Underground mine roadway anchor rod intelligent identification and statistical method and system

Abstract

The invention provides an underground mine roadway anchor rod intelligent identification and statistical method and system, and relates to the technical field of mine engineering, and the method comprises the steps: employing high-precision three-dimensional laser scanning equipment, cooperating with an SLAM system, scanning an underground roadway, and obtaining the point cloud data of the underground roadway before and after the laying of an anchor net and the driving of an anchor rod support; performing point cloud thinning, data labeling, feature optimization and data enhancement processing on the point cloud data to obtain preprocessed high-quality point cloud data; constructing a PointNet network framework as an anchor rod recognition network model, and training and reasoning the preprocessed high-quality point cloud data by using a deep convolutional network to realize anchor rod point cloud recognition of the underground roadway; and feature enhancement, form optimization and key point statistics are carried out on the anchor rod point cloud identification result to obtain anchor rod parameters, including the external length, the inclination angle, the form, the number and the spacing. According to the invention, the problems of low anchor rod acceptance efficiency and insufficient accuracy are solved, and the efficiency and quality of underground mine roadway support anchor rod inspection are improved.

Description

Technical Field

[0001] This invention relates to the field of mining engineering technology, and more particularly to a method and system for intelligent identification and statistical analysis of anchor bolts in underground mine roadways. Background Technology

[0002] With the continuous expansion of mining scale and the increase in mining depth, geological conditions gradually deteriorate, fractured rock masses increase, and ground stress intensifies, leading to a significant increase in the difficulty of maintaining underground roadways and a heightened risk of instability and failure. Roadway support is a key measure to ensure the stability and safety of roadways and plays a vital role in mining production. Currently, mine roadway support mainly adopts a combination of anchor bolts, anchor mesh, and shotcrete. Among these, the number, density, and shape of anchor bolts, as well as the size and laying density of the anchor mesh, directly affect the support quality. However, traditional support quality inspection methods mainly rely on manual observation and statistics, which suffer from low efficiency, strong subjectivity, and large errors, making it difficult to meet the needs of modern mines for efficient and accurate support acceptance.

[0003] In existing technologies, the inspection of roadway support quality mainly relies on manual inspection and recording, using visual observation or simple measuring tools to statistically analyze parameters such as the number, spacing, and angle of anchor bolts. This method is not only time-consuming and labor-intensive, but also susceptible to factors such as the experience of the operators and lighting conditions, making it difficult to guarantee the accuracy and consistency of the inspection results. Furthermore, manual statistics cannot cover a large area of ​​roadway support, resulting in low acceptance efficiency and failing to meet the needs of rapid mine advancement. Although some studies in recent years have attempted to introduce image recognition technology to assist in inspection, due to the complex underground environment, uneven lighting, dust interference, and other factors, the recognition accuracy and stability of existing algorithms still have significant limitations, making it difficult to achieve automated and intelligent statistical analysis of anchor bolts and other support structures. CN118196126A proposes a deep learning-based method for extracting the boundary lines of underground roadway point clouds. This method uses lidar to collect roadway point cloud data, combines it with neural networks to identify anchor bolts and their center points, and then draws the roadway boundary lines. However, this method mainly focuses on roadway boundary extraction and does not involve the identification and statistics of anchor bolt parameters (such as length, angle, and spacing). Its function is relatively simple and cannot meet the needs of automatic detection of multi-dimensional anchor bolt parameters in support quality acceptance. CN117189254A discloses an intelligent monitoring and early warning method for underground roadway deformation. This method uses a mobile 3D laser scanner to acquire roadway point cloud data, combines SLAM technology to construct a roadway map, and identifies deformation areas by comparing historical data to achieve deformation monitoring and early warning. Although this method has advantages in roadway deformation detection, it does not involve the identification and parameter extraction of anchor bolts in the support structure, and cannot be used for the evaluation and acceptance of anchor bolt support quality, thus limiting its application scenarios. CN116517632B proposes an intelligent evaluation model and method for coal mine roadway safety. It evaluates the support status of roadways at different stages (tunneling, mining) through the fusion of multi-source data such as lidar, stress gauges, and delamination instruments. While this method involves monitoring anchor bolt stress and position changes, it primarily relies on manually deployed sensors and fails to achieve automatic anchor bolt identification and geometric parameter extraction based on point cloud data. It also falls short in terms of automation, identification accuracy, and adaptability. Therefore, existing technologies for anchor bolt support quality inspection generally suffer from problems such as limited identification dimensions, low automation, and weak parameter extraction capabilities, making it difficult to meet the practical needs of efficient, accurate, and intelligent identification of anchor bolt support status in complex underground environments. Thus, there is an urgent need to develop an intelligent detection method that can adapt to complex underground environments and possesses high-precision identification capabilities to improve the efficiency and accuracy of roadway support acceptance. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a method and system for intelligent identification and statistical analysis of anchor bolts in underground mine roadways. By automatically collecting roadway point cloud data, intelligently identifying anchor bolts, and accurately extracting key support parameters such as their length, angle, quantity, and spacing, this invention enables rapid and accurate acceptance of roadway support quality.

