Tunnel disease judgment method and system based on large model identification of tunnel point cloud data

Abstract

This invention proposes a method and system for judging tunnel defects based on large-scale model identification of tunnel point cloud data, belonging to the field of tunnel engineering inspection. The method includes: acquiring the original 3D point cloud data of the tunnel and preprocessing it into a standardized point cloud dataset; segmenting the standardized point cloud dataset and constructing image multimodal data pairs; inputting the image multimodal data pairs and dedicated prompting instructions into the constructed defect detection large-scale model to obtain preliminary identification and evaluation results; mapping the results back to the original 3D point cloud coordinate system through a view projection matrix to construct a 3D defect point cloud; constructing a temporal-spatial knowledge graph based on the 3D defect point cloud, generating early warning information when preset risk conditions are met; and generating tunnel defect detection results based on the 3D defect point cloud and early warning information. This invention achieves full-process automation and intelligence in tunnel defect detection, significantly improving detection efficiency and accuracy.

Description

Technical Field

[0001] This invention belongs to the field of tunnel engineering detection technology, and in particular relates to a method and system for judging tunnel defects based on large model identification of tunnel point cloud data. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] As a core infrastructure in transportation, water conservancy, energy, and other fields, the structural integrity and operational safety of tunnels directly affect public safety and engineering benefits, and are a key link in ensuring the stable operation of various engineering systems. With the continuous expansion of tunnel construction scale and the extension of operating life, tunnel structures are prone to various defects such as cracks, water leakage, lining deformation, and spalling. Therefore, accurate and efficient detection and evaluation have become an important requirement in the engineering field.

[0004] Point cloud data, with its high-precision and high-fidelity three-dimensional spatial information representation capabilities, can comprehensively capture the geometric morphology and surface features of tunnel structures, and has become the core data carrier for tunnel inspection. It is typically acquired through equipment such as LiDAR and 3D scanning. Currently, tunnel point cloud data processing and judgment methods mainly rely on traditional machine learning algorithms (such as SVM and random forest) or manual interactive analysis. These methods have played a certain role in the initial stages of tunnel defect detection, but with the increase in engineering complexity and the upgrading of inspection requirements, their limitations are becoming increasingly prominent, specifically manifested in the following ways: First, traditional machine learning algorithms require the construction of separate feature systems for different types of tunnel defects such as cracks, water leakage, and lining deformation. This is not only time-consuming and labor-intensive, but also results in extremely poor model generalization ability, making it unable to adapt to diverse tunnel scenarios with varying geological conditions, construction techniques, and operating environments. Second, the complex internal environment of tunnels, variable lighting conditions, noise interference during equipment data acquisition, and obstructions from pipelines and ancillary facilities all contribute to decreased detection accuracy. Especially in high-noise or sparse point cloud data scenarios, the accuracy of traditional methods is difficult to guarantee, easily leading to missed detections and false positives. Furthermore, when new types of defects emerge in engineering projects or the detection standards for existing defects change, traditional machine learning models require redesigning features, adjusting network structures, and undergoing full retraining. This process is time-consuming and costly, making it difficult to quickly respond to actual engineering needs and severely limiting the flexibility and scalability of the method. Moreover, they face the problems of strong subjectivity and high rates of missed and false positives, especially in the inspection of long tunnels, making it difficult to meet the needs of real-time processing of large-scale data. Finally, although large models have made progress in related fields, when applied to tunnel point cloud data, they face technical bottlenecks such as unstructured point cloud data, high spatial dimensionality, and complex semantic information of tunnel scenes. They also lack a mature and complete intelligent judgment mechanism from data preprocessing to result output, which limits the application of large models in the field of tunnel engineering inspection. Summary of the Invention

[0005] To overcome the shortcomings of the existing technologies, this invention provides a method and system for judging tunnel defects based on large model identification of tunnel point cloud data. By introducing a large point cloud model based on the Transformer architecture, the entire process from data preprocessing to anomaly judgment is automated, overcoming the limitations of traditional methods and improving the accuracy and efficiency of defect identification. This provides an efficient and reliable intelligent judgment mechanism for tunnel construction monitoring and operation and maintenance.

