Reference template construction method and system for roll mill roll face stud detection

Abstract

The present application belongs to the field of industrial measurement and machine vision, and relates to a reference template construction method and system for roll surface stud detection of a roller mill. The method comprises the following steps: S1, converting three-dimensional point cloud data collected by the roller mill roll surface into a depth image and a reference gray image used for stud target identification; S2, based on stud design parameters and arrangement rules, automatically identifying and positioning all studs on the reference gray image, determining the theoretical position and theoretical contour of each stud, and generating a stud position template image; S3, mapping the theoretical contour area of each stud in the template image back to the depth image, and calculating the statistical average value of the pixel depth values in each area as the stud reference height; S4, associating and storing the theoretical position, theoretical contour and reference height of each stud to form a structured reference template. The present application can efficiently process initial three-dimensional point cloud, automatically identify all studs and record their two-dimensional position and three-dimensional reference height, and generate a standardized template file.

Description

Technical Field

[0001] This invention belongs to the field of industrial measurement and machine vision technology, specifically relating to a method and system for constructing a reference template for detecting pins on the roller surface of a roller mill. Background Technology

[0002] The carbide studs on the roller surface of a roller mill are critical and vulnerable parts, and accurate and efficient inspection of their health is essential for equipment maintenance. Existing inspection methods, such as manual inspection or two-dimensional visual inspection, suffer from low efficiency, insufficient accuracy, or poor robustness. Intelligent inspection methods based on three-dimensional contour scanning are considered an effective solution, and the core premise of this method is having a precise reference template containing the initial position and three-dimensional shape information of the studs.

[0003] Currently, the creation of benchmark templates largely relies on manual point-by-point annotation or simple threshold segmentation based on two-dimensional grayscale images. The former is extremely inefficient and prone to subjective errors when dealing with thousands of regularly arranged pins on a roller surface; the latter, while capable of automatic identification, only obtains two-dimensional positional information, failing to acquire and correlate the initial three-dimensional height information of the pins, and is sensitive to image quality such as lighting and contrast, exhibiting poor robustness. This results in incomplete template information and limited accuracy, making it unreliable as a benchmark for subsequent precise wear quantification detection based on three-dimensional comparison. Summary of the Invention

[0004] This invention aims to overcome the shortcomings of existing technologies in terms of low automation and insufficient accuracy in creating reference templates, and provides a method and system for constructing reference templates for detecting pins on the surface of roller mills. This method can efficiently process initial 3D point clouds, automatically identify all pins and record their precise 2D positions and 3D reference heights, generating standardized template files.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A method for constructing a benchmark template for detecting pins on the roller surface of a roller mill includes: Step S1: Convert the complete three-dimensional point cloud data collected from the roller surface of the roller mill in its initial healthy state into a depth image reflecting height information and a reference grayscale image for pin target recognition. Step S2: Based on the known design parameters and arrangement rules of the roller mill pins, automatically identify and locate all pins on the reference grayscale image, determine the theoretical position and theoretical outline of each pin, and generate a pin position template image; Step S3: Map the theoretical contour area of ​​each post in the template image back to the depth image, and calculate the statistical average value of the pixel depth values ​​in each area as the reference height of the corresponding post. Step S4: Associate and store the theoretical position, theoretical outline, and reference height of each post pin to form a structured reference template.

[0006] Furthermore, in step S1, before the new roller is put into use while the roller surface is in its initial healthy state, the roller is controlled to rotate at a low speed for one revolution. The rotary encoder installed on the roller shaft is used to synchronously trigger the three-dimensional contour scanning device to collect high-precision three-dimensional point cloud data covering the entire roller surface as reference point cloud data. The reference point cloud data is preprocessed, including removing background interference points and outlier noise points through pass-through filtering and performing coordinate system normalization.

[0007] Further, in step S1, the preprocessed reference point cloud data is projected onto a two-dimensional plane corresponding to the unfolded outer cylindrical surface of the roller to generate a 32-bit floating-point depth image reflecting height information; by setting a two-dimensional preset mapping parameter corresponding to the height range of the pin, the depth value of the depth image is linearly mapped to 0-255 gray levels to generate an 8-bit reference gray level image to enhance the contrast of the pin target. The preset mapping parameters are used to linearly map the range of depth values ​​corresponding to the pin height in the depth image to the full dynamic range of the grayscale image.

