Mountain tunnel surrounding rock grade automatic identification method, system and device based on image-point cloud fusion, and storage medium

Abstract

The invention relates to the technical field of tunnel engineering intelligent perception and surrounding rock grading, and discloses a mountain tunnel surrounding rock grade automatic identification method, system and device based on image-point cloud fusion, and a storage medium. The automatic mountain tunnel surrounding rock grade identification method based on image-point cloud fusion comprises the following steps of multi-source surrounding rock perception data acquisition, surrounding rock image apparent semantic feature extraction, surrounding rock point cloud geometric state feature extraction, surrounding rock evaluation unit construction and image-point cloud feature fusion, surrounding rock engineering state comprehensive evaluation and grade identification. And surrounding rock grade zoning and result output. According to the invention, through image-point cloud fusion, a continuous evaluation unit is constructed along the axial direction of the tunnel, and the surrounding rock grade is automatically identified and output; according to the method, the objectivity and consistency of surrounding rock grade identification are improved, continuous sensing and grading of surrounding rock states are realized, the real-time performance of acquiring surrounding rock grade information is improved, deep fusion and engineering output of multi-source information are realized, and the method has good engineering applicability.

Description

An automatic identification method, system, device, and storage medium for the surrounding rock grade of mountain tunnels based on image-point cloud fusion. Technical Field

[0001] This application belongs to the field of intelligent sensing and surrounding rock classification technology for tunnel engineering, specifically involving an automatic identification method, system, device, and storage medium for the surrounding rock classification of mountain tunnels based on image-point cloud fusion. Background Technology

[0002] The surrounding rock grade is a crucial basis for determining the support form and parameters during the design and construction of mountain tunnels. Its determination directly affects tunnel construction safety, project quality, and construction costs. Currently, the surrounding rock grade is primarily determined based on geological survey data, combined with the experience of technical personnel during construction regarding the integrity of the surrounding rock at the tunnel face, the development of joints and fissures, lithological characteristics, and stability.

[0003] However, this traditional method, which relies heavily on human experience, has several drawbacks in practical applications: First, the grading results are highly dependent on the subjective experience and judgment of technical personnel, resulting in strong subjectivity and poor consistency in the grading results. Second, the grading results are usually based on only a small number of discrete cross-sections, making it difficult to reflect the continuous changes of the surrounding rock along the tunnel axis and to accurately capture the gradual or abrupt changes in the surrounding rock conditions, thus resulting in low accuracy. Third, the manual analysis and grading process is time-consuming, the information on the surrounding rock grade is updated late, the real-time performance is poor, and it is difficult to provide timely feedback to guide construction, which is not conducive to the dynamic adjustment of support parameters and risk warning.

[0004] With the development of image perception and 3D laser scanning technology, existing technologies such as using digital image processing technology to identify the surface features of rock masses and using 3D laser scanning to obtain high-precision point cloud data of tunnel surrounding rock have been widely used. However, there are still obvious limitations: First, it focuses on the identification and measurement of a single information source or a single target: relying solely on image data can only obtain the apparent semantic features of the surrounding rock (such as cracks and joints), and relying solely on point cloud data can only analyze the geometric morphological features of the surrounding rock (such as over-excavation and under-excavation), and cannot achieve a comprehensive assessment of the overall engineering state of the surrounding rock; Second, it lacks a judgment method directly related to the surrounding rock grade: it only identifies and measures local geometric objects such as joints and cracks in the surrounding rock, and its output results are geometric parameters (such as displacement and crack width) or statistical indicators (such as fracture rate), without constructing an evaluation system for the overall engineering state of the surrounding rock, and cannot directly output surrounding rock grade results that can be directly used in engineering; Third, it does not solve the problem of continuous spatial classification: the classification results are also based on discrete cross-sections or local areas as analysis objects, and it fails to construct a continuous, spatially integrated analysis unit and evaluation framework along the tunnel axis, and cannot output continuously distributed surrounding rock grades, resulting in limited accuracy of the classification results. Summary of the Invention

[0005] To address or partially address the problems existing in related technologies, this application provides an automatic identification method, system, device, and storage medium for the surrounding rock grade of mountain tunnels based on image-point cloud fusion. By using image-point cloud fusion, constructing continuous evaluation units along the tunnel axis, and automatically identifying and outputting the surrounding rock grade, it solves the shortcomings of traditional manual grading and existing image perception and three-dimensional laser scanning technology grading.

