Intelligent Inspection Data Processing Methods and Systems

Abstract

This application relates to a method and system for processing intelligent inspection data. The method includes: analyzing infrared images in an inspection scenario to obtain the image resolution, temperature matrix, and pseudo-color image of the infrared images; synchronizing the annotation information of the infrared images and visible light images in the inspection scenario to obtain multimodal annotation data; performing defect quantification analysis on the multimodal annotation data based on preset reference object information to generate a quantitative result of the true physical size of the defects; and storing the image resolution, temperature matrix, pseudo-color image, and quantification result in association. This application can achieve full-process optimization of inspection data, effectively improving processing stability, defect detection accuracy, and system practicality.

Description

Technical Field

[0001] This application relates to the field of intelligent inspection defect detection technology, and in particular to a method and system for processing intelligent inspection data. Background Technology

[0002] With the continuous deepening of industrial digital transformation, industrial equipment operation and maintenance has become a core link in ensuring production continuity and reducing safety risks, placing higher demands on the intelligence, efficiency, and precision of operation and maintenance. Intelligent inspection technology, by overcoming the time and space limitations of manual inspection and improving defect identification efficiency, is widely used in many key fields such as power, chemical, intelligent manufacturing, rail transportation, and energy.

[0003] In existing intelligent inspection technology systems, multimodal data fusion solutions have become one of the mainstream development directions due to their ability to integrate equipment information from different dimensions. These solutions typically integrate various types of inspection data, such as image-based, sensor-based, and operational condition-based data. By complementing the information advantages of different data sources, they achieve comprehensive and multi-dimensional detection and analysis of equipment defects. Compared to single-modal data processing, they offer significant improvements in defect identification coverage and early hazard detection capabilities, making them an important technology choice for equipment operation and maintenance in complex industrial scenarios.

[0004] However, the current multimodal inspection data processing suffers from significant heterogeneity due to data from different sources and of different types, resulting in low data fusion efficiency. In addition, multimodal data annotation relies heavily on manual intervention, with inconsistent annotation rules and high costs, making it difficult to meet the needs of large-scale applications. These problems pose a serious challenge to the precise and efficient development of industrial equipment operation and maintenance. Summary of the Invention

[0005] Based on this, it is necessary to address the above-mentioned problems. This application provides a method and system for processing intelligent inspection data, which achieves standardized and intelligent processing of the entire inspection data chain by establishing unified data processing specifications, optimizing multimodal data collaboration mechanisms and adaptive storage architecture.

[0006] In a first aspect, this application provides a method for processing intelligent inspection data. The method includes: parsing an infrared image in an inspection scene to obtain the image resolution, temperature matrix, and pseudo-color image of the infrared image; synchronizing the annotation information of the infrared image and the visible light image in the inspection scene to obtain multimodal annotation data; performing defect quantification analysis on the multimodal annotation data based on preset reference object information to generate a quantification result of the true physical size of the defect; and storing the image resolution, temperature matrix, pseudo-color image, and quantification result in association.

[0007] Optionally, in this embodiment, parsing the infrared image in the inspection scenario to obtain the image resolution, temperature matrix, and pseudo-color image of the infrared image includes: encapsulating the infrared thermal imaging processing TSDK using JNA technology, constructing a parsing channel for R-JPEG format data in a Java environment to initialize an independent parsing context containing a TSDK handle and an image pointer; based on the independent parsing context, calling the resolution extraction interface of the TSDK to read the image resolution of the infrared thermal image; based on the independent parsing context, calling the temperature matrix generation interface of the TSDK to extract the temperature matrix of the infrared thermal image; receiving pseudo-color mode parameters, and based on the independent parsing context, calling the pseudo-color conversion interface of the TSDK to convert the infrared thermal image and generate the pseudo-color image; thus achieving accurate parsing of infrared image data and outputting three core data types: image resolution, temperature matrix, and pseudo-color image, supporting subsequent defect quantitative analysis and visualization.

