Feature identification labeling method and system applied to three-dimensional tunnel modeling

Abstract

The embodiment of the invention discloses a feature recognition labeling method and system applied to three-dimensional tunnel modeling, and the method comprises the steps: firstly carrying out the preprocessing of an obtained tunnel construction design drawing and point cloud scanning data of a corresponding construction stage, and obtaining a three-dimensional tunnel model data set with a space coordinate alignment relation; the three-dimensional tunnel model data set comprises a multi-source data fused structural point cloud unit and a design drawing vectorized contour line unit; secondly, performing three-dimensional structure feature extraction on the data set to obtain geometric morphology features of a structure point cloud unit and structure correlation features of a contour line unit; then carrying out conjoint analysis on the two features through a deep learning network to generate a feature analysis result containing feature response distribution; and finally, based on the result, identifying and labeling a key structure region in the three-dimensional tunnel model data set, generating a feature identification and labeling result containing feature type labels and spatial position information, and providing more accurate analysis and management basis for tunnel construction.

Description

Technical Field

[0001] The embodiments of the present invention relate to the technical field of computer data processing, and particularly to a feature recognition and annotation method and system applied to three-dimensional tunnel modeling. Background Art

[0002] During the tunnel construction process, accurately identifying the key structural areas of the tunnel and making effective annotations are crucial for ensuring construction safety, improving construction quality, and optimizing construction progress. Currently, there are various different ways in the prior art to process tunnel construction-related data to achieve structural area recognition and annotation.

[0003] A common prior art is to simply rely on the tunnel construction design drawings for analysis. However, the construction design drawings reflect the theoretical design situation. In the actual construction process, due to the complexity of geological conditions, differences in construction techniques, and unforeseen factors, etc., the actual structure of the tunnel may deviate from the design drawings. The drawings cannot reflect the dynamic changes of the structure during the construction process in real time, and it is difficult to detect problems such as local deformation and structural damage during construction, thus unable to provide timely and effective guidance for the construction process. Another prior art is to only use point cloud scan data for tunnel structure analysis. However, the point cloud data itself is discrete and massive, lacking clear semantic information and structural association information. Just relying on the point cloud data, it is very difficult to determine the specific functions of each part of the tunnel structure and the relationships between them, and it is difficult to accurately identify some internal structures or complex structural connections. Moreover, the processing and analysis of point cloud data require a large amount of computing resources and time, with low efficiency. There are also some prior arts that attempt to combine tunnel construction design drawings and point cloud scan data for analysis, but there are deficiencies in data fusion and feature extraction, lacking systematic processing of the data, and unable to fully explore the potential relationships between various data sources, resulting in incomplete and inaccurate extracted features.

[0004] Therefore, how to optimize the modeling feature recognition and annotation during the tunnel construction process to provide a more accurate analysis and management basis for tunnel construction is a technical problem that needs to be overcome currently. Summary of the Invention

[0005] The embodiments of the present invention provide a feature recognition and annotation method and system applied to three-dimensional tunnel modeling.

[0006] In a first aspect, the embodiments of the present invention provide a feature recognition and annotation method applied to three-dimensional tunnel modeling, which is applied to a feature recognition and annotation system. The method includes: Preprocess the obtained tunnel construction design drawings and point cloud scanning data of the corresponding construction stage to obtain a three-dimensional tunnel model dataset with spatially aligned coordinate relationships. The three-dimensional tunnel model dataset includes a structural point cloud unit with multi-source data fusion and a contour line unit obtained by vectorizing the design drawings. Perform three-dimensional structural feature extraction processing on the three-dimensional tunnel model dataset to obtain the geometric shape features of the structural point cloud unit and the structural association features of the contour line unit. Perform joint analysis processing on the geometric shape features and the structural association features through a deep learning network to generate a feature analysis result including a feature response distribution. Based on the feature analysis result, identify and label the key structural areas in the three-dimensional tunnel model dataset to generate a feature recognition and labeling result including feature type labels and spatial position information.

[0007] In a second aspect, an embodiment of the present invention provides a feature recognition and labeling system, including: A processor; A storage device on which a computer program is stored. When the computer program is executed by the processor, the processor implements any of the feature recognition and labeling methods applied to three-dimensional tunnel modeling.

[0008] An embodiment of the present invention provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the feature recognition and labeling method applied to three-dimensional tunnel modeling are implemented.

[0009] In the embodiment of the present invention, by processing the tunnel construction design drawings and the point cloud scanning data of the corresponding construction stage, a three-dimensional tunnel model dataset with spatially aligned coordinate relationships is obtained, realizing the effective fusion of multi-source data, and enabling the structural point cloud unit and the contour line unit obtained by vectorizing the design drawings to complement each other. Three-dimensional structural feature extraction is performed on the three-dimensional tunnel model dataset to obtain the geometric shape features of the structural point cloud unit and the structural association features of the contour line unit respectively, deeply analyzing the tunnel structure from different angles and providing rich feature information for accurately identifying key structural areas. Through joint analysis processing of these two features by a deep learning network, the powerful feature learning ability of deep learning is fully utilized, potential relationships between features can be mined, and a feature analysis result including a feature response distribution is generated, thereby accurately judging the importance of different structural features. Based on the feature analysis result, the key structural areas are identified and labeled to generate a feature recognition and labeling result including feature type labels and spatial position information, providing an intuitive and detailed basis for the monitoring, management, and maintenance of tunnel construction, helping to improve construction efficiency and quality, and ensuring the safe and stable operation of the tunnel. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 This is a flowchart of a feature recognition and annotation method applied to 3D tunnel modeling provided by an embodiment of the present invention.

[0011] Figure 2 This is a schematic diagram of the basic structure of a feature recognition and annotation system provided by an embodiment of the present invention. Detailed implementation manners

[0012] To make the above objects, features, and advantages of the present invention more obvious and understandable, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings and specific implementation manners.

[0013] Refer to Figure 1 As shown, this figure is a flowchart of a feature recognition and annotation method applied to 3D tunnel modeling provided by an embodiment of the present invention, and this method can be applied to a feature recognition and annotation system. As Figure 1 shown, this method may include step 110 - step 140.

[0014] Step 110: Preprocess the obtained tunnel construction design drawings and point cloud scan data of the corresponding construction stage to obtain a 3D tunnel model dataset with a spatial coordinate alignment relationship. The 3D tunnel model dataset includes a structural point cloud unit with multi-source data fusion and a contour line unit with vectorized design drawings.

[0015] In a tunnel construction scenario, first, obtain the tunnel construction design drawings and the point cloud scan data of the corresponding construction stage. The construction design drawings detail the design structure, dimensions, and other information of the tunnel, while the point cloud scan data is the point cloud data of the actual space of the tunnel obtained through technologies such as 3D laser scanning.

[0016] Then, preprocess the above two types of data. The preprocessing process includes vectorizing the construction design drawings, converting the graphics in the drawings into vector graphics that can be recognized and processed by a computer to form a contour line unit with vectorized design drawings. For the point cloud scan data, perform denoising, filtering, and other operations to remove the noise points and outliers in it and improve the data quality. Through a spatial coordinate alignment algorithm, unify the coordinate systems of the construction design drawings and the point cloud scan data so that the two have an alignment relationship in space, and finally obtain a 3D tunnel model dataset composed of a structural point cloud unit with multi-source data fusion and a contour line unit with vectorized design drawings. For example, by vectorizing the construction design drawings, the contour line segments of each part of the tunnel can be obtained, and after processing the point cloud scan data, it can accurately reflect the actual spatial structure of the tunnel.

[0017] Step 120: Perform three-dimensional structural feature extraction processing on the three-dimensional tunnel model dataset to obtain the geometric shape features of the structural point cloud unit and the structural association features of the contour line unit.

[0018] After obtaining the three-dimensional tunnel model dataset, it is necessary to perform three-dimensional structural feature extraction on it. For the structural point cloud unit, its geometric shape features need to be analyzed, that is, the external shape, surface undulation and other features of the actual tunnel structure. For the contour line unit, its structural association features are mainly extracted, that is, the connection relationship and interaction between various structural components of the tunnel. By extracting the above two features, the structural characteristics and internal relationships of the tunnel can be understood more deeply. For example, the geometric shape features of the structural point cloud unit can reflect whether there are deformations, protrusions, etc. in the tunnel, and the structural association features of the contour line unit can judge whether the connections between various parts of the tunnel are firm and reasonable.

[0019] In an optional embodiment, the performing three-dimensional structural feature extraction processing on the three-dimensional tunnel model dataset to obtain the geometric shape features of the structural point cloud unit and the structural association features of the contour line unit includes: Step 121: Perform noise filtering processing on the structural point cloud unit for outliers and redundant points to obtain a smoothed subset of the structural point cloud.

[0020] In the actual implementation process, there may be outliers and redundant points in the structural point cloud unit, which will affect the accuracy of subsequent feature extraction. Therefore, noise filtering processing needs to be performed on the structural point cloud unit. Outliers refer to points with significantly abnormal distances from surrounding points, and redundant points refer to points that are repeated or similar in space. Through noise filtering algorithms, such as density-based noise filtering algorithms, outliers and redundant points are removed to obtain a smoothed subset of the structural point cloud. For example, in point cloud data, there may be some outliers caused by scanning errors, which are far from the actual structure of the tunnel. Through noise filtering processing, these points can be removed, making the point cloud data more smoothly and accurately reflect the actual structure of the tunnel.

[0021] Step 122: Perform neighborhood analysis processing on the smoothed subset of the structural point cloud, calculate the local curvature parameters and normal vector directions of each point cloud point, and generate geometric shape features reflecting the surface undulation state of the structure. The geometric shape features include the curvature distribution pattern of the point cloud points and the normal vector consistency index.

[0022] After obtaining the smoothed structural point cloud subset, neighborhood analysis is further performed. For each point cloud point, analyze the distribution of points within its neighborhood, and calculate the local curvature parameter and normal vector direction of this point. The local curvature parameter can reflect the surface curvature degree at the position where the point is located, and the normal vector direction represents the normal direction of the surface where the point is located. By statistically analyzing the local curvature parameters and normal vector directions of all point cloud points, a curvature distribution pattern and a normal vector consistency index of the point cloud points are generated, and these indices together constitute the geometric morphological features reflecting the undulation state of the structural surface. For example, at the vault part of a tunnel, the local curvature parameters and normal vector directions of the point cloud points will present corresponding distribution rules, and by analyzing these rules, the surface undulation of the tunnel vault can be accurately understood.

