Coal rock identification method

Abstract

The invention discloses a coal rock identification method, which is based on an X-ray source, a Compton scattering imaging technology of a photon counting detector, a motion support, a data processing and identification module and the like, through a multi-channel energy spectrum fusion and boundary regression fitting algorithm, in combination with third-order B-spline interpolation and Savitzky-Golay smooth filtering, a confidence correction mechanism is introduced, and the coal rock identification accuracy is improved. A high-precision and anti-noise continuous coal rock interface track is generated, and the recognition accuracy and robustness are improved. According to the invention, multi-energy-spectrum resolution of scattered rays of a measured object can be realized, and a coal seam and a thin-layer dirt band can be accurately distinguished in combination with a multi-channel energy spectrum fusion algorithm.

Description

Technical Field

[0001] This invention belongs to the technical fields of mining, mine design and construction, and mine safety, and relates to a method for identifying coal-rock interfaces and a method for determining coal cutting trajectories. Background Technology

[0002] Coal-rock interface identification is crucial for intelligent coal mining, directly impacting coal quality, resource utilization, and the lifespan of mining machines. Due to complex coal seam geological conditions, mining machines often cut through rock or interbedded rock due to unclear coal-rock interfaces, leading to accelerated wear of cutting teeth, decreased efficiency, and even equipment damage. Existing identification technologies include gamma-ray, infrared, image, radar, vibration, acoustic signals, and multi-sensor fusion AI methods, but all have limitations: gamma-ray signals are weak, infrared is easily affected by motion parameters, image recognition is environmentally constrained, radar suffers severe attenuation in thick coal seams, vibration and acoustic methods are susceptible to noise interference, and multi-sensor fusion methods have high hardware requirements, large processing delays, and information conflicts. Furthermore, most methods are ill-suited to the harsh underground environment of high noise, high dust, and strong electromagnetic interference, and suffer from computational errors, low reliability, and response lag, hindering their widespread application in actual production. Summary of the Invention

[0003] In view of the shortcomings of the existing technology, the purpose of this invention is to provide a coal and rock identification method to solve the deficiencies of the existing technology, improve the mining efficiency and safety of coal mines, be able to work stably in harsh environments, have high accuracy and good cost-effectiveness, and have important practical significance and broad application prospects for promoting intelligent and unmanned coal mining.

[0004] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: A coal and rock identification method is disclosed. This method is based on a coal and rock identification device to identify the coal and rock interface. The coal and rock identification device includes an X-ray source, a photon counting detector, a motion support, and a data processing and identification module. The X-ray source and the photon counting detector are both mounted on the motion support and placed on the same side of the coal and rock to be detected. The data processing and identification module can control the working sequence of the X-ray source and the photon counting detector, and process and analyze the measurement data to identify the coal and rock. The method includes the following steps: Step 1: Move the X-ray source and photon counting detector to the front of the coal and rock area to be inspected using a motion support; Step 2: The X-ray source emits high-frequency pulsed X-rays, i.e., primary rays I0, according to the set parameters to irradiate the area to be detected; Step 3: The photon counting detector receives the Compton backscattered rays, i.e., Compton backscattered photons, emitted from the area to be detected. I bsMulti-energy spectral scattering data is collected, multi-channel energy spectral features are extracted, and the data is transmitted to the data processing and recognition module. Step 4, the data processing and recognition module determines the originating ray I0 and the Compton backscattered photons based on the data. I bs Based on multi-channel energy spectrum characteristics, a coal-rock interface generation algorithm based on multi-channel energy spectrum fusion and boundary regression fitting is used. Combined with third-order B-spline interpolation and Savitzky-Golay smoothing filtering, and confidence correction is introduced, the material of the measured area is analyzed to identify coal and rock. Step 5: Move the X-ray source and photon counting detector to different areas using a motion support, and repeat the above steps for multiple parallel sections; integrate the coal-rock interface trajectories on multiple sections to generate a three-dimensional spatial distribution map of the coal-rock interface, providing more comprehensive information for mining planning and geological analysis.

[0005] The present invention also includes the following technical features: Specifically, the X-ray source is used to emit X-rays, and the energy of the X-rays is adjusted by regulating the operating voltage of the X-ray source; The photon counting detector is used to receive Compton backscattered rays and measure their radiation dose and energy distribution; the photon counting detector is equipped with a collimator module, and by adjusting the direction of the collimator, it can receive and measure rays in a specific direction to achieve the measurement of matter in a specific area and at a specific depth. The motion support is used to support the X-ray source and the photon counting detector, and to enable detection of different areas by moving the support.

[0006] Specifically, the X-ray emitting end face of the X-ray source and the signal acquisition end face of the photon counting detector are both facing the coal and rock detection area. After the X-rays interact with the coal and rock material, Compton backscattering occurs, and the photon counting detector acquires and detects the backscattered X-ray signal. The angle θ between the central axis of the X-ray source and the central axis of the photon counting detector is selected within the range of 150° to 180°.