[0006] The first aspect of this application provides an automatic identification method for the surrounding rock grade of mountain tunnels based on image-point cloud fusion, comprising at least the following steps: S1: Multi-source surrounding rock perception data acquisition. During tunnel excavation, after each excavation cycle, visible light image data and three-dimensional laser scanning point cloud data of the newly exposed surrounding rock face, arch, and sidewall areas are acquired within the same time window, and a unified timestamp and tunnel mileage information are recorded; S2: Surrounding rock image apparent semantic feature extraction. The visible light image data acquired in step S1 is preprocessed, and a deep learning semantic segmentation model is used to automatically identify the apparent semantic features of the surrounding rock. The apparent semantic features include at least one of the following: fracture distribution and continuity features, weathering degree and surface fragmentation features, and lithological texture and color distribution features, to form an image feature vector F for characterizing the development degree of the surrounding rock's apparent structure. i S3: Extraction of geometric state features of surrounding rock point cloud. The 3D laser scanning point cloud data acquired in step S1 is preprocessed, and geometric state features of the surrounding rock are extracted based on normal estimation and local neighborhood analysis. These geometric state features include at least one of the statistical characteristics of surrounding rock roughness and / or undulation, and geometric disturbance characteristics of over-excavation and under-excavation of the tunnel cross-section, to form a point cloud feature vector F characterizing the overall geometric stability of the surrounding rock. p S4: Construction of surrounding rock evaluation units and image-point cloud feature fusion. After each excavation cycle, the newly exposed surrounding rock is divided into several continuous evaluation units along the tunnel axis according to mileage segments. Within each evaluation unit, the corresponding image feature vector F obtained in step S2 is... i The corresponding point cloud feature vector F obtained in step S3 p Alignment and fusion are performed to construct the multi-dimensional engineering state feature vector F of the surrounding rock for each evaluation unit. f S5: Comprehensive assessment and grade identification of surrounding rock engineering condition, based on the multi-dimensional engineering condition feature vector F of surrounding rock constructed in step S4. fThe rule-based discrimination model is used to comprehensively evaluate the surrounding rock engineering status of each evaluation unit, and the evaluation results are mapped to the corresponding surrounding rock grade intervals according to the preset surrounding rock grade identification rules to complete the automatic identification of the surrounding rock grade of each evaluation unit; S6: Surrounding rock grade partitioning and result output. Based on the surrounding rock grade identification results of each evaluation unit generated in step S5, the tunnel surrounding rock is spatially partitioned, and the surrounding rock grade distribution information along the tunnel axis is generated and output, which corresponds to the tunnel mileage and construction section number.

[0007] In one alternative, step S2 includes preprocessing of the visible light image data, including denoising, distortion correction, and brightness normalization preprocessing. The deep learning semantic segmentation model is a convolutional neural network or a Transformer-based visual model.

[0008] In one alternative, step S3 includes outlier removal, voxel downsampling, and coordinate unification preprocessing for the 3D laser scanning point cloud data.

[0009] In one alternative, in step S2, the fissure distribution and continuity characteristics are quantified by calculating the fissure area ratio, the weathering degree and surface fragmentation characteristics are quantified by calculating the fissure length density, and the lithological texture and color distribution characteristics are quantified by calculating the texture complexity.

[0010] In one alternative, in step S3, the statistical features are quantified by statistical point cloud surface roughness index and undulation index, and the geometric disturbance features are quantified by registering the point cloud with the design cross-section model to calculate the over-excavation and under-excavation deviations.

[0011] In one optional scheme, step S4 specifically includes: S4.1: Construction of surrounding rock evaluation units. A tunnel mileage coordinate axis s is established along the tunnel axis, and the newly exposed surrounding rock after each excavation cycle is divided into several continuous surrounding rock evaluation units U along the tunnel axis according to a preset length ΔL. i The i-th evaluation unit is defined as: U i ={p|s i ≤s(p) i +ΔL}, where s(p) represents the mileage position corresponding to the acquired image pixels and / or point cloud points, s i S4.2: Multi-source data mapping within the evaluation unit, for each evaluation unit U i The point cloud data that satisfies s(p)∈U i The point set is mapped to a point cloud subset P. i Mapping image data to U through camera extrinsic parameters and projection relationships i The pixel region within the spatial range is mapped to a subset I of the image. i ​Thus, an evaluation unit U is established. i With image subset I i Point cloud subset P i One-to-one correspondence between them; S4.3: Feature statistics within each evaluation unit U i Within, for image subset I i And point cloud subset P i Feature statistics are performed separately, including the proportion of crack area, crack length density, and texture complexity index, to form an image feature vector F. i Point cloud feature vector F is formed by statistically analyzing the surface roughness, undulation, and over-excavation / under-excavation deviation of the surrounding rock. p The image feature vector F i The point cloud feature vector F p As a quantitative input for the surrounding rock engineering state of each evaluation unit; S4.4: Multi-source feature fusion for evaluation unit U i Image feature vector F i With point cloud feature vector F p Weighted fusion is performed to construct a unified multi-dimensional engineering state feature vector F of the surrounding rock. f The multidimensional engineering state feature vector F of the i-th evaluation unit f Defined as: F fi =[α·F i ,β·F p In the formula, α and β represent feature weight coefficients, satisfying α+β=1.

[0012] The second aspect of this application provides an automatic identification system for the surrounding rock grade of mountain tunnels based on image-point cloud fusion. Performing the aforementioned method, the system includes: a data acquisition module, used to simultaneously acquire visible light image data and three-dimensional laser scanning point cloud data of the newly exposed surrounding rock at the tunnel face, arch, and sidewall areas after each excavation cycle during tunnel excavation, and record a unified timestamp and tunnel mileage information; and an image feature extraction module, used to process the image data, automatically identify the apparent semantic features of the surrounding rock using a deep learning semantic segmentation model, calculate quantitative indicators, and form an image feature vector F. i The point cloud feature extraction module processes the 3D laser scanning point cloud data, extracts the geometric state features of the surrounding rock based on normal estimation and local neighborhood analysis, calculates quantitative indicators, and forms a point cloud feature vector F. p The evaluation unit construction and feature fusion module is used to divide the newly exposed surrounding rock after each excavation cycle into several continuous evaluation units along the tunnel axis according to mileage segments, and to integrate the image feature vector F within each evaluation unit. i With the point cloud feature vector F pAlignment and fusion are performed to construct the multi-dimensional engineering state feature vector F of the surrounding rock for each evaluation unit. f The surrounding rock grade identification module is used to identify the surrounding rock grade based on the multi-dimensional engineering state feature vector F. f The system uses a rule-based discrimination model to automatically identify the surrounding rock grade of each evaluation unit. The result output module is used to spatially partition the tunnel surrounding rock based on the surrounding rock grade identification results of each evaluation unit, and generate and output continuous surrounding rock grade distribution information along the tunnel axis.

