CN119992303B - Transaction anti-counterfeiting identification method based on deep learning - Google Patents
Transaction anti-counterfeiting identification method based on deep learning Download PDFInfo
- Publication number
- CN119992303B CN119992303B CN202510457478.5A CN202510457478A CN119992303B CN 119992303 B CN119992303 B CN 119992303B CN 202510457478 A CN202510457478 A CN 202510457478A CN 119992303 B CN119992303 B CN 119992303B
- Authority
- CN
- China
- Prior art keywords
- low
- model
- dimensional vector
- transaction
- counterfeiting identification
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Landscapes
- Image Analysis (AREA)
- Collating Specific Patterns (AREA)
Abstract
The application relates to a transaction anti-counterfeiting identification method based on deep learning, and relates to the technical field of anti-counterfeiting identification. The application obtains low-dimensional vector representation of an initial transaction identification vector by processing the initial transaction identification vector through a diffusion model, the diffusion model processes the low-dimensional vector representation according to an inverse denoising process to obtain low-dimensional vector representation conforming to Gaussian distribution, anti-fake identification marks are added into a low-frequency region of a low-dimensional vector representation frequency domain, a lova model is configured for a Unet model based on attention, the trained lova model is matched with a Unet model based on attention to denoise the low-dimensional vector representation added with the anti-fake identification marks so as to hide the anti-fake identification marks and obtain a reduction result of the low-dimensional vector representation, a variation self-decoder of the diffusion model is based on the obtained transaction identification vector, in the identification process, the anti-fake identification marks are extracted from the transaction identification vector, and whether the similarity between the extracted and constructed anti-fake identification marks exceeds a set threshold value is compared to perform anti-fake.
Description
Technical Field
The invention relates to the technical field of transaction anti-counterfeiting authentication, in particular to a transaction anti-counterfeiting authentication method based on deep learning.
Background
Steganography is a technique for anti-counterfeiting and authentication by embedding specific information into an information carrier. For example, image steganography refers to hiding secret information in an image on the premise of not changing the visual quality and characterization of the image.
The steganographic embedding may be by modifying pixel values, frequency domain transforms (e.g., DCT, DFT), or steganographic methods based on deep learning. When authentication is needed, the embedded information is extracted from the image through a specific algorithm, and the authenticity of the embedded information is verified. In the steganography method based on deep learning, a diffusion model steganography technology based on prompt words is a novel steganography method, the randomness and the diversity of images are generated by using diffusion model generating capacity, and secret information is hidden in noise or details of the images. The information is embedded and extracted by adjusting noise distribution or sampling strategy in the generation process. The process needs to design effective prompt words to realize steganography, needs higher technology and experience, has difficult optimization convergence of the prompt words and has poor adaptability.
It should be noted that the information disclosed in the above background section is only for enhancing understanding of the background of the present disclosure and thus may include information that does not constitute prior art known to those of ordinary skill in the art.
Disclosure of Invention
In order to solve the technical problems or at least partially solve the technical problems, the invention provides a transaction anti-counterfeiting identification method based on deep learning.
The invention provides a transaction anti-counterfeiting identification method based on deep learning, which comprises the following steps:
processing initial transaction identification vectors by a variational self-encoder of a pre-trained diffusion model Low-dimensional vector characterization to obtain initial transaction identification vector;
Characterization of low-dimensional vectors by an attention-based Unet model of the diffusion model according to an inverse denoising processProcessing to obtain low-dimensional vector representation compliant with Gaussian distributionWherein the low-dimensional vector characterizesN-th step inverse denoising generation implemented by an attention-based Unet model;
adding anti-fake identification mark to low-dimensional vector characterization In the low frequency region of the frequency domain;
A learning lora model is configured for a Unet model based on attention, the low-dimensional vector representation added with anti-counterfeiting identification marks is subjected to denoising process by the trained lora model matched with a Unet model based on attention, so that the anti-counterfeiting identification marks are hidden by the lora model matched with the denoising process realized by a Unet model based on attention, and the low-dimensional vector representation is obtained Is the reduction result of (2)The variational self-decoder of the diffusion model can be based onObtaining transaction authentication vector;
In the identification process, the variation self-encoder of the diffusion model processes the transaction identification vectorAnd then the lora model is matched with a Unet model based on attention, the low-dimensional vector representation of the added anti-counterfeiting identification mark is restored through an inverse denoising process, the anti-counterfeiting identification mark is extracted from the low-dimensional vector representation, and whether the similarity between the extracted anti-counterfeiting identification mark and the constructed anti-counterfeiting identification mark exceeds a set threshold value is compared, so that anti-counterfeiting identification is realized.
