Deep forgery detection method and system, computer equipment and storage medium

Abstract

The invention relates to a deep forgery detection method and system, computer equipment and a storage medium, and the method comprises the steps: obtaining a data set containing a deep forgery image, the data set having a plurality of feature domains; performing causal relationship analysis about features on the data set to obtain a feature causal graph; constructing a deep neural network model for deep counterfeiting detection; and inputting to-be-detected image data into the deep neural network model for deep forgery detection, and outputting a deep forgery detection result corresponding to the image data. According to the method, a causal inference theory is used for guiding data deviation analysis, an invariant risk minimization strategy is introduced to serve as a loss constraint of a detection model, meanwhile, a detection task is divided into a plurality of subtasks, specific information is reserved and irrelevant deviation is eliminated through feature alignment and a joint training strategy, and the detection accuracy is improved. According to the invention, efficient feature extraction and fusion are realized, and the detection precision of the deep counterfeiting detection model in a complex and changeable scene is improved.

Description

Technical Field

[0001] The present invention is applicable to the field of artificial intelligence technology applications, and in particular relates to a deep fake detection method, system, computer equipment and storage medium. Background Art

[0002] Deepfakes are a technique that superimposes images or videos onto source images or videos, splicing in an individual's voice, facial expressions, and body movements to create fake content. While the widespread use of deepfakes has brought diversity to media creation in recent years, their synthesized content is also susceptible to abuse for identity fraud, dissemination of false information, and digital harassment.

[0003] Most existing deepfake detection methods are based on deep neural networks, which are typically trained using the empirical risk minimization method (ERM). However, this method easily captures irrelevant biased information in the data, resulting in poor generalization performance in real-world settings.

[0004] On the other hand, the construction of data sets used by deep neural networks is usually achieved by simulating fake data from a large amount of real data. However, since the portraits in pictures or videos are affected by deep fake technology of different races and other environmental factors, the data has obvious selective bias and confounding factors, resulting in the existing detection methods being unsatisfactory in practical applications. Summary of the Invention

[0005] The present invention provides a deep fake detection method, system, computer device and storage medium, aiming to solve the technical problems of data bias and poor detection accuracy in existing methods.

[0006] To solve the above technical problems, in a first aspect, the present invention provides a deep fake detection method, comprising the following steps:

[0007] Obtain a dataset comprising a plurality of deep fake images, wherein the dataset has a plurality of feature domains;

[0008] Performing a causal relationship analysis on the features of the data set to obtain a feature causal graph;

[0009] A deep neural network model for deep fake detection is constructed, wherein the deep neural network model includes a detection network, a causal invariance learning module, and a multi-domain causal preservation module, wherein:

[0010] The detection network is used to implement deep fake detection;

[0011] The causal invariance learning module is used to perform deviation analysis based on the feature causal graph, identify causal invariant features, and construct a loss function for the detection network to detect in different feature domains based on an invariant risk minimization method, so that the detection network focuses on the causal invariant features when detecting in different feature domains;

[0012] The multi-domain causality preserving module is used to divide the detection task of the detection network into multiple subtasks, each of which detects a corresponding feature and outputs a corresponding detection sub-result, and all the detection sub-results are integrated and output, wherein the causal invariant features are aligned between different subtasks;

[0013] The image data to be detected is input into the deep neural network model for deep fake detection, and the deep fake detection result corresponding to the image data is output.

[0014] Furthermore, in the causal invariance learning module, the step of performing deviation analysis based on the feature causal graph to identify causal invariance features includes the following sub-steps:

[0015] Dividing the data set according to the feature domain to obtain multiple multi-domain data sets, and performing causal identification on features in the multi-domain data sets according to the feature causal graph to obtain causal features and non-causal features therein;

[0016] Deviation analysis is performed on the causal features and the non-causal features of different multi-domain data sets to obtain features in which the causal relationship remains unchanged in different feature domains, and define them as the causal invariant features.

[0017] Furthermore, the causal invariance learning module is further used to:

[0018] The causal identification process of the feature causal graph is optimized by a gradient weighted method.

[0019] Furthermore, the step of integrating and outputting all the detection sub-results is specifically as follows:

[0020] Assign a weight to each of the detection sub-results, and output the weighted average of the different detection sub-results according to the weight; or

[0021] The fusion layer is used to fuse all the detection sub-results and then output them.

