Dual-energy x-ray based foreign matter identification method and system for coal mine dry separator

Abstract

This invention discloses a method and system for foreign object identification in coal mine dry separators based on dual-energy X-rays, relating to the field of foreign object identification in dry separators. The method includes: acquiring low-energy X-ray images and high-energy X-ray images; simultaneously acquiring three-dimensional contour data using a laser contour measurement device; calculating the absolute thickness value in the X-ray penetration direction based on the three-dimensional contour data; performing thickness effect compensation on the low-energy and high-energy X-ray images to generate compensated low-energy and compensated high-energy X-ray images; inputting the compensated low-energy and compensated high-energy X-ray images into a pre-trained material identification model to obtain the equivalent atomic number of the detected object; and determining whether the detected object is a foreign object based on the equivalent atomic number and a preset foreign object discrimination threshold. This invention solves the technical problems in the prior art where the thickness of the foreign object affects X-ray absorption, resulting in low accuracy and operational reliability in foreign object identification and a high false identification rate.

Description

Technical Field

[0001] This invention relates to the field of foreign object identification in dry separators, and specifically to a method and system for foreign object identification in coal mine dry separators based on dual-energy X-rays. Background Technology

[0002] Dual-energy X-ray dry separators are the core equipment for dry coal preparation. Their basic principle is to utilize the attenuation difference between low-energy and high-energy X-rays when they pass through materials, and to invert the equivalent atomic number of materials by calculating the dual-energy gray-scale ratio, thereby realizing the material differentiation between coal and gangue and foreign objects.

[0003] However, existing dual-energy X-ray dry separators suffer from technical problems such as low accuracy and reliability of foreign object identification and high false identification rate when there are differences in material thickness. Summary of the Invention

[0004] This application provides a method and system for foreign object identification in coal mine dry separators based on dual-energy X-rays, which addresses the technical problems in the prior art where the thickness of foreign objects affects X-ray absorption, resulting in low accuracy and reliability of foreign object identification and a high false identification rate.

[0005] In view of the above problems, this application provides a method and system for foreign object identification in coal mine dry separators based on dual-energy X-rays.

[0006] In a first aspect, this application provides a method for foreign object identification in coal mine dry separators based on dual-energy X-rays, the method comprising: The system acquires low-energy X-ray images and high-energy X-ray images of the object being detected, and simultaneously acquires the three-dimensional contour data of the object being detected through a laser contour measurement device. The absolute thickness of the detected object in the X-ray penetration direction is calculated based on the three-dimensional contour data. Based on the absolute thickness value, thickness effect compensation processing is performed on the low-energy X-ray image and the high-energy X-ray image respectively to generate compensated low-energy X-ray images and compensated high-energy X-ray images; The compensated low-energy X-ray image and the compensated high-energy X-ray image are input into a pre-trained material recognition model to calculate the equivalent atomic number of the detected object. The model then determines whether the detected object is a foreign object based on the equivalent atomic number and a preset foreign object discrimination threshold.

[0007] Secondly, the present invention provides a foreign object identification system for coal mine dry separators based on dual-energy X-rays, the system comprising: The contour data acquisition module is used to acquire low-energy X-ray images and high-energy X-ray images of the object being detected, and to simultaneously acquire the three-dimensional contour data of the object being detected through a laser contour measurement device. The penetration thickness calculation module is used to calculate the absolute thickness value of the detected object in the X-ray penetration direction based on the three-dimensional contour data. The penetration thickness compensation module is used to perform thickness effect compensation processing on the low-energy X-ray image and the high-energy X-ray image respectively according to the absolute thickness value, and generate compensated low-energy X-ray image and compensated high-energy X-ray image. The material identification module is used to input the compensated low-energy X-ray image and the compensated high-energy X-ray image into a pre-trained material identification model, calculate the equivalent atomic number of the detected object, and determine whether the detected object is a foreign object based on the equivalent atomic number and a preset foreign object discrimination threshold.

[0008] One or more technical solutions provided in this application have at least the following technical effects or advantages: This application acquires the three-dimensional contour data of the object under test through laser contour measurement, achieving multimodal heterogeneous data fusion and providing a geometric measurement basis for subsequent thickness elimination. By using a laser contour measurement device and an X-ray detector, pixel-level correspondence is ensured at sub-millimeter spatial accuracy, avoiding compensation errors caused by data misalignment. Secondly, the absolute thickness value of the object under test is calculated using the three-dimensional contour data. Dual-laser differential technology is used to eliminate common-mode errors in thickness measurement caused by belt slippage, providing reliable physical input for subsequent thickness compensation. Thirdly, by acquiring grayscale response images of different thicknesses, the dual-energy grayscale values ​​of materials at any thickness are corrected to the theoretical grayscale level based on a reference thickness normalization. Compensation is then applied using a compensation coefficient to eliminate the strong interference of thickness on grayscale values, preserving the monotonic mapping relationship between the dual-energy ratio and the equivalent atomic number. Finally, by calculating the equivalent atomic number of the object under test and comparing it with a preset foreign object discrimination threshold, an accurate threshold control strategy is obtained. Dynamic threshold configuration adaptively adjusts the discrimination boundary, avoiding the poor adaptability of fixed thresholds in coal source fluctuation scenarios. Attached Figure Description

[0009] Figure 1 This is a flowchart illustrating the foreign object identification method for coal mine dry separators based on dual-energy X-rays, as described in this application. Figure 2 This is a schematic diagram of the foreign object identification system for coal mine dry separators based on dual-energy X-rays, as described in this application.

[0010] In the attached diagram, the components represented by each number are as follows: Contour data acquisition module 11, penetration thickness calculation module 12, penetration thickness compensation module 13, and material identification module 14. Detailed Implementation

[0011] This application provides a method for foreign object identification in coal mine dry separators based on dual-energy X-rays, which specifically solves the technical problems in the prior art where the thickness of foreign objects affects X-ray absorption, resulting in low accuracy and reliability of foreign object identification and a high false identification rate.

[0012] The present invention will now be described in detail with reference to the accompanying drawings.

[0013] Example 1, as Figure 1 As shown, this application provides a method for foreign object identification in coal mine dry separators based on dual-energy X-ray diffraction, the method comprising: S10: Acquire low-energy X-ray images and high-energy X-ray images of the object being detected, and simultaneously acquire the three-dimensional contour data of the object being detected through a laser contour measurement device. In this embodiment, the object being detected is a material on a conveyor belt passing through the detection area of ​​the dry separator; the low-energy X-ray image and the high-energy X-ray image are grayscale images formed by low-energy photons and high-energy photons collected by the detector in a dual-energy X-ray imaging system after penetrating the object being detected; the laser contour measurement device is a three-dimensional scanning device composed of a high-precision laser emitter and a CMOS / CCD receiving unit; the three-dimensional contour data is a two-dimensional matrix indexed by a pixel coordinate system, with each pixel position recording the height value of the object's surface, constituting a digital representation of the three-dimensional shape of the object being detected.

[0014] Specifically, a dual-energy X-ray imaging unit continuously emits a broad-spectrum X-ray beam. After the beam penetrates the object being inspected on the conveyor belt, it is received by a linear array detector. Using a stacked scintillator structure or rapid energy switching technology, low-energy channel image signals and high-energy channel image signals are output separately within the same scanning cycle to form a raw dual-energy image. Simultaneously, a laser contour measurement device, installed across a gantry spanning the conveyor belt, acquires three-dimensional contour data of the belt and material surface.

[0015] Step S10 in the method provided in this application embodiment includes: A first laser profilometer and a second laser profilometer are installed in parallel on an X-ray gantry. The beam of the first laser profilometer is perpendicular to the surface of the conveyor belt and is used to scan the upper surface profile information of the object being inspected. The beam of the second laser profilometer is aligned with the reference surface of the conveyor belt or the reference surface of the idler roller and is used to measure the position information of the reference surface of the conveyor belt in real time. The encoder pulse trigger signal simultaneously triggers the first laser profilometer, the second laser profilometer, and the X-ray detector to achieve sub-millimeter-level spatial synchronous acquisition of the three-dimensional contour data, the low-energy X-ray image, and the high-energy X-ray image. Real-time differential calculation is performed on the upper surface contour information collected by the first laser profilometer and the position information of the conveyor belt reference surface collected by the second laser profilometer to eliminate common mode measurement error caused by conveyor belt jumping. Based on the data obtained from the differential operation, the true upper surface height of the detected object at the position of the X-ray scan line is reconstructed to generate the three-dimensional contour data.

[0016] In this embodiment, the pulse trigger signal is a TTL level pulse generated by the encoder, used to control the acquisition time of the linear array sensor; the common mode measurement error is caused by belt jumping, etc., and is the error in the measurement values ​​of the first and second profilometers; the X-ray scan line position is the spatial cross-sectional position of the X-ray detector linear array.

[0017] Specifically, firstly, two high-precision laser profilometers are installed in parallel on an X-ray gantry, forming a dual-laser ranging system. The first laser profilometer is installed vertically downwards, with its laser line spanning the width of the conveyor belt. Each time material passes by, this profilometer scans and records the height data of the material's upper surface profile. The second laser profilometer aligns its laser line with an unloaded area of ​​the belt, such as an area without material covering the belt edge or directly with the highest point of the idler supporting the belt, to continuously monitor the instantaneous vertical position of the belt's reference surface. This parallel installation ensures that both profilometers and the X-ray detector operate in the same spatial coordinate system, providing the initial measurement basis for subsequent high-precision thickness calculations.

[0018] Secondly, in high-speed operation scenarios of coal mine dry separators, if each sensor acquires data independently, uncontrollable spatial misalignment will exist between the X-ray image and the laser profile data, which can easily lead to errors in subsequent thickness compensation. Therefore, by installing an incremental encoder on the end of the belt driven pulley shaft, dozens of TTL level pulses are generated per millimeter of displacement. Each pulse simultaneously triggers the first laser profiler, the second laser profiler, and the X-ray detector to perform scanning.

