Intelligent data processing system and method based on AI diffusion model

Abstract

The invention discloses an intelligent data processing system and method based on an AI diffusion model, and the system comprises a data collection module, a data transmission module, a cross-modal feature fusion module and an intelligent decision module, and relates to the technical field of data processing. According to the intelligent data processing system and method based on the AI diffusion model, multi-modal data are coded and classified, feature data which are coded and located at the same time node are extracted, the AI diffusion model is used for denoising processing, and the multi-modal data subjected to denoising processing are fused to form a popularization scheme; according to the cross-border digital business data processing system, the data is automatically processed from collection to analysis while anomaly detection is achieved through a double-classification framework and a feature denoising extraction module, the data processing efficiency is improved, the data processing requirement in cross-border digital business can be quickly responded, and the required data in different regions can be effectively extracted and dynamically adapted.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically to an intelligent data processing system and method based on an AI diffusion model. Background Technology

[0002] In the current booming development of cross-border digital export, data processing faces many challenges. With the advent of the AI ​​era, the amount of data is growing explosively, and the types of data are complex and diverse, including structured data and unstructured data, such as product information and user reviews in cross-border e-commerce, and text, images and videos in social media.

[0003] The reference patent is titled: "A Method and Apparatus for Constructing a Multimodal Content Generation AI Model Based on a Diffusion Model" (Patent Publication No.: CN120277531A, Patent Publication Date: 2025-07-08). The method includes: a multi-source data processing module, a modality priority analysis module, a serialized modality processing module, a diffusion parameter adaptive adjustment module, an interactive feedback integration module, and a collaborative output adjustment module. By integrating multimodal data input and dynamically adjusting model parameters, it uses information entropy to evaluate the information content of the data and sets processing priorities accordingly, enhancing the model's ability to identify the importance of data. It adjusts model parameters in real time to adapt to the characteristics of different data modalities and dynamically adjusts generation parameters based on user feedback, making the generated content more in line with the actual needs and preferences of users. By coordinating the style, theme, and emotion of the output content, it ensures the consistency and integration of the content, improves the attractiveness of the content, and enhances the dissemination effect of the content.

[0004] Based on the above-mentioned documents, although existing AI models have also achieved cross-domain data processing, they are difficult to cope with the dynamically changing data distribution and complex cross-domain data environment in cross-border digital export business operations. As a result, when processing data to form a promotion plan, they rely solely on existing data from others and fail to extract valuable data to form the features required by different regions. Therefore, this invention provides an intelligent data processing system and method based on an AI diffusion model. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an intelligent data processing system and method based on an AI diffusion model. This solves the problem that while existing AI models can achieve cross-domain data processing, they are unable to cope with dynamically changing data distributions and complex cross-domain data environments in cross-border digital export business operations. Consequently, when processing data for promotion plans, they rely solely on existing data from others and fail to extract valuable data to form features suitable for different regions.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an intelligent data processing system based on an AI diffusion model, comprising:

[0007] The data acquisition module is used to collect various types of data in cross-border digital business and to realize the real-time acquisition, transmission and storage of data from different regions;

[0008] The data transmission module enables the acquisition end and the analysis end to establish a transmission protocol and perform wireless communication transmission operations;

[0009] The cross-modal feature fusion module encodes and classifies multimodal data, extracts the encoded feature data located at the same time point, and uses an AI diffusion model for denoising. The denoised multimodal data is then fused to form a promotion scheme.

[0010] The intelligent decision-making module, combining promotion plans with preset business rules and strategies, provides decision support for cross-border digital business.

[0011] Preferably, the cross-modal feature fusion module includes:

[0012] The data encoding and classification module is used to convert the collected raw data into feature data suitable for processing by the AI ​​diffusion model;

[0013] The feature denoising and extraction module uses a pre-trained AI diffusion model to process the encoded data. Through the process of gradually adding or removing noise, it realizes the generation, reconstruction and feature extraction of data.

[0014] A multi-source data construction model is used to fuse the extracted data through a data fusion model to form a joint feature scheme for images, text, and audio.

[0015] Preferably, the operation of establishing a transmission protocol in the data transmission module is as follows:

[0016] The transceiver nodes located at the acquisition end and the analysis end select the same transmission protocol. The acquisition end corresponds to the various data acquisition platforms of the data acquisition module, and the analysis end corresponds to the application terminal of the cross-modal feature fusion module.