The diffusion model further comprises a variation self-encoder, a attention-based and Unet-based model and a variation self-decoder, wherein the attention-based Unet model comprises input encoding units consisting of a residual error module, a multi-head self-attention module and a downsampling module, the input encoding units are cascaded, the input encoding unit at the last stage is connected with an intermediate encoding unit, the intermediate encoding unit at the last stage is composed of the residual error module, the multi-head self-attention module and the residual error module, the intermediate encoding unit at the last stage is cascaded, the intermediate encoding unit at the last stage is connected with an output decoding unit, the output decoding unit is composed of the residual error module, the multi-head self-attention module and the upsampling module, the output decoding unit receives the spliced results of the output of the input decoding unit at the same level and the output of the last stage output decoding unit or the intermediate encoding unit, and the output decoding unit at the last stage is connected with an output layer.
Further, in the inverse denoising process realized based on the Unet model of attention, the Unet model of attention gradually characterizes the low-dimensional vector according to the low-dimensional vector characterization of the nth time step and the Gaussian noise of the nth time step in the denoising process, the predicted Gaussian noise is added to the low-dimensional vector characterization obtained in the n-1 time step inverse denoising process, and the inverse denoising process realized based on the Unet model of attention gradually characterizes the low-dimensional vectorDiffusion to low-dimensional vector characterization following gaussian distribution。
Further, the anti-counterfeiting identification mark is added to the low-dimensional vector representationComprises:
Characterization of low-dimensional vectors Performing two-dimensional Fourier transform, and moving the low-frequency component to the center position of the frequency domain through fftshift operation;
constructing anti-counterfeit authentication marks and referencing low-dimensional vector characterization The frequency domain specification encodes the anti-fake identification mark, and the encoded anti-fake identification mark is added to the low-dimensional vector representationA low frequency part in the frequency domain;
And converting the frequency domain result added with the anti-counterfeiting identification mark into a space domain through two-dimensional inverse Fourier transform to obtain the low-dimensional vector representation added with the anti-counterfeiting identification mark.
Further, the convolution layer and output layer in the residual module of the Unet model based on attention configure the lora parameters to form a lora model.
Further, a denoising process for low-dimensional vector characterization of the anti-counterfeiting identification mark and a transaction identification vector are addedLow-dimensional vector characterization of (2)Training a leavable lora model, wherein the diffusion model and the lora model participate in the training are as follows, and the lora model is matched with a Unet model based on attention to carry out denoising treatment on low-dimensional vector characterization iteration added with the anti-counterfeiting identification mark, so as to sequentially obtain a denoising intermediate state of the low-dimensional vector characterization added with the anti-counterfeiting identification mark......And recovery results during trainingWherein the variational self-decoder of the diffusion model can be based on the restoration results in the training processObtaining a transaction identification vector in the training process; obtaining transaction identification vector in training processThen, the transaction identification vector in the training processTransaction identification vector obtained through variable self-encoder processingLow-dimensional vector characterization of (2)Low-dimensional vector characterization by using lora model in combination with Unet model based on attentionIterative inverse denoising processing is carried out, and the method sequentially obtains......,。
Furthermore, in the process of training the learnable lora model, parameters of the learning-based Unet model are unchanged, only parameters of the lora model are adjusted, and a loss function in the constraint sampling training process is as follows:
The lora model is matched with the restoration result of the denoising process realized by the Unet model based on attention And low-dimensional vector characterizationThe distance between them, the variational self-decoder is based on the result of the restorationGenerated transaction authentication vectorAuthentication vector with initial transactionThe distance between the two models is matched with the attention-based Unet model to identify vectors for transactionLow-dimensional vector characterization of (2)Low-dimensional vector characterization by inverse denoising processKL-divergence summation between the distribution of the low-dimensional vector representation of the increased anti-counterfeit authentication mark.