[0005] The technical means employed in this invention are as follows:

[0006] A method for intelligent identification and statistical analysis of anchor bolts in underground mine roadways includes: S1. Using high-precision three-dimensional laser scanning equipment, combined with a SLAM system, the underground roadway is scanned to obtain point cloud data before and after the installation of anchor mesh and anchor bolt support. S2. Perform point cloud thinning, data annotation, feature optimization and data augmentation on the collected point cloud data to obtain high-quality preprocessed point cloud data; S3. Construct the PointNet network framework as the anchor bolt recognition network model. Use deep convolutional networks to train and infer on the preprocessed high-quality point cloud data to realize anchor bolt point cloud recognition in underground tunnels. S4. Perform feature enhancement, morphological optimization, and key point statistics on the anchor bolt point cloud recognition results to obtain anchor bolt parameters, including anchor bolt external length, anchor bolt inclination angle, anchor bolt shape, number of anchor bolts, and anchor bolt spacing.

[0007] Further, in step S1, a high-precision, densely sampled Geo SLAM-ZEBHorizon 3D laser scanner, equipped with a SLAM system, is used to collect point cloud data before and after the roadway anchor support. The SLAM system includes a lidar and a wheeled odometer, wherein: LiDAR is used to acquire environmental information by scanning with lasers to measure obstacle information in the surrounding 360° environment. Wheel-type odometers continuously acquire mileage information from the device via a motor encoder on a moving chassis.

[0008] Furthermore, the lidar is used to provide point cloud data for the mapping algorithm. After the mapping algorithm acquires sufficient point cloud data, it constructs a local map centered on the device and with the radar range as the radius. At the same time, the difference between the lidar observation data and the wheel odometer predicted pose is used as observation information to correct the Bayesian filter's predicted pose and improve the accuracy of the filtered device pose estimation.

[0009] Further, step S2 includes: S21. Perform point cloud thinning on the collected point cloud data, identify and remove redundant point cloud data according to a preset thinning ratio, and reduce the amount of original point cloud data while maintaining the geometric features of the point cloud. Specifically, this includes: Load point cloud data of anchor mesh support in the underground roadway of Yanqianshan Mine, which was acquired by Geo SLAM-ZEB Horizon 3D laser scanner; Set the thinning ratio and use the point cloud thinning algorithm to process the actual collected original 3D point cloud to obtain the thinned point cloud data. S22. Label the point cloud data after point cloud thinning. Manually label the anchor regions in the point cloud data to provide necessary training, validation, and testing data for the subsequent convolutional neural network. Specifically, this includes: Load the point cloud data of the underground tunnels after point cloud thinning; Since the tunnel floor does not contain anchor bolt data, the point cloud data of the tunnel floor is cut off from the original point cloud data, while the remaining tunnel point cloud information is retained. The remaining point cloud of the roadway is rotated in three dimensions to delineate the boundary of the anchor bolts until all anchor bolt areas in the roadway are delineated. Data labels are then set, with the point cloud label value of the delineated anchor bolt area set to 1 and the remaining point cloud set to 0, in order to distinguish the anchor bolts from the background. S23. Perform secondary data annotation on the already labeled anchor bolt areas to further eliminate redundant points and improve the quality of data annotation; S24. Apply rotation, translation, and scaling functions to augment the labeled point cloud data, simulate different perspectives and positional changes of anchor bolts in a real environment, help the anchor bolt recognition network model learn more comprehensive feature representations, expand the dataset, and obtain several .txt format data files. This enables the anchor bolt recognition network model to better identify anchor bolt shapes and process key anchor bolt features, thereby improving the generalization ability and robustness of the anchor bolt recognition network model when processing three-dimensional spatial data.

