Method and system for identifying lithology of a tunnel face considering slurry disturbance

Abstract

The present application belongs to the technical field of the recognition of the rock properties of the working face, and specifically discloses a method and system for recognizing the rock properties of the working face in consideration of the interference of slurry, which comprises the following steps: obtaining a first slurry area recognition result based on thermal infrared spectral data and a three-dimensional geometric model of the working face; using a graph convolutional neural network to perform denoising processing on the multi-band spectral data of the first slurry area; comparing the spectral curve of each pixel in the denoised multi-band spectral data with a standard slurry curve to preliminarily recognize the slurry-containing area; performing a mixed tuning matching filter algorithm on the slurry-containing area, calculating the matching score of each pixel with the slurry endmember spectrum, and the slurry abundance in a single pixel to obtain a second slurry area recognition result; and dividing the working face into a primary rock area and a slurry-shielded area according to the recognition result and respectively performing mineral abundance inversion. The present application solves the problem that the geological truth is physically shielded by artificial materials after the grouting construction.

Description

Technical Field

[0001] This invention relates to the field of lithology identification technology at the working face, and more particularly to a method and system for lithology identification at the working face that takes into account slurry interference. Background Technology

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

[0003] In geological exploration and tunnel construction, accurately identifying the types and contents of primary minerals is a core step in assessing the stability of surrounding rock and preventing geological disasters.

[0004] Traditional geological logging often uses manual spot measurements or laser-induced breakdown spectroscopy (LIBS) to measure mineral composition, but these methods have serious limitations: at the grouting reinforced face, primary minerals are often covered or mixed with grout, and traditional contact spot measurement methods are difficult to remove the interference of grout signals, resulting in the acquisition of mineral composition data containing a lot of grout noise, which cannot truly reflect the mineral types and abundance of the primary rock mass.

[0005] With the development of UAV remote sensing and hyperspectral imaging technologies, remote sensing of mineral composition has become possible. However, existing UAV spectral systems are mostly concentrated in the visible and near-infrared bands, with insufficient acquisition of spectral features in other bands (such as the mid-infrared region from 3μm to 15μm). In addition, interference from lighting, dust, and humidity in actual tunnel environments is severe, limiting the accuracy of spectral recognition; furthermore, traditional spectral processing algorithms cannot effectively separate the spectral pixels of slurry and primary minerals when dealing with large areas of slurry covering the primary rock surface. Summary of the Invention

[0006] To address the aforementioned issues, this invention proposes a method and system for identifying lithology at the working face that considers slurry interference. By utilizing a full-band spectral sensor mounted on an unmanned aerial vehicle (UAV), the system captures the characteristic radiation generated by the vibration of mineral molecules in the mid-infrared band, providing unique spectral information reflecting the chemical composition of the material. Combined with machine learning noise reduction algorithms and hybrid tuned matched filtering (MTMF) technology, the system identifies and eliminates the spectral characteristics of the slurry and spatial image interference, restoring the true abundance and distribution of the original minerals after they have been covered by the slurry.

[0007] In some implementations, the following technical solutions are adopted: A method for lithological identification of the working face considering slurry interference includes: Obtain the spatial geometric point cloud of the working face and establish a three-dimensional geometric model of the working face; Acquire multi-band spectral data of the working face, including near-infrared, mid-infrared, and thermal infrared spectral data; Based on thermal infrared spectral data and a three-dimensional geometric model of the working face, the identification results of the first slurry region were obtained; A graph convolutional neural network is used to denoise the multi-band spectral data of the first slurry region; the spectral curves of each pixel in the denoised multi-band spectral data are compared with the standard slurry curve to preliminarily identify the slurry-containing region; a hybrid tuned matched filtering algorithm is executed on the slurry-containing region, and the matching score between each pixel and the slurry end-member spectrum, as well as the slurry abundance in a single pixel, are calculated to obtain the identification result of the second slurry region; Based on the results of the second slurry region identification, the working face is divided into the primary rock area and the slurry-shrouded area, and mineral abundance inversion is performed separately; mineral and slurry spatial mapping is generated, and geological stability is quantitatively assessed.

[0008] As a further implementation method, based on thermal infrared spectral data and a three-dimensional geometric model of the tunnel face, the identification result of the first slurry region is obtained, specifically as follows: Calculate the rate of curvature change of the local area unit mesh in the three-dimensional geometric model of the tunnel face. When the rate of curvature change is lower than the predetermined range of the average rate of curvature change of the original rock surface, and the image visual texture shows a highly consistent featureless block, the area is defined as a slurry smooth area in terms of geometric morphology. Moisture-rich areas can be delineated by the specific absorption intensities in the short-wave infrared and mid-infrared bands of the full-band spectrum. The spatial coordinates of the identified slurry smooth area and the water-rich area are aligned. The area that simultaneously meets the criteria of slurry smooth area and water-rich area is determined to be a slurry area. The pixels in these areas are marked as 0, and the pixels in the remaining areas are marked as 1.