[0017] The acquisition end sends a matching command to the analysis end, and after confirming that the matching command has been received, the analysis end generates a feedback command to the acquisition end, thus completing the establishment of the current encrypted transmission channel. During the transmission process, the acquisition end encrypts and compresses the data, while the analysis end decrypts and decompresses the received data packets.

[0018] Preferably, the operation in the data encoding and classification module to convert the feature data into data suitable for processing by the AI ​​diffusion model is as follows:

[0019] The dual-classification architecture includes a primary classification node and a secondary classification node. The primary classification node is used to capture the category semantics of the data to ensure that the generated data has an independent category, while the secondary classification node is used to capture the data style characteristics of different regions and corresponds to the subcategories under the independent category.

[0020] The desired style features and data features to be transformed are determined. After extracting the semantic categories based on the desired style features, the data features to be transformed are converted into the desired semantic categories through adversarial learning, thus obtaining the feature data for subsequent AI diffusion model processing.

[0021] Preferably, the feature denoising and extraction module performs the data generation, reconstruction, and feature extraction operations by gradually adding or removing noise as follows:

[0022] Extract the parameter data to be applied, including data features that are located in the same region under the same main category node but are represented in different ways;

[0023] Select an image data feature, use an AI diffusion model to perform advanced training on the encoded image data feature, determine whether the generated data meets the market demand of the application area, and thus derive the basic semantics of the image data feature.

[0024] Extract all semantics from similar image data features, determine the semantics that match the image data features based on the frequency of semantic occurrence, and construct optimized generalization parameters by combining the image data features and the matched semantics.

[0025] Preferably, the operation of using an AI diffusion model to perform advanced training on the encoded image data features is as follows:

[0026] New sample data is added incrementally, and the sample data consists of image data features of the same type with noise from historical data.

[0027] After the sample data is introduced, the denoising features are obtained by comparing the sample data with the current image data features. The denoising features are then removed from the current image data features to form new image data features. This process is continued until the generated image data features meet the market demand of the application area.

[0028] In the generated data that meets the market demand of the application area, the basic semantics of image data features are determined. That is, the similarity threshold of image data for market demand is set as M, and the similarity is obtained by comparing the current optimized image data features with the demand image data. That is, when N < M, the current image data features need to continue to introduce samples for noise reduction. Conversely, when N > M, the current image data meets the market demand of the application area.

[0029] Preferably, the operation of comparing the features of sample data and current image data to obtain the denoising features is as follows:

[0030] Extract sample image data labeled with noise type, and ensure that the sample image data has the same size as the current image data features;

[0031] Reference points are set on the surface of the sample image data and the current image data features. Multiple reference points are set at equal intervals from the lower left corner of the image feature to the upper left corner, and reference points are set at equal intervals to the right based on the multiple reference points.

[0032] The gray values ​​of reference points at corresponding positions in the two image data are extracted and compared. Those with abnormal gray values ​​are marked, and the abnormal marked points are summarized and filled into blank image features of the same size according to their positions to form denoising features.

[0033] Preferably, the operation of determining the semantics that match the image data features based on the frequency of semantic occurrence is as follows:

[0034] Define a semantic recognition window F(x, y), where x represents the horizontal length of the recognition window and y represents the vertical width of the recognition window, and use the semantic recognition window to recognize image data features with basic semantics of the same type;

[0035] The recognition window F(x, y) is adjusted to fit the number of characters in the content. The basic window width of F(x, y) recognizes two characters, and it expands according to the number of characters in the content. The expansion distance is [(number of characters in the content - 2) × character width + character spacing × (number of characters in the content - 3)];

[0036] The identified content is confirmed by classifying the same content and confirming the number of times it appears. The content is then sorted from most frequent to least frequent, with the most frequent content being the final semantics of the current single category.

[0037] Preferably, the multi-source data construction model uses a data fusion model to fuse the extracted data to form a joint feature scheme for images, text, and audio. The operation is as follows:

[0038] When extracting modal feature data, the optimized image feature data is extracted, and the semantic part related to the image feature data is extracted. The semantics are adjusted according to the language rules of the application region. At the same time, the audio data related to the image feature data is extracted. The similarity between the content features of the current audio data and the image feature data meets the image similarity threshold of the application region, and the text content of the audio data conforms to the semantics of the application region.

[0039] Finally, the extracted modal features are mapped to a unified space to form a joint feature scheme for images, text, and audio.