[0008] Further, in step S2, the pin design parameters include at least the nominal diameter, row spacing, and column spacing of the pins; the arrangement rule is a uniform distribution in a row and column matrix; on the reference grayscale image, an image processing algorithm is used to automatically identify and locate all pin targets; the image processing algorithm includes adaptive threshold segmentation, morphological operations, connected component analysis, and circular contour detection.

[0009] Furthermore, in step S2, based on the known prior constraints of the row and column spacing of the posts, the center of the initially identified circular contour is meshed and corrected. All identified posts are numbered, and their pixel-level theoretical center coordinates and theoretical radius are recorded to determine the final theoretical position and theoretical contour area of ​​each post, and to correct false detections or missed detections.

[0010] Further, in step S3, the theoretical contour region of each post is used as a spatial index and mapped back to the 32-bit floating-point depth image generated in step S1. The statistical average of the depth values ​​of all valid pixels within the region is calculated as the reference average height value of the post.

[0011] Further, in step S3, before calculating the statistical average, the depth values ​​in the region are first subjected to median filtering, and outliers other than ±3σ standard deviation are removed; in step S4, the number of each post, the theoretical center coordinates, the theoretical profile parameters and the benchmark average height value calculated after filtering are associated and stored in the database or structured file to complete the construction of the structured benchmark template.

[0012] A benchmark template construction system for detecting pins on the roller surface of a roller mill includes: A 3D scanning unit is used to acquire 3D point cloud data of the initial roller surface; The image conversion and recognition unit is used to convert the three-dimensional point cloud data into a reference grayscale image and automatically identify and locate all the pins based on the pin design parameters. The data association and processing unit is used to map the identified pin outline to the depth image to calculate the reference height, and generate a structured reference template that associates and stores the location, outline and height information. A storage unit for storing the structured reference template.

[0013] Furthermore, the three-dimensional scanning unit includes a rotary encoder and a three-dimensional contour scanning device. The rotary encoder is mounted on the roller shaft and is used to synchronously trigger the three-dimensional contour scanning device to acquire high-precision three-dimensional point cloud data covering the entire roller surface when the roller rotates at low speed. The three-dimensional scanning unit also includes a preprocessing module, which is used to remove background interference points and outlier noise points through pass-through filtering and to perform coordinate system normalization. The image conversion and recognition unit includes a projection mapping module, a grayscale conversion module, a target initial detection module, and a grid correction module: The projection mapping module projects 3D point cloud data onto a 2D plane corresponding to the unfolded outer cylindrical surface of the roller, generating a 32-bit floating-point depth image; the grayscale conversion module linearly maps the depth values ​​in the depth image to 0-255 gray levels by setting a linear mapping interval corresponding to the nominal height range of the pins, generating an 8-bit reference grayscale image; the target initial detection module performs adaptive threshold segmentation, morphological operations, connected component analysis, and circular contour detection on the reference grayscale image to initially identify the pin targets; the grid correction module performs grid fitting and correction on the center of the initially identified circular contour based on the known prior constraints of the pin row and column spacing, determining and numbering the final theoretical position and theoretical contour region of each pin.

[0014] Furthermore, the data association and processing unit includes: The height calculation module is used to map the theoretical contour area of ​​each post back to the 32-bit floating-point depth image as a spatial index, perform median filtering on the depth values ​​of all valid pixels in the area, and calculate the statistical average value after removing outliers other than ±3σ standard deviation, which is used as the reference average height value of the post. The template generation module is used to associate and package the number, theoretical center coordinates, theoretical contour parameters and benchmark average height value of each post pin to generate the structured benchmark template and store it in a database or structured file.

[0015] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method described above.

[0016] A computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the above-described method.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: High degree of automation: It realizes fully automatic processing from raw point cloud to structured template without manual intervention, which greatly improves the efficiency of template construction.