[0013] A third aspect of this application provides an automatic identification device for the surrounding rock grade of mountain tunnels based on image-point cloud fusion, comprising: a three-dimensional laser scanner for acquiring point cloud data of the surrounding rock face, arch, and sidewall areas; an industrial camera for acquiring visible light image data of the surrounding rock face, arch, and sidewall areas; a processor; and a memory storing a computer program that can run on the processor; when the processor executes the computer program, it implements the aforementioned method.

[0014] A fourth aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned method.

[0015] The beneficial effects of this application are as follows: The automatic identification method for surrounding rock grade of mountain tunnels based on image-point cloud fusion in this application solves the defects of traditional manual grading and existing image perception and three-dimensional laser scanning technology by using image-point cloud fusion, constructing continuous evaluation units along the tunnel axis, and automatically identifying and outputting the surrounding rock grade. It improves the objectivity and consistency of surrounding rock grade identification, realizes continuous perception and grading of surrounding rock status, improves the real-time acquisition of surrounding rock grade information, and realizes deep fusion and engineering output of multi-source information. It has strong practicality and good engineering applicability and promotion value.

[0016] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

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

[0018] Figure 1 is a schematic diagram of the workflow of the automatic rock grade identification method in one embodiment of this application; Figure 2 is a schematic diagram of a sub-process of the automatic rock grade identification method in one embodiment of this application; Figure 3 is a structural block diagram of the automatic rock grade identification system in one embodiment of this application. Detailed Implementation

[0019] The specific embodiments of this application will be further described in detail below with reference to the accompanying drawings and examples. The following examples are used to illustrate this application, but are not intended to limit the scope of this application. Similarly, the following examples are only some embodiments of this application, not all embodiments. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0020] In this application, the terms "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to a specific feature, structure, material, or characteristic described in connection with that embodiment or example, which is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0021] To address the aforementioned problems, this application proposes improvements and innovations, including the following embodiments.

[0022] Example 1: Please refer to Figures 1 and 2. The first aspect of this application provides an automatic identification method for the surrounding rock grade of mountain tunnels based on image-point cloud fusion, including the following steps: S1: Multi-source surrounding rock perception data acquisition. During the tunnel excavation construction process, after each excavation cycle, within the same time window, visible light image data and three-dimensional laser scanning point cloud data of the newly exposed surrounding rock face, arch, and sidewall areas are acquired, and a unified timestamp and tunnel mileage information are recorded.

[0023] Specifically, image data is acquired using industrial cameras with a resolution of no less than 12 megapixels and a lens focal length of 24–35mm to ensure no significant distortion in the covered area. 3D laser scanning point cloud data is acquired using a 3D laser scanner with a scanning density set to 5–20mm as needed to ensure the capture of the geometric features of the surrounding rock. The industrial cameras and 3D laser scanners are deployed in a stable area behind the arch frame. During tunnel excavation, after each excavation cycle, within the same time window, visible light image data is acquired using industrial cameras on the surfaces of the newly exposed surrounding rock face, arch crown, and sidewalls, while 3D laser scanning point cloud data is acquired using the 3D laser scanner. A unified timestamp and corresponding tunnel mileage information are assigned to both the image and point cloud data to ensure temporal and spatial synchronization.

[0024] S2: Extraction of apparent semantic features from surrounding rock images. The visible light image data acquired in step S1 is preprocessed, and a deep learning semantic segmentation model is used to automatically identify the apparent semantic features of the surrounding rock. These apparent semantic features include fracture distribution and continuity characteristics, weathering degree and surface fragmentation characteristics, and lithological texture and color distribution characteristics, to form an image feature vector F that characterizes the degree of development of the surrounding rock's apparent structure. i .

[0025] Specifically, the visible light image data acquired in step S1 undergoes preprocessing including denoising, distortion correction, and brightness normalization. Denoising involves using median filtering and Gaussian filtering algorithms. First, median filtering is used to eliminate salt-and-pepper noise by selecting the median of neighboring pixels, and then Gaussian filtering is used to eliminate Gaussian noise by weighted averaging of neighboring pixels. Distortion correction involves using the Zhang Zhengyou calibration method to obtain camera intrinsic parameters and distortion coefficients, and correcting the image pixel coordinates based on the inverse distortion model, thereby eliminating the geometric distortion problem caused by image stretching and distortion due to lens distortion. Brightness normalization involves using an adaptive histogram equalization algorithm to improve the problem of local underexposure or overexposure caused by uneven illumination in the tunnel, enhance the overall contrast, and make features such as cracks and textures clearer.

