AI vision-based coking coal particle size statistical method, device, equipment and medium

Abstract

This application discloses a method, apparatus, equipment, and medium for coking coal particle size statistics based on AI vision, relating to the field of image recognition technology. The method includes: acquiring images of coking coal conveying and error compensation data; establishing a pixel-to-physical coordinate mapping through coordinate transformation; and determining the presence of particles using a target detection model. After particle identification, contour segmentation and coordinate transformation are performed, the equivalent particle size and mass percentage of each particle size range are calculated, and finally, the results are corrected using error compensation data to output the final statistical data. This method enables online automated detection of coking coal particle size, avoiding the drawbacks of manual operation, and effectively improving detection accuracy through coordinate calibration and error compensation.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and in particular to a method, apparatus, equipment and medium for statistical analysis of coking coal particle size based on AI vision. Background Technology

[0002] Currently, the mainstream testing methods in the industry fall into two categories. The first is traditional manual testing, where staff use square-hole sieves of different sizes to sieve the sampled coking coal step by step, then weigh the material in each particle size range and calculate the corresponding particle size percentage. The second is basic image visual inspection, which uses camera equipment to capture images of the coking coal and relies on simple image analysis methods to identify particles and estimate particle size. In addition, the test results are also cross-validated on-site using both methods.

[0003] Manual screening relies entirely on repetitive manual operations, resulting in high labor intensity and low testing efficiency. It cannot match the continuous conveyor belt pace of industrial production. Furthermore, human factors such as screening force, particle placement angle, and operator skill significantly affect testing accuracy, leading to poor data repeatability. This method also only allows for offline sampling and cannot achieve real-time online monitoring. Conventional image-based visual inspection solutions lack precise calibration of camera pixel coordinates with on-site physical coordinates and do not incorporate error compensation mechanisms. This results in insufficient accuracy in particle contour segmentation and particle size conversion, leading to significant detection deviations and failing to meet the high-precision particle size detection requirements of steel metallurgical production.

[0004] Therefore, how to achieve automated, online, and high-precision statistics of coking coal particle size while reducing manual input and human detection errors has become an urgent problem to be solved. Summary of the Invention

[0005] The main purpose of this application is to provide a method, device, equipment and medium for statistical analysis of coking coal particle size based on AI vision, aiming to solve the technical problem of how to achieve automated online high-precision statistical analysis of particle size during continuous coking coal transportation.

[0006] To achieve the above objectives, this application proposes an AI vision-based method for calculating the particle size of coking coal, comprising: Acquire images of coking coal conveying and error compensation data captured by camera equipment; The coordinates of the pixel positions in the coking coal conveying image are converted to obtain the pixel physical mapping relationship; The image of the coking coal being transported is input into a target detection model to determine the presence of particles, and the result of particle presence is obtained. When the particle presence result indicates the presence of coking coal particles, the coking coal conveying image is segmented into particle contours, and the particle contour coordinates are converted according to the pixel physical mapping relationship to obtain the actual contour coordinates of each coking coal particle. The equivalent particle size of each coking coal particle is calculated based on its actual contour coordinates, and the mass percentage of each particle size range is determined based on its equivalent particle size. The equivalent particle size of each coking coal particle and the mass ratio of each particle size range are corrected based on the error compensation data to obtain the statistical results of coking coal particle size.

[0007] In one embodiment, the step of calculating the equivalent particle size of each coking coal particle based on its actual contour coordinates, and determining the mass percentage of each particle size range based on its equivalent particle size, includes: Based on the actual contour coordinates of each coking coal particle, the maximum abscissa, minimum abscissa, maximum ordinate, and minimum ordinate of the contour of each coking coal particle are extracted to obtain the physical corner points of the circumscribed rectangle of each coking coal particle. The length of the diagonal of the outer rectangle of each coking coal particle is calculated based on the physical corner points of the outer rectangle of each coking coal particle. The equivalent particle size of each coking coal particle is calculated based on the length of the diagonal of the circumscribed rectangle of each coking coal particle and the preset equivalent conversion coefficient, and the equivalent particle size of each coking coal particle is obtained. The preset equivalent conversion coefficient is the ratio of the length of the diagonal of the circumscribed rectangle to the side length of the equivalent cube. The equivalent particle size of each coking coal particle is classified according to the preset particle size range to obtain a particle set for each particle size range. Based on the particle set of each particle size range, the cubic equivalent particle size of each coking coal particle is used as the volume reference value of each coking coal particle. The proportion of the total volume reference value of all particles in each particle size range to the total volume reference value of all particles is calculated to obtain the volume ratio of each particle size range. Based on the uniform density of coking coal particles, the volume ratio of each particle size range is taken as the mass ratio of each particle size range.

[0008] In one embodiment, the step of correcting the equivalent particle size and mass percentage of each coking coal particle based on the error compensation data to obtain the statistical results of coking coal particle size includes: The detection deviation value of each particle size interval is calculated based on the manual standard sieving data in the error compensation data and the mass ratio of each particle size interval, and the deviation data of each particle size interval is obtained. According to the preset weight allocation rules, the corresponding weight coefficients are assigned to each particle size range to obtain the compensation weight of each particle size range. Based on the deviation data of each particle size interval and the compensation weight of each particle size interval, the comprehensive compensation value is calculated by weighted average calculation to obtain the comprehensive compensation value of the particle size interval. Based on the comprehensive compensation value of the particle size interval, the equivalent particle size of each coking coal particle falling into each particle size interval is corrected for interval assignment deviation to obtain the corrected equivalent particle size set. Based on the comprehensive compensation value of the particle size range, the mass ratio of each particle size range is corrected for proportional deviation to obtain the corrected mass ratio set. Based on the corrected equivalent particle size set and the corrected mass percentage set, the statistical results of coking coal particle size are generated.

[0009] In one embodiment, the step of acquiring the coking coal conveying images and error compensation data collected by the camera device includes: Send an image acquisition command to the camera device so that the camera device can acquire images of coking coal conveying according to a preset image frame interval; Receive the video stream returned by the camera device, and extract the current image frame from the video stream to obtain the coking coal conveying image; A predetermined number of particle images are randomly selected from the collected images of coking coal conveying. The particle size and mass percentage of the actual coking coal particles in the corresponding images are measured and statistically analyzed using a manual standard sieving method to obtain error compensation data, which includes the manual standard sieving data.

[0010] In one embodiment, the step of performing coordinate conversion on the pixel positions in the coking coal conveying image to obtain the pixel physical mapping relationship includes: The grid size of the checkerboard calibration plate is determined according to the preset detection accuracy level, and the specification parameters of the calibration plate are obtained. Adjust the position of the checkerboard calibration board according to the calibration board specifications to obtain the calibration board placement state; The camera device is controlled to acquire images of the calibration board from different angles and positions based on the placement state of the calibration board, thereby obtaining a multi-view calibration image set; Based on the multi-view calibration image set, the intrinsic parameter matrix, extrinsic parameter matrix, and distortion coefficients of the camera device are calculated using the Zhang Zhengyou calibration algorithm to obtain the initial calibration parameters. The coordinate transformation accuracy of the preset verification point set is verified based on the initial calibration parameters to obtain the transformation error verification result. When the conversion error verification result is less than the preset error threshold, the initial calibration parameters are determined as camera device calibration data, and a pixel physical mapping relationship is established based on the camera device calibration data.

[0011] In one embodiment, the step of inputting the coking coal conveying image into a target detection model to determine the presence of particles and obtain the particle presence result includes: The image frame to be detected is extracted from the coking coal conveying image according to the preset detection time interval to obtain the current detection frame; The current detection frame is input into the target detection model for multi-scale feature extraction to obtain a multi-scale feature map; Based on the multi-scale feature map, candidate boxes are generated and confidence scores are assigned to the particle regions in the current detection frame to obtain a set of particle candidate boxes and a corresponding set of confidence scores. The set of confidence scores is filtered based on a preset confidence threshold, and candidate boxes with confidence scores greater than the preset confidence threshold are retained as valid granular regions, thus obtaining a set of valid granular regions. Based on the set of effective particle regions, it is determined that there are coking coal particles in the current detection frame, and the particle presence result is obtained.

