A Power Transmission Intelligent Inspection Image Defect Recognition and Intelligent Annotation System

Abstract

This invention provides a power transmission intelligent inspection image defect recognition and intelligent annotation system. The system includes a data layer, an application layer, a system layer, and a hardware layer. The data layer is used to manage and store data. The application layer is used to implement business logic and user interaction functions. The system layer is used to manage the operating environment and security framework. The hardware layer is used to provide underlying computing power, storage, and communication support. This invention integrates multimodal inspection data with multi-round time series analysis to construct an integrated intelligent inspection system with high-precision recognition, trend perception, and visual early warning capabilities.

Description

Technical Field

[0001] This invention relates to the field of power line inspection and image processing technology, and in particular to a power transmission intelligent inspection image defect recognition and intelligent annotation system. Background Technology

[0002] With the continuous expansion of my country's power grid and the increasing complexity of the operating environment for power equipment, the operation and maintenance management of transmission lines is facing unprecedented challenges. Traditional manual inspection methods suffer from low efficiency, reliance on experience for identification accuracy, and difficulty in generating structured data, making them unable to meet the high-frequency, high-reliability inspection requirements of modern transmission lines.

[0003] In recent years, with the development of UAV inspection technology and the popularization of multimodal perception methods such as image recognition and point cloud modeling, intelligent inspection of power transmission lines has gradually become a research hotspot. However, existing systems often suffer from the following problems: First, the processing capabilities of the collected images and point cloud data are limited, making it difficult to achieve high-precision fusion and spatial positioning; second, defect identification mainly relies on the results of a single inspection, lacking a cross-inspection information integration mechanism, making it difficult to accurately capture the trend of risk evolution; third, the early warning and operation and maintenance decision-making links still rely on manual judgment, with a low level of automation, making it difficult to form a closed-loop system from data collection to early warning output.

[0004] Therefore, there is an urgent need to build an intelligent system that can integrate multimodal inspection data, has high-precision defect identification capabilities, and can realize multi-round aggregation of defect information and visualization of risk assessment, so as to comprehensively improve the level of intelligent inspection and risk prediction capabilities of transmission lines.

[0005] A review of publicly available technical solutions reveals that CN115131686A proposes an intelligent power line inspection method based on active learning and semi-supervised learning. This method acquires image data of transmission lines in batches through drone inspections and fixed-point camera photography. A portion of the obtained image data is manually labeled to obtain image labels. The labeled samples are then data-augmented and mixed with unlabeled samples. A task model is then trained using semi-supervised learning. Active learning is used to determine the inconsistency between unlabeled samples and their data-augmented samples; samples with high uncertainty are re-labeled manually. This model is then applied... This solution identifies the condition of various components in power grid inspection scenarios and detects foreign objects in the power grid. Based on anomaly detection algorithms, it analyzes abnormal components or foreign objects to make judgments and conduct defect analysis, and prompts personnel to take safe handling measures. This solution can improve the efficiency and quality of power grid inspection, reduce the consumption of manpower and material resources for inspection, and ensure the safe operation of the power grid system. However, this solution mainly relies on image data for defect identification and lacks multimodal data fusion and spatial positioning capabilities, making it difficult to accurately reflect the distribution characteristics of defects in three-dimensional structures. At the same time, its risk assessment mechanism does not introduce multi-round inspection trend analysis and visualization, and cannot achieve comprehensive perception and intuitive early warning of the defect evolution process. Summary of the Invention

[0006] The purpose of this invention is to address the shortcomings of current systems by proposing an intelligent power transmission inspection image defect recognition and intelligent annotation system.

[0007] The present invention adopts the following technical solution:

[0008] A power transmission intelligent inspection image defect recognition and intelligent annotation system is disclosed. The system includes a data layer, an application layer, a system layer, and a hardware layer. The data layer is used to manage and store data. The application layer is used to implement business logic and user interaction functions. The system layer is used to manage the operating environment and security framework. The hardware layer is used to provide underlying computing power, storage, and communication support.

[0009] The data layer includes a local database and a cloud-based collaborative database; the local database is used to store locally collected inspection multimodal images and point cloud data, and the cloud-based collaborative database is used to complete data sharing between the local and cloud environments.

