Artificial intelligence generated image detection model optimization method, system, device and medium

Abstract

The application discloses an artificial intelligence generated image detection model optimization method, system, device and medium, which are corresponding solutions, and in the solutions: with the help of the limited available generated model, a dynamic interactive optimization mechanism is constructed between the generated model and the to-be-optimized detection model, the performance of the detection model is continuously optimized by simulating the generated model-detection model alternate evolution path in the real world, so that the detection model can maintain stable detection performance in the real scene; the method can be applied to various current mainstream detection model architectures, and has wide application range; meanwhile, the method can detect images generated by various latest generated models, has high generalization and strong practicability, and the average discrimination accuracy in the experiment can reach more than 85%.

Description

Technical Field

[0001] This invention relates to the field of generated image detection technology, and in particular to an optimization method, system, device and medium for an artificial intelligence generated image detection model. Background Technology

[0002] Artificial intelligence-generated images refer to images artificially generated using visual generative AI techniques such as diffusion models. Due to the continuous development of generation technologies, existing generation models exhibit strong representational capabilities and high generation efficiency. At the same time, a large number of realistic visually generated images have emerged on current internet platforms. These indistinguishable images often raise a series of potential content security issues. Therefore, research on the detection of visually generated AI images is a key challenge in the current content security field.

[0003] Existing research on generative image detection often follows a "one-and-done" training paradigm, which involves training the detection model on a closed training set and then deploying it in real-world scenarios for detection. Clearly, faced with the constantly iterating and evolving generative models in the real world, this single, static detection model often struggles to keep pace with the rapid updates and iterations of generative technologies. Outdated forgery detection knowledge often manifests as a precipitous drop in detection accuracy when encountering new generative models.

[0004] In view of this, the present invention is hereby proposed. Summary of the Invention

[0005] The purpose of this invention is to provide an optimization method, system, device and medium for artificial intelligence-generated image detection models. By simulating the alternating evolution path of the generation model and the detection model in the real world, the performance of the detection model is continuously optimized, thereby ensuring that the detection model maintains stable detection performance in real-world scenarios.

[0006] The objective of this invention is achieved through the following technical solution: An optimization method for an AI-generated image detection model includes: Obtain the initial detection model and the initial generation model; A multi-round iterative optimization strategy is adopted to interactively optimize the initial detection model and the initial generation model. This includes: In the nth round of optimization, reinforcement learning is used to construct a reward function based on the detection results of the image samples generated by the generation model after the (n-1)th round of optimization, and then optimizes the generation model after the (n-1)th round of optimization to obtain the generation model after the nth round of optimization. New image samples are then generated from this model and combined with the basic training samples to form the training data for the nth round. Using the training data from the nth round, the detection model after the (n-1)th round of optimization is further optimized through a course-based learning approach to obtain the detection model after the nth round of optimization. The optimization continues for the (n+1)th round until the set number of rounds is reached. The basic samples contain all training data from previous rounds of optimization. If n=1, the generation model after the (n-1)th round of optimization is the initial generation model. The basic samples contain a pre-selected batch of real image samples and image samples generated by the initial generation model. If n>1, the basic samples also contain new image samples from each round.

[0007] An AI-generated image detection model optimization system includes: The initial model acquisition unit is used to acquire the initial detection model and the initial generation model; The interactive optimization unit is used to interactively optimize the initial detection model and the initial generation model using a multi-round iterative optimization strategy. This includes: In the nth round of optimization, reinforcement learning is used to construct a reward function based on the detection results of the image samples generated by the generation model after the (n-1)th round of optimization, and then optimizes the generation model after the (n-1)th round of optimization to obtain the generation model after the nth round of optimization. New image samples are then generated from this model and combined with the basic training samples to form the training data for the nth round. Using the training data from the nth round, the detection model after the (n-1)th round of optimization is further optimized through a course-based learning approach to obtain the detection model after the nth round of optimization. The optimization continues for the (n+1)th round until a set number of rounds is reached. The basic samples contain all training data from previous rounds of optimization. If n=1, the generation model after the (n-1)th round of optimization is the initial generation model. The basic samples contain a pre-selected batch of real image samples and image samples generated by the initial generation model. If n>1, the basic samples also contain new image samples from each round.

[0008] A processing device includes: one or more processors; and a memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the aforementioned method.

[0009] A readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned method.