[0006] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions: The first aspect of this invention provides a method for judging tunnel defects based on large model identification of tunnel point cloud data; A method for identifying tunnel defects based on large model-based tunnel point cloud data includes: Obtain the raw three-dimensional point cloud data of the tunnel, and preprocess the raw point cloud data to obtain a standardized point cloud dataset; The standardized point cloud dataset is segmented and image multimodal data pairs are constructed; A large defect detection model integrating the SIDNet model and a set of dedicated prompts for tunnel defect identification and assessment are constructed. The image multimodal data pairs and the dedicated prompts are input into the large defect detection model to obtain preliminary identification and assessment results. The obtained preliminary identification and evaluation results are mapped back to the original three-dimensional point cloud coordinate system through the view projection matrix to construct a three-dimensional defect point cloud with detailed disease semantic labels. Based on the three-dimensional defect point cloud, a temporal-spatial knowledge graph is constructed to analyze the coupling relationship and spatial clustering characteristics between different disease types within a preset continuous segment. When preset risk conditions are met, early warning information is generated. Based on the three-dimensional defect point cloud and early warning information, tunnel defect detection results are generated.

[0007] As a further technical solution, the original point cloud data is preprocessed, including statistical filtering for noise reduction, voxel downsampling, ICP registration, and mileage marker labeling.

[0008] As a further technical solution, the standardized point cloud dataset is segmented and image multimodal data pairs are constructed, including: The standardized point cloud dataset is segmented according to a preset vertical interval; For each segment of point cloud, a high-resolution depth map and a pseudo-color map are generated by rendering from multiple preset perspectives. The depth map uses grayscale mapping to represent distance, and the pseudo-color map is based on intensity or color channel mapping. Construct point cloud-multi-view image multimodal data pairs based on point cloud data, depth maps, and pseudo-color maps.

[0009] As a further technical solution, a pseudo-color image is generated for each segment of the point cloud, including: The segmented point cloud is projected onto a two-dimensional image plane from a preset viewpoint, and the occlusion relationship is handled by using depth buffering. Extract the laser return intensity value or RGB color channel of each point in the point cloud as a scalar source; The scalar source is normalized. A preset pseudo-color mapping table is used to map normalized scalar values ​​to the RGB color space, creating a high-contrast gradient effect. The projected pixels are interpolated and filled to generate a high-resolution pseudo-color image.

[0010] As a further technical solution, the image multimodal data pairs and the dedicated prompting instructions are input into the defect detection large model to obtain preliminary identification and evaluation results, including: Visual feature vectors are obtained by extracting features from depth maps and pseudo-color maps using a multimodal encoder. The dedicated prompt instructions are converted into prompt embeddings using a text encoder; The calculation formula is as follows: Visual features and cue embeddings are fused using a cross-attention mechanism.

[0011] Where Q represents the query vector, which is derived from the hint embedding, and K and V represent the key vector and value vector, respectively, which are derived from the visual feature vector; The decoder outputs preliminary identification results, including defect segmentation mask, convergent deformation measurement value, and risk assessment score.

[0012] As a further technical solution, the obtained preliminary identification and evaluation results are mapped back to the original 3D point cloud coordinate system through a view projection matrix to construct a 3D defect point cloud with detailed disease semantic labels, including: The projection matrix of each view is recorded during the rendering phase; For the defect image, the depth value is obtained, and the corresponding world coordinates are calculated using the inverse projection formula; Collect multi-view labels, assign them to the original point cloud through majority voting or distance-weighted fusion, add semantic labels to each point, remove isolated noise points and merge connected components.