[0007] Specifically, step 3 includes: Step 3.1: The photon counting detector receives Compton backscattered photons. I bs ; Step 3.2: Distribute the detected Compton backscattered photons into n preset energy channels according to their energy levels; Step 3.3, divide the surface to be inspected into... i List, j The sampling grid of the row; Step 3.4, at each sampling grid point ( xi , z j At point ), the number of photons detected by each energy channel is recorded, forming a multi-channel energy spectrum counting vector: (3).

[0008] Specifically, step 4 includes: Step 4.1, Classifier identifies coal / rock and confidence level: The classifier is used to determine whether a substance is coal or rock. Its input is the multi-channel energy spectrum counting vector f extracted in step 3. ij The output is category labels. l ij and confidence level p ij ; Step 4.2, Boundary transition detection: Based on the classification labels output in step 4.1 l ij The location where the coal and rock classification labels of adjacent scan points at different depths in the same column suddenly change is defined as a jump point; based on the confidence level output in step 4.1... p ij The coordinates of reliable transition points, filtered by confidence level, are defined as boundary points; the detection yields the first... i Column boundary point set B i ; Step 4.3, Boundary point regression fitting and smoothing filtering: Using the set of boundary points of all columns detected in step 4.2 as input nodes, a continuous line representing the coal-rock interface as the horizontal position is generated by third-order B-spline interpolation. x Continuous function of change f(x) ; Then the continuous function f(x) The Savitzky-Golay filter is used for local smoothing to obtain the smoothed coal-rock interface trajectory function. f smoothed ( x ); Step 4.4, Confidence Adjustment: For trajectory function f smoothed ( x For boundary points on the boundary, if their classification confidence is lower than the boundary repair confidence threshold and the confidence of adjacent boundary points is not lower than the boundary repair confidence threshold, then weighted interpolation is used to correct the point value; otherwise, the boundary is retained. f smoothed (x) The trajectory function after confidence-weighted interpolation correction is obtained. fcorrected (x) ; Step 4.5, Output trajectory: The corrected trajectory function z = f corrected (x) As the final coal-rock interface trajectory, this high-precision, smooth, and continuous coal-rock interface trajectory enables the identification of the coal-rock interface and can guide the cutting path of the coal mining machine.

[0009] Specifically, in step 4.1, a classifier is used to determine whether a substance is coal or rock, and the input feature vector is 𝑓. 𝑖𝑗 Use the trained classifier to make predictions: (4) Among them, category labels 0 represents coal seams, and 1 represents rock strata; confidence level This quantifies the reliability of the classification; a larger value indicates a more reliable classification result.

[0010] Specifically, in step 4.2, the transition point is the location where the classification label undergoes a sudden change. x i , z j ( ), satisfying scan points at different depths in the same column l ij ≠ l i(j+1) ; The boundary points satisfy max( p ij , p i(j+1) )> T high , where max( p ij , p i(j+1) ) are two adjacent points ( x i , z j ) and( x i , z j+1 The maximum confidence level of ) T high Filter thresholds for boundary confidence; The first i Column boundary point set B i for: (6) in:x i The x-coordinate of the sampling point; For the first i The ordinates of the boundary points detected in the column; row index j* The ordinate of the corresponding sampling point.

[0011] Specifically, in step 4.3, the set of boundary points for all columns is as follows: (7) Where M is the number of columns in the sampling grid; x i The x-coordinate of the sampling point; For the first i The ordinates of the boundary points detected in the column; The continuous function f ( x ): (8) in: k The index of the B-spline basis function; This represents the control coefficient of the B-spline curve; The basis functions representing B-spline curves; The smoothed trajectory function f smoothed (x) : (10) in, w The width of a local window is half its width. a m is the SG kernel coefficient; m is the relative index within the filter window.

[0012] Specifically, in step 4.4, if the boundary point x i Classification confidence p i < T low and adjacent boundary points x i-1 and x i+1 Confidence level at p i-1 ≥ T low and p i+1 ≥ T low , T low If a confidence threshold is set for boundary repair, then it is considered that... xi The trajectory value at a point is unreliable; weighted interpolation is used to correct the value at that point. Trajectory function corrected by confidence-weighted interpolation f corrected (x) : (11) in, α，β These are the weighting coefficients.

[0013] Specifically, in step 5, the motion support is controlled to move along the vertical Y-axis, for different... y Multiple parallel cross-sections of the value were repeatedly detected, and multiple values ​​were integrated. y The trajectory of the coal-rock interface on the cross section z= f corrected ( x, y This generates a three-dimensional spatial distribution map of the coal-rock interface.

[0014] Compared with the prior art, the present invention has the following technical effects: This invention employs Compton scattering imaging technology based on photon counting detectors to achieve multi-energy spectrum resolution of the scattered rays of the object under test. Combined with a multi-channel energy spectrum fusion algorithm, it can accurately distinguish between coal seams and thin interlayers of gangue (such as 0.5cm rock layers).

