticity of the certificate, ensuring that the certificate has not been tampered with or forged, thereby providing a more reliable anti-counterfeiting effect. The digital watermark data of the certificate is watermarked and embedded to generate mixed domain watermark embedded image data. The watermark embedding process integrates the digital watermark information into a specific position or frequency domain of the certificate image. This embedding method makes the watermark more concealed and difficult to be detected by the human eye, thereby increasing the security of the anti-counterfeiting system and its ability to resist tampering.

[0059] As an example of the present invention, refer to Figure 3 As shown, Figure 1 Detailed implementation steps of step S2 in the embodiment are shown in the flowchart. In this embodiment, step S2 includes:

[0060] Step S21: Calculate the high-dimensional image feature data using a weighted feature fusion regularization formula to generate a weighted distribution feature data set;

[0061] Preferably, the weighted feature fusion regularization formula in step S21 is specifically:

[0062]

[0063] Among them, WFD refers to the weighted feature dataset, m refers to the number of selected features, and w i refers to the weight corresponding to the i-th selected feature, f i (x) refers to the function corresponding to the i-th selected feature, which is a real function, λ refers to a regularization parameter, and x refers to a sample in the high-dimensional image feature data.

[0064] The present invention creates a weighted feature fusion regularization formula, which uses weighted feature fusion technology to perform different degrees of weighted processing on different types of features in high-dimensional image feature data to highlight important features and suppress redundant features, thereby achieving the purpose of feature selection and dimensionality reduction. For example, for certificate image data, features such as face, name, gender, etc. may be more distinguishable and recognizable than features such as background and border, so a higher weight coefficient w can be given. i For other types of image data, the weight coefficient w may need to be dynamically adjusted according to different application scenarios and requirements. i The value of . Through weighted feature fusion technology, effective information in image data can be effectively extracted and the amount of calculation and storage space can be reduced. The main part of the formula is This is a linear combination, where each feature f i (x) according to its corresponding weight w i Weighted. This linear combination ensures that each feature contributes according to their importance (represented by the weight). The formula uses regularization technology to smooth and denoise high-dimensional image feature data to reduce the impact of noise and outliers in the image data on the recognition results and prevent overfitting and underfitting problems. The regularization parameter λ is used to control the degree of smoothing and denoising, and minimize the mean square error between the image data after weight allocation and the original image data, thereby achieving optimization and stabilization of image data.

[0065] Step S22: performing hybrid domain mapping processing on the weighted distribution feature data set to generate a hybrid domain mapping data set;

[0066] In the embodiment of the present invention, the weight allocation feature data set is preprocessed, including data cleaning, denoising, missing value filling and other operations, and the preprocessed weight allocation feature data set is processed using a hybrid domain mapping algorithm. The hybrid domain mapping algorithm combines the mapping methods of multiple domains or models, maps and fuses the features in different domains, and generates a hybrid domain feature data set according to the calculation results of the hybrid domain mapping algorithm;

[0067] Step S23: Performing digital watermark generation processing based on the certificate mixed domain mapping data set to generate certificate digital watermark data;

[0068] In an embodiment of the present invention, the pattern and rules of the digital watermark are designed according to the characteristic information and requirements of the hybrid domain mapping data set of the certificate. The digital watermark may include text, images, identifiers and other forms, which are used to mark the authenticity and anti-counterfeiting of the certificate. In the design process, it is necessary to consider factors such as the embeddability, anti-attack and visibility of the watermark, and use a specific digital watermark generation algorithm to convert the designed watermark pattern and rules into specific digital watermark data. The generation algorithm can be implemented based on methods such as image processing technology, information hiding algorithm or encryption algorithm. The digital watermark generation algorithm is discrete cosine transform (DCT). The generated digital watermark data is embedded in the hybrid domain mapping data set of the certificate. The embedding process can use spatial domain embedding, and the final digital watermark data of the certificate is generated according to the hybrid domain data set obtained by the embedding process and the embedded digital watermark.

[0069] Step S24: Perform watermark embedding processing on the digital watermark data of the certificate to generate mixed domain watermark embedded image data.

[0070] In the embodiment of the present invention, the digital watermark data of the certificate is preprocessed, including operations such as image enhancement, denoising, and adjusting brightness contrast, and a suitable watermark embedding algorithm is selected according to the requirements and application scenarios. Commonly used watermark embedding algorithms include spatial domain embedding, and the digital watermark is embedded into the certificate image using the selected watermark embedding algorithm. The embedding process needs to combine image processing and watermark embedding algorithm to integrate the digital watermark information into a specific position or frequency domain of the image, and generate the final mixed domain watermark embedded image data based on the certificate image with the digital watermark obtained by the embedding process.

