Artificial intelligence forged content detection method based on multi-expert model joint decision

Abstract

The invention discloses a multi-expert-model joint decision-making artificial intelligence forged content detection method, which comprises the following steps of: constructing and training a general expert model, a specialized expert model and a robust expert model which adopt ConvNeXt as a backbone network and adopt ConvNeXt as a backbone network; when the three expert models are trained, a cross entropy loss function is used as a constraint; inputting multimedia content to be detected into each type of specialized expert model, and directly skipping to enter a confidence coefficient calculation link when the detection of a certain type of specialized expert model exceeds a forgery threshold value; if the to-be-detected content passes all the specialized expert model detection, entering a general expert model detection process; entering general expert model detection, inputting to-be-detected content into a general expert model, and if the to-be-detected content reaches a forgery threshold value, entering a confidence coefficient calculation link after marking; and after the detection process of the first two types of expert models, confidence calculation is carried out.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an artificial intelligence forged content detection method with joint decision-making of multiple expert models. Background Art

[0002] With the rapid development of generative artificial intelligence technology, various deep fake content has been widely spread in social networks. Although these technologies have positive applications in creative design and life entertainment, fake content also brings ethical, privacy and security risks to society. Malicious content such as fake news and sensitive topics has greatly affected social stability and national security. Therefore, it is imperative to invent detection algorithms for fake content. In the real environment, the forgery methods are complex and diverse, making it impossible for a specialized single detection model to effectively identify all fake content. In addition, fake content is often spread multiple times through social networks and often undergoes multiple post-processing operations, such as compression, cropping, scaling, etc., which further increases the difficulty of detecting fake content.

[0003] Generative artificial intelligence content (AIGC) is an important branch of the field of artificial intelligence. It generates multimodal content such as images, text, audio, and video through deep learning technology. AIGC technology has not only greatly promoted the development of creative design, film and television production, but also shown broad prospects in application scenarios such as medical imaging and virtual reality. In recent years, based on the innovation of technologies such as generative adversarial networks (GAN) and diffusion models, AIGC technology has developed rapidly, and the quality, fidelity, and diversity of generated content have been continuously improved.

[0004] As a representative method of forgery generation, the generative adversarial network consists of two parts: the generator and the discriminator. Through the game-style training of the two, the generator is continuously optimized to generate realistic content, while the discriminator improves detection capabilities by distinguishing generated content from real content. This adversarial mechanism has brought the content generated by GAN to a new level of quality and authenticity. In recent years, the StyleGAN series of models has become one of the representative works of GAN. It introduces technologies such as hierarchical feature control and style encoding, which not only make the generated images have higher resolution, but also enable fine-grained content editing, such as changing the expression, age or background style of the face. In addition, the BigGAN model has significantly improved the diversity and clarity of the generated content through the optimization of the model architecture and larger-scale training. The innovation of these methods has greatly expanded the application scenarios of GANs, such as virtual human production, anime image generation, image restoration, and super-resolution reconstruction. However, this improvement in generation capabilities has also facilitated the production of deep fake content, making forged images and videos more difficult to identify.

[0005] The diffusion model is a generation method based on stepwise denoising, which generates realistic images through a gradual restoration process from the noise distribution to the target data distribution. The principle is to gradually add Gaussian noise to the input data to generate noise data, and then remove the noise through the inverse process to finally restore the target data. In recent years, the diffusion model has become an emerging mainstay of AIGC technology with its stable training process and excellent generation quality. Methods such as DALL-E and Stable Diffusion are typical representatives of the diffusion model. They combine the Transformer architecture and use large-scale multimodal data training to generate target images based on text descriptions. This type of diffusion model surpasses the traditional GAN ​​model in generating complex textures and fine details, and is widely used in advertising design, game scene creation and other purposes. However, after compression, cropping or other post-processing, the content generated by the diffusion model has more concealed forgery features, which puts higher requirements on forgery detection.

[0006] Existing generative model detection methods can be divided into two categories: manual feature based and deep learning based methods.