[0013] As a further technical solution, based on the three-dimensional defect point cloud, a temporal-spatial knowledge graph is constructed to analyze the coupling relationship and spatial clustering characteristics between different defect types within a preset continuous segment. When preset risk conditions are met, early warning information is generated, including: The disease instances in the 3D defect point cloud are used as nodes in the knowledge graph. Each node's attributes include disease type, mileage location coordinates, severity quantification value, and individual disease risk score. Edges between nodes are constructed based on temporal and spatial relationships, and coupling edges are added according to domain rules; Extract subgraphs that cover a preset continuous paragraph, and perform correlation analysis and risk cluster identification on the subgraphs; When the density of a sub-map exceeds a preset threshold or the total risk score exceeds a preset threshold, it is identified as a risk cluster area, and an early warning message is automatically generated.

[0014] The second aspect of this invention provides a tunnel defect judgment system based on large model identification of tunnel point cloud data.

[0015] A tunnel defect assessment system based on large-scale model identification of tunnel point cloud data includes: The data acquisition and preprocessing module is configured to: acquire the raw three-dimensional point cloud data of the tunnel, preprocess the raw point cloud data, and obtain a standardized point cloud dataset; The point cloud multi-scale segmentation and multi-view rendering module is configured to: segment the standardized point cloud dataset and construct image multimodal data pairs; The large model inference module is configured to: construct a large defect detection model integrating the SIDNet model and a set of dedicated prompts for tunnel defect identification and assessment; input the image multimodal data pairs and the dedicated prompts into the large defect detection model to obtain preliminary identification and assessment results; The three-dimensional back projection module is configured to: map the obtained preliminary identification and evaluation results back to the original three-dimensional point cloud coordinate system through the view projection matrix, and construct a three-dimensional defect point cloud with detailed disease semantic labels; The intelligent judgment and early warning module is configured to: construct a time-space knowledge graph based on the three-dimensional defect point cloud, analyze the coupling relationship and spatial clustering characteristics between different disease types in a preset continuous segment, and generate early warning information when preset risk conditions are met; The detection result output module is configured to generate tunnel defect detection results based on the three-dimensional defect point cloud and early warning information.

[0016] A third aspect of the present invention provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the steps of the tunnel defect judgment method based on large model identification tunnel point cloud data as described in the first aspect of the present invention.

[0017] A fourth aspect of the present invention provides an electronic device, including a memory, a processor, and a program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps in the tunnel defect judgment method based on large model identification tunnel point cloud data as described in the first aspect of the present invention.

[0018] The above one or more technical solutions have the following beneficial effects: This invention employs a multimodal large-scale model architecture integrating the SIDNet model, fully leveraging the powerful feature self-learning capabilities of the large-scale model and the high accuracy of the SIDNet model in segmenting spalling defects in tunnel lining. Based on dual-channel input of laser intensity and depth, the SIDNet model can accurately capture the feature information of spalling defects. Combined with the comprehensive analysis capabilities of the multimodal large-scale model for complex data, it effectively overcomes the shortcomings of traditional methods, such as strong reliance on manual feature engineering, weak generalization ability, and poor adaptability to complex environments. This significantly improves the accuracy and reliability of the detection results, providing precise data support for tunnel structural condition assessment.

[0019] This invention, through its Prompt design, guides large models to rapidly adapt to the detection needs of new defects under zero-sample or few-sample conditions, eliminating the need for full retraining of the entire model. This significantly reduces model upgrade costs and enables rapid response to dynamic needs such as changes in defect types and adjustments to detection standards in actual engineering projects. Simultaneously, by constructing a temporal-spatial knowledge graph, it can correlate multiple defects such as cracks, water leakage, lining deformation, and spalling, accurately identifying risk-accumulating areas and automatically triggering high-level early warnings. This allows for the early detection of potential safety hazards in tunnel structures, avoiding misjudgments caused by detecting a single defect.