[0071] Preferably, step S3 comprises the following steps:

[0072] Step S31: Perform high-dimensional projection space construction processing on the mixed domain watermark embedded image data to generate a high-dimensional projection space data set;

[0073] Step S32: Perform holographic visual mapping processing based on the high-dimensional projection space data set to generate a certificate holographic visual mapping data set;

[0074] Step S33: performing high-order feature analysis processing on the certificate holographic visual mapping data set to generate a certificate high-order feature analysis data set;

[0075] Step S34: Calculate the high-order feature analysis data of the license using Gaussian compression and multi-view fusion calculation formula to generate multi-view feature fusion data of the license.

[0076] The present invention generates a high-dimensional projection space data set by constructing a high-dimensional projection space for the hybrid domain watermark embedded image data, and maps the reduced-dimensional features to the new high-dimensional projection space, which helps to map the original data from a complex high-dimensional space to a more understandable and visualized low-dimensional space. It can better observe and analyze the features of the hybrid domain watermark embedded image data, and discover the potential patterns and laws of the data. Holographic visual mapping processing is performed based on the high-dimensional projection space data set to generate a holographic visual mapping data set for certificates. The holographic image has complex optical properties and a high degree of image reconstruction capability, and has a certain resistance to forgery and tampering. After the holographic visual mapping data set for certificates is generated, by analyzing and comparing the image information under different viewing angles, it is easier to detect forgery, thereby improving the reliability and security of certificate anti-counterfeiting. The holographic visual mapping data set for certificates is processed with high-order features to generate a high-order feature analysis data set for certificates. By applying high-order feature extraction algorithms (such as convolutional neural networks) and feature analysis algorithms (such as wavelet transforms or singular value decomposition), the high-order feature analysis data set for certificates can extract more abstract and complex feature representations. These features not only contain basic features of the image (such as edges and textures), but also capture more advanced semantic information of the image. Through high-order feature parsing, the key features in the certificate image can be highlighted and emphasized, that is, those features that have an important impact on the image content and information. This helps to extract and utilize key information in the certificate image.

[0077] Step S31: Perform high-dimensional projection space construction processing on the mixed domain watermark embedded image data to generate a high-dimensional projection space data set;

[0078] In the embodiment of the present invention, for the mixed domain watermark embedded image data, the linear discriminant analysis (LDA) of the feature dimensionality reduction algorithm is used to convert the high-dimensional features into low-dimensional representations to reduce the data dimension and retain key information. The reduced-dimensional features are mapped to a new high-dimensional projection space using a mapping algorithm, such as local linear embedding (LLE) or isometric mapping (Isomap), thereby generating a high-dimensional projection space data set.

[0079] Step S32: Perform holographic visual mapping processing based on the high-dimensional projection space data set to generate a certificate holographic visual mapping data set;

[0080] In the embodiment of the present invention, holographic encoding technology, such as digital holography or optical holography, is used to convert the certificate image data into a holographic image. The holographic image records the optical information of the certificate in the form of holographic interference, has a high spatial resolution and image reconstruction capability, and uses a perspective generation algorithm, such as multi-view imaging or viewpoint cloud rendering, to generate multiple images of different perspectives from the holographic image, thereby generating a certificate holographic visual mapping data set.

[0081] Step S33: performing high-order feature analysis processing on the certificate holographic visual mapping data set to generate a certificate high-order feature analysis data set;

[0082] In the embodiment of the present invention, a high-order feature extraction algorithm (convolutional neural network (CNN)) is used to extract more abstract and complex feature representations from the multi-view image data of the license. These features have higher semantic expression capabilities and can capture more details and important information in the license image. Feature analysis algorithms such as wavelet transform or singular value decomposition are applied to further analyze and process the high-order features, thereby generating a license high-order feature analysis data set.

[0083] Step S34: Calculate the high-order feature analysis data of the license using Gaussian compression and multi-view fusion calculation formula to generate multi-view feature fusion data of the license;

[0084] Preferably, the Gaussian compression and multi-view fusion calculation formula in step S34 is specifically:

[0085]

[0086] Among them, F(y) refers to the multi-view feature fusion data of the license, y refers to a sample in the high-order feature analysis data set, k refers to the dimension of the feature space, α j refers to the weight coefficient of the jth feature space, σ j refers to the mean of the jth feature space, μ j refers to the standard deviation of the j-th feature space, β j refers to the compression index of the j-th feature space, γ refers to the weight coefficient of feature fusion, δ refers to an infinite decimal approaching zero, and D refers to the multi-view feature distance function of sample y.

