Image file information automatic extraction method and system based on agent workflow

Abstract

The invention discloses an intelligent agent workflow-based image file information automatic extraction method and system, and relates to the technical field of artificial intelligence and image file processing, and the method comprises the steps of inputting a to-be-processed image file into a visual model to recognize text information; the recognized text information is input into an AI Agent to extract field information, and the information extraction process is dynamically optimized through Agent Workflow; and fusing the field information extracted by each Agent in the AI Agent, and generating structured output data. According to the method, the high-precision character recognition capability of the visual model on the image file is utilized, and the knowledge generalization capability of the AI Agent large language model is combined, so that efficient and accurate information extraction is realized under the condition that data does not need to be labeled; meanwhile, the intelligent agent can intelligently schedule the workflow according to task requirements, the extraction process is dynamically optimized, and the processing time is effectively controlled while high accuracy is guaranteed.

Description

Technical Field

[0001] The present application relates to a method and system that combines artificial intelligence (AI), especially computer vision (CV) technology, and an intelligent agent workflow mechanism, and specifically to an automatic extraction method and system for image file information based on an intelligent agent workflow. Background Art

[0002] Image file information extraction is one of the key technologies for processing non-institutional data. The extracted information can be specific fields, keywords, values, etc. It has wide applications in industries such as finance, law, and medicine that need to process massive documents. Traditional file information extraction methods usually rely on manual annotation and fixed rule-based algorithms. This method is not only inefficient, but also highly dependent on annotated data, which makes the model lack generalization and ability to handle complex scenarios. Especially when facing diverse and complex documents, a large amount of manual intervention and custom rules are often required, which not only increases costs, but also makes it difficult to adapt to rapidly changing application scenarios. In addition, when processing large-scale data, traditional methods usually find it difficult to balance extraction accuracy and efficiency at the same time, which becomes a bottleneck in practical applications. In addition, directly using large visual models to extract information from image files, due to the serious hallucination problem of current large visual models, it is impossible to directly extract a large amount of field information in image files, and there is a problem of low accuracy. Summary of the invention

[0003] In order to solve the technical problems existing in the prior art, the present application proposes a method and system for automatic extraction of image file information based on an intelligent agent workflow. The present application utilizes the high-precision text recognition capability of the visual model for image files, combined with the knowledge generalization capability of the AI ​​Agent large language model, to achieve efficient and accurate information extraction without the need for labeled data; at the same time, the intelligent agent can intelligently schedule the workflow according to task requirements, dynamically optimize the extraction process, and effectively control the processing time while ensuring high accuracy.

[0004] In the first aspect, the present application is implemented through the following technical solutions:

[0005] A method for automatically extracting image file information based on an agent workflow, the method comprising:

[0006] Inputting the image file to be processed into the visual model to identify the text information of the image file;

[0007] Input the identified text information into the AI ​​Agent to extract field information, and dynamically optimize the information extraction process through AgenticWorkflow;

[0008] The field information extracted by each AI Agent is integrated to generate structured output data.

[0009] In some embodiments, the method further comprises:

[0010] The image file to be processed is obtained from various cloud storage platforms or local file systems through a RESTful API interface, supporting multiple file formats.

[0011] In some implementations, the step of inputting the image file to be processed into the visual model to identify the text information of the image file specifically includes:

[0012] The image file to be processed is input into the Paddle OCR visual model for text recognition; during the recognition process, the text is marked according to the position coordinates to generate complete structured text data.

[0013] In some implementations, inputting the identified text information to the AI ​​Agent to extract field information specifically includes:

[0014] The identified text information is input into the AI ​​Agent to extract field information. Different agents in the AI ​​Agent are responsible for extracting specific fields.

