Slope half-hole rate calculation method and system based on binocular vision and deep learning

Abstract

The application provides a kind of based on binocular vision and deep learning's side slope half-hole rate calculation method and system, belong to the field of blasting technology, this method includes: through binocular camera obtains the left and right image data of slope surface after blasting, and through sensor collects environmental data;Left and right image data are input into stereo matching algorithm, and disparity map is obtained, and based on the parameter configuration of binocular camera, point cloud data is calculated by triangulation principle;Point cloud data is fused with pre-stored geological data and feature extraction is carried out, and multi-feature point cloud data is obtained;Multi-feature point cloud data is input into pre-trained blast hole recognition model, and blast hole area distribution is obtained;Euclidean clustering algorithm is used to segment blast hole area distribution, and blast hole point cloud is obtained;Based on the geometric features of blast hole point cloud and the preset blasting design parameters, the half-hole rate is calculated, and the half-hole rate is obtained, and the half-hole rate evaluation result is generated and output based on the half-hole rate. The application improves the accuracy of half-hole rate calculation.

Description

Technical Field

[0001] This application relates to the field of blasting technology, and in particular to a method and system for calculating the half-pore ratio of slopes based on binocular vision and deep learning. Background Technology

[0002] The quality of blasting on open-pit slopes is a core guarantee for the safety and stability of construction projects such as open-pit mines and transportation infrastructure, and the half-hole ratio is a key indicator characterizing the blasting effect. Traditional half-hole ratio detection is mostly based on manual on-site observation and two-dimensional image estimation, which suffers from problems such as strong subjectivity, low efficiency, and poor accuracy. Moreover, the complex environment of the slope after blasting, with factors such as dust and uneven lighting interfering with the detection results, cannot meet the control requirements of large-scale projects. With the development of binocular vision and deep learning technologies, three-dimensional detection methods have gradually emerged. However, existing technologies often suffer from shortcomings such as insufficient accuracy of point cloud data, susceptibility of borehole identification to geological conditions, and insufficient fusion of multi-source data, which limits the accuracy of half-hole ratio calculation and its environmental adaptability.

[0003] Therefore, there is an urgent need to develop a method and system for calculating the half-pore ratio of slopes based on binocular vision and deep learning. Summary of the Invention

[0004] To address the aforementioned technical issues, this application provides a method and system for calculating the semi-permeability of slopes based on binocular vision and deep learning.

[0005] A first aspect of this application provides a method for calculating the partial porosity of a slope based on binocular vision and deep learning, including: The left and right image data of the slope surface after blasting are obtained by using a binocular camera, and environmental data is collected by sensors. The left and right image data are input into a stereo matching algorithm to obtain a disparity map; Based on the disparity map and the parameter configuration of the binocular camera, point cloud data is calculated using the principle of triangulation. The point cloud data is fused with the geological data pre-stored on the blasted slope to obtain the target point cloud data; Feature extraction is performed on the target point cloud data to obtain multi-feature point cloud data; The multi-feature point cloud data is input into a pre-trained borehole recognition model to obtain the borehole region distribution; the Euclidean clustering algorithm is used to segment the borehole region distribution to obtain the borehole point cloud. The half-hole ratio is calculated based on the geometric features of the borehole point cloud and the preset blasting design parameters. Based on the semi-porosity, semi-porosity evaluation results are generated and output for post-explosion quality assessment and parameter feedback optimization.

[0006] A second aspect of this application provides a slope partial porosity calculation system based on binocular vision and deep learning, comprising: The data acquisition module is used to acquire left and right image data of the slope surface after blasting through a binocular camera, and to collect environmental data through sensors; The data processing module is used to input the left and right image data into the stereo matching algorithm to obtain a disparity map; based on the disparity map and the parameter configuration of the binocular camera, point cloud data is calculated using the triangulation principle. The data fusion module is used to fuse the point cloud data with the geological data pre-stored on the blasted slope to obtain the target point cloud data; The feature engineering module is used to extract features from the target point cloud data to obtain multi-feature point cloud data. The intelligent recognition module is used to input the multi-feature point cloud data into a pre-trained borehole recognition model to obtain the borehole region distribution; and to segment the borehole region distribution using a Euclidean clustering algorithm to obtain the borehole point cloud. The porosity calculation module is used to calculate the half porosity based on the geometric features of the borehole point cloud and the preset blasting design parameters, and obtain the half porosity. The report output module is used to generate and output half-porosity evaluation results based on the half-porosity for post-explosion quality assessment and parameter feedback optimization.

[0007] A third aspect of this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the steps of the above-described method for calculating the semi-pore ratio of slopes based on binocular vision and deep learning.

[0008] A fourth aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described method for calculating the half-pore ratio of a slope based on binocular vision and deep learning.

[0009] The beneficial effects of the slope half-pore ratio calculation method and system based on binocular vision and deep learning provided in this application are as follows: This application uses a binocular camera to acquire stereo images and generates point cloud data based on the principle of triangulation, while simultaneously integrating pre-stored geological data, thus improving the completeness and accuracy of the three-dimensional data and avoiding the limitations of single visual data. Furthermore, based on a pre-trained borehole recognition model and Euclidean clustering algorithm, automated identification and accurate segmentation of borehole areas are achieved, improving adaptability to large-scale detection needs in complex slope environments. At the same time, the coupled calculation based on borehole geometric features and blasting design parameters improves the accuracy of the half-pore ratio results. This application effectively improves the efficiency and accuracy of slope half-pore ratio calculation. Attached Figure Description

[0010] Figure 1 A flowchart illustrating a slope half-porosity calculation method based on binocular vision and deep learning provided in an embodiment of this application; Figure 2 A structural block diagram of a slope semi-pore ratio calculation system based on binocular vision and deep learning provided in an embodiment of this application; Figure 3 This is a schematic block diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0011] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0012] To make the purpose, technical solution, and advantages of this application clearer, the following will be described in conjunction with the appendix. Figure 1-3 The following is an explanation using specific examples.

[0013] Please refer to Figure 1 , Figure 1 This is a flowchart illustrating a slope partial porosity calculation method based on binocular vision and deep learning, provided in an embodiment of this application. The method includes: S101: Acquire left and right image data of the slope surface after blasting using a binocular camera, and collect environmental data using sensors.

[0014] In this embodiment, a binocular camera is an imaging device that simulates the principle of human binocular vision. It can simultaneously acquire two images (left and right image data) of the same target scene through a synchronous triggering mechanism. In slope blasting detection, a binocular camera is mounted on a stable support outside the safe distance of the blasting area, with its lens facing the exposed slope surface after blasting. The left and right image data acquired by the binocular camera include two-dimensional texture information and spatial parallax information of the slope surface.

[0015] The environmental data collected by sensors refers to various quantitative parameters characterizing the on-site environmental conditions after slope blasting, mainly including meteorological data, illumination data, and residual vibration data. These data affect subsequent data processing. For example, the large amount of dust generated after blasting can cause image blurring, rainwater can interfere with texture recognition, and illumination can cause abnormal image exposure. Meteorological data includes air humidity, visibility, and the presence of fog or dust; illumination data includes ambient light intensity and the angle of incidence of light.