[0018] High precision and complete information: The depth image is obtained by projecting a 3D point cloud, which ensures the homogeneity and accurate correspondence between the 2D position and 3D height information. The reference height value recorded in the template comes directly from 3D measurement, which is highly accurate.

[0019] High robustness: Combining grayscale image processing with mesh fitting based on prior knowledge, it can effectively cope with the situation where individual post pins are not ideally imaged, and accurately identify and locate all post pins.

[0020] Standardized output: The generated template file has a clear structure and contains multi-dimensional information such as position, outline, and height, providing a unified and reliable input for standardized subsequent inspection processes.

[0021] Supports intelligent detection: It provides an essential high-quality data foundation for subsequent advanced applications in this field, such as template-based online detection of pin wear and defect identification. Attached Figure Description

[0022] Figure 1 This is a flowchart of the baseline template construction method of the present invention.

[0023] Figure 2 This is a schematic diagram of an 8-bit reference grayscale image obtained by depth-to-grayscale mapping of the original roller surface point cloud.

[0024] Figure 3 This is a schematic diagram of the pin position template automatically generated by the method of the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0026] This invention provides a method and system for constructing a reference template for detecting pins on the surface of a roller mill. Under the initial healthy condition of a new roller, high-precision point clouds are acquired using 3D contour scanning and encoder synchronous triggering technology. The 3D data is then reduced to a 2D image domain through cylindrical surface unfolding projection to quickly identify the pins, and prior arrangement rules are used to correct the identification results. Finally, the reference height is calculated by backtracking to 3D depth data, thereby constructing a structured reference template containing position, contour, and height, providing a highly reliable comparison benchmark for subsequent wear quantification assessment.

[0027] Hardware deployment: A 3D line laser profile scanner is installed on the side of the roller mill. An incremental encoder is installed at the end of the roller shaft to determine the starting zero position of the roller rotation, and an absolute encoder is used to provide high-resolution angular position signals in real time to trigger the scan. The scanner is connected to a host computer.

[0028] Example like Figure 1-3 As shown, the reference template construction method for detecting pins on the roller surface of a roller mill in this embodiment includes the following steps.

[0029] Step S1: Convert the complete three-dimensional point cloud data collected from the roller surface of the roller mill in its initial healthy state into a depth image reflecting height information and a reference grayscale image for pin target recognition.

[0030] Specifically, this step includes the following sub-steps: Before a new roller is put into use, and with the roller surface in its initial healthy state, the roller is controlled to rotate at a low speed of 1-2 revolutions per minute for one revolution. An incremental encoder triggers the scanner at the zero position. A rotary encoder mounted on the roller shaft synchronously triggers the 3D contour scanning device, ensuring that a scan is triggered every time the roller rotates a small angle, thereby acquiring high-precision 3D point cloud data covering the entire roller surface as reference point cloud data. The rotary encoder is a combination of an incremental encoder and an absolute encoder, and the 3D contour scanning device can be a 3D line laser contour scanner. The reference point cloud data is preprocessed, including removing background interference points and outlier noise points through pass-through filtering, and performing coordinate system unification, unifying all point clouds to a roller coordinate system with the roller's central axis as the Z-axis.

[0031] The preprocessed reference point cloud data is projected onto a two-dimensional plane corresponding to the unfolded outer cylindrical surface of the roller, generating a 32-bit floating-point depth image reflecting height information. In this embodiment, the nominal radius of the roller can be set to R. For any point P(x, y, z) in the preprocessed point cloud, its unfolded coordinates are calculated: circumferential arc length. If the angle is negative, add 2πR to ensure continuity; the axial coordinate is V=z; calculate its radial depth value. In this embodiment, the physical resolution is set to 0.1 mm / pixel, U and V are mapped to the column and row coordinates of the image, the depth value D is filled into the corresponding details, and a 32-bit floating-point depth image is generated. At this time, the depth value is the actual physical offset value of the point relative to the roller reference surface.

[0032] By setting two-dimensional preset mapping parameters corresponding to the pin height range, the depth values ​​of the depth image are linearly mapped to gray levels of 0-255, generating an 8-bit reference grayscale image to enhance the contrast of the pin target. The preset mapping parameters are used to linearly map the depth value range corresponding to the pin height in the depth image to the full dynamic range of the grayscale image.