[0020] This invention constructs a fully automated processing system from data acquisition and preprocessing to report generation. Combining a multi-segment, multi-disease coupled intelligent judgment mechanism with a one-click visual report generation function, it achieves highly efficient operation of the detection process. Through a time-series-spatial knowledge graph, it can automatically complete the correlation analysis and risk cluster identification of multiple diseases within continuous segments, triggering high-level early warnings without manual intervention. Simultaneously, it can directly output three-dimensional defect heatmaps, disease lists, and standardized reports conforming to technical specifications, improving detection efficiency.

[0021] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0022] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0023] Figure 1 This is a flowchart of the method in the first embodiment.

[0024] Figure 2 This is a schematic diagram of point cloud multi-scale segmentation and multi-view rendering in the first embodiment.

[0025] Figure 3 The result of the first embodiment is a 3D back projection and a visualization of the defect point cloud.

[0026] Figure 4 This is a schematic diagram of a multi-segment, multi-disease coupled knowledge graph for the first embodiment.

[0027] Figure 5 This is a schematic diagram of a sample page of the automatically generated detection report for the first embodiment.

[0028] Figure 6 This is a system structure diagram of the second embodiment. Detailed Implementation

[0029] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0030] It should be noted that the terminology used herein is for the purpose of describing particular implementations only and is not intended to limit the exemplary implementations of the present invention.

[0031] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0032] Example 1 This embodiment discloses a method for judging tunnel defects based on large model identification of tunnel point cloud data. By multi-scale segmentation and multi-view rendering, the three-dimensional point cloud is converted into multimodal image data pairs. Combined with the multimodal large model integrating the SIDNet module and dedicated prompts such as defect segmentation, quantitative measurement, and risk assessment, intelligent identification and quantitative analysis are performed. The results are then reconstructed into a three-dimensional defect point cloud with semantic labels by back projection. Subsequently, multi-defect coupled early warning is realized based on spatiotemporal knowledge graph. Finally, a standardized detection report is automatically generated, realizing full-process automation, high-precision identification, and intelligent judgment of tunnel defect detection, significantly improving detection efficiency and reliability.

[0033] Specifically, such as Figure 1 As shown, the method for identifying tunnel defects based on large model-based tunnel point cloud data includes: Step S1: Obtain the original three-dimensional point cloud data of the tunnel, and preprocess the original point cloud data to obtain a standardized point cloud dataset.

[0034] Mobile scanning is performed using vehicle-mounted LiDAR such as the Trimble MX9 platform or Leica Pegasus series equipment, or handheld LiDAR such as the Faro Focus or GeoSLAM ZEB Horizon equipment is used for handheld mobile scanning to collect raw 3D color point cloud data of the entire tunnel cross section. The output point cloud is in LAS or LAZ format, which includes XYZ coordinates, intensity and RGB color features.

[0035] The collected point cloud data undergoes statistical filtering for noise reduction, voxel downsampling, ICP registration, and mileage marker labeling to construct a high-density, low-noise standardized point cloud dataset. In the statistical filtering process, for each point, 50 neighboring points are searched, and the average distance and standard deviation of these neighboring points are calculated. Points whose distance deviates from the average distance by more than twice the standard deviation are identified as noise points and removed. This step effectively removes discrete noise points, retaining the valid point cloud data corresponding to the main tunnel structure.

[0036] The denoised point cloud data is subjected to voxel downsampling processing, which divides the point cloud space into a uniform voxel grid. Only one representative point is retained in each voxel, which not only ensures the sparsity optimization of the data, but also does not lose the feature information of minor defects such as cracks and peeling.

[0037] If multiple segments of point cloud data exist during the acquisition process, the Iterative Closest Point (ICP) algorithm is used for coordinate registration. Using the first segment of point cloud data as the reference coordinate system, subsequent segments of point cloud data are registered with the reference point cloud to achieve coordinate unification of the entire tunnel point cloud data.

