AIGC deep pseudo detection method and system based on synchronous coupling of double cranial nerves

Abstract

The invention discloses an AIGC deep pseudo detection method based on synchronous coupling of double cranial nerves, and the method comprises the following steps: S1, designing a double-person super-scanning experiment normal form under the stimulation of a digital multimedia material, synchronously collecting double-person electroencephalogram signals, and constructing a double-person electroencephalogram database generated by real and AIGC under the stimulation of the multimedia material; s2, constructing a depth separable neural network based on an attention mechanism; and S3, filtering the collected double electroencephalogram signals to obtain four signals with different frequency bands, calculating a phase lock value PLV, stacking to obtain a three-dimensional vector, inputting the three-dimensional vector into the deep separable neural network based on the attention mechanism, and predicting the authenticity of the content of the multimedia stimulation material. The invention further discloses an AIGC deep pseudo detection system based on synchronous coupling of double cranial nerves. According to the method, the information gain and neural synchronization algorithm of double brain electricity are fully utilized, the synchronous response of double brain nerves is analyzed, and AIGC deep pseudo detection inspired by a neural coupling mechanism is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of authenticity detection of artificial intelligence generated content (AIGC) in the cross - direction of computer ethics, brain - computer interface, and information security, and particularly relates to a method and system for AIGC deepfake detection based on dual - brain nerve synchronous coupling. Background Art

[0002] Currently, the authenticity detection of AIGC (the process of using artificial intelligence technology to generate content) mainly relies on the following technical means, but there are significant limitations: 1. Traditional visual / auditory detection methods: rely on machine learning models to analyze the pixel or spectral features of images / audio, such as GAN generation trace detection. Traditional methods usually rely on manually designed features or simple machine learning models to extract the features of images or audio. These features often fail to capture sufficient details and patterns when facing complex AIGC - generated content, resulting in a decline in detection accuracy. With the rapid development of AIGC technology, new generation models such as diffusion models continue to emerge, and traditional methods seem inadequate in dealing with these new technologies; 2. Single - subject (single - person) EEG detection: Decode the neural response differences of AIGC through individual electroencephalogram signals, such as classification based on event - related potentials or spectral features. Although there have been studies using single - subject EEG for authenticity detection, this method has certain limitations. Single - subject detection cannot capture the interaction effects and group dynamics among multiple people. In a social environment with frequent interpersonal communication and interaction, the authenticity detection of AIGC needs to consider the interactions and emotional exchanges among multiple people, while single - subject EEG detection is difficult to provide information in this regard. Facing materials generated by AIGC, the human eye cannot distinguish between true and false, but there may be differences in neuronal responses and induced brain waves. Dual - brain electroencephalogram plays an effect of data fusion and information gain, and dual - brain neural synchronous response coupling becomes the key differentiating information.

[0003] Therefore, there is an urgent need to provide a new method and system for AIGC deepfake detection based on dual - brain nerve synchronous coupling to solve the above problems. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a method and system for AIGC deepfake detection based on dual - brain nerve synchronous coupling, which can enhance the frequency - band representation of electroencephalogram signals, capture the dual - brain neural interaction responses and dynamics, and provide a technology for detecting the authenticity of multimedia information with the help of dual - brain responses and deep learning.

[0005] To solve the above technical problems, a technical solution adopted by the present invention is: to provide an AIGC deepfake detection method based on dual-brain nerve synchronous coupling, including the following steps: S1: Design a dual-person hyperscanning experimental paradigm under digital multimedia material stimulation, synchronously collect the electroencephalogram (EEG) signals of two persons, and construct a dual-person EEG database under real and AIGC-generated multimedia material stimulation; S2: Construct a depthwise separable neural network based on the attention mechanism, and combine the attention mechanism with depthwise separable convolution to process the EEG signal features; S3: Filter the collected dual-person EEG signals to obtain signals in four different frequency bands, then calculate the phase locking value (PLV) of the four frequency band signals, and finally stack the four frequency band PLVs spatially to obtain a three-dimensional vector and input it into the depthwise separable neural network based on the attention mechanism to predict the authenticity of the hyperscanning data content.