[0033] In this embodiment, a linear mapping interval is set based on the nominal protrusion height of the stud and the possible depressions in the substrate. The linear mapping interval [D] is defined. min D max ]. Through formula The 32-bit depth image was linearly mapped to a 0-255 grayscale level, and values ​​outside the range were truncated to generate an 8-bit baseline grayscale image. This image significantly enhances the contrast between the post and the substrate.

[0034] Step S2: Based on the known design parameters and arrangement rules of the roller mill pins, automatically identify and locate all pins on the reference grayscale image, determine the theoretical position and theoretical outline of each pin, and generate a pin position template diagram.

[0035] The pin design parameters include at least the nominal diameter, row spacing, and column spacing of the pins; the arrangement rule is a uniform distribution in a row and column matrix; on the reference grayscale image, an image processing algorithm is used to automatically identify and locate all pin targets; the image processing algorithm includes adaptive threshold segmentation, morphological operations, connected component analysis, and circular contour detection.

[0036] In step S2, based on the known prior constraints of the row and column spacing of the post, the center of the initially identified circular contour is meshed and corrected. All identified post are numbered and their pixel-level theoretical center coordinates and theoretical radius are recorded to determine the final theoretical position and theoretical contour area of ​​each post, and to correct false detections or missed detections.

[0037] Specifically, this step includes the following sub-steps: Preliminary identification and positioning: Obtain the known pin design parameters of the roller mill, including at least the nominal diameter d and row spacing L. r and column spacing L c The arrangement follows a uniform distribution in a row and column matrix. On an 8-bit baseline grayscale image, Gaussian weighted adaptive thresholding is used for local adaptive thresholding to separate the pins from the substrate. Morphological opening and closing operations are performed on the segmented binary image to remove noise caused by surface oil or speckle. Candidate target regions are initially extracted through connected component analysis, and the minimum circumcircle method of the Hough circular detection algorithm is used to initially fit the circular contours of each candidate target, obtaining preliminary center coordinates and radii.

[0038] Mesh Fitting and Correction: Based on the known prior constraints of the row and column spacing of the pins, a mesh fitting is performed on the centers of all initially fitted contours. Specifically, the RANSAC algorithm or the least squares method can be used to fit the initial group of center points with an ideal equidistant matrix mesh. Through fitting, isolated noise points that are falsely detected are eliminated, and missed points caused by reflection or occlusion are filled in. The pins at all mesh intersections are numbered in row and column order, such as P. ij And record its corrected pixel-level theoretical center coordinates (u i v j The theoretical radius r is usually taken as the pixel radius corresponding to the nominal diameter. This determines the theoretical position and theoretical contour area of ​​all the pins and generates a pin position template map.

[0039] Step S3: Map the theoretical contour area of ​​each post in the template image back into the depth image, and calculate the statistical average value of the pixel depth values ​​in each area as the reference height of the corresponding post.

[0040] In step S3, the theoretical contour region of each post is used as a spatial index and mapped back to the 32-bit floating-point depth image generated in step S1. The statistical average of the depth values ​​of all valid pixels within the region is calculated as the reference average height value of the post.

[0041] Before calculating the statistical average, the depth values ​​within the region are first filtered by median and outliers other than ±3σ standard deviation are removed.

[0042] Specifically, for each pin recorded in the template diagram i，j With its theoretical center (u i v j The circular mask, determined by the theoretical radius r, is used as a spatial index and mapped back to the 32-bit floating-point depth image generated in step S12. The set of all valid pixel depth values ​​within this mask region is then extracted. . Before calculating the statistical average, to eliminate the influence of laser edge flash and speckle noise, the depth value set is first subjected to median filtering, and its mean μ and standard deviation σ are calculated. Outliers other than those deviating from the mean ±3σ are removed. The arithmetic mean of the remaining effective depth values ​​is calculated as the reference average height H of the post. baseline .

[0043] Step S4: Associate and store the theoretical position, theoretical outline, and reference height of each post pin to form a structured reference template.