[0009] As a further implementation, a graph convolutional neural network is used to denoise the multi-band spectral data of the first slurry region, specifically as follows: The gridded pixels of the first slurry region are defined as graph nodes, and edge connections are established using the spatial adjacency relationship and spectral similarity between pixels to transform discrete sampling points into a topological graph structure. The topological graph structure is used as the input of a graph convolutional neural network. Using the feature aggregation operator of the graph convolutional neural network, in each convolution iteration, the target pixel automatically absorbs the spectral information of related pixels in its spatial neighborhood. Through Laplacian smoothing and weighted filtering, isolated dust scattering points and random electromagnetic noise are identified and canceled out. Combined with spectral similarity, the edge sharpness of the rock and slurry is preserved. Specifically, edge connections are established using spatial adjacency and spectral similarity between pixels. When the physical distance between two pixels is less than a set threshold, a spatial adjacency edge is established between the two pixels. The angle or correlation score between the spectral vectors of the two pixels is calculated, and a spectral similarity edge between the two pixels is determined based on the angle or score.

[0010] As a further implementation method, the spectral curves of each pixel in the denoised multi-band spectral data are compared with the standard slurry curve to preliminarily identify the slurry-containing region, specifically: Compare the slope, intercept, and peak position of the spectral curve of each pixel in the denoised multi-band spectral data with the standard slurry curve. If the following conditions are met simultaneously: The relative deviation of the slope of the spectral curve of each pixel from that of the standard slurry curve is lower than the set threshold. The difference between the intercept of each pixel and the standard slurry reference value is controlled within a predetermined range; The deviation between the center position of the absorption peak of each pixel in the moisture-sensitive absorption band and the center wavelength of the absorption peak of the standard slurry is less than the single-band resolution of the spectrometer. It was initially identified as a slurry-bearing area.

[0011] As a further implementation, the matching score between each pixel and the slurry endmember spectrum is calculated, specifically as follows: ; in, x Represents the spectral vector of the pixel to be measured. s Represents the spectral vector of the standard slurry endmember. μ This represents the average value vector of the background spectrum across the entire field.

[0012] As a further implementation, the first slurry region identification result is obtained by calculating the matching score between each pixel and the slurry endmember spectrum, as well as the slurry abundance within a single pixel, specifically: If the matching score between the pixel and the slurry endmember spectrum is greater than a set first threshold, and the slurry abundance value within the pixel is greater than a set second threshold, then the region corresponding to the pixel is identified as a high-confidence slurry region. Simultaneously, contour lines showing the most dramatic changes in slurry abundance gradients are extracted, and combined with matching scores, the physical boundaries of slurry regions are determined.

[0013] As a further implementation, the method further includes: for all pixels with matching scores greater than a set first threshold, performing a second verification on the pixels using infeasibility scores, identifying pixels with infeasibility scores greater than a set value as false feature signals and removing them.

[0014] As a further implementation method, based on the results of the second slurry region identification, the tunnel face is divided into a primary rock area and a slurry-shrouded area, specifically: The overlapping area between the first and second slurry areas is marked as the slurry-covered area, and the remaining area is marked as the primary rock area.

[0015] As a further implementation method, mineral abundance inversion is performed separately for the primary rock zone and the slurry-shrouded zone, specifically as follows: For primary rock regions, mineral abundance is calculated directly; For the slurry-shrouded area, the slurry abundance is calculated as a weight, and the weight is multiplied by the standard slurry spectrum. The mineral abundance is calculated by subtracting the product of the weight and the standard slurry spectrum from the measured spectral vector of the slurry-shrouded area, and then applying coefficient compensation to the result.

[0016] In other embodiments, the following technical solutions are adopted: A lithological identification system for the working face that takes into account slurry interference includes: The model building module is used to acquire the spatial geometric point cloud of the working face and build a three-dimensional geometric model of the working face; The data acquisition module is used to acquire multi-band spectral data of the tunnel face, including near-infrared, mid-infrared and thermal infrared spectral data. The image dimension slurry recognition module is used to obtain the recognition result of the first slurry region based on thermal infrared spectral data and the three-dimensional geometric model of the working face; The spectral dimension slurry identification module is used to denoise the multi-band spectral data of the first slurry region using a graph convolutional neural network; compare the spectral curve of each pixel in the denoised multi-band spectral data with the standard slurry curve to preliminarily identify the slurry-containing region; and perform a hybrid tuned matched filtering algorithm on the slurry-containing region to obtain the identification result of the second slurry region by calculating the matching score between each pixel and the slurry end-member spectrum, as well as the slurry abundance in a single pixel. The mineral inversion module is used to combine the results of the second slurry region identification to divide the working face into primary rock area and slurry-shrouded area, and perform mineral abundance inversion separately; generate mineral and slurry spatial mapping, and conduct a quantitative assessment of geological stability.

[0017] Compared with the prior art, the beneficial effects of the present invention are: (1) Some flat rock surfaces or water-soaked areas on the working face may be highly similar to slurry under images and thermal infrared, which may lead to misidentification. This invention integrates near-infrared, mid-infrared and thermal infrared multispectral data to identify slurry areas from both image and spectral dimensions. First, the three-dimensional geometric model (surface smoothness) and thermal infrared data (moisture enrichment characteristics) are used to identify suspected areas where slurry may exist. Then, the unique identification capability of the chemical composition of the substance by the mid-infrared spectral band is used to confirm whether the suspected area really contains slurry components. This solves the problem of low accuracy in identifying minerals and slurry under complex environments and improves identification efficiency.