[0087] The present invention creates a Gaussian compression and multi-view fusion calculation formula, which uses multi-view fusion technology to fuse features of different viewpoints in high-order feature analysis data of certificates with different weights, so as to increase the diversity and integrity of the feature space, thereby achieving the purpose of feature enhancement and security. For example, for high-order feature analysis data of certificates, since it comes from certificate image data of different types and angles, it may contain feature information of multiple viewpoints and dimensions, and these feature information may reflect different attributes and characteristics of the certificate image data, such as face, name, gender, etc. Through multi-view fusion technology, the logarithmic derivative function can be used to describe the rate of change of each viewpoint feature, and the degree of fusion can be controlled by adjusting the fusion weight coefficient γ, and the difference between the fused feature data and the original feature data is maximized, thereby achieving feature data enhancement and security, and using the weighting coefficient α jThe high-level mathematical parameters such as the regularization parameter δ are used to flexibly and finely adjust and control the high-order feature analysis data of the license plate to adapt to the license plate recognition tasks under different scenarios and requirements, thereby achieving the purpose of parameter adaptation and intelligence. The formula performs Gaussian compression on the feature space, which is done through the Gaussian function. The shape of the Gaussian function is affected by its mean μ j and standard deviation σ j Therefore, by changing these two parameters, the feature space can be compressed to different degrees. In addition, the degree of compression is also related to the compression index β j Related, β j The size of will directly affect the shape of the compressed feature space. In this way, lossy compression of the original high-order feature analysis data can be achieved, reducing its data size while retaining its main features as much as possible.

[0088] Preferably, step S4 comprises the following steps:

[0089] Step S41: performing feature hierarchical processing and multimodal mapping processing on the multi-view feature fusion data of the license to generate hierarchical multimodal feature mapping data of the license;

[0090] Step S42: performing chaotic data distortion and compression processing based on the certificate hierarchical multimodal feature mapping data to generate certificate chaotic compression feature data;

[0091] Step S43: performing self-encoding driven deep anomaly detection processing on the chaotic compression feature data of the license and credential to generate deep anomaly marking data of the license and credential;

[0092] Step S44: Perform pattern recognition processing based on graph neural network on the certificate depth abnormality mark data to generate certificate pattern recognition result data.

[0093] The present invention generates hierarchical multimodal feature mapping data of certificates by performing feature hierarchization and multimodal mapping processing on multi-view feature fusion data of certificates. The feature hierarchization and multimodal mapping processing can reveal the implicit relationship and correlation between certificate features. By mapping and fusing features of different levels and modes, the potential association between features can be discovered, and the understanding and analysis of the intrinsic information of certificates can be further deepened. Chaotic data distortion and compression processing is performed based on the hierarchical multimodal feature mapping data of certificates to generate chaotic compressed feature data of certificates. Through chaotic data distortion processing, the hierarchical multimodal feature mapping data of certificates are introduced into the chaotic mapping system, and the sensitivity and nonlinear characteristics of the chaotic system are used to distort the data. This nonlinear data distortion increases the complexity and unpredictability of the data, thereby providing a more challenging data foundation for the certificate anti-counterfeiting data processing method. The chaotic compressed feature data of certificates is subjected to deep anomaly detection processing driven by self-encoder to generate deep anomaly labeling data of certificates. The anomaly detection using the autoencoder model can quickly and efficiently identify anomalies in the chaotic compressed feature data of certificates. Compared with traditional rule-based or statistical anomaly detection technologies, the autoencoder model can learn complex feature representations of data, can more accurately capture abnormal patterns, and improve the accuracy and efficiency of anomaly detection. The deep abnormality labeled data of certificates and licenses is processed with pattern recognition based on graph neural networks to generate certificate pattern recognition result data. During the training process, the graph neural network learns the difference between normal and abnormal patterns and has strong abnormal pattern detection capabilities. By performing pattern recognition processing on the deep abnormality labeled data of certificates and licenses, it is possible to effectively detect and identify abnormal situations in certificates and licenses, such as tampered, forged, or tampered parts, etc., and improve the reliability and security of the certificate and license anti-counterfeiting data processing method.

[0094] Step S41: performing feature hierarchical processing and multimodal mapping processing on the multi-view feature fusion data of the license to generate hierarchical multimodal feature mapping data of the license;

[0095] In an embodiment of the present invention, the multi-view feature fusion data of the certificate is subjected to hierarchical feature processing. Features at different levels are extracted and represented. A hierarchical clustering method is used to organize and decompose the original features to form a hierarchical feature structure. Based on the result of the feature hierarchy, multimodal mapping processing is performed. Features at different levels and in different modes are mapped and fused to obtain more comprehensive and rich feature information. Using multimodal fusion technology methods, the representations of different features are integrated and mapped to generate multimodal feature mapping data, and the final certificate hierarchical multimodal feature mapping data is generated based on the results of feature hierarchy and multimodal mapping.

[0096] Step S42: performing chaotic data distortion and compression processing based on the certificate hierarchical multimodal feature mapping data to generate certificate chaotic compression feature data;

[0097] In an embodiment of the present invention, chaotic data distortion is applied to the hierarchical multimodal feature mapping data of the certificate. Chaotic data distortion is a method of data processing using the characteristics of a chaotic system. The characteristic of a chaotic system is that it is sensitive to initial conditions, that is, a small difference in initial values ​​can cause a huge change in the system state. First, a suitable chaotic map, such as a Lorentz chaotic map, is selected. Then, the hierarchical multimodal feature mapping data of the certificate is used as the initial value of the chaotic map. The subsequent values ​​of the chaotic map are calculated in an iterative manner to generate distorted data, followed by compression processing. The purpose of this step is to reduce the redundancy and noise of the data while retaining the main characteristics of the data. A commonly used method is principal component analysis (PCA), which can project data onto a new set of orthogonal axes through linear transformation. These axes are arranged according to the size of the variance, and the axis with the largest variance is retained, thereby achieving data compression, thereby generating chaotic compressed feature data of the certificate.