[0023] Step 123: Perform line segment connectivity analysis on the contour line unit, identify the contour line segments in the tunnel construction design drawing, and extract the length parameter and included angle parameter of the contour line segment as the basic structural features.

[0024] For the contour line unit, line segment connectivity analysis is required. First, identify the contour line segments in the tunnel construction design drawing, and these line segments represent the boundaries of various structural components of the tunnel. Then, extract the length parameter and included angle parameter of these contour line segments. The length parameter reflects the length of the line segment, and the included angle parameter reflects the relative positional relationship between the line segments. These length parameters and included angle parameters constitute the basic structural features. For example, at the corner of a tunnel, the length and included angle of the contour line segment will change significantly. By extracting these parameters, the structural characteristics at the tunnel corner can be accurately described.

[0025] Step 124: Based on the basic structural features, perform structural correlation modeling on the contour line unit, analyze the intersection relationship and adjacency relationship between different contour line segments, and generate structural correlation features reflecting the connection mode of the tunnel structural components. The structural correlation features include the distribution density of the line segment intersection points and the angle matching degree of the adjacent line segments.

[0026] Based on the extracted basic structural features, perform structural correlation modeling on the contour line unit. Analyze the intersection relationship and adjacency relationship between different contour line segments, that is, determine which line segments intersect each other and which line segments are adjacent. By analyzing these relationships, calculate the distribution density of the line segment intersection points and the angle matching degree of the adjacent line segments. The distribution density of the line segment intersection points reflects the connection complexity between the tunnel structural components, and the angle matching degree of the adjacent line segments reflects the rationality and stability of the connection. These indices together constitute the structural correlation features reflecting the connection mode of the tunnel structural components. For example, at the support structure of a tunnel, different contour line segments will intersect and be adjacent to each other. By analyzing the distribution density of the line segment intersection points and the angle matching degree of the adjacent line segments, the firmness and rationality of the connection of the support structure can be evaluated.

[0027] On the basis of steps 121 to 124, the method further includes: Step 125: Perform unified dimension conversion processing on the geometric features and the structural association features to obtain a target feature set with the same feature dimension. The information entropy calculation result of the target feature set is used to adjust the feature fusion weight parameters of the deep learning network in the joint analysis process.

[0028] After obtaining the geometric features and structural association features, since the feature dimensions of the two may be different, unified dimension conversion processing is required. Through the dimension conversion algorithm, the geometric features and structural association features are converted into a target feature set with the same feature dimension. Then, the information entropy of the target feature set is calculated, and the information entropy can reflect the uncertainty and information content of the feature. According to the calculation results of the information entropy, the feature fusion weight parameters of the deep learning network in the joint analysis process are adjusted. For example, if the information entropy of a feature is large, it means that the feature contains more information, and it can be given a larger weight when the feature is fused to improve the accuracy of the analysis.

[0029] Step 130: Perform joint analysis and processing on the geometric features and the structural association features through a deep learning network to generate a feature analysis result including a feature response distribution.

[0030] The obtained geometric features and structural association features are input into the deep learning network for joint analysis. The deep learning network can mine the potential relationship between features based on its feature learning and analysis capabilities. In the deep learning network, the geometric features and structural association features are processed and analyzed at multiple levels and scales. Through operations such as convolution, pooling, and activation, a feature analysis result containing feature response distribution is finally generated. The feature response distribution reflects the importance of different structural features in the analysis, that is, the activation value and feature type identification corresponding to each feature dimension. For example, in the deep learning network, through the joint analysis of geometric features and structural association features, it can be found that the feature response of some key structural parts of the tunnel is strong, indicating that these parts play an important role in the tunnel structure.

[0031] As an optional embodiment, the geometric morphological features and the structural association features are jointly analyzed and processed by a deep learning network to generate a feature analysis result including a feature response distribution, including: Step 131: Input the geometric features and the structural association features into the multi-scale feature fusion layer of the deep learning network, perform multi-resolution downsampling processing on the features in combination with preset scale division rules, and generate a multi-scale feature stream.

[0032] In an embodiment of the present invention, the geometric morphological features and structural association features are input into the multi-scale feature fusion layer of the deep learning network. In the multi-scale feature fusion layer, according to the preset scale division rule, the features are processed by multi-resolution downsampling. Multi-resolution downsampling can analyze the features at different scales, extract the feature information at different scales, and generate multi-scale feature streams. For example, through different scale divisions, the global information and local information of the features can be obtained, and the multi-scale feature streams can contain the feature information at different scales.

[0033] Step 132: For the local detail feature stream in the multi-scale feature stream, use convolutional kernels with different receptive fields to perform feature extraction processing, capture the local curvature change details of the point cloud in the geometric morphological features and the line segment adjacency relationship details in the structural association features, and obtain the local detail feature output.

[0034] In the multi-scale feature stream, for the local detail feature stream, use convolutional kernels with different receptive fields to perform feature extraction on it. Convolutional kernels with different receptive fields can capture feature information in different ranges. For the geometric morphological features, focus on capturing the local curvature change details of the point cloud, that is, the local bending change of the tunnel surface. For the structural association features, pay attention to the line segment adjacency relationship details, that is, the specific relationship of the adjacent line segments between the structural components of the tunnel. Through the processing of the convolutional kernels, the local detail feature output reflecting these detail information is obtained. For example, using a convolutional kernel with a small receptive field can capture the tiny changes in the local curvature of the point cloud, while a convolutional kernel with a large receptive field can obtain more extensive line segment adjacency relationship information.

[0035] In a preferred embodiment, the step of using convolutional kernels with different receptive fields to perform feature extraction processing on the local detail feature stream in the multi-scale feature stream, capturing the local curvature change details of the point cloud in the geometric morphological features and the line segment adjacency relationship details in the structural association features, and obtaining the local detail feature output includes: Step 1321: Perform window division processing on the geometric morphological features of the local detail feature stream to generate multiple local point cloud window units each containing a preset number of point cloud points.

[0036] Perform window division on the geometric morphological features in the local detail feature stream. According to the preset window size and quantity, divide the geometric morphological features into multiple local point cloud window units, and each window unit contains a preset number of point cloud points. For example, set the window size as a fixed spatial range, and within this range, a certain number of point cloud points are included, so that the geometric morphological features can be localised.

[0037] Step 1322: Conduct statistical analysis on the curvature distribution pattern and normal vector consistency index of the point cloud points within each local point cloud window unit, and calculate the average value and standard deviation of the curvature within the window as the local curvature change statistical features.

[0038] For each local point cloud window unit, conduct statistical analysis on the curvature distribution pattern and normal vector consistency index of the point cloud points therein. Calculate the average value and standard deviation of the curvature of the point cloud points within the window. The average value can reflect the overall level of curvature within the window, while the standard deviation can reflect the degree of dispersion of the curvature. These two indicators together constitute the local curvature change statistical features, which can accurately describe the curvature change of the local point cloud. For example, if the standard deviation of the curvature within a window is large, it indicates that the surface undulation in this area is relatively severe.

[0039] Step 1323: Perform line segment segmentation processing on the structural association features of the local detail feature stream to generate multiple local line segment group units containing continuous line segments.

[0040] For the structural association features in the local detail feature stream, perform line segment segmentation. Divide the contour line segments into multiple local line segment group units according to certain rules, and each unit contains continuous line segments. For example, according to the connection relationship and spatial position of the line segments, adjacent and continuous line segments are divided into a local line segment group unit. In this way, the structural association features can be localised for facilitating the analysis of the line segment adjacency relationship.

[0041] Step 1324: Conduct sequence analysis on the line segment length parameter and included angle parameter within each local line segment group unit, and extract the increasing and decreasing pattern of the line segment length and the periodic change pattern of the included angle as the local line segment adjacency relationship features.

[0042] Conduct sequence analysis on the line segment length parameter and included angle parameter within each local line segment group unit. Analyze the change trend of the line segment length and extract its increasing and decreasing pattern, that is, determine whether the line segment length is gradually increasing or decreasing. At the same time, analyze the change of the included angle and extract its periodic change pattern. These patterns constitute the local line segment adjacency relationship features, which can reflect the characteristics of the adjacency relationship between local line segments. For example, if the line segment length within a local line segment group unit shows an increasing pattern and the included angle has a certain periodic change, it indicates that the structure in this area has a certain regularity.

[0043] Step 1325: Input the local curvature change statistical features and the local line segment adjacency relationship features into the local feature enhancement sub-network to generate an enhanced local feature vector containing multi-source local detail information.

[0044] In the embodiment of the present invention, the local curvature change statistical features and the local line segment adjacency relationship features are input into the local feature enhancement subnet, which is a specially designed neural network module that can further process and enhance the input features and mine the potential relationship between the features. Through the processing of the local feature enhancement subnet, an enhanced local feature vector containing multi-source local detail information is generated, which integrates the local detail information of geometric morphological features and structural association features.

[0045] Step 1326: Perform nonlinear activation processing on the enhanced local feature vector to obtain a local detail feature output reflecting the details of the local curvature change of the point cloud and the details of the line segment adjacency relationship.

[0046] Optionally, the enhanced local feature vector is subjected to nonlinear activation processing, and the enhanced local feature vector is nonlinearly transformed through an activation function, such as a ReLU function, to highlight important features and suppress unimportant features. After the activation processing, a local detail feature output reflecting the details of the local curvature change of the point cloud and the details of the line segment adjacency relationship is obtained. For example, the activation function can make some features with larger eigenvalues more prominent, thereby improving the accuracy of the analysis.

[0047] Step 133: For the global context feature stream in the multi-scale feature stream, the geometric features and the structural association features are globally modeled through a self-attention mechanism to generate a context feature vector that reflects the overall structural layout of the tunnel.

[0048] Optionally, for the global context feature stream in the multi-scale feature stream, a self-attention mechanism is used to model global dependencies. The self-attention mechanism can automatically calculate the correlation between features and capture the global dependencies of features. The geometric morphological features and structural association features are used as inputs, and the attention weights between different feature positions are calculated through the self-attention mechanism to establish a global dependency model. Finally, a context feature vector reflecting the overall structural layout of the tunnel is generated, which can reflect the mutual relationship and overall layout between the various parts of the tunnel. For example, in different parts of the tunnel, some features may have strong dependencies, and these relationships can be accurately captured through the self-attention mechanism.