[0018] (2) This invention utilizes the high sensitivity of the mid-infrared band (3μm to 15μm) to mineral molecular vibrations and slurry chemical components in the spectral dimension to extract the unique absorption / emission characteristic peaks of the slurry. The slurry abundance within a pixel is calculated by a hybrid tuned matched filter algorithm, and the spectral interference of the slurry is effectively removed by vector subtraction. Compared with traditional detection techniques, it can penetrate the grouting cover layer and extract the original mineral characteristics of the covered area, completely solving the problem that the geological truth is "physically shielded" by artificial materials after grouting construction.

[0019] (3) This invention effectively eliminates nonlinear interference caused by tunnel dust and water vapor through machine learning noise reduction preprocessing; and significantly improves the accuracy and reliability of digital geological mapping of the tunnel face by utilizing the "infeasibility" evaluation mechanism of the hybrid tuned matched filter algorithm.

[0020] Other features and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0021] Figure 1 This is a flowchart of the lithology identification method for the working face considering slurry interference in an embodiment of the present invention; Figure 2 This is a schematic diagram of the UAV scanning the tunnel face in an embodiment of the present invention. Detailed Implementation

[0022] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0023] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0024] Example 1 In one or more embodiments, a method for identifying the lithology of a working face that takes into account slurry interference is disclosed, combined with... Figure 1 Specifically, it includes the following processes: S101: Obtain the spatial geometric point cloud of the working face and establish a three-dimensional geometric model of the working face.

[0025] In this embodiment, three-dimensional laser scanning (LiDAR) technology is used to acquire the spatial geometric point cloud of the working face and a high-precision three-dimensional geometric model of the working face is established in 3DMine software.

[0026] S102: Acquire multi-band spectral data of the working face, including near-infrared, mid-infrared and thermal infrared spectral data.

[0027] Combination Figure 2 The tunnel face or geological areas to be tested are scanned by using drones equipped with full-band spectral imaging equipment.

[0028] The full-band spectral imaging equipment includes near-infrared (0.75μm to 2.5μm) detectors, mid-infrared (3μm to 15μm) detectors, and thermal infrared (8μm to 14μm) detectors.

[0029] Mid-infrared detectors are used to detect and record specific spectral absorption and emission characteristics of object surfaces. Substances with different chemical compositions exhibit radiation characteristics in the mid-infrared band that correspond to their molecular structures, and these characteristics are directly related to the chemical composition of the substance. Mid-infrared imaging utilizes hyperspectral sensors to capture spectral signals in these bands, converting them into images of mineral composition distribution and displaying them in the form of spectral maps. This can be used to detect differences in the types of minerals in underground rocks and the chemical composition of grouting materials.

[0030] Thermal infrared detectors are used to detect and record the infrared radiation emitted by objects. All objects radiate infrared radiation at a certain temperature, and this radiation is proportional to the object's temperature. Thermal infrared imaging uses an infrared camera to capture this infrared radiation, converts it into a thermal image, and displays it in the form of a thermal image. It can be used to detect temperature differences in underground structures.

[0031] Near-infrared detectors primarily capture the characteristic absorption peaks of water molecules and hydroxyl groups (-OH) in the 0.7 μm and 2.5 μm wavelength bands. Near-infrared images typically have higher spatial resolution than mid-infrared images and are less affected by environmental dust. By capturing these "moisture fingerprints," near-infrared detectors can identify the initial water-bearing physical boundaries of slurries and typical alteration minerals (such as sericite and kaolinite).

[0032] Meanwhile, by performing a preliminary scan of the tunnel face, standard diagnostic spectral characteristics and thermal infrared radiation information of the original rock and various slurries are obtained, a mapping relationship between material composition and spectral response is established, and the data is stored in the tunnel area spectral library to provide end-member benchmarks for subsequent stripping algorithms.

[0033] S103: Based on thermal infrared spectral data and a three-dimensional geometric model of the working face, the identification result of the first slurry region is obtained.

[0034] In this embodiment, in the image dimension, the geometric morphological differences (such as the smooth area after slurry filling) and the water-rich areas on the face of the tunnel are identified by combining the three-dimensional spatial model and the thermal infrared radiation gradient.

[0035] Specifically, geometrical differences are primarily identified through a combination of microscopic roughness analysis of 3D laser point clouds and image texture feature extraction. Smooth areas refer to regions exhibiting extremely low "normal vector change rates" relative to the original rock surface. Original rock, affected by excavation and blasting, typically has surfaces covered with fine bumps, fissures, and sharp edges, manifesting as high-frequency geometric noise and severe normal vector fluctuations in point cloud data. Slurry, with its fluid properties, fills these tiny pores and fissures on the rock surface after spraying or applying, due to gravity and surface tension, forming a relatively continuous and rounded cover. The system calculates the curvature distribution within the local grid. When the rate of curvature change is between 15% and 30% of the average rate of curvature change of the original rock surface, and the image visual texture exhibits highly consistent "feature-free blocks," this region is defined as a geometrically smooth slurry area.

