Home guest PON checking agent method and system based on large model and target detection

Abstract

The invention relates to the technical field of transmission network resource management, in particular to a home guest PON checking agent method and system based on a large model and target detection, and the method comprises the steps: constructing the core capability of a checking agent, and building a checking agent hardware and software architecture; the method has the advantages that automatic analysis of user intentions is achieved through large-model semantic analysis, the optical splitter is rapidly positioned by combining ElasticSearch and spatial position calculation, the port state is automatically recognized by means of the target detection technology, manual check item by item is not needed in the whole process, the checking period is remarkably shortened, and the method is particularly suitable for large-scale PON network checking scenes.

Description

Technical Field

[0001] This invention relates to the field of transmission network resource management technology, specifically to a method and system for intelligent agent investigation of home PON based on large models and target detection. Background Technology

[0002] In the operation and maintenance of communication networks, the inventory of PON network resources is a key step in ensuring accurate management of network resources and improving operation and maintenance efficiency.

[0003] Currently, PON (Positioning and Networking) inspections largely rely on manual labor, which has the following significant drawbacks: First, the inspection process requires manual verification of information entered into the integrated resource management system and device searches, making the operation cumbersome and inefficient, especially during large-scale inspections in various cities, resulting in lengthy inspection cycles. Second, the identification of splitter SN codes and the determination of port occupancy status depend on manual visual verification, which is easily affected by environmental interference and human negligence, leading to a high data error rate. Third, manual inspections require operators to possess professional knowledge, and different personnel may have different operating standards, making it difficult to guarantee the consistency and accuracy of the inspection results. Fourth, manual inspections cannot automatically plan the work sequence, easily leading to duplicate or missed inspections, further reducing the quality and efficiency of the operation. With breakthroughs in large-scale model technology in the field of semantic understanding and the maturity of target detection technology in the field of image recognition, the automation and intelligentization of PON inspections have become possible.

[0004] Therefore, an intelligent device that integrates the above technologies is needed to solve the problems of low efficiency, large error and poor consistency in the existing manual inspection mode. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for intelligent agent investigation of home PON based on large model and target detection, so as to solve the defects of low efficiency, insufficient accuracy and cumbersome operation caused by the reliance on manual labor in home PON investigation.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a home customer PON clearance agent method based on large model and target detection, comprising the following steps: constructing the core capabilities of the clearance agent: Semantic analysis capability training: A dedicated prompt word system is designed for PON clearance scenarios, covering the input parameters and output results of different clearance scenarios. The prompt words are used as training data to input into the large model to complete the model fine-tuning training, so that the clearance agent has the semantic analysis function to accurately interpret the user's clearance intention and related input information. Image target detection capability training: Collect on-site photos of beam splitters of different models and under different environments, construct an image dataset containing annotation information, and use this dataset to train a large model, enabling the inspection agent to automatically identify the image target detection function of beam splitter port occupancy status from uploaded photos.

[0007] Preferably, this also includes building the hardware and software architecture for the investigation intelligent agent: Hardware modules include a processor, memory, voice acquisition module, text input module, image acquisition module, positioning module, QR code scanning module, and communication module. The processor is used to run the large model and the investigation logic program, the memory is used to store training data, model parameters, and investigation data, and the communication module is used to interact with the integrated resource management system. The software modules include an intent parsing module, a data interaction module, an image recognition module, a survey planning module, and a result output module. The intent parsing module processes user input information based on a trained semantic analysis model to determine the survey intent. The data interaction module interfaces with the integrated resource management system through the communication module to realize data query and upload. The image recognition module calls the target detection model to process the spectrometer photos and outputs the port status. The survey planning module compares user input with data from the asset management system and generates the survey order. The result output module provides feedback on the survey results in a visual or structured format.