[0040] This invention discloses an intelligent data processing method based on an AI diffusion model, specifically including the following steps:

[0041] S1. Collect data from different regions used for cross-border digital business and achieve wireless transmission and storage;

[0042] S2. After completing the data classification and processing, the AI ​​diffusion model is used to optimize the data, and the optimized multimodal data is extracted and fused to form a promotion plan.

[0043] S3. Generate decisions for cross-border digital business based on the promotion plan, and the plan is displayed through a visual interface.

[0044] This invention provides an intelligent data processing system and method based on an AI diffusion model. Compared with existing technologies, it has the following advantages:

[0045] 1. This intelligent data processing system and method based on an AI diffusion model encodes and classifies multimodal data, extracts the encoded feature data located at the same time point, and uses an AI diffusion model for denoising. The denoised multimodal data is then fused to form a promotion scheme. With a dual-classification architecture and a feature denoising extraction module, it achieves anomaly detection while automating the data processing from collection to analysis, improving data processing efficiency. It can quickly respond to the data processing needs in cross-border digital business and effectively extract and dynamically adapt the required data for different regions.

[0046] 2. This intelligent data processing system and method based on an AI diffusion model uses an AI diffusion model to perform progressive training on the encoded image data features, determines whether the generated data meets the market demand of the application area, and thereby derives the basic semantics of the image data features. All semantics of similar matching image data features are extracted, and semantics that match the image data features are determined based on the frequency of semantic occurrence. The image data features and the matched semantics are then used to construct optimized generalization parameters, realizing semantic alignment and collaborative generation of multimodal data. This allows for the extraction of valuable data from the data, improving the dynamic adaptability, accuracy, and efficiency of the intelligent data processing system.

[0047] 3. This intelligent data processing system and method based on an AI diffusion model extracts optimized image feature data and semantic components related to the image feature data, and adjusts the semantics according to the language rules of the application region. Simultaneously, it extracts audio data related to the image feature data. The similarity between the content features of the audio data and the image feature data meets the image similarity threshold of the application region, and the text content of the audio data conforms to the semantics of the application region. Finally, the extracted modal features are mapped to a unified space to form a joint feature scheme of image, text, and audio, realizing the integration of data processing, providing intelligent decision support for cross-border digital business, helping enterprises to better cope with market differences in different regions, and formulate more scientific and reasonable business strategies. Attached Figure Description

[0048] Figure 1 This is a schematic diagram of the data processing system of the present invention;

[0049] Figure 2 This is a flowchart illustrating the operation of the data processing method of the present invention. Detailed Implementation

[0050] 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 scope of protection of the present invention.

[0051] Please see Figures 1-2 This invention provides two technical solutions:

[0052] Example 1: An intelligent data processing system based on an AI diffusion model, comprising:

[0053] The data acquisition module is used to collect various types of data in cross-border digital business and realize the real-time acquisition, transmission and storage of data from different regions, including user behavior data, market data, commodity data and so on from different countries and regions. This module supports multiple data formats and data sources and can realize real-time and batch data acquisition.

[0054] The data transmission module enables the acquisition end and the analysis end to establish a transmission protocol and perform wireless communication transmission operations;

[0055] The cross-modal feature fusion module encodes and classifies multimodal data, extracts the encoded feature data located at the same time point, and uses an AI diffusion model for denoising. The denoised multimodal data is then fused to form a promotion scheme.

[0056] The intelligent decision-making module, combining promotion plans with preset business rules and strategies, provides decision support for cross-border digital business.

[0057] By encoding and classifying multimodal data, extracting the encoded feature data located at the same time point, and using an AI diffusion model for denoising, the denoised multimodal data is fused to form a promotion scheme. With a dual classification architecture and feature denoising extraction module, anomaly detection is achieved while automating the data processing from collection to analysis, improving data processing efficiency, enabling rapid response to data processing needs in cross-border digital business, and effectively extracting and dynamically adapting the required data for different regions.

[0058] In this embodiment of the invention, the cross-modal feature fusion module includes:

[0059] The data encoding and classification module is used to convert the collected raw data into feature data suitable for processing by the AI ​​diffusion model;

[0060] The feature denoising and extraction module uses a pre-trained AI diffusion model to process the encoded data. Through the process of gradually adding or removing noise, it realizes the generation, reconstruction and feature extraction of data.

[0061] A multi-source data construction model is used to fuse the extracted data through a data fusion model to form a joint feature scheme for images, text, and audio.