[0019] Secondly, during the synchronous data acquisition process, the first profilometer collects the surface profile information (i.e., surface profile height) of a row of materials, while the second profilometer collects the position information (i.e., belt reference surface height) of the belt reference surface. The X-ray detector simultaneously collects low-energy and high-energy grayscale data. Subsequently, real-time differential calculations are performed on the surface profile information collected by the first laser profilometer and the position information of the conveyor belt reference surface collected by the second laser profilometer. That is, immediately after each frame of data acquisition, the height value of the first profilometer corresponding to the same pulse number is subtracted from the height value of the second profilometer pixel-by-pixel or as a whole to obtain the real-time difference. This eliminates errors in the measurements of the first and second profilometers caused by factors such as belt jump, vibration, and tension changes. Notably, the distance values ​​measured by the first and second profilometers on the same rigid gantry will produce common-mode measurement errors with the same direction and similar amplitude.

[0020] Finally, based on the contour sequence generated after differential processing, the true upper surface height of the detected object is obtained by differential correction at the spatial cross-sectional position of the X-ray detector linear array. Subsequently, three-dimensional contour data is generated with the horizontal axis representing the position in the conveying direction, the vertical axis representing the position in the belt width direction, and the pixel value representing the true upper surface height at that location. The three-dimensional contour data includes the upper surface contour information of the detected object, i.e., the true upper surface height, and the position information of the conveyor belt reference surface.

[0021] In this embodiment, a first laser profilometer and a second laser profilometer are installed in parallel on an X-ray gantry, and a data acquisition device for collecting contour data is arranged. This parallel installation provides a structural foundation for subsequent synchronous data acquisition. Subsequently, the first laser profilometer, the second laser profilometer, and the X-ray detector are simultaneously triggered by an encoder pulse trigger signal to achieve simultaneous acquisition of images with sub-millimeter spatial resolution in the spatial dimension, improving the real-time performance and reliability of the system. Then, the real-time difference between the two profilometers is calculated based on the position information of the belt reference plane. The difference is then used to eliminate common-mode measurement errors, providing a correction basis for constructing three-dimensional contour data and improving the anti-interference capability of data measurement. Finally, the difference is used to restore the true upper surface height of the detected object at the position of the X-ray scan line, generating three-dimensional contour data and providing reliable data input for subsequent operations.

[0022] S20: Calculate the absolute thickness value of the detected object in the X-ray penetration direction based on the three-dimensional contour data; In this embodiment, the X-ray penetration direction is the propagation path of the X-ray beam emitted from the focal point of the X-ray tube, penetrating the object being detected, and reaching the detector; the absolute thickness value is the actual material path length traveled by the X-ray beam inside the object being detected along the X-ray penetration path.

[0023] Specifically, laser ranging values ​​and belt reference surface laser ranging values ​​are extracted from the 3D contour data. An initial vertical thickness value is obtained by subtracting values ​​pixel by pixel. Subsequently, a planar fit is performed on the neighborhood point cloud surrounding each pixel, and finally, the length of the ray along the actual penetration path, i.e., the absolute thickness value, is obtained.

[0024] Step S20 in the method provided in this application embodiment includes: The vertical height value corresponding to each pixel on the surface of the detected object is extracted from the three-dimensional contour data, wherein the vertical height value is the vertical distance from the laser emission point to the upper surface of the detected object; Extract the height value of the conveyor belt reference surface corresponding to the same pixel position from the three-dimensional contour data, wherein the height value of the conveyor belt reference surface is the vertical distance from the laser emission point to the belt surface or the idler roller reference surface; The vertical height value is subtracted pixel by pixel from the height value of the conveyor belt reference surface to obtain the initial value of the vertical thickness of the detected object. Local neighborhood point cloud analysis is performed on the three-dimensional contour data to calculate the normal vector at each pixel on the surface of the detected object and determine the local tilt angle of the material surface at each pixel. Based on the local tilt angle, the initial vertical thickness value is corrected to the actual X-ray penetration path length, and the actual X-ray penetration path length is used as the absolute thickness value, wherein the actual X-ray penetration path length is equal to the initial vertical thickness value divided by the cosine of the local tilt angle.

[0025] In this embodiment of the application, local neighborhood point cloud analysis is a method of constructing a local point cloud, fitting the geometric characteristics of the local region, and performing three-dimensional geometric processing; the actual X-ray penetration path length is the geometric distance that an X-ray photon actually travels inside the material from entering the material surface to exiting the lower surface of the material.

[0026] Specifically, firstly, through dual-laser differential and spatial methods, three-dimensional contour data that is strictly aligned with the X-ray image has been constructed, where x is the conveying direction coordinate and y is the belt width direction coordinate. The vertical height value corresponding to each pixel on the surface of the object being detected is extracted from the three-dimensional contour data, that is, the vertical distance from the laser emission point to the upper surface of the object being detected.

[0027] Secondly, the pixel positions with the same reference surface height value and vertical height value are extracted from the three-dimensional contour data to obtain the vertical distance from the laser emission point to the belt surface or idler roller reference surface, which is the reference surface height value of the conveyor belt.

[0028] Next, the vertical height value is subtracted from the reference surface height value of the conveyor belt pixel by pixel. For each pixel position, the vertical height value is subtracted from the reference surface height value to obtain the initial value of the vertical thickness of the object being detected. The initial value of the vertical thickness reflects the absolute thickness of the object being detected.

[0029] Simultaneously, local neighborhood point cloud analysis is performed on the 3D contour data. For each pixel in the 3D contour data, a local window is selected centered on that pixel, and all height values ​​within the local window are converted into 3D spatial point coordinates, forming a small point cloud. Subsequently, principal component analysis (PCA) or least squares plane modeling is used to fit the local neighborhood point cloud, obtaining the eigenvector corresponding to the minimum eigenvalue of each pixel on the surface of the detected object, which is the normal vector. Finally, the angle between the normal vector and the vertical direction is calculated to obtain the local tilt angle of the material surface at each pixel.

[0030] Specifically, the 3D contour data is treated as a local neighborhood point cloud. For each pixel (x, y), a local window is extracted. Plane fitting is performed on the 3D coordinates (Xi, Yi, Zi) of all points within the local window to obtain the equation of the best-fit plane: Ax + By + Cz + D = 0, resulting in the normal vector n = (A, B, C). The local tilt angle θ = arccos(|C| / sqrt(A)) is then calculated by performing a dot product operation between the normal vector and the vertical unit vector v = (0, 0, 1). 2 +B 2 +C 2 )).

[0031] For example, taking pixel (5000, 512) as the center, and taking 9 points including the pixel itself and its surrounding 3×3 neighborhood, we convert the pixel coordinates of each point into physical coordinates. We then perform plane fitting on these 9 three-dimensional coordinates to obtain the normal vector n = (0.05, 0.12, 0.99). We calculate the tilt angle: θ = arccos(0.99 / sqrt(0.05)). 2 +0.12 2 +0.99 2 ))≈5.7°.

[0032] Finally, based on the local tilt angle, the initial vertical thickness is corrected to the actual X-ray penetration path length. When X-rays are incident vertically, the penetration path is the initial vertical thickness. When the material surface is tilted, the path of X-rays inside the material extends along the normal direction of the tilted surface. For each pixel, the initial vertical thickness is calculated by dividing it by the cosine of the local tilt angle to obtain the actual X-ray penetration path length: actual X-ray penetration path length = initial vertical thickness / cosine of the local tilt angle. If the initial vertical thickness is d, then the actual X-ray penetration path length = d / cosθ. Finally, the actual X-ray penetration path length is used as the absolute thickness value.

[0033] In this embodiment, a pixel-level index relationship is established between 3D contour data and X-ray image data to directly obtain the vertical height value corresponding to each pixel on the surface of the detected object from the 3D image, laying a reliable data foundation for subsequent high-precision thickness compensation. Subsequently, through local neighborhood point cloud analysis, the normal vector and local tilt angle of the local neighborhood point cloud are calculated to achieve a preliminary calculation of the penetration path length of each pixel. Then, through the local tilt angle correction of the normal vector, the vertically measured thickness is corrected to obtain the true penetration path length, so that the absolute thickness value is completely matched with the physical model of X-ray attenuation, which effectively corrects the subsequent grayscale compensation and improves the accuracy of foreign object identification.

[0034] S30: Based on the absolute thickness value, perform thickness effect compensation processing on the low-energy X-ray image and the high-energy X-ray image respectively to generate a compensated low-energy X-ray image and a compensated high-energy X-ray image; In this embodiment, the thickness effect is the phenomenon that when X-rays penetrate a material, there is a difference in grayscale values ​​between thick and thin materials of the same material. The thickness effect compensation process maps the grayscale values ​​under different thicknesses to the equivalent grayscale values ​​under a certain reference thickness, thereby eliminating the interference of thickness variables on the signal. The compensated X-ray image is an image in which the grayscale value is related to the equivalent atomic number after thickness correction.

[0035] Specifically, for the current coal type being selected, low-energy and high-energy grayscale responses of different thicknesses under static conditions are prepared, and low-energy and high-energy calibration functions are obtained through curve fitting. During compensation processing, a reference thickness is first set according to process requirements, and the absolute thickness value is compensated to obtain the compensated image.

[0036] Step S30 in the method provided in this application embodiment includes: A thickness-grayscale calibration curve library based on the currently selected coal type is pre-constructed, wherein the thickness-grayscale calibration curve library includes low-energy channel grayscale response curves and high-energy channel grayscale response curves of different coal types at different thicknesses; Based on the coal type identifier of the selected coal type, the corresponding low-energy calibration curve function and high-energy calibration curve function are loaded from the thickness-gray scale calibration curve library, and the reference thickness value of the selected coal type is set, wherein the reference thickness value is configured according to the median particle size distribution of the selected coal type material. For each pixel of the detected object, based on the absolute thickness value, the theoretical low-energy grayscale value and the theoretical high-energy grayscale value corresponding to the absolute thickness value are queried through the low-energy calibration curve function and the high-energy calibration curve function, respectively. Calculate the low-energy compensation coefficient, wherein the low-energy compensation coefficient is the ratio of the theoretical low-energy gray value under the reference thickness value to the theoretical low-energy gray value under the current absolute thickness value; Calculate the high-energy compensation coefficient, wherein the high-energy compensation coefficient is the ratio of the theoretical high-energy gray value under the reference thickness value to the theoretical high-energy gray value under the current absolute thickness value; The low-energy compensation coefficient is multiplied by the pixel grayscale value of the original low-energy X-ray image to generate the compensated low-energy X-ray image. The compensated high-energy X-ray image is generated by multiplying the high-energy compensation coefficient by the pixel grayscale value of the original high-energy X-ray image.