[0010] As can be seen from the technical solution provided by the present invention, by using the limited available generative models, a dynamic interactive optimization mechanism is constructed between the generative model and the detection model to be optimized. By simulating the alternating evolution path of the generative model and the detection model in the real world, the detection model can be continuously iteratively optimized and upgraded, with high flexibility. Furthermore, the present invention is applicable to various mainstream detection model architectures, with a wide range of applications. At the same time, it can detect images generated by various latest generative models, with high generalization and strong practicality. In experiments, the average discrimination accuracy can reach more than 85%. Attached Figure Description

[0011] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a flowchart of an artificial intelligence-generated image detection model optimization method provided in an embodiment of the present invention.

[0013] Figure 2 This is a schematic diagram of the overall framework of an artificial intelligence-generated image detection model optimization method provided in an embodiment of the present invention.

[0014] Figure 3 This is a schematic diagram of an artificial intelligence-generated image detection model optimization system provided in an embodiment of the present invention.

[0015] Figure 4 This is a schematic diagram of a processing device provided in an embodiment of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of the present invention.

[0017] First, the following explanations are provided for the terms that may be used in this article: The terms "comprising," "including," "containing," "having," or other similar semantic descriptions should be interpreted as non-exclusive inclusion. For example, including a technical feature element (such as raw material, component, ingredient, carrier, dosage form, material, size, part, component, mechanism, device, step, process, method, reaction conditions, processing conditions, parameter, algorithm, signal, data, product or article of manufacture, etc.) should be interpreted as including not only the expressly listed technical feature element, but also other technical feature elements that are not expressly listed and are well-known in the art.

[0018] The term "composed of" excludes any technical features not expressly listed. When used in a claim, it closes the claim to exclude all technical features other than those expressly listed, except for associated conventional impurities. If the term appears only in a clause of a claim, it limits the claim to the elements expressly listed in that clause; elements recited in other clauses are not excluded from the overall claim.

[0019] The following provides a detailed description of the artificial intelligence-generated image detection model optimization method, system, device, and medium provided by this invention. Contents not described in detail in the embodiments of this invention are prior art known to those skilled in the art. Where specific conditions are not specified in the embodiments of this invention, they are performed according to conventional conditions in the art or conditions recommended by the manufacturer. Instruments used in the embodiments of this invention, unless otherwise specified by the manufacturer, are all commercially available conventional products.

[0020] Example 1 This invention provides an optimization method for an artificial intelligence-generated image detection model, such as... Figure 1 As shown, it mainly includes the following steps: Step 1: Obtain the initial detection model and the initial generation model.

[0021] In this embodiment of the invention, both the detection model and the generation model can be selected from existing models according to the actual situation. Generally speaking, the initial detection model has poor detection performance, while the initial generation model has relatively high image generation performance.

[0022] Step 2: Use a multi-round iterative optimization strategy to interactively optimize the initial detection model and the initial generation model.

[0023] The main process of this step is as follows: In the nth round of optimization, reinforcement learning is used to construct a reward function based on the detection results of the image samples generated by the generative model after the (n-1)th round of optimization, and to optimize the generative model after the (n-1)th round of optimization to obtain the generative model after the nth round of optimization. New image samples are then generated from this model and combined with the basic training samples to form the training data for the nth round. Using the training data of the nth round, the detection model after the (n-1)th round of optimization is optimized through a course-based learning approach to obtain the detection model after the nth round of optimization. The optimization continues for the (n+1)th round until the set number of rounds is reached. The basic samples contain all the training data from the previous rounds of optimization. If n=1, the generative model after the (n-1)th round of optimization is the initial generative model. The basic samples contain a pre-selected batch of real image samples and the image samples generated by the initial generative model. If n>1, the basic samples also contain the new image samples from each round.

[0024] In this embodiment of the invention, the step of employing reinforcement learning, combining the detection results of the image samples generated by the optimized generation model in the (n-1)th round with the optimized detection model in the (n-1)th round to construct a reward function, and optimizing the optimized generation model in the (n-1)th round to obtain the optimized generation model in the nth round includes: Let the detection model after the (n-1)th round of optimization be ? Let the generative model after the (n-1)th round of optimization be ? ; For the generative model after the (n-1)th round of optimization The generated image samples are processed by the detection model optimized in the (n-1)th round. Obtain detection confidence ,in This represents the detection model after the (n-1)th round of optimization. The probability of an image sample being fake is determined; a fake image sample indicates that it was generated by the generative model, while a real image sample indicates that it is a real image. This is combined with the visual evaluation score of the image sample. Construct the following reward function : ; in, and These are preset weighting coefficients; By combining the reward function and optimizing the generative model after the (n-1)th round of optimization, we obtain the generative model after the nth round of optimization. .