[0062] In this embodiment of the invention, the operation of establishing a transmission protocol in the data transmission module is as follows:

[0063] The transceiver nodes located at the acquisition end and the analysis end select the same transmission protocol. The acquisition end corresponds to the various data acquisition platforms of the data acquisition module, and the analysis end corresponds to the application terminal of the cross-modal feature fusion module.

[0064] The acquisition end sends a matching command to the analysis end, and after confirming that the matching command has been received, the analysis end generates a feedback command to the acquisition end, thus completing the establishment of the current encrypted transmission channel. During the transmission process, the acquisition end encrypts and compresses the data, while the analysis end decrypts and decompresses the received data packets.

[0065] In this embodiment of the invention, the operation of converting feature data into data suitable for processing by the AI ​​diffusion model in the data encoding and classification module is as follows:

[0066] The dual-classification architecture includes a primary classification node and a secondary classification node. The primary classification node is used to capture the category semantics of the data to ensure that the generated data has an independent category, while the secondary classification node is used to capture the data style characteristics of different regions and corresponds to the subcategories under the independent category.

[0067] The desired style features and data features to be transformed are determined. After extracting the semantic categories based on the desired style features, the data features to be transformed are converted into the desired semantic categories through adversarial learning, thus obtaining the feature data for subsequent AI diffusion model processing.

[0068] Among them, adversarial learning data features are used to transform style differences between different sub-classification nodes to maintain category consistency. Adversarial learning is an existing mature method, which is a machine learning method that improves the robustness of the model by constructing an adversarial game framework. Its core applications are reflected in the fields of security defense and generative adversarial networks (GAN).

[0069] In this embodiment of the invention, the feature denoising and extraction module performs the following operations to generate, reconstruct, and extract features by gradually adding or removing noise:

[0070] Extract the parameter data to be applied, including data features that are located in the same region under the same main category node but are represented in different ways;

[0071] Select an image data feature, use an AI diffusion model to perform advanced training on the encoded image data feature, determine whether the generated data meets the market demand of the application area, and thus derive the basic semantics of the image data feature.

[0072] Extract all semantics from similar image data features, determine the semantics that match the image data features based on the frequency of semantic occurrence, and construct optimized generalization parameters by combining the image data features and the matched semantics.

[0073] By employing an AI diffusion model to perform advanced training on the encoded image data features, it is determined whether the generated data meets the market demands of the application region. This leads to the deriving of the basic semantics of the image data features. All semantics of similar matching image data features are extracted, and semantics consistent with the image data features are determined based on the frequency of semantic occurrence. The image data features and the matched semantics are then combined to construct optimized generalization parameters, achieving semantic alignment and collaborative generation of multimodal data. This process uncovers valuable data within the data and improves the dynamic adaptability, accuracy, and efficiency of the intelligent data processing system.

[0074] In this embodiment of the invention, the operation of using an AI diffusion model to perform advanced training on the encoded image data features is as follows:

[0075] New sample data is added incrementally, and the sample data consists of image data features of the same type with noise from historical data.

[0076] After the sample data is introduced, the denoising features are obtained by comparing the sample data with the current image data features. The denoising features are then removed from the current image data features to form new image data features. This process is continued until the generated image data features meet the market demand of the application area.

[0077] In the generated data that meets the market demand of the application area, the basic semantics of image data features are determined. That is, the similarity threshold of image data for market demand is set as M, and the similarity is obtained by comparing the current optimized image data features with the demand image data. That is, when N < M, the current image data features need to continue to introduce samples for noise reduction. Conversely, when N > M, the current image data meets the market demand of the application area.

[0078] In this embodiment of the invention, the operation of obtaining denoising features by comparing sample data and current image data features is as follows:

[0079] Extract sample image data labeled with noise type, and ensure that the sample image data has the same size as the current image data features;

[0080] Reference points are set on the surface of the sample image data and the current image data features. Multiple reference points are set at equal intervals from the lower left corner of the image feature to the upper left corner, and reference points are set at equal intervals to the right based on the multiple reference points.

[0081] The gray values ​​of reference points at corresponding positions in the two image data are extracted and compared. Those with abnormal gray values ​​are marked, and the abnormal marked points are summarized and filled into blank image features of the same size according to their positions to form denoising features.