[0037] In this embodiment, the current coal type being selected is the category of raw coal being processed in the current production batch of the coal preparation plant; the coal type identifier is a code or character that identifies the coal type information, which is usually entered by the operator on the system's human-machine interface or automatically issued by the host system.

[0038] Specifically, firstly, before formally conducting foreign object identification, a thickness-grayscale calibration curve library for the currently selected coal type is established through specialized calibration experiments. This library includes low-energy channel grayscale response curves and high-energy channel grayscale response curves. The low-energy channel grayscale response curve is a function curve showing the relationship between the grayscale value recorded by the detector and the penetration thickness after X-rays penetrate standard test blocks of the coal type of different thicknesses at a fixed tube voltage. Due to the beam hardening effect, it usually exhibits a non-linear attenuation trend. The high-energy channel grayscale response curve is a grayscale-thickness relationship curve measured under similar conditions at a higher tube voltage.

[0039] Furthermore, based on the coal type identification being processed, the corresponding calibration curve function is loaded from the thickness-ash-scale calibration curve library. A baseline thickness value is set as the reference standard thickness value during the thickness compensation process. Through the baseline thickness value, the ash-scale values ​​of all materials with different thicknesses can be normalized to the theoretical ash-scale level, i.e., the median of common particle sizes for coal materials. Simultaneously, historical particle size analysis data or online particle size monitoring results for the coal type are used.

[0040] Furthermore, the system iterates through each pixel position of the detected object and reads the corresponding stored absolute thickness value. Then, it substitutes the absolute thickness value into the low-energy calibration curve function to calculate the theoretical low-energy grayscale value; and substitutes it into the high-energy calibration curve function to calculate the theoretical high-energy grayscale value. If the material at the pixel position is pure coal with the same composition as the calibrated coal type, the expected grayscale level range under the current measured thickness can be expressed as: [theoretical low-energy grayscale value - theoretical high-energy grayscale value]. The theoretical low-energy grayscale value and the theoretical high-energy grayscale value are used as reference benchmarks for subsequent compensation coefficient calculations.

[0041] Furthermore, the low-energy compensation coefficient is calculated: for each pixel, the ratio of the theoretical low-energy grayscale value under the calculated reference thickness value to the theoretical low-energy grayscale value under the current absolute thickness value is used as the low-energy compensation coefficient for that pixel.

[0042] For example, assuming a reference thickness d0 = 45 mm, substituting into the low-energy calibration curve function: f L =0.0012×91125-0.35×2025-12.5×45+3100≈1938, current pixel absolute thickness d1=35.38mm, theoretical low-energy grayscale value G L =2273, low-energy compensation coefficient k L =1938 / 2273≈0.85.

[0043] Furthermore, the high-energy compensation coefficient is calculated: for each pixel, the ratio of the theoretical high-energy grayscale value under the reference thickness value to the theoretical high-energy grayscale value under the current absolute thickness value is used as the high-energy compensation coefficient for that pixel.

[0044] For example, assuming a reference thickness d0 = 45 mm, substituting it into the high-energy calibration curve function f H =1700, obtaining the current pixel absolute thickness d2=35.38mm, theoretical high-energy grayscale value G H =1950, High-energy compensation coefficient k H =1700 / 1950≈0.8718.

[0045] Furthermore, each pixel of the original low-energy image is traversed, the original grayscale value is read, and multiplied by the low-energy compensation coefficient of the corresponding pixel. The product result is then written into the output image buffer as the grayscale value of the corresponding pixel in the new image. Finally, the processing of the entire image is completed, and a compensated low-energy X-ray image is obtained.

[0046] For example, in the original low-energy image, the grayscale value of pixel (5000, 512) is 2150, and the low-energy compensation coefficient is 0.8526. The compensated grayscale value = 2150 × 0.8526 = 1833. Compensation calculations are performed on all pixels of the original low-energy image to obtain the compensated low-energy X-ray image.

[0047] Finally, each pixel of the original high-energy image is traversed, the original grayscale value of the pixel is read, and multiplied by the high-energy compensation coefficient of the corresponding pixel. The product result is then written into the output image buffer as the grayscale value of the corresponding pixel in the new image. Finally, the processing of the entire image is completed, and a compensated high-energy X-ray image is obtained.

[0048] For example, in the original high-energy image, the grayscale value of pixel (5000, 512) is 1820. The high-energy compensation coefficient is 0.87, and the compensated grayscale value = 1820 × 0.8718 = 1587.

[0049] In step S30 of the method provided in this application embodiment, a thickness-grayscale calibration curve library based on the current selected coal type is pre-constructed, including: Prepare standard stepped test blocks with atomic number characteristics similar to the currently selected coal type, wherein the standard stepped test blocks contain no less than five continuously increasing thickness levels, the thickness levels covering the minimum to maximum particle size of the selected coal material; The standard stepped test block was placed at the X-ray detection center line, and the grayscale response values ​​of the low-energy channel and the high-energy channel were collected at different thickness levels. For each thickness level, continuous acquisition of no less than one thousand scan lines is performed, and the arithmetic mean is taken as the grayscale response characteristic value of that thickness level. Curve fitting is performed on the thickness-grayscale discrete data points of the low-energy channel to generate the low-energy calibration curve function; Curve fitting is performed on the thickness-grayscale discrete data points of the high-energy channel to generate the high-energy calibration curve function; The low-energy calibration curve function and the high-energy calibration curve function are associated and bound with the current coal type identifier and stored in the thickness-grayscale calibration curve library.

[0050] In this embodiment, the X-ray detection centerline position is the projection position of the line connecting the X-ray source focus and the detector array center point on the belt plane; the association binding is to establish a corresponding mapping relationship between the coal type identifier and the parameter sets of the two calibration curve functions, so that the calibration curve exclusive to the coal type can be uniquely retrieved through the coal type identifier; the grayscale response characteristic value is obtained after statistical averaging and has a high signal-to-noise ratio representative grayscale value.

[0051] Specifically, firstly, standard stepped test blocks with specific geometric dimensions and physical properties, and with equivalent atomic number characteristics similar to the selected coal type, are prepared using physical or chemical methods. Each standard stepped test block contains no fewer than five continuously increasing thickness levels, with thickness values ​​continuously distributed from small to large. The thickness levels cover the minimum to maximum particle size of the selected coal material. Five thickness levels are the minimum requirement for nonlinear curve fitting. The more points, the higher the fitting accuracy, covering the particle size range of the actual coal material to ensure that the calibration curve is effective throughout the entire working range.

[0052] For example, assuming the coal being processed has a particle size range of 10–70 mm and a D50 of 45 mm, representative raw coal is selected, crushed, and ground to obtain fine coal powder, which is then used to prepare stepped test blocks. For instance, test blocks with thicknesses of 10 mm, 20 mm, 30 mm, 40 mm, and 50 mm are prepared.

[0053] Next, stepped test blocks of different thicknesses are placed individually in the X-ray optical path, with exposure and acquisition performed on only one known thickness at a time to avoid interference between different thickness regions. Each thickness level of test block is placed under the X-ray gantry to completely cover the central field of view of the X-ray detector. The X-ray source is activated, and exposure is performed according to preset low-energy and high-energy conditions. The detector outputs one or more frames of corresponding low-energy and high-energy images. Finally, pixels in the projected area of ​​the test block in the image are selected, and the average pixel grayscale value is calculated to obtain the low-energy channel grayscale response value and high-energy channel grayscale response value for that thickness.

[0054] Next, for each thickness level, in X-ray scanning imaging mode, the detector outputs one line of image data each time it is triggered, continuously acquiring no less than 1000 lines of independent image data. Since the test block itself does not move, no less than 1000 lines of independent image data reflect the repeated measurement results at the same thickness. The grayscale values ​​of all measurements are averaged pixel by pixel to obtain the grayscale feature value at the corresponding thickness.

[0055] Furthermore, the obtained low-energy data points are plotted in a two-dimensional coordinate system, with thickness as the horizontal axis and grayscale value as the vertical axis. As the thickness increases, the grayscale value decreases monotonically, and the decay rate gradually slows down. A function form capable of representing nonlinear decay is chosen, such as a quadratic or cubic polynomial. The polynomial coefficients are solved using the least squares method to minimize the sum of squared errors between the fitted curve and each data point. Finally, a low-energy calibration curve function f, for example, a cubic polynomial, is obtained. L =a·d 3 +b·d 2 +c·d+e, and save the coefficients a, b, c, e as parameters of the low-energy calibration curve function.

[0056] Simultaneously, the same method was used to perform curve fitting on the high-energy data points, and the resulting high-energy data points were plotted in a two-dimensional coordinate system with thickness as the horizontal axis and grayscale value as the vertical axis. The polynomial coefficients were solved using the least squares method to minimize the sum of squared errors between the fitted curve and each data point, thus obtaining the high-energy calibration curve function f. H =gd 3 -hd 2 -md+n, through a high-energy calibration curve function, describes the change in grayscale value of a high-energy channel with thickness. Because high-energy rays have stronger penetrating power, their grayscale attenuation curve is usually flatter than that of low-energy rays.

[0057] Finally, a mapping relationship is established between the current coal type identifier, low-energy curve function coefficient, and high-energy curve function: [current coal type identifier → low-energy curve function coefficient → high-energy curve function]. This relationship is then stored as a set of data, written to a storage medium, and stored in the thickness-grayscale calibration curve library.