[0007] The manual feature method extracts features with anti-counterfeiting capabilities by analyzing forgery traces or specific patterns in AIGC content, and uses traditional machine learning models for classification. For example, by studying forgery traces, inconsistent lighting and geometric relationships in the image frequency domain, or detecting statistical anomalies caused by specific generation models, such as color distribution, edge smoothness, pixel consistency, etc. The main feature of this type of method is that it relies on the knowledge of domain experts to design feature engineering and has good interpretability. Its advantages include intuitive and clear feature design, good performance for small-scale data sets, and low computational complexity. However, the manual feature method cannot adapt to new forgery models, and the detection effect will be significantly reduced when the content is post-processed by compression, cropping, scaling, etc. In addition, as the quality of AIGC generation improves, forgery features often become more hidden, and it is difficult to meet high-precision detection requirements by relying solely on manual features.

[0008] The neural network-based method automatically extracts the latent features of AIGC content through a deep learning model and trains a forgery detection network with classification capabilities. For example, the convolutional neural network (CNN) is used to detect artifacts in forged images, and the time series model (such as RNN) is used to analyze the temporal characteristics of forged videos. Compared with the manual feature method, the advantage of the neural network method lies in its powerful feature extraction ability and adaptability to multimodal complex data. Deep features in the neural network can be automatically extracted only through training, without relying on complex manual feature design. In addition, the deep model can learn subtle patterns in forged content that are difficult for humans to detect, and has better detection capabilities for high-quality AIGC content. In addition, the neural network method can be trained end-to-end using large-scale datasets to optimize performance. However, its disadvantages include high demand for computing resources, long training and inference time, and weak model interpretability. More importantly, when the forged content is compressed, cropped, and other operations, the detection performance of the neural network method will also decline. At the same time, the generalization ability of the deep model to unseen forgery types or generative models may be limited.

[0009] According to the above description, the existing technical solutions mainly have the following defects: Current forgery detection technologies have poor robustness when facing real-world scenarios. Forged content is usually spread through social media, which often compresses multimedia content and often undergoes scaling and cropping by users during dissemination. This series of post-processing operations reduces the robustness of the model, resulting in poor detection performance.

[0010] Existing methods are usually based on single-model discrimination, resulting in low fault tolerance. Due to the variety of forged content, a single model usually cannot cover the entire feature space, making the model vulnerable to attacks from adversarial samples. Forged content spreads quickly and causes great harm, and the model's missed judgments often bring huge losses.

[0011] Current detection methods usually only target specific scenarios and lack system-level design considerations. Due to the rapid development and innovation of counterfeiting technology, a single detection method is difficult to cover all application scenarios. In addition, the detection method does not take into account the system interface design and scalability design, making it difficult for new detection methods to be integrated into existing systems, resulting in difficulty in iterating the detection system and being detrimental to the long-term maintenance of the system.

[0012] Therefore, the present invention designs an artificial intelligence forged content detection system with joint decision-making of multiple expert models, constructs a joint decision-making framework of expert models from a system-level perspective, gives full play to the advantages of different models in identifying forged content, and improves the detection system's adaptability to diverse forgery types. At the same time, for the three common post-processing operations of cropping, compression, and scaling, their features are integrated into the expert model to improve the robustness and accuracy of the detection system in the face of post-processed content. This innovative method helps to achieve more robust forged content identification in complex and changeable real-world scenarios, and also provides a system-level solution for the task of forged content detection. Summary of the invention

[0013] In view of the deficiencies of the prior art, the present invention proposes an artificial intelligence forged content detection method with joint decision-making of multiple expert models, constructs a joint decision-making framework of expert models from a system-level perspective, gives full play to the advantages of different models in identifying forged content, and improves the detection system's adaptability to diverse forgery types. At the same time, for the three common post-processing operations of cropping, compression, and scaling, their features are integrated into the expert model to improve the robustness and accuracy of the detection system in the face of post-processed content. By combining a multi-expert collaborative mechanism with robustness optimization, the present invention aims to establish a forged content detection system with high scalability and reliability, which can achieve efficient detection and comprehensive judgment of multiple types of forged content in complex application scenarios.