[0008] Preferably, the PON port clearance process based on the clearance agent includes the following steps: Intent triggering: The user inputs the inspection command via voice or text. The intent parsing module of the inspection agent performs semantic analysis on the command. If the intent is determined to be "PON port inspection", then the process is initiated. Splitter Information Input: The intelligent agent prompts the user to provide some splitter-related information through one or more methods such as voice analysis, text input, splitter latitude and longitude input, and QR code scanning; Splitter Search: The intent parsing module determines the current sub-intent as "splitter search" based on the user input information. The data interaction module calls the ElasticSearch search engine to perform fuzzy search and combines the latitude and longitude data obtained by the positioning module to perform spatial location calculation. It then matches and obtains detailed information of the target splitter from the integrated resource management system and feeds it back to the user for selection and confirmation. SN Code Recognition and PON Port Verification: After the user selects the target optical splitter, they provide the SN code information by entering the SN code text or scanning the device's SN code QR code. The intent parsing module determines the sub-intent as "SN code recognition" and calls the semantic analysis model to parse the accurate SN code data. The data interaction module transmits the SN code to the PON gateway alarm interface to obtain the PON port connection status and alarm information corresponding to the optical splitter, completes the PON port check, and outputs the results.

[0009] Preferably, the splitter port inspection process based on the inspection agent includes the following steps: Intent Trigger: The intent parsing module analyzes the user's command. If the intent is determined to be "splitter port cleanup", then this process is initiated. Splitter Photo Upload: The intelligent agent prompts the user to take a photo of the on-site splitter and upload it through the image acquisition module, or directly calls the camera to acquire photos in real time; Port status recognition: The image recognition module calls the trained target detection model to process the uploaded photos, automatically identify the occupancy status of each port of the splitter, and generate structured port status data. Inventory sequence planning: The data interaction module obtains the port status data maintained by the splitter from the integrated resource management system. The inventory planning module compares the port status identified by the intelligent agent with the data in the asset management system, marks the ports with inconsistent or unclear status, and automatically plans the optimal inventory sequence. Port check execution: When the user enters the command "Start splitter port check", after the intent parsing module confirms the sub-intent, the agent guides the user to complete the detailed check of each port in the planned order. First, it calls the alarm interface to query the fiber plugging status, then detects the user's fiber unplugging behavior, and obtains the port check results when the fiber is successfully unplugged. Finally, the check results are synchronized to the integrated resource management system through the data interaction module.

[0010] Preferably, the input parameters for different inspection scenarios include text commands, voice information, latitude and longitude data, and QR code information; the output results include a list of device information, SN code text, and port status data; the annotation information of the image dataset includes the appearance features of the beam splitter, port location, and port occupancy status (idle / occupied).

[0011] A home-based PON (Positioning and Activation Network) clearance agent system based on large models and object detection includes: a core capability building module for the clearance agent, used to endow the clearance agent with core capabilities, specifically including: The semantic analysis capability training unit designs a dedicated prompt word system for PON inspection scenarios. This system covers input parameters for different inspection scenarios, such as splitter search, SN code recognition, PON port inspection, splitter port inspection, and start of splitter port inspection. Input parameters include text commands, voice information, latitude and longitude data, and QR code information. Output results include a device information list, SN code text, and port status data. The above prompt words are used as training data to input into the large model to complete the model fine-tuning training, enabling the inspection agent to have the semantic analysis function of accurately interpreting the user's inspection intention and related input information. The image target detection capability training unit collects on-site photos of beam splitters of different models and under different environments, and constructs an image dataset containing labeled information such as beam splitter appearance features, port positions, and port occupancy status (idle / occupied). The large model is trained using this dataset, enabling the inspection agent to automatically identify the image target detection function of beam splitter port occupancy status from uploaded photos.

[0012] Preferably, it also includes: a hardware module for the intelligent agent of the investigation, comprising a processor, a memory, a voice acquisition module, a text input module, an image acquisition module, a positioning module, a QR code scanning module, and a communication module; wherein, the processor is used to run the large model and the investigation logic program, the memory is used to store training data, model parameters, and investigation data, and the communication module is used to interact with the integrated resource management system.