[0058] In step S30 of the method provided in this application embodiment, the synchronous acquisition of the three-dimensional contour data of the detected object through the laser contour measuring device further includes a spatiotemporal synchronization error compensation step: An incremental encoder is installed at the driven pulley shaft end of the conveyor belt, wherein the incremental encoder outputs a pulse signal for each fixed rotation angle; The pulse signal of the incremental encoder is used as the main trigger source to simultaneously trigger the first laser profilometer, the second laser profilometer, and the X-ray detector to perform line scanning acquisition, ensuring that the three sensors acquire material information from the same spatial cross section; A laser Doppler velocimeter is installed on the gantry to measure the actual running speed of the conveyor belt in real time. The speed calculated by the incremental encoder is compared with the actual running speed measured by the laser Doppler velocimeter in real time to calculate the belt slippage rate. When the belt slippage rate exceeds a preset threshold, it automatically switches to software resampling compensation mode. In software resampling compensation mode, the acquisition timestamp of each frame of X-ray image and each frame of laser contour is recorded. The spatial position of the image sequence is re-interpolated according to the actual running speed to reconstruct the low-energy X-ray image and the high-energy X-ray image that are strictly aligned with the three-dimensional contour data.

[0059] In this embodiment, the conveyor belt is a ring-shaped rubber belt in a coal mine dry separator responsible for carrying and transporting raw coal and foreign objects; the driven wheel shaft end is the end of the roller shaft at the tail of the conveyor belt where no power is applied; the main trigger source refers to the signal source that serves as the common time reference for the acquisition actions of all devices in a multi-sensor collaborative acquisition system; the laser Doppler velocimeter is a non-contact speed measurement device based on the laser Doppler effect.

[0060] Specifically, firstly, encoder pulses are used as the trigger source to achieve synchronous acquisition between the laser profilometer and the X-ray detector. An incremental encoder is first installed on the driven pulley shaft. When the belt is running, the driven pulley shaft rotates with the belt, and the encoder outputs a pulse every time it rotates a certain angle. The belt displacement corresponding to each pulse is output as a pulse signal.

[0061] Secondly, the pulse signal from the incremental encoder is used as the main trigger source, and the pulse signal from the driven wheel encoder is connected to the trigger input terminals of the first laser profilometer, the second laser profilometer, and the X-ray detector respectively through a hardware splitter. When each pulse rising edge arrives, the three devices simultaneously start line scanning, obtaining material information from the same spatial cross-section collected by the three sensors.

[0062] Next, a laser Doppler velocimeter is installed on the gantry crane, with the laser beam projected vertically onto the area of ​​the belt surface without material coverage. The velocimeter then outputs the actual running speed of the belt at a certain encoder pulse interval. Simultaneously, the decoding speed of the incremental encoder and the actual running speed measured by the laser Doppler velocimeter are calculated. The ratio of the difference between the encoder pulse interval and the actual running speed to the actual running speed is taken as the belt slip rate, i.e., belt slip rate = [(encoder pulse interval - difference in actual running speed) / actual running speed].

[0063] For example, if the encoder's calculation speed is 1.85 m / s and the actual operating speed measured by the Doppler tachometer is 1.8 m / s, the belt slippage rate is (1.85 - 1.8) / 1.85 = 2.7%.

[0064] Ultimately, when the belt slippage rate exceeds a preset threshold, encoder triggering is insufficient to maintain sub-millimeter spatial alignment, necessitating software compensation. In software resampling compensation mode, the encoder pulse is abandoned as the sole spatial index; instead, spatial position reconstruction is performed based on high-precision timestamps and actual speed. First, the time value of each frame or line of contour data is acquired. Then, the actual operating speed measured by the laser Doppler velocimeter is used as a velocity function to integrate the spatial position of the acquired sequence. Through spatial position re-interpolation, the image sequence originally acquired at equal time intervals is resampled into a sequence arranged at equal spatial intervals. Low-energy X-ray and high-energy X-ray images with the same spatial coordinate system and sampling interval as the 3D contour data from the laser contour data are output, and the resulting new low-energy and high-energy X-ray images are perfectly aligned with the 3D contour data that has also undergone spatial position resampling.

[0065] In this embodiment, standard stepped test blocks with atomic number characteristics consistent with the selected coal type are prepared. A thickness-grayscale calibration curve function is established using a thousand-row statistical average and nonlinear curve fitting, providing a theoretical grayscale benchmark for thickness compensation. Subsequently, a thickness-grayscale calibration curve library is constructed to classify and store calibration models for multiple coal types. Grayscale levels are calculated for materials of arbitrary thickness, generating compensation coefficients to eliminate the interference of thickness on dual-energy grayscale while maintaining the mapping relationship between the dual-energy ratio and the equivalent atomic number. Furthermore, a laser Doppler velocimeter is used as an independent velocity benchmark for spatiotemporal registration. When the slippage rate exceeds the limit, timestamp re-interpolation is performed, and a dual-energy image strictly aligned with the three-dimensional contour is reconstructed based on the actual velocity. This ensures that the thickness compensation module obtains spatially aligned input data under any operating condition, completely eliminating the interference of thickness factors on dual-energy material identification.

[0066] S40: Input the compensated low-energy X-ray image and the compensated high-energy X-ray image into a pre-trained material recognition model, calculate the equivalent atomic number of the detected object, and determine whether the detected object is a foreign object based on the equivalent atomic number and a preset foreign object discrimination threshold.

[0067] In the embodiments of this application, the equivalent atomic number is a numerical value that indicates which atomic number the detected object is equivalent to in terms of X-ray attenuation characteristics.

[0068] Specifically, the thickness-compensated low-energy X-ray image and the compensated high-energy X-ray image are input into the material identification model. The material identification model calculates the equivalent atomic number of the detected object, and then calls the foreign object detection logic to determine if a foreign object is present. If a foreign object is detected, a shutdown alarm is triggered. Step S40 in the method provided in this application embodiment includes: The compensated low-energy X-ray image and the compensated high-energy X-ray image are registered at the pixel level to ensure that the same pixel position corresponds to the same spatial point of the detected object; Calculate the ratio of the gray value of the compensated low-energy X-ray image to the gray value of the compensated high-energy X-ray image at each pixel location, and generate a dual-energy ratio feature map; Extract the morphological features of the detected object, wherein the morphological features include area, perimeter, aspect ratio, and circularity; The dual-energy ratio feature map and the morphological features are fused to form a multi-dimensional feature vector, and the multi-dimensional feature vector is input into a pre-trained support vector machine classifier. The classifier outputs the probability distribution of the material category to which the detected object belongs and calculates the corresponding equivalent atomic number estimate as the equivalent atomic number of the detected object.

[0069] In this embodiment, firstly, for the compensated low-energy X-ray image and the compensated high-energy X-ray image, the same pixel points in the two images are precisely aligned to the same spatial coordinate position through the image processing algorithm, so that the pixels with the same coordinates in different images correspond to the exact same spatial point position of the detected object.

[0070] Specifically, using the low-energy compensation image as a reference, spatial transformation optimization is performed on the high-energy compensation image to construct a standardized cross-correlation similarity function. A quadratic interpolation model is iteratively calculated within the sub-pixel search space to solve for the translation parameters (Δx, Δy) that maximize the correlation between the two images, typically achieving an accuracy of 0.1 pixels. Then, bilinear interpolation is used to resample the high-energy compensation image, generating a new high-energy image that is strictly aligned with the low-energy compensation image.

[0071] Secondly, after thickness compensation and pixel-level registration, the dual-energy grayscale readings at the same pixel location are the grayscale values ​​of the compensated low-energy X-ray image and the compensated high-energy X-ray image. The ratio of the compensated low-energy X-ray image grayscale value to the compensated high-energy X-ray image grayscale value is calculated for each pixel location. The value is monotonically positively correlated with the equivalent atomic number of the material; the higher the atomic number, the stronger the low-energy attenuation relative to the high-energy attenuation, and the larger the ratio. The ratio calculation is performed for all pixels, resulting in a two-dimensional matrix composed of the ratios of the compensated low-energy X-ray image grayscale value to the compensated high-energy X-ray image grayscale value for all pixels. This matrix has the same size as the original image and reflects the material properties of the material.

[0072] Next, morphological features of the detected object are extracted. These features include area, perimeter, aspect ratio, and roundness. Area is the total number of pixels the target object occupies in the image, converted to pixel equivalents to obtain the physical area, reflecting the size of the material particles. Perimeter is the total length of the target object's outline boundary, reflecting the complexity of the material's edge. Aspect ratio is the ratio of the longer side to the shorter side of the target object's smallest bounding rectangle. Since coal blocks are usually irregularly shaped, their aspect ratio is close to 1-1.5; however, some foreign objects, such as anchor bolts and wires, can have aspect ratios exceeding 5, allowing for foreign object identification. Roundness can be calculated using the formula: 4π × area / perimeter. 2 The dimensionless parameter has a value range of (0,1). The closer the roundness is to 1, the closer the shape of the object is to a standard circle. Coal blocks, after long-term transportation and collision, have rounded edges and corners, resulting in relatively high roundness. In contrast, some foreign objects, after being crushed into gangue, have sharp edges and corners, resulting in lower roundness. By using the morphological characteristics of the detected object, coal blocks can be distinguished and identified from other foreign objects.

[0073] Finally, the dual-energy ratio feature map and morphological features are fused to form a multi-dimensional feature vector, which is then input into a pre-trained support vector machine classifier. The classifier outputs the probability distribution of the material category to which the detected object belongs and calculates the corresponding equivalent atomic number estimate as the equivalent atomic number of the detected object.

[0074] Specifically, through feature fusion, a multi-dimensional feature vector is constructed for each candidate object: First, statistical features of the R-value are extracted from the mask region of the object in the dual-energy ratio feature map, including the mean, standard deviation, minimum, and maximum values. Then, these features are concatenated with morphological features to form a multi-dimensional feature vector.