[0015] Compared with conventional Compton scattering imaging methods, this method employs X-ray Compton backscattering imaging based on a photon counting detector and uses a coal-rock interface generation algorithm based on multi-channel energy spectrum fusion and boundary regression fitting to identify the coal-rock interface trajectory. By combining energy-resolution detection, multi-energy spectrum fusion analysis, and dynamic trajectory fitting, this method solves the problems of low accuracy and poor anti-interference in the identification of complex coal seams (including thin interlayers of gangue), and combines high accuracy with adaptability to harsh environments.

[0016] This invention employs a CdTe photon counting detector to replace the traditional integrating detector, enabling individual counting and multi-energy spectral analysis of Compton backscattered photons. This overcomes the limitation of traditional methods, which can only count the total number of photons, and significantly improves the resolution of electron density and composition in coal / rock.

[0017] This invention proposes a multi-channel energy spectrum fusion and boundary regression fitting algorithm. The algorithm inputs the feature vectors of the multi-channel energy spectrum into the classifier and outputs coal / rock labels and confidence scores. Combining third-order B-spline interpolation and Savitzky-Golay smoothing filtering, a confidence score correction mechanism is introduced into the algorithm to generate high-precision, noise-resistant continuous coal-rock interface trajectories, thereby improving the accuracy and robustness of identification.

[0018] This invention employs an innovative boundary regression fitting algorithm. By detecting jump points, it quickly locates the boundary and uses third-order B-spline interpolation + SG filtering to generate a smooth and continuous trajectory (z=f(x)). The processing speed is 40% faster than traditional methods, and it is also resistant to local noise interference. Attached Figure Description

[0019] Figure 1 The diagrams show two X-ray imaging methods: (a) a schematic diagram of conventional transmission X-ray absorption imaging, and (b) a schematic diagram of X-ray Compton backscatter imaging.

[0020] Figure 2 This is a schematic diagram of the device layout of the present invention.

[0021] Figure 3 This is a flowchart of the data processing of the present invention.

[0022] Figure 4 This is a diagram showing the angular relationship between the X-ray source and the detector.

[0023] Figure 5 This is a diagram showing the backscattering angle distribution of X-rays with different energies.

[0024] The meanings of the labels in the diagram are as follows: 1. X-ray source, 2. Photon counting detector, 3. Motion support. Detailed Implementation

[0025] In recent years, research on techniques based on the Compton backscattering principle of X-rays, using kV-level X-rays to detect the flux distribution and energy spectrum of backscattered photons for coal-rock interface detection has been increasing. The main forms of X-ray interaction with matter include the photoelectric effect, the Compton effect, and coherent scattering due to electron-pair effects. The probability of different interaction forms occurring depends on factors such as X-ray energy and the atomic number of the absorbing material. For example, for water, the Compton effect dominates in the energy range of 30 keV to 25 MeV. Unlike conventional transmission-type X-ray absorption imaging, Compton scattering imaging allows the X-ray source and detector to be placed on the same side of the object being measured (e.g.,...). Figure 1 Its formula is: (1) Where: N0 is the photon flux of the original beam; This is the differential section for Compton scattering; denoted as θ, where θ is the measurement angle range of the detector relative to the object being measured; n is the electron density of the object being measured; L is the path length of the original beam in the object being measured; and S is the photon flux detected by the detector.

[0026] Because Compton scattering imaging can place the X-ray source and detector on the same side of the object being measured, this technology has promising applications in coal-rock interface identification. However, conventional integrating detectors can only count the total photon flux and cannot distinguish the X-ray energy spectrum distribution. They also cannot analyze the composition of the medium by separately processing the attenuation results of X-rays interacting with the medium at different energy levels. Therefore, the accuracy of Compton scattering imaging based on conventional integrating detectors in complex coal seams (such as those containing thin interlayers of gangue or mixed zones) is limited, restricting its practical application in coal-rock interface identification.

[0027] Based on this, the present invention develops a coal-rock interface identification technology that can improve the mining efficiency and safety of coal mines, can work stably in harsh environments, has high accuracy and good cost-effectiveness, and has important practical significance and broad application prospects for promoting intelligent and unmanned coal mining.

[0028] Specifically, the novel coal-rock identification method provided by this invention is a novel approach to coal-rock identification based on X-ray Compton backscattering imaging using a photon counting detector, and employing a coal-rock interface generation algorithm that combines multi-channel energy spectrum fusion and boundary regression fitting. This method utilizes a high-frequency pulsed X-ray source to irradiate the coal-rock interface region, and uses a photon counting detector made of CdTe semiconductor material to count and analyze the backscattered photons one by one. The coal-rock interface generation algorithm, employing multi-channel energy spectrum fusion and boundary regression fitting, accurately calculates and distinguishes between coal seams and rock strata.