[0036] The moisture-rich region was calculated using specific absorption intensities in the short-wave infrared and mid-infrared bands of the full-spectrum spectrum. The high moisture content was defined by typical water molecule resonance absorption peaks near 1.45 μm, 1.94 μm, and 2.9 μm in the spectral curve. Because the grout contains significantly more moisture than the dry primary rock during mixing and construction, its reflectance in these bands decreases dramatically, forming deep and wide absorption valleys.

[0037] The system aligns the identified "geometrically smooth areas" with the "moisture-rich areas" using spatial coordinates. Only when an area meets the physical morphology of curvature smoothness and the chemical property of high moisture absorption will it be officially identified as a slurry-containing area. The pixels in these areas are marked as 0, and the pixels in the remaining areas are marked as 1; thus, the first slurry area identification result is obtained.

[0038] This embodiment first uses physical morphological features to preliminarily define which pixels belong to the high-susceptibility slurry-containing region from the image dimension, and generates a preliminary binary spatial mask.

[0039] However, some flat rock surfaces or water-soaked areas on the tunnel face may appear highly similar to slurry under images and thermal infrared light, potentially leading to misjudgment of slurry areas and loss of underlying geological information. Therefore, this embodiment, guided by the spatial characteristics of the macroscopic recognition results in the image dimension, further verifies whether the area marked as 0 (i.e., the first identified slurry area) truly contains slurry components from a spectral dimension. The specific processing procedure for the spectral dimension is detailed in step S104.

[0040] S104: Denoise the multi-band spectral data of the first slurry region using a graph convolutional neural network; compare the spectral curves of each pixel in the denoised multi-band spectral data with the standard slurry curve to preliminarily identify the slurry-containing region; perform a hybrid tuned matched filtering algorithm on the slurry-containing region, and obtain the identification result of the second slurry region by calculating the matching score between each pixel and the slurry end-member spectrum, as well as the slurry abundance in a single pixel.

[0041] In this embodiment, a graph convolutional neural network (GCN) is used to perform deep preprocessing on the full-band raw data of the first slurry region to obtain a "clean" dataset with high spatial consistency and high spectral fidelity.

[0042] Specifically, the gridded pixels obtained from the face scanning are defined as graph nodes, and edge connections are established using the spatial adjacency relationship and spectral similarity between pixels, thereby transforming discrete sampling points into a topological graph structure.

[0043] The determination of spatial adjacency relationships involves using the 3D laser point cloud coordinates acquired by the geological information acquisition system. The system treats each gridded pixel on the tunnel face as a coordinate point in 3D space. By setting a physical distance threshold, it performs a nearest neighbor search on any two pixel nodes. This threshold is set to 1.2 to 1.5 times the average point spacing of the point cloud to ensure that each pixel node can establish physical connections with at least 4 to 8 of its nearest neighbors, thereby constructing a continuous topological grid covering the entire tunnel face and avoiding isolated points in the graph structure due to an excessively small threshold. When the actual physical distance between two pixels is less than this threshold, the system determines that they are spatially adjacent and establishes a spatial adjacency edge between the two pixels.

[0044] Determining spectral similarity involves quantifying node relationships at the attribute level. The system utilizes full-band spectral data to extract the intrinsic spectral vector of each pixel node in a specific band (such as the mid-infrared molecular vibration region). The consistency of material composition is measured by calculating the angle or correlation score (e.g., spectral angle mapping, SAM) between the spectral vectors of two nodes; a smaller angle or a correlation score closer to 1 indicates greater spectral similarity. The establishment of a spectral similarity edge between two pixels is determined based on the angle or score.

[0045] Based on the constructed topology graph, the feature aggregation operator of the graph convolutional network (GCN) is used. In each convolutional iteration, the target pixel automatically absorbs the spectral information of related pixels in its spatial neighborhood. Through Laplacian smoothing and weight filtering, isolated dust scattering points and random electromagnetic noise are identified and canceled out.

[0046] Laplacian smoothing is a core internal operator of Graph Convolutional Networks (GCNs). During propagation at each layer of a GCN, nodes aggregate features from their neighbors using the Laplacian matrix. This process essentially performs low-pass filtering in the graph domain, making the attributes of adjacent nodes more consistent and thus eliminating spectral noise. Weight filtering is achieved through edge weight operators in the graph connections, located in the weighted summation stage of the convolutional kernel. The system dynamically adjusts weights based on the spatial distance and spectral similarity between nodes. Neighbors with significantly different attributes from the central node (such as noise points at the edge of a slurry area) are assigned extremely low weights, thus being naturally "filtered" during feature propagation. This design ensures that the model can extract group features during learning without being disturbed by individual anomalous pixels.

[0047] At the boundary between rock and slurry, the starkly different spectral characteristics of the two materials cause the established node associations to weaken, preventing the smooth diffusion of slurry features into the rock region. This results in Laplace smoothing occurring only within the same type of material. Consequently, in the final generated recognition map, the boundary between rock and slurry does not become smoothed by the algorithm but accurately preserves the complex intrusion contact relationships, thus achieving high-fidelity edge sharpness restoration.

[0048] Ultimately, while preserving the sharpness of the rock and slurry edges, a non-linear improvement in the signal-to-noise ratio of the entire band data was achieved.