[0008] Optionally, in this embodiment of the application, before parsing the infrared image in the inspection scenario, the ThreadLocal mechanism is used to bind the independent parsing context to the current processing thread, and the independent parsing context instance of the infrared image is bound to the current processing thread to ensure that the independent parsing contexts of different threads are isolated from each other.

[0009] Optionally, in this embodiment, synchronizing the annotation information of the infrared image and the visible light image in the inspection scenario to obtain multimodal annotation data includes: extracting feature points from the visible light image and the infrared image respectively; matching the feature points of the visible light image and the infrared image to obtain a matching result; determining the homography matrix between the visible light image and the infrared image based on the matching result; mapping the annotation coordinates in the visible light image annotation data to the infrared image based on the homography matrix to obtain infrared image annotation coordinates; generating infrared image annotation data corresponding to the visible light image annotation data based on the infrared image annotation coordinates and the annotation type and defect attributes in the visible light image annotation data; associating the visible light image annotation data with the infrared image annotation data to generate the multimodal annotation data; ensuring multimodal annotation consistency through feature matching and coordinate mapping, achieving accurate synchronization of infrared and visible light image annotation information, reducing annotation costs, and providing a unified benchmark for defect quantification.

[0010] Optionally, in this embodiment of the application, matching feature points of the visible light image and the infrared image to obtain matching results includes: filtering feature points with low response values ​​in the features of the visible light image and the infrared image to obtain high-quality feature points; matching the high-quality feature points of the visible light image and the infrared image using a FLANN matcher to obtain high-quality matching pairs whose distance meets preset conditions, so as to generate feature point matching results accurately and efficiently.

[0011] Optionally, in this embodiment, based on preset reference information, defect quantification analysis is performed on the multimodal annotation data to generate a quantification result of the true physical size of the defect. This includes: using an AI model to identify defects in the multimodal annotation data to obtain the bounding box information and defect type of the defect; obtaining the pixel size of the defect in the visible light image and infrared image based on the bounding box information of the defect; determining the actual physical size of the defect in the two image dimensions of the visible light image and infrared image based on the reference information and the pixel size; generating a quantification result of the true physical size of the defect based on the actual physical size in the two image dimensions; by combining AI recognition and reference calibration, and fusing the pixel size data of the dual-modal image, the problem of inconsistent quantization benchmarks is solved, and the accuracy of the calculation of the true physical size of the defect is improved.

[0012] Optionally, in this embodiment, the preset reference object information includes the actual physical size of the reference object and its pixel size in the visible light image and infrared image; the method of obtaining the reference object information includes: associating the preset reference object with the visible light image and infrared image through a three-level binding mechanism; wherein, the three-level binding mechanism includes data source level binding, data level binding and default level binding; matching the reference object associated with the visible light image and infrared image where the defect is located based on the priority order of the data source level, data level and default level to obtain the successfully matched reference object information.

[0013] Optionally, in this embodiment, the image resolution, temperature matrix, pseudo-color image, and quantization result are associated and stored together, including: assigning a unique association identifier to the image resolution, temperature matrix, pseudo-color image, and quantization result in the same inspection scenario; storing the visible light image, infrared image, and pseudo-color image to a file server and recording the storage path; binding the storage path with the unique association identifier and storing them together; establishing a data link through the unique identifier, clarifying the association logic between structured and unstructured data, improving data traceability and management efficiency, and achieving standardized associated storage of multiple types of inspection data.

[0014] In the above implementation process, the Spring Boot Starter framework is used for database access, caching, and access control to improve data access efficiency and ensure data security; and a MySQL to PostgreSQL intelligent migration component is used to store and migrate business data related to the inspection scenario, solving the data storage and migration problem across database platforms; ensuring the integrity and consistency of business data, and effectively improving the maintainability and scalability of the system.

[0015] Secondly, this application also provides a data processing system, the data processing system comprising: An infrared image analysis module is used to analyze infrared images in an inspection scenario to obtain the image resolution, temperature matrix, and pseudo-color image of the infrared image; a multimodal synchronous calibration module is used to synchronize the annotation information of the infrared image and the visible light image in the inspection scenario to obtain multimodal annotation data; a defect quantification analysis module is used to perform defect quantification analysis on the multimodal annotation data based on preset reference object information to generate the true physical size quantification result of the defect; and an associated storage module is used to associate and store the image resolution, temperature matrix, pseudo-color image, and quantification result.