[0016] S102: Input the left and right image data into the stereo matching algorithm to obtain the disparity map; based on the disparity map and the parameter configuration of the binocular camera, calculate the point cloud data through the principle of triangulation.

[0017] In this embodiment, the essence of the stereo matching algorithm is to find a corresponding matching point for each pixel in the left (or right) image by comparing the left and right image data acquired by the stereo camera, and then calculate the pixel position deviation between the two matching points. This deviation value is called disparity. Finally, the disparity values ​​of all pixels are integrated into a two-dimensional image, namely a disparity map. In the disparity map, the gray value (or numerical value) of a pixel represents the magnitude of the disparity at that location. The disparity value is negatively correlated with the distance from the target object to the camera. For example, the larger the disparity value, the closer the object is to the camera; the smaller the disparity value, the farther the object is from the camera.

[0018] Due to the complex features of the slope surface, such as rock textures, borehole depressions, and undulating terrain, a robust stereo matching algorithm, such as the SGM semi-global matching algorithm, was chosen to address the matching challenges in sparse and repetitive texture regions. The input left and right image data for this stereo matching algorithm require preprocessing such as distortion correction and stereo correction. The SGM semi-global matching algorithm employs a two-tier architecture of local matching + global optimization. Specifically, the bottom layer is the local matching layer, which calculates the similarity cost of candidate matching points in the left and right images pixel-by-pixel using a cost calculation module (AD-Census cost function) to generate an initial cost volume. The upper layer is the global optimization layer, which constructs path energy functions along multiple fixed directions (8 or 16 directions) to minimize the global energy of the initial cost volume, balancing the uniqueness and continuity of the matching, and finally outputting the optimal disparity map. This hierarchical architecture retains the efficiency of local matching while resolving matching ambiguities in sparse and repetitive texture regions through global constraints. The parameters of the SGM semi-global matching algorithm are set as follows: In the cost calculation stage, the AD-Census window size is set to 9×9; the disparity search range is set to 0-128 pixels based on the baseline distance and shooting distance of the binocular camera (including the depth change from near to far on the slope); in the global optimization stage, the smoothing term penalty parameters P1 and P2 are set to 10 and 120 respectively, where P1 controls the penalty value when the disparity change of adjacent pixels is 1, and P2 controls the penalty value when the disparity change is greater than 1; in the post-processing stage, uniqueness constraints and left-right consistency detection are enabled, where the uniqueness threshold for uniqueness constraints is set to 1.5; and the consistency threshold for left-right consistency detection is set to 1.

[0019] The triangulation principle in this embodiment utilizes the geometric imaging model of a binocular camera to convert two-dimensional disparity information in a disparity map into three-dimensional spatial coordinates. The logic is based on the hardware parameter configuration of the binocular camera, such as the baseline distance between the two cameras (the straight-line distance between the optical centers of the two cameras), camera intrinsic parameters (focal length, principal point coordinates, etc., obtained from camera calibration), and based on the disparity value of each pixel in the disparity map, the three-dimensional coordinates (X, Y, Z) of the target point in the real world corresponding to that pixel are calculated through the geometric relationship of similar triangles.

[0020] Specifically, let the baseline distance of the binocular camera be B, the camera focal length be f, and the disparity value of a certain pixel be d. Then, the depth distance Z from that point to the camera can be calculated using the formula Z=(B×f) / d. Furthermore, based on the mapping relationship between the pixel's two-dimensional coordinates (u, v) in the image and the camera's intrinsic parameters, the X and Y coordinates of that point can be derived. After traversing all valid pixels in the disparity map and completing the three-dimensional coordinate calculation, this set of discrete three-dimensional coordinate points constitutes point cloud data. The form of the three-dimensional coordinate points in the point cloud data reconstructs the three-dimensional geometric shape of the slope surface after blasting. Each point includes not only spatial location information but also the texture and color information of the original image.

[0021] S103: The point cloud data is fused with the pre-stored geological data of the blasted slope to obtain the target point cloud data.

[0022] In this embodiment, point cloud data cannot reflect key geological features of the slope, such as lithological differences and structural surface distribution. These features affect the accuracy of borehole identification and the engineering rationality of half-hole ratio calculation. Specifically, the impact on borehole identification accuracy is that differences in surface reflection characteristics in different lithological regions may interfere with borehole contour judgment; the impact on the engineering rationality of half-hole ratio calculation is that the integrity of boreholes in areas with developed structural surfaces is affected by geological conditions. Therefore, point cloud data is fused with pre-stored geological data. Specifically, pre-stored geological data of the blasting slope is acquired, including rock type distribution data, structural surface attitude data, and a topographic digital elevation model. Next, the point cloud data and the topographic digital elevation model are spatially registered to the same coordinate system, and the spatial position residual is calculated. Then, the point cloud data is corrected based on the spatial position residual to generate corrected point cloud data. Finally, the rock type distribution data and structural surface attitude data are used as constraints to perform regional division and feature fusion on the corrected point cloud data to obtain the target point cloud data. Among them, the pre-stored geological data refers to the basic geological information of the slope obtained through geological surveys, drilling, remote sensing and other means before blasting construction.

[0023] S104: Extract features from the target point cloud data to obtain multi-feature point cloud data.

[0024] In this embodiment, multi-dimensional feature extraction is performed on the target point cloud data to obtain multi-feature point cloud data that integrates geometric morphological features, geological correlation features, and texture reflection features. For example, the feature extraction process adopts a hierarchical extraction strategy, covering three dimensions: bottom-level geometric features, middle-level correlation features, and high-level semantic features.

[0025] For the underlying geometric features, a neighborhood space for each point is constructed based on the K-nearest neighbor algorithm. The normal vector, curvature value, concavity / convexity parameter, and point cloud density feature of the point set within the neighborhood are calculated. The normal vector and curvature value are used to characterize the undulation of the local surface. For example, the arcuate depression in the borehole area corresponds to a sudden change in curvature value. The concavity / convexity parameter is used to distinguish between protruding rock edges and concave borehole openings. The point cloud density feature can help identify the differences between fractured rock areas and intact borehole areas.

[0026] For the geological correlation characteristics of the middle layer, based on the labels of the geological data fused in the target point cloud data, the normalized value of lithological reflection intensity and the proximity characteristics of structural planes are extracted: the normalized value of lithological reflection intensity is obtained by calibrating the differences in point cloud reflectivity of different lithological regions; the proximity characteristics of structural planes are based on the pre-stored structural plane occurrence data, the spatial distance from each point to the nearest structural plane is calculated, and the point cloud features in the area affected by the structural plane are marked.

[0027] For high-level texture reflection features, the neighborhood geometry of each point is encoded by a local feature descriptor (FPFH fast point feature histogram) to generate a feature vector with rotation invariance and scale invariance, thereby realizing the semantic representation of typical borehole structures (circular orifice, columnar channel). Finally, the features of the bottom, middle and top layers are spliced ​​and normalized to assign multi-dimensional feature labels to each point in the target point cloud data, forming multi-feature point cloud data. This multi-feature point cloud data includes not only the spatial coordinate information of the point cloud data, but also its geometric shape, geological attributes and local structural features.