[0038] By combining the vehicle mileage or handheld scanning trajectory mileage information recorded during the collection process, the registered point cloud data is labeled with mileage markers, so that each point cloud data is associated with a unique mileage marker, providing a location basis for subsequent segmentation by fixed mileage.

[0039] Step S2: Segment the standardized point cloud dataset and construct image multimodal data pairs.

[0040] The standardized point cloud dataset is longitudinally segmented at fixed intervals of 0.5–2m. Each segment is rendered from multiple preset viewpoints to generate high-resolution depth maps and pseudo-color maps. Specifically, 6–12 viewpoint combinations are selected from eight basic viewpoints: top view, front view, left view, right view, 45° left oblique view, 45° right oblique view, rear view, and bottom view. The rendering resolution is set to 2048×2048 or higher to ensure consistency in rendering parameters across different segments and viewpoints.

[0041] Depth map rendering is based on the 3D spatial coordinates of segmented point clouds. A grayscale mapping rule is used to represent the distance between the point cloud and the rendering camera; the closer the distance, the higher the grayscale value; the farther the distance, the lower the grayscale value. During rendering, invalid point clouds in occluded areas are automatically filtered out, and anti-aliasing is applied to the occluded edges to ensure the edge clarity of the depth map. This accurately reflects the 3D contour differences of the tunnel cross-section, providing support for the identification of depth features of defects such as cracks and deformations.

[0042] The pseudo-color image is generated by mapping based on the inherent features of the segmented point cloud. Specifically, the segmented point cloud is projected onto a two-dimensional image plane from a preset viewpoint, and occlusion relationships are handled using depth buffering; the laser return intensity value or RGB color channel of each point in the point cloud is extracted as a scalar source; the scalar source is normalized to the range of 0-1; a preset pseudo-color mapping table (such as a Jet or Hot mapping table) is applied to map the normalized scalar values ​​to the RGB color space to form a high-contrast gradient effect; the projected pixels are interpolated and filled to generate a high-resolution pseudo-color image, and optional contrast enhancement post-processing can be performed to highlight the potential disease features corresponding to areas of abnormal intensity.

[0043] The depth map uses grayscale mapping to represent distance, while the pseudo-color map is based on intensity or color channel mapping. A point cloud-multi-view image multimodal data pair is constructed based on point cloud data, the depth map, and the pseudo-color map, such as... Figure 2 As shown, multimodal data pairs transform complex 3D point clouds into intuitive 2D image forms, improving the visualization capabilities for occluded areas and subtle defects.

[0044] Step S3: Construct a large defect detection model integrating the SIDNet model and a set of dedicated prompts for tunnel defect identification and assessment. Input the image multimodal data pairs and the dedicated prompts into the large defect detection model to obtain preliminary identification and assessment results.

[0045] The large-scale defect detection model integrates the SIDNet model as the core defect detection backbone network. This model is a deep convolutional neural network based on laser intensity and depth information, used for high-precision segmentation of spalling defects in tunnel lining. Dedicated prompting instructions are designed for defect segmentation, convergence deformation measurement, and comprehensive risk assessment.

[0046] The image multimodal data and the dedicated prompt instruction are input into the defect detection large model. First, the depth map and pseudo-color map are extracted by a multimodal encoder to obtain a visual feature vector. Then, the dedicated prompt instruction is converted into a prompt embedding by a text encoder. Next, the visual features and cue embeddings are fused using a cross-attention mechanism, calculated as follows:

[0047] Where Q represents the query vector, which is derived from the hint embedding, and K and V represent the key vector and value vector, respectively, which are derived from the visual feature vector.

[0048] Finally, the decoder outputs preliminary identification results, including defect segmentation mask, convergent deformation measurement value, and risk assessment score.

[0049] Step S4: The obtained preliminary identification and evaluation results are mapped back to the original three-dimensional point cloud coordinate system through the view projection matrix to construct a three-dimensional defect point cloud with detailed disease semantic labels.