[0026] The preprocessed image data is automatically used by a deep learning semantic segmentation model to identify the apparent semantic features of the surrounding rock. This model can employ a convolutional neural network (CNN) or a Transformer-based visual model, outputting pixel-level segmentation maps of various feature regions. The model uses a pre-labeled dataset of surrounding rock images as training samples, with a minimum of 1000 samples covering different lithologies and fracture development levels to ensure accurate model recognition. For example, a U-Net++-based CNN is used as the semantic segmentation model. The encoder module downsamples to extract deep image features, while the decoder module upsamples to restore image resolution. Skip connections are added between encoding and decoding to preserve detailed features such as fractures and lithological textures. The input is the preprocessed surrounding rock image, and the output is pixel-level semantic segmentation results. Alternatively, a VisionTransformer model is used. The surrounding rock image is divided into fixed-size image blocks, which are then converted into feature vectors. A self-attention mechanism captures global features, and a convolutional module extracts local texture features, achieving accurate identification of the apparent semantic features of the surrounding rock.

[0027] The preprocessed image data is input into the trained semantic segmentation model, which outputs pixel-level semantic segmentation results for fracture regions, fracture zones, and lithological texture regions. Based on these results, quantification metrics for each feature are calculated within each evaluation unit. Fracture distribution and continuity features: The ratio of the number of pixels in the fracture region to the total number of pixels in the image is calculated to obtain the fracture area proportion, quantifying the degree of fracture development. Weathering degree and surface fragmentation features: Morphological refinement is performed on the segmented fracture regions to obtain a single-pixel-wide fracture skeleton. The total pixel length of all fracture skeletons is calculated and divided by the image area to obtain the fracture length density, quantifying the degree of fragmentation. Lithological texture and color distribution features: A gray-level co-occurrence matrix is ​​calculated on the grayscale image of the image region. Entropy is selected as a measure of texture complexity, quantifying lithological differences; a higher entropy value indicates a more complex rock texture. The three quantification metrics—fracture area proportion, fracture length density, and texture complexity—are arranged in order to construct an image feature vector F. i =[a1, a2, a3], where a1 is the proportion of crack area, a2 is the crack length density, and a3 is the texture complexity.

[0028] S3: Extraction of geometric features of surrounding rock point cloud. The 3D laser scanning point cloud data acquired in step S1 is preprocessed, and geometric features of the surrounding rock are extracted based on normal estimation and local neighborhood analysis. These geometric features include statistical characteristics of the surface roughness and / or undulation of the surrounding rock, and geometric disturbance characteristics of over-excavation and under-excavation of the tunnel cross-section, to form a point cloud feature vector F characterizing the overall geometric stability of the surrounding rock. p .

[0029] Specifically, the 3D laser scanning point cloud data acquired in step S1 undergoes preprocessing including outlier removal, voxel downsampling, and coordinate unification. Outlier removal involves using a statistical filtering algorithm to calculate the average distance between each point and its neighboring points. The number of neighboring points, k, is set to 20–50. Outliers whose distance exceeds the mean ± 2 standard deviations are removed to eliminate scanning noise. Voxel downsampling involves using a voxel grid filter to divide the point cloud space into multiple cubic grids (voxels) with a side length of 0.01m. The centroid of all points within each voxel represents that voxel, reducing the amount of data while maintaining the overall morphology of the surrounding rock surface, thereby improving processing efficiency. Coordinate unification involves using the positioning and orientation system of a total station or scanner to uniformly convert all point cloud data to a tunnel mileage coordinate system with the tunnel design centerline as the Z-axis and the horizontal and vertical directions of the tunnel face as the X and Y axes, ensuring the spatial consistency between the point cloud data and the tunnel mileage.

[0030] After preprocessing, the geometric features of the surrounding rock are extracted from the 3D laser scanning point cloud data. Quantitative indices for each feature are calculated within each evaluation unit. Specifically, the statistical features of the surrounding rock surface roughness and undulation are calculated as follows: For the point cloud within the evaluation unit, a k-neighborhood is selected for each point (k = 30). The neighborhood plane is fitted using the least squares method, and the standard deviation of the distance from the point to the plane is calculated. This value represents the single-point roughness. The average roughness of all points within the evaluation unit is used as the surface undulation of that unit. The geometric disturbance features of over-excavation and under-excavation in the tunnel cross-section are also calculated: The point cloud data is projected onto the tunnel design cross-section model, and the vertical distance from each point to the design cross-section outline is calculated. Positive values ​​represent over-excavation deviation, and negative values ​​represent under-excavation deviation. The average deviation of all points within the evaluation unit is used as the over-excavation / under-excavation deviation for that unit. The three quantitative indices—roughness, undulation, and over-excavation / under-excavation deviation—are arranged in order to construct an image feature vector F. p =[b1, b2, b3], where b1 is roughness, b2 is undulation, and b3 is over- or under-excavation deviation.

[0031] S4: Construction of surrounding rock evaluation units and image-point cloud feature fusion. After each excavation cycle, the newly exposed surrounding rock is divided into several continuous evaluation units along the tunnel axis according to mileage segments. Within each evaluation unit, the corresponding image feature vector F obtained in step S2 is... i The corresponding point cloud feature vector F obtained in step S3 p Alignment and fusion are performed to construct the multi-dimensional engineering state feature vector F of the surrounding rock for each evaluation unit. f This serves as a representation of the engineering status of each evaluation unit.

