A belt conveying ore blockiness lightweight identification method and system

Abstract

The application relates to a kind of conveyor belt ore block size lightweight identification method and system, the method comprises: obtaining the motion information of conveyor belt and collecting moving ore image to obtain ore image sequence according to it;Determine the motion blur kernel according to motion information, and the non-blind deblurring is carried out to image sequence;Ore block identification is carried out to deblurred image, and cross-frame association is obtained to obtain ore block trajectory;According to the trajectory, the multiple frame identification results of the same ore block are deduplicated;The profile of the blocked ore block is completed, and the block size statistical weight is corrected;After mapping each ore block size to physical size, the block size distribution is obtained by statistics.The application can still accurately obtain the ore block size distribution when the conveyor belt is moving at high speed and the material flow is stacked and blocked by deblurring, cross-frame deduplication and blocking correction.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and in particular to a method and system for lightweight identification of ore blocks on a conveyor belt. Background Technology

[0002] In the primary crushing stage, ore is transported by conveyor belt. The distribution of ore block size is an important basis for evaluating the crushing effect and adjusting subsequent operations. Therefore, it is necessary to perform online block size identification and statistics on the moving ore on the conveyor belt. Existing methods for ore block size identification on conveyor belts mostly involve identifying blocks frame by frame in the acquired images and then statistically analyzing the block size distribution based on the identification results.

[0003] However, the high-speed movement and stacking of ore on the conveyor belt make the block size distribution obtained by existing methods unreliable. Because the ore moves at high speed with the conveyor belt, motion blur occurs during exposure of a single frame, making it difficult to accurately identify block boundaries. Furthermore, because each frame is counted independently, the same ore block may appear repeatedly in multiple consecutive frames, causing statistical bias in the block size distribution. Additionally, because the material is stacked on the conveyor belt, ore blocks at the bottom are obscured by the top layers and cannot be fully identified, causing the statistically calculated block size distribution to favor larger blocks. The combined effects of motion blur, repeated statistics, and obscured statistics ultimately lead to unreliable block size distribution measurements of the moving material flow.

[0004] Therefore, it is desirable to propose a lightweight identification method for conveyor belt ore size that can accurately obtain the ore size distribution under the working conditions of high-speed conveyor belt movement and material flow stacking and obstruction. Summary of the Invention

[0005] In order to accurately obtain the ore block size distribution under the working conditions of high-speed conveyor belt movement and material flow stacking and obstruction, this application provides a lightweight identification method and related device for ore block size in conveyor belts.

[0006] Firstly, this application provides a method for lightweight identification of ore block size on a conveyor belt, which adopts the following technical solution: A method for lightweight identification of ore blocks on a conveyor belt includes the following steps: S1. Acquiring motion information of the conveyor belt and collecting images of moving ore on the conveyor belt according to the motion information to obtain an ore image sequence; S2. Determining a motion blur kernel based on the motion information and performing non-blind deblurring on the ore image sequence based on the motion blur kernel to obtain a sharpened image sequence; S3. Identifying ore blocks in the sharpened image sequence to obtain the identification results of each frame, and performing cross-frame association of the identification results of adjacent frames based on the motion information to obtain the trajectory of the ore blocks; S4. Identifying the same ore block in multiple frames based on the trajectory of the ore blocks. The identification results are associated and deduplicated, so that each ore block is counted in the block size statistics only once, resulting in a deduplicated set of ore blocks; S5. For the obscured ore blocks in the deduplicated set of ore blocks, the complete size is completed based on the visible outline and convex prior of the obscured ore blocks, and the block size statistics weight is corrected according to the position of the ore block at the top or bottom of the material flow on the conveyor belt, resulting in a corrected set of ore blocks; S6. The size of each ore block in the corrected set of ore blocks is mapped to the physical size, the block size distribution of the ore on the conveyor belt is statistically obtained, and the block size distribution is output.