[0029] The Compton backscattering imaging method (Equation 1) of conventional integrating detectors is based on several simplified assumptions, such as X-rays being monoenergetic, the detector being unable to distinguish the energy spectrum, and only counting the count rate. This method is suitable for integrating energy-indistinguishable detectors (such as NaI scintillators, ion chambers, etc.). In this invention, a novel scheme combining X-ray Compton backscattering imaging with a photon counting detector is adopted. The core advantage of the photon counting detector lies in its energy resolution capability, which can simultaneously acquire photon energy distribution information. Therefore, in this scheme, Equation (1) is modified as follows: (2) in: E Incident X-ray energy; S ( E The energy reaching the detector after scattering is '): E 'Photon flux; P ( E → E '): Energy transfer probability (Compton energy transfer function * detector energy response) N 0(E ): X The energy spectrum of the radiation source (broad energy spectrum); E max , E min : The X-ray energy range, and the meanings of other characters are the same as those in formula (1).

[0030] The following are specific embodiments of the present invention. It should be noted that the present invention is not limited to the following specific embodiments, and all equivalent transformations made on the basis of the technical solutions of this application fall within the protection scope of the present invention.

[0031] Embodiment 1: This embodiment provides a new type of coal and rock identification device, including an X-ray source, a photon counting detector, a moving bracket, and a data processing and identification module, with the structure as shown in Figure 2 ; Among them: The function of the X-ray source is to emit X-rays, and the energy of the X-rays can be adjusted by adjusting the working voltage of the X-ray source.

[0032] The function of the photon counting detector is to receive Compton backscattered rays and measure their ray dose and energy distribution. A collimator module is installed on the detector, and the direction of the collimator can be adjusted to receive and measure the rays in a specific direction, so as to realize the measurement of substances in a specific area and at a specific depth.

[0033] The function of the moving bracket is to support the X-ray source and the detector, and to detect different areas by moving.

[0034] The data processing and identification module is the supporting system software of this system. Its functions are to control the working timing of the X-ray source and the detector, measure the X-ray signal, process and analyze the measurement data, analyze the substances in the measured area, and perform coal and rock identification. The main working process of the data processing and identification module is as follows: Pulse modulation: The data processing and identification module controls the working timing of the X-ray source and the detector by modulating the synchronous pulse signal.

[0035] Signal acquisition: Under the control of the pulse signal, the X-ray source is exposed to emit X-rays, and the detector receives the X-ray signal to complete the signal acquisition of the X-ray backscattered rays.

[0036] Data processing: According to the exposure parameters of the X-ray source (tube current, tube voltage, etc.), calculate the parameters of the rays emitted by the X-ray source (i.e., the original ray I0). According to the X-ray signal received by the detector, calculate the Compton backscattered ray I bs .

[0037] Identification and analysis: According to the original ray I0 and the collected Compton backscattered ray I bsThe material in the tested area is analyzed to identify coal and rock. For example... Figure 3 This is a flowchart of the data processing and recognition module software.

[0038] In this embodiment, both the X-ray source and the photon counting detector are fixed on a moving support, located on one side of the coal and rock detection area. The emitting end face of the X-ray source and the signal acquisition end face of the photon counting detector both face the coal and rock detection area. After the X-rays interact with the coal and rock material, Compton backscattering occurs, and the photon counting detector collects and detects the reflected X-ray backscattered signal. The angular relationship between the X-ray source and the photon counting detector is as follows: Figure 4 and Figure 5 To optimize the intensity and signal-to-noise ratio of the Compton backscattered signal, while considering the typical depth range of the coal-rock interface and the spatial constraints of the working face, the angle θ between the central axis of the X-ray source and the central axis of the photon counting detector (see...) is... Figure 4 The range is selected within 150° to 180°. This range is chosen with reference to the Compton scattering differential section (see...). Figure 5 Within this angular range, it exhibits a high scattering probability, taking into account both the collimation angle limitation of the detector and the geometric constraints of the working surface. The specific included angle θ value can be determined through simulation or experiment based on the actual application scenario (such as expected detection depth and spatial resolution requirements).

[0039] Example 2: This embodiment provides a novel coal and rock identification method, which is based on the coal and rock identification device in Embodiment 1, and includes the following steps: Step 1: Move the X-ray source and photon counting detector to the front of the area to be detected, i.e., the target detection area, using a motion support. The specific models, configurations, and parameters of each component in this embodiment are as follows: (1.1) X-ray source assembly: A high-frequency pulsed X-ray tube (maximum voltage 250kV, current 5mA) is used, installed in an explosion-proof stainless steel housing. Specific parameters: -Pulse frequency: 10Hz (adjustable), pulse width 20~100ms; – Focal point size: 0.8mm × 0.8mm; – Beam angle: 40° conical beam.

[0040] (1.2) Detector array: High-sensitivity CdTe semiconductor photon counting detector (size 20mm×20mm×5mm), with multi-channel energy spectrum counting and real-time energy resolution functions; specific parameters: – Spectral resolution: better than 2keV (@60keV); –Number of energy channels: 32 channels can be set; -Integrated collimator module, collimation angle adjustable ±15° (stepper motor driven, accuracy 0.5°).

[0041] (1.3) The motion support is a three-dimensional motion platform support: a hydraulically driven three-axis slide system (X, Y, Z); – Travel range: X-axis 2m (along the working surface direction), Z-axis 1m (vertical direction); – Positioning accuracy: ±2mm (laser ranging feedback); – Maximum load: 50kg (including radiation source and detector).