[0075] For example, the steps to construct a support vector machine classifier include: First, under the actual operating conditions of the coal mine dry separator, a large amount of material image data is simultaneously acquired through a dual-energy X-ray imaging system and a laser profilometry device, covering all possible material types that may appear in the coal being processed, including clean coal, gangue, wood, plastic, rubber, conveyor belt fragments, anchor bolts, bolts, wire, and other engineering foreign objects. After thickness compensation and dual-energy ratio calculation, the dual-energy ratio feature map of each sample object is obtained, and the pixel mask region of the object is extracted using an image segmentation algorithm. The statistical and morphological features of the dual-energy ratio are calculated for each sample object. Subsequently, standardization is performed using the Z-score method: the mean and standard deviation of each feature dimension on the training set are calculated, and the original feature value is subtracted from the mean of that dimension and divided by the standard deviation, so that each feature dimension follows a standard normal distribution with a mean of 0 and a variance of 1. Finally, process experts assign a real material category label to each sample, and the samples are randomly divided into a training set and a validation set at a ratio of 7:3.

[0076] A sequential minimum optimization algorithm is used iteratively to solve the problem. In each iteration, two Lagrange multipliers that most severely violate the KKT conditions are selected for analytical optimization until all multipliers meet the convergence condition. A combination of grid search and five-fold cross-validation is used to jointly optimize the parameters γ and C of the radial basis function kernel. The search range for γ is set to [0.001, 0.01, 0.1, 1, 10], and the search range for C is set to [0.1, 1, 10, 100, 1000], resulting in 25 parameter combinations. For each (γ, C) combination, the training set is randomly divided into 5 equal parts. Four parts are used for training and one part for validation in turn. The average of the five validation accuracies is used as the evaluation metric for that parameter combination. After iterative optimization, the parameter combination that maximizes the average cross-validation accuracy is selected as the final model parameters. For K classes, a total of K×(K-1) / 2 classifiers need to be trained. During prediction, all classifiers vote on the input sample, and the class with the highest vote is the final prediction result.

[0077] For example, for bolt foreign objects, the binary classifier voting results are: 13 votes for metal, 2 votes for gangue, and 0 votes for other categories. Therefore, the foreign object is determined to be of the metal category.

[0078] Step S40 in the method provided in this application embodiment further includes: An upper limit threshold for the equivalent atomic number of coal and a lower limit threshold threshold for the equivalent atomic number of gangue are preset, wherein the upper limit threshold for the equivalent atomic number of coal and the lower limit threshold threshold for the equivalent atomic number of gangue are configured according to the ash content and calorific value of the coal currently being selected. When the equivalent atomic number of the detected object is lower than the upper limit threshold of the equivalent atomic number of the coal body, the detected object is determined to be clean coal and the injection action is not performed. When the equivalent atomic number of the detected object is between the upper threshold of the equivalent atomic number of the coal body and the lower threshold of the equivalent atomic number of the gangue, the detected object is determined to be an intermediate product, and whether to perform the injection action is selected according to the coal preparation process requirements. When the equivalent atomic number of the detected object is lower than the lower limit threshold of the equivalent atomic number of the gangue, the detected object is determined to be gangue or high-density foreign matter, a spray valve control signal is generated, and a high-speed jet actuator is driven to blow the foreign matter away from the original conveying trajectory, so that the foreign matter enters the selected gangue flow. When the equivalent atomic number of the detected object is higher than the preset metal foreign object threshold, the detected object is determined to be a metal foreign object, and an emergency shutdown alarm signal is generated.

[0079] In this embodiment, before online foreign object identification, the upper limit threshold of the media equivalent atomic number and the lower limit threshold of the gangue equivalent atomic number are set through offline analysis or process parameter configuration. These thresholds can be configured based on the ash content and calorific value of the selected coal type. The ash content is the mass percentage of non-combustible minerals in the coal; the higher the ash content, the larger the equivalent atomic number of the coal. The calorific value is the heat released by the complete combustion of a unit mass of coal; the calorific value is positively correlated with the carbon content and negatively correlated with the ash content; the lower the calorific value, the higher the equivalent atomic number.

[0080] Secondly, when the equivalent atomic number of the detected object is lower than the upper limit threshold of the equivalent atomic number of the coal body, the detected object can be determined to be clean coal. The spray valve control system receives the instruction and does not generate any spray valve control signal for this object, allowing it to enter the clean coal collection bin along the original conveying trajectory.

[0081] Secondly, when the equivalent atomic number of the tested object falls between the upper threshold of the equivalent atomic number of coal and the lower threshold of the equivalent atomic number of gangue, the material within this range does not meet the purity standard of clean coal or the typical density of gangue. Therefore, the tested object is determined to be an intermediate product, i.e., an intergrowth of coal and gangue, or low-quality coal with high ash content but still possessing a certain calorific value. The decision to perform pulse-jet injection is based on the coal preparation process requirements, including full pulse-jet injection or no pulse-jet injection.

[0082] Simultaneously, when the equivalent atomic number of the detected object is lower than the lower limit threshold of the equivalent atomic number of gangue, it indicates that the atomic number of the material has reached or exceeded the typical lower limit of gangue, and it has the characteristics of high density and high ash content. The detected object is determined to be gangue or high-density foreign matter, and a spray valve control signal is generated, including control instructions for the spray valve number, opening delay, and opening duration. The control instructions are sent to the spray valve drive controller, which drives the solenoid valve to open. Compressed air is instantly ejected through the nozzle, forming a high-speed airflow. The airflow acts on the side of the foreign matter, changing the direction of the foreign matter's movement, causing it to deviate from its original parabolic falling trajectory. The gangue and foreign matter separated by the spray fall into a special gangue collection chute or gangue bin, separating from the clean coal product.

[0083] Finally, when the equivalent atomic number of the detected object is higher than the preset metal foreign object threshold, the detected object is determined to be a metal foreign object, and an emergency shutdown alarm signal is generated.

[0084] Specifically, the preset metal foreign object threshold is a fixed value much higher than the gangue discrimination threshold, usually set to 18-20, which is the equivalent atomic number of common metals. Setting the threshold to 20 can cover most non-ferrous and ferrous metals. When the equivalent atomic number of the detected object is higher than the preset metal foreign object threshold, the audible and visual alarm is activated, issuing an alarm on-site, or the dry separator is activated by an emergency stop signal, cutting off the feed, stopping the conveyor belt, and shutting off the X-ray source to prevent metal foreign objects from entering downstream equipment and causing major accidents.

[0085] For example, assuming the calculated equivalent atomic number is 24.5, which far exceeds the metal foreign object threshold of 20, an emergency shutdown alarm will be triggered.

[0086] In step S40 of the method provided in this application embodiment, a spray valve control signal is generated, and a high-speed jet actuator is driven to blow the foreign object away from the original conveying trajectory, so that the foreign object enters the selected gangue flow, including: The centroid coordinates and lateral distribution position on the conveyor belt of the detected object are calculated based on the three-dimensional contour data. The time window for the detected object to move to the spray valve execution area is calculated based on the conveyor belt running speed and the encoder pulse accumulation value. Based on the lateral distribution position of the detected object, the corresponding target spray valve number is mapped from the spray valve array; Calculate the valve opening delay and opening duration, wherein the valve opening delay is corrected based on the time window and system response delay; Generate a valve control command that includes the target valve number, valve opening delay, and opening duration. Send the valve control command to the valve drive controller to drive the corresponding valve to open instantaneously when the foreign object reaches the blowing position, so as to accurately blow the foreign object away from the original conveying trajectory and fall into the sorted gangue collection bin.

[0087] In this embodiment of the application, firstly, the coordinates of the center of mass of the detected object in three-dimensional space are calculated based on the three-dimensional contour data, as well as the coordinate range and distribution characteristics of the detected object in the width direction of the belt. This typically includes the Y coordinate of the center of mass, the Y coordinates of the left and right edges of the object, and the span of the object in the width direction.

[0088] Secondly, based on the conveyor belt speed and the encoder pulse accumulation value, the time window for the detected object to move to the spray valve execution area is calculated. Given the current position of the detected object in the conveying direction, the fixed distance D between the center line of the spray valve execution area and the X-ray detection center line, and the real-time belt speed V, the time required for the object to move from its current position to the center line of the spray valve is t = (DX) / V, where X is the relative position of the current object to the detection center line.

[0089] Since the object being located continues to move forward on the belt as it passes through the X-ray detection area, by updating the encoder pulse accumulation value, when the object is identified as a foreign object, the time t = t0 + T to reach the center line of the spray valve is calculated based on the current time t0 and position X0, and the time window is obtained: [t0, t].

[0090] Secondly, the calibration table of the spray valve array is stored in advance, including the spray centerline coordinates of each spray valve in the belt width direction and its effective working width. Based on the lateral distribution position of the object being detected, the spray centerline coordinates and spray valve spacing are obtained, and the corresponding target spray valve number is mapped from the spray valve array.

[0091] Simultaneously, based on the time t when the foreign object reaches the centerline of the spray valve and the current system time t', the theoretical remaining time T' = t - t' is derived. Directly using the theoretical remaining time as the delay would cause a delay in the actual spraying time, because there is a fixed response delay td between the command being issued and the spray valve actually opening. The spray valve opening delay is calculated as: Td = T' - td. Subsequently, based on the lateral width ΔY of the foreign object and the belt speed V, the spray valve opening duration TS = k × (ΔY / V) + T0 is calculated, where k is the coverage coefficient (1.0~1.5) and T0 is the minimum pulse width.

[0092] Finally, a valve control command is generated, which includes the target valve number, valve opening delay, and opening duration. The valve control command is sent to the valve drive controller, which drives the corresponding valve to open the valve core instantaneously when the foreign object reaches the injection position, so that compressed air is sprayed out instantly, accurately blowing the foreign object away from the original conveying trajectory, ensuring that it falls into the designated gangue collection bin rather than the clean coal bin. Finally, the material stream that has been sorted and identified as gangue or foreign object and successfully separated is finally collected in the gangue bin or gangue conveying system.