[0010] Further, step S3 includes: S31. Parameter initialization: Before training, configure the core training parameters of the anchor recognition network model through command line parameters, including GPU, log path, number of model points, maximum number of rounds, batch size, initial learning rate, optimizer type and momentum, learning rate decay step size and ratio, and test region number. S32. Divide the preprocessed data files into several .txt format files, using one part of the data as the test set, another part as the validation set, and the remaining data as the training set. S33. Load and merge training / test data and labels from the specified path using the `provider.loadDataFile` function to construct a complete dataset; S34. Train the anchor bolt recognition network model. During training, the data is randomly shuffled and input into the anchor bolt recognition network model in batches. Calculate the loss / accuracy and update the parameters. At the same time, log the data and save the model parameters periodically. S35. After each round of training, the generalization ability of the anchor bolt recognition network model is evaluated through the test set. The loss / accuracy is recorded and compared with the training set to detect whether overfitting occurs. Finally, the model with the highest accuracy is retained as the final recognition model. S36. Select any one of the roadway data files from the collected roadway point cloud data, and use the final recognition model to identify the anchor bolts. Observe the recognition results of the final recognition model and the actual manually labeled anchor bolt results to find that the final recognition model can identify the point cloud area of ​​the anchor bolts.

[0011] Furthermore, in step S4, feature enhancement is performed on the anchor bolt point cloud recognition results, including: Alpha Shape Construction: By constructing a continuous surface between the convex hull (completely containing all data points) and the minimum boundary (completely contracted), the accuracy and surface smoothness are balanced, solving the problem of disordered and unclear shape of anchor point clouds after PointNet network recognition. It generates a continuous smooth surface model, enhances point cloud density, retains the main features, and improves the accuracy of subsequent clustering analysis. Fast Euclidean Clustering (FEC): Based on the Euclidean distance threshold, point clouds are divided into different clusters, which efficiently processes noisy data with large differences in point cloud spacing. It can realize anchor bar number statistics and point cloud segmentation, mark anchor bars with different colors, and save them as independent pcd files for subsequent analysis.

[0012] Furthermore, in step S4, the morphological optimization of the anchor point cloud recognition results after feature enhancement is performed, including: RANSAC model fitting: Randomly sample and fit a RANSAC model (such as the anchor plate plane) from the noisy point cloud. Iteratively evaluate the consistency between the in-points (valid data) and the out-points (noise) to robustly estimate the RANSAC model parameters. This allows for accurate extraction of the anchor plate plane parameters even in the presence of noise, separation of the anchor plate plane point cloud (in-points) and redundant noise (out-points), and optimization of the anchor morphology. External point cloud segmentation: Based on spatial location thresholds, points in the external point cloud with a distance greater than 0 from the internal point cloud are identified as valid anchor data, while the rest are redundant points. The anchor shape is optimized to obtain accurate anchor point cloud data.

[0013] Furthermore, in step S4, the key parameters of the anchor bolt are detected on the precise point cloud data of the anchor bolt obtained after morphological optimization, including: Anchor bolt shape fitting: Generate a minimum convex polyhedron to tightly wrap the anchor bolt point cloud, closely approximating the actual anchor bolt shape, and obtain the anchor bolt geometric model for measuring length, shape deviation, and included angle parameters, thereby improving the efficiency and accuracy of parameter detection; Anchor bolt length calculation: The volume is obtained using the hull.get_volume algorithm, the length of the cone is derived based on the convex hull volume formula, and the actual length is calculated in combination with the support plate radius. The calculation formula is as follows:

[0014] in, The length of the anchor rod. For anchor point cloud volume, The radius of the pallet; Anchor bolt and support plate angle calculation: Extract key points, including vertices, the point farthest from the support plate plane in the point cloud, the support plate center (average position of all points in the inner point cloud plane), and the support plate edge points (the intersection of the vertices translated downwards to the inner point plane). Calculate the spatial angle using the three-dimensional coordinates of the vertices, support plate center, and edge points to assess stability. The calculation formula is as follows:

[0015] in, Let A be the angle between the anchor bolt and the support plate, and let A be the coordinates of the anchor bolt vertex. , , The center point coordinates of the pallet are B( , , The coordinates of the pallet edge point are C( , , ).

[0016] Further, in step S4, anchor spacing is detected based on the detected key anchor parameters. The spacing is obtained by calculating the average nearest neighbor distance between two anchor point clouds, including: The trained anchor bolt recognition network model is used to identify anchor bolt point clouds in the roadway point cloud; the anchor bolt point cloud is converted to .txt format, and the nearest neighbor distance between any two anchor bolts is calculated; the spacing between all adjacent anchor bolts is counted.