[0082] In this embodiment of the invention, the operation of determining the semantics that match the image data features based on the frequency of semantic occurrence is as follows:

[0083] Define a semantic recognition window F(x, y), where x represents the horizontal length of the recognition window and y represents the vertical width of the recognition window, and use the semantic recognition window to recognize image data features with basic semantics of the same type;

[0084] The recognition window F(x, y) is adjusted to fit the number of characters in the content. The basic window width of F(x, y) recognizes two characters, and it expands according to the number of characters in the content. The expansion distance is [(number of characters in the content - 2) × character width + character spacing × (number of characters in the content - 3)];

[0085] The identified content is confirmed by classifying the same content and confirming the number of times it appears. The content is then sorted from most frequent to least frequent, with the most frequent content being the final semantics of the current single category.

[0086] In this embodiment of the invention, the multi-source data construction model fuses the extracted data through a data fusion model to form a joint feature scheme for images, text, and audio. The operation is as follows:

[0087] When extracting modal feature data, the optimized image feature data is extracted, and the semantic part related to the image feature data is extracted. The semantics are adjusted according to the language rules of the application region. At the same time, the audio data related to the image feature data is extracted. The similarity between the content features of the current audio data and the image feature data meets the image similarity threshold of the application region, and the text content of the audio data conforms to the semantics of the application region.

[0088] Finally, the extracted modal features are mapped to a unified space to form a joint feature scheme for images, text, and audio.

[0089] By extracting optimized image feature data and semantic components related to it, and adjusting the semantics according to the language rules of the application region, while simultaneously extracting audio data related to the image feature data, the similarity between the content features of the audio data and the image feature data meets the image similarity threshold of the application region, and the text content of the audio data conforms to the semantics of the application region, the extracted modal features are finally mapped to a unified space to form a joint feature scheme of image, text, and audio. This achieves the integration of processed data, provides intelligent decision support for cross-border digital business, and helps enterprises better cope with market differences in different regions and formulate more scientific and reasonable business strategies.

[0090] Example 2 differs from Example 1 in that: This invention discloses an intelligent data processing method based on an AI diffusion model, specifically including the following steps:

[0091] S1. Collect data from different regions used for cross-border digital business and achieve wireless transmission and storage;

[0092] S2. After completing the data classification and processing, the AI ​​diffusion model is used to optimize the data, and the optimized multimodal data is extracted and fused to form a promotion plan.

[0093] S3. Generate decisions for cross-border digital business based on the promotion plan, and the plan is displayed through a visual interface.

[0094] Furthermore, any content not described in detail in this specification is existing technology known to those skilled in the art.

