AIGC large model image detection training system

Through the AIGC big model image detection training system, images are preprocessed and feature extracted using diversified data sets and deep learning algorithms, which solves the problem of incomplete data information in deep learning model training, and improves the training accuracy and generalization ability of the model .

CN119090826APending Publication Date: 2024-12-06HANGZHOU KAILIN CULTURE TECHNOLOGY CO LTD
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202411132875.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-16
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

In the prior art, after the deep learning model tries the image blocks on the original image, the remaining part is not used as input samples, resulting in the model training being unable to contain all the original data information of the original image, affecting the model training accuracy.

Method used

A AIGC large-model image detection training system is designed, including model image data, data processing system, data training system and data evaluation system. By collecting real images and false images, diversified data sets are constructed, data preprocessing, feature extraction and Model training and improve performance through model tuning.

Benefits of technology

By training the model image data as a whole, the training accuracy and generalization ability of the model are improved, ensuring that the model can be trained using all data information.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119090826A_ABST
    Figure CN119090826A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of model image training, and discloses an AIGC large model image detection training system which comprises model image data, a data processing system, a data training system, a data evaluation system and a data output module. And the processed data set is transmitted to a data training system, so that the training system can train model image data, and during data training, useful feature information is extracted from the image by using an image processing technology and a deep learning algorithm, and a proper deep learning model architecture is selected according to requirements. According to the method, the marked image data are trained through the model architecture, the data are integrated through the image data processing system, the model image data are integrally trained through the model architecture, and the precision of model training is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of model image training, and in particular to an AIGC large model image detection training system. Background Art

[0002] Generative AI - AIGC (Artificial Intelligence Generated Content) refers to the technology based on artificial intelligence technical methods such as generative adversarial networks and large pre-trained models. It generates relevant content with appropriate generalization capabilities through learning and identifying existing data. The core idea of ​​AIGC technology is to use artificial intelligence algorithms to generate content with certain creativity and quality. Through training models and learning from a large amount of data, AIGC can generate relevant content based on input conditions or guidance;

[0003] In the prior art, several image blocks are randomly cropped from the original image as input samples of the deep learning model for training. The disadvantage of this method is that the remaining part after cropping the image blocks from the original image will not be used as the input sample of the model. Therefore, there is a problem that the input sample of the deep learning model cannot contain all the original data information of the original image, which in turn affects the accuracy of model training. Summary of the invention

[0004] In view of the deficiencies in the prior art, the present invention provides an AIGC large model image detection training system, which solves the problems raised in the above background technology.

[0005] To solve the above technical problems, according to one aspect of the present invention, more specifically, an AIGC large model image detection training system includes model image data, a data processing system, a data training system, a data evaluation system, and a data output module, which is characterized by: Model image data: used to collect real images and false images generated by different AI models to construct a diversified data set;

[0006] Data processing system: used to pre-process the model image data set and transmit the processed data set to the data training system to train the model image, wherein the data processing system includes a data receiving module, a data cleaning module, a data enhancement module and a data labeling module;

[0007] Data training system: Improves the performance of model images by learning and analyzing model image data, including feature extraction module, model component module and model training module;

[0008] Data evaluation system: evaluates the test set obtained by the data training system, and tunes the model graph based on the evaluation, adjusts the model structure, parameter settings, and training strategies to further improve performance. This includes the model test set performance evaluation module and the model performance evaluation module.

[0009] Data output module: used to output the trained and evaluated data to the outside so that operators can organize and store the data.

[0010] Furthermore, the data processing system includes a data receiving module, a data cleaning module, a data enhancement module and a data annotation module;

[0011] Data receiving module: used to receive model image data;

[0012] Data cleaning module: used to remove noise, duplicate or damaged images to ensure data quality;

[0013] Data enhancement module: Increase the diversity of the data set and improve the generalization ability of the model through rotation, scaling, cropping, and flipping operations;

[0014] Data annotation module: used to annotate images so that the model can learn during training.

[0015] Furthermore, the data training system includes a feature extraction module, a model component module and a model training module;

[0016] Feature extraction module: uses image processing technology and deep learning algorithms to extract useful feature information from images;

[0017] Model component module: select the appropriate deep learning model architecture according to needs;

[0018] Model training module: Use labeled image data to train the model, calculate the loss through forward propagation, and update the model's weights and biases through back propagation.