[0044] In step S4, the ID number of each post pin is... i,j Theoretical center coordinates (u) i v j The theoretical profile parameter radius r and its corresponding reference average height H baseline The data is packaged and stored in a database or structured file, such as ISON or XML format. The resulting structured benchmark model fully records the precise position and three-dimensional morphological benchmark of each stud in a healthy state.

[0045] This embodiment also provides a reference template construction system for detecting pins on the roller surface of a roller mill, including a three-dimensional scanning unit, an image conversion and recognition unit, a data association and processing unit, and a storage unit that are connected in sequence.

[0046] 3D scanning unit: includes a 3D line laser profile scanner mounted above the roller and a rotary encoder on the roller shaft, used to synchronously trigger the acquisition of high-precision 3D point cloud data covering the entire roller surface when the roller surface rotates at low speed initially; Image conversion and recognition unit: It is used to receive data from the 3D scanning unit, perform pass-through filtering to remove noise, generate 32-bit depth map and 8-bit grayscale map by cylindrical surface unfolding projection, and automatically identify and locate all pins on the grayscale map based on the pin design parameters and grid fitting algorithm, and output position template map; Data association and processing unit: used to map the pin outline in the location template map to a 32-bit depth map, execute a noise reduction algorithm and calculate the baseline average height of each pin; Storage unit: Used to receive the output of the data association and processing unit, and associate and store the pin number, position, outline and height data as JSON or database file to form a structured benchmark template.

[0047] The 3D scanning unit is responsible for acquiring and initially cleaning the original high-precision 3D data. It includes incremental and absolute encoders mounted on the roller shaft and a 3D contour scanning device mounted above the roller surface. When the new roller rotates at low speed in its initial healthy state, the rotary encoder outputs hardware trigger pulses at equal angular intervals, which synchronously triggers the 3D contour scanning device to acquire single-line contours, thereby ensuring that the splicing covers the entire roller surface without any omissions or overlapping distortions.

[0048] The preprocessing module receives the complete point cloud generated by the hardware triggering module, performs pass-through filtering to remove large-scale background interference points outside the roller, and uses statistical filtering to remove outlier noise points. Then, the denoised point cloud is unified to the roller coordinate system with the roller center axis as the Z-axis through coordinate transformation, and the reference point cloud data is output to the image conversion and recognition unit.

[0049] The image conversion and recognition unit receives preprocessed reference point cloud data, and its core functions include 3D-to-2D dimensionality reduction and precise target localization. The projection mapping module receives the reference point cloud, calculates the circumferential arc length U, axial coordinate V, and radial depth value D for each point. Using U as the x-coordinate, V as the y-coordinate, and D as the pixel value, it generates a 32-bit floating-point depth image reflecting the true physical height information.

[0050] The grayscale conversion module is synchronized with the projection mapping module. It sets the linear mapping range according to the nominal height range of the post, such as [-1mm, 6mm]. It linearly maps the floating-point depth value of the 32-bit depth map and truncates it to 0-255 grayscale levels to generate an 8-bit reference grayscale image with enhanced contrast.

[0051] The target initial detection module works on an 8-bit baseline grayscale image, sequentially performing local adaptive threshold segmentation, morphological opening and closing operations for noise reduction, connected component analysis to extract candidate regions, and using Hough circle detection to fit a preliminary circular contour center and radius.

[0052] The grid correction module receives the preliminary center coordinates output by the target initial detection module and, combined with the prior constraints of row and column spacing of the post input by the system, performs least-squares grid fitting on the preliminary center. Through fitting, false detection noise points are eliminated, missed centers are filled in, and the module outputs the number of all post, the corrected theoretical center coordinates, and the theoretical contour area, generating a post position template map.

[0053] The data association and processing unit is responsible for backtracking the 2D recognition results to 3D space and extracting high-precision morphological benchmarks.

[0054] The height calculation module receives the 2D theoretical contour mask output by the mesh correction module and the 32-bit floating-point depth image output by the projection mapping module. Using the theoretical contour mask as a spatial index, it extracts all valid pixel depth values ​​within the corresponding region of the 32-bit depth map. Before calculating the statistical average, the set is first smoothed by median filtering, and outliers deviating from the mean ± 3σ standard deviation are removed. Finally, the arithmetic mean of the remaining valid values ​​is calculated as the baseline average height H for the high noise resistance of the post. baseline .