[0098] Step S43: performing self-encoding driven deep anomaly detection processing on the chaotic compression feature data of the license and credential to generate deep anomaly marking data of the license and credential;

[0099] In an embodiment of the present invention, an autoencoder model is used to train the chaotic compression feature data of the license. The autoencoder is an unsupervised learning model that can learn high-level feature representations of data. By training the autoencoder, the intrinsic feature patterns of normal data can be captured. The trained autoencoder model is used to detect anomalies in the chaotic compression feature data of the license. By inputting the data into the autoencoder, the data can be reconstructed and the reconstruction error can be calculated. A higher reconstruction error indicates that the data is significantly different from the training data, and an anomaly may exist. A threshold can be set to determine whether the data is abnormal, and the trained autoencoder model is used to detect anomalies in the chaotic compression feature data of the license. By inputting the data into the autoencoder, the data can be reconstructed and the reconstruction error can be calculated. A higher reconstruction error indicates that the data is significantly different from the training data, and an anomaly may exist. A threshold can be set to determine whether the data is abnormal, and the final license deep anomaly labeling data is generated based on the results of the anomaly labeling.

[0100] Step S44: Perform pattern recognition processing based on graph neural network on the certificate depth abnormality mark data to generate certificate pattern recognition result data.

[0101] In an embodiment of the present invention, a graph neural network model is constructed for certificate pattern recognition. A graph neural network is a deep learning model that can process graph data and can make full use of the characteristics of the graph structure for pattern recognition. According to the specific task requirements, a suitable graph neural network model is selected, such as a graph convolutional network (GCN). The deep anomaly labeled data of the certificate is trained using the graph neural network model. The graph neural network is trained by inputting data and corresponding labels to learn the patterns and feature representations in the graph data. During the training process, backpropagation algorithms and optimization algorithms, such as stochastic gradient descent (SGD), can be used to perform pattern recognition on the deep anomaly labeled data of the certificate using the trained graph neural network model. The data is input into the trained model, and the output result is calculated by the forward propagation algorithm. The model will classify or identify the data according to the characteristics of the input data and the learned weights, obtain the result of pattern recognition, and generate the final certificate pattern recognition result data based on the result of pattern recognition.

[0102] Preferably, step S5 comprises the following steps:

[0103] Step S51: performing pattern recognition result encoding processing on the certificate multi-view feature fusion data to generate a certificate pattern recognition code set;

[0104] Step S52: Perform distributed data allocation processing based on the certificate pattern recognition code set to generate a distributed pattern recognition result data set;

[0105] Step S53: Perform blockchain data encapsulation processing on the distributed pattern recognition result data set to generate a license blockchain encapsulation data set;

[0106] Step S54: Perform consensus algorithm anti-counterfeiting verification processing based on the certificate blockchain encapsulation data set to generate a certificate blockchain consensus anti-counterfeiting verification data set.

[0107] The present invention performs pattern recognition result encoding processing on the multi-view feature fusion data of the license and generates a license and pattern recognition encoding set. The encoded pattern recognition result can be used to quickly retrieve and match key features and accelerate the execution speed of the pattern recognition algorithm. Due to the compactness and efficiency of the encoded data set, the system can perform feature matching and similarity calculation more quickly, improve the efficiency and responsiveness of the pattern recognition process, and hide the detailed information of the original features by encoding the pattern recognition results, thereby enhancing the confidentiality of the data. The encoded data set only contains an abstract representation of the features, which is not easily interpreted and understood by unauthorized people or systems, thereby improving the security and privacy protection of the license and certificate data. Distributed data allocation processing is performed based on the license and certificate pattern recognition encoding set to generate a distributed pattern recognition result data set. By allocating the license and certificate pattern recognition encoding set to multiple computing nodes for parallel processing, distributed computing resources can be fully utilized to achieve efficient processing of large-scale data sets. Parallel computing can significantly improve the computing speed and processing power of the algorithm, thereby accelerating the pattern recognition process and saving time and resource costs. Blockchain data encapsulation processing is performed on the distributed pattern recognition result data set to generate a license and certificate blockchain encapsulation data set, and blockchain technology is used to construct a license and certificate blockchain structure, thereby realizing decentralized data management. Data is no longer controlled by a single central organization, but distributed across multiple nodes, each with a complete copy of the data. This decentralized nature makes data more reliable and dependable, and is not affected by single point failures or centralized attacks. Based on the certificate blockchain encapsulated data set, consensus algorithm anti-counterfeiting verification processing is performed to generate a certificate blockchain consensus anti-counterfeiting verification data set, which is generated based on the verification results of the consensus algorithm. The anti-tampering and anti-counterfeiting capabilities of the certificate data can be ensured by verifying and confirming the nodes participating in the consensus verification according to the rules and protocols of the consensus algorithm. Any tampering of the blockchain data will be detected by the consensus verification system, thereby providing efficient anti-counterfeiting functions.