[0007] By adopting the above technical solution, the motion information of the conveyor belt is used as a priori throughout the entire process. First, the motion blur kernel is parsed from the motion information to perform non-blind deblurring of the image sequence without blind kernel estimation, so that real-time processing at high frame rate and clear boundary can be taken into account. Then, an inter-frame displacement model is established from the motion information to associate and deduplicate the recognition results of the same ore block in multiple consecutive frames, eliminating the repeated statistical bias caused by independent statistics frame by frame. Furthermore, the size of the occluded ore block is completed with visible contour and convex priors and the statistical weight is corrected according to the top and bottom layer positions to eliminate the distribution bias of large blocks caused by material flow stacking. Thus, an accurate ore block size distribution is obtained under the conditions of high-speed movement and stacking occlusion.

[0008] Optionally, in S4, the identification results of the same ore block in multiple frames are associated and deduplicated based on the trajectory of the ore block, including: establishing an inter-frame displacement model based on motion information, determining that the identification results in adjacent frames whose positions conform to the inter-frame displacement model belong to the same ore block, and using only one size for the identification results of multiple frames that are determined to belong to the same ore block in the block size statistics.

[0009] By adopting the above technical solution, the inter-frame displacement model established by the uniform motion of the conveyor belt is used as the identity criterion, so that the deduplication is based on the laws of physical motion rather than appearance similarity, thereby improving the reliability of deduplication in high-speed material flow.

[0010] Optionally, the motion information includes the conveyor belt's speed. S1 includes sub-steps S11-S13: S11. Obtain the conveyor belt's speed. S12. Adjust the image acquisition exposure time and frame rate according to the speed to ensure that the displacement of the ore block within a single frame does not exceed a preset displacement threshold. S13. Acquire images of the moving ore on the conveyor belt according to the adjusted exposure time and frame rate to obtain an ore image sequence.

[0011] By adopting the above technical solution, the exposure time and frame rate are inversely solved by using the displacement within a single frame as a constraint, so that motion blur is suppressed within a controllable range during the acquisition stage, providing clearer input for subsequent deblurring and recognition.

[0012] Optionally, in S2, the motion blur kernel is determined based on the motion information, including: determining the direction of the motion blur kernel as the running direction of the conveyor belt, and determining the length of the motion blur kernel as the displacement length obtained based on the motion speed and exposure time.

[0013] By adopting the above technical solution, the direction and length of the blur kernel can be directly obtained from the known motion speed and exposure time, so that the deblurring is transformed from blind estimation to non-blind processing with known parameters, saving the iterative overhead of kernel estimation and meeting the real-time requirement.

[0014] Optionally, S3 includes sub-steps S31-S32: S31. Using a lightweight recognition model incorporating an attention mechanism, ore block recognition is performed on the sharpened image sequence to obtain the recognition results for each frame. S32. Based on motion information, the position of each ore block in adjacent frames is predicted. Temporal consistency is checked on the recognition results of adjacent frames, and recognition results that do not meet the temporal consistency requirements are removed. The recognition results that pass the check are then correlated across frames to obtain the trajectory of the ore block.

[0015] By adopting the above technical solutions, a lightweight recognition model is used to ensure real-time performance, and motion information-driven temporal consistency verification is used to eliminate misidentifications, making the associated ore block trajectory more stable.

[0016] Optionally, in S6, the dimensions of each ore block in the corrected ore block set are mapped to physical dimensions, including: establishing a mapping relationship from pixel dimensions to physical dimensions based on the known size reference of the conveyor belt and the installation angle and installation height configuration parameters of the image acquisition device, and mapping the dimensions of each ore block to physical dimensions according to the mapping relationship.

[0017] By adopting the above technical solution, and by using the known reference of the conveyor belt and the correspondence between the pixels and physical dimensions of the acquisition device installation configuration, the block size distribution can be expressed in terms of physical dimensions, and adapt to changes in the installation position of the acquisition device.

[0018] Optionally, the block size distribution is statistically obtained in S6, including: using a preset length of material flow on the conveyor belt as the statistical unit, and using a sliding window to incrementally update the block size distribution, which includes a cumulative distribution curve and a uniformity index.