[0014] In order to solve the above technical problems, the technical solution of the present invention is: An artificial intelligence forged content detection method with joint decision-making of multiple expert models includes the following steps: Step 1: construct and train a general expert model group, wherein the general expert model group includes several general expert models, wherein the general expert model uses ConvNeXt as a backbone network, and a cross entropy loss function is used as a constraint during the training of the general expert model; Step 2, construct and train a specialized expert model group, wherein the specialized expert model group includes several specialized expert models, wherein the specialized expert model uses ConvNeXt as a backbone network, and the training data set used to train the specialized expert model includes real samples and forged samples, wherein the forged samples are marked with corresponding forged type labels, and during the training process, a corresponding specialized expert model is trained for each forged type label, and a cross entropy loss function is used as a constraint; Step 3: Obtain the multimedia content to be detected. First, it will be tested by the specialized expert model group. For the multimedia content to be detected, it will be input into each type of specialized expert model. When the detection of a certain type of specialized expert model exceeds the forgery threshold, it will directly jump to the confidence calculation link; if the content to be detected passes the detection of all specialized expert models, it will enter the general expert model detection process; Step 4: After passing the specialized expert model detection, enter the general expert model group detection, input the content to be detected into the general expert model group, and if it reaches the forgery threshold, it will be marked and enter the confidence calculation stage; Step 5: After the content to be detected has passed the detection process of the first two types of expert model groups, the confidence calculation is performed and finally fed back to the user.

[0015] Preferably, the general expert model group and the specialized expert model corresponding to each forgery type are respectively introduced into robust expert models.

[0016] Preferably, the robust expert model uses ConvNeXt as a backbone network, and uses a cross entropy loss function as a constraint during training.

[0017] Preferably, the general expert model, the specialized expert model and the robust expert model all adopt any one of the three variant models of ConvNeXt-Small, ConvNeXt-Base and ConvNeXt-XL.

[0018] Preferably, in step 1, the training method of the general expert model is: obtaining a training data set , and for the training data set Preprocessing is performed by horizontal flipping, vertical flipping and rotation. During training, given the samples in the training set , 0 represents a real image, 1 represents a fake image, is the predicted probability of the model, and the loss function can be expressed by the following formula: ; in, The number of samples is expressed by minimizing the loss function so that when real samples are input, the value tends to 0, and when forged samples are input, the value tends to 1, that is, distinguishing real content from forged content.

[0019] As a preference, for the robust expert model introduced in the general expert model group, the training data set Obtaining training data sets through data augmentation , using the training dataset The robust expert model introduced in the general expert model group is trained using the same training method as the general expert model.

[0020] Preferably, in step 2, a training data set is obtained. , and for the training data set The training dataset is preprocessed by horizontal flipping, vertical flipping and rotation. It includes real samples and forged samples, and the forged type label corresponding to the forged sample is recorded as , all the fake sample data sets By counterfeit label Divide into several subsets, namely , for each forgery type during training Train a corresponding specialized expert model .

[0021] Preferably, for each forgery type, the specialized expert model introduced by the robust expert model is trained with the dataset Obtaining training data sets through data augmentation , using the training dataset The robust expert model introduced by the specialized expert model corresponding to each forgery type is trained in the same way as the general expert model.

[0022] Preferably, the training data set The real samples and training data sets in shared.

[0023] Preferably, in step 5, when the input content is determined to be forged content by the specialized expert model or robust expert model corresponding to the forgery type, the confidence The average confidence value of this group of specialized expert models is adopted.

[0024] Preferably, the confidence The calculation method is: record the group of specialized expert models Include submodels , take the average model confidence in the group, and the calculation method is as follows: ; in is the input sample, Group The output confidence of the expert model, is the total number of specialized expert models and robust expert models in this group.

[0025] Preferably, when the input content is not judged as forged content by the specialized expert model or its corresponding robust expert model, the confidence level is output. The output confidence of the general expert model group and its corresponding robust expert model is averaged and the specific formula is as follows: ; in is the input sample, is the output confidence of the general expert model, is the total number of the set of general expert models and their corresponding robust expert models.