[0013] Preferably, it also includes: a smart agent software module for the investigation, including an intent parsing module, a data interaction module, an image recognition module, an investigation planning module, and a result output module; The intent parsing module processes user input information and determines the investigation intent based on the trained semantic analysis model; The data interaction module interfaces with the integrated resource management system through the communication module to enable data querying and uploading; The image recognition module calls the target detection model to process the beam splitter image and outputs the port status; The inventory planning module compares user input with data from the asset management system and generates an inventory sequence. The results output module provides feedback on the investigation results in a visual or structured format.

[0014] Preferably, the system executes the PON port clearance process, which specifically includes the following steps: Intent triggering: The user inputs the inspection command via voice or text. The intent parsing module of the inspection agent performs semantic analysis on the command. If the intent is determined to be "PON port inspection", then the process is initiated. Splitter Information Input: The intelligent agent prompts the user to provide some splitter-related information through one or more methods, such as voice analysis, text input, splitter latitude and longitude input, and QR code scanning. The relevant information includes device name keywords, approximate location, and QR code identification. Splitter Search: The intent parsing module determines the current sub-intent as "splitter search" based on the user input information. The data interaction module calls the ElasticSearch search engine to perform fuzzy search and combines the latitude and longitude data obtained by the positioning module to perform spatial location calculation. It then matches and obtains detailed information of the target splitter from the integrated resource management system and feeds it back to the user for selection and confirmation. SN Code Recognition and PON Port Verification: After the user selects the target optical splitter, they provide the SN code information by entering the SN code text or scanning the device's SN code QR code. The intent parsing module determines the sub-intent as "SN code recognition" and calls the semantic analysis model to parse the accurate SN code data. The data interaction module transmits the SN code to the PON gateway alarm interface to obtain the PON port connection status and alarm information corresponding to the optical splitter, completes the PON port check, and outputs the results.

[0015] Preferably, the system executes a splitter port clearance procedure, which specifically includes the following steps: Intent Trigger: The intent parsing module analyzes the user's command. If the intent is determined to be "splitter port cleanup", then this process is initiated. Splitter Photo Upload: The intelligent agent prompts the user to take a photo of the on-site splitter and upload it through the image acquisition module, or directly calls the camera to acquire photos in real time; Port status identification: The image recognition module calls the trained target detection model to process the uploaded photos, automatically identify the occupancy status (idle / occupied) of each port of the splitter, and generate structured port status data; Inspection sequence planning: The data interaction module obtains the port status data maintained by the splitter from the integrated resource management system. The inspection planning module compares the port status identified by the intelligent agent with the data in the asset management system, marks the ports with inconsistent or unclear status, and automatically plans the optimal inspection sequence, prioritizing the inspection of ports with inconsistent status. Port check execution: When the user enters the command "Start splitter port check", after the intent parsing module confirms the sub-intent, the agent guides the user to complete the detailed check of each port in the planned order. First, it calls the alarm interface to query the fiber plugging status, then detects the user's fiber unplugging behavior, and obtains the port check results when the fiber is successfully unplugged. Finally, the check results are synchronized to the integrated resource management system through the data interaction module.

[0016] Compared with the prior art, the beneficial effects of the present invention are: The present invention proposes a smart agent method and system for home PON inspection based on large model and target detection. It realizes automatic parsing of user intent through semantic analysis of large model, quickly locates the splitter by combining ElasticSearch and spatial location calculation, and automatically identifies the port status with the help of target detection technology. The entire process does not require manual item-by-item verification, which significantly shortens the inspection cycle and is especially suitable for large-scale PON network inspection scenarios.

[0017] Improving the accuracy of inventory data: This invention uses SN code recognition with dual protection through semantic analysis and QR code parsing. Port status recognition is based on a trained target detection model, reducing the error of manual visual judgment. At the same time, by comparing with data from the asset management system, the inventory sequence is planned to avoid omissions or duplicates, thus improving data consistency.