[0055] The template generation module associates and packages the unique number of each post, the theoretical center coordinates and contour parameters from the mesh correction module, and the baseline average height from the height calculation module into a complete data record.

[0056] The storage unit receives the data record set output by the template generation module, and writes it into a structured reference template that can be directly called in the subsequent wear detection stage, either in a structured format such as JSON or XML files or into a database.

[0057] In the system described in this embodiment of the invention, the hardware triggering module in the 3D scanning unit can use a commercially available incremental rotary encoder and a 3D line laser contour sensor, achieving hardware-level synchronous triggering through a PLC or data acquisition card; the preprocessing module and image conversion and recognition unit can be implemented using C++ language combined with open-source point cloud libraries and open-source computer vision libraries, where the pass-through filtering calls the PCL correlation filter, and image segmentation and Hough circle detection call the corresponding OpenCV operators; the mesh correction module implements least squares matrix fitting based on the Eigen library; the mask clipping and statistical calculation in the data association and processing unit are also implemented based on the OpenCV and Eigen libraries, and the template generation module uses the nlohmann / json library or MySQL API to implement structured storage; the storage unit can be implemented using the local solid-state drive of the industrial control computer or a remote database server. Those skilled in the art, based on the architecture and algorithm logic disclosed in this specification, combined with the aforementioned known hardware and software tools, can completely implement the system of this invention without creative effort.

[0058] This embodiment provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps of the method as described in the embodiment.

[0059] This embodiment also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method as described in the embodiment.

[0060] The above-described embodiments of the present invention solve the problem of cylindrical surface detection distortion by 3D to 2D dimension reduction with precision preservation; overcome the problem of missed and false detections in complex industrial sites by prior mesh correction; obtain a reference height with extremely high precision by 2D mask backtracking 3D noise reduction statistics; and finally construct a structured template to achieve a leap from qualitative inspection to quantitative tracking, which greatly improves the reliability of preventive maintenance of roller mills.

[0061] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for constructing a reference template for detecting pins on the roller surface of a roller mill, characterized in that, include: Step S1: Convert the complete three-dimensional point cloud data collected from the roller surface of the roller mill in its initial healthy state into a depth image reflecting height information and a reference grayscale image for pin target recognition. Step S2: Based on the known design parameters and arrangement rules of the roller mill pins, automatically identify and locate all pins on the reference grayscale image, determine the theoretical position and theoretical outline of each pin, and generate a pin position template image; Step S3: Map the theoretical contour area of ​​each post in the template image back to the depth image, and calculate the statistical average value of the pixel depth values ​​in each area as the reference height of the corresponding post. Step S4: Associate and store the theoretical position, theoretical outline, and reference height of each post pin to form a structured reference template.

2. The method for constructing a reference template for detecting pins on the roller surface of a roller mill according to claim 1, characterized in that, In step S1, before the new roller is put into use, the roller is controlled to rotate at low speed for one revolution while the rotary encoder installed on the roller shaft is used to synchronously trigger the three-dimensional contour scanning device to collect high-precision three-dimensional point cloud data covering the entire roller surface as reference point cloud data. The reference point cloud data is preprocessed, including removing background interference points and outlier noise points through pass-through filtering and performing coordinate system normalization.

3. The method for constructing a reference template for detecting pins on the roller surface of a roller mill according to claim 1, characterized in that, In step S1, the preprocessed reference point cloud data is projected onto a two-dimensional plane corresponding to the unfolded outer cylindrical surface of the roller to generate a 32-bit floating-point depth image reflecting height information. By setting two-dimensional preset mapping parameters corresponding to the height range of the post nail, the depth value of the depth image is linearly mapped to a gray level of 0-255 to generate an 8-bit reference gray level image, thereby enhancing the contrast of the post nail target. The preset mapping parameters are used to linearly map the range of depth values ​​corresponding to the pin height in the depth image to the full dynamic range of the grayscale image.