[0015] In some embodiments, the process of dynamically optimizing information extraction through Agentic Workflow specifically includes:

[0016] Through the execution results of historical tasks, the extraction time, extraction precision and recall rate of each Agent are collected, and the extraction accuracy of each Agent is obtained according to the extraction precision and recall rate. The extraction accuracy is:

[0017]

[0018] Where P i is the extraction accuracy, R i is the recall rate, F 1i is the extraction accuracy;

[0019] According to the extraction accuracy of each Agent, the comprehensive extraction accuracy is:

[0020]

[0021] Where n is the number of agents;

[0022] According to the extraction time and comprehensive extraction accuracy, an objective function is constructed, and the objective function is:

[0023]

[0024] In the formula, (T total ) i represents the extraction time of the i-th iteration; (F 1total ) i represents the comprehensive extraction accuracy of the i-th iteration; ω1 is the time weight coefficient; ω2 is the accuracy weight coefficient; m is the number of iterations; where ω1+ω2=1;

[0025] The agent or agent combination that satisfies the minimum objective function is selected to perform the task.

[0026] In some embodiments, the method further comprises:

[0027] Through the RESTful API interface, business information related to the extracted fields in the business system and / or knowledge base is requested to enrich the background knowledge of each agent, increase the semantic understanding ability of the large language model, and improve the accuracy of agent extraction.

[0028] In the second aspect, the present application proposes an automatic image file information extraction system based on an agent workflow, the system comprising:

[0029] A text recognition module, wherein the text recognition module inputs the image file to be processed into the visual model to recognize the text information of the image file;

[0030] An information extraction module, which inputs the text information identified by the text recognition module to the AI ​​Agent to extract field information, and dynamically optimizes the information extraction process through the Agentic Workflow;

[0031] And, an information integration module, which integrates the field information extracted by each Agent in the AI ​​Agent and generates structured output data.

[0032] In some embodiments, the system further comprises:

[0033] The data acquisition module acquires the image file to be processed from various cloud storage platforms or local file systems through a RESTful API interface and supports multiple file formats.

[0034] In some embodiments, the information extraction module is configured to perform the following process:

[0035] Through the execution results of historical tasks, the extraction time, extraction precision and recall rate of each Agent are collected, and the extraction accuracy of each Agent is obtained according to the extraction precision and recall rate. The extraction accuracy is:

[0036]

[0037] Where P i is the extraction accuracy, R i is the recall rate, F 1i is the extraction accuracy;

[0038] According to the extraction accuracy of each Agent, the comprehensive extraction accuracy is:

[0039]

[0040] Where n is the number of agents;

[0041] According to the extraction time and comprehensive extraction accuracy, an objective function is constructed, and the objective function is:

[0042]

[0043] In the formula, (T total ) i represents the extraction time of the i-th iteration; (F 1total ) i represents the comprehensive extraction accuracy of the i-th iteration; ω1 is the time weight coefficient; ω2 is the accuracy weight coefficient; m is the number of iterations; where ω1+ω2=1;

[0044] The agent or agent combination that satisfies the minimum objective function is selected to perform the task.

[0045] In a third aspect, the present application proposes a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.

[0046] This application proposes an agent-based workflow-based image file information automatic extraction method and system. The method and system aim to achieve efficient information extraction of multi-domain and multi-format files by combining large and small models, using the high-precision text recognition capability of the Paddle OCR visual model for image files, and combining the knowledge generalization capability of the AI ​​Agent's large language model, thereby reducing the need for manual annotation of large amounts of data, and significantly improving the accuracy and scope of extraction. It can be widely used in image file information extraction in scenarios such as finance, medical care, and legal industries.

[0047] This application proposes an automated image file information extraction method and system based on intelligent agent workflow, AI Agent intelligent scheduling and dynamic typesetting algorithm, which optimizes the file information extraction process in real time. Among them, AI Agent is responsible for dynamically adjusting the workflow to adapt to changes in task complexity and data volume, while the dynamic typesetting algorithm adjusts the text structure to improve the efficiency and accuracy of information extraction; at the same time, it can request the business system and / or knowledge base in real time to obtain background information of fields related to information extraction, provide contextual support for information extraction, and further improve the accuracy of image file extraction;

[0048] The present application proposes an automatic image document information extraction method and system based on intelligent workflow, which uses a dynamic optimization algorithm to balance processing speed and accuracy. In the process of large-scale document processing, the extraction strategy is adjusted in real time to optimize the processing speed while ensuring high accuracy, and adapt to document information extraction tasks of different types and formats;