[0025] In this embodiment of the invention, the step of generating new image samples and then combining them with basic training samples to form the nth round of training data includes: New image samples are generated using the optimized generative model from the nth round, forming a set. ; If n=1, then the training data for the nth round is: ,in, Based on the training samples, A set of pre-selected real image samples, A collection of image samples generated for the initial generative model; If n>1, then the training data for the nth round is: Among them, excluding Other data Based on the training samples, It is a set of new image samples from the first round.

[0026] In this embodiment of the invention, optimizing the detection model after the (n-1)th round of optimization through course-based learning to obtain the detection model after the nth round of optimization includes: first using basic training samples to optimize the detection model after the (n-1)th round of optimization. Optimize the model, then gradually add new image samples from the nth round, and finally complete the training of the detection model to obtain the optimized detection model after the nth round. n=1,…,N, where N is the set number of rounds.

[0027] In this embodiment of the invention, through continuous interaction optimization, the desired result is obtained when the set number of rounds N is reached. This is the final optimized detection model, which can be deployed on the data platform of network security regulatory departments to identify and intercept generated images spreading on the Internet. Of course, it can also be applied to other scenarios to identify the authenticity of suspicious images.

[0028] To more clearly demonstrate the technical solution and its effects provided by the present invention, the method provided by the embodiments of the present invention will be described in detail below with reference to specific examples.

[0029] The AI-generated image detection model optimization method provided by this invention constructs an interactive optimization mechanism based on the generation model and the detection model, ultimately aiming to improve the performance of the AI-generated image detection model.

[0030] 1. Optimized settings.

[0031] Assume there is already a poorly performing initial detection model. and a relatively advanced initial generation model This method aims to significantly improve the performance of the detection model by constructing an interactive optimization mechanism between the detection model and the generation model, simulating the iterative optimization path of the detection model in real-world scenarios. The interactive optimization mechanism proposed in this invention is a multi-round iterative strategy, such as... Figure 2 As shown, the overall framework of the optimization method is illustrated. The following section will first introduce the detailed optimization process using the first round as an example.

[0032] In this embodiment of the invention, both the initial detection model and the initial generation model are existing models, which can be selected by the user according to actual needs. This invention does not impose specific limitations. For example, the initial detection model can be a detection model based on the visual Transformer (transformer neural network) architecture, and the initial generation model can be a generation model based on the diffusion (diffusion) architecture.

[0033] 2. First round of optimization process.

[0034] (1) Generation of training hard samples based on adversarial learning.

[0035] For existing generative models One of the most direct approaches to optimizing a detection model is to utilize generative models. Generate a batch of images At the same time, another batch of real images were collected. The two are mixed as training data. For detection model Fine-tuning can be performed to improve the model's detection performance.

[0036] However, this approach cannot maximize the model's detection performance. On the one hand, in real-world scenarios, generative technologies are experiencing explosive iterative updates, while when optimizing detectors, only a limited number of generative models can often be collected to assist in optimization, resulting in generated images... This approach cannot fully represent the distribution of generated images in real-world scenarios. Furthermore, previous research has indicated a discrepancy between human perception of generated images and detection models; that is, generated images that are difficult for the human eye to distinguish from real images are often easy samples for detection models to differentiate. This suggests that generated images oriented towards "generating images that appear realistic to the human eye" are ineffective. On the contrary, these simple training samples are redundant for the detection model and do not significantly improve its performance. To address the above discussion, this invention proposes a method for generating difficult training samples based on adversarial learning. Firstly, the generation model... An optimization is performed to encourage the generative model to produce more challenging generated images, thereby helping the detection model improve its performance.

[0037] This invention employs reinforcement learning to optimize the generative model. First, we introduce the basic principles of reinforcement learning. Most existing reinforcement learning methods can be abstracted into a preference alignment problem. That is, for the target model (in this embodiment, the target model is a generative model), we design a reward function to make the model's output satisfy the reward requirement as much as possible, thereby aligning the model's preferences. In this embodiment, the reward can be set as the image samples generated by the generative model for the detection model. The increased difficulty in detection prompts generative models to generate more difficult samples that are hard to detect.