[0019] By adopting the above technical solution, the block size distribution is updated incrementally for continuous material flow using a sliding window, so that the cumulative distribution curve and uniformity index are output in real time with the material flow, adapting to continuous operation of the conveyor belt.

[0020] Optionally, images of the moving ore on the conveyor belt are acquired by a high-speed camera with a global shutter.

[0021] By adopting the above technical solution, the motion distortion caused by progressive exposure is avoided by using a global shutter, and motion blur is further suppressed at the hardware level.

[0022] Optionally, the image acquisition device for performing image acquisition is mounted on a vibration isolation base and housed in a dustproof and waterproof enclosure with constant temperature and heat dissipation.

[0023] By adopting the above technical solutions, conveyor belt vibration is isolated to reduce imaging offset and adapt to the on-site environment of high temperature, high humidity and high dust.

[0024] Optionally, the image acquisition device for performing image acquisition is mounted above the conveyor belt via a modular mounting bracket with adjustable angle and height.

[0025] By adopting the above technical solution, the field of view can be adapted to conveyor belts of different widths, and known installation configuration parameters are provided for the calibration of pixels and physical dimensions.

[0026] Optionally, the block size distribution can be visualized on a web page and exported as a report file.

[0027] By adopting the above technical solutions, the block size distribution can be easily viewed and preserved on-site.

[0028] Optionally, the identification model used for ore block identification, the equipment for performing image acquisition, and the user interface for providing the display are configured to be upgraded independently.

[0029] By adopting the above technical solutions, each part can be maintained and updated independently without affecting the overall operation.

[0030] Optionally, a coarse detection is first performed on the sharpened image sequence to locate high-density areas where the density of the ore blocks exceeds a preset density, and then the high-density areas are cropped, upsampled, and fine detection is performed.

[0031] By adopting the above technical solution, the resolution of small dense areas is locally improved under the full field of view, thereby improving the recognition accuracy of small objects.

[0032] Optionally, the dust haze level can be estimated based on the contrast of the images in the sharpened image sequence, and the detection threshold for ore block recognition can be adjusted based on the dust haze level.

[0033] By adopting the above technical solution, the recognition can be adaptively adjusted according to the change of dust concentration on site, thus coping with the imaging degradation under dust conditions.

[0034] Optionally, the overall block size distribution can be estimated based on the area ratio of material flow and texture spectrum in the ore image sequence, and the overall block size distribution can be cross-validated with the statistically obtained block size distribution.

[0035] By adopting the above technical solution, the overall statistical path and the individual identification path are used to corroborate each other, and abnormal statistical results are covered.

[0036] Optionally, the feed rate or discharge port size of the upstream crusher on the conveyor belt can be adjusted based on the block size distribution feedback.

[0037] By adopting the above technical solution, the block size measurement results are used to regulate the crushing operation in a closed loop, so that the crushing effect is dynamically optimized according to the block size distribution.

[0038] Secondly, this application provides a lightweight identification system for ore block size on conveyor belts, which adopts the following technical solution: A lightweight ore block identification system for conveyor belts includes: a motion acquisition module configured to acquire motion information of the conveyor belt; an image acquisition module connected to the motion acquisition module configured to acquire images of moving ore on the conveyor belt according to the motion information, obtaining an ore image sequence; a deblurring module connected to the image acquisition module configured to determine a motion blur kernel based on the motion information, and perform non-blind deblurring on the ore image sequence based on the motion blur kernel, obtaining a sharpened image sequence; an identification association module connected to the deblurring module configured to perform ore block identification on the sharpened image sequence to obtain the identification results of each frame, and perform cross-frame association on the identification results of adjacent frames based on the motion information to obtain the ore block trajectory; and a deduplication module connected to the identification association module configured to... The system associates and deduplicates the identification results of the same ore block across multiple frames based on the ore block trajectory, ensuring that each ore block is counted in the block size statistics only once, resulting in a deduplicated set of ore blocks. An occlusion correction module, connected to the deduplication module, is configured to complete the size of occluded ore blocks in the deduplicated set based on their visible outlines and convex priors. It also corrects the block size statistics weights based on the ore block's position at the top or bottom of the material flow on the conveyor belt, resulting in a corrected set of ore blocks. A block size statistics module, connected to the occlusion correction module, is configured to map the size of each ore block in the corrected set to its physical size, statistically analyze the block size distribution of the ore on the conveyor belt, and output the block size distribution.