[0026] The present invention designs a multi-expert model training strategy for forged content detection, which improves the limitation of a single model in detecting forged content, gives full play to the advantages of different models in identifying forged content, and improves the adaptability of the detection system to diverse forgery types. A counterfeit content identification mechanism based on the collaboration of multiple expert models was designed, and a joint decision-making framework of expert models was constructed from a system-level perspective. This fully utilized the advantages of specialized expert models and general expert models in detection and improved the overall detection speed and performance of the system.

[0027] By integrating the three post-processing operations of scaling, cropping, and compression into the expert model, the system can better match the real scene and improve the robustness of the system.

[0028] (1) Use multiple expert models to collaboratively detect forged content, improve the limitations of a single model in detecting forged content, and enhance the generalization of the system.

[0029] (2) Add specialized models for compression, cropping, and scaling operations to cope with common user behaviors in real scenarios and improve the robustness of the system.

[0030] (3) The invention is not just for a single scenario, but is designed from a system-level perspective. The unique organizational structure of the system allows it to dynamically add and change models during operation, enabling the system to cope with new counterfeiting methods that are constantly evolving in the future. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0032] Figure 1 Schematic diagram of the process of the general expert model in the embodiment of the present invention; Figure 2 It is a flow chart of the specialized patent model in the embodiment of the present invention; Figure 3 Schematic diagram of the process of joint decision-making of multiple expert models in an embodiment of the present invention. DETAILED DESCRIPTION

[0033] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0034] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", and the like are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, features defined as "first", "second", and the like may explicitly or implicitly include one or more of the features. In the description of the present invention, unless otherwise specified, "multiple" means two or more.

[0035] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood by specific circumstances. Example

[0036] This embodiment provides an artificial intelligence forged content detection method with joint decision-making of multiple expert models, which consists of two parts: multiple expert model training and multiple expert model joint decision-making.

[0037] (1) Multi-expert model training: Multi-expert model training aims to formulate appropriate training methods, divide different expert models, and prepare appropriate data for each model for training, so as to achieve generalized and specialized detection of counterfeit content and improve the reliability and robustness of the system.

[0038] (1.1) General expert model training: This type of expert model is designed for forged content detection in general scenarios and relies on large-scale and diverse general data sets for training. The model extracts texture features from the data and analyzes the differences between forged content and real content, thereby accurately determining whether multimedia content is forged or not. Figure 1 As shown, the specific training steps are as follows: (1.1.1) First, for the training data Preprocessing operations are performed, including horizontal flipping, vertical flipping, rotation, etc. These operations generate diverse data samples to provide richer learning materials for subsequent model training and improve the robustness and adaptability of the model.

[0039] (1.1.2) Then, the data is sent to the model for training. Since the goal at this stage is to train several general expert models to identify various types of forged content, and the forged content usually differs from the real content in texture information, ConvNeXt is used as the backbone network. The ConvNeXt network is based on convolution operations, which can well capture local texture details in the image and identify subtle differences between forged content and real content. At the same time, ConvNeXt has multiple variants, suitable for different computing resources and task requirements. In this invention, three variants, ConvNeXt-Small, ConvNeXt-Base, and ConvNeXt-XL, are used to provide corresponding solutions for applications of different scales.

[0040] For the problem of forged content detection, the task can be regarded as a binary classification problem: real content and forged content. Therefore, the cross entropy loss function is used as a constraint during the training process. Given , 0 represents a real image, 1 represents a fake image, is the predicted probability of the model, and the loss function can be expressed by the following formula:

[0041] Among them, the number of samples is expressed. By minimizing the loss function, when real samples are input, the value tends to 0, and when forged samples are input, the value tends to 1, that is, to distinguish real content from forged content.

[0042] It is understandable that the neural network can distinguish samples of different categories after optimizing the loss function. Minimizing the loss function is the optimization method. The loss is used to evaluate the degree of error in the training samples, that is, the quality of the neural network training. The greater the loss, the more training samples are misjudged by the neural network, which means poor training. Minimizing the loss function means that the training samples have the least errors, so the better the proof effect, the better the neural network training.