[0042] (1.4) Protective structure: Double-layer lead-tungsten alloy shield (10mm Pb equivalent in the front and 5mm Pb equivalent in the side).

[0043] (1.5) Data processing and recognition module: Embedded edge computing module (NVIDIA Jetson Xavier) to realize functions such as synchronous signal acquisition, multi-channel energy spectrum decoding, classifier recognition, and boundary trajectory fitting.

[0044] More specifically, in this embodiment, the motion support is moved to the set target point coordinates x=3200mm, z=1200mm; X-ray source operating parameters: tube voltage 180kV, tube current 3mA, pulse operating mode, pulse width 50ms; photon counting detector parameters: energy detection range 30keV ~ 150keV, divided into 32 energy channels (n = 32); core algorithm parameters of the data processing and recognition module: boundary confidence screening threshold. T high (e.g., 0.85); Boundary repair confidence threshold T low (e.g., 0.6); Savitzky-Golay (SG) smoothing filter parameters (e.g., local window half-width w = 5); weighted interpolation weight coefficients. α，β (Calculated based on the spatial distance between adjacent boundary points).

[0045] Step 2: The X-ray source emits high-frequency pulsed X-rays (i.e., primary rays I0) of corresponding energy and intensity according to the parameters set by the system, and irradiates the area to be detected; the system software sends a synchronization pulse signal. Step 3: The photon counting detector receives Compton backscattered radiation (Ibs) emitted from the region to be detected and collects multi-energy spectral scattering data for multi-channel energy spectral feature extraction. In this embodiment, data acquisition lasts for 200ms, including approximately 20 pulse cycles. The system accumulates or averages data from multiple pulse cycles to improve the signal-to-noise ratio and transmits the dose values ​​to the system software data processing and recognition module. Specifically, this includes: Step 3.1: The photon counting detector receives Compton backscattered photons. I bs ; Step 3.2: Divide the detected backscattered photons into n discrete energy ranges (channels) according to their energy levels, and assign each detected photon to a preset n (32 in this embodiment) energy channel according to its energy. Step 3.3, divide the detection surface (corresponding to the coal face section) into... i Column (horizontal direction) j The sampling grid along the line (depth direction); Step 3.4, at each sampling grid point ( x i , z j At position ), the number of photons detected in each energy channel is recorded. The vector formed by the number of photons detected in these n channels constitutes the multi-channel energy spectrum counting vector: (3) Step 4: The identification module analyzes the material in the measured area and performs coal and rock identification based on the original radiation I0, Compton backscattered radiation Ibs, and multi-channel energy spectrum characteristics, using a coal-rock interface generation algorithm based on multi-channel energy spectrum fusion and boundary regression fitting; specifically including: Step 4.1: Classifier identifies coal / rock and confidence level A classifier is used to determine whether a sample belongs to a coal seam (label 0) or a rock seam (label 1) by analyzing the multi-channel energy spectrum features collected by a photon counting detector. The classifier outputs the confidence level (reliability quantification) of the prediction result. In this embodiment, the classifier uses existing technologies, such as being built based on an open-source framework (e.g., Scikit-learn SVM), and is trained and calibrated using downhole measured energy spectrum data for engineering adaptation. Its input is the multi-channel energy spectrum counting vector f extracted in step 3. ij The output is a classification label and a confidence score. In this embodiment, the input feature vector is 𝑓. 𝑖𝑗 Use the trained classifier to make predictions: (4) in: Category tags; Category tags 0 represents the coal seam, and 1 represents the rock strata. This is directly used for boundary jump detection in subsequent steps. l ij ≠ l i(j+1) (Time-triggered jump point judgment).

[0046] : Confidence level of the label; confidence level A larger value indicates a more reliable classification result, and it is used to screen reliable boundary points and for confidence correction in subsequent steps. p ij < T low Interpolation repair is initiated at the time.

[0047] Step 4.2: Boundary jump detection (local label change + high confidence) Based on the classification labels output in step 4.1 l ij (0 = coal, 1 = rock) The location where the coal and rock labels of adjacent scan points (different depths in the same column) abruptly change is defined as a jump point, characterizing the starting point of a possible transition zone at the coal-rock interface. Specifically, the location where the classification label jumps is detected ( x i , z j ): That is, adjacent depth points satisfy l ij ≠ l i(j+1) .

[0048] Based on the confidence level output in step 4.1 p ij (Classification reliability quantification) The reliable transition points selected by confidence level are those that satisfy max( p ij , p i(j+1) )> T high The coordinates of the transition point are defined as the boundary point; specifically, the coordinates of the two adjacent points involved in the transition point are calculated. x i , z j ) and( x i , z j+1 The maximum confidence level of ) max( p ij , p i(j+1) Only retain those that satisfy max( p ij , p i(j+1) ) > T high The jump point.