[0049] In this embodiment, the precise extraction and localization of slurry components are mainly achieved through the following process during spectral dimension identification: S1041: Compare the spectral curves of each pixel in the denoised multi-band spectral data with the standard slurry curve to preliminarily identify the slurry-bearing area.

[0050] Specifically, the standard spectral library pre-stored in the tunnel area spectral database is used to extract the characteristics of the target slurry in the mid-infrared band (3μm-15μm), identifying the unique absorption peaks and emissivity characteristics of the slurry. By comparing the slope, intercept, and peak position of the spectral curves of each pixel in the original data with the standard slurry curve, the slurry-bearing area is preliminarily located. The specific comparison process is as follows: Compare the slope, intercept, and peak position of the spectral curve of each pixel in the denoised multi-band spectral data with the standard slurry curve. If the following conditions are met simultaneously: (1) The relative deviation of the slope of the spectral curve of each pixel from that of the standard slurry curve is lower than the set threshold (e.g., 15%); the high synchronization of the slope indicates that the pixel and the slurry have the same spectral evolution rate.

[0051] (2) The difference between the intercept of each pixel and the standard slurry reference value is controlled within a predetermined range; In spectral linear regression analysis, the difference between the intercept term of the measured pixel and the standard slurry reference value must be controlled within ±5% of the maximum spectral reflectance value. This indicator is used to exclude overall brightness shift caused by ambient light or sensor gain, ensuring that the signal is correlated at the quantization level.

[0052] (3) The deviation between the center position of the absorption peak of each pixel in the moisture-sensitive absorption band and the center wavelength of the absorption peak of the standard slurry is less than the single-band resolution of the spectrometer (usually the error is required to be within ±10nm).

[0053] It was initially identified as a slurry-bearing area.

[0054] S1042: The composition inversion system based on hybrid tuned matched filtering (MTMF) performs the MTMF algorithm on full-band data. This algorithm combines the advantages of hybrid pixel decomposition and matched filtering. During the calculation, the system projects the denoised spectral data into a high-dimensional space and calculates the matching score (MF Score) between each pixel and the slurry endmember spectrum. The closer the matching score is to 1, the higher the probability that the pixel contains slurry.

[0055] The matching score represents the degree of correlation between the pixel spectrum and the standard endmember of the slurry, and the value is usually distributed between 0 and 1; the matching score is calculated as follows: ; Where x is the spectral vector of the pixel to be measured (containing reflectance data of all bands), s is the spectral vector of the standard slurry endmember, μ is the average vector of the background spectrum across the entire field, and ∑ -1 The inverse of the covariance matrix representing the background statistical characteristics.

[0056] The system uses linear or nonlinear unmixing models to calculate the physical proportion (i.e. abundance) of slurry within a single pixel, thereby identifying the slurry content even in the "transition zone" where rock and slurry are mixed.

[0057] The specific formula for calculating the serous abundance within a single pixel is as follows: ; in, For the mixed pixel reflectance vector, For the endmember spectral matrix, This represents the abundance value of the slurry. This is the fitting residual / error term.

[0058] If the matching score between a pixel and the slurry end-member spectrum is greater than a set first threshold (e.g., 0.85), and the slurry abundance value within the pixel is greater than a set second threshold (e.g., 0.60), then the region corresponding to the pixel is identified as a high-confidence slurry region. If the matching score between a pixel and the slurry end-member spectrum is lower than a set threshold (e.g., 0.30), then it is directly identified as a "primitive rock region".

[0059] Simultaneously, contour lines showing the most dramatic changes in slurry abundance gradients are extracted, and combined with matching scores, the physical boundaries of the slurry region are determined. Since slurry abundance gradually decreases as it spreads from the center to the edge, the system captures abrupt changes in abundance values ​​from high to zero, and defines these abrupt changes, combined with matching scores, as the physical boundaries of the slurry. This eliminates spurious signals generated by background noise and, through sub-pixel-level abundance calculations, reconstructs the complex intrusion morphology at the interface between the slurry and the primary rock.

[0060] S1043: Use the "Infeasibility Score" to perform false positive removal.

[0061] As an optional implementation, the aforementioned Hybrid Tuned Matched Filtering (MTMF) algorithm also outputs an infeasibility score. The system performs a secondary verification on pixels with high matching scores (e.g., greater than 0.85) but also high infeasibility scores (e.g., greater than 25.0) by setting a dynamic threshold. Since the infeasibility score is typically measured in noise standard deviation (DN value), a value exceeding 25.0 indicates that the spectrum is at an extremely marginal outlier position in the mixing space, making it physically highly unlikely to be composed of grout. This anomaly is usually caused by environmental specular reflection or mineral alteration interference, which the system classifies as a false signal. The system automatically eliminates such false alarms, ensuring that the identification results only point to the actual grouting area.

[0062] This embodiment, based on the suspected slurry region results obtained from image-level identification, utilizes the unique chemical component identification capability of the mid-infrared band to confirm whether the suspected region truly contains slurry components. Simultaneously, it verifies the morphological boundaries identified from the image with the component abundance identified from the spectrum to determine the final slurry region boundary; this solves the problem of low accuracy in mineral and slurry identification under complex environments, while also improving identification efficiency.