Further, in the training process, after the transaction identification vector is generated, interference operation is carried out on the transaction identification vector, and then the identifiability of the anti-counterfeiting identification mark is verified, wherein the interference operation adopts any one or a combination of the following operations of compressing interference, adding Gaussian noise interference, gaussian blur interference, scaling interference, partial cutting interference and rotating interference, wherein each element of the transaction identification vector is added with constant interference, and each element of the transaction identification vector is multiplied by constant coefficient interference.
Further, in the authentication process, the product side processes the transaction authentication vector by using the variation self-encoder of the diffusion modelObtaining the low-dimensional vector representation thereofAnd then the private lota model is matched with a Unet model based on attention to restore the low-dimensional vector representation added with the anti-counterfeiting identification mark through an inverse denoising process, two-dimensional Fourier transformation is carried out on the low-dimensional vector representation added with the anti-counterfeiting identification mark to obtain a frequency domain result, the anti-counterfeiting identification mark is extracted from the frequency domain result, and whether the similarity between the extracted anti-counterfeiting identification mark and the constructed anti-counterfeiting identification mark exceeds a set threshold value is compared to realize anti-counterfeiting identification.
Further, the similarity between the extraction of the anti-counterfeiting identification mark and the construction of the anti-counterfeiting identification mark is as follows:
;
wherein, the Respectively extracting the anti-fake identification mark and constructing the average value of the anti-fake identification mark,The variances of the anti-fake identification mark extraction and the anti-fake identification mark construction are respectively,To extract the anti-counterfeit authentication mark and construct the covariance of the anti-counterfeit authentication mark,Is a constant for stabilizing the similarity.
Compared with the prior art, the technical scheme provided by the embodiment of the invention has the following advantages:
the application characterizes the low-dimensional vector according to the inverse denoising process through the Unet model based on the attention of the diffusion model Processing to obtain low-dimensional vector representation compliant with Gaussian distributionLow-dimensional vector characterization to follow gaussian distribution for inverse denoising processAdds anti-fake identification mark instead of low-dimensional vector characterization which is obtained by diffusion process and obeys Gaussian distributionThe anti-counterfeiting identification mark is added, and noise predicted by the Unet model based on attention is added, so that the anti-counterfeiting identification mark is easier to converge in the subsequent training process matched with the lora model of the Unet model based on attention, and the training efficiency is improved.
The application adds the anti-counterfeiting identification mark to the low-dimensional vector representationThe application adds the anti-counterfeiting identification mark code to the low frequency part, and when the high frequency part of the visible representation is determined to change, the anti-counterfeiting identification mark code information of the low frequency part is not lost.
A learning lora model is configured for a Unet model based on attention, the low-dimensional vector representation added with anti-counterfeiting identification marks is subjected to denoising process by the trained lora model matched with a Unet model based on attention, so that the anti-counterfeiting identification marks are hidden by the lora model matched with the denoising process realized by a Unet model based on attention, and the low-dimensional vector representation is obtainedIs the reduction result of (2)The variational self-decoder of the diffusion model can be based onObtaining transaction authentication vector. The original variational self-encoder, variational self-decoder and lora model and the attention-based Unet model in the application actually form a new diffusion model which can use the generating capacity to hide the anti-fake identification mark in the denoising process to obtain the low-dimensional vector characterizationIs the reduction result of (2)The variational self-decoder can be based onObtaining transaction authentication vectorGenerated transaction authentication vectorThe characterization of the initial transaction identification vector is preserved. In the steganography of the image class, the original representation of the carrier image can be reserved. In the identification process, the variation self-encoder of the diffusion model processes the transaction identification vectorAnd then the lora model is matched with a Unet model based on attention, the low-dimensional vector representation of the added anti-counterfeiting identification mark is restored through an inverse denoising process, the anti-counterfeiting identification mark is extracted from the low-dimensional vector representation, and whether the similarity between the extracted anti-counterfeiting identification mark and the constructed anti-counterfeiting identification mark exceeds a set threshold value is compared, so that anti-counterfeiting identification is realized. Because the product side reserves the lora model and the constructed anti-counterfeiting identification mark, even if the diffusion model is an existing model, the privacy of identification can be realized, and the identification is not easy to crack. The method adopts the lora model to participate in the denoising and inverse denoising processes, fully utilizes the robustness of the model, improves the robustness of the encoding and decoding processes of the anti-counterfeiting identification mark, and improves the anti-jamming capability of the anti-counterfeiting identification mark. The lota can directly finely adjust the parameters of the Unet model based on the attention through low-rank matrix decomposition, so that the generated result can be controlled more accurately, and the adaptation capability of the anti-counterfeiting identification mark is stronger compared with the existing prompting word optimization processing.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and together with the description, serve to explain the principles of the invention.