[0017] This invention also provides an intelligent identification and statistical system for anchor bolts in underground mine roadways based on the above-mentioned intelligent identification and statistical method, comprising: The data acquisition module uses high-precision 3D laser scanning equipment, combined with a SLAM system, to scan the underground roadway and acquire point cloud data before and after the installation of anchor mesh and anchor bolt support. The data preprocessing module performs point cloud thinning, data annotation, feature optimization, and data augmentation on the collected point cloud data to obtain high-quality preprocessed point cloud data. An anchor bolt point cloud recognition module: Construct a PointNet network framework as the anchor bolt recognition network model. Use a deep convolutional network to train and infer on the preprocessed high-quality point cloud data to realize anchor bolt point cloud recognition in underground tunnels. The automated anchor bolt parameter statistics module performs feature enhancement, morphological optimization, and key point statistics on the anchor bolt point cloud recognition results to obtain anchor bolt parameters, including anchor bolt external length, anchor bolt inclination angle, anchor bolt shape, number of anchor bolts, and anchor bolt spacing.

[0018] Compared with the prior art, the present invention has the following advantages: 1. This invention uses three-dimensional laser scanning technology to realize the automatic collection and positioning of tunnel point cloud data without the need for manual intervention in coordinate calibration, reducing the technical threshold for operators and making the acceptance of support quality more convenient and standardized.

[0019] 2. This invention significantly reduces data processing time through point cloud thinning technology. The automated parameter statistics function avoids the tedious process of manual measurement, providing efficient technical support for safe mine production.

[0020] 3. This invention uses the PointNet network framework combined with data augmentation technology to automatically identify anchor bolt point clouds and extract key anchor bolt parameters, thereby changing the subjectivity and inefficiency of traditional manual acceptance and improving the level of intelligent acceptance.

[0021] 4. This invention utilizes Alpha shape construction, RANSAC model fitting, and convex hull algorithm to solve the problems of noise interference and shape ambiguity in point cloud data, thereby improving the accuracy of anchor bolt parameter detection. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 A flowchart illustrating an intelligent identification and statistical method for anchor bolts in underground mine roadways, provided as an embodiment of the present invention.

[0024] Figure 2 This is a flowchart of anchor bolt point cloud recognition provided in an embodiment of the present invention.

[0025] Figure 3 This is a local anchor point cloud distribution map provided for an embodiment of the present invention.

[0026] Figure 4This invention provides a method for acquiring point clouds of adjacent anchor bolts in an embodiment of the invention.

[0027] Figure 5 The calculation of the point cloud spacing between adjacent anchor bolts provided in the embodiments of the present invention, wherein, Figure 5 (a) shows the calculated anchor spacing. Figure 5 (b) shows the statistical results of the spacing between adjacent anchor bolts in the roadway. Detailed Implementation

[0028] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0029] It should be noted that the terms "comprising" and "having" and any variations thereof in the specification, claims and accompanying drawings of this invention are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to such processes, methods, products or devices.

[0030] like Figure 1 As shown, this invention provides a method for intelligent identification and statistical analysis of anchor bolts in underground mine roadways, including: S1. Using high-precision three-dimensional laser scanning equipment, combined with a SLAM system, the underground roadway is scanned to obtain point cloud data before and after the installation of anchor mesh and anchor bolt support. S2. Perform point cloud thinning, data annotation, feature optimization and data augmentation on the collected point cloud data to obtain high-quality preprocessed point cloud data; S3. Construct the PointNet network framework as the anchor bolt recognition network model. Use deep convolutional networks to train and infer on the preprocessed high-quality point cloud data to realize anchor bolt point cloud recognition in underground tunnels. S4. Perform feature enhancement, morphological optimization, and key point statistics on the anchor bolt point cloud recognition results to obtain anchor bolt parameters, including anchor bolt external length, anchor bolt inclination angle, anchor bolt shape, number of anchor bolts, and anchor bolt spacing.

[0031] In a specific implementation, as a preferred embodiment of the present invention, in step S1, a high-precision, densely sampled Geo SLAM-ZEB Horizon 3D laser scanner, equipped with a SLAM system, is used to collect point cloud data of different areas of the Yanqianshan Iron Mine before and after anchor bolt support in underground roadways. The SLAM system includes a lidar and a wheeled odometer, wherein: LiDAR is used to acquire environmental information by scanning the surrounding 360° environment to measure obstacle information. In this embodiment, the LiDAR plays two main roles: firstly, it provides point cloud data for the mapping algorithm. Once the mapping algorithm acquires sufficient point cloud data, it can construct a local map centered on the device with the radar range as its radius; secondly, the difference between the LiDAR observation data and the wheel odometer's predicted pose can be used as observation information to correct the Bayesian filter's predicted pose, thereby improving the accuracy of the filtered device pose estimation.