[0019] Furthermore, the model training module includes setting hyperparameters, setting training codes and training models;

[0020] Setting hyperparameters: Train the model by setting parameters and adjust the hyperparameters through grid search, random search, and Bayesian optimization methods to find the optimal model configuration;

[0021] Set up training code: Write training code according to the set hyperparameters, and the training code includes data loading, model forward propagation, loss calculation, back propagation, and parameter update steps;

[0022] Training model: Generate a training model through training code, and use the validation set to monitor the training process of the model to prevent overfitting.

[0023] Furthermore, the data training system affects the performance of model training according to the model covariate shift effect, wherein the calculation formula of the model covariate shift rate is:

[0024]

[0025] Where θ is the model covariate shift effect; i, j are the coordinates of the output feature map, m, n are the sizes of the convolution kernel; N is the total number of pixels in the model image; I i is the gray value of each pixel in the model image; μ is the mean value of the pixel values ​​in the model image.

[0026] Furthermore, the data processing system normalizes and standardizes the model image data, which helps to accelerate model training and improve model performance, wherein:

[0027] The calculation formula for normalizing the model image data is:

[0028]

[0029] In the formula, G is the normalized pixel value of the model image; α is the original pixel value of the model image; β is the minimum value of the model image pixel; γ is the maximum value of the model image pixel;

[0030] The calculation formula for model image data standardization is:

[0031]

[0032] Where B is the standardized pixel value of the model image; α is the original pixel value of the model image; ω is the mean value of the model image pixels; σ is the standard deviation of the model image pixel values.

[0033] Furthermore, the data evaluation system includes a model test set performance evaluation module and a model performance evaluation module;

[0034] Model test set performance evaluation module: used to evaluate the data set obtained by the data training module to reflect the generalization ability of the model on unseen data;

[0035] Model performance evaluation module: used to analyze the model's prediction results and error cases, understand the model's strengths and weaknesses, and adjust the model structure and hyperparameters based on the evaluation results to improve model performance.

[0036] A method for using an AIGC large model image detection training system comprises the following steps:

[0037] S1. Collect model image data: collect real images and fake images generated by different AI models and build a diverse dataset;

[0038] S2. Model image data processing: preprocessing the model image data set and transmitting the processed data set to the data training system so that the training system can train the model image data;

[0039] S3. Model image data training: Use image processing technology and deep learning algorithms to extract useful feature information from images, select a suitable deep learning model architecture based on demand, and train the model on labeled image data through the model architecture;

[0040] S4, model image training evaluation: evaluate the test set obtained by training the data training system, and tune the model graph based on the evaluation, adjust the model structure, parameter settings, and training strategy to further improve performance;

[0041] S5. Data output: Output the trained and evaluated data to the outside, and organize and store the data.

[0042] The beneficial effects of the AIGC large model image detection training system of the present invention are:

[0043] The present invention preprocesses the model image data set through the model image data system, and transmits the processed data set to the data training system so that the training system can train the model image data. During the data training, image processing technology and deep learning algorithms are used to extract useful feature information from the image, and a suitable deep learning model architecture is selected according to the needs. The model is trained on the labeled image data through the model architecture. The present invention integrates the data through the image data processing system, and uses the model framework to train the model image data as a whole, thereby increasing the accuracy of the model training. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The present invention is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0045] Figure 1 It is a structural schematic diagram of the present invention. DETAILED DESCRIPTION

[0046] The present invention will be described in detail below with reference to the accompanying drawings and in combination with embodiments. It should be noted that the embodiments and features in the embodiments of the present application can be combined with each other without conflict.

[0047] like Figure 1As shown, according to one aspect of the present invention, there is provided an AIGC large model image detection training system, including model image data, a data processing system, a data training system, a data evaluation system and a data output module, characterized in that: Model image data: used to collect real images and false images generated by different AI models to construct a diversified data set;

[0048] Data processing system: used to pre-process the model image data set and transmit the processed data set to the data training system to train the model image, wherein the data processing system includes a data receiving module, a data cleaning module, a data enhancement module and a data labeling module;

[0049] Data training system: Improves the performance of model images by learning and analyzing model image data, including feature extraction module, model component module and model training module;

[0050] Data evaluation system: evaluates the test set obtained by the data training system, and tunes the model graph based on the evaluation, adjusts the model structure, parameter settings, and training strategies to further improve performance. This includes the model test set performance evaluation module and the model performance evaluation module.