[0018] Lowering the operational threshold: Users can trigger the investigation process with simple operations such as voice, text, and photo. The intelligent agent automatically completes complex data analysis and logical processing, without requiring operators to have in-depth professional knowledge, thus reducing the skill requirements for maintenance personnel.

[0019] Standardize the inventory process: By using a preset prompt system and a standardized inventory process, ensure that the operating standards of different operators are consistent and the inventory results are in a uniform format, which facilitates subsequent resource management and data statistical analysis.

[0020] It has good scalability: the designed prompt word system and model training method can be flexibly adjusted according to new investigation scenarios (such as equipment failure investigation) without large-scale modification of the device hardware, and adapt to the upgrade needs of different models of optical splitters. Attached Figure Description

[0021] Figure 1 This is a system architecture diagram of the present invention; Figure 2 This is a schematic diagram of the PON port inspection process of the present invention; Figure 3 This is a flowchart of the process for checking the port of the optical splitter in this invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the present invention clear and complete, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only some, not all, embodiments of the present invention, and are merely illustrative of the embodiments of the present invention. They are not intended to limit 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.

[0023] Example 1: This invention provides a technical solution: a home customer PON inventory intelligent agent method based on large model and target detection. By designing an inventory intelligent agent device with semantic analysis and image target detection capabilities, and constructing a standardized inventory process, intelligent inventory of home customer PON resources is achieved. The overall architecture is shown in the attached figure. Figure 1 As shown, the specific technical solution is as follows: 1. Investigate and build the core capabilities of intelligent agents. Semantic analysis capability training: A dedicated prompting word system was designed for PON inspection scenarios, including input parameters (such as text commands, voice information, latitude and longitude data, QR code information, etc.) and output results (such as device information list, SN code text, port status data, etc.) for different inspection scenarios such as splitter search, SN code recognition, PON port inspection, splitter port inspection, and start of splitter port inspection. These prompting words were used as training data to input into a large model, completing model fine-tuning training, enabling the inspection agent to possess the semantic analysis function to accurately interpret the user's inspection intent and related input information.

[0024] Image target detection capability training: Collect on-site photographs of beam splitters of different models and under different environments to construct an image dataset containing labeled information such as beam splitter appearance features, port locations, and port occupancy status (idle / occupied). Use this dataset to train a large model, enabling the inspection agent to automatically identify image target detection capabilities based on uploaded photos, recognizing the port occupancy status of beam splitters.

[0025] 2. Investigate the hardware and software architecture of the intelligent agent. The hardware module includes a processor, memory, voice acquisition module, text input module, image acquisition module, positioning module, QR code scanning module, and communication module. The processor runs the large model and the verification logic program; the memory stores training data, model parameters, and verification data; and the communication module interacts with the integrated resource management system.

[0026] The software modules include an intent parsing module, a data interaction module, an image recognition module, a survey planning module, and a results output module. The intent parsing module processes user input based on a trained semantic analysis model to determine the survey intent; the data interaction module interfaces with the integrated resource management system via a communication module to enable data querying and uploading; the image recognition module uses a target detection model to process spectrometer images and outputs port status; the survey planning module compares user input with data from the asset management system and generates a survey order; and the results output module provides feedback on the survey results in a visual or structured format.

[0027] 3. PON clearance process based on clearance agent PON port inspection process, as attached Figure 2The diagram shown is a flowchart of the PON port inspection process, which includes the following steps: a) Intent Trigger: The user inputs the inspection command via voice, text, etc. The intent parsing module of the inspection agent performs semantic analysis on the command. If the intent is determined to be "PON port inspection", then this process is entered.

[0028] b) Splitter Information Input: The intelligent agent prompts the user to provide some splitter-related information (such as device name keywords, approximate location, QR code identification, etc.) through one or more methods such as voice analysis, text input, splitter latitude and longitude input, and QR code scanning.

[0029] c) Splitter Search: The intent parsing module determines the current sub-intent as "splitter search" based on the user input information. The data interaction module calls the ElasticSearch search engine to perform fuzzy search and combines the latitude and longitude data obtained by the positioning module to perform spatial location calculation. It then matches and obtains detailed information of the target splitter from the integrated resource management system and feeds it back to the user for selection and confirmation.