[0006] In a preferred embodiment of the present invention, in step S1, the design method of the dual-person hyperscanning experimental paradigm includes: The experimental conditions are manipulated by three factors: modality, valence, and life type. The modality factor includes three levels: image, text, and audio. The valence condition includes three levels: positive, neutral, and negative. The life type factor distinguishes living and non-living according to the nature of the stimulus. The living experimental objects include human faces, human voices, and texts containing conversations. The non-living stimulus materials include photos of daily items, natural sounds, and texts without any description of people or conversations; The main task of the participants is to evaluate the authenticity of various presented materials, including images, texts, and audio clips. The participants are required to identify whether each stimulus is real or forged by AIGC based on their subjective evaluation.

[0007] Furthermore, the dual-person hyperscanning experimental paradigm uses the inter-subject correlation (ISC) to measure the synchrony of the brain responses of two persons when receiving the same external audio-visual stimuli, and verify the effectiveness of the experimental paradigm design.

[0008] In a preferred embodiment of the present invention, in step S2, the construction method of the depthwise separable neural network based on the attention mechanism includes: S201: The input vector X1 is first globally average pooled into a one-dimensional vector of size 1*1*C, where C is the feature dimension of X1. Then, the average pooled one-dimensional vector is put into two fully connected layers to learn the non-linear relationship between channels. In the last layer, a weight vector for each channel is generated through the Sigmoid activation function. Finally, the weight vector and the original feature map are multiplied channel by channel to obtain a weighted feature vector of N*N*4, where N is the number of electrodes; S202: Convolve the four dimensions of the X2 feature vector with four convolutional kernels of size 1*N respectively to obtain a feature vector X3 of size N*1*4; S203: Convolve the feature vector X3 with four convolutional kernels of size 1*1 to obtain a feature vector X4; S204: Flatten the feature vector X4 to obtain a vector of size (N*4)*1, and input this vector into a softmax classifier for binary classification to obtain 0 or 1, which is the authenticity of the EEG induced by the multimedia stimulus material.

[0009] In a preferred embodiment of the present invention, in step S3, the phase locking value PLV is selected as the synchronization index of the depthwise separable neural network model based on the attention mechanism. The calculation method of the phase locking value PLV is as follows: First, calculate the phase locking value PLV of each subject at each time point and each frequency band:

[0010] where θ1(t, n) and θ2(t, n) are the instantaneous phases of the nth channel of each subject, N is the number of channels, and j is the imaginary unit; Then, average the total PLV over a period of time at all time points.

[0011] To solve the above technical problems, another technical solution adopted by the present invention is: to provide an AIGC deepfake detection system based on dual-brain neural synchronization coupling, including: A dual-person AIGC EEG database construction module, which is used to design a dual-person hyperscanning experimental paradigm under the stimulation of digital multimedia materials, synchronously collect dual-person EEG signals, and construct a dual-person EEG database under the stimulation of real and AIGC-generated multimedia materials; A depthwise separable neural network construction module based on the attention mechanism, which is used to construct a depthwise separable neural network based on the attention mechanism and combine the attention mechanism with depthwise separable convolution to process EEG signal features; A module for predicting data authenticity, which is used to filter the collected dual-person EEG signals to obtain signals in four different frequency bands, then calculate the phase locking value PLV of the four frequency band signals, and finally stack the four frequency band PLVs spatially to obtain a three-dimensional vector and input it into the depthwise separable neural network based on the attention mechanism to predict the authenticity of the hyperscanning data content.

[0012] In a preferred embodiment of the present invention, the method for the dual-person AIGC EEG database construction module to design the dual-person hyperscanning experimental paradigm includes: The experimental conditions were manipulated by three factors: modality, valence, and type of life. The modality factor included three levels: image, text, and audio. The valence condition included three levels: positive, neutral, and negative. The type of life factor distinguished between living and non-living objects according to the nature of the stimuli. The living experimental objects included human faces, human voices, and texts containing conversations. The non-living stimulus materials included photos of daily objects, natural sounds, and texts without any description of people or conversations. The main task of the participants was to evaluate the authenticity of various presented materials, including images, texts, and audio clips. The participants were required to identify whether each stimulus was real or generated by AIGC based on their subjective evaluation.

[0013] Furthermore, the dual-person hyperscanning experimental paradigm used the inter-subject correlation (ISC) to measure the synchrony of the brain responses of two people when receiving the same external audiovisual stimuli, and to verify the effectiveness of the experimental paradigm design.