In order to more clearly illustrate the embodiments of the invention or the technical solutions of the prior art, the drawings which are used in the description of the embodiments or the prior art will be briefly described, and it will be obvious to a person skilled in the art that other drawings can be obtained from these drawings without inventive effort.
Fig. 1 is a flowchart of a transaction anti-counterfeiting authentication method based on deep learning according to an embodiment of the present disclosure.
Fig. 2 is a schematic diagram of an overall architecture of a transaction anti-counterfeiting authentication method based on deep learning according to an embodiment of the present disclosure.
Fig. 3 is a schematic diagram of an attention-based Unet model provided by an embodiment of the present disclosure.
FIG. 4 is a representation of the addition of anti-counterfeit authentication markers to low-dimensional vectors provided by an embodiment of the present disclosureIs a schematic diagram of the frequency domain of (a).
FIG. 5 is a flow chart of a process involved in training a model in a lora model process provided by an embodiment of the present disclosure.
Fig. 6 is a schematic diagram of a transaction anti-counterfeiting authentication device based on deep learning according to an embodiment of the present disclosure.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments of the present invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises an element.
Example 1
Referring to fig. 1 and fig. 2 in combination, an embodiment of the present invention provides a transaction anti-counterfeiting authentication method based on deep learning, including:
s100, processing the initial transaction identification vector through a variation self-encoder of a pre-trained diffusion model Low-dimensional vector characterization to obtain initial transaction identification vectorInitial transaction authentication vectorThe parameter distribution of the vector is of a high-dimensional vector, and has a set meaning, such as identification for representing a transaction product.
The pre-trained diffusion model comprises a variance self-encoder, a attention and Unet based model and a variance self-decoder, the variance self-encoder identifies an initial transaction vectorMapping to a low-dimensional space for processing, reducing the complexity of operation in the processing process and reducing the hardware configuration requirement. The attention-based Unet model supports subsequent inverse denoising and denoising processes in a low-dimensional space. The variance self-decoder restores the initial transaction identification vector according to the output result of the attention-based Unet model in the noise reduction process.
As shown in fig. 2, during the diffusion of the diffusion model, iterative low-dimensional vector characterization of the initial transaction identification vectorAdding Gaussian noise, and sequentially obtaining low-dimensional vector characterization......,Wherein the low-dimensional vector characterizesObeying gaussian distribution. In the denoising process, the Unet model based on attention characterizes the predicted noise according to time step coding and the current low-dimensional vector, the predicted noise corresponds to the added noise of the corresponding step in the diffusion process, and the noise is reduced by modeling the predicted noise based on the Unet model based on attention, so that the predicted noise is characterized according to the low-dimensional vectorIterative noise reduction to obtain low-dimensional vector characterization......,. Wherein, the Characterization for low-dimensional vectorsIs characterized by a low-dimensional vector from a decoderIs the reduction result of (2)Restoring the initial transaction authentication vector. Time-step coding is a conventional technique involved in a diffusion model, and prediction based on a Unet model of attention is related to time steps, so that time-step coding is introduced to control prediction.
In the specific implementation process, as shown in fig. 3, the Unet model based on attention comprises an input coding unit consisting of a residual error module, a multi-head self-attention module and a downsampling module, multiple stages of input coding units are cascaded, a last stage of input coding unit is connected with an intermediate coding unit, the intermediate coding unit consists of the residual error module, the multi-head self-attention module and the residual error module, the multiple stages of intermediate coding units are cascaded, the last stage of intermediate coding unit is connected with an output decoding unit, the output decoding unit consists of the residual error module, the multi-head self-attention module and the upsampling module, the output decoding unit receives the splicing result of the output of the same-stage input decoding unit and the output of the last stage of output decoding unit or the intermediate coding unit for decoding, and the last stage of output decoding unit is connected with an output layer. In a specific implementation process, the multi-head self-attention module can be optimized into other types of attention modules such as cross attention and linear attention according to requirements.