[0049] The present application proposes a method and system for automatic extraction of image file information based on intelligent workflow. Compared with traditional methods, it shows a higher accuracy rate (up to 99%) when processing long texts and complex documents, and the extraction speed is increased by 30%. The present invention adopts a modular architecture design, so that the coupling degree between the components is low, which is easy to expand and maintain. It is worth noting that the information extraction module can be integrated into the third-party business system through the API interface without making any changes to the latter, which simplifies the deployment process and promotes the widespread application of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] The drawings described herein are used to provide a further understanding of the embodiments of the present application, constitute a part of the present application, and do not constitute a limitation on the embodiments of the present application. In the drawings:

[0051] Figure 1 A schematic diagram of the method flow proposed in the embodiment of the present application;

[0052] Figure 2 This is a block diagram of the system principles proposed in the embodiments of the present application. DETAILED DESCRIPTION

[0053] In order to make the objectives, technical solutions and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with examples and drawings. The illustrative implementation scheme of the present application and its description are only used to explain the present application and are not intended to limit the present application.

[0054] Example:

[0055] The existing image file information extraction methods have the following shortcomings: (1) they are highly dependent on a large amount of annotated data, and the annotation process is time-consuming and laborious. The system has poor generalization ability and is difficult to adapt to different application scenarios and files of different formats; (2) there is a lack of dynamic optimization mechanism, which makes it difficult for the system to simultaneously meet the requirements of accuracy and timeliness when facing data of different scales and complexities; (3) traditional methods have limitations on document length, and usually can only extract paragraph-level information, and it is difficult to handle more complex text structures; (4) large visual models have hallucination problems for scenarios that require the extraction of a large amount of field information, making it difficult to ensure accuracy and cannot be directly used in business scenarios. In response to the above problems, this embodiment proposes an automatic image file information extraction method based on an agent workflow. This embodiment combines a visual model with a large-scale pre-trained model, and through the intelligent scheduling workflow of an agent, dynamically optimizes the extraction process to achieve automatic information extraction of image files. Among them, the core of the agent workflow is composed of multiple agents. These agents can autonomously perceive, reason and execute tasks through the collaboration of natural language processing and large-scale pre-trained models, thereby realizing the automation and intelligence of the information extraction process.

[0056] like Figure 1 As shown, the extraction method proposed in this embodiment specifically includes the following steps:

[0057] Step 1: Input the image file to be processed into the visual model to identify the text information of the image file.

[0058] Furthermore, in step 1, the image file to be processed can be obtained from the local file system or cloud storage service, for example, the image file to be processed can be obtained from various cloud storage platforms (such as AWS S3, Alibaba Cloud OSS, etc.) or the local file system through the RESTful API interface, and the supported file formats include but are not limited to PDF, scanned images, Word documents, etc. The specific input is the file URL or storage path Filepath, and the output is the binary data stream Filebuf of the image file.

[0059] The acquired image file is input into the visual model (such as Paddle OCR) for text recognition. During the model recognition process, the text is marked according to the position coordinates to generate complete structured text data. Specifically, the acquired image file data stream Filebuf is input into the Paddle OCR visual model to identify the text content corresponding to the coordinates. During this process, the text content is identified according to the position coordinates, and all the contents are spliced ​​in a loop to generate complete text information file_content.

[0060] Step 2: Input the recognized text information into the AI ​​Agent to extract field information, and dynamically optimize the information extraction process through AgenticWorkflow.

[0061] Furthermore, in step 2, according to the fields to be identified, AI Agent built based on a large language model and prompt words is used for extraction. AI Agent can optimize the workflow based on real-time feedback, gradually learn how to better process file information of different complexity, and improve the processing efficiency of the overall system. When facing large-scale information extraction, AI Agent can simultaneously dispatch multiple information extraction agent nodes for parallel processing, automatically adjust the extraction strategy according to the current task and the complexity of the extracted field, maximize the use of computing resources, speed up document processing, and ensure that the system can still maintain a high processing speed and accuracy in large-scale data processing. Specifically, the generated file_content is input into the AI ​​Agent to extract field information. Different Agents in the AI ​​Agent are responsible for extracting specific fields, that is, each Agent is responsible for performing a task, and multiple Agents are strung together to complete the entire extraction task, for example: Agent responsible for extracting basic information such as name, gender, ID number, address, etc.; Agent responsible for extracting financial related information such as bank account number, transaction amount, transaction date, balance, etc. Traditional methods rely on rules and annotated data for extraction, which takes a lot of time and has limited application scenarios. In this embodiment, Agent is used for information extraction. The large language model is pre-trained with massive data and has excellent semantic understanding and context learning capabilities. It can efficiently extract key information without annotated data. At the same time, the text processing window of the large language model can support long text input, which greatly improves processing efficiency and information coverage.