[0014] In a preferred embodiment of the present invention, the depthwise separable neural network based on the attention mechanism constructed by the depthwise separable neural network construction module based on the attention mechanism includes: A channel attention mechanism module, which first globally average pools the input vector X1 into a one-dimensional vector of size 1*1*C, where C is the feature dimension of X1. Then, the average-pooled one-dimensional vector is put into two fully connected layers to learn the non-linear relationship between channels. In the last layer, a weight vector for each channel is generated through the Sigmoid activation function. Finally, the weight vector is multiplied with the original feature map channel by channel to obtain a weighted N*N*4 feature vector, namely X2, where N is the number of electrodes. A depth convolution module, which convolves each of the four dimensions of the X2 feature vector output by the channel attention mechanism module with four convolution kernels of size 1*N to obtain a N*1*4 feature vector X3. A pointwise convolution module, which convolves the feature vector X3 output by the depth convolution module with four convolution kernels of size 1*1 to obtain a feature vector X4. A flattening module, which flattens the feature vector X4 output by the pointwise convolution module to obtain a (N*4)*1 vector. A classifier module, which inputs the (N*4)*1 vector output by the flattening module into a softmax classifier for binary classification to obtain 0 or 1, which is the authenticity of the electroencephalogram induced by the multimedia stimulus material.

[0015] In a preferred embodiment of the present invention, the method for the prediction data authenticity module to calculate the phase locking value (PLV) is as follows: First, calculate the phase-locking value (PLV) of each subject at each time point and for each frequency band:

[0016] where θ1(t, n) and θ2(t, n) are the instantaneous phases of the nth channel of each subject, N is the number of channels, and j is the imaginary unit.

[0017] Then, average the total PLV over a period of time at all time points.

[0018] The beneficial effects of the present invention are as follows: (1) The present invention proposes an AIGC deepfake content recognition technology based on a dual-brain EEG synchronization hyperscanning experimental paradigm, neural synchronization decoupling, and an improved depthwise separable neural network. By collecting dual-brain EEG signals, an EEG database of dual-person synchronous recordings under real and AIGC forged materials stimuli is constructed, enabling EEG detection to capture the interaction effects between the two persons and accurately detect AIGC forged materials; (2) The present invention develops an advanced neural architecture, namely a depthwise separable neural network based on the attention mechanism. The attention mechanism is combined with depthwise separable convolution to process PLV features. The adaptive reweighting amplifies the feature channels related to the task while attenuating those less important for the classification task. After the reweighting driven by the attention mechanism, the refined feature map is fed into the depthwise separable convolutional layer. This efficient convolutional architecture separates spatial filtering and channel filtering, enhancing the network's ability to capture subtle inter-frequency dependencies while maintaining computational efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 is a flowchart of the AIGC deepfake detection method based on dual-brain neural synchronization coupling of the present invention; Figure 2 is a diagram of the dual-person hyperscanning experimental paradigm designed by the present invention; Figure 3 is a structural diagram of the depthwise separable neural network based on the attention mechanism; Figure 4 is a schematic diagram of the inter-subject correlation (ISC) results of four frequency bands in different modalities between the real task and the AIGC task; Figure 5 is a schematic diagram of the inter-subject correlation (ISC) results under different task conditions based on valence; Figure 6 is a schematic diagram of the inter-subject correlation (ISC) results based on the classification of living and non-living materials; Figure 7 is a structural block diagram of the AIGC deepfake detection system based on dual-brain neural synchronization coupling. Specific embodiments

[0020] The following elaborates on the preferred embodiments of the present invention in conjunction with the accompanying drawings, so that the advantages and features of the present invention can be more easily understood by those skilled in the art, thereby making the protection scope of the present invention more clearly defined.

[0021] Please refer to Figure 1 , the embodiments of the present invention include: An AIGC deepfake detection method based on dual-brain neural synchronous coupling, comprising the following steps: S1: Design a dual-person hyperscanning experimental paradigm under digital multimedia material stimulation, synchronously collect the electroencephalogram (EEG) signals of two people, and construct a dual-person EEG database under real and AIGC-generated multimedia material stimulation; As Figure 2 shown, the experimental conditions are manipulated through three key factors: (1) modality, (2) valence, and (3) type of life. The main task of the participants is to evaluate the authenticity of various materials presented to them. These materials include images, texts, and audio clips, which may be real or generated by AIGC. The participants are required to identify whether each stimulus is real or forged by AIGC based on their subjective evaluation. The modality factor includes three levels: image, text, and audio. The valence condition includes three levels: positive, neutral, and negative. The type-of-life factor distinguishes between living and non-living entities according to the nature of the stimulus. Living experimental objects include human faces, human voices, and texts containing conversations. Non-living stimulus materials include photos of everyday objects, natural sounds (such as ambient sounds), and texts without any description of people or conversations. Therefore, the experiment adopts a 3 (modality: image, text, audio) × 3 (valence: positive, neutral, negative) × 2 (type of life: living, non-living) design. The above real images, texts, and audios are all from open-source datasets, and the AI images, texts, and audios are all from the generation results of open-source models.