[0063] S105: Based on the results of the second slurry area identification, the working face is divided into the primary rock area and the slurry-shrouded area, and mineral abundance inversion is performed separately; mineral and slurry spatial mapping is generated, and geological stability is quantitatively assessed.

[0064] In this embodiment, the overlapping area between the first slurry area and the second slurry area is marked as a slurry-masked area, and its spatial mask is updated to 0; the remaining area is marked as a primary rock area, and its spatial mask is updated to 1.

[0065] Mineral abundance inversion was performed on the slurry-shrouded area and the primary rock area, respectively. The specific process is as follows: (1) For primary rock areas, mineral abundance inversion is performed directly.

[0066] Mineral abundance refers to the area proportion of a specific mineral component within a single pixel. The system employs fully constrained least squares to transform the net rock region spectral vector with a spatial mask of 1. Decomposed into a linear combination of m standard alteration mineral endmembers: ; in, Let be the standard endmember spectral vector of the j-th mineral. To fit the residuals, the system requires the root mean square error (RMSE) to be less than 0.02 to ensure accuracy; Let the abundance of the mineral to be solved be in the range [0,1], representing the proportion of the j-th mineral in this rock, and satisfying: Sum That is, the sum of the abundances of all minerals is 1.

[0067] Mineral abundance is subject to a mandatory "non-negativity constraint" during the solution process. ≥0) and "unified constraint" ( = 1.0), ensuring the calculation results conform to physical laws. By continuously adjusting the weight values, the similarity between the reconstructed spectrum and the measured spectrum in the characteristic absorption band is maximized. When the root mean square error of the calculation residual RMSE < 0.02 and the iteration deviation < 10, the calculation is considered successful. -6 At that time, the current The value is locked as the abundance weight of that mineral.

[0068] (2) For the slurry-covered area, the abundance weights generated by the Mixed Tuned Matched Filter (MTMF) are called, and the spectral vector subtraction operation is performed to deduct the spectral features of the standard slurry endmembers from the mixed pixels proportionally, thereby eliminating the superposition and masking of the spectral feature peaks of the original rock by the slurry.

[0069] The specific formula for spectral vector subtraction is as follows:

[0070] in, The slurry abundance calculated by the MTMF algorithm. For standard slurry spectra, This represents the measured spectral vector of the mixed pixels; The spectral vector after eliminating the slurry spectrum.

[0071] As a signal compensation coefficient, its function is to restore the rock reflection signal weakened by slurry to 100% of the exposure level, thereby compensating for energy loss.

[0072] Based on the spectral vector after eliminating the slurry spectrum Mineral abundance inversion is performed to output a high-fidelity primary mineral distribution map and rock alteration evaluation parameters after "penetrating reduction". This output completely eliminates the interference of slurry and accurately restores the mineral types, contents and spatial topological relationships of the obscured areas.

[0073] As a further implementation method, in areas with a mask of 0, the system combines the occurrence constraint logic of the geological knowledge graph and performs intelligent topological reasoning based on the lithological orientation of the surrounding unmasked areas to reconstruct a continuous mineral vein distribution layer.

[0074] The geological knowledge graph is a system that automatically extracts geological entities and their relationships from historical exploration reports, borehole columnar sections, core logging, and regional geological records of mining areas using natural language processing technology. Based on geological ontology, a semantic network is constructed to associate "specific alteration minerals" with "geological environments." For example, the graph pre-determines that if a chloritization area with an abundance weight W > 0.45 is present, then there is a high probability of hydrothermal vein-like infill around it. Finally, dynamic calibration is performed using real-time acquired spectral features of the working face. When the matching probability between the measured data and a certain type of geological evolution model in the graph exceeds 92%, the system automatically activates the reasoning logic in that context, providing prior criteria for missing areas.

[0075] Occurrence constraint logic refers to using the strike, dip, and dip angle of rock strata or veins in three-dimensional space as geometric boundary conditions for topological extrapolation. In actual operation, when the grouting mask (logic value "0") largely obscures the rock surface, the system retrieves the vein boundary vector calculated from the surrounding unobscured areas (logic value "1") as the starting boundary. The system sets a spatial deflection threshold of ±5°. If the average strike of the veins in the exposed areas on both sides is 135°, then the mineral abundance filling in the middle mask area must be linearly or nonlinearly reconstructed along this direction line. This logic ensures that in the three-dimensional mapping system, the distribution layers of veins blocked by grout are not randomly generated, but strictly conform to the physical continuity of real geological structures.

[0076] As a further implementation method, a visualization terminal converts spectral data into a 3D geological map. Combined with a 3D laser point cloud model, the original mineral attributes are transformed using spatial coordinate operators to ensure that each geological attribute point (such as quartz content or alteration degree) can find its corresponding geographic coordinates in the 3D grid of the tunnel face. The complex mapping process is broken down into two logical layers: one is a real-time visible slurry diffusion distribution layer, and the other is a primary lithology base layer after removing interference. This provides a spatially continuous digital criterion for assessing the stability of the surrounding rock and providing disaster early warning at the construction site.