[0108] Step S51: performing pattern recognition result encoding processing on the certificate multi-view feature fusion data to generate a certificate pattern recognition code set;

[0109] In an embodiment of the present invention, key pattern recognition features are extracted from the multi-view feature fusion data of the certificate. The local binary pattern (LBP) in the feature extraction algorithm and technology is used to capture the statistical laws and spatial structure of the features, and extract features with distinguishing degree and importance. The extracted pattern recognition features are encoded and converted into a processable data format. Coding techniques such as hash coding and vector quantization can be used to represent the features as a set of encoded data, and the encoded pattern recognition results are compressed. Lossless compression algorithms and dictionary compression algorithms in compression algorithms are used to reduce the volume of data, and the final certificate pattern recognition code set is generated according to the results of the coding compression.

[0110] Step S52: Perform distributed data allocation processing based on the certificate pattern recognition code set to generate a distributed pattern recognition result data set;

[0111] In an embodiment of the present invention, the certificate pattern recognition code set is segmented according to certain rules and strategies. Distributed algorithms and technologies, such as hash functions, are used to segment the data set into multiple subsets. The segmentation process can be flexibly adjusted according to the distributed computing environment and data sharing requirements. The segmented certificate pattern recognition code set is allocated to different nodes or devices for processing. Using a distributed computing framework, such as Apache Hadoop, Spark, etc., the data is assigned to multiple computing nodes for parallel processing. The allocation process can make full use of distributed computing resources and generate a final distributed pattern recognition result data set based on the results of distributed computing.

[0112] Step S53: Perform blockchain data encapsulation processing on the distributed pattern recognition result data set to generate a certificate blockchain encapsulation data set.

[0113] In an embodiment of the present invention, each data block in the distributed pattern recognition result data set is hashed. A hash algorithm, such as SHA-256, MD5, etc., is used to convert the data block into a hash value of fixed length. Based on blockchain technology, a data structure of a license blockchain is constructed. This includes generating initial block header information, creating a genesis block and a block structure on the chain, etc. A blockchain platform or development framework, such as Hyperledger Fabric, is used to construct a blockchain structure with distributed, tamper-proof and decentralized characteristics, and the hashed distributed pattern recognition result data set is encapsulated into the blockchain structure. The hash value of each data block is recorded in the transaction record of the blockchain as part of the transaction data. The data encapsulation process is to link the data blocks to the blockchain one by one to form an unalterable chain structure, and generate the final license blockchain encapsulation data set based on the result of data encapsulation.

[0114] Step S54: Perform consensus algorithm anti-counterfeiting verification processing based on the certificate blockchain encapsulation data set to generate a certificate blockchain consensus anti-counterfeiting verification data set.

[0115] In the embodiment of the present invention, an applicable consensus algorithm is selected to verify the certificate blockchain encapsulated data set. Byzantine Fault Tolerance (BFT) and the like in the consensus algorithm are adopted. Under the selected consensus algorithm, consensus verification is performed on the certificate blockchain encapsulated data set. The nodes participating in the consensus verification will verify and confirm the transaction data in the blockchain according to the rules and protocols of the consensus algorithm, and generate the final certificate blockchain consensus anti-counterfeiting verification data set according to the verification results of the consensus algorithm.

[0116] Preferably, step S6 comprises the following steps:

[0117] Step S61: Perform data standardization processing on the certificate blockchain consensus anti-counterfeiting verification data set to generate a standardized blockchain verification data set;

[0118] Step S62: Perform feature engineering transformation processing based on the standardized blockchain verification data set to generate feature-based blockchain verification data;

[0119] Step S63: Perform anti-fraud model training based on the characterized blockchain verification data to generate license anti-fraud model data;

[0120] The present invention generates a standardized blockchain verification data set by performing data standardization processing on the consensus anti-counterfeiting verification data set of the certificate blockchain. By performing data standardization processing on the consensus anti-counterfeiting verification data set of the certificate blockchain, the dimension difference in the data can be eliminated, so that the data with different characteristics have consistent statistical laws. The standardized blockchain verification data set generated in this way can maintain the consistency of the data and improve the accuracy of data analysis and comparison. Feature engineering transformation processing is performed based on the standardized blockchain verification data set to generate characterized blockchain verification data. The feature selection and feature transformation methods can extract key information and features from the original standardized blockchain verification data, and reduce the influence of redundant and noisy data. Through this information extraction process, the generated characterized blockchain verification data has more information and can better express the intrinsic characteristics and patterns of the data. Anti-fraud model training processing is performed based on the characterized blockchain verification data to generate certificate anti-fraud model data. Anti-fraud decision generation based on model interpretation and analysis can generate personalized decision suggestions for different certificate data. According to the prediction results and interpretation information of the model for the certificate, suggestions on the authenticity recognition, fraud risk, etc. of the certificate can be determined. Such targeted decision suggestions can help users more accurately evaluate the credibility and risk of the certificate and take corresponding measures.