In the application, S200, the low-dimensional vector is characterized according to the inverse denoising process through the Unet model based on the attentionProcessing to obtain low-dimensional vector representation compliant with Gaussian distributionWherein the low-dimensional vector characterizesThe nth step inverse denoising result is implemented by the attention-based Unet model. In the inverse denoising process realized based on the Unet model of attention, the Unet model of attention is used for gradually characterizing the low-dimensional vector according to the low-dimensional vector characterization of the nth time step and the predicted Gaussian noise of the nth time step in the denoising process, the predicted Gaussian noise is added into the low-dimensional vector characterization obtained in the n-1 time step inverse denoising process, and the low-dimensional vector is characterized in the inverse denoising process realized based on the Unet model of attentionDiffusion to low-dimensional vector characterization following gaussian distribution。
S300, adding the anti-counterfeiting identification mark to the low-dimensional vector representationIn the low frequency region of the frequency domain. In the specific implementation process, the product party constructs the anti-counterfeiting identification mark and reserves the constructed anti-counterfeiting identification mark for subsequent identification.
In the specific implementation process, as shown in fig. 4, the above process includes:
Characterization of low-dimensional vectors Performing two-dimensional Fourier transform, and moving the low-frequency component to the center position of the frequency domain through fftshift operation;
constructing anti-counterfeit authentication marks and referencing low-dimensional vector characterization The frequency domain specification encodes the anti-fake identification mark, and the encoded anti-fake identification mark is added to the low-dimensional vector representationThe application adds the anti-counterfeiting identification mark code to the low frequency part, and the anti-counterfeiting identification mark code of the low frequency part is not lost when the high frequency part of the visible representation is determined to change.
And converting the frequency domain result added with the anti-counterfeiting identification mark into a space domain through two-dimensional inverse Fourier transform to obtain the low-dimensional vector representation added with the anti-counterfeiting identification mark.
Because the anti-counterfeiting identification mark is added in the low-dimensional vector representation added with the anti-counterfeiting identification mark, the low-dimensional vector representation added with the anti-counterfeiting identification mark cannot be obtained by denoising the low-dimensional vector representation based on the Unet model of attentionAnd accordingly, the variational self-decoder of the diffusion model cannot generate the initial transaction identification vectorIs a result of the reduction of (a). The application aims to enable the diffusion model to obtain the transaction identification vector based on the low-dimensional vector representation of the added anti-counterfeiting identification mark, and the transaction identification vector is consistent with the visible representation of the initial transaction identification vector, so that the transaction identification vector can hide the anti-counterfeiting identification mark while preserving the meaning of the initial transaction identification vector. Therefore, the product provider can restore the low-dimensional vector representation added with the anti-counterfeiting identification mark from the transaction identification vector by using the decoding key reserved in the hand, and further extract the anti-counterfeiting identification mark for anti-counterfeiting identification.
To achieve the above object, the present application performs the following operations:
S400, as shown in fig. 2, a learning-capable lora model is configured to the attention-based Unet model, and the lora model does not affect the parameters of the attention-based Unet model, and the parameters of the lora model exist independently of the attention-based Unet model. In the implementation process, a convolution layer and an output layer in a residual error module of the Unet model based on attention are configured with the lora parameters to form a lora model.
The trained lora model is matched with a Unet model based on attention to carry out a denoising process on the low-dimensional vector representation added with the anti-counterfeiting identification mark, the anti-counterfeiting identification mark is hidden in the denoising process realized by the lora model matched with a Unet model based on attention, and the low-dimensional vector representation is obtainedIs the reduction result of (2),And (3) withWith consistency, the variational self-decoder of the diffusion model can be based onAnd obtaining a transaction identification vector, wherein the obtained transaction identification vector is consistent with the visible representation of the initial transaction identification vector.