[0016] The intelligent inspection data processing method and system of this application, by integrating thread resource allocation mechanism, multimodal annotation collaboration and standardized data management scheme, realizes full-process optimization of multimodal inspection data parsing, quantitative analysis and storage migration; it can adapt to the processing needs of large-scale, multi-dimensional inspection data in industrial scenarios, and effectively improve the overall reliability and performance of the system. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the conventional technology, the drawings used in the description of the embodiments or the conventional technology will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of a method for processing intelligent inspection data provided in one embodiment of this application; Figure 2 This is a structural block diagram of an intelligent inspection data processing system provided in another embodiment of this application; Figure labeling: Infrared image analysis module-10; Multimodal synchronous calibration module-20; Defect quantification analysis module-30; Correlation storage module-40; Data processing system-200. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0020] In one embodiment, see Figure 1 This application provides a method for processing intelligent inspection data, which includes the following steps: S11~S14.

[0021] S11: Analyze the infrared image in the inspection scenario to obtain the image resolution, temperature matrix and pseudo-color image of the infrared image.

[0022] S12: Synchronize the annotation information of the infrared image and the visible light image in the inspection scene to obtain multimodal annotation data.

[0023] S13: Based on preset reference information, perform defect quantification analysis on the multimodal annotation data to generate the true physical size quantification result of the defect.

[0024] S14: The image resolution, temperature matrix, pseudo-color image, and quantization result are stored together.

[0025] The intelligent inspection data processing method of this application achieves full-process optimization of multimodal inspection data parsing, quantitative analysis and storage migration by integrating thread resource allocation mechanism, multimodal annotation collaboration and standardized data management scheme; it can adapt to the processing needs of large-scale, multi-dimensional inspection data in industrial scenarios and effectively improve the overall reliability and performance of the system.

[0026] In step S11, please refer to Figure 1 In step S11, the infrared image in the inspection scenario is analyzed to obtain the image resolution, temperature matrix and pseudo-color image of the infrared image.

[0027] As an example, before parsing the infrared image in the inspection scenario, the method further includes: using the ThreadLocal mechanism to bind the independent parsing context to the current processing thread.

[0028] Specifically, before executing the infrared image parsing step, an independent parsing context can be bound to the current processing thread using the ThreadLocal mechanism. First, an independent infrared image parsing context structure containing a TSDK handle, an image pointer, and a unique image identifier is defined in the Java code, and a corresponding ThreadLocal instance is created to achieve thread-local storage. Then, the JNA-encapsulated TSDK initialization interface is called to obtain a valid TSDK operation handle. Storage memory for the infrared image is allocated through the TSDK memory allocation interface, and an image pointer is obtained. The handle, image pointer, and the unique identifier of the infrared image to be parsed are encapsulated into an independent infrared image parsing context instance. The ThreadLocal set() method is called to bind the independent infrared image parsing context instance to the current processing thread, ensuring that the independent parsing contexts of different threads are isolated from each other.

[0029] Specifically, during the independent resolution context initialization process, TSDK interface call exceptions can be caught simultaneously using try-catch blocks in the programming language. If an exception occurs, the resource cleanup process is immediately triggered, calling the TSDK handle release interface and memory release interface to reclaim allocated resources, and clearing the current thread's context storage using the ThreadLocal remove() method.

[0030] As an example, step S11 may include the following steps: S111~S114.

[0031] S111: The infrared thermal imaging processing TSDK is encapsulated using JNA technology to build a parsing channel for R-JPEG format data in a Java environment, in order to initialize an independent parsing context containing the TSDK handle and image pointer.

[0032] S112: Based on the independent parsing context, call the resolution extraction interface of the TSDK to read the image resolution of the infrared thermal imaging.

[0033] S113: Based on the independent parsing context, call the temperature matrix generation interface of the TSDK to extract the temperature matrix of the infrared thermal imaging.

[0034] S114: Receive pseudo-color mode parameters, and based on the independent parsing context, call the pseudo-color conversion interface of the TSDK to convert the infrared thermal image and generate the pseudo-color image.

[0035] Specifically, based on this independent parsing context, the corresponding function interfaces of TSDK are called sequentially; the resolution extraction interface is called, the handle and image pointer are input, the metadata of the R-JPEG image header is parsed, and the image resolution composed of width and height pixels is obtained; the temperature matrix generation interface is called, the temperature data area in the image memory is accessed, and the temperature value corresponding to each pixel is converted into a two-dimensional floating-point temperature matrix; at the same time, the pseudo-color mode parameters passed by the user are received, the pseudo-color conversion interface of TSDK is called, and according to the temperature value of each pixel in the temperature matrix and the color mapping rules of the pseudo-color mode, RGB color is assigned to each pixel (various colors are mixed by the principle of superposition of the three primary colors of light), generating an intuitive pseudo-color image, and finally outputting three core data: image resolution, temperature matrix, and pseudo-color image, to support subsequent defect quantitative analysis and visualization.

[0036] In step S12, please refer to Figure 1 In step S12, the annotation information of the infrared image and the visible light image in the inspection scene is synchronized to obtain multimodal annotation data.

[0037] As an example, step S12 may include the following steps: S121~S126.

[0038] S121: Extract feature points from the visible light image and the infrared image respectively.

[0039] S122: Match the feature points of the visible light image and the infrared image to obtain the matching result.

[0040] S123: Based on the matching results, determine the homography matrix between the visible light image and the infrared image.

[0041] S124: Based on the homography matrix, map the annotation coordinates in the visible light image annotation data to the infrared image to obtain the infrared image annotation coordinates.

[0042] S125: Based on the infrared image annotation coordinates and the annotation type and defect attributes in the visible light image annotation data, generate infrared image annotation data corresponding to the visible light image annotation data.

[0043] S126: Associate the visible light image annotation data with the infrared image annotation data to generate the multimodal annotation data.

[0044] As an example, matching feature points of the visible light image and the infrared image to obtain a matching result includes: Filter out feature points with low response values ​​from the features of the visible light image and infrared image to obtain high-quality feature points; The FLANN matcher is used to match high-quality feature points of the visible light image and infrared image to obtain high-quality matching pairs whose distance meets the preset conditions, and the matching result is generated.

[0045] Specifically, the ORB algorithm is first used to extract feature points from the visible light and infrared images of the inspection scene, with each feature point accompanied by location, scale, and orientation descriptions. Then, the FLANN algorithm is used to match the feature points of the two images, and the RANSAC algorithm is used to remove incorrect matches, selecting high-quality pairs of identically named feature points. Next, based on these high-quality identically named feature point pairs, the homography matrix describing the projection transformation relationship between the infrared and visible light images is obtained using the least squares method. After the visible light image is annotated (including defect bounding box coordinates, annotation type, and defect attributes), the annotation coordinates from the visible light annotation data are substituted into the homography matrix, and the corresponding defect's annotation coordinates in the infrared image are calculated through projection transformation. Based on these infrared annotation coordinates, and using the type and attribute information of the visible light annotations, infrared image annotation data with a consistent format is generated. Finally, the visible light and infrared annotation data are associated through image pair identification, forming multimodal annotation data containing defect information of both modes.

[0046] In step S13, please refer to Figure 1 In step S13, based on preset reference information, defect quantification analysis is performed on the multimodal annotation data to generate the true physical size quantification result of the defect.

[0047] As an example, step S13 may include the following steps: S131~S134.

[0048] S131: Use an AI model to identify defects in the multimodal labeled data to obtain the bounding box information and defect type of the defects.

[0049] S132: Based on the bounding box information of the defect, obtain the pixel size of the defect in the visible light image and the infrared image, respectively.

[0050] S133: Based on the reference object information and the pixel size, determine the actual physical size of the defect in both visible light and infrared image dimensions.

[0051] S134: Based on the actual physical size in the two image dimensions, generate the true physical size quantification result of the defect.

[0052] As an example, the preset reference object information includes the actual physical size of the reference object and its pixel size in visible light and infrared images; the reference object information is obtained in the following ways: The preset reference object is associated with the visible light image and the infrared image through a three-level binding mechanism; wherein, the three-level binding mechanism includes data source level binding, data level binding and default level binding; Based on the priority order of the data source level, data level, and default level, the reference objects associated with the visible light image and infrared image where the defect is located are matched to obtain the information of the successfully matched reference objects.

[0053] Specifically, first, the AI ​​defect recognition model that supports multimodal input is started, and the multimodal annotation data generated by S12 is input; the model outputs the bounding box information and defect type (such as crack) of each defect by fusing the texture and temperature features of the two images; and calculates the defect pixel size in the two images according to the bounding box coordinates respectively. For example, the defect pixel width in the visible light image is 300-100=200px, and the pixel width in the infrared image is 305-105=200px.

[0054] Specifically, the reference object matching process can be initiated by prioritizing "data source level → data level → default level". If the current image belongs to the "Device A Inspection Task" bound to the data source level, then the preset reference object information on Device A is matched, such as a nameplate with an actual physical width of 50mm, a pixel width of 100px in the visible light image, and a pixel width of 95px in the infrared image. Then, the scale of the two images is calculated: visible light image scale = 50mm / 100px = 0.5mm / px, infrared image scale = 50mm / 95px ≈ 0.53mm / px. The missing data is then converted based on the scale. The actual physical size of the defect is calculated as follows: in the visible light image, the physical width of the defect is 200px × 0.5mm / px = 100mm; in the infrared image, the physical width of the defect is 200px × 0.53mm / px ≈ 106mm. Finally, the results can be fused using a weighted average method (for example, weights can be assigned according to the recognition confidence of the two images: the confidence of the visible light image is 0.9, and the confidence of the infrared image is 0.8, so the weight ratio is 9:8) to calculate the final quantized size = (100×9 + 106×8) / (9 + 8) ≈ 102.8mm, generating a quantized result that includes the defect type and the actual physical size of the defect.

[0055] In step S14, please refer to Figure 1 In step S14, the image resolution, temperature matrix, pseudo-color image, and quantization result are associated and stored.

[0056] As an example, step S14 may include the following steps: S141~S143.

[0057] S141: Assign a unique association identifier to the image resolution, temperature matrix, pseudo-color image and quantization result under the same inspection scenario.

[0058] S142: Store the visible light image, infrared image, and pseudo-color image to a file server and record the storage path.

[0059] S143: Bind the storage path to the unique association identifier and associate it with storage.

[0060] Specifically, firstly, unique association identifiers are assigned to structured data (image resolution, temperature matrix, defect quantification results) and unstructured data (visible light images, infrared images, pseudo-color images) in the same inspection scenario; unstructured image files are stored on a file server and the storage path is recorded, while structured data is organized into a standardized format; then, a unique identifier is established in the database to map the image storage path and structured data, completing the unified association storage of the two types of data.

[0061] As an example, the method for processing the intelligent inspection data also includes: The Spring Boot Starter framework is used for database access, caching, and access control, and the MySQL to PostgreSQL intelligent migration component is used to store and migrate business data related to the inspection scenario.

[0062] Specifically, in the process of processing intelligent inspection data, the Spring Boot Starter framework integrates MyBatis to achieve associated storage and fast CRUD operations (Create, Read, Update, Delete) of structured and unstructured data paths, greatly simplifying the read and write configuration of inspection data association tables; it integrates Redis to cache frequently accessed inspection data (such as recent defect quantification results and image resolution of commonly used equipment), reducing database query pressure, improving query response speed during data processing, and adapting to the need for real-time inspection data viewing; at the same time, it combines Spring Security to implement role-based access control, setting permission thresholds for each stage of data processing to ensure the security of inspection data during processing and flow. The MySQL to PostgreSQL intelligent migration component focuses on cross-database compatibility and scalability of data processing, addressing potential database environment switching needs during inspection data processing. It intelligently adapts data types (e.g., converting MySQL's JSON-formatted temperature matrix to PostgreSQL's JSONB type, and converting DATETIME-formatted timestamps to TIMESTAMP) to ensure no format loss in structured data and associated unstructured data paths during cross-database migration. It fully migrates and covers historical inspection data processing results (e.g., stored defect physical dimensions and image association markers), incrementally capturing new content added in real-time data processing, such as real-time generated pseudo-color image paths and newly detected defect annotation data, ensuring continuity and consistency in cross-database scenarios. Furthermore, data integrity verification after migration further ensures the reliability of subsequent secondary data processing (e.g., defect tracing and multi-dimensional analysis), achieving efficient end-to-end support for inspection data processing from storage management to cross-database expansion.

[0063] In another embodiment, this application also provides a data processing system 200, which may include: an infrared image analysis module 10, a multimodal synchronous calibration module 20, a defect quantification analysis module 30, and an associated storage module 40; wherein the infrared image analysis module 10 is used to analyze infrared images in an inspection scenario to obtain the image resolution, temperature matrix, and pseudo-color image of the infrared images; the multimodal synchronous calibration module 20 is used to synchronize the annotation information of the infrared images and the visible light images in the inspection scenario to obtain multimodal annotation data; the defect quantification analysis module 30 is used to perform defect quantification analysis on the multimodal annotation data based on preset reference object information to generate a quantification result of the true physical size of the defects; and the associated storage module 40 is used to associate and store the image resolution, temperature matrix, pseudo-color image, and quantification result.

[0064] The data processing system of this application can achieve full-process optimization of multimodal inspection data parsing, quantitative analysis and storage migration by integrating the infrared image analysis module 10, multimodal synchronous calibration module 20, defect quantification analysis module 30 and associated storage module 40, and by integrating thread resource allocation mechanism, multimodal annotation collaboration and standardized data management scheme; it can adapt to the processing needs of large-scale and multi-dimensional inspection data in industrial scenarios, and effectively improve the overall reliability and performance of the system.

[0065] As an example, the data processing system of this application can be used to perform, for example... Figure 1 And the method for processing intelligent inspection data in related embodiments.

[0066] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features of the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0067] The above embodiments merely illustrate several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for processing intelligent inspection data, characterized in that, The method for processing the intelligent inspection data includes: Analyze infrared images in an inspection scenario to obtain the image resolution, temperature matrix, and pseudo-color image of the infrared images; Synchronize the annotation information of the infrared image and the visible light image in the inspection scenario to obtain multimodal annotation data; Based on preset reference object information, defect quantification analysis is performed on the multimodal annotation data to generate a quantification result of the true physical size of the defect. This includes: using an AI model to identify defects in the multimodal annotation data to obtain the bounding box information and defect type of the defect; obtaining the pixel size of the defect in visible light and infrared images based on the bounding box information; determining the actual physical size of the defect in both visible light and infrared image dimensions based on the reference object information and the pixel size; and generating a quantification result of the true physical size of the defect based on the actual physical size in both image dimensions. The preset reference object information includes the actual physical size of the reference object and its position in the visible light image. The reference object information is obtained by: associating a preset reference object with the visible light image and the infrared image through a three-level binding mechanism; wherein, the three-level binding mechanism includes data source level binding, data level binding, and default level binding; matching the reference object associated with the visible light image and the infrared image where the defect is located based on the priority order of the data source level, data level, and default level binding to obtain the successfully matched reference object information; calculating the visible light image scale and the infrared image scale respectively using the actual physical size of the successfully matched reference object and its pixel size in the visible light image and the infrared image; and converting the pixel size of the defect into its actual physical size using the visible light image scale and the infrared image scale. The image resolution, temperature matrix, pseudocolor image, and quantization result are stored together.