[0032] Specifically, step S4 includes: S4.1: Construction of surrounding rock evaluation units. A tunnel mileage coordinate axis s is established along the tunnel axis. Taking the tunnel design centerline as the Z-axis, the newly exposed surrounding rock after each excavation cycle is divided into several continuous surrounding rock evaluation units U along the tunnel axis according to a preset length ΔL. i The value of ΔL is consistent with the length of a single support during construction, and the i-th evaluation unit is defined as: U i ={p|s i ≤s(p) i +ΔL}, where s(p) represents the mileage position corresponding to the acquired image pixels and / or point cloud points, s i This represents the starting mileage of the i-th evaluation unit.

[0033] The construction of the surrounding rock evaluation unit follows the principle of axial continuity: the evaluation unit is continuously divided along the tunnel axis to reflect the gradual changes in the surrounding rock conditions with mileage; the principle of engineering interpretability: the scale of the evaluation unit should match the scale of the construction section and the adjustment of the support design parameters; and the principle of data stability: each evaluation unit should contain no less than a preset threshold of image feature samples and point cloud points. Specifically, the number of image feature samples should be no less than 100 and the number of point cloud points should be no less than 1000 to ensure the reliability of the statistical features.

[0034] S4.2: Multi-source data mapping within the evaluation unit, for each evaluation unit U i The point cloud data that satisfies s(p)∈U i The point set is mapped to a point cloud subset P. i Mapping image data to U through camera extrinsic parameters and projection relationships i The pixel region within the spatial range is mapped to a subset I of the image. i Thus, an evaluation unit U is established. i With image subset I i Point cloud subset P i A one-to-one correspondence between them.

[0035] S4.3: Feature statistics within each evaluation unit U i Within, for image subset I i And point cloud subset P i Feature statistics are performed separately, including the proportion of crack area, crack length density, and texture complexity index, to form an image feature vector F. i =[a1, a2, a3]; Point cloud feature statistics include the surface roughness, undulation, and over-excavation / under-excavation deviation indices of the surrounding rock, forming the point cloud feature vector F. p =[b1, b2, b3]; and then the above image feature vector F i The point cloud feature vector F p ​This serves as a quantitative input for the surrounding rock engineering state of each evaluation unit.

[0036] S4.4: Multi-source feature fusion for evaluation unit U i Image feature vector F i With point cloud feature vector F p Weighted fusion is performed to construct a unified multi-dimensional engineering state feature vector F of the surrounding rock. f The multidimensional engineering state feature vector F of the i-th evaluation unit f Defined as: F fi =[α·F i ,β·F p In the formula, α and β represent feature weight coefficients, satisfying α+β=1.

[0037] For the weighting coefficients, under standard operating conditions, α=β=0.5. When the surrounding rock has obvious apparent fissures, severe fractures, and significant lithological differences, α is taken as 0.55–0.65 and β as 0.35–0.45. When the surrounding rock has large surface roughness, large undulations, and obvious over-excavation or under-excavation, α is taken as 0.35–0.45 and β as 0.55–0.65. In specific applications, the weighting coefficients α and β can be determined by expert scoring and / or machine learning optimization. Specifically, expert scoring involves inviting experts in tunnel geology and construction to score the importance of image features and point cloud features based on their experience in surrounding rock classification, and then using a weighted average method to calculate the α and β values. Machine learning optimization involves using engineering data with known surrounding rock grades as the training set, using α and β as optimization parameters, and using the accuracy of grade identification as the optimization objective, to iteratively solve for the optimal weighting values ​​using a gradient descent algorithm.

[0038] S5: Comprehensive assessment and grade determination of surrounding rock engineering condition, based on the multi-dimensional engineering condition feature vector F of surrounding rock constructed in step S4. f The rule-based discrimination model is used to comprehensively evaluate the surrounding rock engineering status of each evaluation unit, and the evaluation results are mapped to the corresponding surrounding rock grade intervals according to the preset surrounding rock grade identification rules, thereby completing the automatic identification of the surrounding rock grade of each evaluation unit.

[0039] Specifically, based on the multi-dimensional engineering state feature vector Ff of the surrounding rock constructed in step S4, an automatic identification model is used to automatically identify the surrounding rock grade. Specifically, firstly, a surrounding rock grading rule base is established. For example, using the proportion of fracture area, surface roughness, and over- or under-excavation deviation as indicators, the surrounding rock grades are divided into grades I–V, and the threshold values ​​for the judgment indicators of grades I–V are defined: Grade I surrounding rock (fracture area proportion ≤ 1%, surface roughness ≤ 0.5 mm, over- or under-excavation deviation ≤ 50 mm), Grade II surrounding rock (fracture area proportion > 1% and ≤ 5%, surface roughness > 0.5 and ≤ 1.0 mm, over- or under-excavation deviation > 50 and ≤ 100 mm), Grade III surrounding rock (fracture area proportion > 5% and ≤ 15%, surface roughness > 1.0 and ≤ 2.0 mm, over- or under-excavation deviation > 100 and ≤ 150 mm). The evaluation process involves several steps: First, the surrounding rock is classified as Class IV (fracture area > 15% and ≤ 30%, surface roughness > 2.0 and ≤ 5.0 mm, over-excavation / under-excavation deviation > 150 and ≤ 200 mm); second, the multi-dimensional engineering state feature vector Ff of each evaluation unit is compared with the rule base threshold, and a weighted scoring method is used to calculate the total score, with the weights consistent with α and β. Then, the initial surrounding rock grade is determined based on the interval of the total score, thereby completing the automatic identification of the surrounding rock grade of each evaluation unit along the tunnel axis after each excavation cycle.