[0012] In one embodiment, the step of performing particle contour segmentation on the coking coal conveying image and converting the particle contour coordinates according to the pixel physical mapping relationship to obtain the actual contour coordinates of each coking coal particle when the particle presence result is that coking coal particles are present includes: When the result indicates the presence of coking coal particles, the coking coal conveying image is input into the particle segmentation model. Contour segmentation and extraction are performed on each coking coal particle to obtain the initial set of contour pixel coordinates for each particle. The initial contour pixel coordinate set is subjected to abnormal contour removal processing according to the preset contour filtering rules to obtain the contour pixel coordinate set of each particle. Based on the pixel physical mapping relationship, the pixel coordinates in the set of outline pixel coordinates of each particle are converted into actual physical coordinates one by one to obtain the actual outline coordinates of each coking coal particle.

[0013] Furthermore, to achieve the above objectives, this application also proposes an AI vision-based coking coal particle size counting device, which includes: The data acquisition module is used to acquire images of coking coal conveying and error compensation data collected by the camera equipment; The coordinate conversion module is used to perform coordinate conversion on the pixel positions in the coking coal conveying image to obtain the pixel physical mapping relationship; The particle detection module is used to input the coking coal conveying image into the target detection model to determine the presence of particles and obtain the particle presence result; The contour segmentation module is used to perform particle contour segmentation on the coking coal conveying image when the particle presence result is that coking coal particles are present, and to convert the particle contour coordinates according to the pixel physical mapping relationship to obtain the actual contour coordinates of each coking coal particle. The particle size statistics module is used to calculate the equivalent particle size of each coking coal particle based on the actual contour coordinates of each coking coal particle, and to determine the mass percentage of each particle size range based on the equivalent particle size of each coking coal particle. The error correction module is used to correct the equivalent particle size of each coking coal particle and the mass ratio of each particle size interval based on the error compensation data, so as to obtain the statistical results of the coking coal particle size.

[0014] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the AI ​​vision-based coking coal particle size statistical method described above.

[0015] In addition, to achieve the above objectives, this application also provides a computer program product, which includes a computer program that, when executed by a processor, implements the steps of the AI ​​vision-based coking coal particle size statistical method described above.

[0016] This application acquires images of coking coal conveying and error compensation data, establishes a pixel-to-physical coordinate mapping through coordinate transformation, and uses a target detection model to determine the presence of particles. After particle identification, contour segmentation and coordinate transformation are performed, the equivalent particle size and the mass proportion of each particle size range are calculated, and finally, the results are corrected using error compensation data to output the final statistical data. This enables online automated detection of coking coal particle size, avoiding the drawbacks of manual operation, and effectively improving detection accuracy through coordinate calibration and error compensation. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating the first embodiment of the coking coal particle size statistical method based on AI vision in this application. Figure 2 This is a flowchart illustrating the second embodiment of the coking coal particle size statistical method based on AI vision in this application. Figure 3 This is a schematic diagram of the module structure of the coking coal particle size counting device based on AI vision in this application; Figure 4 This is a schematic diagram of the equipment structure of the hardware operating environment involved in the AI ​​vision-based coking coal particle size statistics method in the embodiments of this application.

[0019] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0020] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0021] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0022] Currently, the mainstream testing methods in the industry fall into two categories. The first is traditional manual testing, where workers use square-hole sieves of different sizes to sieve the sampled coking coal step by step, then weigh the material in each particle size range and calculate the corresponding particle size percentage. The second is basic image visual inspection, which uses camera equipment to capture images of the coking coal and relies on simple image analysis methods to identify particles and estimate particle size. In addition, the test results are cross-validated using both methods on-site. Manual screening relies entirely on repetitive manual operations, resulting in high labor intensity and low testing efficiency. It cannot match the pace of continuous conveyor belt transport in industrial production. Furthermore, human factors such as screening force, particle placement angle, and operator skill can significantly affect the testing accuracy, leading to poor data repeatability. This method can only perform offline sampling and cannot achieve real-time online monitoring. Conventional image visual inspection solutions, on the other hand, do not achieve accurate calibration of the camera's pixel coordinates with the physical coordinates on-site, nor do they have an error compensation mechanism. The accuracy of particle contour segmentation and particle size conversion is insufficient, resulting in large detection deviations and making it difficult to meet the high-precision particle size detection requirements of steel metallurgical production. Therefore, how to achieve automated, online, and high-precision statistics of coking coal particle size while reducing manual input and human detection errors has become an urgent problem to be solved.

[0023] Based on the above, this application also provides an AI vision-based method for calculating the particle size of coking coal, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the coking coal particle size statistical method based on AI vision in this application.

[0024] In this embodiment, the AI ​​vision-based method for calculating the particle size of coking coal includes steps S10 to S60: Step S10: Obtain images of coking coal conveying and error compensation data captured by the camera equipment.

[0025] It should be noted that step S10 includes: sending an image acquisition command to the camera device so that the camera device can acquire coking coal conveying images at a preset image frame interval; receiving the video stream returned by the camera device and extracting the current image frame from the video stream to obtain the coking coal conveying image; randomly selecting a preset number of particle image images from the acquired coking coal conveying images, and measuring the particle size and mass percentage of the actual coking coal particle samples in the corresponding images using a manual standard sieving method to obtain error compensation data, wherein the error compensation data includes the manual standard sieving data.

[0026] It's important to understand that the preset image frame interval is a pre-defined time length between two adjacent acquired images, determined based on the on-site production rhythm and actual testing requirements. The coking coal conveying image is the content captured by the camera equipment, showing coking coal particles on the conveyor. The current image frame is a single static image extracted from a continuous video stream, forming the basic unit of the video stream, and can be directly used for subsequent particle detection and analysis. Manual standard sieving is a traditional particle size detection method commonly used in the industry, relying on sieves of different sizes to sieve the material step by step, followed by weighing to complete data statistics. The coking coal particle sample is the actual coking coal extracted from the corresponding particle image location; the particle size of the actual particles is consistent with the particles presented in the image. Error compensation data is reference data used to correct visual inspection results. This type of data is generated based on manual standard inspection results to offset deviations generated during visual inspection. Manual standard sieving data is the value obtained after measurement and statistics using the manual standard sieving method, including the particle size values ​​of each particle size range of coking coal and their corresponding mass percentage values.