To achieve the above object, a denoising process for low-dimensional vector characterization of anti-counterfeit authentication mark needs to be added, and transaction authentication vectorLow-dimensional vector characterization of (2)Training the learner-based lota model. As shown in figures 2 and 5, in training, the diffusion model and the lora model participate in the process that the lora model is matched with a Unet model based on attention to perform denoising treatment on low-dimensional vector characterization iteration added with the anti-counterfeiting identification mark, so that a denoising intermediate state of the low-dimensional vector characterization added with the anti-counterfeiting identification mark is sequentially obtained......And recovery results during trainingWherein the variational self-decoder of the diffusion model can be based on the restoration results in the training processObtaining transaction identification vector in training processObtaining the transaction identification vector in the training processThen, the transaction identification vector in the training processTransaction identification vector in training process is obtained through variation self-encoder processing of diffusion modelLow-dimensional vector characterization of (2)Low-dimensional vector characterization by the lora model in combination with the attention-based Unet modelIterative inverse denoising processing is carried out, and the method sequentially obtains......And low-dimensional vector characterization with added anti-counterfeiting identification mark in training process。
In training, parameters of a Unet model based on attention are unchanged, only parameters of a lora model are adjusted, and a loss function of a constraint sampling training process is a reduction result of a denoising process realized by matching the lora model with a Unet model based on attentionAnd low-dimensional vector characterizationThe distance between them, the variational self-decoder is based on the result of the restorationGenerated transaction authentication vectorAuthentication vector with initial transactionThe distance between the two models is matched with the attention-based Unet model to identify vectors for transactionLow-dimensional vector characterization of (2)Low-dimensional vector characterization by inverse denoising processKL-divergence summation between the distribution of the low-dimensional vector representation of the increased anti-counterfeit authentication mark. Parameters of the lora model are adjusted with the aim of minimizing the loss function. Through the training process, the lora model is matched with the Unet model based on attention to realize hiding of the anti-counterfeiting identification mark in the denoising process, the hiding can be used as a 'decoding key' lora model and a Unet model based on attention to decode in the inverse denoising process, and in the process of hiding of the anti-counterfeiting identification mark, a loss function constrains a reduction resultAnd low-dimensional vector characterizationDistance between them, constrained the variational self-decoder based on the result of the restorationGenerated transaction authentication vectorAuthentication vector with initial transactionDistance between each other, ensure generated transaction identification vectorBasic and initial transaction authentication vectorAnd consistent.
In the specific training process, after the transaction identification vector is generated, interference operation is carried out on the transaction identification vector, and then the identifiability of the anti-counterfeiting identification mark is verified, so that the identification effect of the generated transaction identification vector after being interfered is verified. The interference operation adopts any one or a combination of several operations as follows:
Compressing interference, adding Gaussian noise interference, gaussian blur interference, scaling interference, partially clipping interference, rotating interference, adding a constant interference to each element of the transaction identification vector, and multiplying each element of the transaction identification vector by a constant coefficient interference. Through verification, the parameters of the lora model are ensured to have robustness on the interfered transaction identification vector.
After the transaction identification vector is generated, the transaction identification vector is transmitted to a buyer together with the product in the transaction process. In the specific implementation process, after a buyer in a transaction obtains a product, a corresponding transaction identification vector is provided for a product party, and the product Fang Li extracts an anti-counterfeiting identification mark by using a diffusion model and a lora model in a hand to realize identification.
In the authentication process, the product party processes the transaction authentication vector by using a variation self-encoder of a diffusion modelObtaining the low-dimensional vector representation thereofAnd then the private lota model is matched with a Unet model based on attention to restore the low-dimensional vector representation added with the anti-counterfeiting identification mark through an inverse denoising process, two-dimensional Fourier transformation is carried out on the low-dimensional vector representation added with the anti-counterfeiting identification mark to obtain a frequency domain result, the anti-counterfeiting identification mark is extracted from the frequency domain result, and whether the similarity between the extracted anti-counterfeiting identification mark and the constructed anti-counterfeiting identification mark exceeds a set threshold value is compared to realize anti-counterfeiting identification.
An example similarity is as follows:
;
wherein, the Respectively extracting the anti-fake identification mark and constructing the average value of the anti-fake identification mark,The variances of the anti-fake identification mark extraction and the anti-fake identification mark construction are respectively,To extract the anti-counterfeit authentication mark and construct the covariance of the anti-counterfeit authentication mark,Is a constant for stabilizing the similarity.