[0022] The present invention uses intersubject correlation (ISC) to measure the synchrony of brain responses of participants when viewing the same stimulus. This is particularly suitable for a naturalistic stimulus paradigm that conforms to real-life scenarios because it allows measuring the consistency of neural responses between two individuals to natural stimulus multimedia materials without the need to pre-define the periodic occurrence of stimulus events in the form of event locking, which is the typical event-related potential (ERP) paradigm for cognitive psychology research in laboratory scenarios. First, for each pair of participants k and l, calculate the cross-electrode covariance matrix , which is:

[0023] Among them, X k (t) and X l (t) respectively represent the electroencephalogram signals of two individuals k and l at time point t. μ k and μ l respectively represent the means of individuals k and l in the entire time series, usually the average of each individual's signal at all time points. T is the total number of time points of the signal. For each time point t, calculate the signals X k (t) and X l (t) and their respective means μ k and μ l of the deviations. The sum of the products of these two deviations reflects the change synchronization of the two individual signals at each time point. The average of the products of all time points is calculated to obtain the covariance C kl .

[0024] Then, use the following formula to calculate the ISC value for each pair of participants:

[0025] Here, is the i-th eigenvector, usually from principal component analysis (PCA) or other feature extraction methods, used to represent the features of the signal. T is the total number of time points, usually the length of the signal or the number of time frames in the experiment. For each pair of time points t, calculate the dot product of each individual signal and the eigenvector X k (t)*Vi and X l (t)*V i , respectively sum the squares of these two dot products (the numerator and denominator) to obtain their respective weighted signal strengths. The numerator part is the sum of the products of the weighted signals of the two individuals at each time point, reflecting their signal synchronization. The denominator is the product of the weighted squared sums of the signals for normalization. Finally, the result of ISC is obtained by calculating the ratio of the two, reflecting the degree of synchronization of the two individuals in time under the given eigenvector V i .

[0026] Through the study of inter-subject correlation (ISC), it is found that the inter-subject correlations (ISC) corresponding to the four frequency bands (δ, θ, α, β) under different conditions all show significant differences, indicating that the four frequency bands show different neural synchronization responses when the brain discriminates true and false materials at the level of social cognitive interpersonal neuroscience.

[0027] S2: Construct a depthwise separable neural network based on the attention mechanism, and combine the attention mechanism with depthwise separable convolution to process the electroencephalogram signal features; The present invention selects PLV (Phase Locking Value) as the synchronization index of the model:

[0028] where θ1(t, n) and θ2(t, n) are the instantaneous phases of the nth channel of each subject, N is the number of channels, and j is the imaginary unit. Finally, the total PLV of the experiment is averaged over all time points. The PLV value ranges from 0 to 1, and the higher the value, the stronger the synchronization.

[0029] S3: Filter the collected EEG signals of two people to obtain signals in four different frequency bands, then calculate the phase locking value PLV for the four frequency band signals. The PLV calculated for each frequency band is a matrix with a size of N*N. Finally, stack the four frequency band PLVs spatially to obtain a three-dimensional vector of N*N*4 and input it into the depthwise separable neural network based on the attention mechanism to predict the authenticity of the hyperscanning data content. In this example, N is taken as 32.

[0030] Specifically, the depthwise separable neural network based on the attention mechanism (SEDSC) is as Figure 3 shown, which combines the attention mechanism with depthwise separable convolution to process these PLV features.