[0028] S105: Input the multi-feature point cloud data into the pre-trained borehole recognition model to obtain the borehole region distribution; use the Euclidean clustering algorithm to segment the borehole region distribution to obtain the borehole point cloud.

[0029] In this embodiment, multi-feature point cloud data is input into a pre-trained borehole recognition model to obtain the borehole area distribution. The pre-trained borehole recognition model uses a YOLOv8-PointNet++ hybrid model as its framework to process the multi-feature point cloud data. This borehole recognition model includes four layers: an input layer, a dual-branch feature extraction layer, a cross-modal feature fusion layer, and a detection and segmentation output layer. The input layer receives two types of standardized data: one is multi-feature point cloud data that has undergone voxel downsampling and feature normalization, including spatial coordinates, geometric morphology, and geological attribute feature vectors; the other is two-dimensional image data scaled to 640×640 pixels and enhanced with Mosaic. In the dual-branch feature extraction layer, the PointNet++ branch completes feature embedding through three fully connected layers (mapping the feature dimension from 12 to 64 dimensions), and then outputs 128-dimensional point-level three-dimensional features and 256-dimensional global three-dimensional features through a three-scale bundled feature extraction module. The YOLOv8 branch then integrates the first layer of the backbone network... The convolutional kernels are increased from 64-dimensional to 96-dimensional. Through the C2f module (the backbone network layer of the 2D feature extraction branch) and the FPN+PAN layer (the feature fusion sub-layer of the 2D feature extraction branch), 2D feature maps of scales of 80×80, 40×40, and 20×20 are output. The cross-modal feature fusion layer maps the point-level 3D features to a 3D point feature map (80×80×128) that matches the 2D feature map through coordinate projection, and maps the global 3D features to a global feature matrix of the same size. After being weighted by the CBAM attention mechanism, it is concatenated with the 2D feature map in the channel dimension (outputting fused feature maps of 80×80×512, etc.). The detection and segmentation output layer adopts the YOLOv8 decoupled head structure, and outputs the coordinates of the borehole bounding boxes according to CIoU loss (bounding box regression), BCEWithLogitsLoss (classification), and DiceLoss (segmentation), thus obtaining the distribution of the borehole region.

[0030] This embodiment sets the parameters of the borehole recognition model. Specifically, the PointNet++ branch uses pure point cloud borehole data for training, with the parameters set as batch size=32 and learning rate=1e. -3 The YOLOv8 branch is trained using pure image-based borehole data, with parameters set to batch size = 16 and learning rate = 1e. -3 Iterate for 100 epochs.

[0031] Based on the point cloud data of the borehole area distribution, a KD-Tree spatial index is constructed. A seed point is selected from the point cloud data of the borehole area distribution, and the KD-Tree spatial index is used to search for neighboring points within a preset neighborhood search radius threshold. If the number of neighboring points is greater than or equal to a preset minimum point cloud quantity threshold for a single class, the neighboring points are marked as a cluster. If the number of neighboring points is less than the minimum point cloud quantity threshold for a single class, a repeated search mechanism is executed. The distance between clusters is calculated, and adjacent clusters with a distance less than a preset second threshold are merged to obtain the borehole point cloud.

[0032] S106: The half-hole ratio is calculated based on the geometric features of the borehole point cloud and the preset blasting design parameters.

[0033] In this embodiment, the preset blasting design parameters are engineering technical standards formulated before blasting construction. They include the total number of design holes in the slope area (the total number of blast holes planned in the blasting scheme, denoted as M), the theoretical spatial position of a single blast hole (three-dimensional coordinates (X0, Y0, Z0), marked by the blasting design drawings), and the theoretical radius (the design diameter of the blast hole construction, denoted as R0), which serve as a reference benchmark for blast hole detection.

[0034] Secondly, the geometric features of the borehole point cloud are extracted, including: 1) Borehole opening diameter: the diameter D is calculated by fitting the minimum circumcircle of the borehole point cloud (the average of multiple fitting results is taken); 2) Hole trace depth: the vertical distance H from the borehole opening to the bottom along the borehole axis (calculated by point cloud projection and depth integration); 3) Hole trace integrity: the ratio of the actual retained hole trace length to the designed hole depth (denoted as C, with a value range of [0,1], the designed hole depth is provided by the blasting design parameters); 4) Three-dimensional coordinates of the center point of the borehole point cloud: the position coordinates (X,Y,Z) of the actual borehole are obtained by calculating the spatial geometric center of the borehole point cloud set.

[0035] Subsequently, the position coordinates (X,Y,Z) of each actual blast hole are compared with the theoretical spatial position (X0,Y0,Z0) in the blasting design parameters. A spatial matching threshold is set (0.3-0.5m, the specific value is adjusted according to the blast hole layout density). If the Euclidean distance between two points is less than or equal to the spatial matching threshold, it is determined to be an actual blast hole that is retained. If no matching item is found after traversing all theoretical positions, or if no corresponding actual blast hole point cloud is found for a certain theoretical position, it is determined to be a blast hole missing after blasting. The design number, theoretical spatial position and quantity of the missing blast hole are recorded simultaneously (denoted as M1).

[0036] Next, based on the dual indicators of borehole opening diameter and borehole trace integrity, the actual retained boreholes are classified and judged. Specifically, integrity thresholds are set: diameter matching threshold: the ratio of borehole opening diameter D to theoretical radius R0 must be within the range of [0.8, 1.2] (allowing small borehole diameter deviations caused by construction and blasting); borehole trace integrity threshold: the borehole trace integrity C is greater than or equal to 0.7 (ensuring that the main structure of the borehole is retained). If the actual retained borehole meets both of the above integrity threshold requirements, it is judged as a complete half-hole (the number is recorded as N); if only one of them is met or neither is met, it is judged as an incomplete half-hole, and the geometric feature defects (excessive borehole diameter deviation, broken borehole trace, insufficient integrity, etc.) and the number (recorded as M2) of the incomplete half-hole are recorded.

[0037] Finally, the half-porosity is calculated: the formula for calculating the half-porosity (η) is η=(N / M)×100%, where M is the total number of holes in the blasting design parameters and satisfies M=N+M1+M2.

[0038] S107: Based on the semi-porosity, generate and output semi-porosity evaluation results for post-explosion quality assessment and parameter feedback optimization.

[0039] In this embodiment, a half-porosity evaluation result is generated and output based on the half-porosity for post-blast quality assessment and parameter feedback optimization. The goal is to use the half-porosity as an evaluation conclusion and optimization suggestion with engineering guidance significance. A higher half-porosity indicates more intact blast hole traces remaining on the borehole wall, and the degree of rock fragmentation and contour regularity after blasting are more in line with design requirements. Conversely, a lower half-porosity indicates deviations in blasting parameters or insufficient consideration of the influence of geological data. Reasons for deviations in blasting parameters include unreasonable explosive dosage, charge structure, and detonation sequence.