[0032] A wheeled odometer continuously acquires mileage information from the device via a motor encoder on a moving chassis. In this embodiment, the primary function of the wheeled odometer is to provide mileage information to the SLAM system.

[0033] In specific implementation, as a preferred embodiment of the present invention, such as Figure 2 As shown, step S2 includes: S21. Perform point cloud thinning on the collected point cloud data, identify and remove redundant point cloud data according to a preset thinning ratio, and reduce the amount of original point cloud data while maintaining the geometric features of the point cloud. Specifically, this includes: Load point cloud data of anchor mesh support in the underground roadway of Yanqianshan Mine, which was acquired by Geo SLAM-ZEB Horizon 3D laser scanner. The point cloud contains approximately 27 million data points. The thinning ratio was set to 0.004, and a point cloud thinning algorithm was used to process the actual acquired original 3D point cloud to obtain thinned point cloud data. In this embodiment, the number of data points decreased from the original 27 million to approximately 10 million, representing about 37% of the original number. Simultaneously, the data size also decreased from 1.8GB to 0.64GB. Therefore, it can be concluded that the number of data points decreased after thinning, and the data size also decreased. Furthermore, although the data density decreased after point cloud thinning, the data integrity was maintained.

[0034] S22. Label the point cloud data after point cloud thinning. Manually label the anchor regions in the point cloud data to provide necessary training, validation, and testing data for the subsequent convolutional neural network. Specifically, this includes: Load the point cloud data of the underground tunnels after point cloud thinning; Since the tunnel floor does not contain anchor bolt data, the point cloud data of the tunnel floor is cut off from the original point cloud data, while the remaining tunnel point cloud information is retained. The remaining point cloud of the roadway is rotated in three dimensions to delineate the boundary of the anchor bolts until all anchor bolt areas in the roadway are delineated. Data labels are then set, with the point cloud label value of the delineated anchor bolt area set to 1 and the remaining point cloud set to 0, in order to distinguish the anchor bolts from the background. S23. Perform secondary data annotation on the already labeled anchor bolt areas to further eliminate redundant points and improve the quality of data annotation; S24. Rotation, translation, and scaling functions are applied to augment the labeled point cloud data, simulating different perspectives and positional changes of anchor bolts in a real-world environment. This helps the anchor bolt recognition network model learn more comprehensive feature representations, expanding the dataset to obtain 12 .txt format data files, each with no fewer than 13 million data points and a total data size of approximately 21GB. This enables the anchor bolt recognition network model to better identify anchor bolt shapes and process key anchor bolt features, improving its generalization ability and robustness when processing 3D spatial data.

[0035] In a specific implementation, as a preferred embodiment of the present invention, step S3 includes: S31. Parameter Initialization: Before training, configure the core training parameters of the anchor recognition network model via command-line parameters, including GPU, log path, number of model points, maximum number of epochs, batch size, initial learning rate, optimizer type and momentum, learning rate decay step size and ratio, and test region number; as shown in Table 1 below: Table 1 Core Training Parameter Configuration

[0036] S32. Divide the 12 .txt format data files obtained after preprocessing into a test set, randomly select 60% as the test set, 20% as the validation set, and the remaining 20% ​​as the training set. S33. Load and merge training / test data and labels from the specified path using the `provider.loadDataFile` function to construct a complete dataset; S34. Train the anchor bolt recognition network model. During training, the data is randomly shuffled and input into the anchor bolt recognition network model in batches. Calculate the loss / accuracy and update the parameters. At the same time, log the data and save the model parameters periodically. S35. After each round of training, the generalization ability of the anchor bolt recognition network model is evaluated through the test set. The loss / accuracy is recorded and compared with the training set to detect whether overfitting occurs. Finally, the model with the highest accuracy is retained as the final recognition model. S36. Select any data file of a roadway from the collected point cloud data of the roadways, and use the final recognition model to identify anchor bolts. Observe the recognition results of the final recognition model and the actual manually labeled anchor bolt results to find that the recognition results of the final recognition model can identify the point cloud areas of all anchor bolts.