[0051] Data output module: used to output the trained and evaluated data to the outside so that operators can organize and store the data.

[0052] In this embodiment, the data processing system includes a data receiving module, a data cleaning module, a data enhancement module and a data annotation module;

[0053] Data receiving module: used to receive model image data;

[0054] Data cleaning module: used to remove noise, duplicate or damaged images to ensure data quality;

[0055] Data enhancement module: Increase the diversity of the data set and improve the generalization ability of the model through rotation, scaling, cropping, and flipping operations;

[0056] Data annotation module: used to annotate images so that the model can learn during training.

[0057] In this embodiment, the data training system includes a feature extraction module, a model component module and a model training module;

[0058] Feature extraction module: uses image processing technology and deep learning algorithms to extract useful feature information from images;

[0059] Model component module: select the appropriate deep learning model architecture according to needs;

[0060] Model training module: Use labeled image data to train the model, calculate the loss through forward propagation, and update the model's weights and biases through back propagation.

[0061] In this embodiment, the model training module includes setting hyperparameters, setting training codes, and training models;

[0062] Setting hyperparameters: Train the model by setting parameters and adjust the hyperparameters through grid search, random search, and Bayesian optimization methods to find the optimal model configuration;

[0063] Set up training code: Write training code according to the set hyperparameters, and the training code includes data loading, model forward propagation, loss calculation, back propagation, and parameter update steps;

[0064] Training model: Generate a training model through training code, and use the validation set to monitor the training process of the model to prevent overfitting.

[0065] In this embodiment, the data training system affects the performance of model training according to the model covariate shift effect, wherein the calculation formula of the model covariate shift rate is:

[0066]

[0067] Where θ is the model covariate shift effect; i, j are the coordinates of the output feature map, m, n are the sizes of the convolution kernel; N is the total number of pixels in the model image; I i is the gray value of each pixel in the model image; μ is the mean value of the pixel values ​​in the model image.

[0068] In this embodiment, the data processing system normalizes and standardizes the model image data, which helps to accelerate model training and improve model performance, wherein:

[0069] The calculation formula for normalizing the model image data is:

[0070]

[0071] In the formula, G is the normalized pixel value of the model image; α is the original pixel value of the model image; β is the minimum value of the model image pixel; γ is the maximum value of the model image pixel;

[0072] The calculation formula for model image data standardization is:

[0073]

[0074] Where B is the standardized pixel value of the model image; α is the original pixel value of the model image; ω is the mean value of the model image pixels; σ is the standard deviation of the model image pixel values.

[0075] In this embodiment, the data evaluation system includes a model test set performance evaluation module and a model performance evaluation module;

[0076] Model test set performance evaluation module: used to evaluate the data set obtained by the data training module to reflect the generalization ability of the model on unseen data;

[0077] Model performance evaluation module: used to analyze the model's prediction results and error cases, understand the model's strengths and weaknesses, and adjust the model structure and hyperparameters based on the evaluation results to improve model performance.

[0078] A method for using an AIGC large model image detection training system comprises the following steps:

[0079] S1. Collect model image data: collect real images and fake images generated by different AI models and build a diverse dataset;

[0080] S2. Model image data processing: preprocessing the model image data set and transmitting the processed data set to the data training system so that the training system can train the model image data;

[0081] S3. Model image data training: Use image processing technology and deep learning algorithms to extract useful feature information from images, select a suitable deep learning model architecture based on demand, and train the model on labeled image data through the model architecture;

[0082] S4, model image training evaluation: evaluate the test set obtained by training the data training system, and tune the model graph based on the evaluation, adjust the model structure, parameter settings, and training strategy to further improve performance;

[0083] S5. Data output: Output the trained and evaluated data to the outside, and organize and store the data.

[0084] Of course, the above description is not a limitation of the present invention, and the present invention is not limited to the above examples. Changes, modifications, additions or substitutions made by ordinary technicians in this technical field within the essential scope of the present invention also fall within the protection scope of the present invention.