[0022] Furthermore, in the causal invariance learning module, when aligning the causal invariant features between different subtasks, maximum mean difference alignment is used.

[0023] Furthermore, before the step of inputting the image data to be detected into the deep neural network model for deep forgery detection and outputting the deep forgery detection result corresponding to the image data, the method further includes:

[0024] The image data is preprocessed, wherein the preprocessing method includes at least one of data enhancement, label annotation and normalization.

[0025] In a second aspect, the present invention further provides a deep fake detection system, comprising:

[0026] a data acquisition unit, configured to acquire a data set comprising a plurality of deep fake images, wherein the data set has a plurality of feature domains;

[0027] A causal construction unit, configured to perform causal relationship analysis on features of the data set to obtain a feature causal graph;

[0028] A model building unit is used to build a deep neural network model for deep fake detection, wherein the deep neural network model includes a detection network, a causal invariance learning module, and a multi-domain causal preservation module, wherein:

[0029] The detection network is used to implement deep fake detection;

[0030] The causal invariance learning module is used to perform deviation analysis based on the feature causal graph, identify causal invariant features, and construct a loss function for the detection network to detect in different feature domains based on an invariant risk minimization method, so that the detection network focuses on the causal invariant features when detecting in different feature domains;

[0031] The multi-domain causality preserving module is used to divide the detection task of the detection network into multiple subtasks, each of which detects a corresponding feature and outputs a corresponding detection sub-result, and all the detection sub-results are integrated and output, wherein the causal invariant features are aligned between different subtasks;

[0032] The detection unit is used to input the image data to be detected into the deep neural network model for deep fake detection, and output the deep fake detection result corresponding to the image data.

[0033] Furthermore, the causal invariance learning module is further used to:

[0034] Dividing the data set according to the feature domain to obtain multiple multi-domain data sets, and performing causal identification on features in the multi-domain data sets according to the feature causal graph to obtain causal features and non-causal features therein;

[0035] Deviation analysis is performed on the causal features and the non-causal features of different multi-domain data sets to obtain features in which the causal relationship remains unchanged in different feature domains, and define them as the causal invariant features.

[0036] In a third aspect, the present invention also provides a computer device comprising: a memory, a processor, and a deep fake detection program stored on the memory and runnable on the processor, wherein when the processor executes the deep fake detection program, the steps of the deep fake detection method as described in any one of the above embodiments are implemented.

[0037] In a fourth aspect, the present invention also provides a storage medium, on which a deep fake detection program is stored. When the deep fake detection program is executed by a processor, the steps of the deep fake detection method described in any one of the above embodiments are implemented.

[0038] The beneficial effect achieved by the present invention is that a deep fake detection method is proposed, which uses causal inference theory to guide data deviation analysis, introduces an invariant risk minimization strategy as the loss constraint of the detection model, and at the same time, decomposes the detection task into multiple subtasks. It retains specific information and eliminates irrelevant deviations through feature alignment and joint training strategies, thereby realizing efficient feature extraction and fusion, and improving the detection accuracy of the deep fake detection model in complex and changing scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 1 is a schematic diagram of the steps of the deep fake detection method provided by an embodiment of the present invention;

[0040] Figure 2 This is a schematic diagram of a cause-and-effect diagram provided by an embodiment of the present invention;

[0041] Figure 3 Schematic diagram of the logical structure of the deep neural network model provided by an embodiment of the present invention;

[0042] Figure 4 is a schematic structural diagram of a deep fake detection system provided by an embodiment of the present invention;

[0043] Figure 5 It is a structural diagram of a computer device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0044] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0045] Please refer to Figure 1 , Figure 1 : This is a flowchart of the steps of a deep fake detection method provided by an embodiment of the present invention, which includes the following steps:

[0046] S101. Obtain a dataset containing multiple deep fake images, where the dataset has multiple feature domains.

[0047] A feature domain defines and describes a specific aspect of data. Each feature domain represents a specific attribute or feature type of the data. Since it is nearly impossible to collect large-scale, unbiased data directly from the real world, existing deepfake image datasets are generated by applying different fabrication methods to real images. Consequently, these datasets often contain multi-domain characteristics, influenced by factors such as the ethnicity of the individuals depicted, the fabrication method, and the camera equipment used. In other words, each domain has a unique feature distribution.