[0031] In the channel attention mechanism module, the input vector X1 is first globally average pooled (Squeeze, i.e., Fsq(·)) into a one-dimensional vector of size 1*1*C. This operation can aggregate spatial information and capture the global distribution characteristics of the channels, enabling subsequent operations to adjust the channel weights based on the global receptive field (C is the dimension 4); then perform the Fex(·, W) operation, put the averaged one-dimensional vector into two fully connected layers to learn the non-linear relationship between channels, and generate a weight vector for each channel through the Sigmoid activation function in the last layer. W is the parameter matrix of the fully connected layer; thus, a one-dimensional weight vector is obtained. Finally, multiply it with the original feature map channel by channel to obtain a weighted feature vector of N*N*4, namely X2.

[0032] In the depth convolution module, the four dimensions of the X2 feature vector are respectively convolved with four convolution kernels of size 1*N to obtain a feature vector X3 of N*1*4. Convolve X3 with four convolution kernels of size 1*1 to obtain X4. Then flatten X3 to obtain a vector of (N*4)*1. Finally, input this vector into the softmax classifier for binary classification to obtain 0 or 1 (corresponding to false or true).

[0033] In this neural network, adaptive reweighting amplifies the task-related feature channels while attenuating those that are less important for the classification task. After the attention mechanism-driven reweighting, the refined feature map is fed into a depthwise separable convolutional layer. This efficient convolutional architecture separates spatial filtering and channel filtering, enhancing the network's ability to capture fine-grained frequency dependencies while maintaining computational efficiency.

[0034] Using the method described in the present invention, three conditions (modality, valence, and type of life) all show obvious detection results: (1)Inter-subject correlations in four frequency bands under different modalities Figure 4 It shows significant differences in inter-subject correlations (ISC) between real tasks and AIGC tasks, presenting different activation patterns under different conditions. In text and audio tasks, ISC is always higher in real tasks than in AIGC tasks. Specifically, in the β and α bands, the ISC values in real tasks are significantly greater than those in AIGC tasks (p < 0.0001). In audio tasks, there are differences in both the low-frequency and high-frequency bands. The difference in the δ band is significant (p < 0.01), and the difference in the β band is also significant (p < 0.05). However, in image tasks, the pattern is reversed: ISC is significantly higher during AIGC tasks than in real tasks, especially in the β band (p < 0.01).

[0035] (2)Inter-subject correlations in four frequency bands under different valences Figure 5 It shows the results of inter-subject correlations (ISC) under different task conditions based on valence. In negative tasks, there are significant differences in ISC caused by real materials in all frequency bands, and the ISC generated by real materials is always higher than that of AIGC materials. Among them, the difference in ISC in the β band is extremely significant (p < 0.01), and the differences in ISC in the α, θ, and δ bands are all significant (p < 0.05). In positive tasks, the ISC induced by real materials shows strong significance in both the θ and α bands (p < 0.01), while the ISC induced by AIGC materials is weaker. In neutral tasks, there is no significant difference in ISC between real materials and AIGC materials (p > 0.05). However, in all tasks, the ISC induced by real materials is always higher than that induced by AIGC materials.

[0036] (3)Inter-subject correlations in four frequency bands under different valences Figure 6 It shows the results of inter-subject correlations (ISC) based on the classification of living and non-living materials. The ISC induced by real materials in the living group has significant differences in both the θ and α bands, with the significance level being p < 0.05.

[0037] In addition, the network SEDSC (Queeze-and-Excitation-Enhanced Depth- wise Separable Convolutional Network) proposed in the present invention was compared with RFDSC (Row-Filtered Depthwise Separable Convolution Network) and a traditional CNN (with a convolutional kernel size of 3), and five-fold cross-validation was used, as shown in Table 1.

[0038] Table 1 Classification results of SEDSC compared with other models

[0039] The model described in the present invention achieved high classification accuracy under 5 different data input formats. The accuracy was always better than that of CNN and RFDSC, and the highest classification accuracy reached 92.42%, with an increase of nearly 10% in all frequency bands. This shows that the addition of the attention mechanism enhanced the representation of the frequency bands, thus improving the performance of SEDSC in the classification task.