[0062] This embodiment introduces a large language model and AI Agent to achieve the optimal effect of the AI ​​Agent dynamic workflow typesetting algorithm. The specific implementation process is as follows:

[0063] First, through the execution results of historical tasks, collect each information extraction agent A i The extraction time T i And the extraction accuracy P i , recall rate R i And the extraction accuracy F calculated based on precision and recall 1i . Specifically expressed as:

[0064]

[0065] Among them, A i represents the i-th information extraction agent (Agent), TP means that the system correctly extracts the correct information, that is, the information extraction agent Ai Identify and extract the information that should be extracted from the image file. TN means that the system correctly identifies the information that does not need to be extracted, that is, the information extraction agent A i The information that does not need to be extracted from the image file is not extracted. FP means that the system mistakenly extracts the information that does not need to be extracted from the image file, that is, the information extraction agent A i FN refers to the system missing the information that should be extracted, that is, the information extraction agent A i No information that needs to be extracted is extracted. In addition, P i It measures the proportion of information extracted by the system that is correct. The higher the precision, the fewer cases of incorrect extraction. The recall rate R i It is used to measure the ratio of the number of correct information extracted to the total number of documents. The higher the recall rate, the fewer missed cases. In order to comprehensively evaluate the performance of information extraction, the harmonic mean F of precision and recall is used. 1i , combines the advantages of these two indicators and is used to measure the performance of the system in balancing precision and recall. 1total It represents the harmonic mean of the total precision and recall of all information extracted from the image file, that is, each intelligent information extraction agent A i The sum of the harmonic mean of precision and recall. total It represents the total time to extract all the information from the image file, that is, each information extraction agent A i The extraction time is comprehensive, and n represents the number of information extraction agents.

[0066] In this embodiment, the Agentic Workflow dynamically selects the best agent or combination of agents to optimize the extraction time and extraction accuracy (F1 value). The specific optimization goal can be expressed as a dual-objective optimization problem. For tasks with high timeliness requirements, the extraction time is set as the main optimization goal, otherwise the accuracy (F1 value) is set as the main optimization goal:

[0067]

[0068] In the formula, (T total ) i represents the extraction time of the i-th iteration; (F 1total ) irepresents the comprehensive extraction accuracy of the i-th iteration; ω1 is the time weight coefficient, ranging from [0,1], which is used to reflect the timeliness of the task; ω2 is the accuracy weight coefficient, ranging from [0,1], which is used to reflect the comprehensive extraction accuracy of the task. Among them, ω1+ω2=1, by adjusting the values ​​of ω1 and ω2 to adapt to the needs of different tasks, the system can achieve a balance and optimization between efficiency and accuracy.

[0069] This embodiment dynamically selects the best agent or multiple agents to perform the task based on the model performance and task priority. The selection is based on the extraction time T of each agent. i And the extraction accuracy F 1i The process of agent selection is: select the agent or combination of agents that satisfies the minimum objective function E. Assuming that the image file information extraction consists of n agents, it can be expressed as: A all =(A1,A2,…A n ).

[0070] Through the above optimization model, the system can dynamically select the best agent or combination of agents to perform tasks according to the timeliness and accuracy requirements of the task, achieving the best balance between efficiency and accuracy. The continuous performance feedback mechanism ensures the self-optimization and adaptability of the model to meet changing task requirements.

[0071] Furthermore, step 2 can also request business information related to the extracted fields in the business system and / or knowledge base through the RESTful API interface, enrich the background knowledge of each agent node, increase the semantic understanding ability of the large language model, and improve the accuracy of agent extraction. For example, for the bank card information extraction agent, you can request the knowledge base to obtain the bank card knowledge of XX bank; for the case role judgment agent, you can request the business system to obtain role information.