Of course, direct comparison and extraction of the anti-counterfeiting identification mark and coding similarity of the anti-counterfeiting identification mark are supported, and the decoding process of the anti-counterfeiting identification mark is omitted.
Example 2
Referring to fig. 6, the invention provides a transaction anti-counterfeiting authentication device based on deep learning, which comprises a sending end and a receiving end, wherein the sending end and the receiving end both comprise a processing unit, a storage unit and a communication unit which are interconnected through a bus, the communication units of the sending end and the receiving end are connected, the storage unit stores a computer program, and the transaction anti-counterfeiting authentication method based on deep learning is realized when the processing unit reads and executes the computer program.
In the specific implementation process, the sending end is mastered by the buyer, the buyer sends the transaction identification vector of the product to the receiving end through the sending end, the receiving end is mastered by the product party, and the product party processes the transaction identification vector by utilizing a variation self-encoder of the diffusion model after receiving the transaction identification vectorObtaining the low-dimensional vector representation thereofAnd then the private lota model is matched with a Unet model based on attention to restore the low-dimensional vector representation added with the anti-counterfeiting identification mark through an inverse denoising process, two-dimensional Fourier transformation is carried out on the low-dimensional vector representation added with the anti-counterfeiting identification mark to obtain a frequency domain result, the anti-counterfeiting identification mark is extracted from the frequency domain result, and whether the similarity between the extracted anti-counterfeiting identification mark and the constructed anti-counterfeiting identification mark exceeds a set threshold value is compared.
Of course, the storage unit in the transaction anti-counterfeiting authentication device based on deep learning provided by the embodiment of the invention is not limited to the method operation described above, and the related operation in the transaction anti-counterfeiting authentication method based on deep learning provided by any embodiment of the invention can be executed.
Example 3
The embodiment of the invention provides a computer readable storage medium, which stores computer instructions, and the computer instructions realize the transaction anti-counterfeiting identification method based on deep learning when being executed by a processor, and the method comprises the following steps:
processing initial transaction identification vectors by a variational self-encoder of a pre-trained diffusion model Low-dimensional vector characterization to obtain initial transaction identification vector;
Characterization of low-dimensional vectors by an attention-based Unet model of the diffusion model according to an inverse denoising processProcessing to obtain low-dimensional vector representation compliant with Gaussian distributionWherein the low-dimensional vector characterizesN-th step inverse denoising generation implemented by an attention-based Unet model;
adding anti-fake identification mark to low-dimensional vector characterization In the low frequency region of the frequency domain;
A learning lora model is configured for a Unet model based on attention, the low-dimensional vector representation added with anti-counterfeiting identification marks is subjected to denoising process by the trained lora model matched with a Unet model based on attention, so that the anti-counterfeiting identification marks are hidden by the lora model matched with the denoising process realized by a Unet model based on attention, and the low-dimensional vector representation is obtained Is the reduction result of (2)The variational self-decoder of the diffusion model can be based onObtaining transaction authentication vector;
In the identification process, the variation self-encoder of the diffusion model processes the transaction identification vectorAnd then the lora model is matched with a Unet model based on attention, the low-dimensional vector representation of the added anti-counterfeiting identification mark is restored through an inverse denoising process, the anti-counterfeiting identification mark is extracted from the low-dimensional vector representation, and whether the similarity between the extracted anti-counterfeiting identification mark and the constructed anti-counterfeiting identification mark exceeds a set threshold value is compared, so that anti-counterfeiting identification is realized.
Of course, the computer readable storage medium provided by the embodiment of the invention stores the computer program which is not limited to the method operation described above, and can also execute the related operation in the transaction anti-counterfeiting identification method based on deep learning provided by any embodiment of the invention.
In the embodiments provided in the present invention, it should be understood that the disclosed structures and methods may be implemented in other manners. For example, the structural embodiments described above are merely illustrative, and for example, the division of the units is merely a logical function division, and there may be other manners of division in actual implementation, for example, multiple units or components may be combined or integrated into another system, or some features may be omitted, or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be an indirect coupling or communication connection via interfaces, structures or units, which may be in electrical, mechanical or other forms.
The units described as separate units may or may not be physically separate, and units shown as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
In addition, each functional unit in the embodiments of the present invention may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit. The integrated units may be implemented in hardware or in software functional units.