[0010] The application layer includes a defect identification and annotation module, a fusion information output module, an early warning module, a suggestion generation module, and a user interface. The defect identification and annotation module identifies transmission line defects based on multimodal inspection images and point cloud data, and intelligently annotates each predefined area in three-dimensional space. The fusion information output module aggregates the annotation information from multiple rounds of defect identification and annotation modules to generate a structured multimodal report for each predefined area. The early warning module aggregates the annotation information from multiple rounds of defect identification and annotation modules, outputs the risk level of each predefined area, and visualizes it through a three-dimensional heatmap. The suggestion generation module generates maintenance suggestion strategies based on the structured multimodal reports. The user interface provides a three-dimensional visual operation panel to facilitate information interaction between the user and the system.

[0011] Furthermore, the specific workflow of the defect identification and annotation module is as follows:

[0012] S11: Acquire inspection multimodal images and point cloud data, establish the mapping relationship between multimodal images and point cloud data, and generate an image-point cloud association structure;

[0013] S12: Preprocess the image-point cloud association structure. The preprocessing includes standardizing the inspection multimodal image and filtering the point cloud data.

[0014] S13: Map the image-point cloud association structure processed in the previous step to a predefined region; the predefined region is a transmission line area branch pre-set based on the line ledger information;

[0015] S14: Input the image-point cloud association structure within each predefined region into the pre-established defect recognition model to complete the defect recognition and annotation within each predefined region. The annotation content includes the defect type within each predefined region and a text description of the defect type.

[0016] Furthermore, the specific workflow of the fused information output module is as follows:

[0017] Perform the following for each predefined region:

[0018] S21: Obtain the annotation content of multiple rounds of inspections, including the timestamp of the inspection round, the defect type, and the text description of the defect type;

[0019] S22: Establish a defect type-defect description library, which contains a fixed standardized description template related to each defect type, and map the content obtained in the previous step to the library for extraction based on semantic similarity;

[0020] S23: Extract the fusion weight value of each defect type and its corresponding fixed standardized description template in multi-round inspections:

[0021]

[0022] Among them, W key f is the fusion weight value for a certain defect type and its corresponding fixed standardized description template. i n represents the frequency of occurrence of this defect type and its corresponding fixed standardized description template in the i-th inspection; n is the total number of inspection rounds.

[0023] t now t is the timestamp of the current time. i γ is the timestamp of the i-th inspection, T is the preset maximum time span, and γ is the time decay factor used to control the decay rate of historical data, which is set through pre-experimentation.

[0024] S24: Sort each defect type and its corresponding fixed standardized description template according to the fusion weight value from high to low to form a defect list. The defect list contains each defect type, its corresponding fixed standardized description template, and its fusion weight value. Call the DeepSeek model and input the defect list into the DeepSeek model for context integration and natural language generation to generate a structured multimodal report for the predefined region.

[0025] Furthermore, the specific workflow of the early warning module is as follows:

[0026] S31: Establish a defect type mapping relationship library and a risk level description library. The defect type mapping relationship library contains a basic fault risk factor corresponding to each defect type. The basic fault risk factor is used to describe the potential hazard level of the corresponding defect type in the transmission line. The risk level description library contains the risk impact weights corresponding to the degree descriptive words describing the defect type.

[0027] S32: Obtain the annotation content of multiple rounds of inspections, and extract the degree description words corresponding to each defect type from the text description of the defect type;

[0028] S33: Combine the defect type mapping relationship library and the risk degree description library to find the corresponding basic failure risk factors and risk impact weights for the defect type and degree descriptive words obtained in the previous step;

[0029] S34: Calculate the risk level assessment value for each predefined area;

[0030] S35: Generate a risk heatmap for each predefined area based on the risk level assessment value of each predefined area; the heatmap maps the position of each predefined area in the three-dimensional point cloud to its corresponding risk level assessment value, and sets a color gradient in the heatmap according to the specific value of the risk level assessment value. The higher the risk level assessment value, the higher the corresponding color gradient, thereby realizing a visual representation of the risk level of each predefined area.

[0031] Furthermore, the suggestion generation module calls the DeepSeek model and inputs the structured multimodal report into the DeepSeek model to complete the generation of maintenance suggestion strategies.

[0032] Furthermore, the system layer includes an operating system framework, a secure communication framework, and a resource scheduling engine; the operating system framework is used to implement device drivers, file management, and hardware resource scheduling; the secure communication framework is used to ensure data communication security; and the resource scheduling engine is used to dynamically allocate computing power within the system.