[0049] In another preferred embodiment, the global context feature stream in the multi-scale feature stream is subjected to global dependency modeling processing on the geometric features and the structural association features through a self-attention mechanism to generate a context feature vector reflecting the overall structural layout of the tunnel, including: Step 1331: Map the geometric morphological features and the structural correlation features into query vectors, key vectors, and value vectors respectively. The query vector is used to represent the feature information at the target feature position, the key vector is used to represent the feature information at non-target feature positions, and the value vector is used to represent the feature values at non-target feature positions.

[0050] Optionally, map the geometric morphological features and the structural correlation features separately to obtain query vectors, key vectors, and value vectors. The query vector represents the feature information at the target feature position, that is, the information at the feature position that needs to be focused on currently. The key vector represents the feature information at non-target feature positions and is used to match with the query vector. The value vector represents the feature values at non-target feature positions and is used for subsequent weighted summation. For example, in the self-attention mechanism, by matching the query vector and the key vector, other feature positions related to the target feature position can be found, and then the value vector is used for information fusion.

[0051] Step 1332: Calculate the dot product similarity between the query vector and the key vector to generate an attention weight matrix reflecting the degree of association between feature positions.

[0052] In this step, calculate the dot product similarity between the query vector and the key vector. The dot product similarity can reflect the similarity degree between two vectors. By calculating the dot product similarities of all query vectors and key vectors, an attention weight matrix is generated, which reflects the degree of association between different feature positions. For example, if the dot product similarity between a certain query vector and a certain key vector is large, it indicates that the association between these two feature positions is strong, and a larger weight can be given to it in subsequent processing.

[0053] Step 1333: Perform softmax normalization processing on the attention weight matrix to obtain the attention distribution probabilities of each feature position with respect to the target feature position.

[0054] Among them, the softmax function can convert the values in the attention weight matrix into a probability distribution, making the sum of the attention distribution probabilities of each feature position with respect to the target feature position equal to 1. Through normalization processing, the attention distribution probabilities of each feature position with respect to the target feature position are obtained, and these probabilities can more accurately represent the degree of association between features. For example, if the attention distribution probability of a feature position is large, it indicates that the influence of this position on the target feature position is large.

[0055] Step 1334: Perform weighted summation processing on the value vector based on the attention distribution probabilities to generate a context-enhanced feature vector containing global feature position dependency relationships.

[0056] Further, according to the attention distribution probability, weighted summation processing is performed on the value vectors. Each value vector is multiplied by the corresponding attention distribution probability, and then all the results are added together to generate a context-enhanced feature vector that includes the global feature position dependence relationship. This vector integrates the information of different feature positions and reflects the dependence relationship of features within the global scope. For example, if the attention distribution probability of a certain feature position is relatively large, then the corresponding value vector will account for a larger proportion in the weighted summation and contribute more to the context-enhanced feature vector.

[0057] Step 1335: Perform a residual connection process on the context-enhanced feature vector, the geometric shape feature, and the structural association feature to obtain an initial context feature vector that integrates the global dependence relationship.

[0058] Optionally, perform a residual connection on the context-enhanced feature vector, the geometric shape feature, and the structural association feature. The residual connection can retain the information of the original features and at the same time add the information of the global dependence relationship. By adding the context-enhanced feature vector to the original features, an initial context feature vector that integrates the global dependence relationship is obtained. For example, the residual connection can avoid problems such as gradient disappearance during the information transmission process and ensure the effective transmission of information.

[0059] Step 1336: Perform dimensionality compression processing on the initial context feature vector to generate a context feature vector that reflects the overall structural layout of the tunnel.

[0060] In the embodiment of the present invention, the dimensionality compression processing can be implemented through a dimensionality compression algorithm, such as principal component analysis, etc., to reduce the dimension of the feature vector while retaining its main information. Finally, a context feature vector that reflects the overall structural layout of the tunnel is generated. This vector is more concise in dimension and is convenient for subsequent processing and analysis. For example, dimensionality compression can remove redundant information in the feature vector, improve the calculation efficiency and the accuracy of analysis.

[0061] Step 134: Input the local detail feature output and the context feature vector into a feature interaction network, and perform information complement through an element-wise product operation to generate an interaction feature map with spatio-temporal consistency constraints.

[0062] In the embodiment of the present invention, after inputting the local detail feature output and the context feature vector into the feature interaction network, information complement is performed through an element-wise product operation. The local detail feature output contains detailed information about the local part of the tunnel, such as the details of the local curvature change of the point cloud and the details of the line segment adjacency relationship, etc., while the context feature vector reflects the information of the overall structural layout of the tunnel. By multiplying element-wise, the local information and the global information can be integrated, so that the features at each position not only contain local details but also consider their positions and associations in the overall structure.

[0063] When performing the element-wise multiplication operation, ensure that the dimensions of the local detailed feature output and the context feature vector match. If the dimensions do not match, corresponding adjustments may be required, such as padding the feature with a lower dimension or cropping the feature with a higher dimension, to ensure that the two can be multiplied smoothly. After the element-wise multiplication operation, the resulting outcome is the interaction feature map with spatio-temporal consistency constraints. This interaction feature map synthesizes local and global information and can more accurately reflect the characteristics of the tunnel structure.

[0064] For example, at a key structural part of the tunnel, the local detailed feature output may show that the point cloud curvature at this part changes significantly, indicating obvious surface undulations. The context feature vector may show that this part has a high importance in the overall structure and is closely related to the surrounding structures. Through the element-wise multiplication, the interaction feature map can reflect the information of both aspects simultaneously.

[0065] Step 135: Perform channel normalization on the interaction feature map, and splice the normalized interaction feature map along the channel dimension to generate a feature fusion vector containing multi-scale interaction information.

[0066] It can be understood that after obtaining the interaction feature map, channel normalization is performed on it. The purpose of channel normalization is to standardize the data of each channel so that the data of different channels have a similar distribution range, which is convenient for subsequent processing and comparison. During the channel normalization process, the mean and standard deviation of the data of each channel need to be calculated, and then the data of each channel is subtracted by the mean and divided by the standard deviation to obtain the normalized channel data.

[0067] After completing the channel normalization, splice the normalized interaction feature map along the channel dimension. The splicing operation arranges the data of different channels in sequence along the channel dimension to form a new feature vector. This feature vector contains multi-scale interaction information and integrates the information of each channel of the interaction feature map after channel normalization.

[0068] For example, the interaction feature map may contain multiple channels, and each channel reflects information in different aspects. Through channel normalization and splicing processing, the information in these different aspects can be integrated together to form a more comprehensive and richer feature fusion vector.

[0069] Step 136: Invoke the response distribution generation layer of the deep learning network to perform activation function mapping processing on the feature fusion vector to generate a feature response distribution reflecting the importance degree of different structural features. The feature response distribution contains the activation values of each feature dimension and the corresponding feature type identifiers.

[0070] Optionally, the feature fusion vector is input into the response distribution generation layer of the deep learning network, where the feature fusion vector is mapped using an activation function. The role of the activation function is to perform a non-linear transformation on each element in the feature fusion vector, so that the output result can reflect the importance of different structural features.

[0071] After being mapped by the activation function, a feature response distribution is generated. The feature response distribution contains the activation values of each feature dimension and the corresponding feature type identifiers. The activation value indicates the importance of the feature dimension in the analysis. The larger the activation value, the more important the structural feature corresponding to the feature dimension. The feature type identifier is used to distinguish different types of structural features, such as the main tunnel structure, ancillary facility structure, etc.

[0072] For example, in the feature response distribution, if the activation value of a certain feature dimension is large and the corresponding feature type identifier is the main tunnel structure, it indicates that the structural feature represented by this feature dimension has a high importance in the entire tunnel structure. Through the feature response distribution, it is possible to quickly identify which structural features in the tunnel are critical.

[0073] Step 140: Based on the feature analysis result, identify and label the key structural regions in the three-dimensional tunnel model dataset, and generate a feature recognition and labeling result including feature type labels and spatial location information.

[0074] In the embodiment of the present invention, after obtaining the feature analysis result, the key structural regions in the three-dimensional tunnel model dataset are identified and labeled according to this result. The feature response distribution in the feature analysis result can reflect the importance of different structural features. By analyzing the feature response distribution, it is possible to determine which structural regions corresponding to the feature dimensions are critical.

[0075] For the identified key structural regions, feature type labels need to be assigned to them and spatial location information needs to be determined. The feature type label is used to describe the structural attributes and functional attributes of the key structural region, such as the main tunnel structure, ancillary facility structure, etc. The spatial location information is used to clarify the specific location of the key structural region in the three-dimensional tunnel model dataset, including the spatial coordinate range and the contour line segment coverage range, etc.

[0076] For example, through the feature response distribution, it is found that the activation value of a certain feature dimension is high, and the corresponding structural region may be the support structure of the tunnel. Assign the feature type label of "tunnel support structure" to this region, and determine its spatial coordinate range in the point cloud scanning data and the contour line segment coverage range in the tunnel construction design drawings, so as to complete the identification and labeling of this key structural region.

[0077] As an implementation, identifying and labeling the key structural regions in the 3D tunnel model dataset based on the feature analysis results, and generating a feature recognition and labeling result including feature type labels and spatial location information, includes: Step 141: According to the feature response distribution in the feature analysis results, extract the feature dimensions with activation values exceeding a preset threshold as key feature dimensions, and the key feature dimensions correspond to the key structural regions of the tunnel.

[0078] In the feature response distribution of the feature analysis results, set a preset threshold. Extract the feature dimensions with activation values exceeding this preset threshold, and these feature dimensions are defined as key feature dimensions. The structural regions corresponding to the key feature dimensions are the key structural regions of the tunnel because the features of these regions show relatively high importance in the analysis.

[0079] For example, the preset threshold is a relatively large activation value standard. When the activation value of a certain feature dimension exceeds this threshold, it indicates that the structural region represented by this feature dimension plays a key role in the tunnel structure, such as the entrances and exits of the tunnel, important connection nodes, etc. By extracting the key feature dimensions, the key structural regions of the tunnel can be quickly located.

[0080] Step 142: According to the feature type identifiers corresponding to the key feature dimensions, assign corresponding feature type labels to each key structural region, and the feature type labels include structural component types and functional attribute categories.

[0081] After determining the key feature dimensions, assign feature type labels to each key structural region according to the corresponding feature type identifiers. The feature type identifiers are the identifiers corresponding to each feature dimension in the feature response distribution, used to distinguish different types of structural features. The feature type labels are further refined and include structural component types and functional attribute categories.

[0082] For example, if the feature type identifier corresponding to the key feature dimension is the main structure of the tunnel, then more detailed feature type labels can be assigned to this key structural region according to specific feature information, such as "tunnel vault structure", where "tunnel vault" is the structural component type and "structure" reflects its functional attribute category.