[0027] Specifically, firstly, the high-definition camera is fixedly installed on a bracket directly above the coking coal conveyor belt. The lens plane of the high-definition camera is adjusted to be parallel to the conveyor belt's transport plane. This is because when the lens plane is parallel to the transport plane, there is no angle between the camera's imaging plane and the physical plane of the conveyor belt, which avoids perspective distortion causing stretching or compression of particle edges and ensures that the ratio between pixel coordinates and physical coordinates remains consistent throughout the entire frame during subsequent coordinate conversion. Secondly, the field of view of the high-definition camera is adjusted to completely cover the width of the conveyor belt, with pre-set redundant space on both sides of the width direction. This is to prevent the coking coal particles in the edge area from remaining intact in the frame even if the conveyor belt deviates slightly during long-term operation, avoiding missed detections or distortion in particle size calculations due to particles being cut off by the image. Next, the installation angle of the high-definition camera is adjusted so that the running direction of the conveyor belt is perpendicular to the horizontal axis of the pixel coordinates in the high-definition camera's image. This ensures that the movement direction of the particles on the conveyor belt is consistent with the vertical axis of the pixel coordinates. When extracting the maximum and minimum horizontal coordinates of the particle outlines, the difference in the horizontal coordinates directly corresponds to the physical size of the particle in the width direction, simplifying the rotational transformation calculations for coordinate conversion. Then, the installation height of the high-definition camera is set according to the width of the conveyor belt and the focal length of the high-definition camera lens. Too low an installation height would result in insufficient field of view to cover the entire width of the conveyor belt, while too high an installation height would result in too few pixels occupied by a single particle in the image, leading to loss of image detail. Determining the installation height through bandwidth and focal length matching calculations achieves a balance between framing completeness and image clarity. Finally, an image acquisition command is sent to the camera to acquire images of the coking coal conveying process at preset image frame intervals. This ensures real-time online detection during continuous conveying while avoiding the waste of computing power and data redundancy caused by full-frame acquisition. Next, the video stream returned by the camera is received, and the current image frame is extracted from the video stream to obtain the coking coal conveying image. This is because the video stream is transmitted continuously, and single frames need to be extracted from it at preset time intervals as input for subsequent processing. Finally, a preset number of particle images are randomly selected from the historical storage data of the acquired coking coal conveying images. The particle size and mass percentage of the actual coking coal particle samples in the corresponding images are measured and statistically analyzed using the manual standard sieving method to obtain error compensation data. The error compensation data includes the manually measured mass percentage of each particle size range. The reason for random selection rather than fixed selection is to avoid sample bias and ensure that the error compensation data can objectively reflect the particle distribution characteristics of different conveying periods. The manual standard sieving method is an industry benchmark measurement method, and its measurement results are regarded as true values, which are used to compare with the visual inspection results to calculate the deviation.

[0028] Step S20: Perform coordinate conversion on the pixel positions in the coking coal conveying image to obtain the pixel physical mapping relationship.

[0029] It should be noted that step S20 includes: determining the grid size of the checkerboard calibration board according to the preset detection accuracy level to obtain the calibration board specification parameters; adjusting the position of the checkerboard calibration board according to the calibration board specification parameters to obtain the calibration board placement state; controlling the camera device to acquire calibration board images from different angles and positions according to the calibration board placement state to obtain a multi-view calibration image set; calculating the intrinsic parameter matrix, extrinsic parameter matrix, and distortion coefficient of the camera device using the Zhang Zhengyou calibration algorithm based on the multi-view calibration image set to obtain the initial calibration parameters; verifying the coordinate transformation accuracy of the preset verification point set according to the initial calibration parameters to obtain the transformation error verification result; when the transformation error verification result is less than the preset error threshold, determining the initial calibration parameters as the camera device calibration data, and establishing the pixel physical mapping relationship based on the camera device calibration data.

[0030] It's important to understand that the preset detection accuracy level refers to a pre-defined level of precision. Different levels correspond to different stringent requirements for particle size detection results on-site and serve as a reference for selecting the calibration plate specifications. The checkerboard calibration plate is a specialized calibration tool with a surface distribution of alternating black and white squares; it's a common tool used in the vision field to calibrate camera equipment parameters. The preset verification point set consists of several pre-selected coordinate reference points distributed on the transport plane, specifically used to verify the accuracy of coordinate transformation. The camera equipment calibration data consists of the final imaging parameters determined after accuracy verification; it is the core basis for achieving pixel-to-physical coordinate transformation.

[0031] Specifically, firstly, the grid size of the checkerboard calibration plate is determined according to the preset detection accuracy level to obtain the calibration plate specification parameters. This is because the higher the detection accuracy level, the smaller the grid size of the calibration plate and the more grids are required, so as to provide denser corner coordinates for calibration calculation. For example, a small grid calibration plate of 5 mm × 5 mm is selected when millimeter-level detection accuracy is required, while a large grid calibration plate of 20 mm × 20 mm can be selected when centimeter-level accuracy is required. By matching the accuracy level with the grid size, it is ensured that the calibration results can meet the accuracy requirements of subsequent particle size calculation. Secondly, the checkerboard calibration board is laid flat on the conveyor belt's transport plane, and its position is adjusted according to the calibration board's specifications to ensure it is completely within the camera's field of view and without any tilt or obstruction. This is because the calibration board must be located on the actual detection plane (i.e., the conveyor belt plane) to establish the pixel-physical mapping relationship of that plane. If the calibration board is tilted, the corner coordinates will not be in the same plane, introducing additional projection errors. If there is any obstruction, some corner points will not be recognized, resulting in missing calibration data. Then, based on the placement of the calibration board, the camera device is controlled to acquire images of the calibration board at different angles and positions, resulting in a multi-view calibration image set. This is because Zhang Zhengyou's calibration algorithm needs to utilize the imaging differences of the calibration board under different poses to calculate the intrinsic and distortion parameters of the camera device. Images from a single angle and position cannot provide sufficient constraints. Typically, 10 to 15 images covering different translation and rotation angles need to be acquired. For example, the calibration board is placed sequentially at the center, left, right, near, and far ends of the conveyor belt and tilted at different angles to ensure that the corner points of the calibration board cover the entire image area. Next, based on the multi-view calibration image set, Zhang Zhengyou's calibration algorithm calculates the intrinsic parameter matrix, extrinsic parameter matrix, and distortion coefficients of the camera device to obtain the initial calibration parameters. The intrinsic parameter matrix describes the focal length and principal point position of the camera device lens, the extrinsic parameter matrix describes the pose relationship of each calibration image relative to the camera device, and the distortion coefficients describe the radial and tangential distortion of the lens. These three parameters together constitute the mathematical basis for the transformation from pixel coordinates to physical coordinates. Next, the coordinate transformation accuracy of the preset verification point set is verified according to the initial calibration parameters to obtain the transformation error verification result. Specifically, several verification points with known physical coordinates are selected on the conveyor belt plane, and their pixel coordinates are converted into physical coordinates using the initial calibration parameters. The deviation is calculated by comparing them with the actual physical coordinates. For example, the four corner points of the conveyor belt edge are selected as verification points. If the deviation between the converted coordinates and the measured coordinates exceeds the preset error threshold, it indicates that the calibration accuracy is insufficient.Finally, when the conversion error verification result is less than the preset error threshold, the initial calibration parameters are determined as the camera equipment calibration data, and the pixel physical mapping relationship is established based on the camera equipment calibration data. This is because only the calibration parameters that have passed the accuracy verification can ensure the reliability of the subsequent particle size calculation. If the verification fails, the calibration image needs to be re-acquired or the grid size of the calibration plate needs to be adjusted. Once the pixel physical mapping relationship is established, any pixel coordinates can be directly converted into actual physical coordinates on the conveyor belt plane, providing a conversion basis for the physical size calculation of the particle profile.

[0032] Step S30: Input the coking coal conveying image into the target detection model to determine the presence of particles and obtain the particle presence result.

[0033] It should be noted that step S30 includes: extracting the image frame to be detected from the coking coal conveying image according to a preset detection time interval to obtain the current detection frame; inputting the current detection frame into the target detection model for multi-scale feature extraction to obtain a multi-scale feature map; generating candidate boxes and scoring confidence scores for particle regions in the current detection frame based on the multi-scale feature map to obtain a set of particle candidate boxes and a set of corresponding confidence scores; filtering the set of confidence scores based on a preset confidence threshold, retaining candidate boxes with confidence scores greater than the preset confidence threshold as valid particle regions to obtain a set of valid particle regions; and determining the presence of coking coal particles in the current detection frame based on the set of valid particle regions to obtain a particle presence result.

[0034] It's important to understand that the preset detection time interval is the pre-defined time span between two consecutive image detection operations, a value determined based on on-site production conditions and detection requirements. The image frame to be detected is a single static image selected from the coking coal conveying images for particle identification analysis; it is the direct processing object in the detection process. The target detection model is a trained intelligent algorithm model specifically designed to locate a specified target object in the image; here, it is used to identify coking coal particles.