[0043] (1.2) Specialized expert model training: This type of expert model is mainly used to detect specific types of forgeries, providing reliable judgments in specific scenarios, so that the system can maintain reliable judgments when facing various known types of forgeries, such as Figure 2 As shown, the specific steps are as follows: (1.2.1) First, for the training data Preprocessing operations are performed, including horizontal flipping, vertical flipping, rotation, etc. These operations generate diverse data samples to provide richer learning materials for subsequent model training and improve the robustness and adaptability of the model.

[0044] (1.2.2) Then, the data is fed into the model for training. To extract the texture information of fake content, which is usually different from the real content, ConvNeXt is used as the backbone network. Consistent with what is mentioned in (1.1.2), three variants, ConvNeXt-Small, ConvNeXt-Base, and ConvNeXt-XL, are used to provide corresponding solutions for applications of different scales.

[0045] Since the goal of this stage is to train several specialized expert models to identify a specified type of forged content, it is necessary to prepare specific data for each expert model. The specific division rules are as follows: for real samples, all expert models share this part of data so that they can widely learn the common features of real samples; for forged samples, the forged type label corresponding to the forged sample is recorded as , all the fake sample data sets By counterfeit label Divide into several subsets, namely , for each forgery type during training Train a corresponding specialized expert model , the corresponding data sets include and The loss function used in training uses the cross entropy loss function, and the formula is as follows:

[0046] in represents the sample size. Given , 0 represents a real image, 1 represents a fake image, is the predicted probability of the model. By minimizing this loss function, the model can effectively distinguish between real content and fake content.

[0047] (2) Joint decision-making of multiple expert models: The joint detection of multiple expert models aims to efficiently and accurately detect the input image by combining various expert models through a certain mechanism, and give the final detection confidence through a certain strategy. The specific process is as follows Figure 2 Shown: (2.1) Specialized expert model detection: The multimedia content to be detected will first be detected by a specialized expert model. This type of model has extremely high detection accuracy for known forgery types. The specific process is as follows: For the multimedia content to be detected, it is input into each type of specialized expert model. When a certain type of specialized expert model detection exceeds the forgery threshold, it will directly jump to the confidence calculation stage. If the content to be detected passes all the specialized expert model detections, it will enter the general expert model detection process.

[0048] It should be noted that the forgery threshold of the specialized expert model is the probability value output by the specialized expert model, that is, the confidence level. In this embodiment, the forgery threshold of the specialized expert model is set to 0.5.

[0049] (2.2) General expert model detection: After passing the specialized expert detection model, enter the general expert model detection. This type of expert model is designed for general scene detection. Compared with the specialized expert model, although it weakens the detection ability of known forgery types, it can analyze general forgery traces and better detect general forgery content. The content to be detected is input into the general expert model. If it reaches the forgery threshold, it will be marked and enter the confidence calculation stage.

[0050] It should be noted that the forgery threshold of the general expert model is the probability value output by the general expert model, that is, the confidence level. In this embodiment, the forgery threshold of the general expert model is set to 0.5.

[0051] (2.3) Confidence calculation: After the content to be detected has gone through the detection process of the first two types of expert models, the confidence is obtained through a certain mechanism and finally fed back to the user. The specific confidence calculation method is as follows: When the input content is judged as forged content by the specialized expert model, the confidence The average confidence value of the expert model is used. Specifically, Include submodels . Take the average confidence level of the model in the group. The formula is as follows:

[0052] in is the input sample, Group The output confidence of the expert model, is the total number of specialized expert models in this group.

[0053] When the input content is not judged as forged content by the specialized expert model, the output confidence It is obtained by taking the average of the output confidence of the general expert model. The specific formula is as follows:

[0054] in is the input sample, is the output confidence of the general expert model, is the total number of general expert models in this group. Example

[0055] The difference between this embodiment and the first embodiment is that the general expert model group and the specialized expert model corresponding to each forgery type respectively introduce robust expert models.