[0083] As a preferred embodiment, the assigning corresponding feature type labels to each key structural region according to the feature type identifiers corresponding to the key feature dimensions includes: Step 1421: Establish a mapping relationship between the feature type identifiers and a preset label library, and the preset label library includes main structure labels of the tunnel, auxiliary facility structure labels, and potential disease structure labels.

[0084] First, establish the mapping relationship between the feature type identifier and the preset tag library. The preset tag library is a predefined set of tags, including different types of tags such as tunnel main structure tags, accessory facility structure tags, and potential disease structure tags. By establishing the mapping relationship, the feature type identifier can be corresponded to specific tags.

[0085] For example, if the feature type identifier is "related to the main structure", it can be mapped to the "tunnel main structure tag" in the preset tag library. Thus, the corresponding tag category can be quickly found according to the feature type identifier.

[0086] Step 1422: Extract the peak position of the activation value in the feature response distribution of the key feature dimension, and determine the core feature type identifier corresponding to the key feature dimension; search for the corresponding basic tag category in the preset tag library according to the core feature type identifier, and the basic tag category includes the first-level tag of the structural component type and the second-level tag of the functional attribute category.

[0087] For the feature response distribution of the key feature dimension, extract the peak position of the activation value. The feature type identifier corresponding to the peak position of the activation value is the core feature type identifier, which represents the most important feature type of the key feature dimension. According to the core feature type identifier, search for the corresponding basic tag category in the preset tag library. The basic tag category includes the first-level tag of the structural component type and the second-level tag of the functional attribute category.

[0088] For example, if the feature type identifier corresponding to the peak position of the activation value in the feature response distribution of the key feature dimension is "tunnel support", the corresponding basic tag category can be found in the preset tag library, such as the first-level tag is "support structure" and the second-level tag is "load-bearing function".

[0089] Step 1423: Based on the sub-peak position in the feature response distribution of the key feature dimension, extract the auxiliary feature type identifier reflecting the additional attributes of the feature; search for the corresponding supplementary tag information in the preset tag library according to the auxiliary feature type identifier, and the supplementary tag information includes additional tags of the structural material type and the construction technology category.

[0090] In addition to the peak position of the activation value, also pay attention to the sub-peak position in the feature response distribution of the key feature dimension. The feature type identifier corresponding to the sub-peak position is the auxiliary feature type identifier, which reflects the additional attributes of the feature. According to the auxiliary feature type identifier, search for the corresponding supplementary tag information in the preset tag library. The supplementary tag information includes additional tags of the structural material type and the construction technology category.

[0091] For example, the feature type identifier corresponding to the secondary peak position is "concrete material", and the corresponding supplementary label information can be found in the preset label library, such as additional labels like "concrete structural material" and "cast-in-place construction technology".

[0092] Step 1424: Combine and splice the basic label category with the supplementary label information to generate a feature type label containing multiple label levels, where the multiple label levels are used to describe the structural attributes and functional attributes of the key structural area.

[0093] Combine the basic label category and the supplementary label information through combination and splicing to form a feature type label with multiple label levels. The multiple label levels can describe the structural attributes and functional attributes of the key structural area in more detail.

[0094] For example, combine and splice the basic label category "support structure - load-bearing function" and the supplementary label information "concrete structural material - cast-in-place construction technology" to obtain the feature type label "support structure - load-bearing function - concrete structural material - cast-in-place construction technology", which comprehensively describes the attributes and functions of the key structural area.

[0095] Step 143: Extract the structural point cloud units and contour line units associated with the key feature dimension from the three-dimensional tunnel model dataset, and determine the spatial coordinate range of the key structural area in the point cloud scan data and the contour line coverage range in the tunnel construction design drawings.

[0096] After determining the key feature dimension, extract the structural point cloud units and contour line units associated with the key feature dimension from the three-dimensional tunnel model dataset. The structural point cloud units reflect the actual spatial structure of the tunnel, while the contour line units reflect the contour information of the tunnel in the design drawings.

[0097] By analyzing the extracted structural point cloud units and contour line units, determine the spatial coordinate range of the key structural area in the point cloud scan data and the contour line coverage range in the tunnel construction design drawings. The spatial coordinate range can be determined by calculating the minimum bounding box of the point cloud units, and the contour line coverage range can be obtained by extracting the relevant contour lines and determining their starting and ending point coordinates.

[0098] For example, in the point cloud scan data, by analyzing the point cloud units associated with the key feature dimension, calculate the vertex coordinates of its minimum bounding box to determine the spatial coordinate range of the key structural area. In the tunnel construction design drawings, extract the relevant contour lines, record their starting and ending point coordinates, and obtain the contour line coverage range.

[0099] Further, extracting the structural point cloud units and contour line units associated with the key feature dimension from the three-dimensional tunnel model dataset, and determining the spatial coordinate range of the key structural area in the point cloud scanning data and the contour line segment coverage range in the tunnel construction design drawings, includes: Step 1431: Perform feature dimension association analysis processing on the structural point cloud units in the three-dimensional tunnel model dataset, establish the correspondence between the geometric shape features of each structural point cloud point and the key feature dimension; screen out the point cloud points whose geometric shape features contain the key feature dimension to form the point cloud subset of the key structural area, and calculate the vertex coordinates of the minimum bounding box of the point cloud subset as the spatial coordinate range.

[0100] Perform feature dimension association analysis on the structural point cloud units in the three-dimensional tunnel model dataset. Establish the correspondence between the geometric shape features of each structural point cloud point and the key feature dimension, that is, determine whether the geometric shape features of each point cloud point contain the key feature dimension. Screen out the point cloud points whose geometric shape features contain the key feature dimension, and these points form the point cloud subset of the key structural area.

[0101] For the point cloud subset, calculate the vertex coordinates of its minimum bounding box. The minimum bounding box is the smallest cube or cuboid that can completely enclose the point cloud subset, and its vertex coordinates represent the spatial coordinate range of the key structural area in the point cloud scanning data.

[0102] For example, through feature dimension association analysis, it is found that the geometric shape features of some point cloud points match the key feature dimension, and these points are screened out to form a point cloud subset. Then calculate the vertex coordinates of the minimum bounding box of the point cloud subset, and these coordinates can determine the spatial range of the key structural area in the point cloud scanning data.

[0103] Step 1432: Perform feature dimension association analysis processing on the contour line units in the three-dimensional tunnel model dataset, establish the correspondence between the structural association features of each contour line segment and the key feature dimension; screen out the contour line segments whose structural association features contain the key feature dimension to form the line segment subset of the key structural area, and extract the starting point coordinates and ending point coordinates of the line segment subset as the contour line segment coverage range.

[0104] Perform feature dimension correlation analysis on the contour line elements in the three-dimensional tunnel model dataset. Establish the correspondence between the structural correlation features of each contour line segment and the key feature dimensions, and determine whether the structural correlation features of each contour line segment contain the key feature dimensions. Screen out the contour line segments whose structural correlation features contain the key feature dimensions, and these line segments form a subset of the line segments in the key structural area. Extract the starting point coordinates and ending point coordinates of the line segment subset, and these coordinates determine the contour line segment coverage range of the key structural area in the tunnel construction design drawing. For example, through feature dimension correlation analysis, screen out the contour line segments whose structural correlation features match the key feature dimensions. Record the starting point coordinates and ending point coordinates of these line segments, so as to determine the contour coverage range of the key structural area in the design drawing.

[0105] Based on this, the method further includes: Step 1433: Perform spatial position alignment processing on the spatial coordinate range and the contour line segment coverage range, so that the spatial coordinates in the point cloud scan data and the contour line segment coordinates in the tunnel construction design drawing have a corresponding relationship in the same coordinate system, and obtain a spatial position alignment result.

[0106] After obtaining the spatial coordinate range of the key structural area in the point cloud scan data and the contour line segment coverage range in the tunnel construction design drawing, perform spatial position alignment processing on the two. Since the point cloud scan data and the tunnel construction design drawing may use different coordinate systems, it is necessary to unify their coordinate systems so that the two have a corresponding relationship in the same coordinate system.

[0107] During the spatial position alignment process, operations such as coordinate transformation, translation, and rotation may be required to ensure that the spatial coordinates in the point cloud scan data and the contour line segment coordinates in the tunnel construction design drawing can accurately correspond. Through spatial position alignment processing, a spatial position alignment result is obtained. For example, the origin of the coordinate system of the point cloud scan data may be different from the origin of the coordinate system of the tunnel construction design drawing. Through coordinate translation operations, the origins of the two are aligned so that the spatial coordinates and the contour line segment coordinates can be compared and analyzed in the same coordinate system.

[0108] Step 1434: Based on the spatial position alignment result, generate bimodal position description information including the point cloud coordinate range and the line segment coverage range, and the bimodal position description information is used to co-locate the key structural area from the perspective of multi-source data.

[0109] Generate bimodal position description information according to the spatial position alignment result. The bimodal position description information includes the point cloud coordinate range and the line segment coverage range, and co-locates the key structural area from the perspective of multi-source data. By combining the information of the point cloud scan data and the tunnel construction design drawing, the position of the key structural area can be determined more accurately.

[0110] For example, the bimodal position description information can be expressed as "Point cloud coordinate range: (x1, y1, z1) - (x2, y2, z2); Line segment coverage range: starting point (x3, y3) - ending point (x4, y4)". Such description information can simultaneously reflect the positions of the key structural areas in the point cloud scan data and the tunnel construction design drawings.

[0111] Step 144: Based on the spatial coordinate range and the contour line segment coverage range, calculate the geometric center coordinates and the boundary bounding box parameters of the key structural area as the spatial position information of the key structural area.

[0112] Using the spatial coordinate range and the contour line segment coverage range of the key structural area, calculate its geometric center coordinates and the boundary bounding box parameters. The geometric center coordinates can represent the central position of the key structural area, while the boundary bounding box parameters can describe the boundary range of this area.

[0113] When calculating the geometric center coordinates, it is necessary to comprehensively consider the spatial coordinate range in the point cloud scan data and the contour line segment coverage range in the tunnel construction design drawings. For the boundary bounding box parameters, in the point cloud scan data, they can be determined by calculating the size parameters of the minimum bounding box of the spatial coordinate range, and in the tunnel construction design drawings, they can be determined by calculating the total length of the line segments and the maximum included angle parameters of the contour line segment coverage range.