[0050] Efficient batch inference is performed on multiple multimodal data pairs. The high-precision segmentation output of the SIDNet model is utilized, and the inference results are accurately mapped back to the original 3D point cloud coordinate system using a view projection matrix. This constructs a 3D defect point cloud with detailed semantic labels for the defects. Figure 3 As shown. The projection matrix of each view is recorded during the rendering phase. Where K is the camera intrinsic parameter matrix, For external reference.

[0051] For a defective pixel (u, v), obtain the depth value d, and calculate the corresponding world coordinates using the inverse projection formula:

[0052] in, Used as world coordinates.

[0053] Collect multi-view labels and assign them to the original point cloud points through majority voting or distance-weighted fusion. Add semantic labels to each point, including disease type, location coordinates, severity quantification value, and risk level. Post-processing removes isolated noise points and merges connected components.

[0054] Step S5: Based on the three-dimensional defect point cloud, construct a temporal-spatial knowledge graph, analyze the coupling relationship and spatial clustering characteristics between different disease types within a preset continuous segment, and generate early warning information when preset risk conditions are met.

[0055] like Figure 4 As shown, a temporal-spatial knowledge graph is constructed on the three-dimensional defect point cloud to perform correlation analysis and risk clustering identification on multiple defects within a continuous range of 10 to 50 meters. When the preset risk clustering conditions are met, a high-level early warning is automatically triggered to form an intelligent early warning result for potential risks. The knowledge graph associates the coupling relationships of multiple defects such as cracks, water leakage, lining deformation, and spalling.

[0056] Edges between nodes are constructed based on temporal relationships (the difference in mileage between adjacent lesions is less than a preset threshold, such as 5m) and spatial relationships (the Euclidean distance between lesion centers is less than a preset threshold, such as 2m). Coupled relationship edges are added according to domain rules (such as connecting crack nodes and leakage nodes, with edge weights based on induction probability, for example, the weight increases by 0.8 when the crack width is greater than 5mm). Extract sub-images that cover a preset continuous segment; The subgraph is subjected to association analysis and risk cluster identification. Specifically, the subgraph density is calculated by the ratio of the number of nodes to the subgraph volume, and the total risk score is calculated by the weighted sum of the node risk score and the edge weight. When the density of a sub-map exceeds a preset threshold or the total risk score exceeds a preset threshold, it is identified as a risk cluster area, and a high-level early warning message is automatically generated, including the warning location, coupled disease type, risk cause analysis, and treatment suggestions.

[0057] Step S6: Generate tunnel defect detection results based on the three-dimensional defect point cloud and early warning information.

[0058] Based on the aforementioned 3D defect point cloud and early warning results, a 3D defect heat map, a defect list table, and a Word or PDF inspection report conforming to the technical specifications for highway tunnel maintenance are automatically generated. Figure 5As shown, the report includes heatmaps, checklists, and statistical charts, forming an efficient and consistent visualization and documentation output.

[0059] Example 2 This embodiment discloses a tunnel defect judgment system based on large model identification of tunnel point cloud data; like Figure 6 As shown, the tunnel defect identification system based on large model identification tunnel point cloud data includes: The data acquisition and preprocessing module is configured to: acquire the raw three-dimensional point cloud data of the tunnel, preprocess the raw point cloud data, and obtain a standardized point cloud dataset; The point cloud multi-scale segmentation and multi-view rendering module is configured to: segment the standardized point cloud dataset and construct image multimodal data pairs; The large model inference module is configured to: construct a large defect detection model integrating the SIDNet model and a set of dedicated prompts for tunnel defect identification and assessment; input the image multimodal data pairs and the dedicated prompts into the large defect detection model to obtain preliminary identification and assessment results; The three-dimensional back projection module is configured to: map the obtained preliminary identification and evaluation results back to the original three-dimensional point cloud coordinate system through the view projection matrix, and construct a three-dimensional defect point cloud with detailed disease semantic labels; The intelligent judgment and early warning module is configured to: construct a time-space knowledge graph based on the three-dimensional defect point cloud, analyze the coupling relationship and spatial clustering characteristics between different disease types in a preset continuous segment, and generate early warning information when preset risk conditions are met; The detection result output module is configured to generate tunnel defect detection results based on the three-dimensional defect point cloud and early warning information.