[0038] Specifically, using generative models Generate a batch of images For one of the generated images, it is first input into the detection model. Obtaining detection confidence ,in This represents the detection model. The probability of classifying it as fake is calculated, with higher scores indicating a stronger certainty that it is fake (i.e., a generated image). Based on this, the reward function of reinforcement learning... It can be set to: ; Among them, the first item The first term represents minimizing the model's detection score, i.e., increasing the model's detection difficulty, while the second term... The visual evaluation score represents the generated image, used to ensure that the generated image still meets the most basic visual aesthetic requirements, while stabilizing the optimization process of reinforcement learning.

[0039] In this embodiment of the invention, the visual evaluation score can be obtained using a pre-trained image scoring model. For example, the PickScore model (which is a model for evaluating the quality of text-to-image generation, outputting a score that reflects human preferences by calculating the semantic similarity between the text prompt and the generated image) can be used. Specifically, the image and the text prompt used to generate the image are input into the pre-trained image scoring model to obtain a visual evaluation score that simulates human preference evaluation, where a high score represents conformity to human aesthetics and a low score represents non-conformity to human aesthetics.

[0040] Through the reinforcement learning optimization described above, the generative model can be successfully optimized. Optimized generative model More detection models can be generated. Difficult samples that are hard to detect, among which, symbols This represents the optimization process. Based on this, using... Generate a new batch of training data The training data can also be expanded accordingly, and represented as... .

[0041] (2) Optimization of the detection model based on course learning.

[0042] Training data generated based on the aforementioned steps Continue to work on the detection model Optimize (fine-tune the training). Consider the training data. It contains generated image samples of varying difficulty, namely and Based on the generated image samples, a detection model optimization method based on curriculum learning is proposed, which gradually broadens the cognitive boundary of the detection model from easy to difficult.

[0043] Specifically, initially only low-difficulty training data is used. For detection model Optimize and encourage detection models Enhance the basic understanding of this type of generated image sample. Furthermore, gradually incorporate more challenging generated image samples into the optimization process. The generated image samples further enhance the model's capabilities. Meanwhile, the model still retains... This is used to stabilize the optimization direction of the detection model. Based on the above process, the generative model can be successfully optimized. .

[0044] 3. Multi-round optimization process.

[0045] Previously, we used the first round of optimization as an example to introduce the single-round optimization process, which can optimize the generative model and the detection model in one round. , This process can then be repeated for a total of N rounds. Finally, in the Nth round, the generative model can be optimized. Expanding the training data Correspondingly, it can be achieved by using To optimize the detection model , here This is the final optimized detection model.

[0046] Through the above description of the embodiments, those skilled in the art can clearly understand that the above embodiments can be implemented by software, or by using software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solutions of the above embodiments can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, mobile hard drive, etc.), including several instructions to cause a computer device (such as a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0047] Example 2 This invention also provides an artificial intelligence-generated image detection model optimization system, which is mainly used to implement the methods provided in the foregoing embodiments, such as... Figure 3 As shown, the system mainly includes: The initial model acquisition unit is used to acquire the initial detection model and the initial generation model; The interactive optimization unit is used to interactively optimize the initial detection model and the initial generation model using a multi-round iterative optimization strategy. This includes: In the nth round of optimization, reinforcement learning is used to construct a reward function based on the detection results of the image samples generated by the generation model after the (n-1)th round of optimization, and then optimizes the generation model after the (n-1)th round of optimization to obtain the generation model after the nth round of optimization. New image samples are then generated from this model and combined with the basic training samples to form the training data for the nth round. Using the training data from the nth round, the detection model after the (n-1)th round of optimization is further optimized through a course-based learning approach to obtain the detection model after the nth round of optimization. The optimization continues for the (n+1)th round until a set number of rounds is reached. The basic samples contain all training data from previous rounds of optimization. If n=1, the generation model after the (n-1)th round of optimization is the initial generation model. The basic samples contain a pre-selected batch of real image samples and image samples generated by the initial generation model. If n>1, the basic samples also contain new image samples from each round.

[0048] In this embodiment of the invention, the step of employing reinforcement learning, combining the detection results of the image samples generated by the optimized generation model in the (n-1)th round with the optimized detection model in the (n-1)th round to construct a reward function, and optimizing the optimized generation model in the (n-1)th round to obtain the optimized generation model in the nth round includes: Let the detection model after the (n-1)th round of optimization be ? Let the generative model after the (n-1)th round of optimization be ? ; For the generative model after the (n-1)th round of optimization The generated image samples are processed by the detection model optimized in the (n-1)th round. Obtain detection confidence ,in This represents the detection model after the (n-1)th round of optimization. The probability of an image sample being fake is determined; a fake image sample indicates that it was generated by the generative model, while a real image sample indicates that it is a real image. This is combined with the visual evaluation score of the image sample. Construct the following reward function : ; in, and These are preset weighting coefficients; The generative model optimized in the (n-1)th round is further optimized using the reward function to obtain the generative model optimized in the nth round. .