[0048] S102: Perform causal relationship analysis on features of the data set to obtain a feature causal graph.

[0049] Causal analysis is a method used to determine the causal relationship between features, aiming to understand how one feature affects another, as well as the mechanism and extent of this influence. In deep fake images, certain features show causal patterns that are different from normal images. For example, the lighting effects of objects in fake images and the lighting characteristics of the surrounding environment may not conform to the natural causal relationship. By performing causal analysis on these features, it is helpful to provide a basis for subsequent detection of whether the image has been tampered with and the location and method of tampering. A causal diagram is a graphical tool for intuitively displaying the causal relationship between variables. It graphically presents the causal connection and influence path between various factors. A schematic diagram of a causal diagram is as follows: Figure 2 As shown, Figure 2 It shows the causal relationship between features Z, S, X, and Y. The solid arrows from Z to X and Y indicate that Z is the cause of X and Y; the solid arrows from S to X S and Y solid arrows, indicating that S is X S and Y; and from X V Pointing to Y and from X S The dotted arrow pointing to Y may represent a potential association that has not yet been clarified or needs further exploration, in contrast to the clear causal relationship represented by the solid arrow.

[0050] In general, the feature causal diagram can intuitively reflect the causal relationship between various features, which is conducive to data analysis.

[0051] S103. Construct a deep neural network model for deep fake detection. The deep neural network model includes a detection network, a causal invariance learning module, and a multi-domain causal preservation module. Specifically, Figure 3 As shown, Figure 3 : This is a schematic diagram of the logical structure of the deep neural network model provided by an embodiment of the present invention, wherein:

[0052] The detection network is used to implement deep fake detection;

[0053] The causal invariance learning module is used to perform deviation analysis based on the feature causal graph, identify causal invariant features, and construct a loss function for the detection network to detect in different feature domains based on an invariant risk minimization method (IRM), so that the detection network focuses on the causal invariant features when detecting in different feature domains;

[0054] The multi-domain causal preservation module is used to divide the detection task of the detection network into multiple subtasks, each of which detects a corresponding feature and outputs a corresponding detection sub-result, and all the detection sub-results are integrated and output, wherein the causal invariant features are aligned between different subtasks.

[0055] In an embodiment of the present invention, the detection network serves as the backbone network for image detection and discrimination, and can be implemented based on existing model architectures, such as generative adversarial networks (GANs) and convolutional neural networks (CNNs). In an embodiment of the present invention, a Causal Invariance Learning (CIL) module and a Multi-domain Causal Preservation (MCP) module are added to the backbone network to achieve optimization.

[0056] Specifically, in the causal invariance learning module, the step of performing deviation analysis based on the feature causal graph to identify causal invariance features includes the following sub-steps:

[0057] Dividing the data set according to the feature domain to obtain multiple multi-domain data sets, and performing causal identification on features in the multi-domain data sets according to the feature causal graph to obtain causal features and non-causal features therein;

[0058] Deviation analysis is performed on the causal features and the non-causal features of different multi-domain data sets to obtain the causal invariant features in which the causal relationship remains unchanged in different feature domains.

[0059] Deviation analysis is a data processing method, including statistical analysis, variance analysis, etc. In an embodiment of the present invention, the original data set is first divided into multiple multi-domain data sets by clarifying the different feature domains present in the data set, such as the color, texture, semantics, geometry and other information that a deep fake image may include. Then, based on the feature causal graph, the features represented by each node in the graph and the direction of the causal relationship represented by the edge are clarified for analysis, thereby clarifying the causal features and non-causal features. Among them, causal features are features that have a causal effect on other features, while non-causal features are features that have no causal effect on other features. Although non-causal features themselves do not directly participate in causal relationships, they can be used as control variables or covariates to help eliminate interference from other factors, thereby more accurately inferring causal relationships.

[0060] Afterwards, deviation analysis and comparison are performed based on the different multi-domain data sets to obtain causal invariant features. The causal invariant features refer to causal features in which the causal relationship remains relatively stable in the multi-domain data sets with different feature domains.

[0061] It is understandable that the causal invariant features identified in different multi-domain data sets have a certain degree of universality and stability between different data sets. Therefore, detecting and identifying the causal invariant features can effectively improve the robustness of the detection model.