[0039] Thirdly, the computer device provided in this application adopts the following technical solution: A computer device comprising: One or more processors; Memory; One or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the one or more processors, the one or more applications being configured to: The above-described method for lightweight identification of ore blocks on conveyor belts is applied.

[0040] Fourthly, the computer-readable storage medium provided in this application adopts the following technical solution: A computer-readable storage medium storing a computer program that can be loaded by a processor and executed as described above.

[0041] The storage medium stores at least one instruction, at least one program, a code set, or an instruction set, wherein the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by the processor to implement the following: The above-mentioned method for identifying lightweight ore blocks on conveyor belts.

[0042] In summary, this application includes at least one of the following beneficial technical effects: 1. Using conveyor belt motion information as a unified prior for deblurring, cross-frame association and deduplication, motion fuzzing is suppressed into non-blind processing with known parameters, and the recurrence of multiple frames of the same ore block is associated for deduplication, so that the block size distribution on the moving material flow can both eliminate the statistical bias of repetition and maintain clear boundary identification.

[0043] 2. Complete the size of the obscured ore blocks with visible outlines and convex shapes a priori, and adjust the statistical weights according to their position at the top or bottom layer in the material flow to correct the systematic bias of the block size distribution that is biased towards larger blocks due to the stacking of the material flow.

[0044] 3. By combining speed-adaptive exposure, lightweight recognition model and sliding window incremental statistics, and supplemented by coarse and fine detection focusing, haze adaptive parameter adjustment and block size distribution closed-loop feedback on the crusher, the method maintains real-time performance under high-speed and dusty conditions, and the measurement results are used to optimize crushing operations. Attached Figure Description

[0045] Figure 1 This is a flowchart illustrating a method for identifying lightweight ore blocks on a conveyor belt, provided in some embodiments of this application. Figure 2 This is a structural block diagram of a lightweight ore block size identification system for conveyor belts provided in some embodiments of this application; Figure 3 This is a schematic diagram of the sub-processes for cross-frame deduplication and occlusion correction provided in some embodiments of this application; Figure 4 This is a field deployment diagram of the conveyor belt ore block size lightweight identification system provided in some embodiments of this application; Figure 5 This is a schematic diagram of the cumulative distribution curve of the block size distribution provided in some embodiments of this application.

[0046] Explanation of reference numerals in the attached figures: 100. Lightweight ore block size identification system for conveyor belts; 110. Motion acquisition module; 120. Image acquisition module; 130. Deblurring module; 140. Recognition and association module; 150. Deduplication module; 160. Occlusion correction module; 170. Block size statistics module; 200. Conveyor belt; 210. Image acquisition device; 220. Modular mounting bracket; 230. Vibration isolation base; 240. Dustproof and waterproof enclosure; 250. Computer equipment. Detailed Implementation

[0047] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. The described embodiments are only some embodiments of this application, and not all embodiments.

[0048] Before detailing the embodiments of this application, some terms will be explained. "Block size" refers to the size of a single ore block, while "block size distribution" refers to the statistical distribution of the sizes of ore blocks in a batch of ore, serving as the basis for evaluating the crushing effect and adjusting subsequent operations during the initial crushing stage. "Uniformity index" is an indicator that measures the dispersion of block size distribution, and "cumulative distribution curve" is the cumulative percentage curve of ore block sizes from smallest to largest.

[0049] Motion information refers to the operating status information of the conveyor belt, including at least its speed and possibly its direction. The motion blur kernel describes the direction and degree of blurring caused by the movement of the subject during exposure. Non-blind deblurring refers to the restoration process performed on the image when the blur kernel is known, which is different from blind deblurring, which requires the blur kernel to be estimated first.