[0093] In this embodiment, a thickness-compensated dual-energy image is used as input. Pixel-level registration ensures sub-pixel accuracy in calculating the dual-energy ratio. The dual-energy ratio feature map is then fused with morphological features such as area, perimeter, aspect ratio, and roundness in a multi-dimensional manner. A support vector machine is used as a classifier to output a continuous estimate of the equivalent atomic number of the detected object. Subsequently, based on the equivalent atomic number, foreign object discrimination is performed on the detected object, establishing a dynamic mapping relationship between coal quality indicators and discrimination thresholds, and dynamically adjusting the upper limit of coal mass and the lower limit of gangue. Then, based on the geometric information of the three-dimensional contour data, delay correction is performed to achieve precise pre-compensation for the physical injection timing at the time of command issuance. Matching of injection energy and foreign object size is achieved through the coupling calculation of lateral distribution width and belt speed. Finally, two special handling channels are set up: an independent category for intermediate products and an emergency shutdown channel for metal foreign objects. This enables accurate control of sorting density, establishes a injection safety protection mechanism for high-risk metal foreign objects, and eliminates the possibility of metal foreign objects entering downstream crushing equipment.

[0094] The application embodiments, through the above specific implementation methods, achieve the following technical effects: In this embodiment, firstly, a first laser profilometer and a second laser profilometer are installed in parallel on an X-ray gantry to arrange the acquisition device for collecting contour data. This parallel installation provides a structural foundation for subsequent synchronous data acquisition. Then, the first laser profilometer, the second laser profilometer, and the X-ray detector are simultaneously triggered by an encoder pulse trigger signal to achieve simultaneous acquisition of images with sub-millimeter spatial resolution in the spatial dimension, improving the system's real-time performance and reliability. Next, the real-time difference between the two profilometers is calculated based on the position information of the belt reference plane. This difference is then used to eliminate common-mode measurement errors, providing a correction basis for constructing three-dimensional contour data and improving the anti-interference capability of the data measurement. Finally, the difference is used to restore the true upper surface height of the detected object at the X-ray scan line position, generating three-dimensional contour data and providing reliable data input for subsequent operations.

[0095] Secondly, a pixel-level index relationship is established between 3D contour data and X-ray image data to directly obtain the vertical height value corresponding to each pixel on the surface of the detected object from the 3D image, laying a reliable data foundation for subsequent high-precision thickness compensation. Subsequently, through local neighborhood point cloud analysis, the normal vector and local tilt angle of the local neighborhood point cloud are calculated to achieve a preliminary calculation of the penetration path length of each pixel. Then, through the local tilt angle correction of the normal vector, the vertically measured thickness is corrected to obtain the true penetration path length, so that the absolute thickness value is completely matched with the physical model of X-ray attenuation, which effectively corrects the subsequent grayscale compensation and improves the accuracy of foreign object identification.

[0096] Secondly, by preparing standard stepped test blocks with atomic number characteristics consistent with the currently selected coal types, a thickness-grayscale calibration curve function was established using thousand-row statistical averaging and nonlinear curve fitting, providing a theoretical grayscale benchmark for thickness compensation. Subsequently, a thickness-grayscale calibration curve library was constructed to classify and store calibration models for multiple coal types. Grayscale levels were calculated for materials of arbitrary thickness, generating compensation coefficients to eliminate the interference of thickness on dual-energy grayscale while maintaining the mapping relationship between the dual-energy ratio and the equivalent atomic number. Furthermore, a laser Doppler velocimeter was used as an independent velocity benchmark for spatiotemporal registration. When the slippage rate exceeded the limit, timestamp re-interpolation was performed, and a dual-energy image strictly aligned with the three-dimensional contour was reconstructed based on the actual velocity. This ensured that the thickness compensation module obtained spatially aligned input data under any operating condition, completely eliminating the interference of thickness factors on dual-energy material identification.

[0097] Finally, using the thickness-compensated dual-energy image as input, pixel-level registration ensures sub-pixel accuracy in calculating the dual-energy ratio. The dual-energy ratio feature map is then fused with morphological features such as area, perimeter, aspect ratio, and roundness for multi-dimensional feature fusion. A support vector machine is used as a classifier to output continuous estimates of the equivalent atomic number of the detected object. Subsequently, based on the equivalent atomic number, foreign object discrimination is performed on the detected object, establishing a dynamic mapping relationship between coal quality indicators and discrimination thresholds, and dynamically adjusting the upper limit of coal mass and the lower limit of gangue. Then, based on the geometric information of the three-dimensional contour data, delay correction is performed to achieve precise pre-compensation for the physical injection timing at the command issuance time. Matching of injection energy and foreign object size is achieved through the coupling calculation of lateral distribution width and belt speed. Finally, a two-level special handling channel is set up: an independent category for intermediate products and an emergency shutdown channel for metal foreign objects. This enables accurate control of sorting density, establishes a injection safety protection mechanism for high-risk metal foreign objects, and eliminates the possibility of metal foreign objects entering downstream crushing equipment.

[0098] Example 2, as Figure 2 As shown, based on the same inventive concept as the dual-energy X-ray-based foreign object identification method for coal mine dry separators provided in Embodiment 1, this embodiment of the invention also provides a dual-energy X-ray-based foreign object identification system for coal mine dry separators, comprising: The contour data acquisition module 11 is used to acquire low-energy X-ray images and high-energy X-ray images of the object being detected, and to simultaneously acquire the three-dimensional contour data of the object being detected through a laser contour measurement device. The penetration thickness calculation module 12 is used to calculate the absolute thickness value of the detected object in the X-ray penetration direction based on the three-dimensional contour data. The penetration thickness compensation module 13 is used to perform thickness effect compensation processing on the low-energy X-ray image and the high-energy X-ray image respectively according to the absolute thickness value, and generate compensated low-energy X-ray image and compensated high-energy X-ray image. The material identification module 14 is used to input the compensated low-energy X-ray image and the compensated high-energy X-ray image into a pre-trained material identification model, calculate the equivalent atomic number of the detected object, and determine whether the detected object is a foreign object based on the equivalent atomic number and a preset foreign object discrimination threshold.

[0099] In one embodiment, the contour data acquisition module 11 is used for: A first laser profilometer and a second laser profilometer are installed in parallel on an X-ray gantry. The beam of the first laser profilometer is perpendicular to the surface of the conveyor belt and is used to scan the upper surface profile information of the object being inspected. The beam of the second laser profilometer is aligned with the reference surface of the conveyor belt or the reference surface of the idler roller and is used to measure the position information of the reference surface of the conveyor belt in real time. The encoder pulse trigger signal simultaneously triggers the first laser profilometer, the second laser profilometer, and the X-ray detector to achieve sub-millimeter-level spatial synchronous acquisition of the three-dimensional contour data, the low-energy X-ray image, and the high-energy X-ray image. Real-time differential calculation is performed on the upper surface contour information collected by the first laser profilometer and the position information of the conveyor belt reference surface collected by the second laser profilometer to eliminate common mode measurement error caused by conveyor belt jumping. Based on the data obtained from the differential operation, the true upper surface height of the detected object at the position of the X-ray scan line is reconstructed to generate the three-dimensional contour data.

[0100] In one embodiment, the penetration thickness calculation module 12 is used for: The vertical height value corresponding to each pixel on the surface of the detected object is extracted from the three-dimensional contour data, wherein the vertical height value is the vertical distance from the laser emission point to the upper surface of the detected object; Extract the height value of the conveyor belt reference surface corresponding to the same pixel position from the three-dimensional contour data, wherein the height value of the conveyor belt reference surface is the vertical distance from the laser emission point to the belt surface or the idler roller reference surface; The vertical height value is subtracted pixel by pixel from the height value of the conveyor belt reference surface to obtain the initial value of the vertical thickness of the detected object. Local neighborhood point cloud analysis is performed on the three-dimensional contour data to calculate the normal vector at each pixel on the surface of the detected object and determine the local tilt angle of the material surface at each pixel. Based on the local tilt angle, the initial vertical thickness value is corrected to the actual X-ray penetration path length, and the actual X-ray penetration path length is used as the absolute thickness value, wherein the actual X-ray penetration path length is equal to the initial vertical thickness value divided by the cosine of the local tilt angle.

[0101] In one embodiment, the penetration thickness compensation module 13 is used for: A thickness-grayscale calibration curve library based on the currently selected coal type is pre-constructed, wherein the thickness-grayscale calibration curve library includes low-energy channel grayscale response curves and high-energy channel grayscale response curves of different coal types at different thicknesses; Based on the coal type identifier of the selected coal type, the corresponding low-energy calibration curve function and high-energy calibration curve function are loaded from the thickness-gray scale calibration curve library, and the reference thickness value of the selected coal type is set, wherein the reference thickness value is configured according to the median particle size distribution of the selected coal type material. For each pixel of the detected object, based on the absolute thickness value, the theoretical low-energy grayscale value and the theoretical high-energy grayscale value corresponding to the absolute thickness value are queried through the low-energy calibration curve function and the high-energy calibration curve function, respectively. Calculate the low-energy compensation coefficient, wherein the low-energy compensation coefficient is the ratio of the theoretical low-energy gray value under the reference thickness value to the theoretical low-energy gray value under the current absolute thickness value; Calculate the high-energy compensation coefficient, wherein the high-energy compensation coefficient is the ratio of the theoretical high-energy gray value under the reference thickness value to the theoretical high-energy gray value under the current absolute thickness value; The low-energy compensation coefficient is multiplied by the pixel grayscale value of the original low-energy X-ray image to generate the compensated low-energy X-ray image. The compensated high-energy X-ray image is generated by multiplying the high-energy compensation coefficient by the pixel grayscale value of the original high-energy X-ray image.