Claims

1. An AIGC large model image detection training system, comprising model image data, a data processing system, a data training system, a data evaluation system and a data output module, characterized in that: Model image data: used to collect real images and fake images generated by different AI models to build a diverse dataset; Data processing system: used to pre-process the model image data set and transmit the processed data set to the data training system to train the model image, wherein the data processing system includes a data receiving module, a data cleaning module, a data enhancement module and a data labeling module; Data training system: Improves the performance of model images by learning and analyzing model image data, including feature extraction module, model component module and model training module; Data evaluation system: evaluates the test set obtained by the data training system, and tunes the model graph based on the evaluation, adjusts the model structure, parameter settings, and training strategies to further improve performance. This includes the model test set performance evaluation module and the model performance evaluation module. Data output module: used to output the trained and evaluated data to the outside so that operators can organize and store the data.

2. The AIGC large model image detection training system according to claim 1, characterized in that: The data processing system includes a data receiving module, a data cleaning module, a data enhancement module and a data annotation module; Data receiving module: used to receive model image data; Data cleaning module: used to remove noise, duplicate or damaged images to ensure data quality; Data enhancement module: Increase the diversity of the data set and improve the generalization ability of the model through rotation, scaling, cropping, and flipping operations; Data annotation module: used to annotate images so that the model can learn during training.

3. The AIGC large model image detection training system according to claim 1, characterized in that: The data training system includes a feature extraction module, a model component module and a model training module; Feature extraction module: uses image processing technology and deep learning algorithms to extract useful feature information from images; Model component module: select the appropriate deep learning model architecture according to needs; Model training module: Use labeled image data to train the model, calculate the loss through forward propagation, and update the model's weights and biases through back propagation.

4. The AIGC large model image detection training system according to claim 3 is characterized by: The model training module includes setting hyperparameters, setting training codes and training models; Setting hyperparameters: Train the model by setting parameters and adjust the hyperparameters through grid search, random search, and Bayesian optimization methods to find the optimal model configuration; Set up training code: Write training code according to the set hyperparameters, and the training code includes data loading, model forward propagation, loss calculation, back propagation, and parameter update steps; Training model: Generate a training model through training code, and use the validation set to monitor the training process of the model to prevent overfitting.

5. The AIGC large model image detection training system according to claim 3 is characterized by: The data training system affects the performance of model training according to the model covariate shift effect, wherein the calculation formula of the model covariate shift rate is: Where θ is the model covariate shift effect; i, j are the coordinates of the output feature map, m, n are the sizes of the convolution kernel; N is the total number of pixels in the model image; I i is the gray value of each pixel in the model image; μ is the mean value of the pixel values ​​in the model image.

6. The AIGC large model image detection training system according to claim 1, characterized in that: The data processing system normalizes and standardizes the model image data, which helps to accelerate model training and improve model performance, wherein: The calculation formula for normalizing the model image data is: In the formula, G is the normalized pixel value of the model image; α is the original pixel value of the model image; β is the minimum value of the model image pixel; γ is the maximum value of the model image pixel; The calculation formula for model image data standardization is: Where B is the standardized pixel value of the model image; α is the original pixel value of the model image; ω is the mean value of the model image pixels; σ is the standard deviation of the model image pixel values.

7. The AIGC large model image detection training system according to claim 1, characterized in that: The data evaluation system includes a model test set performance evaluation module and a model performance evaluation module; Model test set performance evaluation module: used to evaluate the data set obtained by the data training module to reflect the generalization ability of the model on unseen data; Model performance evaluation module: used to analyze the model's prediction results and error cases, understand the model's strengths and weaknesses, and adjust the model structure and hyperparameters based on the evaluation results to improve model performance.

8. An AIGC large model image detection training system, comprising a method for using the AIGC large model image detection training system according to any one of claims 1 to 7, characterized in that: The following steps are involved: S1. Collect model image data: collect real images and fake images generated by different AI models and build a diverse dataset; S2. Model image data processing: preprocessing the model image data set and transmitting the processed data set to the data training system so that the training system can train the model image data; S3. Model image data training: Use image processing technology and deep learning algorithms to extract useful feature information from images, select a suitable deep learning model architecture based on demand, and train the model on labeled image data through the model architecture; S4, model image training evaluation: evaluate the test set obtained by training the data training system, and tune the model graph based on the evaluation, adjust the model structure, parameter settings, and training strategy to further improve performance; S5. Data output: Output the trained and evaluated data to the outside, and organize and store the data.

Citation Information

Cited By

  • AI image model training data integration processing system for multi-source data

    CN120375120A