[0037] In a specific implementation, as a preferred embodiment of the present invention, step S4 involves feature enhancement of the anchor bolt point cloud recognition results, including: Alpha Shape Construction: By constructing a continuous surface between the convex hull (completely containing all data points) and the minimum boundary (completely contracted), the accuracy and surface smoothness are balanced, solving the problem of disordered and unclear shape of anchor point clouds after PointNet network recognition. It generates a continuous smooth surface model, enhances point cloud density, retains the main features, and improves the accuracy of subsequent clustering analysis. Fast Euclidean Clustering (FEC): Based on the Euclidean distance threshold, point clouds are divided into different clusters, which efficiently processes noisy data with large differences in point cloud spacing. It can realize anchor bar number statistics and point cloud segmentation, mark anchor bars with different colors, and save them as independent pcd files for subsequent analysis.

[0038] In a specific implementation, as a preferred embodiment of the present invention, step S4 involves morphological optimization of the anchor point cloud recognition results after feature enhancement, including: RANSAC model fitting: Randomly sample and fit a RANSAC model (such as the anchor plate plane) from the noisy point cloud. Iteratively evaluate the consistency between the in-points (valid data) and the out-points (noise) to robustly estimate the RANSAC model parameters. This allows for accurate extraction of the anchor plate plane parameters even in the presence of noise, separation of the anchor plate plane point cloud (in-points) and redundant noise (out-points), and optimization of the anchor morphology. External point cloud segmentation: Based on spatial location thresholds, points in the external point cloud with a distance greater than 0 from the internal point cloud are identified as valid anchor data, while the rest are redundant points. The anchor shape is optimized to obtain accurate anchor point cloud data.

[0039] In a preferred embodiment of the invention, step S4 involves detecting key anchor parameters on the precise point cloud data obtained after morphological optimization. This includes: anchor morphological fitting: generating a minimum convex polyhedron to tightly wrap the anchor point cloud, closely approximating the actual anchor shape, and obtaining an anchor geometric model for measuring length, shape deviation, and angle parameters, thereby improving parameter detection efficiency and accuracy. In this embodiment, after obtaining precise point cloud data through anchor morphological optimization, a convex hull algorithm is used to construct a geometric model that closely matches the actual anchor. This model can not only quantify key parameters such as anchor length, shape deviation, and angle with the support plate, but its minimum convex polyhedron characteristics can also significantly improve the accuracy and efficiency of parameter detection. The convex hull is processed, where both triangular meshes and wireframe representations can be used for shape analysis, with the latter facilitating a more intuitive understanding of the anchor morphological characteristics.

[0040] Anchor bolt length calculation: The volume is obtained using the hull.get_volume algorithm, the length of the cone is derived based on the convex hull volume formula, and the actual length is calculated in combination with the support plate radius. The calculation formula is as follows:

[0041] in, The length of the anchor rod. For anchor point cloud volume, The radius of the pallet; Anchor bolt and support plate angle calculation: Extract key points, including vertices, the point farthest from the support plate plane in the point cloud, the support plate center (average position of all points in the inner point cloud plane), and the support plate edge points (the intersection of the vertices translated downwards to the inner point plane). Calculate the spatial angle using the three-dimensional coordinates of the vertices, support plate center, and edge points to assess stability. The calculation formula is as follows:

[0042] in, Let A be the angle between the anchor bolt and the support plate, and let A be the coordinates of the anchor bolt vertex. , , The center point coordinates of the pallet are B( , , The coordinates of the pallet edge point are C( , , ).

[0043] In a specific implementation, as a preferred embodiment of the present invention, in step S4, anchor spacing is detected based on the detected key anchor parameters, and the spacing is obtained by calculating the average nearest distance between two anchor point clouds, including: The trained anchor bolt recognition network model is used to identify anchor bolt point clouds in the tunnel point cloud, such as... Figure 3As shown. Convert the anchor point cloud to .txt format, and iterate through it to calculate the nearest neighbor distance between any two anchors, as shown. Figure 4 As shown. The spacing between all adjacent anchor bolts is counted, as follows: Figure 5 As shown, where Figure 5 (a) gives the calculation results of the anchor bolt spacing. Figure 5 (b) provides the statistical results of the spacing between adjacent anchor bolts in the roadway.