[0114] For example, by calculating the vertex coordinates of the spatial coordinate range, the geometric center coordinates are obtained. At the same time, calculate the length, width, and height of the minimum bounding box of the spatial coordinate range as the boundary bounding box parameters in the point cloud scan data, and calculate the total length of the line segments and the maximum included angle of the contour line segment coverage range as the boundary bounding box parameters in the design drawings.

[0115] In the following steps, the calculating the geometric center coordinates and the boundary bounding box parameters of the key structural area based on the spatial coordinate range and the contour line segment coverage range as the spatial position information of the key structural area includes: Step 1441: Perform a mean calculation process on the vertex coordinates of the minimum bounding box of the spatial coordinate range to obtain the first geometric center coordinates of the key structural area in the point cloud scan data; perform a midpoint calculation process on the starting point coordinates and the ending point coordinates of the contour line segment coverage range to obtain the second geometric center coordinates of the key structural area in the tunnel construction design drawings; perform a weighted average process on the first geometric center coordinates in the point cloud scan data and the second geometric center coordinates in the design drawings to generate the global geometric center coordinates that fuse multi-source data, and the weight coefficients of the weighted average are determined according to the credibility evaluation results of the multi-source data.

[0116] For the vertex coordinates of the minimum bounding box of the spatial coordinate range, calculate the mean value. Add the corresponding dimension values of all vertex coordinates and then divide by the number of vertices to obtain the first geometric center coordinate of the key structural area in the point cloud scan data. For the starting point coordinate and the ending point coordinate of the contour line segment coverage range, calculate their midpoint coordinate, that is, add the corresponding dimension values of the starting point coordinate and the ending point coordinate and divide by 2 to obtain the second geometric center coordinate of the key structural area in the tunnel construction design drawing.

[0117] Then, perform weighted average processing on the first geometric center coordinate and the second geometric center coordinate. The weight coefficients of the weighted average are determined according to the credibility evaluation results of multi-source data. If the credibility of the point cloud scan data is relatively high, then a larger weight can be given to the first geometric center coordinate during weighted average; conversely, if the credibility of the tunnel construction design drawing is relatively high, then a larger weight is given to the second geometric center coordinate. Through weighted average, the first geometric center coordinate and the second geometric center coordinate are fused to generate the global geometric center coordinate that integrates multi-source data. The global geometric center coordinate can comprehensively consider the information of the point cloud scan data and the tunnel construction design drawing and more accurately represent the central position of the key structural area. For example, if the acquisition accuracy of the point cloud scan data is high and the data quality is good, and its credibility is determined to be relatively high after credibility evaluation, then during weighted average, the weight of the first geometric center coordinate will be relatively large, making the global geometric center coordinate more biased towards the central position reflected by the point cloud scan data.

[0118] Step 1442: Calculate the size parameters of the minimum bounding box of the spatial coordinate range as the boundary bounding box parameters in the point cloud scan data; calculate the total line segment length and the maximum included angle parameter of the contour line segment coverage range as the boundary bounding box parameters in the design drawing.

[0119] Optionally, for the spatial coordinate range in the point cloud scan data, calculate the size parameters of its minimum bounding box. The size parameters of the minimum bounding box include information such as length, width, and height, which can describe the boundary range of the key structural area in the point cloud scan data. When calculating, first determine the maximum and minimum coordinate values of the minimum bounding box in each coordinate axis direction, and then calculate the differences between them to obtain the length, width, and height. For example, in the X-axis direction, find the maximum and minimum X coordinates of all points in the spatial coordinate range, and the difference between the two is the length of the minimum bounding box in the X-axis direction.

[0120] For the coverage range of the contour line segments in the tunnel construction design drawings, calculate the total length of the line segments and the maximum included angle parameter. The total length of the line segments is the sum of the lengths of all line segments within the coverage range of the contour line segments, reflecting the overall length scale of the key structural area in the design drawings. The maximum included angle parameter is the maximum included angle between the line segments within the coverage range of the contour line segments, reflecting the shape characteristics of the key structural area. For example, if the maximum included angle is close to 180 degrees, it indicates that the shape of the area is relatively long and narrow; if the maximum included angle is small, it indicates that the shape of the area is relatively compact.

[0121] Step 1443: Combine and process the global geometric center coordinates with the bounding box parameters in the point cloud scan data and the bounding box parameters in the tunnel construction design drawings to generate spatial position information including the center position and the boundary range, and the spatial position information is used to mark the spatial distribution of the key structural area.

[0122] Combine the global geometric center coordinates obtained by fusing multi-source data with the bounding box parameters in the point cloud scan data and the bounding box parameters in the tunnel construction design drawings. Through the combination process, integrate the center position and boundary range information of the key structural area to form spatial position information including the center position and the boundary range, and this spatial position information can comprehensively describe the distribution of the key structural area in the three-dimensional space. For example, the spatial position information can be expressed as "Global geometric center coordinates: (x0, y0, z0); Point cloud scan data bounding box parameters: length l1, width w1, height h1; Tunnel construction design drawing bounding box parameters: Total line segment length L, Maximum included angle α", and this information can be used to accurately mark the spatial distribution of the key structural area in the three-dimensional tunnel model.

[0123] Step 145: Perform an association and binding process on the feature type label and the spatial position information to generate a preliminary annotation result including the label-coordinate correspondence.

[0124] Associate the feature type labels assigned to the key structure regions with the calculated spatial location information. The association and binding process is to establish a one-to-one correspondence between the feature type labels and the spatial location information, so that the feature type of each key structure region and its position in the three-dimensional space can correspond to each other. Through the association and binding, a preliminary annotation result containing the label-coordinate correspondence relationship is generated. For example, for a key structure region, its feature type label is "tunnel ventilation opening structure", and the spatial location information is "global geometric center coordinates: (x0, y0, z0); point cloud scan data bounding box parameters: length l1, width w1, height h1; tunnel construction design drawing bounding box parameters: total line segment length L, maximum angle α". After associating and binding the two, the preliminary annotation result can be recorded as "label: tunnel ventilation opening structure; coordinate information: (x0, y0, z0), point cloud boundary (l1, w1, h1), drawing boundary (L, α)". This preliminary annotation result provides a basis for subsequent processing, enabling the features and location information of each key structure region to be clearly presented.

[0125] Step 146: Perform duplicate region detection processing on the preliminary annotation result, merge the annotation information of key structure regions with overlapping spatial coordinate ranges, and generate a deduplicated feature recognition annotation result.

[0126] Optionally, there may be key structure regions with overlapping spatial coordinate ranges in the preliminary annotation result, and these overlapping regions may be caused by data acquisition errors, inaccurate feature extraction, etc. To improve the accuracy and conciseness of the annotation result, duplicate region detection processing needs to be performed on the preliminary annotation result.

[0127] As an embodiment, the performing duplicate region detection processing on the preliminary annotation result, merging the annotation information of key structure regions with overlapping spatial coordinate ranges, and generating a deduplicated feature recognition annotation result includes: Step 1461: Perform boundary expansion processing on the spatial coordinate range of each key structure region in the preliminary annotation result to generate an extended coordinate range including a buffer region, and the width of the buffer region is adjusted according to the size of the key structure region.

[0128] Expand the spatial coordinate ranges of each key structural region in the preliminary annotation results. The purpose of boundary expansion is to more accurately detect the overlap between regions and avoid missed detections caused by data precision issues. During the expansion process, an expanded coordinate range including a buffer region is generated. The width of the buffer region should be adjusted according to the size of the key structural region. Generally speaking, the larger the size of the key structural region, the larger the width of the buffer region will be. For example, for a key structural region with a relatively large size, the width of its buffer region can be set relatively wide to ensure the potential overlap with other regions can be detected; while for a key structural region with a relatively small size, the width of the buffer region can be appropriately reduced. Through boundary expansion processing, the spatial relationship between key structural regions can be considered more comprehensively.

[0129] Step 1462: Calculate the ratio of the intersection area to the union area between the expanded coordinate ranges of different key structural regions as the regional overlap degree index; screen out the pairs of key structural regions whose regional overlap degree index exceeds the preset overlap threshold to form a set of duplicate regions to be merged; perform a consistency verification process on the feature type labels of the key structural regions in the set of duplicate regions, retain the pairs of duplicate regions with exactly the same labels, and delete the pairs of duplicate regions with inconsistent labels; perform a union calculation process on the spatial coordinate ranges of the retained pairs of duplicate regions to generate a merged spatial coordinate range, and the merged spatial coordinate range includes all the point cloud points and contour line segments of the original two regions.

[0130] Calculate the ratio of the intersection area to the union area between the expanded coordinate ranges of different key structural regions as the regional overlap degree index. The regional overlap degree index can quantify the overlap degree between different key structural regions. By comparing the regional overlap degree index with the preset overlap threshold, screen out the pairs of key structural regions whose regional overlap degree index exceeds the preset overlap threshold, and these pairs of key structural regions form a set of duplicate regions to be merged. The preset overlap threshold is a pre-set standard used to judge whether the overlap between regions reaches the level that needs to be merged. For example, if the preset overlap threshold is 0.5, when the regional overlap degree index of two key structural regions is greater than 0.5, it is considered that their overlap degree is relatively high and needs to be merged.

[0131] For the key structural regions in the set of duplicate regions to be merged, perform consistency verification on their feature type labels. Only when the feature type labels of two key structural regions are exactly the same, the duplicate region pair is retained; if the labels are inconsistent, the duplicate region pair is deleted because inconsistent labels may indicate that these two regions actually represent different types of structures and should not be merged. For example, if the feature type label of one region is "Tunnel main structure - Vault" and the feature type label of another region is "Tunnel auxiliary facilities - Lighting equipment", since the labels of these two regions are inconsistent, they should be deleted from the set of duplicate regions to be merged.

[0132] For the retained duplicate region pairs, perform union calculation on their spatial coordinate ranges. Union calculation combines all the point cloud points and contour line segments of the original two regions together to generate the merged spatial coordinate range. The merged spatial coordinate range can contain all the spatial information of the original two regions, avoiding information loss. For example, through union calculation, the point cloud points and contour line segments of two overlapping regions are integrated into a new spatial coordinate range, and this new range represents the spatial range of the merged key structural region.

[0133] Step 1463: Perform deduplication and merging processing on the feature type labels of the retained duplicate region pairs to generate a unified feature type label; perform association and binding processing on the merged spatial coordinate range and the unified feature type label to generate the deduplicated feature recognition annotation result.

[0134] Perform deduplication and merging processing on the feature type labels of the retained duplicate region pairs. Since the feature type labels of these duplicate region pairs have passed consistency verification, they can be directly merged to generate a unified feature type label. For example, if the feature type labels of two duplicate regions are both "Tunnel main structure - Support column", then the unified feature type label after merging is still "Tunnel main structure - Support column".