[0040] In practical applications, machine learning classification models can also be used, based on the multi-dimensional engineering state feature vector F of the surrounding rock constructed in step S4. f A comprehensive assessment of the surrounding rock engineering condition of each evaluation unit is conducted. Specifically, a training dataset is first constructed, engineering case data of different surrounding rock grades are collected, and the F-value of each case is extracted. fi The feature vectors are used as input samples, and the corresponding actual surrounding rock grades are used as labels, with a sample size of no less than 1000 groups. Model training and optimization are then performed using a random forest classification model with 100 decision trees. Five-fold cross-validation is used to optimize model parameters, ensuring a model accuracy of ≥95%. Finally, the trained model is deployed by embedding it into the system, inputting the feature vector F of the unit to be evaluated. fi This allows for the real-time output of surrounding rock grade results.

[0041] S6: Surrounding rock grade zoning and result output. Based on the surrounding rock grade identification results of each evaluation unit generated in step S5, the surrounding rock of the tunnel is spatially zoned, and the surrounding rock grade distribution information along the tunnel axis is generated and output, corresponding to the tunnel mileage and construction section number.

[0042] Specifically, based on the surrounding rock grade identification results of each evaluation unit, all evaluation units U iThe identification results are arranged in mileage order, forming a continuous sequence of surrounding rock grade zones along the tunnel axis, clearly defining the start and end mileage of each zone. This allows for the graphical output of surrounding rock grade distribution information, generating a "tunnel surrounding rock grade longitudinal profile map." With tunnel mileage as the horizontal axis and surrounding rock grade as the vertical axis, different colored or patterned strip areas visually display the distribution of surrounding rock grades in each evaluation unit of newly exposed surrounding rock along the tunnel axis after the excavation cycle. Simultaneously, data results are output, including the actual length of each surrounding rock grade zone, the corresponding tunnel mileage, and the corresponding construction section number, forming a surrounding rock grade report. The output information is also directly transmitted to the tunnel construction management platform, thereby using the surrounding rock grade information to guide subsequent dynamic adjustments of support parameters, construction methods, and construction risk warnings.

[0043] This application presents an automatic rock mass classification method for mountain tunnels based on image-point cloud fusion. By integrating image and point cloud data, constructing continuous evaluation units along the tunnel axis, and automatically identifying and outputting the rock mass classification, this method overcomes the shortcomings of traditional manual classification and existing image perception and 3D laser scanning technologies. It offers the following advantages: Improved objectivity and consistency in rock mass classification: By fusing image and point cloud data and extracting quantitative features, and using a data-driven model for classification, the method significantly reduces subjective reliance on human experience, resulting in highly repeatable and consistent classification results across different times and personnel. Continuous perception and classification of rock mass conditions: By constructing continuous evaluation units along the tunnel axis, the method changes the traditional discrete section-based analysis mode, enabling precise characterization of rock mass conditions along the tunnel axis. The gradual or abrupt changes in rock mass characteristics are analyzed to output continuous rock mass classification maps, providing refined data guidance for subsequent dynamic adjustments of support parameters, construction methods, and construction risk warnings. The real-time acquisition of rock mass classification information is improved: compared to manual analysis and classification, the time from data acquisition to classification output is greatly reduced, enabling real-time updates of rock mass classification information, which is beneficial for timely feedback and construction guidance. Deep integration and engineering output of multi-source information are achieved: the rock mass classification fully integrates various apparent semantic features and geometric state features, constructing a comprehensive evaluation system for the overall engineering state of the rock mass. This system can provide a more comprehensive characterization of the rock mass engineering state and output rock mass classification results that can be directly used in engineering, seamlessly connecting with engineering practice needs. It is highly practical and has good engineering applicability and promotional value.

[0044] Example 2: Referring to Figure 3, corresponding to the aforementioned example of the automatic identification method for the surrounding rock grade of mountain tunnels based on image-point cloud fusion, this example also provides an automatic identification system for the surrounding rock grade of mountain tunnels based on image-point cloud fusion. Performing the aforementioned method, the system includes: a data acquisition module, used to simultaneously acquire visible light image data and three-dimensional laser scanning point cloud data of the newly exposed surrounding rock at the tunnel face, arch, and sidewall areas after each excavation cycle during tunnel excavation, and record a unified timestamp and tunnel mileage information; and an image feature extraction module, used to process the image data, and automatically identify the apparent semantic features of the surrounding rock using a deep learning semantic segmentation model and calculate quantitative indicators to form an image feature vector F. i The point cloud feature extraction module processes the 3D laser scanning point cloud data, extracts the geometric state features of the surrounding rock based on normal estimation and local neighborhood analysis, calculates quantitative indicators, and forms a point cloud feature vector F. p The evaluation unit construction and feature fusion module is used to divide the newly exposed surrounding rock after each excavation cycle into several continuous evaluation units along the tunnel axis according to mileage segments, and to integrate the image feature vector F within each evaluation unit. i With the point cloud feature vector F p Alignment and fusion are performed to construct the multi-dimensional engineering state feature vector F of the surrounding rock for each evaluation unit. f The surrounding rock grade identification module is used to identify the surrounding rock grade based on the multi-dimensional engineering state feature vector F. f The system uses a rule-based discrimination model to automatically identify the surrounding rock grade of each evaluation unit. The result output module is used to spatially partition the tunnel surrounding rock based on the surrounding rock grade identification results of each evaluation unit, and generate and output continuous surrounding rock grade distribution information along the tunnel axis.