[0040] The half-hole ratio evaluation results are generated based on blasting design standards and actual engineering scenarios. For example, firstly, the half-hole ratio value is compared with the acceptable threshold of 80% or greater and the excellent threshold of 90% or greater, set for slope grade and rock mass type, to obtain the blasting quality grade, which includes excellent, acceptable, and unacceptable. Secondly, for areas with unacceptable half-hole ratios, the root cause of the problem is traced based on previously integrated geological data and borehole distribution characteristics. For example, if the half-hole ratio in a certain area is unacceptable and the corresponding geological data shows the presence of thick, weak interlayers, it can be determined that the poor blasting effect is due to geological conditions. Finally, based on the root cause of the problem, targeted parameter optimization suggestions are proposed, such as adjusting the explosive consumption per unit, optimizing the ratio of charge length to plugging length, and adjusting the initiation time difference.

[0041] The output of the half-pore rate evaluation results includes visualization charts, quality assessment reports, and parameter optimization schemes. The visualization charts include a heat map of the half-pore rate distribution on the slope and a correlation analysis diagram between the half-pore rate and blasting parameters. The quality assessment report includes the overall blasting quality grade, key problem areas, and analysis of the causes of the problems. The parameter optimization scheme includes the adjustment range and basis of the blasting parameters. The heat map of the half-pore rate distribution on the slope shows the differences in half-pore rate in the same area.

[0042] As can be seen from the above, this application improves the completeness and accuracy of 3D data by using a binocular camera to acquire stereo images and generating point cloud data based on the principle of triangulation, while simultaneously integrating pre-stored geological data, thus avoiding the limitations of single-vision data. Furthermore, based on a pre-trained borehole recognition model and Euclidean clustering algorithm, automated identification and precise segmentation of borehole areas are achieved, improving adaptability to large-scale inspection needs in complex slope environments. At the same time, the coupled calculation based on borehole geometric features and blasting design parameters improves the accuracy of the half-hole ratio results. This application effectively improves the efficiency and accuracy of slope half-hole ratio calculation.

[0043] In one embodiment of this application, before inputting the left and right image data into the stereo matching algorithm to obtain the disparity map, the method further includes: Based on the pre-calibrated intrinsic parameters and distortion coefficients of the binocular camera, distortion correction is performed on the left and right image data respectively; Stereo correction is performed on the distortion-corrected left and right image data to obtain corrected left and right image data; The left and right image data are processed based on an adaptive processing strategy to obtain the target left and right image data; The adaptive processing strategy includes: Based on meteorological data in the environmental data, determine whether there is atmospheric scattering-type deterioration in the left and right image data caused by explosion dust, natural fog or rainwater adhesion; If atmospheric scattering-related degradation exists, an image dehazing algorithm based on dark channel priors and atmospheric scattering physical models is used for processing. If there is no atmospheric scattering-type degradation, the light intensity in the environmental data is calculated, and the left and right image data corresponding to light intensities less than a preset first threshold are processed using the enhancement algorithm based on Retinex theory to obtain the target left and right image data.

[0044] In this embodiment, firstly, manufacturing errors and installation deviations of the binocular camera lens can cause radial and tangential distortion in the image, thereby affecting the spatial accuracy of the pixels. However, the distortion coefficients of the camera's internal components, which are obtained in advance through methods such as Zhang's calibration, can be used to correct the actual position of each pixel through geometric transformation formulas, thus eliminating the image distortion caused by the distortion.

[0045] Secondly, the two cameras in a stereo camera cannot perfectly guarantee that their optical axes are parallel and their imaging planes are coplanar. Directly using this for stereo matching would significantly increase the difficulty of finding corresponding points. Stereo correction, however, uses the epipolar constraint principle to project the left and right images onto the same virtual imaging plane, ensuring that pixels on the same epipolar line are in the same row in both images. This achieves epipolar alignment and reduces the computational complexity of stereo matching algorithms.

[0046] Finally, the corrected left and right image data are processed based on an adaptive processing strategy to obtain the target left and right image data. This adaptive processing strategy is to adjust the image optimization scheme according to the post-blasting environment. Specifically, meteorological data is extracted from the environmental data collected by the sensor, and it is determined whether there is atmospheric scattering degradation caused by blasting dust, natural fog, or rainwater. This atmospheric scattering degradation will reduce image contrast and blur details. An image dehazing algorithm based on dark channel prior and atmospheric scattering physical model is used to restore clear image texture by estimating atmospheric light value and transmittance. If there is no atmospheric scattering degradation, the light intensity in the environmental data is further read. When the light intensity is less than a preset first threshold, an enhancement algorithm based on Retinex theory is used to process the image. By separating the illuminance component and reflection component of the image, details in low-light areas are enhanced and overexposure in high-light areas is suppressed, resulting in target left and right image data with uniform brightness and clear details. If the light intensity meets the requirements, the corrected left and right image data is used as the target left and right image data. The method for determining the first threshold is based on the imaging characteristics of the binocular camera, the image acquisition requirements of the slope blasting scene, and the statistical laws of ambient lighting. The calculation formula is as follows: I1 = a × I2 + b × c + I3 Wherein, I1 is the first threshold; I2 is the minimum illumination intensity required for the binocular camera to achieve effective texture recognition; c is the standard deviation of historical illumination intensity at the slope blasting site; I3 is the basic compensation value, used to compensate for the attenuation effect of blasting dust and slight fog on illumination, and is taken as 5%-10% of the average illumination intensity when there is no degradation at the site; a is the minimum effective illumination weight coefficient, with a value range of 0.8-1.0, and the specific value is determined based on the camera hardware capabilities and actual scene requirements; b is the illumination fluctuation compensation coefficient, with a value range of 0.2-0.5, and the specific value is determined based on the impact of illumination fluctuation on the threshold, where the larger the fluctuation (the larger c), the larger b value.

[0047] As can be seen from the above, this embodiment eliminates imaging distortion caused by camera hardware through distortion correction, achieves epipolar alignment through stereo correction to reduce matching difficulty, and specifically addresses image degradation caused by environmental factors such as blasting dust, fog, and uneven lighting through adaptive processing strategies. This not only ensures the geometric accuracy of the image at the hardware level, but also improves the clarity and detail integrity of the image at the environmental adaptation level, effectively avoiding parallax map calculation errors caused by image quality defects.

[0048] In one embodiment of this application, after processing the corrected left and right image data based on an adaptive processing strategy to obtain the target left and right image data, the method further includes: Based on the preset blasting design drawing, the slope surface area including all blast holes is located and extracted from the left and right image data of the target as the region of interest in the image; The left and right image data of the target within the region of interest in the image are processed separately, including: Anisotropic diffusion filtering is used to smooth the edges of the region of interest in the image while preserving them, and local contrast adaptive histogram equalization is used to enhance the texture of the smoothed image, resulting in enhanced left and right image data. Consistency calibration is performed on the enhanced left and right image data to obtain the left and right image data input to the stereo matching algorithm. Consistency calibration includes histogram matching to eliminate bilateral color differences.

[0049] In this embodiment, firstly, based on a preset blasting design drawing, the slope surface area including all blast holes is located and extracted from the left and right image data of the target as the region of interest. The blasting design drawing indicates the planned locations and distribution range of the blast holes, as well as the slope detection boundaries.

[0050] Secondly, the left and right image data within the region of interest are processed in two steps: The first step uses anisotropic diffusion filtering to smooth the edges. Specifically, this anisotropic diffusion filtering method differs from the uniform blurring characteristic of traditional Gaussian filtering. It can smooth the rock surface, eliminate random noise, and retain the contour information of key structures such as borehole edges and rock joints. Random noise includes particle noise from blasting dust residue and image sensor noise. The second step uses local contrast adaptive histogram equalization to enhance the texture of the smoothed image. The smoothed image is textured by dividing the image into blocks and adaptively adjusting the histogram distribution of each sub-block to obtain enhanced left and right image data.