[0102] This includes pre-constructing a thickness-grayscale calibration curve library based on the currently selected coal type, including: Prepare standard stepped test blocks with atomic number characteristics similar to the currently selected coal type, wherein the standard stepped test blocks contain no less than five continuously increasing thickness levels, the thickness levels covering the minimum to maximum particle size of the selected coal material; The standard stepped test block was placed at the X-ray detection center line, and the grayscale response values ​​of the low-energy channel and the high-energy channel were collected at different thickness levels. For each thickness level, continuous acquisition of no less than one thousand scan lines is performed, and the arithmetic mean is taken as the grayscale response characteristic value of that thickness level. Curve fitting is performed on the thickness-grayscale discrete data points of the low-energy channel to generate the low-energy calibration curve function; Curve fitting is performed on the thickness-grayscale discrete data points of the high-energy channel to generate the high-energy calibration curve function; The low-energy calibration curve function and the high-energy calibration curve function are associated and bound with the current coal type identifier and stored in the thickness-grayscale calibration curve library.

[0103] The step of synchronously acquiring the three-dimensional contour data of the detected object through a laser contour measurement device also includes a spatiotemporal synchronization error compensation step. An incremental encoder is installed at the driven pulley shaft end of the conveyor belt, wherein the incremental encoder outputs a pulse signal for each fixed rotation angle; The pulse signal of the incremental encoder is used as the main trigger source to simultaneously trigger the first laser profilometer, the second laser profilometer, and the X-ray detector to perform line scanning acquisition, ensuring that the three sensors acquire material information from the same spatial cross section; A laser Doppler velocimeter is installed on the gantry to measure the actual running speed of the conveyor belt in real time. The speed calculated by the incremental encoder is compared with the actual running speed measured by the laser Doppler velocimeter in real time to calculate the belt slippage rate. When the belt slippage rate exceeds a preset threshold, it automatically switches to software resampling compensation mode. In software resampling compensation mode, the acquisition timestamp of each frame of X-ray image and each frame of laser contour is recorded. The spatial position of the image sequence is re-interpolated according to the actual running speed to reconstruct the low-energy X-ray image and the high-energy X-ray image that are strictly aligned with the three-dimensional contour data.

[0104] In one embodiment, the substance identification module 14 is used for: The compensated low-energy X-ray image and the compensated high-energy X-ray image are registered at the pixel level to ensure that the same pixel position corresponds to the same spatial point of the detected object; Calculate the ratio of the gray value of the compensated low-energy X-ray image to the gray value of the compensated high-energy X-ray image at each pixel location, and generate a dual-energy ratio feature map; Extract the morphological features of the detected object, wherein the morphological features include area, perimeter, aspect ratio, and circularity; The dual-energy ratio feature map and the morphological features are fused to form a multi-dimensional feature vector, and the multi-dimensional feature vector is input into a pre-trained support vector machine classifier. The classifier outputs the probability distribution of the material category to which the detected object belongs and calculates the corresponding equivalent atomic number estimate as the equivalent atomic number of the detected object.

[0105] In one embodiment, the substance identification module 14 is used for: An upper limit threshold for the equivalent atomic number of coal and a lower limit threshold threshold for the equivalent atomic number of gangue are preset, wherein the upper limit threshold for the equivalent atomic number of coal and the lower limit threshold threshold for the equivalent atomic number of gangue are configured according to the ash content and calorific value of the coal currently being selected. When the equivalent atomic number of the detected object is lower than the upper limit threshold of the equivalent atomic number of the coal body, the detected object is determined to be clean coal and the injection action is not performed. When the equivalent atomic number of the detected object is between the upper threshold of the equivalent atomic number of the coal body and the lower threshold of the equivalent atomic number of the gangue, the detected object is determined to be an intermediate product, and whether to perform the injection action is selected according to the coal preparation process requirements. When the equivalent atomic number of the detected object is lower than the lower limit threshold of the equivalent atomic number of the gangue, the detected object is determined to be gangue or high-density foreign matter, a spray valve control signal is generated, and a high-speed jet actuator is driven to blow the foreign matter away from the original conveying trajectory, so that the foreign matter enters the selected gangue flow. When the equivalent atomic number of the detected object is higher than the preset metal foreign object threshold, the detected object is determined to be a metal foreign object, and an emergency shutdown alarm signal is generated.

[0106] The process includes generating a spray valve control signal and driving a high-speed jet actuator to blow foreign objects away from their original transport trajectory, allowing them to enter the selected gangue flow. The centroid coordinates and lateral distribution position on the conveyor belt of the detected object are calculated based on the three-dimensional contour data. The time window for the detected object to move to the spray valve execution area is calculated based on the conveyor belt running speed and the encoder pulse accumulation value. Based on the lateral distribution position of the detected object, the corresponding target spray valve number is mapped from the spray valve array; Calculate the valve opening delay and opening duration, wherein the valve opening delay is corrected based on the time window and system response delay; Generate a valve control command that includes the target valve number, valve opening delay, and opening duration. Send the valve control command to the valve drive controller to drive the corresponding valve to open instantaneously when the foreign object reaches the blowing position, so as to accurately blow the foreign object away from the original conveying trajectory and fall into the sorted gangue bin.

[0107] Compared to existing technologies, this method firstly uses a contour data acquisition module 11 to install a first laser profilometer and a second laser profilometer in parallel on an X-ray gantry, arranging the acquisition devices for collecting contour data. This parallel installation provides a structural foundation for subsequent synchronous data acquisition. Then, an encoder pulse trigger signal simultaneously triggers the first laser profilometer, the second laser profilometer, and the X-ray detector, enabling simultaneous acquisition of images with sub-millimeter spatial resolution in the spatial dimension, improving the system's real-time performance and reliability. Next, the real-time difference between the two profilometers is calculated based on the position information of the belt reference surface. This difference is then used to eliminate common-mode measurement errors, providing a correction basis for constructing three-dimensional contour data and improving the anti-interference capability of data measurement. Finally, the difference is used to restore the true upper surface height of the detected object at the X-ray scan line position, generating three-dimensional contour data and providing reliable data input for subsequent operations.

[0108] Secondly, through the penetration thickness calculation module 12, a pixel-level index relationship is established between the three-dimensional contour data and the X-ray image data to directly obtain the vertical height value corresponding to each pixel on the surface of the detected object from the three-dimensional image, laying a reliable data foundation for subsequent high-precision thickness compensation. Subsequently, through local neighborhood point cloud analysis, the normal vector and local tilt angle of the local neighborhood point cloud are calculated to achieve a preliminary calculation of the penetration path length of each pixel. Then, through the local tilt angle correction of the normal vector, the vertically measured thickness is corrected to obtain the true penetration path length, so that the absolute thickness value is completely matched with the physical model of X-ray attenuation, which effectively corrects the subsequent grayscale compensation and improves the accuracy of foreign object identification.

[0109] Secondly, through the thickness compensation module 13, standard stepped test blocks with atomic number characteristics consistent with the currently selected coal type are prepared. Using a thousand-row statistical average and nonlinear curve fitting, a thickness-grayscale calibration curve function is established, providing a theoretical grayscale benchmark for thickness compensation. Subsequently, a thickness-grayscale calibration curve library is constructed to classify and store calibration models for multiple coal types. Grayscale levels are calculated for materials of arbitrary thickness, generating compensation coefficients to eliminate the interference of thickness on dual-energy grayscale while maintaining the mapping relationship between the dual-energy ratio and the equivalent atomic number. Furthermore, a laser Doppler velocimeter is used as an independent velocity benchmark for spatiotemporal registration. When the slippage rate exceeds the limit, timestamp re-interpolation is performed, and a dual-energy image strictly aligned with the three-dimensional contour is reconstructed based on the actual velocity. This ensures that the thickness compensation module obtains spatially aligned input data under any operating condition, completely eliminating the interference of thickness factors on dual-energy material identification.

[0110] Finally, through the material identification module 14, using the thickness-compensated dual-energy image as input, pixel-level registration ensures sub-pixel accuracy in calculating the dual-energy ratio. The dual-energy ratio feature map is then fused with morphological features such as area, perimeter, aspect ratio, and roundness in a multi-dimensional manner. A support vector machine is used as a classifier to output a continuous estimate of the equivalent atomic number of the detected object. Subsequently, based on the equivalent atomic number, foreign object discrimination is performed on the detected object, establishing a dynamic mapping relationship between coal quality indicators and discrimination thresholds, and dynamically adjusting the upper limit of coal mass and the lower limit of gangue. Then, based on the geometric information of the three-dimensional contour data, delay correction is performed to achieve precise pre-compensation for the physical injection timing at the command issuance time. Matching of injection energy and foreign object size is achieved through the coupling calculation of lateral distribution width and belt speed. Finally, a two-level special handling channel is set up: an independent category for intermediate products and an emergency shutdown channel for metal foreign objects. This enables accurate control of sorting density, establishes a injection safety protection mechanism for high-risk metal foreign objects, and prevents the possibility of metal foreign objects entering downstream crushing equipment.