[0030] d) SN Code Recognition and PON Port Verification: After selecting the target optical splitter, the user provides the SN code information by inputting the SN code text or scanning the device's SN code QR code. The intent parsing module determines the sub-intent as "SN code recognition" and calls the semantic analysis model to parse the accurate SN code data. The data interaction module transmits the SN code to the PON gateway alarm interface, obtains the PON port connection status and alarm information corresponding to the optical splitter, completes the PON port check, and outputs the results.

[0031] Splitter port troubleshooting procedure, as attached Figure 3 The diagram shown illustrates the process for checking the ports of a splitter, which includes the following steps: a) Intent Trigger: The intent parsing module analyzes the user's command. If the intent is determined to be "splitter port cleanup", then this process is entered.

[0032] b) Upload of beam splitter photos: The intelligent agent prompts the user to take photos of the beam splitter on site and upload them through the image acquisition module, or directly calls the camera to acquire photos in real time.

[0033] c) Port status recognition: The image recognition module calls the trained target detection model to process the uploaded photos, automatically identify the occupancy status (idle / occupied) of each port of the splitter, and generate structured port status data.

[0034] d) Inventory sequence planning: The data interaction module obtains the port status data maintained by the splitter from the integrated resource management system. The inventory planning module compares the port status identified by the intelligent agent with the data in the asset management system, marks the ports with inconsistent or unclear status, and automatically plans the optimal inventory sequence (such as prioritizing the inventory of ports with inconsistent status).

[0035] e) Port check execution: When the user inputs the command "Start splitter port check", after the intent parsing module confirms the sub-intent, the agent guides the user to complete the detailed check of each port in the planned order (first call the alarm interface to query the fiber insertion status, then detect the user's fiber unplugging behavior, obtain the port check results when the fiber is successfully unplugged, and finally synchronize the check results to the integrated resource management system through the data interaction module).

[0036] Example 2, based on Example 1, proposes a home customer PON clearance agent system based on a large model and target detection, including: a core capability building module for the clearance agent, used to endow the clearance agent with core capabilities, specifically including: The semantic analysis capability training unit designs a dedicated prompt word system for PON inspection scenarios. This system covers input parameters for different inspection scenarios, such as splitter search, SN code recognition, PON port inspection, splitter port inspection, and start of splitter port inspection. Input parameters include text commands, voice information, latitude and longitude data, and QR code information. Output results include a device information list, SN code text, and port status data. The above prompt words are used as training data to input into the large model to complete the model fine-tuning training, enabling the inspection agent to have the semantic analysis function of accurately interpreting the user's inspection intention and related input information. The image target detection capability training unit collects on-site photos of beam splitters of different models and under different environments, and constructs an image dataset containing labeled information such as beam splitter appearance features, port positions, and port occupancy status (idle / occupied). The large model is trained using this dataset, enabling the inspection agent to automatically identify the image target detection function of beam splitter port occupancy status from uploaded photos.

[0037] It also includes: a hardware module for the intelligent agent, comprising a processor, memory, voice acquisition module, text input module, image acquisition module, positioning module, QR code scanning module, and communication module; wherein, the processor is used to run the large model and the investigation logic program, the memory is used to store training data, model parameters, and investigation data, and the communication module is used to interact with the integrated resource management system.

[0038] It also includes: a software module for the investigation intelligent agent, comprising an intent parsing module, a data interaction module, an image recognition module, an investigation planning module, and a result output module; The intent parsing module processes user input information and determines the investigation intent based on the trained semantic analysis model; The data interaction module interfaces with the integrated resource management system through the communication module to enable data querying and uploading; The image recognition module calls the target detection model to process the beam splitter image and outputs the port status; The inventory planning module compares user input with data from the asset management system and generates an inventory sequence. The results output module provides feedback on the investigation results in a visual or structured format.