The foregoing is only a specific embodiment of the invention to enable those skilled in the art to understand or practice the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims (10)
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202510457478.5A CN119992303B (en) | 2025-04-14 | 2025-04-14 | Transaction anti-counterfeiting identification method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202510457478.5A CN119992303B (en) | 2025-04-14 | 2025-04-14 | Transaction anti-counterfeiting identification method based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN119992303A CN119992303A (en) | 2025-05-13 |
| CN119992303B true CN119992303B (en) | 2025-06-24 |
Family
ID=95639570
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202510457478.5A Active CN119992303B (en) | 2025-04-14 | 2025-04-14 | Transaction anti-counterfeiting identification method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN119992303B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117376484A (en) * | 2023-12-05 | 2024-01-09 | 北京邮电大学 | A generative steganography method for anti-counterfeiting of electronic certificates |
| CN119129620A (en) * | 2024-08-16 | 2024-12-13 | 江苏诺真防伪科技有限公司 | Trademark identification method and system based on anti-counterfeiting coding |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US12231151B1 (en) * | 2017-10-30 | 2025-02-18 | Atombeam Technologies Inc | Federated large codeword model deep learning architecture with homomorphic compression and encryption |
| CN115643348B (en) * | 2022-12-26 | 2023-04-21 | 中国科学技术大学 | Provably secure natural steganography method and device based on reversible image processing network |
-
2025
- 2025-04-14 CN CN202510457478.5A patent/CN119992303B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117376484A (en) * | 2023-12-05 | 2024-01-09 | 北京邮电大学 | A generative steganography method for anti-counterfeiting of electronic certificates |
| CN119129620A (en) * | 2024-08-16 | 2024-12-13 | 江苏诺真防伪科技有限公司 | Trademark identification method and system based on anti-counterfeiting coding |
Also Published As
| Publication number | Publication date |
|---|---|
| CN119992303A (en) | 2025-05-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Sinhal et al. | Machine learning based blind color image watermarking scheme for copyright protection | |
| Tsougenis et al. | Adaptive color image watermarking by the use of quaternion image moments | |
| CN116645260B (en) | Digital watermark attack method based on conditional diffusion model | |
| Pérez-González et al. | Approaching the capacity limit in image watermarking: a perspective on coding techniques for data hiding applications | |
| Zhang et al. | Robust data hiding using inverse gradient attention | |
| KR101980886B1 (en) | Method and apparatus for embedding watermark | |
| CN117437108B (en) | A watermark embedding method for image data | |
| Zareian et al. | A novel gain invariant quantization-based watermarking approach | |
| CN112465687B (en) | Image hiding method and device | |
| Awasthi et al. | An optimized and secured image watermarking and its dual authentication for internet of medical things | |
| Wei et al. | A robust image watermarking approach using cycle variational autoencoder | |
| CN115829819A (en) | Neural network-based image robust reversible information hiding method, device and medium | |
| Wang et al. | Lossless image steganography: Regard steganography as super-resolution | |
| CN116485622A (en) | Multi-level Network Deep Learning Robust Blind Watermarking Algorithm and System Combined with CNN-Transformer | |
| You et al. | Two-stage watermark removal framework for spread spectrum watermarking | |
| Alia et al. | Robust Image Watermarking using DWT, DCT, and PSO with CNN-Based Attack Evaluation. | |
| CN119992303B (en) | Transaction anti-counterfeiting identification method based on deep learning | |
| CN108711130B (en) | Image watermarking system and method based on compressed sensing noise reconstruction | |
| CN119562067A (en) | Cross-modal steganography method, device, equipment and storage medium | |
| CN115086674B (en) | Image steganography method based on generation of countermeasure network | |
| CN116630124A (en) | High-robustness watermark embedding method based on AI model | |
| Ferik et al. | A blind digital watermarking approach using palmprint-embedded local binary patterns and arnold cat map transformations | |
| CN116402669A (en) | A Nonlinear Blind Watermark Embedding and Extraction Method | |
| CN114493971A (en) | Media data conversion model training and digital watermark embedding method and device | |
| Li et al. | Robust watermarking for diffusion models based on STDM and latent space fine-tuning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |