Efficient map element quality inspection AI system and implementation method
The map quality inspection AI system, which combines multimodal data with deep learning, solves the problems of low efficiency in traditional quality inspection methods and limited functionality in automated quality inspection tools. It achieves efficient and accurate quality inspection of multiple types of maps, has self-learning capabilities and one-click rectification functions, and improves the quality of map data.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHINESE PEOPLES LIBERATION ARMY UNIT 61206
- Filing Date
- 2026-01-29
- Publication Date
- 2026-05-12
AI Technical Summary
Traditional map quality inspection methods are inefficient and susceptible to human factors, failing to meet the demands for large-scale, high-efficiency, and high-precision quality inspection. Furthermore, existing automated quality inspection tools have limited functionality and lack the ability to jointly verify multiple types of map elements.
The system employs a high-efficiency map quality inspection AI system that combines multimodal data with deep learning and knowledge reasoning. It includes a multi-source data access module, an AI quality inspection model training module, a multi-element intelligent verification module, and a visualization report and rectification module. It supports joint verification of multiple types of map elements and automatic generation of dynamic rule bases.
It achieves efficient, accurate, and intelligent map quality inspection, significantly shortens the quality inspection cycle, meets the timeliness requirements of map production and updates, has self-learning capabilities and can flexibly adjust quality inspection rules, improves the accuracy and reliability of quality inspection, and provides intuitive quality inspection reports and one-click rectification functions.
Smart Images
Figure CN122019679A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of map quality inspection system technology, and in particular to an efficient map quality inspection AI system and its implementation method. Background Technology
[0002] In the process of map creation and updating, the quality of map element data is crucial. Traditional map quality inspection methods mainly rely on manual checks, which are not only inefficient but also susceptible to human factors, making it difficult to guarantee the consistency and accuracy of the inspection results. With the rapid development of GIS technology, the volume of map element data has increased dramatically, and the data types have become increasingly complex. Traditional quality inspection methods can no longer meet the needs of large-scale, high-efficiency, and high-precision quality inspection.
[0003] Currently, while some automated quality inspection tools exist, most are single-function, capable of simple verification only for specific types of map features or specific error types. They lack the ability to jointly verify multiple types of map features, and their quality inspection rules are fixed, making it difficult to adapt to different application scenarios and changes in map publishing / production standards. Therefore, a highly efficient AI system for map feature quality inspection is needed. Summary of the Invention
[0004] To address the problems in the background technology, this invention provides an efficient map quality inspection AI system and implementation method, which can automatically identify eight common error types in map drawing. Through multimodal data, combined with deep learning and knowledge reasoning, it achieves efficient, accurate, and intelligent map quality inspection.
[0005] This invention provides an efficient map quality inspection AI system, specifically comprising: The multi-source data access module is used to import various types of map feature data and unify the spatial reference system; The AI quality inspection model training module includes a spatial topology feature extraction network and a quality inspection rule self-learning sub-model, which are used to generate a dynamic rule base. The multi-element intelligent verification module integrates topology verification units, attribute verification units, and association verification units for joint verification of multiple types of map elements. The visualization report and rectification module is used to generate structured quality inspection reports and supports one-click rectification. The system management module is used for user permission management and model iteration and maintenance.
[0006] Furthermore, the multi-source data access module supports the import of mainstream GIS vector data formats such as SHP, GDB, GeoJSON, and PDF, as well as satellite raster base map data. It has a built-in automatic spatial reference system conversion unit, which can uniformly convert map elements of different coordinate systems into the target coordinate system, including CGCS2000 and WGS84.
[0007] Furthermore, the spatial topology feature extraction network is based on the improved PointNet++ algorithm, which introduces a GIS spatial neighborhood relationship weight factor to encode the node coordinates, geometric shape, and adjacent element relationships of map elements in a high-dimensional way. The quality inspection rule self-learning sub-model adopts a reinforcement learning framework, uses historical quality inspection annotation data as training samples, and uses compliance indicators of different map publishing / production standards as reward functions to automatically generate and iteratively update a dynamic verification rule library that adapts to multiple scenarios.
[0008] Furthermore, the topology verification unit is configured with a node micro-fracture recognition threshold, an element overlap area threshold, and a self-intersection judgment algorithm, which can automatically identify implicit topological errors such as node fractures, geometric overlaps, and self-intersections of map elements. The attribute verification unit can check the completeness of the required fields of map elements, the compliance of field value ranges, and the consistency of attribute associations based on a dynamic rule base. The attribute associations include the matching relationship between road grades and speed limits, and the correspondence between bridges across rivers and water system grades. The association verification unit integrates spatial overlay analysis algorithms and attribute matching algorithms, and can conduct joint verification of spatial location rationality and attribute association for multiple element combinations such as road-water system, road-traffic appurtenances, and water system-administrative division.
[0009] Furthermore, the visualization report and rectification module supports generating quality inspection reports in three formats: PDF, Excel, and GIS vector layer. The reports include spatial location annotations of erroneous elements, error type classifications, and rectification suggestions. It is also equipped with a one-click rectification unit, which can automatically correct common errors such as node micro-fractures and non-standard field formats.
[0010] Furthermore, the system management module includes a task log recording unit and a model iteration triggering unit. The task log recording unit stores the data source information, verification rule version, and error correction records for each quality inspection. The model iteration triggering unit can automatically start the iterative training process of the AI quality inspection model based on manually corrected feedback data.
[0011] A method for implementing an efficient map quality inspection AI system includes the following steps: S1. Data Preprocessing: Import map feature data through the multi-source data access module to complete data format parsing, spatial reference system conversion, duplicate feature removal and invalid geometry cleaning; S2.AI Quality Inspection Model Training: Construct a training dataset containing samples of topological errors, attribute errors, and relationship errors. Use an improved PointNet++ algorithm to extract spatial topological features. Train a self-learning sub-model of quality inspection rules through reinforcement learning to generate a dynamic rule base. S3. Multi-element intelligent verification: Input the pre-processed map element data into the pre-trained model, and perform topology verification, attribute verification, and association verification in sequence. Integrate the verification results and mark the unique identifier and spatial location of the erroneous elements. S4. Report Generation and Rectification: The system outputs structured quality inspection reports through a visual report and rectification module, supports one-click correction of common errors, and the rectified data is re-entered into the system for re-inspection; S5. Model Iteration and Optimization: Input the re-inspection results and manual correction feedback data into the AI quality inspection model training module to continuously optimize the model parameters and dynamic rule base.
[0012] Furthermore, the process of constructing the training dataset in step S2 includes collecting map quality inspection data of different scales and application scenarios, and performing three-level annotation on error samples. The three-level annotation includes error type, error severity, and rectification priority.
[0013] Furthermore, the specific process of verifying the association relationship in step S3 is as follows: first, the spatial positional relationship between the two types of elements is determined by the spatial overlay analysis algorithm; then, the consistency of the association fields between the elements is checked by the attribute matching algorithm; and finally, the association relationship is judged to be compliant by comparing with the dynamic rule base.
[0014] Furthermore, the model iteration optimization in step S5 adopts an incremental training mode, which only updates the parameters of the feature dimensions corresponding to the newly added feedback data, without the need to retrain the model, thus greatly shortening the iteration cycle.
[0015] The present invention provides an efficient map quality inspection AI system and implementation method, which has the following beneficial effects: This invention is an intelligent map error detection tool based on a multimodal large model. It can automatically identify eight common error types in map drawing. Through multimodal data, combined with deep learning and knowledge reasoning, it achieves efficient, accurate and intelligent map quality inspection. Compared with traditional manual quality inspection methods, it can process large-scale map element data in a short time, significantly shorten the quality inspection cycle, and meet the timeliness requirements of map production, updating and other processes.
[0016] In addition, the system integrates multiple functions such as multi-source data access, topology verification, attribute verification, and relationship verification, enabling comprehensive and joint verification of various types of map elements. It can not only detect common topology errors and attribute errors, but also accurately identify complex relationship errors, ensuring that map element data meets quality requirements in all aspects.
[0017] In addition, through the AI quality inspection model training module, the system can automatically generate and iteratively update the dynamic rule base, enabling the quality inspection rules to be flexibly adjusted according to different application scenarios and changes in map publishing / production standards. At the same time, the system has self-learning capabilities, which can continuously absorb new data features and error patterns, continuously optimize its own performance, and improve the accuracy and reliability of quality inspection.
[0018] In addition, the visualization report and rectification module provides users with intuitive and detailed quality inspection reports, supporting multiple output formats for easy viewing and sharing; the one-click rectification function greatly simplifies the rectification process, improves rectification efficiency, and enables users to quickly correct errors in map features and improve data quality.
[0019] In addition, the system management module implements user permission management and model iteration maintenance functions, ensuring the security and stability of the system. The task log recording unit can record relevant information for each quality inspection in detail, providing data support for system auditing and model optimization; the model iteration triggering unit can automatically start the model iteration training process based on manual correction feedback data, enabling the system to continuously optimize and improve quality inspection performance. Attached Figure Description
[0020] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings of the embodiments will be briefly described below.
[0021] The accompanying drawings described below are only related to some embodiments of the invention and are not intended to limit the invention.
[0022] In the attached diagram: Figure 1 A flowchart illustrating the system usage steps of the present invention is shown. Detailed Implementation
[0023] To make the objectives, solutions, and advantages of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Unless otherwise stated, the terms used herein have their ordinary meanings in the art. The same reference numerals in the drawings represent the same parts.
[0024] Example 1: This invention proposes an efficient map quality inspection AI system, comprising: The multi-source data access module is used to import various types of map feature data and unify the spatial reference system. The multi-source data access module supports the import of mainstream GIS vector data formats such as SHP, GDB, GeoJSON, and PDF, as well as satellite raster base map data. It has a built-in automatic spatial reference system conversion unit, which can convert map features in different coordinate systems into the target coordinate system, including CGCS2000 and WGS84. The AI quality inspection model training module includes a spatial topology feature extraction network and a quality inspection rule self-learning sub-model, used to generate a dynamic rule base. The spatial topology feature extraction network is based on an improved PointNet++ algorithm, introducing a GIS spatial neighborhood relationship weight factor to encode the node coordinates, geometric shape, and adjacent element relationships of map elements in a high-dimensional feature. The quality inspection rule self-learning sub-model adopts a reinforcement learning framework, using historical quality inspection annotation data as training samples and compliance indicators of different map publishing / production standards as reward functions, to automatically generate and iteratively update a dynamic verification rule base adapted to multiple scenarios. The multi-element intelligent verification module integrates a topology verification unit, an attribute verification unit, and a relationship verification unit for joint verification of multiple types of map elements. The topology verification unit is configured with a node micro-fracture recognition threshold, an element overlap area threshold, and a self-intersection judgment algorithm, which can automatically identify implicit topological errors such as node fractures, geometric overlaps, and self-intersections in map elements. The attribute verification unit can check the completeness of required fields of map elements, the compliance of field value ranges, and the consistency of attribute relationships based on a dynamic rule base. Attribute relationships include the matching relationship between road grades and speed limits, and the correspondence between bridges across rivers and water system grades. The relationship verification unit integrates spatial overlay analysis algorithms and attribute matching algorithms, and can conduct joint verification of spatial location rationality and attribute relationships for multiple element combinations such as road-water system, road-traffic appurtenances, and water system-administrative division. The Visual Reporting and Rectification module is used to generate structured quality inspection reports and supports one-click rectification. The Visual Reporting and Rectification module supports generating quality inspection reports in three formats: PDF, Excel, and GIS vector layer. The reports include spatial location annotations of erroneous elements, error type classifications, and rectification suggestions. It also has a one-click rectification unit that can automatically correct common errors such as node micro-fractures and non-standard field formats. The system management module is used for user permission management and model iteration maintenance. The system management module includes a task log recording unit and a model iteration triggering unit. The task log recording unit stores the data source information, verification rule version, and error correction records for each quality inspection. The model iteration triggering unit can automatically start the iterative training process of the AI quality inspection model based on manually corrected feedback data.
[0025] In this embodiment of the invention, the implementation method of the high-efficiency map quality inspection AI system specifically includes the following steps: S1. Data Preprocessing: Import map feature data through the multi-source data access module to complete data format parsing, spatial reference system conversion, duplicate feature removal and invalid geometry cleaning; S2. AI Quality Inspection Model Training: Construct a training dataset containing samples of topological errors, attribute errors, and relationship errors. Use an improved PointNet++ algorithm to extract spatial topological features. Train a self-learning sub-model of quality inspection rules through reinforcement learning to generate a dynamic rule base. The process of constructing the training dataset includes collecting map quality inspection data of different scales and application scenarios, and performing three-level annotation on error samples. The three-level annotation includes error type, error severity, and rectification priority. S3. Multi-element intelligent verification: Input the pre-processed map element data into the pre-trained model, and perform topology verification, attribute verification, and association verification in sequence. Integrate the verification results and mark the unique identifier and spatial location of the erroneous elements. The specific process of association verification is as follows: First, the spatial positional relationship between the two types of elements is determined by the spatial overlay analysis algorithm. Then, the consistency of the association fields between the elements is checked by the attribute matching algorithm. Finally, the association relationship is judged to be compliant by comparing with the dynamic rule base. S4. Report Generation and Rectification: The system outputs structured quality inspection reports through a visual report and rectification module, supports one-click correction of common errors, and the rectified data is re-entered into the system for re-inspection; S5. Model Iteration Optimization: Input the re-inspection results and manual correction feedback data into the AI quality inspection model training module to continuously optimize model parameters and dynamic rule base; the model iteration optimization adopts incremental training mode, only updating the parameters of the feature dimensions corresponding to the newly added feedback data, without the need for full retraining of the model, which greatly shortens the iteration cycle.
[0026] In embodiments of the present invention, such as Figure 1 As shown, the detailed steps for using the high-efficiency map quality inspection AI system are as follows: Step 1: Select an error detector: Select the type of error to detect from the "Error Detector" drop-down menu; After selection, the system will display two sample images for this error type: Correct example: Demonstrates a drawing method that conforms to the specifications; Incorrect example: Displaying a drawing method that does not conform to the specifications; Step 2: Select the image to be detected: Click the "Select this image" button below the sample image to select the image you want to detect (you can select from either the correct or incorrect examples); After selection, the image will be displayed in the preview area; Step 3: Start the test: Click the "Start Detection" button, and the system will: 1. Use an AI model for detection; 2. Analyze the content of the image; 3. Generate detection results; Step 4: View the results: After the test is completed, the results will be displayed in the "Test Results" area at the bottom of the page; The test results are in Markdown format and include: Observation: Description of the content of the image; Analysis: Standard-based error analysis; Conclusion: Final judgment (correct / incorrect) and recommendations; Detector file location; The trained detector files are saved in the pkls / directory: pkls / ├── detector_qwen_railway_overlap.pkl ├── detector_qwen_road_river_bridge.pkl ├── detector_qwen_house_road_overlap.pkl ├── detector_qwen_road_intersection.pkl ├── detector_qwen_greenhouse_boundary.pkl ├── detector_qwen_street_intersection_overlay.pkl ├── detector_qwen_elevation_label_overlap.pkl └── detector_qwen_intersection_elevation_point.pkl In this embodiment of the invention, eight common error types in map drawing can be automatically identified, as follows:
[0027] The following points should be noted in this article: 1. The accompanying drawings of the embodiments of the present invention only involve the structures involved in the embodiments of the present invention; other structures can refer to general designs.
[0028] 2. Where there is no conflict, the embodiments of the present invention and the features thereof can be combined with each other to obtain new embodiments.
[0029] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.
Claims
1. A high-efficiency map quality inspection AI system, characterized in that, include: The multi-source data access module is used to import various types of map feature data and unify the spatial reference system; The AI quality inspection model training module includes a spatial topology feature extraction network and a quality inspection rule self-learning sub-model, which are used to generate a dynamic rule base. The multi-element intelligent verification module integrates topology verification units, attribute verification units, and association verification units for joint verification of multiple types of map elements. The visualization report and rectification module is used to generate structured quality inspection reports and supports one-click rectification. The system management module is used for user permission management and model iteration and maintenance.
2. The efficient map quality inspection AI system according to claim 1, characterized in that, The multi-source data access module supports the import of mainstream GIS vector data formats such as SHP, GDB, GeoJSON, and PDF, as well as satellite raster base map data. It has a built-in automatic spatial reference system conversion unit, which can uniformly convert map elements in different coordinate systems into the target coordinate system, including CGCS2000 and WGS84.
3. The efficient map quality inspection AI system according to claim 1, characterized in that, The spatial topology feature extraction network is based on the improved PointNet++ algorithm and introduces a GIS spatial neighborhood relationship weight factor to perform high-dimensional feature encoding on the node coordinates, geometric shape, and adjacent feature relationships of map elements. The quality inspection rule self-learning sub-model adopts a reinforcement learning framework, using historical quality inspection annotation data as training samples and compliance indicators of different map publishing / production standards as reward functions to automatically generate and iteratively update a dynamic verification rule library that adapts to multiple scenarios.
4. The efficient map quality inspection AI system according to claim 1, characterized in that, The topology verification unit is equipped with a node micro-fracture recognition threshold, an element overlap area threshold, and a self-intersection judgment algorithm, which can automatically identify implicit topology errors such as node fractures, geometric overlaps, and self-intersections of map elements. The attribute verification unit can check the completeness of the required fields of map elements, the compliance of field value ranges, and the consistency of attribute associations based on a dynamic rule base. The attribute associations include the matching relationship between road grade and speed limit value, and the correspondence between bridges across rivers and water system grades. The association verification unit integrates spatial overlay analysis algorithm and attribute matching algorithm, and can conduct joint verification of spatial location rationality and attribute association for multiple element combinations such as road-water system, road-traffic appurtenances, and water system-administrative division.
5. The efficient map quality inspection AI system according to claim 1, characterized in that, The visualization report and rectification module supports generating quality inspection reports in three formats: PDF, Excel, and GIS vector layer. The reports include spatial location annotations of erroneous elements, error type classifications, and rectification suggestions. It also has a one-click rectification unit that can automatically correct common errors such as node micro-fractures and non-standard field formats.
6. The efficient map quality inspection AI system according to claim 1, characterized in that, The system management module includes a task log recording unit and a model iteration triggering unit. The task log recording unit stores the data source information, verification rule version, and error correction records for each quality inspection. The model iteration triggering unit can automatically start the iterative training process of the AI quality inspection model based on manually corrected feedback data.
7. The implementation method of an efficient map quality inspection AI system according to claims 1-6, characterized in that, Includes the following steps: S1. Data Preprocessing: Import map feature data through the multi-source data access module to complete data format parsing, spatial reference system conversion, duplicate feature removal and invalid geometry cleaning; S2.AI Quality Inspection Model Training: Construct a training dataset containing samples of topological errors, attribute errors, and relationship errors. Use an improved PointNet++ algorithm to extract spatial topological features. Train a self-learning sub-model of quality inspection rules through reinforcement learning to generate a dynamic rule base. S3. Multi-element intelligent verification: Input the pre-processed map element data into the pre-trained model, and perform topology verification, attribute verification, and association verification in sequence. Integrate the verification results and mark the unique identifier and spatial location of the erroneous elements. S4. Report Generation and Rectification: The system outputs structured quality inspection reports through a visual report and rectification module, supports one-click correction of common errors, and the rectified data is re-entered into the system for re-inspection; S5. Model Iteration and Optimization: Input the re-inspection results and manual correction feedback data into the AI quality inspection model training module to continuously optimize the model parameters and dynamic rule base.
8. The implementation method of an efficient map quality inspection AI system according to claim 7, characterized in that, The process of constructing the training dataset in step S2 includes collecting map quality inspection data of different scales and application scenarios, and performing three-level annotation on error samples. The three-level annotation includes error type, error severity, and rectification priority.
9. The implementation method of an efficient map quality inspection AI system according to claim 7, characterized in that, The specific process of verifying the association relationship in step S3 is as follows: First, the spatial positional relationship between the two types of elements is determined by the spatial overlay analysis algorithm; then, the consistency of the association fields between the elements is checked by the attribute matching algorithm; and finally, the association relationship is judged to be compliant by comparing with the dynamic rule base.
10. The implementation method of an efficient map quality inspection AI system according to claim 7, characterized in that, The model iteration optimization in step S5 adopts an incremental training mode, which only updates the parameters of the feature dimensions corresponding to the newly added feedback data, without the need for full retraining of the model, thus greatly shortening the iteration cycle.