[0121] Step S61: Perform data standardization processing on the certificate blockchain consensus anti-counterfeiting verification data set to generate a standardized blockchain verification data set;

[0122] In the embodiment of the present invention, the certificate blockchain consensus anti-counterfeiting verification data set is preprocessed, including data cleaning, missing value processing, outlier processing, etc., and the preprocessed certificate blockchain consensus anti-counterfeiting verification data set is standardized. Standardization can transform data into a standard distribution that conforms to certain statistical laws, eliminate the dimensional differences of different features, and make the data comparable and interpretable. Commonly used standardization methods include mean-variance standardization, and the final standardized blockchain verification data set is generated based on the results of standardization and statistical feature extraction.

[0123] Step S62: Perform feature engineering transformation processing based on the standardized blockchain verification data set to generate feature-based blockchain verification data;

[0124] In the embodiment of the present invention, feature selection is performed on the standardized blockchain verification data set. Feature selection is to select the most representative and distinguishing features from a large number of features, and feature transformation is performed on the standardized blockchain verification data set after feature selection. Feature transformation is to transform the original features into more meaningful and useful feature representations, which is achieved through mathematical transformation methods. Common feature transformation methods include polynomial transformation. On the basis of feature transformation, feature combination operations are performed. Feature combination is to combine different features to form new features, thereby generating feature-based blockchain verification data.

[0125] Step S63: Perform anti-fraud model training based on the characterized blockchain verification data to generate license anti-fraud model data;

[0126] In an embodiment of the present invention, edge node computing processing is performed on the certificate anti-fraud model data to generate certificate node feature matrix data, heterogeneous graph neural network encoding processing is performed based on the certificate node feature matrix data to generate certificate graph encoding vector data, distributed federated learning training processing is performed on the certificate graph encoding vector data to generate certificate federated learning model parameters, model interpretation and anti-fraud decision generation processing are performed based on the certificate federated learning model parameters to generate certificate anti-fraud model data.

[0127] Preferably, step S63 includes the following steps:

[0128] Step S631: Perform edge node calculation processing on the characterized blockchain verification data to generate certificate node characteristic matrix data;

[0129] Step S632: Perform heterogeneous graph neural network encoding processing based on the certificate node feature matrix data to generate certificate graph encoding vector data;

[0130] Step S633: Perform distributed federated learning training on the certificate image encoding vector data to generate certificate federated learning model parameters;

[0131] Step S634: Perform model interpretation and anti-fraud decision generation processing based on the certificate federated learning model parameters to generate certificate anti-fraud model data.

[0132] The present invention generates certificate node feature matrix data by performing edge node calculation processing on the certificate anti-fraud model data, and can identify and extract edge nodes in the certificate graph by performing edge node calculation processing on the characterized blockchain verification data. These edge nodes usually have relatively few connections, so they have a unique position in the topological structure. By analyzing and calculating the edge nodes, the topological structure of the certificate graph can be deeply understood, and the relationship and importance between the nodes can be revealed. Based on the certificate node feature matrix data, heterogeneous graph neural network encoding processing is performed to generate certificate graph encoding vector data. The heterogeneous graph constructed based on the certificate node feature matrix data can comprehensively represent different types of entities and attributes in the certificate and the relationship between them. Through heterogeneous graph neural network encoding processing, these rich feature information can be effectively encoded into the graph encoding vector, providing a more comprehensive and global description and expression capability for the certificate graph data. Distributed federated learning training processing is performed on the certificate graph encoding vector data to generate certificate federated learning model parameters. Distributed federated learning training processing is adopted, and the certificate graph encoding vector data is divided into multiple data subsets, and model training is performed locally on different participants. This distributed training method avoids direct sharing of original data and effectively protects sensitive certificate image information and privacy content. Through distributed federated learning training, each participant has its own local model and trains and updates it based on a local data subset. In each round of training, the participant merges and updates the local model parameters with the model parameters of other participants, and finally generates the parameters of the federated learning model. This training method allows different participants to conduct personalized model training based on the characteristics of local data, which improves the intelligence and adaptability of the model. Based on the certificate federated learning model parameters, model interpretation and anti-fraud decision generation processing are performed to generate certificate anti-fraud model data. Anti-fraud decision generation based on model interpretation and analysis can generate personalized decision recommendations for different certificate data. Based on the model's prediction results and interpretation information for the certificate, recommendations on the authenticity recognition, fraud risk, etc. of the certificate can be determined. Such targeted decision recommendations can help users more accurately assess the credibility and risk of certificates and take corresponding measures.