[0035] Specifically, firstly, the image frame to be detected is extracted from the coking coal conveying image according to a preset detection time interval to obtain the current detection frame. This is because coking coal is continuously conveyed on the conveyor belt, and performing subsequent segmentation and calculation on each frame would result in a significant waste of computing power. By setting a reasonable detection time interval (e.g., extracting one frame every 200 seconds), timely monitoring of raw material quality fluctuations can be ensured while avoiding ineffective processing, achieving a balance between detection efficiency and real-time performance. Secondly, the current detection frame is input into the target detection model, which performs multi-scale feature extraction to obtain a multi-scale feature map. This is because the particle size range of coking coal particles is large, and the scale difference from fine powder to block particles in the image is significant. Multi-scale feature extraction can simultaneously capture the local texture features of small particles and the global shape features of large particles. For example, the model will extract the edge details of fine powder and the contour information of block particles at different resolution levels, avoiding the omission of small particles or the loss of large particle features at a single scale. Then, candidate boxes are generated and confidence scores are assigned to the particle regions in the current detection frame based on the multi-scale feature map, resulting in a set of particle candidate boxes and a corresponding set of confidence scores. Specifically, the target detection model slides anchor boxes of different sizes on the feature map, performs binary classification of the area covered by each anchor box as either particle or non-particle, and outputs the classification confidence score. For example, if the probability of an area covered by a certain anchor box being identified as coking coal particles is 0.92, a candidate box is generated and assigned a confidence score of 0.92, while recording the position coordinates and size of the candidate box. Next, the set of confidence scores is filtered based on a preset confidence threshold. Candidate boxes with confidence scores greater than the preset confidence threshold are retained as valid particle regions, resulting in a set of valid particle regions. This is because low-confidence candidate boxes are often false targets caused by interference factors such as conveyor belt scratches, coal dust, and reflective noise. By setting a confidence threshold (e.g., 0.7) for hard filtering, most false detection areas can be filtered out, and only high-confidence real particle regions are retained for subsequent processing, reducing the invalid calculations in the instance segmentation stage. Finally, based on the set of effective particle regions, it is determined that there are coking coal particles in the current detection frame, and the particle presence result is obtained. This is because as long as the set of effective particle regions is not empty, it means that there is at least one real particle region with high confidence in the current detection frame, and the subsequent instance segmentation process is triggered. If the set of effective particle regions is empty, the subsequent processing is skipped and the next frame is detected directly to avoid invalid segmentation of frames without particles.

[0036] Furthermore, before step S30, the process includes: First, acquiring first-class sample images containing coking coal particles and second-class sample images not containing coking coal particles using a camera device to obtain an initial sample image set. The number of first-class and second-class sample images is equal to the preset sample balance number, for example, 50 images of each. This is because an imbalance in the number of samples will cause the model to favor the class with more samples, affecting the accuracy of particle presence judgment. Second, performing data augmentation processing on the initial sample image set to obtain an enhanced sample image set. The data augmentation processing covers scenarios with varying lighting and particle distribution density. For example, adjusting brightness and contrast, adding Gaussian noise to simulate different lighting conditions, and randomly cropping, rotating, and scaling to simulate different distribution densities. This is because the lighting environment and particle accumulation state are highly variable in actual production, and the enhanced samples enable the model to learn more robust feature representations, avoiding failure under specific lighting or density conditions. Then, based on the enhanced sample image set, a binary classification training process is performed on the first and second class sample images using a pre-defined target detection network structure to obtain the target detection model. This pre-defined target detection network structure employs a single-stage detector architecture, such as the YOLO or SSD series. These models integrate target localization and classification into a single regression problem, achieving millisecond-level inference speed while maintaining detection accuracy, meeting the real-time requirements of continuous coking coal transportation scenarios. During training, the intersection-union ratio (IUU) loss between candidate boxes and ground truth bounding boxes, along with the classification cross-entropy loss, are used as joint optimization objectives. The network weights are iteratively updated using a backpropagation algorithm until the loss converges or the preset number of iterations is reached. Finally, the trained target detection model is deployed to a coking coal particle size statistics system to determine the presence of particles in real-time acquired coking coal transportation images.

[0037] Step S40: When the particle presence result indicates the presence of coking coal particles, the coking coal conveying image is segmented into particle contours, and the particle contour coordinates are converted according to the pixel physical mapping relationship to obtain the actual contour coordinates of each coking coal particle.

[0038] It should be noted that step S40 includes: when the particle presence result indicates the presence of coking coal particles, inputting the coking coal conveying image into the particle segmentation model, performing contour segmentation and extraction on each coking coal particle to obtain the initial contour pixel coordinate set of each particle; performing abnormal contour removal processing on the initial contour pixel coordinate set according to the preset contour filtering rules to obtain the contour pixel coordinate set of each particle; and converting each pixel coordinate in the contour pixel coordinate set of each particle into actual physical coordinates one by one according to the pixel physical mapping relationship to obtain the actual contour coordinates of each coking coal particle.

[0039] Specifically, firstly, when the particle presence result indicates the presence of coking coal particles, the current detection frame is input into the particle segmentation model. Contour segmentation and extraction are performed on each coking coal particle to obtain the initial set of contour pixel coordinates for each particle. The particle segmentation model adopts an instance segmentation network architecture, such as Mask R-CNN or the SOLO series. These models add a pixel-level mask branch on top of object detection, which can simultaneously output the bounding box position of the particle and the accurate contour mask. Unlike semantic segmentation, which only distinguishes pixel categories without distinguishing individual instances, instance segmentation can accurately separate the contact boundaries of adjacent particles, avoiding overlapping particles being incorrectly merged into a single contour. During training, sample images containing coking coal particles are acquired using high-definition camera equipment. Image annotation tools are used to annotate the contour boundaries of each coking coal particle to obtain an annotated sample image set. Then, contour segmentation training is performed through a preset instance segmentation network structure. The mask intersection-union ratio loss and boundary distance loss are used as joint optimization objectives. The network parameters are iteratively optimized until the mask prediction accuracy meets the preset threshold, thus obtaining the final particle segmentation model. Secondly, abnormal contours are removed from the initial contour pixel coordinate set according to preset contour filtering rules to obtain the contour pixel coordinate set of each particle. This is because the initial contours output by instance segmentation may contain artificial noise contours with too small an area, adhesive region contours with too large an area, and narrow fragment contours with abnormal aspect ratios. By setting lower and upper limits for area and aspect ratio ranges, filtering is performed to retain only reasonable contours that conform to the morphological characteristics of coking coal particles, avoiding abnormal contours from entering subsequent particle size calculations and causing statistical distortion. Finally, according to the pixel physical mapping relationship, the pixel coordinates in the contour pixel coordinate set of each particle are converted one by one into actual physical coordinates to obtain the actual contour coordinates of each coking coal particle. This is because contour pixel coordinates only represent the row and column positions in the image, while particle size calculation requires actual physical dimensions. Through the pixel physical mapping relationship established in the calibration stage, each pixel on the contour is transformed from the image coordinate system to the conveyor belt plane coordinate system, thereby ensuring that the subsequent equivalent particle size calculation is based on the real physical dimensions rather than pixel distances.

[0040] Step S50: Calculate the equivalent particle size of each coking coal particle based on the actual contour coordinates of each coking coal particle, and determine the mass percentage of each particle size range based on the equivalent particle size of each coking coal particle.

[0041] It should be noted that step S50 includes: First, extracting the maximum and minimum abscissa, maximum and minimum ordinate of the coking coal particle profile based on the actual contour coordinates of each coking coal particle, thus obtaining the physical corner points of the bounding rectangle of each coking coal particle. Specifically, the actual contour coordinates are a set of physical coordinate points converted from a series of discrete pixels. By traversing the abscissa and ordinate values ​​of all points in this set, the two points with the maximum and minimum abscissas and the two points with the maximum and minimum ordinates are selected. These four extreme points constitute the four corner points of the minimum bounding rectangle that can completely enclose the particle profile. For example, if the extreme abscissas of a certain particle profile are 152 mm and 168 mm, and the extreme ordinates are 203 mm and 221 mm, then the four physical corner points of the bounding rectangle are (152, 203), (168, 203), (152, 221), and (168, 221).