[0072] Step 3: Fusion the field information extracted by each agent in the AI ​​Agent to generate structured output data.

[0073] Furthermore, step 3 integrates the results extracted by multiple agents and converts them into structured data, such as data in JSON or XML format, and the final result will be handed over to the business system for subsequent processing.

[0074] The method proposed in this embodiment uses the high-precision text recognition capability of the Paddle OCR visual model for image files, and combines the text understanding and knowledge generalization capabilities of the AI ​​Agent to achieve efficient information extraction of multi-field and multi-format files using only prompt words, avoiding the need for a large amount of labeled data. The large language model can process image files of various types and lengths through pre-training and transfer learning, thereby improving the accuracy and robustness of information extraction. Compared with traditional NLP methods, information extraction using a large language model is pre-processed through prompt words, without the need for complex steps such as word segmentation, NER, and denoising. The powerful contextual understanding and zero-sample learning capabilities of the large language model enable it to efficiently extract information directly from the original text. This method not only simplifies the preprocessing process, but also can handle complex text structures and semantics, improving the accuracy and flexibility of information extraction. In addition, the method proposed in this embodiment constructs an automated workflow for file information extraction through AgenticWorkflow technology, dynamically optimizes the information extraction process, achieves the best balance between efficiency and accuracy, and meets the needs of different tasks. The continuous performance feedback mechanism further enhances the self-optimization and adaptability of the system, and provides a flexible and automated solution for the field of image file information extraction. Whether in the financial, medical, or legal industries, this technology can greatly improve information processing efficiency and reduce labor costs, and has broad application prospects. In addition, compared with traditional methods, the method proposed in this embodiment shows a higher accuracy rate (up to 99%) when processing long texts and complex documents, and the extraction speed is increased by 30%.

[0075] Based on the same technical concept as above, this embodiment also proposes an automatic image file information extraction system based on intelligent agent workflow, such as Figure 2 As shown, the system proposed in this embodiment includes:

[0076] The text recognition module inputs the image file to be processed into the visual model to recognize the text information of the image file.

[0077] The information extraction module inputs the text information recognized by the text recognition module to the AIAgent to extract field information, and dynamically optimizes the information extraction process through the Agentic Workflow. It should be noted that the dynamic optimization information extraction process is specifically as described in step 2 above, and will not be repeated here.

[0078] And, the information integration module integrates the field information extracted by each agent in the AI ​​Agent and generates structured output data.

[0079] Furthermore, the system proposed in this embodiment also includes:

[0080] The data acquisition module obtains the image files to be processed from various cloud storage platforms or local file systems through the RESTful API interface. The supported file formats include but are not limited to PDF, scanned images, Word documents, etc. The specific input is the file URL or storage path Filepath, and the output is the binary data stream Filebuf of the image file.

[0081] Furthermore, the information extraction module of this embodiment includes multiple agents, each of which is responsible for extracting specific fields, such as an agent responsible for extracting personal information, an agent responsible for extracting contact information, an agent responsible for extracting bank card information, an agent responsible for document type identification, an agent responsible for extracting corporate information, etc.

[0082] Furthermore, the information extraction module of this embodiment can also request business information related to the extracted fields in the business system and / or knowledge base through the RESTful API interface, enrich the background knowledge of each intelligent agent node, increase the semantic understanding ability of the large language model, and improve the accuracy of intelligent agent extraction.

[0083] The system of this embodiment adopts a modular design to make the system architecture clear and concise, with low coupling between modules, which is easy to expand and maintain. At the same time, the information extraction module can be provided to the business system for calling in the form of an API interface without the need to modify the business system.

[0084] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0085] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1A device that provides the functions specified in a block or multiple blocks.

[0086] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0087] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0088] The specific implementation methods described above further illustrate the purpose, technical solutions and beneficial effects of the present application in detail. It should be understood that the above description is only the specific implementation method of the present application and is not intended to limit the scope of protection of the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the scope of protection of the present application.