[0133] Step S631: Perform edge node calculation processing on the characterized blockchain verification data to generate certificate node characteristic matrix data;

[0134] In an embodiment of the present invention, edge nodes are extracted from the characterized blockchain verification data. Edge nodes refer to nodes that are relatively less connected to other nodes in the certificate graph, and usually have important feature information. Graph theory and network analysis methods are used to identify and extract edge nodes in the certificate graph, and the node features of the extracted edge nodes are calculated. Node features are numerical representations that describe node attributes and relationships, and may include indicators such as node degree, clustering coefficient, and betweenness centrality. By calculating node features, the importance and characteristics of nodes can be revealed from a numerical perspective, and the node features of the extracted edge nodes are calculated. Node features are numerical representations that describe node attributes and relationships, and may include indicators such as node degree, clustering coefficient, and betweenness centrality. By calculating node features, the importance and characteristics of nodes can be revealed from a numerical perspective, and the final certificate node feature matrix data is generated based on the constructed node feature matrix.

[0135] Step S632: Perform heterogeneous graph neural network encoding processing based on the certificate node feature matrix data to generate certificate graph encoding vector data;

[0136] In an embodiment of the present invention, a heterogeneous graph is constructed based on the certificate node feature matrix data. A heterogeneous graph is a graph structure composed of different types of nodes and edges. The node type can represent different entities or attributes, and the edge represents the relationship between nodes. According to the information of the node feature matrix, the connection relationship between the node type and the edge is determined, and based on the constructed heterogeneous graph, a heterogeneous graph neural network model is designed and constructed. A heterogeneous graph neural network is a deep learning model that can process heterogeneous graph data, and can make full use of the relationship between node types and edges for encoding and learning. Select a suitable heterogeneous graph neural network model, such as HAN, and use the heterogeneous graph neural network model to perform graph encoding on the certificate node feature matrix data. By inputting the node features and graph structure into the heterogeneous graph neural network model, the graph data is encoded and embedded. The model can learn the low-dimensional representation of the graph data, convert it into a graph encoding vector, and generate the final certificate graph encoding vector data based on the result of the graph encoding.

[0137] Step S633: Perform distributed federated learning training on the certificate image encoding vector data to generate certificate federated learning model parameters;

[0138] In an embodiment of the present invention, the certificate image encoding vector data is segmented and divided into multiple data subsets. Each data subset contains a portion of the certificate image encoding vector data, which is used for distributed federated learning training. In distributed federated learning, each participant (such as different institutions or devices) has its own local model. Before starting training, these local models need to be initialized and the initial parameter values ​​are set. Through the federated learning algorithm, different participants train their own models locally and fuse and update the models through communication and coordination. In each round of training, the participants update the model according to the local data subset and send the updated model parameters to other participants. After receiving the model parameters of other participants, each participant fuses the aggregated received parameters with its own local parameters. Algorithms such as weighted average and FedAvg can be used to aggregate parameters to achieve the update and improvement of the global model, and the parameters of the certificate federated learning model are generated according to the final parameter fusion results.

[0139] Step S634: Perform model interpretation and anti-fraud decision generation processing based on the certificate federated learning model parameters to generate certificate anti-fraud model data.

[0140] In an embodiment of the present invention, the model is interpreted and analyzed based on the generated certificate federated learning model parameters. Model interpretation technology, such as LIME, can be used to interpret the model's prediction results, reveal the model's decision rules and feature importance, and use the explained model to generate anti-fraud decisions for new certificate data. Based on the model's prediction results and interpretation information for the certificate data, an anti-fraud decision recommendation for the certificate is generated. The decision may include whether to recognize the authenticity of the certificate, whether there is a fraud risk, etc., and the final certificate anti-fraud model data is generated based on the generated anti-fraud decision results. This data set contains anti-fraud decision recommendations and related information for different certificates, which can be used as a basis for subsequent anti-counterfeiting verification and risk assessment.