[0050] The inter-frame displacement model describes the positional change of the same ore block between adjacent image frames. The ore block trajectory refers to the positional sequence of the same ore block in multiple consecutive image frames as it moves with the conveyor belt. The material flow refers to the material flow formed when ore is continuously stacked and transported on the conveyor belt.

[0051] Figure 1 A schematic flowchart of a lightweight ore block identification method for conveyor belts according to some embodiments of this application is shown. Figure 1As shown, the method includes steps S1 to S6. This method uses the motion information of the conveyor belt as prior information throughout the entire process. In S1, images of the moving ore are acquired; in S2, motion blur is removed using the motion information; in S3, ore blocks are identified and their trajectories are obtained through cross-frame correlation; in S4, duplicate ore blocks are removed using the motion information; in S5, obscured ore blocks are completed and statistical deviations are corrected; and in S6, physical dimensions are mapped and the block size distribution is statistically output. Each step is described in detail below.

[0052] S1. Obtain the motion information of the conveyor belt, and collect images of the moving ore on the conveyor belt according to the motion information to obtain an ore image sequence.

[0053] The ore is conveyed into the primary crushing stage via conveyor belt 200. An image acquisition device 210 is installed above the conveyor belt to image the ore moving on the conveyor belt frame by frame, obtaining a sequence of ore images composed of multiple frames. For example, the image acquisition device is a high-speed camera with a global shutter. The global shutter allows the entire frame to be exposed at the same time, avoiding image distortion caused by line-by-line exposure under high-speed motion, and suppressing motion blur at the hardware level.

[0054] Specifically, the motion information includes the speed of the conveyor belt, and S1 includes sub-steps S11 to S13.

[0055] S11. Obtain the speed of the conveyor belt. For example, the speed of the conveyor belt is provided in real time by a speed sensor installed on the conveyor belt.

[0056] S12. Adjust the exposure time and frame rate of image acquisition according to the motion speed to ensure that the displacement of the ore block within a single frame does not exceed a preset displacement threshold. The displacement of the ore block within a single frame is... ,in This refers to the displacement per frame. For the speed of movement, For example, the shorter the exposure time, the smaller the displacement and the weaker the motion blur. meters per second seconds, Meters, meaning a single-frame displacement of 2 millimeters; when When the speed is increased to 4 m / s, Shortened to 0.0005 seconds and the frame rate increased accordingly, making It remains at 0.002 meters, or 2 millimeters, which does not exceed the preset displacement threshold.

[0057] S13. Acquire images of the moving ore on the conveyor belt according to the adjusted exposure time and frame rate to obtain an ore image sequence.

[0058] The above describes one implementation of S1, namely, adaptively adjusting the exposure time and frame rate according to the motion speed. In other embodiments, S1 can also use a fixed high frame rate in conjunction with a global shutter for acquisition, that is, without adjusting the exposure parameters with speed, but using a sufficiently high fixed frame rate to control motion blur within an acceptable range. This alternative method can also obtain a sequence of mineral images for subsequent processing, and therefore also belongs to the implementation of S1.

[0059] S2. Determine the motion blur kernel based on the motion information, and perform non-blind deblurring on the ore image sequence based on the motion blur kernel to obtain a sharpened image sequence.

[0060] Since the ore moves at a constant linear speed along the conveyor belt, the motion blur in a single frame image has a definite direction and degree. The motion blur kernel can be directly obtained from the motion information without blindly estimating the blur kernel for each frame. Specifically, S2 includes sub-steps S21 to S22.

[0061] S21. Determine the motion blur kernel based on the motion information, set the direction of the motion blur kernel to the direction of the conveyor belt's movement, and set the length of the motion blur kernel to the displacement length obtained based on the motion speed and exposure time. The length of the motion blur kernel can be determined by the following formula: in, The length of the motion blur kernel, in pixels; The speed of the conveyor belt; The exposure time for image acquisition; This refers to the physical size corresponding to a single pixel. For example, let... meters per second Second, If the length of the motion blur kernel is meters per pixel, then the length of the motion blur kernel is: That is, the length of the motion blur kernel is 4 pixels.