Claims

1. A method for automatically extracting image file information based on intelligent workflow, characterized in that: The method comprises: Inputting the image file to be processed into the visual model to identify the text information of the image file; Input the identified text information into the AI ​​Agent to extract field information, and dynamically optimize the information extraction process through the Agentic Workflow; The field information extracted by each agent in the AI ​​Agent is integrated to generate structured output data.

2. The method for automatically extracting image file information based on agent workflow according to claim 1, characterized in that: The method further comprises: The image file to be processed is obtained from various cloud storage platforms or local file systems through a RESTful API interface, supporting multiple file formats.

3. The method for automatically extracting image file information based on agent workflow according to claim 1, characterized in that: The step of inputting the image file to be processed into the visual model to identify the text information of the image file specifically includes: The image file to be processed is input into the Paddle OCR visual model for text recognition; during the recognition process, the text is marked according to the position coordinates to generate complete structured text data.

4. A method for automatically extracting image file information based on agent workflow according to any one of claims 1 to 3, characterized in that: The step of inputting the identified text information to the AI ​​Agent to extract field information specifically includes: The identified text information is input into the AIAgent to extract field information. Different agents in the AIAgent are responsible for extracting specific fields.

5. The method for automatically extracting image file information based on agent workflow according to claim 4, characterized in that: The process of dynamically optimizing information extraction through Agentic Workflow specifically includes: Through the execution results of historical tasks, the extraction time, extraction precision and recall rate of each Agent are collected, and the extraction accuracy of each Agent is obtained according to the extraction precision and recall rate. The extraction accuracy is: Where P i is the extraction accuracy, R i is the recall rate, F 1i is the extraction accuracy; According to the extraction accuracy of each Agent, the comprehensive extraction accuracy is: Where n is the number of agents; According to the extraction time and comprehensive extraction accuracy, an objective function is constructed, and the objective function is: In the formula, (T total ) i represents the extraction time of the i-th iteration; (F 1total ) i represents the comprehensive extraction accuracy of the i-th iteration; ω1 is the time weight coefficient; ω2 is the accuracy weight coefficient; m is the number of iterations; where ω1+ω2=1; The agent or agent combination that satisfies the minimum objective function is selected to perform the task.

6. The method for automatically extracting image file information based on agent workflow according to claim 5, characterized in that: The method further comprises: Through the RESTful API interface, business information related to the extracted fields in the business system and / or knowledge base is requested to enrich the background knowledge of each agent, increase the semantic understanding ability of the large language model, and improve the accuracy of agent extraction.

7. An automatic image file information extraction system based on intelligent workflow, characterized in that: The system comprises: A text recognition module, wherein the text recognition module inputs the image file to be processed into the visual model to recognize the text information of the image file; An information extraction module, which inputs the text information identified by the text recognition module to the AI ​​Agent to extract field information, and dynamically optimizes the information extraction process through the Agentic Workflow; And, an information integration module, which integrates the field information extracted by each Agent in the AI ​​Agent and generates structured output data.

8. The system for automatically extracting image file information based on agent workflow according to claim 7, characterized in that: The system further comprises: The data acquisition module acquires the image file to be processed from various cloud storage platforms or local file systems through a RESTful API interface and supports multiple file formats.

9. The system for automatically extracting image file information based on agent workflow according to claim 7, characterized in that: The information extraction module is configured to perform the following process: Through the execution results of historical tasks, the extraction time, extraction precision and recall rate of each Agent are collected, and the extraction accuracy of each Agent is obtained according to the extraction precision and recall rate. The extraction accuracy is: Where P i is the extraction accuracy, R i is the recall rate, F 1i is the extraction accuracy; According to the extraction accuracy of each Agent, the comprehensive extraction accuracy is: Where n is the number of agents; According to the extraction time and comprehensive extraction accuracy, an objective function is constructed, and the objective function is: In the formula, (T total ) i represents the extraction time of the i-th iteration; (F 1total ) i represents the comprehensive extraction accuracy of the i-th iteration; ω1 is the time weight coefficient; ω2 is the accuracy weight coefficient; m is the number of iterations; where ω1+ω2=1; The AIAgent or combination of AIAgents that satisfies the minimum objective function is selected to perform the task.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

Images