[0056] (1.3) Robust expert model training: This type of model is mainly aimed at forged content that has undergone post-processing operations such as cropping, compression, and scaling. It is mainly used to enhance the application effect of the first two types of expert models in complex scenarios, making the system more adaptable to real scenarios such as social platforms. This type of expert model is usually integrated into the system as a variant of the first two types of expert models. Therefore, during training, for the robust expert model introduced in the general expert model group, the training data set Obtaining training data sets through data augmentation , using the training dataset The robust expert model introduced in the general expert model group is trained.

[0057] For each forgery type, the specialized expert model introduced by the robust expert model will train the dataset Obtaining training data sets through data augmentation , using the training dataset The robust expert model introduced by the specialized expert model corresponding to each forgery type is trained.

[0058] The specific training steps are as follows: (1.3.1) First, prepare the training data and perform preprocessing operations on the training data, including horizontal flipping, vertical flipping, rotation, etc. This is the same as (1.1.1) and (1.2.1).

[0059] (1.3.2) Perform specified enhancement operations on the data. In real scenarios, compression, cropping, and scaling are the most common post-processing operations, and have a significant impact on detection performance. Therefore, in the training process of this type of expert model, corresponding post-processing operations are applied to the data, and then trained to enable the model to learn the forgery traces after these post-processing operations, thereby enhancing the robustness of the model in real application scenarios.

[0060] (1.3.3) The enhanced data is fed into the model for training. The training model uses ConvNeXt as the backbone network. Consistent with what is mentioned in (1.1.2), three variants, ConvNeXt-Small, ConvNeXt-Base, and ConvNeXt-XL, are used to provide corresponding solutions for applications of different scales.

[0061] In the multi-expert model joint decision-making stage, the multimedia content to be detected will first be tested by the specialized expert model group. For the multimedia content to be detected, it will be input into each type of specialized expert model and its corresponding robust expert model. When the detection of a certain type of specialized expert model or its corresponding robust expert model exceeds the forgery threshold, it will directly jump to the confidence calculation link; if the content to be detected passes the detection of all specialized expert model groups, it will enter the general expert model group detection process; When the input content is judged as forged content by the specialized expert model or robust expert model corresponding to the forgery type, the confidence The average confidence value of this group of specialized expert models is adopted.

[0062] The confidence calculation method is: record the group of specialized expert models Include submodels , take the average model confidence in the group, and the calculation method is as follows: ; in is the input sample, Group The output confidence of the expert model, is the total number of specialized expert models and robust expert models in this group.

[0063] After passing the specialized expert model detection, it enters the general expert model group detection, and the content to be detected is input into the general expert model group. If the general expert model in the general expert model group or its corresponding robust expert model reaches the forgery threshold, it will be marked and enter the confidence calculation link.

[0064] When the input content is not judged as forged content by the specialized expert model or its corresponding robust expert model, the confidence level is output. The output confidence of the general expert model group and its corresponding robust expert model is averaged and the specific formula is as follows: ; in is the input sample, is the output confidence of the general expert model, is the total number of the set of general expert models and their corresponding robust expert models.

[0065] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited to the described embodiments. For those skilled in the art, various changes, modifications, substitutions and variations of these embodiments including components are made without departing from the principles and spirit of the present invention, and still fall within the scope of protection of the present invention.

Claims

1. An artificial intelligence forged content detection method based on joint decision-making of multiple expert models, characterized in that: The steps include: Step 1: construct and train a general expert model group, wherein the general expert model group includes several general expert models, wherein the general expert model uses ConvNeXt as a backbone network, and a cross entropy loss function is used as a constraint during the training of the general expert model; Step 2, construct and train a specialized expert model group, wherein the specialized expert model group includes several specialized expert models, wherein the specialized expert model uses ConvNeXt as a backbone network, and the training data set used to train the specialized expert model includes real samples and forged samples, wherein the forged samples are marked with corresponding forged type labels, and during the training process, a corresponding specialized expert model is trained for each forged type label, and a cross entropy loss function is used as a constraint; Step 3: Obtain the multimedia content to be detected. First, it will be tested by the specialized expert model group. For the multimedia content to be detected, it will be input into each type of specialized expert model. When the detection of a certain type of specialized expert model exceeds the forgery threshold, it will directly jump to the confidence calculation link; if the content to be detected passes the detection of all specialized expert models, it will enter the general expert model detection process; Step 4: After passing the specialized expert model detection, enter the general expert model group detection, input the content to be detected into the general expert model group, and if it reaches the forgery threshold, it will be marked and enter the confidence calculation stage; Step 5: After the content to be detected has passed the detection process of the first two types of expert model groups, the confidence calculation is performed and finally fed back to the user.