[0039] The system executes the PON port cleanup process, which includes the following steps: Intent triggering: The user inputs the inspection command via voice or text. The intent parsing module of the inspection agent performs semantic analysis on the command. If the intent is determined to be "PON port inspection", then the process is initiated. Splitter Information Input: The intelligent agent prompts the user to provide some splitter-related information through one or more methods, such as voice analysis, text input, splitter latitude and longitude input, and QR code scanning. The relevant information includes device name keywords, approximate location, and QR code identification. Splitter Search: The intent parsing module determines the current sub-intent as "splitter search" based on the user input information. The data interaction module calls the ElasticSearch search engine to perform fuzzy search and combines the latitude and longitude data obtained by the positioning module to perform spatial location calculation. It then matches and obtains detailed information of the target splitter from the integrated resource management system and feeds it back to the user for selection and confirmation. SN Code Recognition and PON Port Verification: After the user selects the target optical splitter, they provide the SN code information by entering the SN code text or scanning the device's SN code QR code. The intent parsing module determines the sub-intent as "SN code recognition" and calls the semantic analysis model to parse the accurate SN code data. The data interaction module transmits the SN code to the PON gateway alarm interface to obtain the PON port connection status and alarm information corresponding to the optical splitter, completes the PON port check, and outputs the results.

[0040] The system performs a splitter port check process, which includes the following steps: Intent Trigger: The intent parsing module analyzes the user's command. If the intent is determined to be "splitter port cleanup", then this process is initiated. Splitter Photo Upload: The intelligent agent prompts the user to take a photo of the on-site splitter and upload it through the image acquisition module, or directly calls the camera to acquire photos in real time; Port status identification: The image recognition module calls the trained target detection model to process the uploaded photos, automatically identify the occupancy status (idle / occupied) of each port of the splitter, and generate structured port status data; Inspection sequence planning: The data interaction module obtains the port status data maintained by the splitter from the integrated resource management system. The inspection planning module compares the port status identified by the intelligent agent with the data in the asset management system, marks the ports with inconsistent or unclear status, and automatically plans the optimal inspection sequence, prioritizing the inspection of ports with inconsistent status. Port check execution: When the user enters the command "Start splitter port check", after the intent parsing module confirms the sub-intent, the agent guides the user to complete the detailed check of each port in the planned order. First, it calls the alarm interface to query the fiber plugging status, then detects the user's fiber unplugging behavior, and obtains the port check results when the fiber is successfully unplugged. Finally, the check results are synchronized to the integrated resource management system through the data interaction module.

[0041] 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. A method for intelligent agents to investigate home customer PON based on large models and target detection, characterized in that: Includes the following steps: Building core capabilities for the intelligent agent of the investigation: Semantic analysis capability training: A dedicated prompt word system is designed for PON clearance scenarios, covering the input parameters and output results of different clearance scenarios. The prompt words are used as training data to input into the large model to complete the model fine-tuning training, so that the clearance agent has the semantic analysis function to accurately interpret the user's clearance intention and related input information. Image target detection capability training: Collect on-site photos of beam splitters of different models and under different environments, construct an image dataset containing annotation information, and use this dataset to train a large model, enabling the inspection agent to automatically identify the image target detection function of beam splitter port occupancy status from uploaded photos.

2. The method for intelligent agent-based PON clearing of home customers based on large model and target detection according to claim 1, characterized in that: This also includes building the hardware and software architecture for the investigation intelligent agent: Hardware modules include a processor, memory, voice acquisition module, text input module, image acquisition module, positioning module, QR code scanning module, and communication module. The processor is used to run the large model and the investigation logic program, the memory is used to store training data, model parameters, and investigation data, and the communication module is used to interact with the integrated resource management system. The software modules include an intent parsing module, a data interaction module, an image recognition module, a survey planning module, and a result output module. The intent parsing module processes user input information based on a trained semantic analysis model to determine the survey intent. The data interaction module interfaces with the integrated resource management system through the communication module to realize data query and upload. The image recognition module calls the target detection model to process the spectrometer photos and outputs the port status. The survey planning module compares user input with data from the asset management system and generates the survey order. The result output module provides feedback on the survey results in a visual or structured format.