[0033] Furthermore, the hardware layer includes a GPU computing unit, a multi-core CPU, a high-speed storage device, a network interface module, and a power supply and heat dissipation module; the GPU computing unit is equipped with a 24GB graphics card and supports large model inference and real-time image processing; the multi-core CPU is used to complete task scheduling and computing task execution; the high-speed storage device is used to complete local data caching; the network interface module is equipped with 5G communication and is used to complete the interconnection between the local machine and the cloud; the power supply and heat dissipation module is used to provide power to the system devices.

[0034] The beneficial effects achieved by this invention are as follows:

[0035] This invention constructs a multi-level integrated intelligent transmission line inspection image defect recognition and intelligent annotation system, realizing a closed-loop process from raw data acquisition, defect recognition, semantic fusion, risk assessment to visualization output and intelligent suggestion generation. By spatially mapping and deeply fusing multimodal images with point cloud data, the accuracy of defect recognition and spatial positioning precision are improved. By introducing a fusion mechanism and time decay factor for multi-round inspection data, a defect list is generated by combining each defect type and its corresponding fixed standardized description template. The defect list is then input into the DeepSeek text generation model to complete the semantic integration of defects and the automatic generation of structured inspection reports. This significantly improves the utilization efficiency of multi-round inspection data and the consistency of defect expression, significantly reduces manual intervention, and enhances the system's intelligence level and application effectiveness in actual inspection scenarios. Attached Figure Description

[0036] The invention will be further understood from the following description taken in conjunction with the accompanying drawings. The components in the drawings are not necessarily drawn to scale, but rather the emphasis is on illustrating the principles of the embodiments. In different views, the same reference numerals designate corresponding parts.

[0037] Figure 1 This is a schematic diagram of the overall modules of the present invention.

[0038] Figure 2 This is a schematic diagram of the workflow of the defect identification and annotation module of the present invention.

[0039] Figure 3 This is a schematic diagram of the workflow of the integrated information output module of the present invention.

[0040] Figure 4 This is a schematic diagram of the workflow of the early warning module of the present invention.

[0041] Figure 5 To illustrate the different values ​​of the time decay factor γ in this invention, About functions A schematic diagram of the changing function. Detailed Implementation

[0042] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to its embodiments. It should be understood that the specific embodiments described herein are only for explaining the present invention and are not intended to limit the present invention. Other systems, methods, and / or features of this embodiment will become apparent to those skilled in the art after reviewing the following detailed description. It is intended that all such additional systems, methods, features, and advantages are included within this specification, are included within the scope of the present invention, and are protected by the appended claims. Further features of the disclosed embodiments are described in the following detailed description, and these features will be apparent from the following detailed description.

[0043] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components. In the description of the present invention, it should be understood that if terms such as "upper," "lower," "left," "right," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or component referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting the present patent. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0044] Example 1:

[0045] like Figure 1As shown in the figure, this embodiment provides a power transmission intelligent inspection image defect recognition and intelligent annotation system. The system includes a data layer, an application layer, a system layer, and a hardware layer. The data layer is used to manage and store data. The application layer is used to implement business logic and user interaction functions. The system layer is used to manage the operating environment and security framework. The hardware layer is used to provide underlying computing power, storage, and communication support.

[0046] The data layer includes a local database and a cloud-based collaborative database; the local database is used to store locally collected inspection multimodal images and point cloud data, and the cloud-based collaborative database is used to complete data sharing between the local and cloud environments.

[0047] The application layer includes a defect identification and annotation module, a fusion information output module, an early warning module, a suggestion generation module, and a user interface. The defect identification and annotation module identifies transmission line defects based on multimodal inspection images and point cloud data, and intelligently annotates each predefined area in three-dimensional space. The fusion information output module aggregates the annotation information from multiple rounds of defect identification and annotation modules to generate a structured multimodal report for each predefined area. The early warning module aggregates the annotation information from multiple rounds of defect identification and annotation modules, outputs the risk level of each predefined area, and visualizes it through a three-dimensional heatmap. The suggestion generation module generates maintenance suggestion strategies based on the structured multimodal reports. The user interface provides a three-dimensional visual operation panel to facilitate information interaction between the user and the system.

[0048] Furthermore, such as Figure 2 As shown, the specific workflow of the defect identification and annotation module is as follows:

[0049] S11: Acquire multimodal images and point cloud data for inspection, establish the mapping relationship between multimodal images and point cloud data, and generate an image-point cloud association structure;

[0050] S12: Preprocess the image-point cloud association structure. The preprocessing includes standardizing the inspection multimodal image and filtering the point cloud data.