[0077] Finally, by calculating the matching score between primary minerals and slurry within each grid cell, the mineral abundance can be quantitatively filled. Combined with the reduced dike orientation and alteration characteristics, the surrounding rock integrity parameters are automatically calculated, enabling in-depth analysis and prediction of geological hazard risks.

[0078] This embodiment utilizes a drone equipped with a full-band spectral sensor as the data acquisition vehicle, changing the dangerous traditional method of manual close-to-the-face exploration. It significantly improves data acquisition efficiency and spatial accuracy, and ensures that exploration tasks can be completed safely, efficiently, and without blind spots even in complex conditions such as unstable surrounding rock. Through "mask and vector subtraction" technology, the spectral interference of the grout is effectively eliminated. Compared to traditional exploration techniques, it can penetrate the grouting overlay layer and extract the original mineral characteristics of the concealed areas, completely solving the problem of geological truth being "physically shielded" by artificial materials after grouting construction.

[0079] Example 2 In one or more embodiments, a lithological identification system for a working face that takes into account slurry interference is disclosed, comprising: The model building module is used to acquire the spatial geometric point cloud of the working face and build a three-dimensional geometric model of the working face; The data acquisition module is used to acquire multi-band spectral data of the tunnel face, including near-infrared, mid-infrared and thermal infrared spectral data. The image dimension slurry recognition module is used to obtain the recognition result of the first slurry region based on thermal infrared spectral data and the three-dimensional geometric model of the working face; The spectral dimension slurry identification module is used to denoise the multi-band spectral data of the first slurry region using a graph convolutional neural network; compare the spectral curve of each pixel in the denoised multi-band spectral data with the standard slurry curve to preliminarily identify the slurry-containing region; and perform a hybrid tuned matched filtering algorithm on the slurry-containing region to obtain the identification result of the second slurry region by calculating the matching score between each pixel and the slurry end-member spectrum, as well as the slurry abundance in a single pixel. The mineral inversion module is used to combine the results of the second slurry region identification to divide the working face into primary rock area and slurry-shrouded area, and perform mineral abundance inversion separately; generate mineral and slurry spatial mapping, and conduct a quantitative assessment of geological stability.

[0080] It should be noted that the specific implementation methods of the above modules are exactly the same as those in Example 1, and will not be described in detail again.

[0081] Example 3 In one or more embodiments, a terminal device is disclosed, comprising a processor and a memory, the processor for implementing instructions; the memory for storing multiple instructions adapted to be loaded by the processor and executed by the processor for the lithological identification method of the working face considering slurry interference as described in Embodiment 1.

[0082] It should be understood that in this embodiment, the processor can be a central processing unit (CPU), or it can 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 can be a microprocessor or any conventional processor.

[0083] Memory may include read-only memory and random access memory, and provides instructions and data to the processor. A portion of memory may also include non-volatile random access memory. For example, memory may also store information about the device type.

[0084] In the implementation process, each step of the above method can be completed by the integrated logic circuits in the processor hardware or by software instructions.

[0085] Example 4 In one or more embodiments, a computer-readable storage medium is disclosed, wherein a plurality of instructions are stored, the instructions being adapted to be loaded by a processor of a terminal device and executed by the method for identifying the lithology of the working face considering slurry interference as described in Embodiment 1.

[0086] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A method for identifying the lithology of a working face considering slurry interference, characterized in that, include: Acquire the spatial geometric point cloud of the working face and establish a three-dimensional geometric model of the working face; Acquire multi-band spectral data of the working face, including near-infrared, mid-infrared, and thermal infrared spectral data; Based on thermal infrared spectral data and a three-dimensional geometric model of the working face, the identification results of the first slurry region were obtained; A graph convolutional neural network is used to denoise the multi-band spectral data of the first slurry region; the spectral curves of each pixel in the denoised multi-band spectral data are compared with the standard slurry curve to preliminarily identify the slurry-containing region; a hybrid tuned matched filtering algorithm is executed on the slurry-containing region, and the matching score between each pixel and the slurry end-member spectrum, as well as the slurry abundance in a single pixel, are calculated to obtain the identification result of the second slurry region; Based on the results of the second slurry region identification, the working face is divided into the primary rock area and the slurry-shrouded area, and mineral abundance inversion is performed separately; mineral and slurry spatial mapping is generated, and geological stability is quantitatively assessed.

2. The method for lithological identification of a working face considering slurry interference as described in claim 1, characterized in that, Based on thermal infrared spectral data and a three-dimensional geometric model of the tunnel face, the identification results of the first slurry region were obtained, specifically as follows: Calculate the rate of curvature change of the local area unit mesh in the three-dimensional geometric model of the tunnel face. When the rate of curvature change is lower than the predetermined range of the average rate of curvature change of the original rock surface, and the image visual texture shows a highly consistent featureless block, the area is defined as a slurry smooth area in terms of geometric morphology. Moisture-rich areas can be delineated by the specific absorption intensities in the short-wave infrared and mid-infrared bands of the full-band spectrum. The spatial coordinates of the identified slurry smooth area and the water-rich area are aligned. The area that simultaneously meets the criteria of slurry smooth area and water-rich area is determined to be a slurry area. The pixels in these areas are marked as 0, and the pixels in the remaining areas are marked as 1.