[0049] In this embodiment of the invention, the step of generating new image samples and then combining them with basic training samples to form the nth round of training data includes: New image samples are generated using the optimized generative model from the nth round, forming a set. ; If n=1, then the training data for the nth round is: ,in, Based on the training samples, A set of pre-selected real image samples, A collection of image samples generated for the initial generative model; If n>1, then the training data for the nth round is: Among them, excluding Other data Based on the training samples, It is a set of new image samples from the first round.

[0050] In this embodiment of the invention, training the detection model optimized in the (n-1)th round using a course-based learning approach to obtain the detection model optimized in the nth round includes: First, use the basic training samples to optimize the detection model in the (n-1)th round. Optimize the model, then gradually add new image samples from the nth round, and finally complete the optimization of the detection model to obtain the optimized detection model after the nth round. .

[0051] Those skilled in the art will understand that, for the sake of convenience and brevity, the above-described division of functional modules is used as an example. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the system can be divided into different functional modules to complete all or part of the functions described above.

[0052] Example 3 The present invention also provides a processing device, such as Figure 4 As shown, it mainly includes: one or more processors; a memory for storing one or more programs; wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the method provided in the foregoing embodiments.

[0053] Furthermore, the processing device also includes at least one input device and at least one output device; in the processing device, the processor, memory, input device, and output device are connected via a bus.

[0054] In this embodiment of the invention, the specific types of the memory, input device, and output device are not limited; for example: Input devices can be touchscreens, image acquisition devices, physical buttons, or mice, etc. The output device can be a display terminal; The memory can be random access memory (RAM) or non-volatile memory, such as disk storage.

[0055] Example 4 The present invention also provides a readable storage medium storing a computer program that, when executed by a processor, implements the method provided in the foregoing embodiments.

[0056] In this embodiment of the invention, the readable storage medium is a computer-readable storage medium and can be disposed in the aforementioned processing device, for example, as a memory in the processing device. Furthermore, the readable storage medium can also be any medium capable of storing program code, such as a USB flash drive, portable hard drive, read-only memory (ROM), magnetic disk, or optical disk.

[0057] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims. The information disclosed in the background section is intended only to enhance the understanding of the overall background technology of the present invention and should not be construed as an admission or implication in any way that such information constitutes prior art known to those skilled in the art.

Claims

1. A method for optimizing an artificial intelligence-generated image detection model, characterized in that, include: Obtain the initial detection model and the initial generation model; A multi-round iterative optimization strategy is adopted to interactively optimize the initial detection model and the initial generation model. This includes: In the nth round of optimization, reinforcement learning is used to construct a reward function based on the detection results of the image samples generated by the generation model after the (n-1)th round of optimization, and then optimizes the generation model after the (n-1)th round of optimization to obtain the generation model after the nth round of optimization. New image samples are then generated from this model and combined with the basic training samples to form the training data for the nth round. Using the training data from the nth round, the detection model after the (n-1)th round of optimization is further optimized through a course-based learning approach to obtain the detection model after the nth round of optimization. The optimization continues for the (n+1)th round until the set number of rounds is reached. The basic samples contain all training data from previous rounds of optimization. If n=1, the generation model after the (n-1)th round of optimization is the initial generation model. The basic samples contain a pre-selected batch of real image samples and image samples generated by the initial generation model. If n>1, the basic samples also contain new image samples from each round.

2. The method for optimizing an artificial intelligence-generated image detection model according to claim 1, characterized in that, The method employs reinforcement learning, combining the detection results of the image samples generated by the optimized generative model in the (n-1)th round with the optimized detection model to construct a reward function, and further optimizing the optimized generative model in the (n-1)th round to obtain the optimized generative model in the nth round, including: Let the detection model after the (n-1)th round of optimization be ? Let the generative model after the (n-1)th round of optimization be ? ; For the generative model after the (n-1)th round of optimization The generated image samples are processed by the detection model optimized in the (n-1)th round. Obtain detection confidence ,in This represents the detection model after the (n-1)th round of optimization. The probability of an image sample being fake is determined; a fake image sample indicates that it was generated by the generative model, while a real image sample indicates that it is a real image. This is combined with the visual evaluation score of the image sample. Construct the following reward function : ； in, and These are preset weighting coefficients; The generative model optimized in the (n-1)th round is further optimized using the reward function to obtain the generative model optimized in the nth round. .