[0040] Refer to Figure 7 , and in the example of the present invention, an AIGC deepfake detection system based on dual-brain nerve synchronous coupling is also provided, including: A dual-person AIGC EEG database construction module, which is used to design a dual-person hyperscanning experimental paradigm under digital multimedia material stimulation, synchronously collect dual-person EEG signals, and construct a dual-person EEG database under real and AIGC-generated multimedia material stimulation; A depthwise separable neural network construction module based on the attention mechanism, which is used to construct a depthwise separable neural network based on the attention mechanism, and combine the attention mechanism with depthwise separable convolution to process EEG signal features; A module for predicting data authenticity, which is used to filter the dual-person EEG signals collected by the dual-person AIGC EEG database construction module to obtain signals in four different frequency bands, then calculate the phase-locking value PLV of the four frequency band signals, and finally stack the four frequency band PLVs spatially to obtain a three-dimensional vector of N*N*4 and input it into the depthwise separable neural network based on the attention mechanism to predict the authenticity of the multimedia stimulus material content An AIGC deepfake detection system based on dual-brain nerve synchronous coupling in this example can execute an AIGC deepfake detection method provided by the present invention, can execute any combination implementation steps of the method example, and has the corresponding functions and beneficial effects of the method.

[0041] The above are only embodiments of the present invention, and do not thereby limit the patent scope of the present invention. Any equivalent structural or equivalent process transformations made by using the content of the specification and drawings of the present invention, or directly or indirectly applied in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. An AIGC deepfake detection method based on dual-brain nerve synchronous coupling, characterized in that It includes the following steps: S1: Design a dual - person hyperscanning experimental paradigm under digital multimedia material stimulation, synchronously collect the electroencephalogram (EEG) signals of two persons, and construct a dual - person EEG database under real and AIGC - generated multimedia material stimulation; S2: Construct a depth - separable neural network based on the attention mechanism, and combine the attention mechanism with depth - separable convolution to process the EEG signal features; S3: Filter the dual - person EEG signals collected in step S1 to obtain signals in four different frequency bands, then calculate the phase - locking value (PLV) of the four - band signals, and finally stack the four - band PLVs spatially to obtain a three - dimensional vector and input it into the depth - separable neural network based on the attention mechanism constructed in step S2 to predict the authenticity of the multimedia stimulus material content.

2. The AIGC deepfake detection method based on dual-cerebral nerve synchronous coupling according to claim 1, wherein, In step S1, the design method of the dual - person hyperscanning experimental paradigm includes: The experimental conditions are manipulated by three factors: modality, valence, and type of life. The modality factor includes three levels: image, text, and audio. The valence condition includes three levels: positive, neutral, and negative. The type - of - life factor distinguishes between living and non - living according to the nature of the stimulus. The living experimental objects include human faces, human voices, and texts containing conversations. The non - living stimulus materials include photos of daily items, natural sounds, and texts without any description of people or conversations; The main task of the participants is to evaluate the authenticity of the various presented materials, which include images, texts, and audio clips. The participants are required to identify whether each stimulus is real or forged by AIGC based on their subjective evaluation.

3. The AIGC deepfake detection method based on dual-cerebral nerve synchronous coupling according to claim 2, wherein The dual - person hyperscanning experimental paradigm uses the inter - subject correlation (ISC) to measure the synchrony of the brain responses of two persons when receiving the same external audio - visual stimuli, and verify the effectiveness of the experimental paradigm design.

4. The AIGC deepfake detection method based on dual-brain nerve synchronous coupling according to claim 1, wherein In step S2, the construction method of the depth - separable neural network based on the attention mechanism includes: S201: The input vector X1 is first globally average pooled into a one-dimensional vector of size 1*1*C, where C is the feature dimension of X1. Then, the average-pooled one-dimensional vector is fed into two fully connected layers to learn the non-linear relationships between channels. At the last layer, a weight vector for each channel is generated through the Sigmoid activation function. Finally, the weight vector and the original feature map are multiplied channel by channel to obtain a weighted feature vector of N*N*4, i.e., X2, where N is the number of electrodes; S202: Convolve the four dimensions of the X2 feature vector with four convolutional kernels of size 1*N respectively to obtain a feature vector X3 of size N*1*4; S203: Convolve the feature vector X3 with four convolutional kernels of size 1*1 to obtain the feature vector X4; S204: Flatten the feature vector X4 to obtain a vector of size (N*4)*1, and input this vector into a softmax classifier for binary classification to obtain 0 or 1, which is the authenticity of the EEG induced by the multimedia stimulus material.