[0049] The coal-rock interface is essentially a depth-continuous curved surface. In complex coal seams, the labels of adjacent points may frequently jump (coal → rock → coal). False jumps (such as noise interference) need to be filtered out using confidence scores to locate the true interface. This step uses the original classification results (including noisy labels) obtained in step 4.1, and utilizes confidence scores... T high Low-confidence transitions are eliminated, and the discrete spatial location of the coal-rock interface is extracted; this provides accurate input data for the subsequent B-spline fitting step.

[0050] For each column (fixed 𝑥) 𝑖 Scanning 𝑗 Find the first boundary point: (5) in, The row index j of the column where a coal-rock label jump occurs (i.e., lij ≠ li(j+1)) and the jump has a high confidence level (max(pij, pi(j+1))>Thigh) represents a reliable coal-rock transition location encountered when scanning from this column.

[0051] Define boundary points B i for: (6) in: x i This represents the x-coordinate of the sampling point, i.e., the horizontal coordinate of the column. The row index calculated in formula (5) The corresponding ordinate of the sampling point; Let B be the ordinate of the boundary point detected in column i; and let B be the set of boundary points in formula (6). i It refers to the first i Boundary points in the column.

[0052] Step 4.3: Boundary point regression fitting (B-spline + SG smoothing) The set of discrete boundary points detected in step 4.2 This is fitted into a continuous curve function that characterizes the coal-rock interface as a function of the horizontal position x. z = f(x) .

[0053] Let the set of boundary points be: (7) Where M is the number of columns in the sampling grid; the set of boundary points B in formula (7) is the global collection of all valid boundary points {B1, B2, ..., BM} in all columns.

[0054] Using the set of boundary points as input nodes, a line representing the coal-rock interface is generated by fitting a third-order B-spline interpolation. x Continuous function of change f ( x ): (8) in: k The index of the B-spline basis function; The control coefficients of the B-spline curve are represented. The basis functions represent the B-spline curve; The set of boundary points B = {B1, B2, ..., BM} in formula (7) provides M discrete coal-rock interface points. These points are used as input nodes for trajectory fitting. Formula (8) uses these discrete points to generate a continuous coal-rock interface function f(x), where the index k extends M from 0 to M+2 (because the third-order B spline requires additional control points to handle curve endpoint smoothing). That is, the discrete set B of Formula (7) is directly used as the input data for spline fitting of Formula (8), ensuring a natural transition from discrete boundary points to continuous trajectory.

[0055] To eliminate high-frequency noise disturbances (such as coordinate jitter caused by mechanical vibration and photon count fluctuations) in the B-spline fitted trajectory, while preserving the true geometric abrupt changes in the coal-rock interface (such as the steep boundary of thin-layer interbedded rock), a Savitzky-Golay filter is used for local smoothing. Through local polynomial least squares fitting, an optimal balance is achieved between smoothing noise (improving the cutting stability of the coal mining machine) and protecting details (ensuring the detection rate of thin-layer interbedded rock), thus solving the problem of loss of thin-layer features caused by excessive smoothing in traditional filtering methods (such as Gaussian blur).

[0056] The continuous function generated by third-order B-spline interpolation fitting is input into the Savitzky-Golay filter for local smoothing. (9) in: represents the ordinate of the sampled point after filtering and smoothing; w is the half-width of the local window, where w is a positive integer that defines the size of the filtering window. a m SG kernel coefficients, i.e., the weight coefficients of the filter, are pre-calculated by polynomial least squares fitting; m is the relative index within the filter window, representing the offset relative to the current point.

[0057] In this embodiment, the filter uses a preset window half-width. w (For example w =5, indicating the use of 2 w(+1=11 data points window). Polynomial least squares fitting is performed within each local window to generate a smoothed coal-rock interface trajectory function. f smoothed ( x ).

[0058] Obtain the smoothed trajectory function f smoothed (x) : (10) in: w : Partial window half-width; a m : SG kernel coefficients; w, a m The meaning of m is the same as that of formula (9).

[0059] This step effectively suppresses high-frequency noise caused by photon counting fluctuations or mechanical micro-vibrations, while maintaining the true boundary abrupt changes characteristics of thin-layer silicon inclusions.

[0060] Step 4.4: Confidence Adjustment Although step 4.2 passed T high Even after filtering out low-confidence jump points, the fitted trajectory may still be affected by two types of interference: regions of uneven coal quality and sparse boundary points. Regions of uneven coal quality contain impurities such as pyrite, which can negatively impact the classification confidence of local points. p i The confidence level is low; sparse boundary point areas are caused by missing boundary points in a certain column due to detection blind spots or mechanical vibration. If the fitting results from step 4.3 are used directly, trajectory deviations will occur at such locations. Therefore, confidence correction is required in this step.

[0061] The object of this step is to correct the trajectory function fitted in step 4.3. f smoothed (x) Boundary points x i The value, not the original boundary points or the data before fitting. Step 4.1 output p i This involves a classifier assessing the reliability of coal / rock tags for each original scan point. In step 4.2, the system filters out tags that meet the requirements. p i > T high boundary points Bi Input the data from step 4.3 to perform the fitting, and obtain the confidence level used in this step.