[0062] S22. Perform non-blind deblurring on the ore image sequence based on the motion blur kernel to obtain a sharpened image sequence. For example, perform deconvolution operation on each frame image with a determined motion blur kernel to restore the sharpened image with suppressed motion blur. Since the motion blur kernel is directly given by the physical motion parameters, the deblurring process does not require blind estimation of the blur kernel iteration, and the processing speed can match the real-time requirements of high frame rate acquisition.

[0063] The above describes one implementation of S2, which involves non-blind deblurring after parsing the motion blur kernel from motion information. In other embodiments, S2 can also remove motion blur using multi-frame fusion. This involves acquiring images with shorter exposure times and images with normal exposure times for the same material flow area, fusing the clearer block boundaries in the shorter exposure image with the dark textures in the normal exposure image to obtain an image with suppressed motion blur. This alternative method also yields a sharpened image sequence and is therefore also an implementation of S2.

[0064] S3. Perform ore block recognition on the sharpened image sequence to obtain the recognition results of each frame, and perform cross-frame correlation on the recognition results of adjacent frames based on motion information to obtain the trajectory of the ore block.

[0065] Specifically, S3 includes sub-steps S31 to S32.

[0066] S31. A lightweight recognition model incorporating an attention mechanism is used to identify ore blocks in a sequence of sharpened images, yielding recognition results for each frame. This lightweight model maintains recognition accuracy while having a small number of parameters and computational load, meeting the real-time processing requirements of high-speed conveyor belt movement. The attention mechanism allows the model to focus more on the boundary regions of the ore blocks even in complex environments such as dust and vibration. The recognition results include the position and contour of each ore block in each frame.

[0067] In real-world data acquisition scenarios, the conveyor belt is relatively wide, while individual ore blocks occupy fewer pixels in the full-frame image, making it easy for the recognition accuracy of small blocks to decrease. In some embodiments, in step S31, a coarse detection is first performed on the clarified image to locate high-density areas where the density of ore blocks exceeds a preset density. Then, the high-density areas are cropped, upsampled, and subjected to fine detection, thereby improving the recognition accuracy of small blocks without significantly increasing the overall computational load.

[0068] The dust concentration at the site can affect the imaging quality depending on the working conditions. In some embodiments, the dust haze level is estimated based on the contrast of the sharpened image, and the detection threshold for ore block identification is adjusted according to the dust haze level, so that the identification is adaptive to changes in dust concentration. For example, the detection threshold can be appropriately reduced when the haze level is high to avoid missed detections.

[0069] S32. Based on motion information, predict the position of each ore block in adjacent frames. Perform temporal consistency verification on the recognition results of adjacent frames, discard recognition results that do not meet the temporal consistency requirement, and perform cross-frame association on the recognition results that pass the verification to obtain the ore block trajectory. Since the ore moves at a constant speed with the conveyor belt, the position change of the same ore block in adjacent frames conforms to the pattern given by the motion information. Based on this, its position in the next frame can be predicted. If a certain recognition result does not have a corresponding result in an adjacent frame that matches the predicted position, then the recognition result is judged as a misidentification and discarded, thereby improving the reliability of the associated ore block trajectory.

[0070] The above describes one implementation of ore block recognition in S3, namely, using a lightweight recognition model. In other embodiments, ore block recognition can also employ a combination of edge detection and watershed segmentation. This involves first extracting edges from the image, and then using a watershed algorithm to segment contacting ore blocks, obtaining the outlines of each ore block. This alternative method also yields recognition results for each frame and is therefore also an implementation of ore block recognition in S3.

[0071] Figure 3 A schematic diagram of the sub-processes for cross-frame deduplication and occlusion correction according to some embodiments of this application is shown. Figure 3 As shown, S4 and S5 together constitute the debiasing process for the block size distribution.

[0072] S4. Based on the trajectory of the ore block, the identification results of the same ore block in multiple frames are associated to remove duplicates, so that each ore block is counted in the block size statistics only once, and the deduplicated ore block set is obtained.