[0051] S13: Map the image-point cloud association structure processed in the previous step to a predefined region; the predefined region is a transmission line area branch pre-set based on the line ledger information;

[0052] S14: Input the image-point cloud association structure within each predefined region into the pre-established defect recognition model to complete the defect recognition and annotation within each predefined region. The annotation content includes the defect type within each predefined region and a text description of the defect type.

[0053] Specifically, the defect recognition model is a deep learning model based on a multimodal fusion architecture. The defect recognition model extracts texture features from multimodal images and spatial structure features from point cloud data, performs feature alignment, and then inputs them into a pre-trained multimodal feature decoding network to complete the identification of defect types, three-dimensional localization of defect regions, and text description of corresponding defect types.

[0054] Furthermore, such as Figure 3 , Figure 5 As shown, the specific workflow of the fused information output module is as follows:

[0055] Perform the following for each predefined region:

[0056] S21: Obtain the annotation content of multiple rounds of inspections, including the timestamp of the inspection round, the defect type, and the text description of the defect type;

[0057] S22: Establish a defect type-defect description library, which contains a fixed standardized description template related to each defect type, and map the content obtained in the previous step to the library for extraction based on semantic similarity;

[0058] S23: Extract the fusion weight value of each defect type and its corresponding fixed standardized description template in multi-round inspections:

[0059]

[0060] Among them, W key f is the fusion weight value for a certain defect type and its corresponding fixed standardized description template. i n represents the frequency of occurrence of this defect type and its corresponding fixed standardized description template in the i-th inspection; n is the total number of inspection rounds.

[0061] t now t is the timestamp of the current time. i γ is the timestamp of the i-th inspection, T is the preset maximum time span, and γ is the time decay factor used to control the decay rate of historical data, which is set through pre-experimentation.

[0062] S24: Sort each defect type and its corresponding fixed standardized description template according to the fusion weight value from high to low to form a defect list. The defect list contains each defect type, its corresponding fixed standardized description template, and its fusion weight value. Call the DeepSeek model and input the defect list into the DeepSeek model for context integration and natural language generation to generate a structured multimodal report for the predefined region.

[0063] This solution improves the stability and accuracy of defect identification results by integrating and fusing annotations from multiple rounds of inspections; it standardizes diverse descriptions and enhances consistency by employing a defect type-defect description library combined with semantic similarity mapping; it introduces a time decay weighting mechanism to enable the system to dynamically assess defect development trends and improve sensitivity to recent anomalies; and it combines the defect fusion weight value W... key The sorting and highlighting of key defect information effectively supports the content logic of intelligent report generation; by inputting the fused defect list into the DeepSeek model to complete text integration, a structured multimodal inspection report is automatically generated, reducing manual sorting work and improving work efficiency;

[0064] Furthermore, the suggestion generation module calls the DeepSeek model and inputs the structured multimodal report into the DeepSeek model to complete the generation of maintenance suggestion strategies;

[0065] The system layer includes an operating system framework, a secure communication framework, and a resource scheduling engine; the operating system framework is used to implement device drivers, file management, and hardware resource scheduling; the secure communication framework is used to ensure data communication security; and the resource scheduling engine is used to dynamically allocate computing power within the system.

[0066] The hardware layer includes a GPU computing unit, a multi-core CPU, a high-speed storage device, a network interface module, and a power supply and heat dissipation module. The GPU computing unit is equipped with a 24GB graphics card and supports large model inference and real-time image processing. The multi-core CPU is used to perform task scheduling and computation task execution. The high-speed storage device is used to perform local data caching. The network interface module is equipped with 5G communication and is used to achieve interconnection between the local network and the cloud. The power supply and heat dissipation module is used to provide power to the system devices.

[0067] Example 2:

[0068] This embodiment should be understood to include at least all the features of any of the foregoing embodiments, and to further improve upon them;

[0069] This embodiment provides a power transmission intelligent inspection image defect recognition and intelligent annotation system. The system includes a data layer, an application layer, a system layer, and a hardware layer. The data layer is used to manage and store data. The application layer is used to implement business logic and user interaction functions. The system layer is used to manage the operating environment and security framework. The hardware layer is used to provide underlying computing power, storage, and communication support.

[0070] The application layer includes a defect identification and annotation module, a fusion information output module, an early warning module, a suggestion generation module, and a user interface. The defect identification and annotation module identifies transmission line defects based on multimodal inspection images and point cloud data, and intelligently annotates each predefined area in three-dimensional space. The fusion information output module aggregates the annotation information from multiple rounds of defect identification and annotation modules to generate a structured multimodal report for each predefined area. The early warning module aggregates the annotation information from multiple rounds of defect identification and annotation modules, outputs the risk level of each predefined area, and visualizes it through a three-dimensional heatmap. The suggestion generation module generates maintenance suggestion strategies based on the structured multimodal reports. The user interface provides a three-dimensional visual operation panel to facilitate information interaction between the user and the system.