3. The method for optimizing an artificial intelligence-generated image detection model according to claim 1, characterized in that, The process of generating new image samples and combining them with the basic training samples to form the training data for the nth round includes: New image samples are generated using the optimized generative model from the nth round, forming a set. ; If n=1, then the training data for the nth round is: ,in, Based on the training samples, A set of pre-selected real image samples, A collection of image samples generated for the initial generative model; If n>1, then the training data for the nth round is: Among them, excluding Other data Based on the training samples, This is the set of new image samples from the first round.

4. The method for optimizing an artificial intelligence-generated image detection model according to claim 1 or 3, characterized in that, The optimization of the detection model after the (n-1)th round of optimization through course-based learning to obtain the detection model after the nth round of optimization includes: First, use the basic training samples to optimize the detection model in the (n-1)th round. Optimize the model, then gradually add new image samples from the nth round, and finally complete the optimization of the detection model, obtaining the optimized detection model after the nth round. .

5. An AI-generated image detection model optimization system, characterized in that, include: The initial model acquisition unit is used to acquire the initial detection model and the initial generation model; The interactive optimization unit is used to interactively optimize the initial detection model and the initial generation model using a multi-round iterative optimization strategy. This includes: In the nth round of optimization, reinforcement learning is used to construct a reward function based on the detection results of the image samples generated by the generation model after the (n-1)th round of optimization, and then optimizes the generation model after the (n-1)th round of optimization to obtain the generation model after the nth round of optimization. New image samples are then generated from this model and combined with the basic training samples to form the training data for the nth round. Using the training data from the nth round, the detection model after the (n-1)th round of optimization is further optimized through a course-based learning approach to obtain the detection model after the nth round of optimization. The optimization continues for the (n+1)th round until a set number of rounds is reached. The basic samples contain all training data from previous rounds of optimization. If n=1, the generation model after the (n-1)th round of optimization is the initial generation model. The basic samples contain a pre-selected batch of real image samples and image samples generated by the initial generation model. If n>1, the basic samples also contain new image samples from each round.

6. The artificial intelligence-generated image detection model optimization system according to claim 5, characterized in that, The method employs reinforcement learning, combining the detection results of the image samples generated by the optimized generative model in the (n-1)th round with the optimized detection model to construct a reward function, and further optimizing the optimized generative model in the (n-1)th round to obtain the optimized generative model in the nth round, including: Let the detection model after the (n-1)th round of optimization be ? Let the generative model after the (n-1)th round of optimization be ? ; For the generative model after the (n-1)th round of optimization The generated image samples are processed by the detection model optimized in the (n-1)th round. Obtain detection confidence ,in This represents the detection model after the (n-1)th round of optimization. The probability of an image sample being fake is determined; a fake image sample indicates that it was generated by the generative model, while a real image sample indicates that it is a real image. This is combined with the visual evaluation score of the image sample. Construct the following reward function : ； in, and These are preset weighting coefficients; The generative model optimized in the (n-1)th round is further optimized using the reward function to obtain the generative model optimized in the nth round. .

7. The artificial intelligence-generated image detection model optimization system according to claim 5, characterized in that, The process of generating new image samples and combining them with the basic training samples to form the training data for the nth round includes: New image samples are generated using the optimized generative model from the nth round, forming a set. ; If n=1, then the training data for the nth round is: ,in, Based on the training samples, A set of pre-selected real image samples, A collection of image samples generated for the initial generative model; If n>1, then the training data for the nth round is: Among them, excluding Other data Based on the training samples, This is the set of new image samples from the first round.

8. The artificial intelligence-generated image detection model optimization system according to claim 5 or 7, characterized in that, The optimization of the detection model after the (n-1)th round of optimization through course-based learning to obtain the detection model after the nth round of optimization includes: First, use the basic training samples to optimize the detection model in the (n-1)th round. Optimize the model, then gradually add new image samples from the nth round, and finally complete the optimization of the detection model, obtaining the optimized detection model after the nth round. .

9. A processing device, characterized in that, include: One or more processors; Memory, used to store one or more programs; Wherein, when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the method as described in any one of claims 1 to 4.

10. A readable storage medium storing a computer program, characterized in that, When a computer program is executed by a processor, it implements the method as described in any one of claims 1 to 4.

Images