4. The method for constructing a reference template for detecting pins on the roller surface of a roller mill according to claim 1, characterized in that, In step S2, the pin design parameters include at least the nominal diameter, row spacing, and column spacing of the pins; the arrangement rule is a uniform distribution in a row and column matrix; on the reference grayscale image, an image processing algorithm is used to automatically identify and locate all pin targets; the image processing algorithm includes adaptive threshold segmentation, morphological operations, connected component analysis, and circular contour detection.

5. The method for constructing a reference template for detecting pins on the roller surface of a roller mill according to claim 4, characterized in that, In step S2, based on the known prior constraints of the row and column spacing of the post, the center of the initially identified circular contour is meshed and corrected. All identified post are numbered and their pixel-level theoretical center coordinates and theoretical radius are recorded to determine the final theoretical position and theoretical contour area of ​​each post, and to correct false detections or missed detections.

6. The method for constructing a reference template for detecting pins on the roller surface of a roller mill according to claim 3, characterized in that, In step S3, the theoretical contour region of each post is used as a spatial index and mapped back to the 32-bit floating-point depth image generated in step S1. The statistical average of the depth values ​​of all valid pixels within the region is calculated as the reference average height value of the post.

7. The method for constructing a reference template for detecting pins on the roller surface of a roller mill according to claim 6, characterized in that, In step S3, before calculating the statistical average, the depth values ​​in the region are first filtered by median, and outliers other than ±3σ standard deviation are removed. In step S4, the number of each post, the theoretical center coordinates, the theoretical profile parameters, and the benchmark average height value calculated after filtering are associated and stored in the database or a structured file to complete the construction of the structured benchmark template.

8. A reference template construction system for detecting pins on the roller surface of a roller mill, characterized in that, include: A 3D scanning unit is used to acquire 3D point cloud data of the initial roller surface; The image conversion and recognition unit is used to convert the three-dimensional point cloud data into a reference grayscale image and automatically identify and locate all the pins based on the pin design parameters. The data association and processing unit is used to map the identified pin outline to the depth image to calculate the reference height, and generate a structured reference template that associates and stores the location, outline and height information. A storage unit for storing the structured reference template.

9. The reference template construction system for detecting roller surface pins in a roller mill according to claim 8, characterized in that, The 3D scanning unit includes a rotary encoder and a 3D contour scanning device. The rotary encoder is mounted on the roller shaft and is used to synchronously trigger the 3D contour scanning device to acquire high-precision 3D point cloud data covering the entire roller surface when the roller rotates at low speed. The 3D scanning unit also includes a preprocessing module, which is used to remove background interference points and outlier noise points through pass-through filtering and to perform coordinate system normalization. The image conversion and recognition unit includes a projection mapping module, a grayscale conversion module, a target initial detection module, and a grid correction module: the projection mapping module is used to project three-dimensional point cloud data onto a two-dimensional plane corresponding to the unfolded outer cylindrical surface of the roller, generating a 32-bit floating-point depth image; The grayscale conversion module is used to linearly map the depth values ​​in the depth image to 0-255 grayscale levels by setting a linear mapping interval corresponding to the nominal height range of the post nail, thereby generating an 8-bit reference grayscale image. The target initial detection module is used to perform adaptive threshold segmentation, morphological operations, connected component analysis, and circular contour detection on the reference grayscale image to initially identify the pin targets; the grid correction module is used to perform grid fitting and correction on the center of the initially identified circular contour based on the known prior constraints of the row and column spacing of the pins, to determine the final theoretical position and theoretical contour region of each pin and number it.

10. The reference template construction system for detecting roller surface pins in a roller mill according to claim 8, characterized in that, The data association and processing unit includes: The height calculation module is used to map the theoretical contour area of ​​each post back to the 32-bit floating-point depth image as a spatial index, perform median filtering on the depth values ​​of all valid pixels in the area, and calculate the statistical average value after removing outliers other than ±3σ standard deviation, which is used as the reference average height value of the post. The template generation module is used to associate and package the number, theoretical center coordinates, theoretical contour parameters and benchmark average height value of each post pin to generate the structured benchmark template and store it in a database or structured file.

Images