[0071] Furthermore, such as Figure 4 As shown, the specific workflow of the early warning module is as follows:

[0072] S31: Establish a defect type mapping relationship library and a risk level description library. The defect type mapping relationship library contains a basic fault risk factor corresponding to each defect type. The basic fault risk factor is used to describe the potential hazard level of the corresponding defect type in the transmission line. The risk level description library contains the risk impact weights corresponding to the degree descriptive words describing the defect type.

[0073] Specifically, the defect type mapping database can be evaluated by experts in fields such as power inspection, operation and maintenance, and security. Based on the impact of various defects on the system, scores are assigned to form a classification and mapping to basic fault risk factors.

[0074] Specifically, the risk level description library can collect a large amount of manually annotated or system-generated defect description text, classify and statistically analyze commonly used degree descriptive words, and manually formulate the mapping of degree descriptive words to risk impact weights by combining Chinese semantic strength levels.

[0075] S32: Obtain the annotation content of multiple rounds of inspections, and extract the degree description words corresponding to each defect type from the text description of the defect type;

[0076] S33: Combine the defect type mapping relationship library and the risk degree description library to find the corresponding basic failure risk factors and risk impact weights for the defect type and degree descriptive words obtained in the previous step;

[0077] S34: Calculate the risk level assessment value for each predefined area:

[0078]

[0079] Among them, R 总R is the risk level assessment value for a predefined area. j d represents the cumulative risk assessment value of the j-th defect type in the current region within the predefined region, where l is the total number of defect types appearing in the predefined region; i,j For the basic failure risk factor of the j-th defect type during the i-th inspection, m i,j α represents the risk impact weight of the j-th defect type during the i-th inspection; α is the correction coefficient, used to adjust the impact weight of the time cumulative trend term in the overall risk assessment, and is set through pre-experimentation.

[0080] S35: Generate a risk heatmap for each predefined area based on the risk level assessment value of each predefined area; the heatmap maps the position of each predefined area in the three-dimensional point cloud to its corresponding risk level assessment value, and sets a color gradient in the heatmap according to the specific value of the risk level assessment value. The higher the risk level assessment value, the higher the corresponding color gradient, thereby realizing a visual representation of the risk level of each predefined area.

[0081] By calculating the risk level assessment value for each predefined area using the above method, a comprehensive quantitative assessment of multiple types of defect risks within the predefined area can be achieved; among which, the cumulative risk assessment value R j The first term to calculate More attention is paid to recent defects, the second item It pays more attention to long-term defects and hidden dangers, thereby achieving accurate early warning and dynamic trend judgment of transmission line inspection risks, and improving the actual value of inspection data and the scientific nature of operation and maintenance decisions.

[0082] The content disclosed above is only a preferred and feasible embodiment of the present invention, and is not intended to limit the scope of protection of the present invention. Therefore, all equivalent technical changes made based on the content of the present invention specification and drawings are included within the scope of protection of the present invention. Furthermore, the elements therein can be updated as technology develops.