[0044] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for intelligent identification and statistical analysis of anchor bolts in underground mine roadways, characterized in that, include: S1. Using high-precision three-dimensional laser scanning equipment, combined with a SLAM system, the underground roadway is scanned to obtain point cloud data before and after the installation of anchor mesh and anchor bolt support. S2. Perform point cloud thinning, data annotation, feature optimization and data augmentation on the collected point cloud data to obtain high-quality preprocessed point cloud data; S3. Construct the PointNet network framework as the anchor bolt recognition network model. Use deep convolutional networks to train and infer on the preprocessed high-quality point cloud data to realize anchor bolt point cloud recognition in underground tunnels. S4. Perform feature enhancement, morphological optimization, and key point statistics on the anchor bolt point cloud recognition results to obtain anchor bolt parameters, including anchor bolt external length, anchor bolt inclination angle, anchor bolt shape, number of anchor bolts, and anchor bolt spacing.

2. The intelligent identification and statistical method for anchor bolts in underground mine roadways according to claim 1, characterized in that, In step S1, a high-precision, densely sampled Geo SLAM-ZEB Horizon 3D laser scanner, paired with a SLAM system, is used to collect point cloud data before and after the roadway anchor support. The SLAM system includes a lidar and a wheeled odometer, wherein: LiDAR is used to acquire environmental information by scanning with lasers to measure obstacle information in the surrounding 360° environment. Wheel-type odometers continuously acquire mileage information from the device via a motor encoder on a moving chassis.

3. The intelligent identification and statistical method for anchor bolts in underground mine roadways according to claim 2, characterized in that, The lidar is used to provide point cloud data for the mapping algorithm. After the mapping algorithm acquires enough point cloud data, it constructs a local map with the device as the center and the radar range as the radius. At the same time, the difference between the lidar observation data and the wheel odometer predicted pose is used as observation information to correct the Bayesian filter's predicted pose and improve the accuracy of the filtered device pose estimation.

4. The intelligent identification and statistical method for anchor bolts in underground mine roadways according to claim 1, characterized in that, Step S2 includes: S21. Perform point cloud thinning on the collected point cloud data, identify and remove redundant point cloud data according to a preset thinning ratio, and reduce the amount of original point cloud data while maintaining the geometric features of the point cloud. Specifically, this includes: Load point cloud data of anchor mesh support in the underground roadway of Yanqianshan Mine, which was acquired by Geo SLAM-ZEB Horizon 3D laser scanner; Set the thinning ratio and use the point cloud thinning algorithm to process the actual collected original 3D point cloud to obtain the thinned point cloud data. S22. Perform data annotation on the point cloud data after point cloud thinning, manually annotating the anchor rod regions in the point cloud data, specifically including: Load the point cloud data of the underground tunnels after point cloud thinning; Since the tunnel floor does not contain anchor bolt data, the point cloud data of the tunnel floor is cut off from the original point cloud data, while the remaining tunnel point cloud information is retained. The remaining point cloud of the roadway is rotated in three dimensions to delineate the boundary of the anchor bolts until all anchor bolt areas in the roadway are delineated. Data labels are then set, with the point cloud label value of the delineated anchor bolt area set to 1 and the remaining point cloud set to 0, in order to distinguish the anchor bolts from the background. S23. Perform secondary data annotation on the already labeled anchor bolt areas to further eliminate redundant points and improve the quality of data annotation; S24. Apply rotation, translation, and scaling functions to augment the labeled point cloud data, simulate different perspectives and positional changes of anchor bolts in a real environment, help the anchor bolt recognition network model learn more comprehensive feature representations, expand the dataset, and obtain several .txt format data files, enabling the anchor bolt recognition network model to better identify anchor bolt shapes and process key anchor bolt features.

5. The intelligent identification and statistical method for anchor bolts in underground mine roadways according to claim 1, characterized in that, Step S3 includes: S31. Parameter initialization: Before training, configure the core training parameters of the anchor recognition network model through command line parameters, including GPU, log path, number of model points, maximum number of rounds, batch size, initial learning rate, optimizer type and momentum, learning rate decay step size and ratio, and test region number. S32. Divide the preprocessed data files into several .txt format files, using one part of the data as the test set, another part as the validation set, and the remaining data as the training set. S33. Load and merge training / test data and labels from the specified path using the `provider.loadDataFile` function to construct a complete dataset; S34. Train the anchor bolt recognition network model. During training, the data is randomly shuffled and input into the anchor bolt recognition network model in batches. Calculate the loss / accuracy and update the parameters. At the same time, log the data and save the model parameters periodically. S35. After each round of training, the generalization ability of the anchor bolt recognition network model is evaluated through the test set. The loss / accuracy is recorded and compared with the training set to detect whether overfitting occurs. Finally, the model with the highest accuracy is retained as the final recognition model. S36. Select any one of the roadway data files from the collected roadway point cloud data, and use the final recognition model to identify the anchor bolts. Observe the recognition results of the final recognition model and the actual manually labeled anchor bolt results to find that the final recognition model can identify the point cloud area of ​​the anchor bolts.