[0062] For boundary points x i(As defined in step 4.2), if its classification confidence level p i < T low and adjacent boundary points x i-1 and x i+1 Confidence level at p i-1 ≥ T low and p i+1 ≥ T low , then it is considered that x i The trajectory value at a given point may be unreliable due to localized coal quality inhomogeneity (such as impurities containing pyrite). Weighted interpolation should be used to correct this point value; otherwise, retain the value. f smoothed (x)。

[0063] Use adjacent high-confidence boundary points ( x i-1 , f smoothed ( x i-1 )) and( x i+1 , f smoothed ( x i+1 The smooth trajectory value at point )) is based on a preset weighting coefficient. α and β (usually determined by spatial distance) and (Decision), calculate the trajectory function after confidence-weighted interpolation correction. f corrected (x) : (11) Among them, the weighting coefficient α，β Based on spatial distance calculations, a smooth trajectory is ensured. α，β Weighting coefficients: , .

[0064] Step 4.5: Output Trajectory The fitted curve, smoothed in step 4.3 and corrected in step 4.4, is... z = f corrected (x)As the final coal-rock interface trajectory, the generated high-precision, smooth, and continuous coal-rock interface trajectory is transmitted in real time to the coal mining machine control system via industrial Ethernet to realize the identification of the coal-rock interface, which serves as a key input to guide the cutting path and direct the coal mining machine to cut the path.

[0065] Table 1 Character Definitions in Formulas

[0066] Step 5: Move the X-ray source and photon counting detector to different areas using the motion support. Repeat the above steps. Specifically, control the motion support to move the detector module in the vertical direction (Y-axis) for multiple parallel cross-sections (different... y (Value) Repeat the above steps; the system software integrates multiple y The trajectory of the coal-rock interface on the cross section z= f corrected ( x, y This generates a three-dimensional spatial distribution map of the coal-rock interface, providing more comprehensive information for mining planning and geological analysis.

[0067] The preferred embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the specific details of the above embodiments. Within the scope of the technical concept of the present invention, various simple modifications can be made to the technical solution of the present invention, and these simple modifications all fall within the protection scope of the present invention.

[0068] It should also be noted that the various specific technical features described in the above embodiments can be combined in any suitable manner without contradiction. To avoid unnecessary repetition, the present invention will not describe the various possible combinations separately.

[0069] Furthermore, various different embodiments of the present invention can be combined in any way, as long as they do not violate the spirit of the present invention, they should also be regarded as the content disclosed by the present invention.

Claims

1. A method for coal and rock identification, characterized in that, This method is based on a coal-rock identification device to identify the coal-rock interface. The coal-rock identification device includes an X-ray source, a photon counting detector, a motion support, and a data processing and identification module. The X-ray source and the photon counting detector are both mounted on the motion support and placed on the same side of the coal and rock to be detected. The data processing and identification module can control the working sequence of the X-ray source and the photon counting detector, and process and analyze the measurement data to identify the coal and rock. The method includes the following steps: Step 1: Move the X-ray source and photon counting detector to the front of the coal and rock area to be inspected using a motion support; Step 2: The X-ray source emits high-frequency pulsed X-rays, i.e., primary rays I0, according to the set parameters to irradiate the area to be detected; Step 3: The photon counting detector receives the Compton backscattered rays, i.e., Compton backscattered photons, emitted from the area to be detected. I bs Multi-energy spectral scattering data is collected, multi-channel energy spectral features are extracted, and the data is transmitted to the data processing and recognition module. Step 4, the data processing and recognition module determines the originating ray I0 and the Compton backscattered photons based on the data. I bs Based on multi-channel energy spectrum characteristics, a coal-rock interface generation algorithm based on multi-channel energy spectrum fusion and boundary regression fitting is used. Combined with third-order B-spline interpolation and Savitzky-Golay smoothing filtering, and confidence correction is introduced, the material of the measured area is analyzed to identify coal and rock. Step 5: Move the X-ray source and photon counting detector to different areas using a motion support, and repeat the above steps for multiple parallel sections; integrate the coal-rock interface trajectories on multiple sections to generate a three-dimensional spatial distribution map of the coal-rock interface, providing more comprehensive information for mining planning and geological analysis.

2. The coal and rock identification method as described in claim 1, characterized in that, The X-ray source is used to emit X-rays, and the energy of the X-rays is adjusted by regulating the operating voltage of the X-ray source; The photon counting detector is used to receive Compton backscattered rays and measure their radiation dose and energy distribution; the photon counting detector is equipped with a collimator module, and by adjusting the direction of the collimator, it can receive and measure rays in a specific direction to achieve the measurement of matter in a specific area and at a specific depth. The motion support is used to support the X-ray source and the photon counting detector, and to enable detection of different areas by moving the support.

3. The coal and rock identification method as described in claim 2, characterized in that, The X-ray source's emitting end face and the photon counting detector's signal acquisition end face both face the coal and rock detection area. After the X-rays interact with the coal and rock material, Compton backscattering occurs, and the photon counting detector acquires and detects the reflected X-ray backscattering signal. The angle θ between the central axis of the X-ray source and the central axis of the photon counting detector is selected within the range of 150° to 180°.