[0042] Secondly, the diagonal length of the circumscribed rectangle of each coking coal particle is calculated based on the physical corner points of the circumscribed rectangle. Specifically, the difference between the maximum and minimum abscissa is taken as the rectangle width, and the difference between the maximum and minimum ordinate is taken as the rectangle height. Then, the diagonal length is calculated using the Pythagorean theorem. This diagonal length comprehensively reflects the two-dimensional extension scale of the particle on the conveyor belt plane.

[0043] Next, the equivalent particle size of each coking coal particle is calculated based on the diagonal length of its circumscribed rectangle and a preset equivalent transformation coefficient. It is important to understand that the preset equivalent transformation coefficient is the ratio of the diagonal length of the circumscribed rectangle to the side length of the equivalent cube, and its value is... This is because, assuming coking coal particles are regular cubes, the diameter of the circumscribed sphere of this cube is equal to the length of its spatial diagonal. The diagonal of the circumscribed rectangle is approximately equal to the projection of this spatial diagonal onto the imaging plane. By calculation, the side length of the equivalent cube can be obtained, which is the equivalent particle size. This allows for the standardization and quantification of irregular particle shapes into comparable standard particle size values. The specific formula is as follows: in, This represents the equivalent particle size of coking coal particles (equivalent to the side length of a regular cube). Indicates the maximum x-coordinate. Represents the minimum x-coordinate. Represents the maximum ordinate. This represents the maximum minimum y-coordinate.

[0044] Next, the equivalent particle size of each coking coal particle is classified according to the preset particle size range, resulting in a particle set for each particle size range. Specifically, because the coking process has specific requirements for the proportion of different particle size segments, for example, the particle size is divided into multiple ranges such as 0 to 10 mm, 10 to 25 mm, 25 to 40 mm, and above 40 mm. Each particle falls into the corresponding range according to its equivalent particle size, which facilitates subsequent statistical analysis of the mass proportion by range, resulting in a particle set for each particle size range.

[0045] Then, based on the particle sets of each particle size range, the volume reference value of each coking coal particle is calculated using the cube of its equivalent particle size as the volume reference value. The proportion of the total volume reference value of all particles within each particle size range to the total volume reference value of all particles is then calculated, yielding the volume percentage of each particle size range. Specifically, it is difficult to accurately calculate the true volume of irregular particles, while the volume of the equivalent cubic model is the cube of its side length. Although there is an approximate error in the volume reference value of a single particle, this error tends to be evenly distributed when a large number of particles are counted. The proportional relationship of the volume percentage of each range can stably reflect the true volume distribution. For example, if a range contains 3 particles with equivalent particle sizes of 10 mm, 15 mm, and 20 mm, then the total volume reference value of this range is 1000 + 3375 + 8000, which is 12375 cubic millimeters. If the total volume reference value of all particles is 50000 cubic millimeters, then the volume percentage of this range is 24.75%.

[0046] Finally, based on the uniform density of coking coal particles, the volume percentage of each particle size range is taken as the mass percentage of that range. Specifically, since coking coal is a homogeneous mineral, its density can be considered uniform throughout the batch. According to the physical relationship that mass equals density multiplied by volume, when the density is constant, the mass percentage of each particle size range is directly equal to its volume percentage. The specific formula is as follows: in, Indicates the first The mass percentage of coking coal particles in each particle size range; The total number of pre-divided coking coal particle size ranges; Indicates the first The number of coking coal particles within each particle size range; Indicates the first Within the particle size range, the first The equivalent particle size of each coking coal particle; Indicates the first Within the particle size range, the first The volume of each coking coal particle is determined. This avoids the tedious operation of weighing particles in each section in traditional manual screening methods, and enables non-contact measurement of quality statistics through visual inspection alone.

[0047] Step S60: Correct the equivalent particle size and mass ratio of each coking coal particle according to the error compensation data to obtain the statistical results of coking coal particle size.

[0048] Specifically, firstly, a predetermined number of particle images are randomly selected from the historical storage data of collected coking coal conveying images to obtain a manual comparison sample image set. Random selection, rather than selection at fixed time intervals, is used to avoid sample concentration in a specific conveying state, which could lead to compensation deviations and ensure that the error compensation data objectively reflects the particle distribution characteristics under different operating conditions. Secondly, based on the manual comparison sample image set, actual coking coal particle samples are extracted from the corresponding images. Each actual coking coal particle sample is then measured for particle size and its mass percentage is statistically analyzed using a manual standard sieving method. This manual standard sieving method employs an industry-standard square-hole sieve for step-by-step sieving and weighing. The measurement results are considered the true values ​​and are used for subsequent comparison with visual inspection results. Then, based on the manually measured mass percentage of each particle size interval in the error compensation data and the mass percentage of each particle size interval obtained from visual inspection, the detection deviation value for each particle size interval is calculated. This deviation value reflects the systematic error within that particle size interval. Finally, a corresponding weight coefficient is assigned to each particle size interval according to a preset weight allocation rule. Next, based on the deviation data of each particle size interval and the compensation weight of each particle size interval, a comprehensive compensation value is calculated through weighted averaging to obtain the comprehensive compensation value for each particle size interval. Finally, based on the comprehensive compensation value for each particle size interval, the equivalent particle size of each coking coal particle and the mass proportion of each particle size interval are dynamically corrected. For example, the comprehensive compensation value is mapped to individual particles according to the interval particle size for scale correction. At the same time, a correction term consisting of the product of the deviation value and the weight coefficient is added to the mass proportion of each interval to ensure that the deviation between the corrected particle size distribution statistical results and the manual standard measurement results is controlled within a preset accuracy range, thereby obtaining the statistical results of coking coal particle size.

[0049] This embodiment acquires images of coking coal conveying and error compensation data, establishes a pixel-to-physical coordinate mapping through coordinate transformation, and uses a target detection model to determine the presence of particles. After particle identification, contour segmentation and coordinate transformation are performed, the equivalent particle size and the mass proportion of each particle size range are calculated, and finally, the results are corrected using error compensation data to output the final statistical data. This enables online automated detection of coking coal particle size, avoiding the drawbacks of manual operation, and effectively improving detection accuracy through coordinate calibration and error compensation.

[0050] Based on the first embodiment of this application, in the second embodiment of this application, the content that is the same as or similar to that in Embodiment 1 above can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 2 The AI ​​vision-based method for calculating the particle size of coking coal further includes steps S201-S206 in step S60: Step S201: Calculate the detection deviation value of each particle size interval based on the manual standard sieving data and the mass ratio of each particle size interval in the error compensation data, and obtain the deviation data of each particle size interval.

[0051] Specifically, firstly, manual standard sieving data for each particle size range is extracted from the error compensation data. This data is obtained by manually sieving and weighing actual coking coal particle samples using a standard sieving method, resulting in the mass percentage of each range. For example, the manually measured mass percentage for the 10-25 mm range is 35.2%. Secondly, the output mass percentage for each particle size range is read; for example, the visual inspection mass percentage for the same range is 32.8%. Then, the corresponding range mass percentage obtained from visual inspection is subtracted from the mass percentage of each range in the manual standard sieving data to calculate the detection deviation value for each particle size range. For example, the detection deviation value for the 10-25 mm range is 35.2% minus 32.8%, which equals 2.4%. This positive value indicates a systematic underestimation in this range. Next, the above subtraction operation is repeated for all particle size ranges to obtain complete deviation data for each particle size range. This data is then integrated into a deviation dataset, which quantitatively characterizes the error distribution features across different particle size ranges, providing input for subsequent differentiated weight compensation.