3. The method for lithological identification of a working face considering slurry interference as described in claim 1, characterized in that, The multi-band spectral data of the first slurry region is denoised using a graph convolutional neural network, specifically as follows: The gridded pixels of the first slurry region are defined as graph nodes, and edge connections are established using the spatial adjacency relationship and spectral similarity between pixels to transform discrete sampling points into a topological graph structure. The topological graph structure is used as the input of a graph convolutional neural network. Using the feature aggregation operator of the graph convolutional neural network, in each convolution iteration, the target pixel automatically absorbs the spectral information of related pixels in its spatial neighborhood. Through Laplacian smoothing and weighted filtering, isolated dust scattering points and random electromagnetic noise are identified and canceled out. Combined with spectral similarity, the edge sharpness of the rock and slurry is preserved. Specifically, edge connections are established using spatial adjacency and spectral similarity between pixels. When the physical distance between two pixels is less than a set threshold, a spatial adjacency edge is established between the two pixels. The angle or correlation score between the spectral vectors of the two pixels is calculated, and a spectral similarity edge between the two pixels is determined based on the angle or score.

4. The method for lithological identification of a working face considering slurry interference as described in claim 1, characterized in that, The spectral curves of each pixel in the denoised multi-band spectral data were compared with the standard slurry curve to preliminarily identify the slurry-containing areas, specifically: Compare the slope, intercept, and peak position of the spectral curve of each pixel in the denoised multi-band spectral data with the standard slurry curve. If the following conditions are met simultaneously: The relative deviation of the slope of the spectral curve of each pixel from that of the standard slurry curve is lower than the set threshold. The difference between the intercept of each pixel and the standard slurry reference value is controlled within a predetermined range; The deviation between the center position of the absorption peak of each pixel in the moisture-sensitive absorption band and the center wavelength of the absorption peak of the standard slurry is less than the single-band resolution of the spectrometer. It was initially identified as a slurry-bearing area.

5. The method for lithological identification of a working face considering slurry interference as described in claim 1, characterized in that, Calculate the matching score between each pixel and the slurry endmember spectrum, specifically: ； in, x Represents the spectral vector of the pixel to be measured. s Represents the spectral vector of the standard slurry endmember. μ This represents the average value vector of the background spectrum across the entire field.

6. The method for lithological identification of a working face considering slurry interference as described in claim 1, characterized in that, The first slurry region identification result is obtained by calculating the matching score between each pixel and the slurry endmember spectrum, as well as the slurry abundance within a single pixel. Specifically: If the matching score between the pixel and the slurry endmember spectrum is greater than a set first threshold, and the slurry abundance value within the pixel is greater than a set second threshold, then the region corresponding to the pixel is identified as a high-confidence slurry region. Simultaneously, contour lines showing the most dramatic changes in slurry abundance gradients are extracted, and combined with matching scores, the physical boundaries of slurry regions are determined.

7. The method for lithological identification of a working face considering slurry interference as described in claim 6, characterized in that, Also includes: For all pixels with a matching score greater than a set first threshold, the pixels are then verified a second time using the infeasibility score. Pixels with an infeasibility score greater than the set value are identified as false feature signals and removed.

8. The method for lithological identification of a working face considering slurry interference as described in claim 1, characterized in that, Based on the results of the second slurry region identification, the working face is divided into a primary rock area and a slurry-shrouded area, specifically: The overlapping area between the first and second slurry areas is marked as the slurry-covered area, and the remaining area is marked as the primary rock area.

9. The method for lithological identification of a working face considering slurry interference as described in claim 1, characterized in that, Mineral abundance inversion was performed separately for the primary rock zone and the slurry-shrouded zone, as follows: For primary rock regions, mineral abundance is calculated directly; For the slurry-shrouded area, the slurry abundance is calculated as a weight, and the weight is multiplied by the standard slurry spectrum. The mineral abundance is calculated by subtracting the product of the weight and the standard slurry spectrum from the measured spectral vector of the slurry-shrouded area, and then applying coefficient compensation to the result.

10. A lithological identification system for a working face that takes into account slurry interference, characterized in that, include: The model building module is used to acquire the spatial geometric point cloud of the working face and build a three-dimensional geometric model of the working face; The data acquisition module is used to acquire multi-band spectral data of the tunnel face, including near-infrared, mid-infrared and thermal infrared spectral data. The image dimension slurry recognition module is used to obtain the recognition result of the first slurry region based on thermal infrared spectral data and the three-dimensional geometric model of the working face; The spectral dimension slurry identification module is used to denoise the multi-band spectral data of the first slurry region using a graph convolutional neural network; compare the spectral curve of each pixel in the denoised multi-band spectral data with the standard slurry curve to preliminarily identify the slurry-containing region; and perform a hybrid tuned matched filtering algorithm on the slurry-containing region to obtain the identification result of the second slurry region by calculating the matching score between each pixel and the slurry end-member spectrum, as well as the slurry abundance in a single pixel. The mineral inversion module is used to combine the results of the second slurry region identification to divide the working face into primary rock area and slurry-shrouded area, and perform mineral abundance inversion separately; generate mineral and slurry spatial mapping, and conduct a quantitative assessment of geological stability.

Images