Claims

1. A power transmission intelligent inspection image defect recognition and intelligent annotation system, characterized in that, The system comprises a data layer, an application layer, a system layer, and a hardware layer; the data layer manages and stores data; the application layer implements business logic and user interaction functions; the system layer manages the operating environment and security framework; and the hardware layer provides underlying computing power, storage, and communication support. The data layer includes a local database and a cloud-based collaborative database; the local database is used to store locally collected inspection multimodal images and point cloud data, and the cloud-based collaborative database is used to complete data sharing between the local and cloud environments. The application layer includes a defect identification and annotation module, a fusion information output module, an early warning module, a suggestion generation module, and a user interaction interface; The defect identification and annotation module is used to identify transmission line defects based on inspection multimodal images and point cloud data, and intelligently annotate each predefined area in three-dimensional space; the fusion information output module is used to aggregate the annotation information from multiple rounds of defect identification and annotation modules to generate a structured multimodal report for each predefined area; the early warning module is used to aggregate the annotation information from multiple rounds of defect identification and annotation modules to output the risk level of each predefined area, and visualize it through a three-dimensional heat map; the suggestion generation module is used to generate maintenance suggestion strategies based on the structured multimodal reports; the user interface is used to provide a three-dimensional visual operation panel to complete information interaction between the user and the system; The specific workflow of the fused information output module is as follows: Perform the following for each predefined region: S21: Obtain the annotation content of multiple rounds of inspections, including the timestamp of the inspection round, the defect type, and the text description of the defect type; S22: Establish a defect type-defect description library, which contains a fixed standardized description template related to each defect type, and map the content obtained in the previous step to the library for extraction based on semantic similarity; S23: Extract the fusion weight value of each defect type and its corresponding fixed standardized description template in multi-round inspections: ； in, The fusion weight value for a certain defect type and its corresponding fixed standardized description template. For this type of defect and its corresponding fixed standardized description template, in the [section / section]... Frequency of occurrence during each inspection; For the total number of inspection rounds; The timestamp of the current time. For the first The timestamp of the next inspection. The preset maximum time span; This is the time decay factor, used to control the decay rate of historical data, and is set through pre-experimentation; S24: Sort each defect type and its corresponding fixed standardized description template according to the fusion weight value from high to low to form a defect list. The defect list contains each defect type, its corresponding fixed standardized description template, and its fusion weight value. The DeepSeek model is invoked, and the defect list is input into the DeepSeek model for context integration and natural language generation, generating a structured multimodal report for the predefined region; The specific workflow of the early warning module is as follows: S31: Establish a defect type mapping relationship library and a risk level description library. The defect type mapping relationship library contains a basic fault risk factor corresponding to each defect type. The basic fault risk factor is used to describe the potential hazard level of the corresponding defect type in the transmission line. The risk level description library contains the risk impact weights corresponding to the degree descriptive words describing the defect type. S32: Obtain the annotation content of multiple rounds of inspections, and extract the degree description words corresponding to each defect type from the text description of the defect type; S33: Combine the defect type mapping relationship library and the risk degree description library to find the corresponding basic failure risk factors and risk impact weights for the defect type and degree descriptive words obtained in the previous step; S34: Calculate the risk level assessment value for each predefined area; S35: Generate a risk heatmap for each predefined area based on the risk level assessment value of each predefined area; the heatmap maps the position of each predefined area in the three-dimensional point cloud to its corresponding risk level assessment value, and sets a color gradient in the heatmap according to the specific value of the risk level assessment value. The higher the risk level assessment value, the higher the corresponding color gradient, thereby realizing a visual representation of the risk level of each predefined area.

2. The intelligent power transmission inspection image defect recognition and intelligent annotation system according to claim 1, characterized in that, The specific workflow of the defect identification and annotation module is as follows: S11: Acquire inspection multimodal images and point cloud data, establish the mapping relationship between multimodal images and point cloud data, and generate an image-point cloud association structure; S12: Preprocess the image-point cloud association structure. The preprocessing includes standardizing the inspection multimodal image and filtering the point cloud data. S13: Map the image-point cloud association structure processed in the previous step to a predefined region; the predefined region is a transmission line area branch pre-set based on the line ledger information; S14: Input the image-point cloud association structure within each predefined region into the pre-established defect recognition model to complete the defect recognition and annotation within each predefined region. The annotation content includes the defect type within each predefined region and a text description of the defect type.

3. The intelligent power transmission inspection image defect recognition and intelligent annotation system according to claim 1, characterized in that, The suggestion generation module generates maintenance suggestion strategies by calling the DeepSeek model and inputting the structured multimodal report into the DeepSeek model.

4. The intelligent power transmission inspection image defect recognition and intelligent annotation system according to claim 1, characterized in that, The system layer includes an operating system framework, a secure communication framework, and a resource scheduling engine; the operating system framework is used to implement device drivers, file management, and hardware resource scheduling; the secure communication framework is used to ensure data communication security; and the resource scheduling engine is used to dynamically allocate computing power within the system.

5. The intelligent power transmission inspection image defect recognition and intelligent annotation system according to claim 1, characterized in that, The hardware layer includes a GPU computing unit, a multi-core CPU, a high-speed storage device, a network interface module, and a power supply and heat dissipation module. The GPU computing unit is equipped with a 24GB graphics card and supports large model inference and real-time image processing. The multi-core CPU is used to perform task scheduling and computation task execution. The high-speed storage device is used to perform local data caching. The network interface module is equipped with 5G communication and is used to achieve interconnection between the local network and the cloud. The power supply and heat dissipation module is used to provide power to the system devices.

Images