[0051] Finally, consistency calibration is performed on the enhanced left and right image data. For example, due to slight hardware differences between the two cameras in a stereo camera (sensor sensitivity, lens transmittance), even after initial calibration, inconsistencies in color and brightness may still occur between the left and right images. These differences can affect the search accuracy of corresponding points in stereo matching. Therefore, consistency calibration uses histogram matching technology to map the grayscale distribution of the right image to the histogram distribution characteristics of the left image, eliminating color and brightness differences between the two images and obtaining the left and right image data as input to the stereo matching algorithm.

[0052] From the above, it can be concluded that this embodiment accurately delineates the region of interest on the slope surface of the blast hole using the blasting design drawing, which can eliminate irrelevant background interference and reduce the amount of data processing. The texture enhancement based on anisotropic diffusion filtering for edge smoothing and local contrast adaptive histogram equalization effectively suppresses random noise on the rock surface and highlights features such as the blast hole edges. At the same time, histogram matching is used to achieve consistency calibration of the left and right images after enhancement, eliminating color differences between the two sides and ensuring the accuracy of subsequent disparity map generation and point cloud data calculation.

[0053] In one embodiment of this application, point cloud data is fused with pre-stored geological data of a blasted slope to obtain target point cloud data, including: Acquire pre-stored geological data of the blasting slope, including rock type distribution data, structural surface attitude data, and topographic digital elevation model; Point cloud data and digital elevation model of terrain are spatially registered to the same coordinate system and spatial position residuals are calculated. Corrected point cloud data is generated by correcting the spatial location residuals. Using rock type distribution data and structural plane orientation data as constraints, the modified point cloud data is divided into regions and its features are fused to obtain the target point cloud data.

[0054] In this embodiment, pre-stored geological data of the blasting slope is acquired. The geological data includes rock type distribution data, structural plane attitude data, and topographic digital elevation model. The rock type distribution data includes the spatial distribution range and boundaries of different lithologies such as granite and sandstone. The structural plane attitude data includes the strike, dip, dip angle, and spatial location of geological structures such as joints and faults. The topographic digital elevation model represents the three-dimensional elevation information of the original topography of the slope before blasting.

[0055] This embodiment first uses registration algorithms such as ICP (Iterative Closest Point) to unify point cloud data and terrain digital elevation model under the same geographic coordinate system, and calculates the spatial position residual between the registered point cloud data and terrain digital elevation model, that is, the difference between the elevation and plane coordinates of the corresponding position; then, based on the residual, the original point cloud data is corrected point by point to generate corrected point cloud data.

[0056] Secondly, using rock type distribution data and structural plane attitude data as constraints, the corrected point cloud data is divided into regions and its features are fused. Specifically, based on the boundary information of the rock type distribution data, the corrected point cloud data is divided into regions corresponding to different lithologies, and lithology labels are assigned to the point clouds of each region, resulting in the first sub-region point cloud set. Next, based on the spatial location and attitude elements of structural planes in the structural plane attitude data, structural plane feature points are identified and marked in the corresponding sub-region point clouds, resulting in the second sub-region point cloud set. Finally, the first and second sub-region point cloud sets are fused to obtain the target point cloud data with three-dimensional morphology and geological attributes.

[0057] As can be seen from the above, this embodiment achieves deep integration of point cloud geometric data and geological attribute data by acquiring pre-stored geological data, spatially registering point cloud data with the terrain digital elevation model and correcting residuals, and then completing regional division and feature fusion by using rock type and structural surface attitude as constraints, thus making up for the deficiency of simple point cloud lacking geological background.

[0058] In one embodiment of this application, rock type distribution data and structural plane orientation data are used as constraints to perform regional division and feature fusion on the modified point cloud data to obtain target point cloud data, including: Based on the boundaries of different lithological regions in the rock type distribution data, the corrected point cloud data is segmented to obtain sub-region point cloud sets; For each sub-region point cloud in the sub-region point cloud set, calculate the typical surface reflection characteristic value of the corresponding rock type, and normalize and correct the reflection intensity characteristic value to obtain the first sub-region point cloud set. Based on the spatial location and attitude elements of the structural surfaces in the structural surface attitude data, structural surface feature points are identified and marked in the corresponding sub-region point cloud to obtain the second sub-region point cloud set. The point cloud set of the first sub-region is fused with the point cloud set of the second sub-region to obtain the target point cloud data.

[0059] In this embodiment, the rock type distribution data labels the spatial distribution range and boundary coordinates of different lithologies such as granite, sandstone, and shale. The corrected point cloud data is segmented based on the boundaries of different lithological regions within the rock type distribution data. For example, a region growing algorithm is used to divide the corrected point cloud data into sub-region point cloud sets corresponding one-to-one with lithological regions. Each sub-region point cloud includes point cloud data within its corresponding lithological range. Next, the reflection intensity characteristics of each sub-region point cloud are normalized and corrected. Specifically, typical surface reflection characteristic values ​​of each lithological region are obtained through geological sample testing. The deviation between the reflection intensity in the sub-region point cloud and the typical values ​​is then compared. A linear transformation is used to normalize the reflection intensity characteristic values ​​of all points within the region, ensuring consistency in the reflection characteristics of point clouds within the same lithological region. The processed sub-region point cloud is the first sub-region point cloud set.

[0060] Simultaneously, based on the structural plane attitude data, the identification and labeling of structural plane feature points in the sub-region point cloud are completed. According to the spatial location and attitude elements of structural planes such as joints and faults in the surface attitude data, the point cloud data located at the structural planes are identified in the corresponding sub-region point cloud through algorithms such as normal vector mutation detection and plane fitting. Structural plane attribute labels are added to these feature points to indicate the correlation between the point cloud data and the geological structural planes. The labeled regional point cloud data area is the second sub-region point cloud set.

[0061] Finally, the point cloud sets of the first and second sub-regions are merged. The fusion process is not a simple point cloud overlay, but rather a process that simultaneously assigns lithological reflection intensity normalization features and structural surface attribute labels to each point in the modified point cloud data. This ensures that each point retains basic information such as three-dimensional spatial coordinates and reflection intensity, while also being associated with geological attributes such as its lithological type and whether it is located on a structural surface. Ultimately, this results in a target point cloud data set that integrates geometric morphology, lithological characteristics, and structural surface information.

[0062] From the above, it can be concluded that this embodiment obtains sub-region point cloud sets by segmenting and correcting the point cloud according to the rock type boundary, and completes the reflection intensity normalization correction through reflection characteristic values. Secondly, it identifies and marks structural surface feature points based on surface attitude data to obtain a fused sub-region point cloud. This not only realizes the binding of point cloud data with geological attributes such as rock type and structural surface attitude, giving the point cloud data multi-dimensional geological features, but also eliminates feature distortion caused by lithological differences through reflection intensity normalization, thereby improving the recognition and effectiveness of point cloud data.