4. The coal and rock identification method as described in claim 1, characterized in that, Step 3 includes: Step 3.1: The photon counting detector receives Compton backscattered photons. I bs ; Step 3.2: Distribute the detected Compton backscattered photons into n preset energy channels according to their energy levels; Step 3.3, divide the surface to be inspected into... i List, j The sampling grid of the row; Step 3.4, at each sampling grid point ( x i , z j At point ), the number of photons detected by each energy channel is recorded, forming a multi-channel energy spectrum counting vector: （3）。 5. The coal and rock identification method as described in claim 4, characterized in that, Step 4 includes: Step 4.1, Classifier identifies coal / rock and confidence level: The classifier is used to determine whether a substance is coal or rock. Its input is the multi-channel energy spectrum counting vector f extracted in step 3. ij The output is category labels. l ij and confidence level p ij ; Step 4.2, Boundary transition detection: Based on the classification labels output in step 4.1 l ij The location where the coal and rock classification labels of adjacent scan points at different depths in the same column suddenly change is defined as a jump point; based on the confidence level output in step 4.1... p ij The coordinates of reliable transition points, filtered by confidence level, are defined as boundary points; the detection yields the first... i Column boundary point set B i ; Step 4.3, Boundary point regression fitting and smoothing filtering: Using the set of boundary points of all columns detected in step 4.2 as input nodes, a continuous line representing the coal-rock interface as the horizontal position is generated by third-order B-spline interpolation. x Continuous function of change f(x) ; Then the continuous function f(x) The Savitzky-Golay filter is used for local smoothing to obtain the smoothed coal-rock interface trajectory function. f smoothed ( x ); Step 4.4, Confidence Adjustment: For trajectory function f smoothed ( x For boundary points on the boundary, if their classification confidence is lower than the boundary repair confidence threshold and the confidence of adjacent boundary points is not lower than the boundary repair confidence threshold, then weighted interpolation is used to correct the point value; otherwise, the boundary is retained. f smoothed (x) The trajectory function after confidence-weighted interpolation correction is obtained. f corrected (x) ; Step 4.5, Output trajectory: The corrected trajectory function z = f corrected (x) As the final coal-rock interface trajectory, this high-precision, smooth, and continuous coal-rock interface trajectory enables the identification of the coal-rock interface and can guide the cutting path of the coal mining machine.

6. The coal and rock identification method as described in claim 5, characterized in that, In step 4.1, a classifier is used to determine whether a substance is coal or rock, and the input feature vector is 𝑓. 𝑖𝑗 Use the trained classifier to make predictions: （4） Among them, category labels 0 represents coal seams, and 1 represents rock strata; confidence level This quantifies the reliability of the classification; a larger value indicates a more reliable classification result.

7. The coal and rock identification method as described in claim 5, characterized in that, In step 4.2, the jump point is the location where the classification label undergoes a sudden change. x i , z j ( ), satisfying scan points at different depths in the same column l ij ≠ l i(j+1) ; The boundary points satisfy max( p ij , p i(j+1) ) > T high , where max( p ij , p i(j+1) ) are two adjacent points ( x i , z j ) and( x i , z j+1 The maximum confidence level of ) T high Filter thresholds for boundary confidence; The first i Column boundary point set B i for: （6） in: x i The x-coordinate of the sampling point; For the first i The ordinates of the boundary points detected in the column; row index j* The ordinate of the corresponding sampling point.

8. The coal and rock identification method as described in claim 5, characterized in that, In step 4.3, the set of boundary points for all columns is: （7） Where M is the number of columns in the sampling grid; x i The x-coordinate of the sampling point; For the first i The ordinates of the boundary points detected in the column; The continuous function f ( x ): （8） in: k The index of the B-spline basis function; This represents the control coefficient of the B-spline curve; The basis functions representing B-spline curves; The smoothed trajectory function f smoothed (x) : （10） in, w The width of a local window is half its width. a m is the SG kernel coefficient; m is the relative index within the filter window.

9. The coal and rock identification method as described in claim 5, characterized in that, In step 4.4, if the boundary point x i Classification confidence p i < T low and adjacent boundary points x i-1 and x i+1 Confidence level at p i-1 ≥ T low and p i+1 ≥ T low , T low If a confidence threshold is set for boundary repair, then it is considered that... x i The trajectory value at a point is unreliable; weighted interpolation is used to correct the value at that point. Trajectory function corrected by confidence-weighted interpolation f corrected (x) : （11） in, α，β These are the weighting coefficients.

10. The coal and rock identification method as described in claim 1, characterized in that, In step 5, the motion support is controlled to move along the vertical Y-axis, for different... y Multiple parallel cross-sections of the value were repeatedly detected, and multiple values ​​were integrated. y The trajectory of the coal-rock interface on the cross section z= f corrected ( x, y This generates a three-dimensional spatial distribution map of the coal-rock interface.

Images