3. The method for intelligent agent-based PON clearing of home customers based on large model and target detection according to claim 2, characterized in that: The PON port clearance process based on the clearance agent includes the following steps: Intent triggering: The user inputs the inspection command via voice or text. The intent parsing module of the inspection agent performs semantic analysis on the command. If the intent is determined to be "PON port inspection", the process is initiated. Splitter Information Input: The intelligent agent prompts the user to provide some splitter-related information through one or more methods such as voice analysis, text input, splitter latitude and longitude input, and QR code scanning; Splitter Search: The intent parsing module determines the current sub-intent as "splitter search" based on the user input information. The data interaction module calls the ElasticSearch search engine to perform fuzzy search and combines the latitude and longitude data obtained by the positioning module to perform spatial location calculation. It then matches and obtains detailed information of the target splitter from the integrated resource management system and feeds it back to the user for selection and confirmation. SN Code Recognition and PON Port Verification: After the user selects the target optical splitter, they provide the SN code information by entering the SN code text or scanning the device's SN code QR code. The intent parsing module determines the sub-intent as "SN code recognition" and calls the semantic analysis model to parse out the accurate SN code data. The data interaction module transmits the SN code to the PON gateway alarm interface to obtain the PON port connection status and alarm information corresponding to the optical splitter, completes the PON port check, and outputs the results.

4. The method for intelligent agent-based PON clearing of home customers based on large model and target detection according to claim 3, characterized in that: The splitter port clearance process based on the clearance agent includes the following steps: Intent Trigger: The intent parsing module analyzes the user's command. If the intent is determined to be "splitter port cleanup", then this process is initiated. Splitter Photo Upload: The intelligent agent prompts the user to take a photo of the on-site splitter and upload it through the image acquisition module, or directly calls the camera to acquire photos in real time; Port status recognition: The image recognition module calls the trained target detection model to process the uploaded photos, automatically identify the occupancy status of each port of the splitter, and generate structured port status data. Inventory sequence planning: The data interaction module obtains the port status data maintained by the splitter from the integrated resource management system. The inventory planning module compares the port status identified by the intelligent agent with the data in the asset management system, marks the ports with inconsistent or unclear status, and automatically plans the optimal inventory sequence. Port check execution: When the user enters the command "Start splitter port check", after the intent parsing module confirms the sub-intent, the agent guides the user to complete the detailed check of each port in the planned order. First, it calls the alarm interface to query the fiber plugging status, then detects the user's fiber unplugging behavior, obtains the port check results when the fiber is successfully unplugged, and finally synchronizes the check results to the integrated resource management system through the data interaction module.

5. The method for intelligent agent-based PON clearing of home customers based on large model and target detection according to claim 4, characterized in that: Input parameters for different inspection scenarios include text commands, voice information, latitude and longitude data, and QR code information; output results include a list of device information, SN code text, and port status data; the annotation information of the image dataset includes the appearance features of the beam splitter, port location, and port occupancy status (idle / occupied).

6. A home customer PON inspection intelligent agent system based on large model and target detection, applied to the method described in claim 5, characterized in that: include: The core capability building module for the investigation agent is used to endow the investigation agent with core capabilities, specifically including: The semantic analysis capability training unit designs a dedicated prompt word system for PON inspection scenarios. This system covers input parameters for different inspection scenarios, such as splitter search, SN code recognition, PON port inspection, splitter port inspection, and start of splitter port inspection. Input parameters include text commands, voice information, latitude and longitude data, and QR code information. Output results include a device information list, SN code text, and port status data. The above prompt words are used as training data to input into the large model to complete the model fine-tuning training, enabling the inspection agent to have the semantic analysis function of accurately interpreting the user's inspection intention and related input information. The image target detection capability training unit collects on-site photos of beam splitters of different models and under different environments, and constructs an image dataset containing labeled information such as beam splitter appearance features, port positions, and port occupancy status (idle / occupied). The large model is trained using this dataset, enabling the inspection agent to automatically identify the image target detection function of beam splitter port occupancy status from uploaded photos.