[0063] In one embodiment of this application, after inputting multi-feature point cloud data into a pre-trained borehole recognition model to obtain the borehole region distribution, the method further includes: The spatial location of the borehole area is compared with the physical spatial location of the region of interest in the image; The distribution of borehole regions whose spatial location is consistent with the physical spatial location of the region of interest in the image is used as the borehole region for segmentation using the Euclidean clustering algorithm. The locations of boreholes in the region of interest that were not identified by the borehole recognition model were marked as missed detection areas. The original image data and point cloud data corresponding to the missed detection areas were used as negative samples to train and update the borehole recognition model. The locations of boreholes outside the region of interest in the borehole area distribution are marked as false alarm areas and filtered out.

[0064] In this embodiment, the spatial location of the borehole region distribution is compared with the region of interest (ROI) in the image. The ROI is a slope area encompassing all planned boreholes, defined based on the blasting design drawing. For example, the spatial coordinates of the borehole region distribution output by the borehole recognition model are matched one-to-one with the physical spatial extent of the ROI in the image. Then, borehole regions whose spatial locations perfectly match the ROI are marked as valid borehole regions and used as input for Euclidean clustering segmentation. Simultaneously, borehole locations within the ROI that are not identified by the borehole recognition model—that is, boreholes present in the design plan but not detected by the model—are marked as missed detection regions. The corresponding original image data and point cloud data are extracted as negative samples. These negative samples are labeled as samples that should have been identified but were not, and added to the borehole recognition model's training dataset for iterative training and updating of the model.

[0065] In this embodiment, the distribution of borehole regions output by the borehole recognition model that are outside the region of interest in the image is marked as false alarm regions. False alarm regions are where the borehole recognition model mistakenly identifies rock protrusions and noise point clusters as boreholes, and these false alarm regions are filtered out.

[0066] The iterative training and updating of the borehole recognition model includes parameter adjustment. Specifically, the weights of the dual-branch feature extraction layer in the PointNet++ branch and the backbone network weights in the YOLOv8 branch are frozen. Next, the channel attention weight coefficients and spatial attention convolution kernel size of the CBAM attention mechanism, as well as the classification loss weights and bounding box regression loss weights of the YOLOv8 branch output layer, are adjusted. After adjustment, under the conditions of reducing the learning rate and adjusting the batch size, the sphere neighborhood radius and the number of farthest sampling points in the bundled feature extraction layer of the PointNet++ branch, and the number of convolution kernels in the C2f module and the feature fusion coefficients of the FPN+PAN layer in the YOLOv8 branch are adjusted. For example, the weights of the dual-branch feature extraction layer are frozen first, and only the feature fusion layer and output layer are trained (with batch size=8 and learning rate 1e). -4(Iterate for 50 epochs); then unfreeze all levels and perform end-to-end adjustments (learning rate reduced to 1e). -5 (Iteration 50 epochs). During training, a multi-task loss function is introduced, and a weight coefficient of 5 times is set for the borehole category to alleviate the sample imbalance problem. At the same time, the non-maximum suppression threshold is set to 0.5 to filter redundant bounding boxes.

[0067] Furthermore, the non-maximum suppression threshold and class confidence threshold of the YOLOv8 branch detection and segmentation output layer are adjusted based on the false alarm region to suppress the false alarm response of the borehole recognition model to rock surface protrusions and noise point clusters. Specifically, the original settings of the non-maximum suppression threshold and class confidence threshold in the output results of the YOLOv8 branch detection and segmentation output layer in the false alarm region are first statistically analyzed, along with the confidence score and bounding box overlap data when false alarm samples are identified as boreholes. Then, the class confidence threshold is gradually reduced (from the initial 0.5 to 0.6-0.7) to filter out false borehole recognition results with low confidence, while the non-maximum suppression is adjusted (from the initial 0.5 to 0.4-0.45). During the adjustment process, the false recognition rate of the false alarm region is continuously verified, and finally, the optimal non-maximum suppression and class confidence threshold parameters are determined and updated to the YOLOv8 branch detection and segmentation output layer.

[0068] As can be seen from the above, this embodiment compares the spatial location of the borehole region distribution output by the borehole recognition model with the region of interest in the image, filters out the effective borehole regions with matching locations for clustering and segmentation, and simultaneously marks and uses the missed detection region data to optimize the borehole recognition model and filter out false alarm regions. This not only improves the accuracy of borehole region locking, but also reduces the missed detection rate of the borehole recognition model through iterative training and updating of the borehole recognition model.

[0069] In one embodiment of this application, a Euclidean clustering algorithm is used to segment the distribution of borehole regions to obtain a borehole point cloud, including: Based on the point cloud data of borehole area distribution, a KD-Tree spatial index is constructed. Select a seed point from the point cloud data of the borehole area distribution, and search for neighboring points within a preset neighborhood search radius threshold using the KD-Tree spatial index; If the number of neighboring points is greater than or equal to the preset threshold for the minimum number of point clouds in a single class, then the neighboring points are marked as a cluster. If the number of neighboring points is less than the threshold for the minimum number of point clouds in a single class, a repeated search mechanism is executed to obtain extended neighboring points, and extended neighboring points that are greater than or equal to the threshold for the minimum number of point clouds in a single class are taken as clusters. The distance between clusters is calculated, and adjacent clusters with a distance less than a preset second threshold are merged to obtain the borehole point cloud.

[0070] In this embodiment, the KD-Tree constructs a tree structure by recursively dividing the spatial dimensions (sequentially along the x, y, and z axes), organizing point cloud data into a hierarchical spatial index to obtain the neighborhood range that can locate the target point, reducing the time complexity of neighborhood search from O(n) to O(logn). Unmarked points are selected from the point cloud data distributed in the borehole area as seed points. Using the KD-Tree spatial index, neighborhood points of the seed points are searched within a preset neighborhood search radius threshold. This neighborhood search radius threshold is set based on the actual borehole diameter and the point cloud acquisition density; for example, a borehole diameter of 100mm and a point cloud density of 5 points / mm. 2 At this time, the radius threshold is set to 55mm to ensure complete coverage of the point cloud range of a single borehole. If the number of neighboring points is greater than or equal to the preset minimum point cloud quantity threshold for a single class, the seed point and its neighboring points are marked as an independent cluster. If the number of neighboring points is less than the minimum point cloud quantity threshold for a single class, a repeated search mechanism is executed. Specifically, the neighborhood search radius threshold is adaptively expanded and a re-search is performed to obtain extended neighboring points. Extended neighboring points that are greater than or equal to the minimum point cloud quantity threshold for a single class are used as clusters. If the extended neighboring points are still less than the minimum point cloud quantity threshold for a single class, the seed point is determined to be an isolated noise point and is removed. The minimum point cloud quantity threshold for a single class is determined by the minimum point cloud quantity that a single borehole should include.

[0071] In this embodiment, the coordinates of the center points of all clusters are calculated, and the distance between the center points of adjacent clusters is solved using the Euclidean distance formula. If the distance is less than a preset second threshold, these clusters are determined to belong to the same borehole and are merged into a complete cluster. The second threshold is set according to the minimum value of the borehole design spacing. For example, when the borehole design spacing is 500mm, the second threshold is set to 100mm. This is used to identify multiple sub-clusters of the same borehole formed due to incomplete point cloud segmentation. After merging all clusters, each merged cluster corresponds to a borehole target, and the set of all clusters is the segmented borehole point cloud.