6. The intelligent identification and statistical method for anchor bolts in underground mine roadways according to claim 1, characterized in that, In step S4, feature enhancement is performed on the anchor bolt point cloud recognition results, including: Alpha Shape Construction: By constructing a continuous surface between the convex hull and the minimum boundary, a balance is struck between accuracy and surface smoothness, generating a continuous smooth surface model and enhancing point cloud density; Fast Euclidean Clustering: Based on the Euclidean distance threshold, point clouds are divided into different clusters, efficiently processing noisy data with large differences in point cloud spacing, realizing anchor bar number statistics and point cloud segmentation, annotating anchor bars with different colors, and saving them as independent pcd files.

7. The intelligent identification and statistical method for anchor bolts in underground mine roadways according to claim 1, characterized in that, In step S4, the morphological optimization of the anchor point cloud recognition results after feature enhancement is performed, including: RANSAC model fitting: Randomly sample and fit the RANSAC model from the noisy point cloud, evaluate the consistency between the inner and outer points through iteration, robustly estimate the RANSAC model parameters, separate the anchor plate planar point cloud from redundant noise, and optimize the anchor shape. External point cloud segmentation: Based on spatial location thresholds, points in the external point cloud with a distance greater than 0 from the internal point cloud are identified as valid anchor data, while the rest are redundant points. The anchor shape is optimized to obtain accurate anchor point cloud data.

8. The intelligent identification and statistical method for anchor bolts in underground mine roadways according to claim 1, characterized in that, In step S4, the key parameters of the anchor bolt are detected on the precise point cloud data of the anchor bolt obtained after morphological optimization, including: Anchor bolt shape fitting: Generate a minimum convex polyhedron to tightly wrap the anchor bolt point cloud, closely approximating the actual anchor bolt shape, and obtain the anchor bolt geometric model for measuring length, shape deviation, and included angle parameters, thereby improving the efficiency and accuracy of parameter detection; Anchor bolt length calculation: The volume is obtained using the hull.get_volume algorithm, the length of the cone is derived based on the convex hull volume formula, and the actual length is calculated in combination with the support plate radius. The calculation formula is as follows: in, The length of the anchor rod. For anchor point cloud volume, The radius of the pallet; Anchor bolt and support plate angle calculation: Extract key points, including vertices, the point farthest from the support plate plane in the point cloud, the support plate center, and the support plate edge points. Calculate the spatial angle using the three-dimensional coordinates of the vertices, support plate center, and edge points to assess stability. The calculation formula is as follows: in, Let A be the angle between the anchor bolt and the support plate, and let A be the coordinates of the anchor bolt vertex. , , The center point coordinates of the pallet are B( , , The coordinates of the pallet edge point are C( , , ).

9. The intelligent identification and statistical method for anchor bolts in underground mine roadways according to claim 1, characterized in that, In step S4, anchor spacing is detected based on the detected key anchor parameters. The spacing is obtained by calculating the average nearest distance between two anchor point clouds, including: The trained anchor bolt recognition network model is used to identify anchor bolt point clouds in the roadway point cloud; the anchor bolt point cloud is converted to .txt format, and the nearest neighbor distance between any two anchor bolts is calculated; the spacing between all adjacent anchor bolts is counted.

10. A system for intelligent identification and statistical analysis of anchor bolts in underground mine roadways, based on the intelligent identification and statistical analysis method for anchor bolts in underground mine roadways according to any one of claims 1-9, characterized in that, include: The data acquisition module uses high-precision 3D laser scanning equipment, combined with a SLAM system, to scan the underground roadway and acquire point cloud data before and after the installation of anchor mesh and anchor bolt support. The data preprocessing module performs point cloud thinning, data annotation, feature optimization, and data augmentation on the collected point cloud data to obtain high-quality preprocessed point cloud data. An anchor bolt point cloud recognition module: Construct a PointNet network framework as the anchor bolt recognition network model. Use a deep convolutional network to train and infer on the preprocessed high-quality point cloud data to realize anchor bolt point cloud recognition in underground tunnels. The automated anchor bolt parameter statistics module performs feature enhancement, morphological optimization, and key point statistics on the anchor bolt point cloud recognition results to obtain anchor bolt parameters, including anchor bolt external length, anchor bolt inclination angle, anchor bolt shape, number of anchor bolts, and anchor bolt spacing.

Images