[0060] Example 3 The purpose of this embodiment is to provide a computer-readable storage medium.

[0061] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the tunnel defect judgment method based on large model identification tunnel point cloud data as described in Embodiment 1.

[0062] Example 4 The purpose of this embodiment is to provide an electronic device.

[0063] An electronic device includes a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps in the tunnel defect judgment method based on large model identification tunnel point cloud data as described in Embodiment 1.

[0064] The steps and methods involved in the apparatuses of Embodiments 2, 3, and 4 above correspond to those in Embodiment 1. For specific implementation details, please refer to the relevant description section of Embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood as including any medium capable of storing, encoding, or carrying an instruction set for execution by a processor and enabling the processor to perform any of the methods in this invention.

[0065] Those skilled in the art will understand that the modules or steps of the present invention described above can be implemented using general-purpose computer devices. Optionally, they can be implemented using computer-executable program code, thereby allowing them to be stored in a storage device for execution by a computer device, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. The present invention is not limited to any particular combination of hardware and software.

[0066] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A method for identifying tunnel defects based on large-scale model-based tunnel point cloud data, characterized in that, include: Obtain the raw three-dimensional point cloud data of the tunnel, and preprocess the raw point cloud data to obtain a standardized point cloud dataset; The standardized point cloud dataset is segmented and image multimodal data pairs are constructed; A large defect detection model integrating the SIDNet model and a set of dedicated prompts for tunnel defect identification and assessment are constructed. The image multimodal data pairs and the dedicated prompts are input into the large defect detection model to obtain preliminary identification and assessment results. The obtained preliminary identification and evaluation results are mapped back to the original three-dimensional point cloud coordinate system through the view projection matrix to construct a three-dimensional defect point cloud with detailed disease semantic labels. Based on the three-dimensional defect point cloud, a temporal-spatial knowledge graph is constructed to analyze the coupling relationship and spatial clustering characteristics between different disease types within a preset continuous segment. When preset risk conditions are met, early warning information is generated. Based on the three-dimensional defect point cloud and early warning information, tunnel defect detection results are generated.

2. The tunnel defect judgment method based on large model identification tunnel point cloud data as described in claim 1, characterized in that, The raw point cloud data is preprocessed, including statistical filtering for noise reduction, voxel downsampling, ICP registration, and mileage marker labeling.

3. The tunnel defect judgment method based on large model identification tunnel point cloud data as described in claim 1, characterized in that, The standardized point cloud dataset is segmented and image multimodal data pairs are constructed, including: The standardized point cloud dataset is segmented according to a preset vertical interval; For each segment of point cloud, a high-resolution depth map and a pseudo-color map are generated by rendering from multiple preset perspectives. The depth map uses grayscale mapping to represent distance, and the pseudo-color map is based on intensity or color channel mapping. Construct point cloud-multi-view image multimodal data pairs based on point cloud data, depth maps, and pseudo-color maps.

4. The tunnel defect judgment method based on large model identification tunnel point cloud data as described in claim 3, characterized in that, For each segment of the point cloud, a pseudo-color image is generated, including: The segmented point cloud is projected onto a two-dimensional image plane from a preset viewpoint, and the occlusion relationship is handled by using depth buffering. Extract the laser return intensity value or RGB color channel of each point in the point cloud as a scalar source; The scalar source is normalized. A preset pseudo-color mapping table is used to map normalized scalar values ​​to the RGB color space, creating a high-contrast gradient effect. The projected pixels are interpolated and filled to generate a high-resolution pseudo-color image.