[0141] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be 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 will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A method for processing certificate anti-counterfeiting data based on digital watermark, characterized in that: The following steps are involved: Step S1: Acquire the certificate image data, perform complex feature extraction and encoding processing on the certificate image data, and generate high-dimensional image feature data; Step S2: performing mixing and digital watermark embedding protection processing based on high-dimensional image feature data to generate mixed domain watermark embedded image data; wherein step S2 includes: Step S21: Calculate the high-dimensional image feature data using a weighted feature fusion regularization formula to generate a weighted distribution feature data set; Among them, the weighted feature fusion regularization formula is as follows: Among them, WFD refers to the weighted feature dataset, m refers to the number of selected features, and w i refers to the weight corresponding to the i-th selected feature, f i (x) refers to the function corresponding to the i-th selected feature, which is a real function, λ refers to a regularization parameter, and x refers to a sample in the high-dimensional image feature data; Step S22: performing hybrid domain mapping processing on the weighted distribution feature data set to generate a hybrid domain mapping data set; Step S23: Performing digital watermark generation processing based on the certificate mixed domain mapping data set to generate certificate digital watermark data; Step S24: Perform watermark embedding processing on the digital watermark data of the certificate to generate mixed domain watermark embedded image data; Step S3: Perform multi-view feature fusion extraction processing on the mixed domain watermark embedded image data to generate multi-view feature fusion data of the certificate; Step S4: Perform complex pattern and anomaly detection processing based on the multi-view feature fusion data of the license, and generate license pattern recognition result data; wherein step S4 includes: Step S41: performing feature hierarchical processing and multimodal mapping processing on the multi-view feature fusion data of the license to generate hierarchical multimodal feature mapping data of the license; Step S42: performing chaotic data distortion and compression processing based on the certificate hierarchical multimodal feature mapping data to generate certificate chaotic compression feature data; Step S43: performing self-encoding driven deep anomaly detection processing on the chaotic compression feature data of the license and credential to generate deep anomaly marking data of the license and credential; Step S44: Performing pattern recognition processing based on graph neural network on the depth abnormality mark data of the license and credential to generate license and credential pattern recognition result data; Step S5: Perform distributed blockchain anti-counterfeiting verification and consensus processing on the certificate pattern recognition result data to generate certificate blockchain consensus anti-counterfeiting verification data; Step S6: Perform anti-fraud and data analysis based on the consensus anti-counterfeiting verification data of the certificate blockchain to generate intelligent anti-fraud model data.

2. The method for processing certificate anti-counterfeiting data based on digital watermark according to claim 1, characterized in that: The specific steps of step S1 are: Step S11: Acquire the certificate image data, perform dynamic data acquisition processing on the certificate image data, and generate an original certificate image data set; Step S12: performing unsupervised deep segmentation processing based on the original certificate image data set to generate a segmented image element set; Step S13: performing heterogeneous feature coding processing on the segmented image element set to generate a certificate feature coding data set; Step S14: Perform high-dimensional feature mapping processing on the certificate feature encoding data set to generate high-dimensional image feature data.

3. The method for processing certificate anti-counterfeiting data based on digital watermark according to claim 1, characterized in that: The specific steps of step S31 are: Step S31: Perform high-dimensional projection space construction processing on the mixed domain watermark embedded image data to generate a high-dimensional projection space data set; Step S32: Perform holographic visual mapping processing based on the high-dimensional projection space data set to generate a certificate holographic visual mapping data set; Step S33: performing high-order feature analysis processing on the certificate holographic visual mapping data set to generate a certificate high-order feature analysis data set; Step S34: Calculate the high-order feature analysis data of the license using Gaussian compression and multi-view fusion calculation formula to generate multi-view feature fusion data of the license.

4. The method for processing certificate anti-counterfeiting data based on digital watermark according to claim 3 is characterized in that: The specific calculation formula of Gaussian compression and multi-view fusion in step S34 is: Among them, F(y) refers to the multi-view feature fusion data of the license, y refers to a sample in the high-order feature analysis data set, k refers to the dimension of the feature space, α j refers to the weight coefficient of the jth feature space, σ j refers to the mean of the jth feature space, μ j refers to the standard deviation of the j-th feature space, β j refers to the compression index of the j-th feature space, γ refers to the weight coefficient of feature fusion, δ refers to an infinite decimal approaching zero, and D refers to the multi-view feature distance function of sample y.

5. The method for processing certificate anti-counterfeiting data based on digital watermark according to claim 1, characterized in that: The specific steps of step S5 are: Step S51: performing pattern recognition result encoding processing on the certificate multi-view feature fusion data to generate a certificate pattern recognition code set; Step S52: Perform distributed data allocation processing based on the certificate pattern recognition code set to generate a distributed pattern recognition result data set; Step S53: Perform blockchain data encapsulation processing on the distributed pattern recognition result data set to generate a license blockchain encapsulation data set; Step S54: Perform consensus algorithm anti-counterfeiting verification processing based on the certificate blockchain encapsulation data set to generate a certificate blockchain consensus anti-counterfeiting verification data set.

6. The method for processing certificate anti-counterfeiting data based on digital watermark according to claim 1, characterized in that: The specific steps of step S6 are: Step S61: Perform data standardization processing on the certificate blockchain consensus anti-counterfeiting verification data set to generate a standardized blockchain verification data set; Step S62: Perform feature engineering transformation processing based on the standardized blockchain verification data set to generate feature-based blockchain verification data; Step S63: Perform anti-fraud model training based on the characterized blockchain verification data to generate certificate anti-fraud model data.

7. The method for processing certificate anti-counterfeiting data based on digital watermark according to claim 6, characterized in that: The specific steps of step S63 are: Step S631: Perform edge node calculation processing on the characterized blockchain verification data to generate certificate node characteristic matrix data; Step S632: Perform heterogeneous graph neural network encoding processing based on the certificate node feature matrix data to generate certificate graph encoding vector data; Step S633: Perform distributed federated learning training on the certificate image encoding vector data to generate certificate federated learning model parameters; Step S634: Perform model interpretation and anti-fraud decision generation processing based on the certificate federated learning model parameters to generate certificate anti-fraud model data.

Images