[0062] The causal invariance learning module is further used to:

[0063] The causal identification process of the feature causal graph is optimized by a gradient weighted method.

[0064] The gradient weighted method is a commonly used optimization method. In the process of causal identification of the feature causal graph by the detection model, it can enhance the influence of important features and suppress the interference of minor features by calculating gradients and assigning weights, thereby improving the accuracy and reliability of causal identification.

[0065] The multi-domain causal preservation module splits the detection task of the detection model into multiple subtasks, and shares the global causal invariant features between the subtasks through feature alignment while preserving the local domain features. Figure 3 As shown in the figure, each subtask relies on the loss function constructed based on the invariant risk minimization method to perform the detection task separately and obtain the corresponding detection sub-result.

[0066] The steps of integrating and outputting all the detection sub-results are specifically as follows:

[0067] Assign a weight to each of the detection sub-results, and output the weighted average of the different detection sub-results according to the weight; or

[0068] The fusion layer is used to fuse all the detection sub-results and then output them.

[0069] By integrating all detection sub-results, the overall deep neural network model has both high generalization and sensitivity to local details.

[0070] In the causal invariance learning module, when aligning the causal invariant features between different subtasks, maximum mean difference alignment is used.

[0071] During the implementation process, the constructed deep neural network model needs to undergo a training process based on the loss function and multi-domain training set to obtain a model with accurate detection capabilities.

[0072] S104: Input the image data to be detected into the deep neural network model to perform deep fake detection, and output a deep fake detection result corresponding to the image data.

[0073] Before the step of inputting the image data to be detected into the deep neural network model for deep forgery detection and outputting the deep forgery detection result corresponding to the image data, the method further includes:

[0074] The image data is preprocessed, wherein the preprocessing method includes at least one of data enhancement, label annotation and normalization.

[0075] The beneficial effect achieved by the present invention is that a deep fake detection method is proposed, which uses causal inference theory to guide data deviation analysis, introduces an invariant risk minimization strategy as the loss constraint of the detection model, and at the same time, decomposes the detection task into multiple subtasks. It retains specific information and eliminates irrelevant deviations through feature alignment and joint training strategies, thereby realizing efficient feature extraction and fusion, and improving the detection accuracy of the deep fake detection model in complex and changing scenarios.

[0076] The present invention also provides a deep fake detection system 200, please refer to Figure 4 , Figure 4 : is a schematic diagram of the structure of a deep fake detection system provided by an embodiment of the present invention, which includes:

[0077] A data acquisition unit 201 is configured to acquire a dataset comprising a plurality of deep fake images, wherein the dataset has a plurality of feature domains;

[0078] A causal construction unit 202 is configured to perform a causal relationship analysis on features of the data set to obtain a feature causal graph;

[0079] The model construction unit 203 is configured to construct a deep neural network model for deep fake detection, wherein the deep neural network model includes a detection network, a causal invariance learning module, and a multi-domain causal preservation module, wherein:

[0080] The detection network is used to implement deep fake detection;

[0081] The causal invariance learning module is used to perform deviation analysis based on the feature causal graph, identify causal invariant features, and construct a loss function for the detection network to detect in different feature domains based on an invariant risk minimization method, so that the detection network focuses on the causal invariant features when detecting in different feature domains;

[0082] The multi-domain causality preserving module is used to divide the detection task of the detection network into multiple subtasks, each of which detects a corresponding feature and outputs a corresponding detection sub-result, and all the detection sub-results are integrated and output, wherein the causal invariant features are aligned between different subtasks;

[0083] The detection unit 204 is used to input the image data to be detected into the deep neural network model to perform deep fake detection, and output the deep fake detection result corresponding to the image data.

[0084] Furthermore, the causal invariance learning module is further used to:

[0085] Dividing the data set according to the feature domain to obtain multiple multi-domain data sets, and performing causal identification on features in the multi-domain data sets according to the feature causal graph to obtain causal features and non-causal features therein;

[0086] Deviation analysis is performed on the causal features and the non-causal features of different multi-domain data sets to obtain features in which the causal relationship remains unchanged in different feature domains, and define them as the causal invariant features.

[0087] The deep fake detection system 200 can implement the steps in the deep fake detection method in the above embodiment and can achieve the same technical effects. Please refer to the description in the above embodiment and will not be repeated here.