[0072] From the above, it can be concluded that this embodiment improves the efficiency of neighbor point search by constructing a KD-Tree spatial index. The determination of the minimum number of point clouds in a single class and the repeated search mechanism effectively filter noise and avoid omission of effective borehole clusters. At the same time, the merging of adjacent clusters is completed by determining the distance threshold between clusters, which improves the efficiency and accuracy of borehole region segmentation.

[0073] In one embodiment of this application, a seed point is selected from the point cloud data of the borehole area distribution, and neighborhood points are obtained by searching within a preset neighborhood search radius threshold using the KD-Tree spatial index, including: The density of each point in the point cloud data of the borehole area is calculated to obtain the local point cloud density; Select the point with the highest local point cloud density as the current seed point; Based on the KD-Tree spatial index, a spherical region search is performed within the neighborhood search radius threshold, with the current seed point as the center. Obtain the set of all points within the spherical region as the initial neighborhood point set; The spatial distribution consistency of the initial neighborhood point set is checked, and outliers that deviate from the main distribution area are removed to obtain the neighborhood points.

[0074] In this embodiment, the local point cloud density of each point in the point cloud data of the borehole area is calculated. For example, the K-nearest neighbor density estimation method is used to count the number of points within a preset neighborhood for each point, and the point density of each point within that preset neighborhood is quantified to obtain the local point cloud density value for each point. The preset neighborhood is determined based on the K-nearest neighbor density estimation method, and this neighborhood is the spatial range of a fixed number of K nearest neighbor points around each point.

[0075] In this embodiment, the point with the highest local point cloud density is first selected as the current seed point.

[0076] Secondly, based on the KD-Tree spatial index, a spherical region search is performed within a search radius threshold in the neighborhood, centered on the current seed point. The final set of all points within the search area constitutes the initial neighborhood point set. The spherical search range conforms to the three-dimensional cylindrical shape of the borehole.

[0077] Finally, a spatial distribution consistency check is performed on the preliminary neighborhood point set to remove outliers. For example, by calculating the spatial distance and azimuth distribution of each point in the preliminary neighborhood point set relative to the seed point, dense sub-regions in the point set are identified. Scattered points that deviate from the main dense distribution area are identified as outliers and removed, resulting in neighborhood points with a concentrated spatial distribution that fit the actual contour of the borehole.

[0078] From the above, it can be concluded that this embodiment improves the accuracy of borehole area anchoring by calculating the local point cloud density and selecting the point with the highest density as the seed point; the spherical region search based on KD-Tree spatial index can improve the efficiency of obtaining the initial neighborhood point set, and the spherical range fits the borehole shape to ensure the integrity of the borehole; and outliers are removed by spatial distribution consistency test, so that the neighborhood points have both accuracy and integrity.

[0079] Corresponding to the slope semi-pore ratio calculation method based on binocular vision and deep learning in the above embodiment, Figure 2 This is a structural block diagram of a slope partial porosity calculation system based on binocular vision and deep learning, provided as an embodiment of this application. For ease of explanation, only the parts relevant to the embodiment of this application are shown. References Figure 2 The slope semi-porosity calculation system 20 based on binocular vision and deep learning includes: a data acquisition module 21, a data processing module 22, a data fusion module 23, a feature engineering module 24, an intelligent recognition module 25, a porosity calculation module 26, and a report output module 27.

[0080] Among them, the data acquisition module 21 is used to acquire left and right image data of the slope surface after blasting through a binocular camera, and to collect environmental data through a sensor. Data processing module 22 is used to input left and right image data into the stereo matching algorithm to obtain a disparity map; based on the disparity map and the parameter configuration of the binocular camera, point cloud data is calculated through the principle of triangulation. The data fusion module 23 is used to fuse point cloud data with pre-stored geological data of the blasted slope to obtain target point cloud data; Feature engineering module 24 is used to extract features from target point cloud data to obtain multi-feature point cloud data; The intelligent recognition module 25 is used to input multi-feature point cloud data into a pre-trained borehole recognition model to obtain the borehole area distribution; and to segment the borehole area distribution using a Euclidean clustering algorithm to obtain the borehole point cloud. The porosity calculation module 26 is used to calculate the half porosity based on the geometric features of the borehole point cloud and the preset blasting design parameters, and obtain the half porosity. Report output module 27 is used to generate and output half-porosity evaluation results based on half-porosity for post-explosion quality assessment and parameter feedback optimization.

[0081] See Figure 3 , Figure 3 This is a schematic block diagram of an electronic device provided according to an embodiment of this application. Figure 3 The electronic device 300 in this embodiment may include one or more processors 301, one or more input devices 302, one or more output devices 303, and one or more memories 304. The processors 301, input devices 302, output devices 303, and memories 304 communicate with each other via a communication bus 305. The memories 304 store computer programs, including program instructions. The processors 301 execute the program instructions stored in the memories 304. Specifically, the processors 301 are configured to invoke the program instructions to perform the functions of the modules in the aforementioned device embodiments, for example... Figure 2The data acquisition module 21, data processing module 22, data fusion module 23, feature engineering module 24, intelligent recognition module 25, porosity calculation module 26, and report output module 27 are shown.

[0082] It should be understood that, in the embodiments of this application, the processor 301 may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0083] Input device 302 may include a touchpad, a fingerprint sensor (for collecting the user's fingerprint information and fingerprint orientation information), a microphone, etc., and output device 303 may include a display (LCD, etc.), a speaker, etc.

[0084] The memory 304 may include read-only memory and random access memory, and provides instructions and data to the processor 301. A portion of the memory 304 may also include non-volatile random access memory. For example, the memory 304 may also store device type information.

[0085] In specific implementations, the processor 301, input device 302, and output device 303 described in the embodiments of this application can execute the implementation methods described in any embodiment of the slope half-pore ratio calculation method based on binocular vision and deep learning provided in the embodiments of this application, or they can execute the implementation methods of the electronic devices described in the embodiments of this application, which will not be repeated here.

[0086] In another embodiment of this application, a computer-readable storage medium is provided. This computer-readable storage medium stores a computer program, which includes program instructions. When executed by a processor, the program instructions implement all or part of the processes in the methods described above. Alternatively, the computer program can instruct related hardware to complete the process. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include any entity or device capable of carrying computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0087] The computer-readable storage medium can be an internal storage unit of the electronic device in any of the foregoing embodiments, such as a hard disk or memory of the electronic device. The computer-readable storage medium can also be an external storage device of the electronic device, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the electronic device. Furthermore, the computer-readable storage medium can include both internal and external storage units of the electronic device. The computer-readable storage medium is used to store computer programs and other programs and data required by the electronic device. The computer-readable storage medium can also be used to temporarily store data that has been output or will be output.

[0088] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this application.

[0089] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the electronic devices and units described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0090] In the several embodiments provided in this application, it should be understood that the disclosed electronic devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces or units, or it may be an electrical, mechanical, or other form of connection.

[0091] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of the embodiments of this application, depending on actual needs.