7. The intelligent agent system for home customer PON inspection based on large model and target detection according to claim 6, characterized in that: Also includes: The hardware module for the intelligent agent includes a processor, a memory, a voice acquisition module, a text input module, an image acquisition module, a positioning module, a QR code scanning module, and a communication module. The processor is used to run the large model and the investigation logic program, the memory is used to store training data, model parameters, and investigation data, and the communication module is used to interact with the integrated resource management system.

8. The intelligent agent system for home customer PON inspection based on large model and target detection according to claim 7, characterized in that: Also includes: The intelligent agent software module for the investigation includes an intent parsing module, a data interaction module, an image recognition module, an investigation planning module, and a result output module; The intent parsing module processes user input information and determines the investigation intent based on the trained semantic analysis model; The data interaction module interfaces with the integrated resource management system through the communication module to enable data querying and uploading; The image recognition module calls the target detection model to process the beam splitter image and outputs the port status; The inventory planning module compares user input with data from the asset management system and generates an inventory sequence. The results output module provides feedback on the investigation results in a visual or structured format.

9. A home customer PON clearing intelligent agent system based on large model and target detection according to claim 8, characterized in that: The system executes the PON port cleanup process, which includes the following steps: Intent triggering: The user inputs the inspection command via voice or text. The intent parsing module of the inspection agent performs semantic analysis on the command. If the intent is determined to be "PON port inspection", the process is initiated. Splitter Information Input: The intelligent agent prompts the user to provide some splitter-related information through one or more methods, such as voice analysis, text input, splitter latitude and longitude input, and QR code scanning. The relevant information includes device name keywords, approximate location, and QR code identification. Splitter Search: The intent parsing module determines the current sub-intent as "splitter search" based on the user input information. The data interaction module calls the ElasticSearch search engine to perform fuzzy search and combines the latitude and longitude data obtained by the positioning module to perform spatial location calculation. It then matches and obtains detailed information of the target splitter from the integrated resource management system and feeds it back to the user for selection and confirmation. SN Code Recognition and PON Port Verification: After the user selects the target optical splitter, they provide the SN code information by entering the SN code text or scanning the device's SN code QR code. The intent parsing module determines the sub-intent as "SN code recognition" and calls the semantic analysis model to parse out the accurate SN code data. The data interaction module transmits the SN code to the PON gateway alarm interface to obtain the PON port connection status and alarm information corresponding to the optical splitter, completes the PON port check, and outputs the results.

10. A home customer PON clearing intelligent agent system based on large model and target detection according to claim 9, characterized in that: The system performs a splitter port check process, which includes the following steps: Intent Trigger: The intent parsing module analyzes the user's command. If the intent is determined to be "splitter port cleanup", then this process is initiated. Splitter Photo Upload: The intelligent agent prompts the user to take a photo of the on-site splitter and upload it through the image acquisition module, or directly calls the camera to acquire photos in real time; Port status identification: The image recognition module calls the trained target detection model to process the uploaded photos, automatically identify the occupancy status (idle / occupied) of each port of the splitter, and generate structured port status data; Inspection sequence planning: The data interaction module obtains the port status data maintained by the splitter from the integrated resource management system. The inspection planning module compares the port status identified by the intelligent agent with the data in the asset management system, marks the ports with inconsistent or unclear status, and automatically plans the optimal inspection sequence, prioritizing the inspection of ports with inconsistent status. Port check execution: When the user enters the command "Start splitter port check", after the intent parsing module confirms the sub-intent, the agent guides the user to complete the detailed check of each port in the planned order. First, it calls the alarm interface to query the fiber plugging status, then detects the user's fiber unplugging behavior, obtains the port check results when the fiber is successfully unplugged, and finally synchronizes the check results to the integrated resource management system through the data interaction module.

Images