[0135] Perform association and binding processing on the merged spatial coordinate range and the unified feature type label. Through association and binding, the merged spatial position information and the unified feature type label are corresponding to generate the deduplicated feature recognition annotation result. The deduplicated feature recognition annotation result is more accurate and concise, avoiding the problem of duplicate annotation, and can more clearly reflect the features and location information of the key structural regions in the tunnel. For example, the deduplicated feature recognition annotation result can clearly display the unique feature type label and accurate spatial position information of each key structural region.

[0136] As a non-limiting embodiment, the method further includes: obtaining the correspondence between the spatial position information of each key structural area in the feature recognition annotation result in the point cloud scan data and the coverage range of the contour line segments in the tunnel construction design drawing, and constructing a multimodal annotation mapping table; performing cross-modal feature alignment processing on the multimodal annotation mapping table, and extracting the overlapping area features and the differential area features between the point cloud spatial coordinates and the drawing contour coordinates; based on the overlapping area features and the differential area features, calculating the label consistency score and the position deviation amount between the point cloud annotation and the drawing annotation respectively; inputting the label consistency score and the position deviation amount into a preset credibility evaluation algorithm to generate the multimodal credibility score of each key structural area; screening and correcting the feature recognition annotation result according to the credibility score, retaining the annotation areas with credibility scores higher than the preset threshold, and adjusting the feature type labels or the spatial position information of the annotation areas with credibility scores lower than the threshold to generate a multimodal calibrated feature recognition annotation result.

[0137] First, obtain the correspondence between the spatial position information of each key structural area in the feature recognition annotation result in the point cloud scan data and the coverage range of the contour line segments in the tunnel construction design drawing, and construct a multimodal annotation mapping table. The multimodal annotation mapping table can associate the annotation information of the point cloud scan data and the tunnel construction design drawing, facilitating subsequent processing and analysis. For example, the multimodal annotation mapping table can record the correspondence between the spatial coordinate range of each key structural area in the point cloud scan data and the coverage range of the contour line segments in the tunnel construction design drawing, as well as the corresponding feature type labels.

[0138] Second, perform cross-modal feature alignment processing on the multimodal annotation mapping table. Cross-modal feature alignment is to match and align the point cloud spatial coordinates and the drawing contour coordinates to find the overlapping areas and the differential areas between them. In this process, extract the overlapping area features and the differential area features between the point cloud spatial coordinates and the drawing contour coordinates. The overlapping area features reflect the consistency between the point cloud scan data and the tunnel construction design drawing in certain areas, while the differential area features reflect the differences between the two. For example, in some key structural areas, there may be certain deviations between the coordinates of the point cloud scan data and the tunnel construction design drawing, and the cross-modal feature alignment can accurately find these deviation areas.

[0139] Furthermore, based on the overlapping area features and the differential area features, calculate the label consistency score and the position deviation amount between the point cloud annotation and the drawing annotation respectively. The label consistency score is used to measure the degree of consistency of the feature type labels between the point cloud annotation and the drawing annotation, and the position deviation amount is used to represent the deviation size of the spatial positions between the point cloud annotation and the drawing annotation. For example, if the feature type labels of the point cloud annotation and the drawing annotation are exactly the same, then the label consistency score will be higher; if the spatial position deviation between the two is larger, the position deviation amount will be larger.

[0140] Further, the label consistency score and the position deviation amount are input into a preset credibility evaluation algorithm to generate a multi-modal credibility score for each key structural region. The credibility evaluation algorithm comprehensively considers the label consistency score and the position deviation amount to evaluate the credibility of the annotation information for each key structural region. The multi-modal credibility score can reflect the reliability of the annotation information for each key structural region. For example, if a key structural region has a high label consistency score and a small position deviation amount, then its multi-modal credibility score will be relatively high.

[0141] Finally, the feature recognition annotation results are screened and corrected according to the credibility scores. The annotation regions with credibility scores higher than the preset threshold are retained. The information in these annotation regions is relatively reliable and can be directly retained. For the annotation regions with credibility scores lower than the threshold, the feature type labels or spatial position information need to be adjusted. For example, if the credibility score of a certain key structural region is low and it is found through analysis that it is due to a large position deviation between the point cloud annotation and the drawing annotation, then its spatial position information can be corrected according to the actual situation. Through the screening and correction process, a multi-modal calibrated feature recognition annotation result is generated, and the multi-modal calibrated feature recognition annotation result is more accurate and reliable.

[0142] As another non-limiting embodiment, the method further includes: extracting the feature type labels and spatial position information of all key structural regions from the feature recognition annotation results to construct a structured annotation database including region identifiers, label categories, and spatial coordinates; performing region association analysis processing on the structured annotation database to identify groups of key structural regions having functional cooperation relationships or spatial adjacency relationships; generating an associated feature description reflecting the interaction pattern between regions according to the semantic relevance of the feature type labels and the topological relevance of the spatial positions of the regions within the group of key structural regions; performing a fusion and extension process on the associated feature description and the original annotation information of each region within the group of key structural regions to supplement the group of key structural regions with additional labels including the interaction pattern, and the additional labels are used to indicate the cooperative function or spatial constraint relationship between regions, so as to obtain the fused and extended annotation information; updating the structured annotation database based on the fused and extended annotation information to generate an enhanced feature recognition annotation result including region association information.

[0143] Extract the feature type labels and spatial position information of all key structural regions from the feature recognition annotation results to construct a structured annotation database. The structured annotation database includes information such as region identifiers, label categories, and spatial coordinates, and can store the annotation information of key structural regions in a structured manner, facilitating subsequent query and analysis. For example, the structured annotation database can use the numbers of key structural regions as region identifiers to record the feature type labels and spatial position information of each region.

[0144] Perform regional association analysis on the structured annotation database. By analyzing the feature type tags and spatial location information of key structural regions, identify groups of key structural regions with functional collaboration relationships or spatial adjacency relationships. A functional collaboration relationship means that different key structural regions cooperate and work together functionally, while a spatial adjacency relationship means that they are adjacent or close in space. For example, the ventilation system and lighting system of a tunnel may have a functional collaboration relationship, and two adjacent support structures of the tunnel have a spatial adjacency relationship.

[0145] Generate an association feature description reflecting the interaction pattern between regions based on the semantic relevance of the feature type tags and the topological relevance of the spatial locations of the regions within the group of key structural regions. The association feature description can detail the interaction methods and relationships between regions. For example, if a group of key structural regions includes a ventilation system and a lighting system, the association feature description can explain how the operation of the ventilation system affects the heat dissipation of the lighting system, or how the layout of the lighting system matches the duct design of the ventilation system.

[0146] Perform a fusion and extension process on the association feature description and the original annotation information of each region within the group of key structural regions, and supplement the group of key structural regions with additional tags containing the interaction pattern. The additional tags are used to indicate the collaborative functions or spatial constraint relationships between regions. For example, for a group of key structural regions including a ventilation system and a lighting system, an additional tag "ventilation-lighting collaboration" can be added to indicate their collaborative function. Through the fusion and extension process, the fused and extended annotation information is obtained.

[0147] Update the structured annotation database based on the fused and extended annotation information to generate an enhanced feature recognition annotation result containing regional association information. The enhanced feature recognition annotation result not only includes the basic annotation information of the key structural regions but also adds the association information between regions, and can more comprehensively reflect the overall situation of the tunnel structure. For example, in the enhanced feature recognition annotation result, the association relationship between each key structural region and other regions can be clearly seen, providing richer information for the maintenance, management, and optimization of the tunnel.

[0148] As another non-limiting embodiment, the method further includes: generating a historical annotation sequence arranged in chronological order based on the feature recognition annotation results generated in each construction stage within the full construction cycle of the tunnel; performing a time-series feature encoding process on the historical annotation sequence to extract the evolution trajectory of the feature types and the spatial position migration path of the key structural regions in each time step; inputting the time-series feature encoding results and the currently generated feature recognition annotation results into a dynamic evolution prediction model, and combining time-dependent relationship modeling processing to predict the evolution pattern of the feature types and the spatial position migration trend of each key structural region in the subsequent construction stage; and associating the evolution pattern of the feature types and the spatial position migration trend as dynamic evolution information to the corresponding key structural regions of the currently generated feature recognition annotation results to generate a time-series extended feature recognition annotation result including full-cycle evolution prediction.

[0149] Based on the feature recognition annotation results generated in each construction stage within the full construction cycle of the tunnel, a historical annotation sequence arranged in chronological order is generated. The historical annotation sequence records the annotation information of the key structural regions of the tunnel in different construction stages, including feature type labels and spatial position information. For example, the historical annotation sequence can sequentially record the annotation conditions of the key structural regions in each stage according to the chronological order of the construction stages, forming a time-series data.

[0150] Perform a time-series feature encoding process on the historical annotation sequence. Time-series feature encoding is to convert and encode the information in the historical annotation sequence for subsequent analysis and processing. In this process, the evolution trajectory of the feature types and the spatial position migration path of the key structural regions in each time step are extracted. The evolution trajectory of the feature types reflects the changes in the feature types of the key structural regions in different construction stages, and the spatial position migration path reflects the movement of the spatial position of the key structural regions over time. For example, in the early stage of tunnel construction, a certain key structural region may be a temporary support structure, and as the construction progresses, it may evolve into a permanent structure. The time-series feature encoding can accurately record this evolution of the feature types.

[0151] Input the time-series feature encoding results and the currently generated feature recognition annotation results into a dynamic evolution prediction model, and combine time-dependent relationship modeling processing. The dynamic evolution prediction model is a model based on machine learning or deep learning that can learn the time-dependent relationships in historical data and predict the evolution pattern of the feature types and the spatial position migration trend of each key structural region in the subsequent construction stage according to these relationships. Time-dependent relationship modeling is to analyze the variation laws of the feature types and spatial positions of the key structural regions over time. For example, the spatial positions of some key structural regions may gradually move towards the interior of the tunnel as the construction progresses.

[0152] Taking the characteristic type evolution pattern and the spatial position migration trend as dynamic evolution information, associating them with the corresponding key structure regions in the current feature recognition and annotation results. Through this association, the predicted dynamic evolution information is combined with the current annotation results to generate a time-series extended feature recognition and annotation result that includes full-cycle evolution prediction. This result not only contains the annotation information of the current key structure regions but also adds prediction information about their future evolution.