[0052] Step S202: Assign corresponding weight coefficients to each particle size range according to the preset weight allocation rules to obtain the compensation weight of each particle size range.

[0053] It should be noted that the preset weighting rules are pre-defined based on the degree of influence of each particle size range on coke quality. For example, the particle size range is divided into four ranges: 0 to 10 mm, 10 to 25 mm, 25 to 40 mm, and above 40 mm, corresponding to fine powder, median diameter, coarse particles, and extra-large lumps, respectively. Secondly, the core particle size range affecting coking quality is typically the median diameter range of 10 to 25 mm. This is because particles in this range form the main structure of the coke skeleton during the coking process, directly determining the pore distribution, reactivity, and mechanical strength of the coke. This coefficient is greater than the weighting coefficients of other particle size ranges because the detection deviation in the core range needs to be prioritized to ensure that compensation resources are tilted towards process-sensitive aspects. Therefore, the weighting coefficient of the core particle size range affecting coking quality is greater than the weighting coefficients of other particle size ranges. Finally, a mapping relationship is established between each particle size range and its corresponding weight coefficient to obtain the compensation weight of each particle size range. For example, the 10 to 25 mm range is mapped to 1.5, the 25 to 40 mm range is mapped to 1.0, and the remaining ranges are mapped to 0.8. This compensation weight is used as an amplification or reduction coefficient for the deviation value of each range in the subsequent weighted average calculation, so as to achieve high-intensity correction of the core range and moderate correction of the non-core range.

[0054] Step S203: Based on the deviation data of each particle size interval and the compensation weight of each particle size interval, calculate the comprehensive compensation value through weighted average calculation to obtain the comprehensive compensation value of the particle size interval.

[0055] Specifically, firstly, the deviation data for each particle size range is paired with its corresponding compensation weight for each range. For example, the deviation value for the 0-10 mm range is 0.8%, and the compensation weight is 0.15; the deviation value for the 10-25 mm range is 2.4%, and the compensation weight is 0.45; the deviation value for the 25-40 mm range is -1.2%, and the compensation weight is 0.25; and the deviation value for the range above 40 mm is 0.5%, and the compensation weight is 0.15. This ensures that the deviation value for each range corresponds one-to-one with its process importance weight, and that the sum of all weight coefficients equals 1. Secondly, the deviation value for each range is multiplied by its compensation weight to obtain the weighted deviation value for each range. This multiplication significantly amplifies the deviation contribution of the core particle size range of 10-25 mm, while the deviation contributions of other ranges decrease in order of importance. Finally, the weighted deviation values ​​for all four ranges are summed to obtain the comprehensive compensation value for the particle size range. Finally, the comprehensive compensation value of the particle size range is used as the unified correction benchmark. This is because the weighted summation after weight normalization has essentially completed the averaging process. The comprehensive compensation value not only reflects the high weight influence of the core range, but also partially offsets it through the negative deviation in the 25 to 40 mm range. This allows the correction strategy to focus on the key range of the process while maintaining the overall numerical stability and avoiding the distortion of the distribution pattern caused by over-correction.

[0056] Step S204: Based on the comprehensive compensation value of the particle size interval, the equivalent particle size of each coking coal particle falling into each particle size interval is corrected for interval assignment deviation to obtain the corrected equivalent particle size set.

[0057] Specifically, firstly, based on the comprehensive compensation value for particle size intervals, the equivalent particle size of each coking coal particle falling into each particle size interval is corrected for interval assignment deviation, resulting in a corrected set of equivalent particle sizes. This is because the systematic bias of the visual inspection system can cause some particles to be misclassified when their equivalent particle size is near the interval boundary. For example, a particle with a true particle size of 24.5 mm should be classified into the 25 to 40 mm interval, but the visual inspection value is 23.8 mm, which is incorrectly classified into the 10 to 25 mm interval. By introducing a comprehensive compensation value, the equivalent particle size of all particles in the interval is uniformly corrected. For example, the equivalent particle size of all particles in the 10 to 25 mm interval is multiplied by a correction coefficient, so that the interval as a whole moves closer to the true distribution, thereby reducing the number of particles misjudged at the boundary. Secondly, in the corrected equivalent particle size set, particle sizes that were originally compressed or stretched due to systematic biases are corrected. The corrected equivalent particle size of some particles may migrate across intervals; for example, the aforementioned 23.8 mm particle becomes 24.6 mm after correction, automatically falling into the correct 25-40 mm interval. This migration process requires no manual intervention; the interval assignment is automatically adjusted by the compensation value. Finally, the corrected equivalent particle size set is used as input for subsequent mass percentage statistics. This is because the accuracy of individual particle sizes directly determines the correctness of interval classification. Only by reclassifying based on the corrected equivalent particle sizes can the authenticity of the particle sets in each size interval be ensured to be consistent with the results of manual standard measurements, providing individual dimension assurance for the reliability of the final particle size distribution statistics.

[0058] Step S205: Based on the comprehensive compensation value of the particle size range, the mass ratio of each particle size range is corrected for proportional deviation to obtain the corrected mass ratio set.

[0059] Specifically, firstly, based on the comprehensive compensation value for each particle size interval, the mass proportion of each particle size interval is corrected for proportional deviation, resulting in a corrected set of mass proportions. This is because the visual inspection system exhibits differentiated systematic biases across different particle size intervals. The mass proportion of the core particle size interval is systematically underestimated due to decreased segmentation accuracy caused by dense particle overlap, while the mass proportion of the non-core interval is systematically overestimated due to higher false detection rates caused by reflective interference or blurred edges. By introducing a comprehensive compensation value, the original mass proportion of each interval is corrected in a targeted manner, bringing the distribution of proportions closer to the results of manual standard measurements. Secondly, the specific method of proportional deviation correction is to add a correction term, the product of the detection deviation value and the corresponding weight coefficient, to the original mass proportion of each particle size interval. This correction method prioritizes compressing the deviation in the core particle size interval while avoiding distribution distortion caused by excessive correction in the non-core interval. Then, the above correction operation is repeated for all particle size intervals to obtain a complete set of corrected mass proportions. The sum of the corrected mass proportions of each interval remains 100%, ensuring that the normalization constraint of the statistical results is not violated. Finally, the corrected mass percentage set is used as a component of the statistical results of coking coal particle size. This is because the mass percentage is a core indicator for coal blending control in coking processes. Only by evaluating the raw material quality based on the corrected mass percentage can we ensure the consistency between visual inspection results and manual standard measurement results, and provide reliable data input for subsequent optimization of coking production parameters.

[0060] Step S206: Generate statistical results of coking coal particle size based on the corrected equivalent particle size set and the corrected mass percentage set.

[0061] Specifically, firstly, based on the corrected equivalent particle size set and the corrected mass percentage set, statistical results of coking coal particle size are generated. This is because the corrected individual equivalent particle size and interval mass percentage together constitute a complete description of the particle size distribution, containing both the precise particle size information of each particle and the proportion of each particle size interval; neither can be omitted. Secondly, the coking coal particles in the corrected equivalent particle size set are reclassified according to particle size intervals to obtain updated particle sets for each particle size interval. This is because after compensation and correction, the equivalent particle size of some particles may migrate across intervals, requiring a reclassification of intervals based on the corrected particle size values ​​to ensure consistency between interval classification and the corrected particle size distribution. Then, the updated particle sets for each particle size interval are linked and integrated with the corrected mass percentage set to form structured particle size statistical results. These results include the number of particles in each particle size interval, the equivalent particle size range, the sum of volume reference values, and the mass percentage. This structured result is presented in the form of a data table or report, facilitating direct reading and comparison by process engineers. Next, the statistical results of coking coal particle size are compared and verified with the preset accuracy threshold to obtain the accuracy verification results. This is because the effectiveness of error compensation needs to be verified by quantitative indicators. For example, the deviation between the corrected mass ratio and the manual standard measurement result should not exceed the preset accuracy threshold. If the verification passes, it indicates that the current compensation parameters are effective; if the verification fails, it indicates that there may be new drift factors in the detection system. Finally, when the accuracy verification result shows that the deviation exceeds the preset accuracy threshold, a new round of manual comparison sample collection process is triggered. Particle images are reselected from the latest coking coal conveying images for manual standard sieving measurement, updating the manual measurement mass ratio of each particle size interval in the error compensation data, and recalculating the detection deviation value and comprehensive compensation value. This achieves periodic refreshing of the compensation parameters, thereby establishing a closed-loop optimization mechanism from visual detection, deviation calculation, weighted compensation, result correction to accuracy verification and parameter updating. This ensures that the statistical results of coking coal particle size continuously approach the accuracy of manual standard measurement during long-term continuous operation, providing stable and reliable raw material quality data support for coking production.