[0095] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0096] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An intelligent data processing system based on an AI diffusion model, characterized in that: The application relates to a cross-border digital business intelligent decision support system. The system comprises the following: a data collection module for collecting various types of data in cross-border digital business and realizing real-time collection, transmission and storage of data in different regions; a data transmission module for realizing transmission protocol establishment between the collection end and the analysis end and wireless communication transmission operation; a cross-modal feature fusion module for encoding and classifying multi-modal data, extracting feature data under the same time node after encoding, and carrying out denoising treatment by using an AI diffusion model, and fusing the multi-modal data after denoising treatment to form a promotion scheme; 2.The AI diffusion model-based intelligent data processing system of claim 1, wherein: an intelligent decision module for providing decision support for cross-border digital business by combining the promotion scheme and preset business rules and strategies. The cross-modal feature fusion module comprises: a data encoding and classifying module for converting the collected original data into feature data suitable for AI diffusion model processing; a feature denoising and extracting module for processing the encoded data by using a pre-trained AI diffusion model, realizing data generation, reconstruction and feature extraction through a process of gradually adding or removing noise; 3.The AI diffusion model-based intelligent data processing system of claim 1, wherein: a multi-source data construction model for fusing the extracted data to form a joint feature scheme of images, texts and audios by using a data fusion model. The operation of establishing a transmission protocol in the data transmission module is as follows: the transceiving nodes located at the collection end and the analysis end determine to select the same transmission protocol, the collection end corresponds to various data collection platforms of the data collection module, and the analysis end corresponds to application terminals of the cross-modal feature fusion module; 4.The AI diffusion model-based intelligent data processing system of claim 2, wherein: the collection end sends a matching instruction to the analysis end, the analysis end generates a feedback instruction to the collection end after confirming the reception of the matching instruction, the establishment of the current encrypted transmission channel is completed, and data encryption compression is performed at the collection end during transmission, and the received data packet is decrypted and decompressed at the analysis end. The operation of converting the collected original data into feature data suitable for AI diffusion model processing in the data encoding and classifying module is as follows: a double-classification architecture is set, including a main classification node and an auxiliary classification node, the main classification node is used for capturing the category semantics of data to ensure that the generated data has an independent category, and the auxiliary classification node is used for capturing the data style features of different regions and corresponding subcategories under the independent category; 5.The AI diffusion model-based intelligent data processing system according to claim 4, characterized in that: the required application style features and the required data features are determined, the semantic categories are extracted according to the required application style features, the required data features are converted into the required semantic categories through adversarial learning, and the feature data for subsequent AI diffusion model processing is obtained. The operation of realizing data generation, reconstruction and feature extraction through a process of gradually adding or removing noise in the feature denoising and extracting module is as follows: parameter data required for application is extracted, including data features of different data characteristics in the same region under the same main classification node; an image data feature is selected, an AI diffusion model is used for advanced training of the encoded image data feature, and it is determined whether the generated data meets the market demand of the application region, so that the basic semantics of the image data feature are obtained. Extract all semantics matched with similar image data features, and determine the semantics matched with the image data features according to the number of times of the semantics, and construct the image data features and the matched semantics to form the optimized promotion parameters. 6.The AI diffusion model-based intelligent data processing system according to claim 5, characterized in that: The operation of using the AI diffusion model to further train the coded image data features is: By gradually adding new sample data, and the sample data is the image data features of the same type with noise in the historical data; After introducing the sample data, the noise features are obtained by comparing the sample data and the current image data features, the noise features are removed from the current image data features to form new image data features, and the operation is continued until the generated image data features meet the market demand of the application area; In the generated data that meets the market demand of the application area, the basic semantics of the image data features are determined, that is, the image data similarity threshold of the market demand is set as M, and the similarity N is obtained by comparing the current optimized image data features with the demand image data, that is, N < M, then the current image data features need to continue to introduce sample noise, otherwise N > M, then the current image data meets the market demand of the application area. 7.The AI diffusion model-based intelligent data processing system of claim 6, wherein: The operation of comparing the sample data and the current image data features to obtain the noise features is: Extract the sample image data labeled with noise type, and the size of the sample image data is consistent with that of the current image data features; A reference point is set on the surface of the sample image data and the current image data features, and multiple reference points are set at equal distances from the lower left corner of the image feature to the upper left corner, and reference points are set at equal distances to the right side based on the multiple reference points; The reference point gray values of the corresponding positions of the two image data are compared, and the gray values with abnormalities are marked, and the abnormal marks are summarized and filled into the blank image features of the same size according to the positions to form the noise features. 8.The AI diffusion model-based intelligent data processing system of claim 6, wherein: The operation of determining the semantics matched with the image data features according to the number of times of the semantics is: Set the semantic recognition window F(x, y), x represents the horizontal length of the recognition window, and y represents the vertical width of the recognition window, and use the semantic recognition window to recognize the image data features with basic semantics of the same type; The recognition window F(x, y) is adjusted according to the number of content characters, the window F(x, y) has a basic window width of two characters, and is expanded according to the number of content characters, and the expansion distance is [(content character number-2) × character width + character spacing × (content character number-3)]; Confirm the recognized content, classify the same content and confirm the number of times, sort according to the number of times from many to few, that is, the most times are the final semantics of the current single category. 9.The AI diffusion model-based intelligent data processing system of claim 6, wherein: The operation of the multi-source data construction model fusing the extracted data to form the joint feature scheme of image, text and audio through the data fusion model is: In the modal feature data extraction, the optimized image feature data is extracted, the semantic part related to the image feature data is extracted, the semantic adjustment is performed according to the language rules of the application area, the audio data related to the image feature data is extracted, the similarity of the content features of the current audio data and the image feature data satisfies the image similarity threshold of the application area, and the text content of the audio data conforms to the semantics of the application area; Finally, the extracted modal features are mapped to a unified space to form an image, text and audio joint feature scheme.

10. An intelligent data processing method based on an AI diffusion model, using an intelligent data processing system based on an AI diffusion model according to any one of claims 1-9. Specifically, the following steps are included: S1, collecting different regional data for cross-border digital services to realize wireless transmission and storage; S2, after completing the classification processing of the data, the data is optimized by using an AI diffusion model, and the optimized multi-modal data fusion is extracted to form a promotion scheme; S3, generating a decision for the cross-border digital service based on the promotion scheme, and the scheme is displayed through a visual interface.

Images