[0153] For example, in the current feature recognition and annotation results, a certain key structure region is annotated as "primary support structure". Through the dynamic evolution prediction model, it is predicted that in the subsequent construction stage, this region will gradually evolve into a "secondary lining structure", and its spatial position will shift a certain distance towards the interior of the tunnel as the lining construction progresses. Taking these characteristic type evolution patterns and spatial position migration trends as dynamic evolution information, they are associated with the "primary support structure" key structure region in the current annotation results. In this way, in the time-series extended feature recognition and annotation results, the annotation for this region will not only include the current "primary support structure" label and spatial position information but also additional prediction information about evolving into a "secondary lining structure" in the future and position migration.

[0154] From the perspective of the dynamic evolution prediction model, its construction process needs to consider multiple aspects. First, the input data of the model includes the time-series feature encoding results and the currently generated feature recognition and annotation results. The time-series feature encoding results are obtained by processing the historical annotation sequences and record the feature evolution and position migration information of the key structure regions in past construction stages. The currently generated feature recognition and annotation results reflect the latest situation in the current construction stage.

[0155] In the embodiment of the present invention, the dynamic evolution prediction model internally includes multiple modules. Among them, the feature extraction module is responsible for extracting useful feature information from the input data, such as the feature type, spatial coordinates, time step, etc. of the key structure regions. These features will be further passed to the time-dependent modeling module, which analyzes the time-dependent relationships in the historical data to learn the laws of the features and positions of the key structure regions changing over time. For example, by analyzing the data of multiple construction stages, it is found that there is a certain periodicity in the evolution of the feature types of some key structure regions, or there is a linear relationship between the spatial position migration and the construction progress, etc.

[0156] Based on the learned time-dependent relationships, the prediction module of the model will predict the characteristic type evolution patterns and spatial position migration trends of each key structure region in the subsequent construction stages. The prediction module will combine the current construction state and historical laws to output the prediction results of each key structure region at different future time steps.

[0157] When training a dynamic evolution prediction model, a large amount of historical labeled data is required as training samples. These data can come from the actual records of multiple tunnel construction projects or virtual data generated by simulating the tunnel construction process. During the training process, appropriate training parameters need to be set, such as the learning rate, the number of iterations, etc. The learning rate controls the step size of the model to update parameters in each iteration, and the number of iterations determines the number of rounds of model training. By continuously adjusting the training parameters, the model can accurately learn the time-dependent relationships and improve the prediction accuracy.

[0158] The application of the recognition and annotation results of time-series extended features is also very extensive. In terms of tunnel construction planning, construction personnel can arrange the procurement of construction materials and the deployment of construction equipment in advance according to the predicted evolution pattern of feature types and the migration trend of spatial positions. For example, if it is predicted that a certain key structural area will evolve into a structure that requires special materials in the future, the corresponding materials can be purchased in advance to avoid construction delays.

[0159] In terms of construction quality monitoring, by comparing the actual construction situation with the prediction results, it is possible to timely detect whether there are deviations in the construction process. If the actual feature type or spatial position of a certain key structural area does not match the prediction results, it is necessary to inspect and adjust the construction process to ensure that the construction quality meets the requirements.

[0160] In the tunnel operation and maintenance stage, the recognition and annotation results of time-series extended features can also provide important references. According to the predicted evolution of feature types and location migration information, a reasonable maintenance plan can be formulated to inspect and maintain the areas that may have problems in advance, thereby extending the service life of the tunnel.

[0161] In addition, the recognition and annotation results of time-series extended features can also support the digital management of tunnel construction. Integrating these annotation results into the tunnel construction management system enables real-time monitoring and dynamic management of the tunnel construction process. Through the visualization interface, construction management personnel can intuitively view the current status and future prediction information of each key structural area and make decisions in a timely manner.

[0162] Thus, starting from obtaining the tunnel construction design drawings and point cloud scanning data, through processes such as preprocessing, feature extraction, joint analysis, recognition and annotation, etc., the recognition and annotation results of time-series extended features including full-cycle evolution prediction are finally obtained. This result synthesizes the information of multi-source data and can not only accurately reflect the characteristics and positions of key structural areas in the current tunnel construction stage, but also predict future evolutions, improving the efficiency, quality and safety of tunnel construction.

[0163] In the embodiments of the present invention, data preprocessing and feature extraction can be performed based on standard point cloud processing libraries (such as Point Cloud Library, PCL) and deep learning frameworks (such as PyTorch or TensorFlow) in the prior art. For example, in the noise filtering process of the structural point cloud unit, referring to the statistical outlier removal or radius filtering algorithm in PCL, the neighborhood radius and standard deviation threshold are set according to the point cloud density to eliminate outliers and redundant points, generating a smoothed subset; for the local curvature calculation of geometric shape features, drawing on the surface fitting method of PCL, the curvature parameters and normal vector directions are directly output through the eigenvalue decomposition of the neighborhood point covariance matrix. In the structural association feature modeling of the contour line unit, the line segment detection and topological analysis tools of OpenCV are used to automatically extract the line segment length and angle, and the distribution density is calculated based on the line segment intersection point spatial clustering algorithm. At the same time, the angle histogram matching is combined to verify the adjacency relationship, ensuring the repeatability of feature extraction.

[0164] The implementation of the deep learning network can be based on the convolutional neural network architecture. The multi-scale feature fusion layer uses standard pyramid pooling or dilated convolution operations for downsampling according to the preset resolution rules. The window division of the local detail feature stream directly uses the sliding window or region growing algorithm to segment the point cloud and line segments. The local feature enhancement subnet integrates lightweight network modules for non-linear activation processing and calls the activation function to explicitly select public function types such as ReLU or Sigmoid. The self-attention mechanism of the global context feature stream refers to the standard implementation of the Transformer model. After mapping the features into query-key-value vectors, the similarity is calculated through dot product operations, and the attention weight matrix is generated using softmax normalization. Combining with residual connections to optimize gradient propagation, and at the same time, the dimensionality reduction processing uses principal component analysis or multi-layer perceptron for dimensionality reduction, ensuring the clear definition of the model structure and the reproducibility of the model.

[0165] The calculation of the spatial position of the key structural area can integrate GIS library functions, fuse the point cloud and design drawing data through weighted average coordinates, and the weight coefficient is dynamically set based on the data acquisition accuracy report. The boundary bounding box parameters directly output the size values by calling the bounding box generation algorithm. The spatial position alignment process uses the ICP algorithm or coordinate transformation matrix to unify the coordinate system to achieve dimensional consistency processing. In the repeated area detection process, the width of the buffer is adaptively adjusted according to the point cloud density, the IoU index is used for the calculation of the area overlap degree, and the logical rigor is ensured through label consistency matching. The confidence evaluation combines the feature response distribution and position deviation for multi-modal calibration.

[0166] The embodiment of the present invention processes the tunnel construction design drawings and the point cloud scanning data of the corresponding construction stage to obtain a three-dimensional tunnel model data set with a spatial coordinate alignment relationship, thereby realizing the effective fusion of multi-source data, so that the structural point cloud unit and the contour line unit of the design drawing vectorization complement each other; three-dimensional structural feature extraction is performed on the three-dimensional tunnel model data set to obtain the geometric morphological features of the structural point cloud unit and the structural association features of the contour line unit, respectively, to deeply analyze the tunnel structure from different angles, and to provide rich feature information for accurately identifying key structural areas; the two features are jointly analyzed and processed by a deep learning network, giving full play to the powerful feature learning ability of deep learning, being able to mine the potential relationship between the features, and generating a feature analysis result containing a feature response distribution, so as to accurately judge the importance of different structural features; based on the feature analysis result, the key structural area is identified and labeled, and a feature recognition and labeling result containing a feature type label and spatial position information is generated, which provides an intuitive and detailed basis for the monitoring, management and maintenance of tunnel construction, helps to improve construction efficiency and quality, and ensures the safe and stable operation of the tunnel.

[0167] See also Figure 2 As shown, this figure is a schematic diagram of the basic structure of a feature recognition and annotation system 200 provided by an embodiment of the present invention. The feature recognition and annotation system 200 includes: Processor 201; a storage device 202 on which a computer program 2020 is stored; When the computer program 2020 is executed by the processor 201, the processor 201 implements any of the feature recognition and annotation methods for three-dimensional tunnel modeling.

[0168] Based on the above, a readable storage medium is provided, on which a program or instruction is stored, and when the program or instruction is executed by a processor, the steps of the above method are implemented.

[0169] It should be noted that the various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and the same or similar parts between the various embodiments can be referred to each other. For the system or device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description.

Claims

1. A feature recognition and annotation method applied to three-dimensional tunnel modeling, characterized in that, The method includes: Preprocessing the obtained tunnel construction design drawings and point cloud scanning data of the corresponding construction stage to obtain a three-dimensional tunnel model dataset with spatially coordinate-aligned relationships. The three-dimensional tunnel model dataset includes a structural point cloud unit with multi-source data fusion and a contour line unit with vectorized design drawings; Performing three-dimensional structural feature extraction processing on the three-dimensional tunnel model dataset to obtain the geometric shape features of the structural point cloud unit and the structural association features of the contour line unit; Performing joint analysis processing on the geometric shape features and the structural association features through a deep learning network to generate a feature analysis result including a feature response distribution; Based on the feature analysis result, identifying and labeling key structural regions in the three-dimensional tunnel model dataset to generate a feature recognition and labeling result including feature type labels and spatial position information.

2. The method according to claim 1, wherein The performing three-dimensional structural feature extraction processing on the three-dimensional tunnel model dataset to obtain the geometric shape features of the structural point cloud unit and the structural association features of the contour line unit includes: Performing noise filtering processing on the structural point cloud unit for outliers and redundant points to obtain a smoothed structural point cloud subset; Performing neighborhood analysis processing on the smoothed structural point cloud subset, calculating the local curvature parameters and normal vector directions of each point cloud point, and generating geometric shape features reflecting the undulating state of the structural surface. The geometric shape features include the curvature distribution pattern of the point cloud points and the normal vector consistency index; Performing line segment connectivity analysis processing on the contour line unit, identifying the contour line segments in the tunnel construction design drawings, and extracting the length parameters and angle parameters of the contour line segments as basic structural features; Based on the basic structural features, performing structural association modeling processing on the contour line unit, analyzing the intersection relationship and adjacency relationship between different contour line segments, and generating structural association features reflecting the connection method of tunnel structural components. The structural association features include the distribution density of line segment intersection points and the angle matching degree of adjacent line segments; The method further includes: Performing unified dimension conversion processing on the geometric shape features and the structural association features to obtain a target feature set with the same feature dimension. The information entropy calculation result of the target feature set is used to adjust the feature fusion weight parameters in the joint analysis processing of the deep learning network.