2. The artificial intelligence forged content detection method of multiple expert models joint decision-making according to claim 1 is characterized in that: The general expert model group and the specialized expert model corresponding to each forgery type are respectively introduced into the robust expert model.

3. According to claim 2, the artificial intelligence forged content detection method with joint decision-making of multiple expert models is characterized in that: The robust expert model uses ConvNeXt as the backbone network and uses the cross entropy loss function as a constraint during training.

4. According to claim 3, the artificial intelligence forged content detection method with joint decision-making of multiple expert models is characterized in that: The general expert model, specialized expert model and robust expert model all adopt any one of the three variant models of ConvNeXt-Small, ConvNeXt-Base and ConvNeXt-XL.

5. According to claim 4, the artificial intelligence forged content detection method with joint decision-making of multiple expert models is characterized in that: In step 1, the training method of the general expert model is: obtaining a training data set , and for the training data set Preprocessing is performed by horizontal flipping, vertical flipping and rotation. During training, given the samples in the training set , 0 represents a real image, 1 represents a fake image, is the predicted probability of the model, and the loss function can be expressed by the following formula: ； in, The number of samples is expressed by minimizing the loss function so that when real samples are input, the value tends to 0, and when forged samples are input, the value tends to 1, that is, distinguishing real content from forged content.

6. The artificial intelligence forged content detection method of multiple expert models joint decision-making according to claim 5 is characterized in that: For the robust expert model introduced in the general expert model group, the training data set Obtaining training data sets through data augmentation , using the training dataset The robust expert model introduced in the general expert model group is trained using the same training method as the general expert model.

7. The artificial intelligence forged content detection method of multiple expert models joint decision-making according to claim 6 is characterized in that: In step 2, obtain the training data set , and for the training data set The training dataset is preprocessed by horizontal flipping, vertical flipping and rotation. It includes real samples and forged samples, and the forged type label corresponding to the forged sample is recorded as , all the fake sample data sets By counterfeit label Divide into several subsets, namely , for each forgery type during training Train a corresponding specialized expert model .

8. The artificial intelligence forged content detection method of multiple expert models joint decision-making according to claim 7 is characterized in that: For each forgery type, the specialized expert model introduced by the robust expert model will train the dataset Obtaining training data sets through data augmentation , using the training dataset The robust expert model introduced by the specialized expert model corresponding to each forgery type is trained in the same way as the general expert model.

9. The artificial intelligence forged content detection method of multiple expert models joint decision-making according to claim 7 is characterized in that: The training dataset The real samples and training data sets in shared.

10. The artificial intelligence forged content detection method of multiple expert models joint decision-making according to claim 8 is characterized in that: In step 5, when the input content is determined to be forged content by the specialized expert model or robust expert model corresponding to the forgery type, the confidence The average confidence value of this group of specialized expert models is adopted.

11. The artificial intelligence forged content detection method of multiple expert models joint decision-making according to claim 10 is characterized in that: The confidence The calculation method is: Include submodels , take the average model confidence in the group, and the calculation method is as follows: ； in is the input sample, Group The output confidence of the expert model, is the total number of specialized expert models and robust expert models in this group.

12. The artificial intelligence forged content detection method of multiple expert models joint decision-making according to claim 11 is characterized in that: When the input content is not judged as forged content by the specialized expert model or its corresponding robust expert model, the output confidence is obtained by averaging the output confidences of the general expert model group and its corresponding robust expert model. The specific formula is as follows: ； in is the input sample, is the output confidence of the general expert model, is the total number of general expert models and robust expert models in this group.

Images