5. The AIGC deepfake detection method based on dual-cerebral nerve synchronous coupling according to claim 1, wherein, In step S3, the phase - locking value (PLV) is selected as the synchronization index of the depth - separable neural network model based on the attention mechanism. The calculation method of the phase - locking value (PLV) is: First, calculate the phase - locking value (PLV) of each subject at each time point for each frequency band: ， where θ1(t, n) and θ2(t, n) are the instantaneous phases of the nth channel of each subject, N is the number of channels, and j is the imaginary unit; Then, average the total PLV over a period of time at all time points.

6. An AIGC deepfake detection system based on dual-brain nerve synchronous coupling, characterized in that, It includes: A dual-person AIGC EEG database construction module, which is used to design a dual-person hyperscanning experimental paradigm under the stimulation of digital multimedia materials, synchronously collect the EEG signals of two persons, and construct a dual-person EEG database under the stimulation of real and AIGC-generated multimedia materials; A depthwise separable neural network construction module based on the attention mechanism, which is used to construct a depthwise separable neural network based on the attention mechanism, and combine the attention mechanism with depthwise separable convolution to process the EEG signal features; A module for predicting data authenticity, which is used to filter the EEG signals of two persons collected by the dual-person AIGC EEG database construction module to obtain signals in four different frequency bands, then calculate the phase locking value PLV of the four frequency band signals, and finally stack the four frequency band PLVs spatially to obtain a three-dimensional vector and input it into the depthwise separable neural network based on the attention mechanism to predict the authenticity of the multimedia stimulus material content.

7. The AIGC deepfake detection system based on dual-brain nerve synchronous coupling according to claim 6, wherein The method for the dual-person AIGC EEG database construction module to design the dual-person hyperscanning experimental paradigm includes: The experimental conditions are manipulated by three factors: modality, valence, and type of life. The modality factor includes three levels: image, text, and audio. The valence condition includes three levels: positive, neutral, and negative. The type of life factor distinguishes living and non-living according to the nature of the stimulus. The living experimental objects include human faces, human voices, and texts containing conversations. The non-living stimulus materials include photos of daily items, natural sounds, and texts without any description of people or conversations; The main task of the participants is to evaluate the authenticity of the presented various materials, including images, texts, and audio clips. The participants are required to identify whether each stimulus is recorded from the objective real world or forged by AIGC based on their subjective evaluation.

8. The AIGC deepfake detection system based on dual-cerebral nerve synchronous coupling according to claim 7, wherein, The dual-person hyperscanning experimental paradigm uses the inter-subject correlation (ISC) to measure the synchrony of the brain responses of participants when viewing the same stimulus, and verify the effectiveness of the experimental paradigm design.

9. The AIGC deepfake detection system based on dual-brain nerve synchronous coupling according to claim 6, wherein The depthwise separable neural network based on the attention mechanism constructed by the depthwise separable neural network construction module based on the attention mechanism includes: Channel attention mechanism module, which is used to first globally average pool the input vector X1 into a one-dimensional vector of size 1*1*C, where C is the feature dimension of X1. Then, the average-pooled one-dimensional vector is put into two fully-connected layers to learn the non-linear relationship between channels. At the last layer, a weight vector for each channel is generated through the Sigmoid activation function. Finally, the weight vector is multiplied with the original feature map channel by channel to obtain a weighted feature vector of N*N*4, i.e., X2, where N is the number of electrodes; A depth convolution module, which is used to convolve the four dimensions of the X2 feature vector output by the channel attention mechanism module with four convolution kernels of 1*N respectively to obtain a feature vector X3 of N*1*4; A pointwise convolution module, which is used to convolve the feature vector X3 output by the depth convolution module with four convolution kernels of 1*1 to obtain a feature vector X4; A flattening module, which is used to flatten the feature vector X4 output by the pointwise convolution module to obtain a vector of (N*4)*1; A classifier module, which is used to input the vector of (N*4)*1 output by the flattening module into a softmax classifier for binary classification to obtain 0 or 1, that is, the authenticity of the EEG induced by the multimedia stimulus material.

10. The AIGC deepfake detection system based on dual-brain nerve synchronous coupling according to claim 6, wherein The method for the module for predicting data authenticity to calculate the phase locking value PLV is: First, calculate the phase locking value PLV of each subject at each time point and each frequency band: ， where θ1(t, n) and θ2(t, n) are the instantaneous phases of the nth channel of each subject, N is the number of channels, and j is the imaginary unit; Then, the average of the total PLV over a period of time is calculated at all time points.

Images