[0045] Example 3: Corresponding to the aforementioned examples of automatic identification methods and systems for the surrounding rock grade of mountain tunnels based on image-point cloud fusion, this example also provides an automatic identification device for the surrounding rock grade of mountain tunnels based on image-point cloud fusion, comprising: a 3D laser scanner for acquiring point cloud data of the tunnel face, arch, and sidewall areas; an industrial camera for acquiring visible light image data of the tunnel face, arch, and sidewall areas; a processor; and a memory storing a computer program that can run on the processor; when the processor executes the computer program, the aforementioned automatic identification method for the surrounding rock grade of mountain tunnels based on image-point cloud fusion is implemented.

[0046] As will be understood by those skilled in the art, computer equipment can be a desktop computer, a laptop, a handheld computer, or a cloud server, etc., and computer equipment can interact with users through a keyboard, mouse, remote control, touchpad, or voice control device.

[0047] The memory includes at least one type of readable storage medium, which can be a magnetic medium, such as a floppy disk, hard disk, or magnetic tape; or an optical medium, such as a compact disc read-only memory (CD-ROM) or a digital versatile disc (DVD); or a semiconductor medium, such as a solid-state disk (SSD), random access memory (RAM), read-only memory (ROM), a smart media card (SMC), a secure digital card (SD), a flash card, a register, etc. In some embodiments, the memory can be an internal storage unit of a computer device, such as the hard disk or RAM of the computer device; in other embodiments, the memory can also be an external storage device of the computer device, such as a plug-in hard disk, a smart media card (SMC), a secure digital card (SD), a flash card, etc., provided on the computer device. Of course, the memory can also include both internal storage units and external storage devices of a computer device. In this embodiment, the memory is used to store the operating system and various application software installed on the computer device, such as the program code of the two-level diagnosis method for early faults of wind turbines in various embodiments of this application.

[0048] The processor can be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip. This processor is typically used to control the overall operation of a computer device. In this embodiment, the processor is used to run program code stored in memory or process data, such as running the program code for the two-level early fault diagnosis method for wind turbines according to various embodiments of this application.

[0049] Example 4: Corresponding to the aforementioned examples of automatic identification method, system, and device for mountain tunnel surrounding rock grade based on image-point cloud fusion, this example also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned automatic identification method for mountain tunnel surrounding rock grade based on image-point cloud fusion.

[0050] Finally, it should be noted that although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application, all of which should be included within the protection scope of this application.

Claims

1. An automatic identification method for the surrounding rock grade of mountain tunnels based on image-point cloud fusion, characterized in that, At least the following steps are included: S1: Multi-source surrounding rock sensing data acquisition. During tunnel excavation, after each excavation cycle, visible light image data and 3D laser scanning point cloud data of the newly exposed surrounding rock face, arch, and sidewall areas are acquired within the same time window, and a unified timestamp and tunnel mileage information are recorded. S2: Surrounding rock image apparent semantic feature extraction. The visible light image data acquired in step S1 is preprocessed, and a deep learning semantic segmentation model is used to automatically identify the apparent semantic features of the surrounding rock. These apparent semantic features include at least one of the following: fracture distribution and continuity features, weathering degree and surface fragmentation features, and lithological texture and color distribution features, to form an image feature vector F representing the degree of development of the surrounding rock's apparent structure. i S3: Extraction of geometric state features of surrounding rock point cloud. The 3D laser scanning point cloud data acquired in step S1 is preprocessed, and geometric state features of the surrounding rock are extracted based on normal estimation and local neighborhood analysis. These geometric state features include at least one of the statistical characteristics of surrounding rock roughness and / or undulation, and geometric disturbance characteristics of over-excavation and under-excavation of the tunnel cross-section, to form a point cloud feature vector F characterizing the overall geometric stability of the surrounding rock. p S4: Construction of surrounding rock evaluation units and image-point cloud feature fusion. After each excavation cycle, the newly exposed surrounding rock is divided into several continuous evaluation units along the tunnel axis according to mileage segments. Within each evaluation unit, the corresponding image feature vector F obtained in step S2 is... i The corresponding point cloud feature vector F obtained in step S3 p Alignment and fusion are performed to construct the multi-dimensional engineering state feature vector F of the surrounding rock for each evaluation unit. f S5: Comprehensive assessment and grade identification of surrounding rock engineering condition, based on the multi-dimensional engineering condition feature vector F of surrounding rock constructed in step S4. f The rule-based discrimination model is used to comprehensively evaluate the surrounding rock engineering status of each evaluation unit, and the evaluation results are mapped to the corresponding surrounding rock level range according to the preset surrounding rock level identification rules, so as to complete the automatic identification of the surrounding rock level of each evaluation unit. S6: Surrounding rock grade zoning and result output. Based on the surrounding rock grade identification results of each evaluation unit generated in step S5, the surrounding rock of the tunnel is spatially zoned, and the surrounding rock grade distribution information along the tunnel axis is generated and output, corresponding to the tunnel mileage and construction section number.