Claims

1. A method for foreign object identification in coal mine dry separators based on dual-energy X-ray diffraction, characterized in that, The methods include: The system acquires low-energy X-ray images and high-energy X-ray images of the object being detected, and simultaneously acquires the three-dimensional contour data of the object being detected through a laser contour measurement device. The absolute thickness of the detected object in the X-ray penetration direction is calculated based on the three-dimensional contour data. Based on the absolute thickness value, thickness effect compensation processing is performed on the low-energy X-ray image and the high-energy X-ray image respectively to generate compensated low-energy X-ray images and compensated high-energy X-ray images; The compensated low-energy X-ray image and the compensated high-energy X-ray image are input into a pre-trained material recognition model to calculate the equivalent atomic number of the detected object. The model then determines whether the detected object is a foreign object based on the equivalent atomic number and a preset foreign object discrimination threshold. The three-dimensional contour data of the object being detected is acquired synchronously using a laser contour measurement device, including: A first laser profilometer and a second laser profilometer are installed in parallel on an X-ray gantry. The beam of the first laser profilometer is perpendicular to the surface of the conveyor belt and is used to scan the upper surface profile information of the object being inspected. The beam of the second laser profilometer is aligned with the reference surface of the conveyor belt or the reference surface of the idler roller and is used to measure the position information of the reference surface of the conveyor belt in real time. The encoder pulse trigger signal simultaneously triggers the first laser profilometer, the second laser profilometer, and the X-ray detector to achieve sub-millimeter-level spatial synchronous acquisition of the three-dimensional contour data, the low-energy X-ray image, and the high-energy X-ray image. Real-time differential calculation is performed on the upper surface contour information collected by the first laser profilometer and the position information of the conveyor belt reference surface collected by the second laser profilometer to eliminate common mode measurement error caused by conveyor belt jumping. Based on the data after differential operation, the true upper surface height of the detected object at the position of the X-ray scan line is reconstructed to generate the three-dimensional contour data. Calculating the absolute thickness of the detected object in the X-ray penetration direction based on the three-dimensional contour data includes: The vertical height value corresponding to each pixel on the surface of the detected object is extracted from the three-dimensional contour data, wherein the vertical height value is the vertical distance from the laser emission point to the upper surface of the detected object; Extract the height value of the conveyor belt reference surface corresponding to the same pixel position from the three-dimensional contour data, wherein the height value of the conveyor belt reference surface is the vertical distance from the laser emission point to the belt surface or the idler roller reference surface; The vertical height value is subtracted pixel by pixel from the height value of the conveyor belt reference surface to obtain the initial value of the vertical thickness of the detected object. Local neighborhood point cloud analysis is performed on the three-dimensional contour data to calculate the normal vector at each pixel on the surface of the detected object and determine the local tilt angle of the material surface at each pixel. Based on the local tilt angle, the initial vertical thickness value is corrected to the actual X-ray penetration path length, and the actual X-ray penetration path length is used as the absolute thickness value, wherein the actual X-ray penetration path length is equal to the initial vertical thickness value divided by the cosine of the local tilt angle.

2. The method for foreign object identification in coal mine dry separators based on dual-energy X-rays according to claim 1, characterized in that, Based on the absolute thickness value, thickness effect compensation processing is performed on the low-energy X-ray image and the high-energy X-ray image respectively, including: A thickness-grayscale calibration curve library based on the currently selected coal type is pre-constructed, wherein the thickness-grayscale calibration curve library includes low-energy channel grayscale response curves and high-energy channel grayscale response curves of different coal types at different thicknesses; Based on the coal type identifier of the selected coal type, the corresponding low-energy calibration curve function and high-energy calibration curve function are loaded from the thickness-gray scale calibration curve library, and the reference thickness value of the selected coal type is set, wherein the reference thickness value is configured according to the median particle size distribution of the selected coal type material. For each pixel of the detected object, based on the absolute thickness value, the theoretical low-energy grayscale value and the theoretical high-energy grayscale value corresponding to the absolute thickness value are queried through the low-energy calibration curve function and the high-energy calibration curve function, respectively. Calculate the low-energy compensation coefficient, wherein the low-energy compensation coefficient is the ratio of the theoretical low-energy gray value under the reference thickness value to the theoretical low-energy gray value under the current absolute thickness value; Calculate the high-energy compensation coefficient, wherein the high-energy compensation coefficient is the ratio of the theoretical high-energy gray value under the reference thickness value to the theoretical high-energy gray value under the current absolute thickness value; The low-energy compensation coefficient is multiplied by the pixel grayscale value of the original low-energy X-ray image to generate the compensated low-energy X-ray image. The compensated high-energy X-ray image is generated by multiplying the high-energy compensation coefficient by the pixel grayscale value of the original high-energy X-ray image.

3. The method for foreign object identification in coal mine dry separators based on dual-energy X-rays according to claim 2, characterized in that, A pre-constructed library of thickness-grayscale calibration curves based on the currently selected coal type is included, including: Prepare standard stepped test blocks with atomic number characteristics similar to the currently selected coal type, wherein the standard stepped test blocks contain no less than five continuously increasing thickness levels, the thickness levels covering the minimum to maximum particle size of the selected coal material; The standard stepped test block was placed at the X-ray detection center line, and the grayscale response values ​​of the low-energy channel and the high-energy channel were collected at different thickness levels. For each thickness level, continuous acquisition of no less than one thousand scan lines is performed, and the arithmetic mean is taken as the grayscale response characteristic value of that thickness level. Curve fitting is performed on the thickness-grayscale discrete data points of the low-energy channel to generate the low-energy calibration curve function; Curve fitting is performed on the thickness-grayscale discrete data points of the high-energy channel to generate the high-energy calibration curve function; The low-energy calibration curve function and the high-energy calibration curve function are associated and bound with the current coal type identifier and stored in the thickness-grayscale calibration curve library.

4. The method for foreign object identification in coal mine dry separators based on dual-energy X-rays according to claim 1, characterized in that, The process of synchronously acquiring the three-dimensional contour data of the detected object through a laser contour measurement device also includes a spatiotemporal synchronization error compensation step: An incremental encoder is installed at the driven pulley shaft end of the conveyor belt, wherein the incremental encoder outputs a pulse signal for each fixed rotation angle; The pulse signal of the incremental encoder is used as the main trigger source to simultaneously trigger the first laser profilometer, the second laser profilometer, and the X-ray detector to perform line scanning acquisition, ensuring that the three sensors acquire material information from the same spatial cross section; A laser Doppler velocimeter is installed on the gantry to measure the actual running speed of the conveyor belt in real time. The speed calculated by the incremental encoder is compared with the actual running speed measured by the laser Doppler velocimeter in real time to calculate the belt slippage rate. When the belt slippage rate exceeds a preset threshold, it automatically switches to software resampling compensation mode. In software resampling compensation mode, the acquisition timestamp of each frame of X-ray image and each frame of laser contour is recorded. The spatial position of the image sequence is re-interpolated according to the actual running speed to reconstruct the low-energy X-ray image and the high-energy X-ray image that are strictly aligned with the three-dimensional contour data.

5. The method for foreign object identification in coal mine dry separators based on dual-energy X-rays according to claim 1, characterized in that, The compensated low-energy X-ray image and the compensated high-energy X-ray image are input into a pre-trained material recognition model to calculate the equivalent atomic number of the detected object, including: The compensated low-energy X-ray image and the compensated high-energy X-ray image are registered at the pixel level to ensure that the same pixel position corresponds to the same spatial point of the detected object; Calculate the ratio of the gray value of the compensated low-energy X-ray image to the gray value of the compensated high-energy X-ray image at each pixel location, and generate a dual-energy ratio feature map; Extract the morphological features of the detected object, wherein the morphological features include area, perimeter, aspect ratio, and circularity; The dual-energy ratio feature map and the morphological features are fused to form a multi-dimensional feature vector, and the multi-dimensional feature vector is input into a pre-trained support vector machine classifier. The classifier outputs the probability distribution of the material category to which the detected object belongs and calculates the corresponding equivalent atomic number estimate as the equivalent atomic number of the detected object.

6. The method for foreign object identification in coal mine dry separators based on dual-energy X-rays according to claim 1, characterized in that, Determining whether the detected object is a foreign object based on the equivalent atomic number and a preset foreign object discrimination threshold includes: An upper limit threshold for the equivalent atomic number of coal and a lower limit threshold threshold for the equivalent atomic number of gangue are preset, wherein the upper limit threshold for the equivalent atomic number of coal and the lower limit threshold threshold for the equivalent atomic number of gangue are configured according to the ash content and calorific value of the coal currently being selected. When the equivalent atomic number of the detected object is lower than the upper limit threshold of the equivalent atomic number of the coal body, the detected object is determined to be clean coal and the injection action is not performed. When the equivalent atomic number of the detected object is between the upper threshold of the equivalent atomic number of the coal body and the lower threshold of the equivalent atomic number of the gangue, the detected object is determined to be an intermediate product, and whether to perform the injection action is selected according to the coal preparation process requirements. When the equivalent atomic number of the detected object is lower than the lower limit threshold of the equivalent atomic number of the gangue, the detected object is determined to be gangue or high-density foreign matter, a spray valve control signal is generated, and a high-speed jet actuator is driven to blow the foreign matter away from the original conveying trajectory, so that the foreign matter enters the selected gangue flow. When the equivalent atomic number of the detected object is higher than the preset metal foreign object threshold, the detected object is determined to be a metal foreign object, and an emergency shutdown alarm signal is generated.

7. The method for foreign object identification in coal mine dry separators based on dual-energy X-rays according to claim 6, characterized in that, Generate a spray valve control signal and drive a high-speed jet actuator to blow foreign objects away from their original transport trajectory, allowing them to enter the sorted gangue flow, including: The centroid coordinates and lateral distribution position on the conveyor belt of the detected object are calculated based on the three-dimensional contour data. The time window for the detected object to move to the spray valve execution area is calculated based on the conveyor belt running speed and the encoder pulse accumulation value. Based on the lateral distribution position of the detected object, the corresponding target spray valve number is mapped from the spray valve array; Calculate the valve opening delay and opening duration, wherein the valve opening delay is corrected based on the time window and system response delay; Generate a valve control command that includes the target valve number, valve opening delay, and opening duration. Send the valve control command to the valve drive controller to drive the corresponding valve to open instantaneously when the foreign object reaches the blowing position, so as to accurately blow the foreign object away from the original conveying trajectory and fall into the sorted gangue collection bin.

8. A foreign object identification system for coal mine dry separators based on dual-energy X-rays, characterized in that, For implementing the dual-energy X-ray-based foreign object identification method for coal mine dry separators according to any one of claims 1-7, the system comprises: The contour data acquisition module is used to acquire low-energy X-ray images and high-energy X-ray images of the object being detected, and to simultaneously acquire the three-dimensional contour data of the object being detected through a laser contour measurement device. The penetration thickness calculation module is used to calculate the absolute thickness value of the detected object in the X-ray penetration direction based on the three-dimensional contour data. The penetration thickness compensation module is used to perform thickness effect compensation processing on the low-energy X-ray image and the high-energy X-ray image respectively according to the absolute thickness value, and generate compensated low-energy X-ray image and compensated high-energy X-ray image. The material identification module is used to input the compensated low-energy X-ray image and the compensated high-energy X-ray image into a pre-trained material identification model, calculate the equivalent atomic number of the detected object, and determine whether the detected object is a foreign object based on the equivalent atomic number and a preset foreign object discrimination threshold.

Images