[0088] The embodiment of the present invention also provides a computer device, please refer to Figure 5 , Figure 5 300 is a structural diagram of a computer device provided in an embodiment of the present invention. The computer device 300 includes: a memory 302, a processor 301, and a deep fake detection program stored in the memory 302 and executable on the processor 301.

[0089] The processor 301 calls the deep fake detection program stored in the memory 302 and executes the steps of the deep fake detection method provided by the embodiment of the present invention. Figure 1 , the deep fake detection method specifically includes the following steps:

[0090] S101. Obtain a dataset containing multiple deep fake images, where the dataset has multiple feature domains.

[0091] S102: Perform causal relationship analysis on features of the data set to obtain a feature causal graph.

[0092] S103. Construct a deep neural network model for deep fake detection. The deep neural network model includes a detection network, a causal invariance learning module, and a multi-domain causal preservation module, wherein:

[0093] The detection network is used to implement deep fake detection;

[0094] The causal invariance learning module is used to perform deviation analysis based on the feature causal graph, identify causal invariant features, and construct a loss function for the detection network to detect in different feature domains based on an invariant risk minimization method, so that the detection network focuses on the causal invariant features when detecting in different feature domains;

[0095] The multi-domain causal preservation module is used to divide the detection task of the detection network into multiple subtasks, each of which detects a corresponding feature and outputs a corresponding detection sub-result, and all the detection sub-results are integrated and output, wherein the causal invariant features are aligned between different subtasks.

[0096] Specifically, in the causal invariance learning module, the step of performing deviation analysis based on the feature causal graph to identify causal invariance features includes the following sub-steps:

[0097] Dividing the data set according to the feature domain to obtain multiple multi-domain data sets, and performing causal identification on features in the multi-domain data sets according to the feature causal graph to obtain causal features and non-causal features therein;

[0098] Deviation analysis is performed on the causal features and the non-causal features of different multi-domain data sets to obtain features in which the causal relationship remains unchanged in different feature domains, and define them as the causal invariant features.

[0099] The causal invariance learning module is further used to:

[0100] The causal identification process of the feature causal graph is optimized by a gradient weighted method.

[0101] The steps of integrating and outputting all the detection sub-results are specifically as follows:

[0102] Assign a weight to each of the detection sub-results, and output the weighted average of the different detection sub-results according to the weight; or

[0103] The fusion layer is used to fuse all the detection sub-results and then output them.

[0104] In the causal invariance learning module, when aligning the causal invariant features between different subtasks, maximum mean difference alignment is used.

[0105] S104: Input the image data to be detected into the deep neural network model to perform deep fake detection, and output a deep fake detection result corresponding to the image data.

[0106] Before the step of inputting the image data to be detected into the deep neural network model for deep forgery detection and outputting the deep forgery detection result corresponding to the image data, the method further includes:

[0107] The image data is preprocessed, wherein the preprocessing method includes at least one of data enhancement, label annotation and normalization.

[0108] The computer device 300 provided in an embodiment of the present invention can implement the steps in the deep fake detection method in the above embodiment and can achieve the same technical effects. Please refer to the description in the above embodiment and will not be repeated here.

[0109] An embodiment of the present invention also provides a storage medium, on which a deep fake detection program is stored. When the deep fake detection program is executed by a processor, the various processes and steps in the deep fake detection method provided by an embodiment of the present invention are implemented, and the same technical effects can be achieved. To avoid repetition, they will not be repeated here.

[0110] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by using a deep fake detection program to instruct related hardware (which can be a mobile phone, computer, server, air conditioner, or network equipment, etc.). The program can be stored in a computer-readable storage medium, and when executed, the program can include the processes in the above-described method embodiments. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0111] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or apparatus comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or apparatus comprising the element.

[0112] The embodiments of the present invention are described above in conjunction with the accompanying drawings. What is disclosed is only a preferred embodiment of the present invention. However, the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms and equivalent changes without departing from the scope of protection of the purpose of the present invention and the claims, which are all within the protection of the present invention.