2. The method for automatic identification of surrounding rock grade of mountain tunnels based on image-point cloud fusion according to claim 1, characterized in that: In step S2, the preprocessing of the visible light image data includes denoising, distortion correction, and brightness normalization preprocessing. The deep learning semantic segmentation model is a convolutional neural network or a Transformer-based visual model.

3. The method for automatic identification of surrounding rock grade of mountain tunnels based on image-point cloud fusion according to claim 1, characterized in that: In step S3, the preprocessing of the 3D laser scanning point cloud data includes outlier removal, voxel downsampling, and coordinate unification preprocessing.

4. The method for automatic identification of surrounding rock grade of mountain tunnels based on image-point cloud fusion according to claim 1, characterized in that: In step S2, the distribution and continuity characteristics of the fractures are quantified by calculating the proportion of fracture area, the degree of weathering and surface fragmentation characteristics are quantified by calculating the fracture length density, and the lithological texture and color distribution characteristics are quantified by calculating the texture complexity.

5. The method for automatic identification of surrounding rock grade of mountain tunnels based on image-point cloud fusion according to claim 4, characterized in that: In step S3, the statistical features are quantified by statistical point cloud surface roughness index and undulation index, and the geometric disturbance features are quantified by registering the point cloud with the design cross-section model to calculate the over-excavation and under-excavation deviation.

6. The method for automatic identification of surrounding rock grade of mountain tunnels based on image-point cloud fusion according to claim 5, characterized in that, Step S4 specifically includes: S4.1: Construction of surrounding rock evaluation units. A tunnel mileage coordinate axis s is established along the tunnel axis. After each excavation cycle, the newly exposed surrounding rock is divided into several continuous surrounding rock evaluation units U along the tunnel axis according to a preset length ΔL. i The i-th evaluation unit is defined as: U i ={p|s i ≤s(p) i +ΔL}, where s(p) represents the mileage position corresponding to the acquired image pixels and / or point cloud points, s i S4.2: Multi-source data mapping within the evaluation unit, for each evaluation unit U i The point cloud data that satisfies s(p)∈U i The point set is mapped to a point cloud subset P. i Mapping image data to U through camera extrinsic parameters and projection relationships i The pixel region within the spatial range is mapped to a subset I of the image. i Thus, an evaluation unit U is established. i With image subset I i Point cloud subset P i One-to-one correspondence between them; S4.3: Feature statistics within each evaluation unit U i Within, for image subset I i And point cloud subset P i Feature statistics are performed separately, including the proportion of crack area, crack length density, and texture complexity index, to form an image feature vector F. i Point cloud feature vector F is formed by statistically analyzing the surface roughness, undulation, and over-excavation / under-excavation deviation of the surrounding rock. p The image feature vector F i The point cloud feature vector F p As a quantitative input for the surrounding rock engineering state of each evaluation unit; S4.4: Multi-source feature fusion for evaluation unit U i Image feature vector F i With point cloud feature vector F p Weighted fusion is performed to construct a unified multi-dimensional engineering state feature vector F of the surrounding rock. f The multidimensional engineering state feature vector F of the i-th evaluation unit f Defined as: F fi =[α·F i ,β·F p In the formula, α and β represent feature weight coefficients, satisfying α+β=1.​ 7. An automatic identification system for the surrounding rock grade of mountain tunnels based on image-point cloud fusion, characterized in that, include: The data acquisition module is used to simultaneously collect visible light image data and three-dimensional laser scanning point cloud data of the newly exposed surrounding rock at the tunnel face, arch, and sidewall areas after each excavation cycle during the tunnel excavation process, and record a unified timestamp and tunnel mileage information. The image feature extraction module is used to process the image data and automatically identify the apparent semantic features of the surrounding rock using a deep learning semantic segmentation model, and calculate quantitative indicators to form an image feature vector F. i The point cloud feature extraction module processes the 3D laser scanning point cloud data, extracts the geometric state features of the surrounding rock based on normal estimation and local neighborhood analysis, calculates quantitative indicators, and forms a point cloud feature vector F. p The evaluation unit construction and feature fusion module is used to divide the newly exposed surrounding rock after each excavation cycle into several continuous evaluation units along the tunnel axis according to mileage segments, and to integrate the image feature vector F within each evaluation unit. i With the point cloud feature vector F p Alignment and fusion are performed to construct the multi-dimensional engineering state feature vector F of the surrounding rock for each evaluation unit. f The surrounding rock grade identification module is used to identify the surrounding rock grade based on the multi-dimensional engineering state feature vector F. f The system uses a rule-based discrimination model to automatically identify the surrounding rock grade of each evaluation unit. The result output module is used to spatially partition the tunnel surrounding rock based on the surrounding rock grade identification results of each evaluation unit, and generate and output continuous surrounding rock grade distribution information along the tunnel axis.

8. An automatic identification device for the surrounding rock grade of mountain tunnels based on image-point cloud fusion, characterized in that, include: A 3D laser scanner is used to collect point cloud data of the surrounding rock face, arch, and sidewall areas. Industrial cameras are used to acquire visible light image data of the working face, arch, and sidewall areas of the surrounding rock; processors are also used. A memory storing a computer program executable on the processor; when the processor executes the computer program, it implements the method according to any one of claims 1-6.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, it implements the method according to any one of claims 1-6.

Images