3. The method according to claim 1, characterized in that, The performing joint analysis processing on the geometric shape features and the structural association features through a deep learning network to generate a feature analysis result including a feature response distribution includes: Inputting the geometric shape features and the structural association features into the multi-scale feature fusion layer of the deep learning network, and performing multi-resolution downsampling processing on the features in combination with a preset scale division rule to generate a multi-scale feature stream; For the local detail feature stream in the multi-scale feature stream, using convolutional kernels with different receptive fields to perform feature extraction processing, capturing the local curvature change details of the point cloud in the geometric shape features and the line segment adjacency relationship details in the structural association features, and obtaining a local detail feature output; For the global context feature stream in the multi-scale feature stream, the self-attention mechanism is used to model the global dependency relationship between the geometric shape features and the structural association features, generating a context feature vector reflecting the overall structural layout of the tunnel; The local detail feature output and the context feature vector are input into the feature interaction network, and information complementarity is performed through element-wise product operations to generate an interaction feature map with spatio-temporal consistency constraints; The interaction feature map is subjected to channel normalization processing, and the normalized interaction feature map is concatenated along the channel dimension to generate a feature fusion vector containing multi-scale interaction information; The response distribution generation layer of the deep learning network is called to perform activation function mapping processing on the feature fusion vector, generating a feature response distribution reflecting the importance degree of different structural features, and the feature response distribution includes the activation values of each feature dimension and the corresponding feature type identifiers.

4. The method according to claim 3, wherein For the local detail feature stream in the multi-scale feature stream, convolution kernels with different receptive fields are used for feature extraction processing to capture the local curvature change details of the point cloud in the geometric shape features and the segment adjacency relationship details in the structural association features, obtaining the local detail feature output, including: The geometric shape features of the local detail feature stream are subjected to window division processing to generate multiple local point cloud window units containing a preset number of point cloud points; Statistical analysis processing is performed on the point cloud point curvature distribution pattern and the normal vector consistency index within each local point cloud window unit, and the average value and standard deviation of the curvature within the window are calculated as the local curvature change statistical features; The structural association features of the local detail feature stream are subjected to line segment segmentation processing to generate multiple local line segment group units containing continuous line segments; Sequence analysis processing is performed on the line segment length parameter and the included angle parameter within each local line segment group unit, and the increasing and decreasing pattern of the line segment length and the periodic change pattern of the included angle are extracted as the local line segment adjacency relationship features; The local curvature change statistical features and the local line segment adjacency relationship features are input into the local feature enhancement subnet to generate an enhanced local feature vector containing multi-source local detail information; Nonlinear activation processing is performed on the enhanced local feature vector to obtain the local detail feature output reflecting the local curvature change details of the point cloud and the segment adjacency relationship details; For the global context feature stream in the multi-scale feature stream, the self-attention mechanism is used to model the global dependency relationship between the geometric shape features and the structural association features, generating a context feature vector reflecting the overall structural layout of the tunnel, including: The geometric shape features and the structural association features are respectively mapped into query vectors, key vectors, and value vectors. The query vector is used to represent the feature information at the target feature position, the key vector is used to represent the feature information at the non-target feature position, and the value vector is used to represent the feature value at the non-target feature position; Calculate the dot product similarity between the query vector and the key vector to generate an attention weight matrix reflecting the association degree between feature positions; Perform softmax normalization on the attention weight matrix to obtain the attention distribution probability of each feature position relative to the target feature position; Perform weighted summation on the value vector based on the attention distribution probability to generate a context-enhanced feature vector containing global feature position dependencies; Perform residual connection processing on the context-enhanced feature vector, the geometric shape feature, and the structural association feature to obtain an initial context feature vector integrating global dependencies; Perform dimensionality compression processing on the initial context feature vector to generate a context feature vector reflecting the overall structural layout of the tunnel.

5. The method according to claim 1, wherein Based on the feature analysis results, identify and label key structural regions in the 3D tunnel model dataset to generate a feature recognition and labeling result containing feature type labels and spatial location information, including: According to the feature response distribution in the feature analysis results, extract feature dimensions with activation values exceeding a preset threshold as key feature dimensions, and the key feature dimensions correspond to the key structural regions of the tunnel; According to the feature type identifiers corresponding to the key feature dimensions, assign corresponding feature type labels to each key structural region, and the feature type labels include structural component types and functional attribute categories; Extract the structural point cloud units and contour line units associated with the key feature dimensions in the 3D tunnel model dataset, and determine the spatial coordinate range of the key structural regions in the point cloud scan data and the contour line segment coverage range in the tunnel construction design drawings; Based on the spatial coordinate range and contour line segment coverage range, calculate the geometric center coordinates and boundary bounding box parameters of the key structural regions as the spatial location information of the key structural regions; Perform association and binding processing on the feature type labels and the spatial location information to generate a preliminary labeling result containing the label-coordinate correspondence; Perform duplicate region detection processing on the preliminary labeling result, and merge the labeling information of key structural regions with overlapping spatial coordinate ranges to generate a deduplicated feature recognition and labeling result.

6. The method according to claim 5, characterized in that The step of assigning corresponding feature type labels to each key structural region according to the feature type identifiers corresponding to the key feature dimensions includes: Establish a mapping relationship between the feature type identifiers and a preset label library, and the preset label library includes tunnel main structure labels, accessory facility structure labels, and potential disease structure labels; Extract the peak position of the activation value in the feature response distribution of the key feature dimension to determine the core feature type identifier corresponding to the key feature dimension; Search for the corresponding basic label category in the preset label library according to the core feature type identifier, and the basic label category includes the first-level label of the structural component type and the second-level label of the functional attribute category; Based on the sub-peak position in the feature response distribution of the key feature dimension, extract the auxiliary feature type identifier reflecting the additional attributes of the feature; Search for the corresponding supplementary label information in the preset label library according to the auxiliary feature type identifier, and the supplementary label information includes additional labels of the structural material type and the construction process category; Combine and splice the basic label category and the supplementary label information to generate a feature type label including multiple label levels, where the multiple label levels are used to describe the structural attributes and functional attributes of the key structure area.

7. The method according to claim 5, wherein Extract the structural point cloud units and contour line units associated with the key feature dimension from the three-dimensional tunnel model dataset, and determine the spatial coordinate range of the key structure area in the point cloud scan data and the contour line segment coverage range in the tunnel construction design drawings, including: Perform feature dimension association analysis processing on the structural point cloud units in the three-dimensional tunnel model dataset, establish the correspondence between the geometric shape features of each structural point cloud point and the key feature dimension; screen out the point cloud points whose geometric shape features include the key feature dimension to form the point cloud subset of the key structure area, and calculate the vertex coordinates of the minimum bounding box of the point cloud subset as the spatial coordinate range; Perform feature dimension association analysis processing on the contour line units in the three-dimensional tunnel model dataset, establish the correspondence between the structural association features of each contour line segment and the key feature dimension; screen out the contour line segments whose structural association features include the key feature dimension to form the line segment subset of the key structure area, and extract the starting point coordinates and ending point coordinates of the line segment subset as the contour line segment coverage range; The method further includes: Perform spatial position alignment processing on the spatial coordinate range and the contour line segment coverage range, so that the spatial coordinates in the point cloud scan data and the contour line segment coordinates in the tunnel construction design drawings have a corresponding relationship in the same coordinate system, and obtain the spatial position alignment result; Based on the spatial position alignment result, generate dual-modal position description information including the point cloud coordinate range and the line segment coverage range, and the dual-modal position description information is used to jointly locate the key structure area from the perspective of multi-source data.

8. The method according to claim 5, characterized in that, Based on the spatial coordinate range and the contour line segment coverage range, calculate the geometric center coordinates and the boundary bounding box parameters of the key structure area as the spatial position information of the key structure area, including: Perform mean calculation processing on the vertex coordinates of the minimum bounding box of the spatial coordinate range to obtain the first geometric center coordinates of the key structure area in the point cloud scan data; Perform midpoint calculation processing on the starting point coordinates and the ending point coordinates of the contour line segment coverage range to obtain the second geometric center coordinates of the key structure area in the tunnel construction design drawings; Perform weighted average processing on the first geometric center coordinates in the point cloud scan data and the second geometric center coordinates in the design drawings to generate the global geometric center coordinates integrating multi-source data, and the weight coefficient of the weighted average is determined according to the credibility evaluation result of the multi-source data; Calculate the size parameters of the minimum bounding box of the spatial coordinate range as the boundary bounding box parameters in the point cloud scan data; Calculate the total line segment length and the maximum included angle parameters of the contour line segment coverage range as the boundary bounding box parameters in the design drawings; Combine the global geometric center coordinates with the bounding box parameters in the point cloud scan data and the bounding box parameters in the tunnel construction design drawings to generate spatial position information including the center position and the boundary range, and the spatial position information is used to mark the spatial distribution of key structural areas.

9. The method according to claim 5, wherein Perform duplicate area detection processing on the preliminary annotation results, and merge the annotation information of key structural areas with overlapping spatial coordinate ranges to generate a deduplicated feature recognition annotation result, including: Perform boundary expansion processing on the spatial coordinate range of each key structural area in the preliminary annotation results to generate an expanded coordinate range including a buffer area, and the width of the buffer area is adjusted according to the size of the key structural area; Calculate the ratio of the intersection area to the union area between the expanded coordinate ranges of different key structural areas as the area overlap degree index; Filter out key structural area pairs with an area overlap degree index exceeding a preset overlap threshold to form a set of duplicate areas to be merged; Perform consistency verification processing on the feature type labels of the key structural areas in the set of duplicate areas, retain the duplicate area pairs with exactly the same labels, and delete the duplicate area pairs with inconsistent labels; Perform union calculation processing on the spatial coordinate ranges of the retained duplicate area pairs to generate a merged spatial coordinate range, and the merged spatial coordinate range includes all point cloud points and contour line segments of the original two areas; Perform deduplication and merging processing on the feature type labels of the retained duplicate area pairs to generate a unified feature type label; Perform association binding processing on the merged spatial coordinate range and the unified feature type label to generate a deduplicated feature recognition annotation result.

10. A feature recognition and annotation system, characterized in that, Including: A processor; A storage device on which a computer program is stored, and when the computer program is executed by the processor, the processor implements the feature recognition annotation method for three-dimensional tunnel modeling as described in any one of claims 1-9.

Images