Claims

1. A deep fake detection method, characterized in that: The following steps are involved: Obtain a dataset comprising a plurality of deep fake images, wherein the dataset has a plurality of feature domains; Performing a causal relationship analysis on features of the data set to obtain a feature causal graph; A deep neural network model for deep fake detection is constructed, wherein the deep neural network model includes a detection network, a causal invariance learning module, and a multi-domain causal preservation module, wherein: The detection network is used to implement deep fake detection; The causal invariance learning module is used to perform deviation analysis based on the feature causal graph, identify causal invariant features, and construct a loss function for the detection network to detect in different feature domains based on an invariant risk minimization method, so that the detection network focuses on the causal invariant features when detecting in different feature domains; The multi-domain causality preserving module is used to divide the detection task of the detection network into multiple subtasks, each of which detects a corresponding feature and outputs a corresponding detection sub-result, and all the detection sub-results are integrated and output, wherein the causal invariant features are aligned between different subtasks; The image data to be detected is input into the deep neural network model for deep fake detection, and the deep fake detection result corresponding to the image data is output.

2. The deep fake detection method according to claim 1, characterized in that In the causal invariance learning module, the step of performing deviation analysis based on the feature causal graph to identify causal invariance features includes the following sub-steps: Dividing the data set according to the feature domain to obtain multiple multi-domain data sets, and performing causal identification on features in the multi-domain data sets according to the feature causal graph to obtain causal features and non-causal features therein; Deviation analysis is performed on the causal features and the non-causal features of different multi-domain data sets to obtain features in which the causal relationship remains unchanged in different feature domains, and define them as the causal invariant features.

3. The deep fake detection method according to claim 2, characterized in that The causal invariance learning module is further used to: The causal identification process of the feature causal graph is optimized by a gradient weighted method.

4. The deep fake detection method according to claim 1, characterized in that The steps of integrating and outputting all the detection sub-results are specifically as follows: Assign a weight to each of the detection sub-results, and output the weighted average of the different detection sub-results according to the weight; or The fusion layer is used to fuse all the detection sub-results and then output them.

5. The deep fake detection method according to claim 1, characterized in that In the causal invariance learning module, when aligning the causal invariant features between different subtasks, maximum mean difference alignment is used.

6. The deep fake detection method according to claim 1, characterized in that Before the step of inputting the image data to be detected into the deep neural network model for deep forgery detection and outputting the deep forgery detection result corresponding to the image data, the method further includes: The image data is preprocessed, wherein the preprocessing method includes at least one of data enhancement, label annotation and normalization.

7. A deep fake detection system, characterized in that include: a data acquisition unit, configured to acquire a data set comprising a plurality of deep fake images, wherein the data set has a plurality of feature domains; A causal construction unit, configured to perform causal relationship analysis on features of the data set to obtain a feature causal graph; A model building unit is used to build a deep neural network model for deep fake detection, wherein the deep neural network model includes a detection network, a causal invariance learning module, and a multi-domain causal preservation module, wherein: The detection network is used to implement deep fake detection; The causal invariance learning module is used to perform deviation analysis based on the feature causal graph, identify causal invariant features, and construct a loss function for the detection network to detect in different feature domains based on an invariant risk minimization method, so that the detection network focuses on the causal invariant features when detecting in different feature domains; The multi-domain causality preserving module is used to divide the detection task of the detection network into multiple subtasks, each of which detects a corresponding feature and outputs a corresponding detection sub-result, and all the detection sub-results are integrated and output, wherein the causal invariant features are aligned between different subtasks; The detection unit is used to input the image data to be detected into the deep neural network model for deep fake detection, and output the deep fake detection result corresponding to the image data.

8. The deep fake detection system according to claim 7, characterized in that The causal invariance learning module is further used to: Dividing the data set according to the feature domain to obtain multiple multi-domain data sets, and performing causal identification on features in the multi-domain data sets according to the feature causal graph to obtain causal features and non-causal features therein; Deviation analysis is performed on the causal features and the non-causal features of different multi-domain data sets to obtain features in which the causal relationship remains unchanged in different feature domains, and define them as the causal invariant features.

9. A computer device, characterized in that: include: A memory, a processor, and a deep fake detection program stored on the memory and executable on the processor, wherein the processor implements the steps of the deep fake detection method according to any one of claims 1 to 6 when executing the deep fake detection program.

10. A storage medium, characterized in that: The storage medium stores a deep fake detection program, which, when executed by the processor, implements the steps of the deep fake detection method according to any one of claims 1 to 6.

Images