2. The method according to claim 1, characterized in that, The analysis of infrared images in the inspection scenario yields the image resolution, temperature matrix, and pseudo-color image of the infrared images, including: This paper describes how to encapsulate the infrared thermal imaging processing TSDK using JNA technology and build a Java-based parser for R-JPEG format data. A channel is used to initialize a separate parsing context containing a TSDK handle and an image pointer; Based on the independent parsing context, the resolution extraction interface of the TSDK is called to read the image resolution of the infrared thermal imaging; Based on the independent parsing context, the temperature matrix of the infrared thermal image is extracted by calling the temperature matrix generation interface of the TSDK. The system receives pseudo-color mode parameters and, based on the independent parsing context, calls the pseudo-color conversion interface of the TSDK to convert the infrared thermal image and generate the pseudo-color image.

3. The method according to claim 2, characterized in that, Before parsing the infrared image in the inspection scenario, the method further includes: using the ThreadLocal mechanism to bind the independent parsing context to the current processing thread.

4. The method according to claim 1, characterized in that, The synchronization of annotation information between the infrared image and the visible light image in the inspection scenario yields multimodal annotation data, including: Feature points are extracted from the visible light image and the infrared image respectively; The feature points of the visible light image and the infrared image are matched to obtain the matching result; Based on the matching results, the homography matrix between the visible light image and the infrared image is determined; Based on the homography matrix, the annotation coordinates in the visible light image annotation data are mapped to the infrared image to obtain the infrared image annotation coordinates; Based on the infrared image annotation coordinates, and the annotation type and defect attributes in the visible light image annotation data, infrared image annotation data corresponding to the visible light image annotation data is generated; The visible light image annotation data is associated with the infrared image annotation data to generate the multimodal annotation data.

5. The method according to claim 4, characterized in that, The matching of feature points between the visible light image and the infrared image to obtain a matching result includes: Filter out feature points with low response values ​​from the features of the visible light image and infrared image to obtain high-quality feature points; The FLANN matcher is used to match high-quality feature points of the visible light image and infrared image to obtain high-quality matching pairs whose distance meets the preset conditions, and the matching result is generated.

6. The method according to claim 1, characterized in that, The step of associating and storing the image resolution, temperature matrix, pseudocolor image, and quantization result includes: A unique association identifier is assigned to the image resolution, temperature matrix, pseudo-color image, and quantization result under the same inspection scenario; The visible light image, infrared image, and pseudo-color image are stored on a file server and the storage path is recorded. Bind the storage path to the unique associated identifier and associate the storage.

7. The method according to claim 1, characterized in that, The method for processing the intelligent inspection data also includes: The Spring Boot Starter framework is used for database access, caching, and access control, and the MySQL to PostgreSQL intelligent migration component is used to store and migrate business data related to the inspection scenario.

8. A data processing system, characterized in that, The data processing system includes: The infrared image analysis module is used to analyze infrared images in the inspection scenario to obtain the image resolution, temperature matrix and pseudo-color image of the infrared image. The multimodal synchronous calibration module is used to synchronize the annotation information of the infrared image and the visible light image in the inspection scene to obtain multimodal annotation data; The defect quantification analysis module is used to perform defect quantification analysis on the multimodal annotation data based on preset reference object information, and generate a quantification result of the true physical size of the defect. This includes: using an AI model to identify defects in the multimodal annotation data to obtain the bounding box information and defect type of the defect; obtaining the pixel size of the defect in visible light and infrared images based on the bounding box information; determining the actual physical size of the defect in both visible light and infrared image dimensions based on the reference object information and the pixel size; and generating a quantification result of the true physical size of the defect based on the actual physical size in both image dimensions. The preset reference object information includes the actual physical size of the reference object and... The pixel dimensions in visible light and infrared images; the method for obtaining the reference object information includes: associating a preset reference object with the visible light and infrared images through a three-level binding mechanism; wherein, the three-level binding mechanism includes data source level binding, data level binding, and default level binding; matching the reference objects associated with the visible light and infrared images where the defect is located based on the priority order of the data source level, data level, and default level to obtain the successfully matched reference object information; using the actual physical size of the successfully matched reference object and its pixel dimensions in the visible light and infrared images, calculating the visible light image scale and infrared image scale respectively; using the visible light image scale and infrared image scale, converting the pixel dimensions of the defect into actual physical dimensions; The associated storage module is used to associate and store the image resolution, temperature matrix, pseudo-color image, and quantization result.

Images