[0062] This embodiment calculates the detection deviation for each particle size range by combining manual standard sieving data with the mass ratio of visual inspection. Weighting coefficients are then assigned according to preset rules, and a weighted average is used to obtain a comprehensive compensation value. This compensation value is used to correct the equivalent particle size and the mass ratio of each range, and the two types of corrected data are integrated to generate the final statistical result. By relying on standard detection data for weighted error compensation, the deviation caused by visual inspection is effectively compensated, comprehensively improving the detection accuracy of coking coal particle size and proportion data, and meeting the high-precision detection requirements of coking production.

[0063] Based on the first embodiment of this application, this application also provides an AI vision-based coking coal particle size counting device, please refer to... Figure 3 The device includes: The data acquisition module 10 is used to acquire images of coking coal conveying and error compensation data collected by the camera equipment.

[0064] The coordinate conversion module 20 is used to perform coordinate conversion on the pixel positions in the coking coal conveying image to obtain the pixel physical mapping relationship.

[0065] The particle detection module 30 is used to input the coking coal conveying image into the target detection model to determine the presence of particles and obtain the particle presence result.

[0066] The contour segmentation module 40 is used to perform particle contour segmentation on the coking coal conveying image when the particle presence result is that coking coal particles exist, and to convert the particle contour coordinates according to the pixel physical mapping relationship to obtain the actual contour coordinates of each coking coal particle.

[0067] The particle size statistics module 50 is used to calculate the equivalent particle size of each coking coal particle based on the actual contour coordinates of each coking coal particle, and to determine the mass percentage of each particle size range based on the equivalent particle size of each coking coal particle.

[0068] The error correction module 60 is used to correct the equivalent particle size and mass ratio of each particle size range of each coking coal particle based on the error compensation data, so as to obtain the statistical results of the particle size of coking coal particles.

[0069] The coking coal particle size counting device based on AI vision provided in this application, employing the coking coal particle size counting method based on AI vision in the above embodiments, can solve the technical problem of how to achieve automated online high-precision counting of particle size during continuous coking coal transportation. Compared with the prior art, the beneficial effects of the coking coal particle size counting device based on AI vision provided in this application are the same as those of the coking coal particle size counting method based on AI vision provided in the above embodiments, and other technical features in the coking coal particle size counting device based on AI vision are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0070] This application provides an AI vision-based coking coal particle size counting device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the AI ​​vision-based coking coal particle size counting method in the above embodiment 1.

[0071] The following is for reference. Figure 4The diagram illustrates a structural schematic of an AI-based vision-based coking coal particle size counting device suitable for implementing embodiments of this application. The AI-based vision-based coking coal particle size counting device in the embodiments of this application can include, but is not limited to, mobile terminals such as mobile phones, laptops, digital radio receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), and in-vehicle terminals (e.g., in-vehicle navigation terminals), as well as fixed terminals such as digital TVs and desktop computers. Figure 4 The AI ​​vision-based coking coal particle size counting device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0072] like Figure 4 As shown, the AI ​​vision-based coking coal particle size counting device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage device 1003 into a random access memory (RAM) 1004. The RAM 1004 also stores various programs and data required for the operation of the AI ​​vision-based coking coal particle size counting device. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the AI ​​vision-based coking coal particle size counting device to communicate wirelessly or wiredly with other devices to exchange data. Although various AI vision-based coking coal particle size counting devices are shown in the figures, it should be understood that implementation or possession of all of them is not required. More or fewer may be implemented alternatively.

[0073] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0074] The coking coal particle size counting device based on AI vision provided in this application, employing the coking coal particle size counting method based on AI vision in the above embodiments, can solve the technical problem of how to achieve automated online high-precision counting of particle size during continuous coking coal transportation. Compared with the prior art, the beneficial effects of the coking coal particle size counting device based on AI vision provided in this application are the same as those of the coking coal particle size counting method based on AI vision provided in the above embodiments, and other technical features in this coking coal particle size counting device based on AI vision are the same as those disclosed in the previous embodiment method, and will not be repeated here.

[0075] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0076] The above description is merely a specific embodiment of this application, but the scope of protection of this application 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 this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0077] This application provides a computer-readable medium having computer-readable program instructions (i.e., a computer program) stored thereon, which are used to execute the AI ​​vision-based coking coal particle size statistics method in the above embodiments.

[0078] The computer-readable medium provided in this application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor devices, or any combination thereof. More specific examples of computer-readable media may include, but are not limited to: electrical connections with one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable medium may be any tangible medium containing or storing a program that can be executed by instructions, used by a device, or used in conjunction with it. The program code contained on the computer-readable medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0079] The aforementioned computer-readable medium may be included in an AI vision-based coking coal particle size counting device; or it may exist independently and not assembled into an AI vision-based coking coal particle size counting device.

[0080] The aforementioned computer-readable medium carries one or more programs that, when executed by an AI vision-based coking coal particle size counting device, enable the AI ​​vision-based coking coal particle size counting device to write computer program code for performing the operations of this application in one or more programming languages ​​or a combination thereof. These programming languages ​​include object-oriented programming languages—such as Java, Smalltalk, and C++—and conventional procedural programming languages—such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0081] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of methods and computer program products according to various embodiments of this application. In this regard, all blocks in the flowcharts or block diagrams may represent a module, segment, or portion of code containing one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that all blocks in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using dedicated hardware-based implementations that perform the specified functions or operations, or using a combination of dedicated hardware and computer instructions.

[0082] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0083] The readable medium provided in this application is a computer-readable medium, which stores computer-readable program instructions (i.e., a computer program) for executing the above-described AI vision-based coking coal particle size statistics method. This solves the technical problem of how to achieve automated, online, high-precision particle size statistics during continuous coking coal transportation. Compared with the prior art, the beneficial effects of the computer-readable medium provided in this application are the same as those of the AI ​​vision-based coking coal particle size statistics method provided in the above embodiments, and will not be repeated here.

[0084] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the AI ​​vision-based coking coal particle size statistical method described above.

[0085] The computer program product provided in this application can solve the technical problem of how to achieve automated online high-precision statistical analysis of particle size during continuous coking coal transportation. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as those of the AI ​​vision-based coking coal particle size statistical method provided in the above embodiments, and will not be repeated here.

[0086] The above description is only a part of the embodiments of this application and does not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A method for statistical analysis of coking coal particle size based on AI vision, characterized in that, The method includes: Acquire images of coking coal conveying and error compensation data captured by camera equipment; The coordinates of the pixel positions in the coking coal conveying image are converted to obtain the pixel physical mapping relationship; The image of the coking coal being transported is input into a target detection model to determine the presence of particles, and the result of particle presence is obtained. When the particle presence result indicates the presence of coking coal particles, the coking coal conveying image is segmented into particle contours, and the particle contour coordinates are converted according to the pixel physical mapping relationship to obtain the actual contour coordinates of each coking coal particle. The equivalent particle size of each coking coal particle is calculated based on its actual contour coordinates, and the mass percentage of each particle size range is determined based on its equivalent particle size. The equivalent particle size of each coking coal particle and the mass ratio of each particle size range are corrected based on the error compensation data to obtain the statistical results of coking coal particle size.