5. The tunnel defect judgment method based on large model identification tunnel point cloud data as described in claim 1, characterized in that, The image multimodal data and the dedicated prompting instructions are input into the defect detection large model to obtain preliminary identification and evaluation results, including: Visual feature vectors are obtained by extracting features from depth maps and pseudo-color maps using a multimodal encoder. The dedicated prompt instructions are converted into prompt embeddings using a text encoder; The calculation formula is as follows: Visual features and cue embeddings are fused using a cross-attention mechanism. Where Q represents the query vector, which is derived from the hint embedding, and K and V represent the key vector and value vector, respectively, which are derived from the visual feature vector; The decoder outputs preliminary identification results, including defect segmentation mask, convergent deformation measurement value, and risk assessment score.

6. The tunnel defect judgment method based on large model identification tunnel point cloud data as described in claim 1, characterized in that, The obtained preliminary identification and evaluation results are mapped back to the original 3D point cloud coordinate system through a view projection matrix to construct a 3D defect point cloud with detailed defect semantic labels, including: The projection matrix of each view is recorded during the rendering phase; For the defect image, the depth value is obtained, and the corresponding world coordinates are calculated using the inverse projection formula; Collect multi-view labels, assign them to the original point cloud through majority voting or distance-weighted fusion, add semantic labels to each point, remove isolated noise points and merge connected components.

7. The method for judging tunnel defects based on large model identification tunnel point cloud data as described in claim 1, characterized in that, Based on the aforementioned 3D defect point cloud, a temporal-spatial knowledge graph is constructed. The coupling relationships and spatial clustering characteristics between different defect types within a preset continuous segment are analyzed. When preset risk conditions are met, early warning information is generated, including: The disease instances in the 3D defect point cloud are used as nodes in the knowledge graph. Each node's attributes include disease type, mileage location coordinates, severity quantification value, and individual disease risk score. Edges between nodes are constructed based on temporal and spatial relationships, and coupling edges are added according to domain rules; Extract subgraphs that cover a preset continuous paragraph, and perform correlation analysis and risk cluster identification on the subgraphs; When the density of a sub-map exceeds a preset threshold or the total risk score exceeds a preset threshold, it is identified as a risk cluster area, and an early warning message is automatically generated.

8. A tunnel defect identification system based on large model identification of tunnel point cloud data, characterized in that, include: The data acquisition and preprocessing module is configured to: acquire the raw three-dimensional point cloud data of the tunnel, preprocess the raw point cloud data, and obtain a standardized point cloud dataset; The point cloud multi-scale segmentation and multi-view rendering module is configured to: segment the standardized point cloud dataset and construct image multimodal data pairs; The large model inference module is configured to: construct a large defect detection model integrating the SIDNet model and a set of dedicated prompts for tunnel defect identification and assessment; input the image multimodal data pairs and the dedicated prompts into the large defect detection model to obtain preliminary identification and assessment results; The three-dimensional back projection module is configured to: map the obtained preliminary identification and evaluation results back to the original three-dimensional point cloud coordinate system through the view projection matrix, and construct a three-dimensional defect point cloud with detailed disease semantic labels; The intelligent judgment and early warning module is configured to: construct a time-space knowledge graph based on the three-dimensional defect point cloud, analyze the coupling relationship and spatial clustering characteristics between different disease types in a preset continuous segment, and generate early warning information when preset risk conditions are met; The detection result output module is configured to generate tunnel defect detection results based on the three-dimensional defect point cloud and early warning information.

9. A computer-readable storage medium having a program stored thereon, characterized in that, When executed by the processor, the program implements the steps in the tunnel defect judgment method based on large model identification tunnel point cloud data as described in any one of claims 1-7.

10. An electronic device comprising a memory, a processor, and a program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the tunnel defect judgment method based on large model identification tunnel point cloud data as described in any one of claims 1-7.

Images