[0092] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0093] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for calculating the semi-permeability of slopes based on binocular vision and deep learning, characterized in that, include: The left and right image data of the slope surface after blasting are obtained by using a binocular camera, and environmental data is collected by sensors. The left and right image data are input into a stereo matching algorithm to obtain a disparity map; Based on the disparity map and the parameter configuration of the binocular camera, point cloud data is calculated using the principle of triangulation. The point cloud data is fused with the geological data pre-stored on the blasted slope to obtain the target point cloud data; Feature extraction is performed on the target point cloud data to obtain multi-feature point cloud data; The multi-feature point cloud data is input into a pre-trained borehole recognition model to obtain the borehole region distribution; the Euclidean clustering algorithm is used to segment the borehole region distribution to obtain the borehole point cloud. The half-hole ratio is calculated based on the geometric features of the borehole point cloud and the preset blasting design parameters. Based on the semi-porosity, semi-porosity evaluation results are generated and output for post-explosion quality assessment and parameter feedback optimization.

2. The method for calculating the half-pore ratio of a slope based on binocular vision and deep learning according to claim 1, characterized in that, Before obtaining the disparity map by inputting the left and right image data into the stereo matching algorithm, the algorithm further includes: Based on the pre-calibrated binocular camera intrinsic parameters and distortion coefficients, distortion correction is performed on the left and right image data respectively; Stereo correction is performed on the distortion-corrected left and right image data to obtain corrected left and right image data; The corrected left and right image data are processed based on an adaptive processing strategy to obtain the target left and right image data; The adaptive processing strategy includes: Based on the meteorological data in the environmental data, determine whether the left and right image data are subject to atmospheric scattering-type deterioration caused by blasting dust, natural fog, or rainwater adhesion; If atmospheric scattering-type degradation exists, an image dehazing algorithm based on dark channel priors and atmospheric scattering physical models is used for processing. If atmospheric scattering-type degradation does not exist, the light intensity in the environmental data is calculated, and the left and right image data corresponding to light intensities less than a preset first threshold are processed using the enhancement algorithm based on Retinex theory to obtain the target left and right image data.

3. The method for calculating the half-pore ratio of a slope based on binocular vision and deep learning according to claim 2, characterized in that, After processing the corrected left and right image data based on the adaptive processing strategy to obtain the target left and right image data, the method further includes: Based on the preset blasting design drawing, the slope surface area including all blast holes is located and extracted from the left and right image data of the target as the region of interest of the image; The left and right image data within the region of interest of the image are processed separately, and the processing includes: Anisotropic diffusion filtering is used to smooth the edges of the region of interest in the image while preserving them, and local contrast adaptive histogram equalization is used to enhance the texture of the smoothed image, resulting in enhanced left and right image data. The enhanced left and right image data are subjected to consistency calibration to obtain the left and right image data input to the stereo matching algorithm.

4. The method for calculating the half-pore ratio of a slope based on binocular vision and deep learning according to claim 1, characterized in that, The process of fusing the point cloud data with pre-stored geological data of the blasted slope to obtain target point cloud data includes: Acquire pre-stored geological data of the blasting slope, which includes rock type distribution data, structural surface attitude data, and topographic digital elevation model; The point cloud data and the terrain digital elevation model are spatially registered to the same coordinate system, and the spatial position residuals are calculated. The point cloud data is corrected based on the spatial location residual to generate corrected point cloud data; Using the rock type distribution data and the structural surface orientation data as constraints, the modified point cloud data is divided into regions and fused with features to obtain the target point cloud data.

5. The method for calculating the half-pore ratio of a slope based on binocular vision and deep learning according to claim 4, characterized in that, The step of using the rock type distribution data and the structural plane orientation data as constraints to perform regional division and feature fusion on the modified point cloud data to obtain the target point cloud data includes: Based on the boundaries of different lithological regions in the rock type distribution data, the corrected point cloud data is segmented to obtain sub-region point cloud sets. For each sub-region point cloud in the sub-region point cloud set, calculate the typical surface reflection characteristic value of the corresponding rock type, and normalize and correct the reflection intensity characteristic value to obtain the first sub-region point cloud set. Based on the spatial location and attitude elements of the structural surfaces in the structural surface attitude data, structural surface feature points are identified and marked in the corresponding sub-region point cloud to obtain the second sub-region point cloud set. The target point cloud data is obtained by fusing the first sub-region point cloud set with the second sub-region point cloud set.

6. The method for calculating the half-pore ratio of a slope based on binocular vision and deep learning according to claim 1, characterized in that, The step of segmenting the borehole region distribution using the Euclidean clustering algorithm to obtain a borehole point cloud includes: Based on the point cloud data of the borehole area distribution, a KD-Tree spatial index is constructed. Select a seed point from the point cloud data of the borehole area distribution, and search for neighboring points within a preset neighborhood search radius threshold using the KD-Tree spatial index; If the number of neighborhood points is greater than or equal to a preset threshold for the minimum number of point clouds in a single class, then the neighborhood points are marked as a cluster. If the number of neighborhood points is less than the threshold for the minimum number of point clouds in a single class, a repeated search mechanism is executed to obtain extended neighborhood points, and extended neighborhood points that are greater than or equal to the threshold for the minimum number of point clouds in a single class are taken as clusters. Calculate the distance between the clusters, and merge adjacent clusters whose distance is less than a preset second threshold to obtain the borehole point cloud.

7. The method for calculating the half-pore ratio of a slope based on binocular vision and deep learning according to claim 6, characterized in that, The step of selecting a seed point from the point cloud data of the borehole area distribution, and searching for neighboring points within a preset neighborhood search radius threshold using the KD-Tree spatial index, includes: The density of each point in the point cloud data of the borehole area is calculated to obtain the local point cloud density; The point with the highest local point cloud density is selected as the current seed point; Based on the KD-Tree spatial index, a spherical region search is performed within the neighborhood search radius threshold, with the current seed point as the center. Obtain the set of all points within the spherical region as the initial neighborhood point set; The spatial distribution consistency of the preliminary neighborhood point set is checked, and outliers that deviate from the main distribution area are removed to obtain the neighborhood points.

8. A slope partial porosity calculation system based on binocular vision and deep learning, characterized in that, include: The data acquisition module is used to acquire left and right image data of the slope surface after blasting through a binocular camera, and to collect environmental data through sensors; The data processing module is used to input the left and right image data into the stereo matching algorithm to obtain a disparity map; based on the disparity map and the parameter configuration of the binocular camera, point cloud data is calculated using the triangulation principle. The data fusion module is used to fuse the point cloud data with the geological data pre-stored on the blasted slope to obtain the target point cloud data; The feature engineering module is used to extract features from the target point cloud data to obtain multi-feature point cloud data. The intelligent recognition module is used to input the multi-feature point cloud data into a pre-trained borehole recognition model to obtain the borehole region distribution; and to segment the borehole region distribution using a Euclidean clustering algorithm to obtain the borehole point cloud. The porosity calculation module is used to calculate the half porosity based on the geometric features of the borehole point cloud and the preset blasting design parameters, and obtain the half porosity. The report output module is used to generate and output half-porosity evaluation results based on the half-porosity for post-explosion quality assessment and parameter feedback optimization.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 7.

Images