2. The method as described in claim 1, characterized in that, The steps of calculating the equivalent particle size of each coking coal particle based on its actual contour coordinates, and determining the mass percentage of each particle size range based on its equivalent particle size, include: Based on the actual contour coordinates of each coking coal particle, the maximum abscissa, minimum abscissa, maximum ordinate, and minimum ordinate of the contour of each coking coal particle are extracted to obtain the physical corner points of the circumscribed rectangle of each coking coal particle. The length of the diagonal of the outer rectangle of each coking coal particle is calculated based on the physical corner points of the outer rectangle of each coking coal particle. The equivalent particle size of each coking coal particle is calculated based on the length of the diagonal of the circumscribed rectangle of each coking coal particle and the preset equivalent conversion coefficient, and the equivalent particle size of each coking coal particle is obtained. The preset equivalent conversion coefficient is the ratio of the length of the diagonal of the circumscribed rectangle to the side length of the equivalent cube. The equivalent particle size of each coking coal particle is classified according to the preset particle size range to obtain a particle set for each particle size range. Based on the particle set of each particle size range, the cubic equivalent particle size of each coking coal particle is used as the volume reference value of each coking coal particle. The proportion of the total volume reference value of all particles in each particle size range to the total volume reference value of all particles is calculated to obtain the volume ratio of each particle size range. Based on the uniform density of coking coal particles, the volume ratio of each particle size range is taken as the mass ratio of each particle size range.

3. The method as described in claim 1, characterized in that, The step of correcting the equivalent particle size and mass percentage of each coking coal particle based on the error compensation data to obtain the statistical results of coking coal particle size includes: The detection deviation value of each particle size interval is calculated based on the manual standard sieving data in the error compensation data and the mass ratio of each particle size interval, and the deviation data of each particle size interval is obtained. According to the preset weight allocation rules, the corresponding weight coefficients are assigned to each particle size range to obtain the compensation weight of each particle size range. Based on the deviation data of each particle size interval and the compensation weight of each particle size interval, the comprehensive compensation value is calculated by weighted average calculation to obtain the comprehensive compensation value of the particle size interval. Based on the comprehensive compensation value of the particle size interval, the equivalent particle size of each coking coal particle falling into each particle size interval is corrected for interval assignment deviation to obtain the corrected equivalent particle size set. Based on the comprehensive compensation value of the particle size range, the mass ratio of each particle size range is corrected for proportional deviation to obtain the corrected mass ratio set. Based on the corrected equivalent particle size set and the corrected mass percentage set, the statistical results of coking coal particle size are generated.

4. The method as described in claim 1, characterized in that, The steps for acquiring the coking coal conveying images and error compensation data collected by the camera equipment include: Send an image acquisition command to the camera device so that the camera device can acquire images of coking coal conveying according to a preset image frame interval; Receive the video stream returned by the camera device, and extract the current image frame from the video stream to obtain the coking coal conveying image; A predetermined number of particle images are randomly selected from the collected images of coking coal conveying. The particle size and mass percentage of the actual coking coal particles in the corresponding images are measured and statistically analyzed using a manual standard sieving method to obtain error compensation data, which includes the manual standard sieving data.

5. The method as described in claim 1, characterized in that, The step of performing coordinate conversion on the pixel positions in the coking coal conveying image to obtain the pixel physical mapping relationship includes: The grid size of the checkerboard calibration plate is determined according to the preset detection accuracy level, and the specification parameters of the calibration plate are obtained. Adjust the position of the checkerboard calibration board according to the calibration board specifications to obtain the calibration board placement state; The camera device is controlled to acquire images of the calibration board from different angles and positions based on the placement state of the calibration board, thereby obtaining a multi-view calibration image set; Based on the multi-view calibration image set, the intrinsic parameter matrix, extrinsic parameter matrix, and distortion coefficients of the camera device are calculated using the Zhang Zhengyou calibration algorithm to obtain the initial calibration parameters. The coordinate transformation accuracy of the preset verification point set is verified based on the initial calibration parameters to obtain the transformation error verification result. When the conversion error verification result is less than the preset error threshold, the initial calibration parameters are determined as camera device calibration data, and a pixel physical mapping relationship is established based on the camera device calibration data.

6. The method as described in claim 1, characterized in that, The step of inputting the coking coal conveying image into the target detection model to determine the presence of particles and obtain the particle presence result includes: The image frame to be detected is extracted from the coking coal conveying image according to the preset detection time interval to obtain the current detection frame; The current detection frame is input into the target detection model for multi-scale feature extraction to obtain a multi-scale feature map; Based on the multi-scale feature map, candidate boxes are generated and confidence scores are assigned to the particle regions in the current detection frame to obtain a set of particle candidate boxes and a corresponding set of confidence scores. The set of confidence scores is filtered based on a preset confidence threshold, and candidate boxes with confidence scores greater than the preset confidence threshold are retained as valid granular regions, thus obtaining a set of valid granular regions. Based on the set of effective particle regions, it is determined that there are coking coal particles in the current detection frame, and the particle presence result is obtained.

7. The method as described in claim 1, characterized in that, The step of performing particle contour segmentation on the coking coal conveying image and converting the particle contour coordinates according to the pixel physical mapping relationship to obtain the actual contour coordinates of each coking coal particle when the particle presence result is coking coal particles includes: When the result indicates the presence of coking coal particles, the coking coal conveying image is input into the particle segmentation model. Contour segmentation and extraction are performed on each coking coal particle to obtain the initial set of contour pixel coordinates for each particle. The initial contour pixel coordinate set is subjected to abnormal contour removal processing according to the preset contour filtering rules to obtain the contour pixel coordinate set of each particle. Based on the pixel physical mapping relationship, the pixel coordinates in the set of outline pixel coordinates of each particle are converted into actual physical coordinates one by one to obtain the actual outline coordinates of each coking coal particle.

8. A coking coal particle size counting device based on AI vision, characterized in that, The device includes: The data acquisition module is used to acquire images of coking coal conveying and error compensation data collected by the camera equipment; The coordinate conversion module is used to perform coordinate conversion on the pixel positions in the coking coal conveying image to obtain the pixel physical mapping relationship; The particle detection module is used to input the coking coal conveying image into the target detection model to determine the presence of particles and obtain the particle presence result; The contour segmentation module is used to perform particle contour segmentation on the coking coal conveying image when the particle presence result is that coking coal particles are present, and to convert the particle contour coordinates according to the pixel physical mapping relationship to obtain the actual contour coordinates of each coking coal particle. The particle size statistics module is used to calculate the equivalent particle size of each coking coal particle based on the actual contour coordinates of each coking coal particle, and to determine the mass percentage of each particle size range based on the equivalent particle size of each coking coal particle. The error correction module is used to correct the equivalent particle size of each coking coal particle and the mass ratio of each particle size interval based on the error compensation data, so as to obtain the statistical results of the coking coal particle size.

9. A coking coal particle size counting device based on AI vision, characterized in that, The device includes: a memory, a processor, and an AI vision-based coking coal particle size statistics program stored in the memory and running on the processor, the AI ​​vision-based coking coal particle size statistics program being configured to implement the steps of the AI ​​vision-based coking coal particle size statistics method as described in any one of claims 1-7.

10. A storage medium, characterized in that, The storage medium stores a coking coal particle size statistics program based on AI vision. When the coking coal particle size statistics program based on AI vision is executed by the processor, it implements the steps of the coking coal particle size statistics method based on AI vision as described in any one of claims 1-7.

Images