[0073] Since the same ore block will appear repeatedly in multiple consecutive frames of images as it moves along the conveyor belt, if the recognition results of each frame are counted independently, the size of the same ore block will be counted repeatedly, causing a statistical bias in the block size distribution. S4 removes duplicates from the multi-frame recognition results of the same ore block based on the ore block trajectory, including: establishing an inter-frame displacement model based on motion information, determining that recognition results in adjacent frames whose positions conform to the inter-frame displacement model belong to the same ore block, and only using the size to count the block size statistics once for the multi-frame recognition results determined to belong to the same ore block.

[0074] The inter-frame displacement model gives the positional offset of the same ore block between adjacent frames, and this offset is... ,in This is the inter-frame offset. For the speed of movement, This is the time interval between two adjacent frames. For example, If the speed is meters per second and the image acquisition frame rate is 100 frames per second, then Second, The measurement is calculated in meters, meaning the positional offset of the same ore block between adjacent frames along the direction of travel is 2 centimeters. If the position of an ore block in the first frame, after adding this offset, matches its position in the second frame, then the two identification results in the first and second frames are determined to belong to the same ore block. For example, if an ore block appears in five consecutive frames, its size is counted five times before deduplication, but only once after deduplication, thus eliminating the bias caused by the ore block being counted four times repeatedly.

[0075] To further improve the reliability of block size distribution, in some embodiments, the overall block size distribution is estimated based on the area ratio of material flow and texture spectrum in the ore image sequence. This overall block size distribution is then cross-validated with the block size distribution obtained by statistical analysis of the deduplicated ore block set. When the deviation between the two exceeds a preset range, a prompt is made to review the current statistical results, thereby providing a safety net for abnormal results in individual identification statistics.

[0076] The above describes one implementation of S4, namely, deduplication based on an inter-frame displacement model. In other embodiments, S4 can also deduplicatize based on the spatiotemporal overlap of the detection boxes of each recognition result in adjacent frames, that is, determining that detection boxes in adjacent frames with an overlap exceeding a preset overlap threshold belong to the same ore block. This alternative method can also obtain a deduplicated set of ore blocks, and therefore also belongs to the implementation of S4.

[0077] S5. For the obscured ore blocks in the deduplicated ore block set, complete the full size based on the visible outline and convex prior of the obscured ore blocks, and correct the block size statistical weight according to the position of the ore block at the top or bottom of the material flow on the conveyor belt, to obtain the corrected ore block set.

[0078] As the material flows and stacks on the conveyor belt, the ore blocks at the bottom are partially obscured by the ore blocks at the top, only revealing their outlines in the image. Because smaller ore blocks are more likely to be completely obscured by adjacent larger ore blocks and thus missed in the image, directly calculating the size based on the visible outlines will underestimate the size of the obscured ore blocks. Furthermore, some ore blocks will be missed due to complete obscuration, causing the statistically calculated block size distribution to be biased towards larger blocks.

[0079] S5 performs size completion on the occluded ore blocks. Since the ore block's shape is approximately convex, a complete outline can be fitted from the visible contour, estimating the size of the occluded portion and obtaining the complete size of the ore block. For example, if only about 70% of the outline of an occluded ore block is visible in the image, the complete size obtained after fitting and completing the image based on its visible outline and convex prior is larger than the size obtained solely based on the visible outline, thus avoiding an underestimation of the ore block's size.

[0080] S5, on the other hand, corrects the statistical weighting of ore size. Ore blocks at different layers in the material flow have varying probabilities of being collected; top-layer ore blocks are more likely to be collected intact, while bottom-layer ore blocks are more likely to be obscured and missed. Therefore, the statistical weighting of top-layer ore blocks is lowered, while the statistical weighting of bottom-layer ore blocks is increased. This allows the statistical weighting to compensate for the differences in collection probability between ore blocks at different layers, correcting the systematic bias of ore size distribution towards larger blocks, resulting in a corrected set of ore blocks.

[0081] The above describes one implementation of size completion in S5, namely, completion based on a priori fitting of the visible contour and convex shape. In other embodiments, size completion can also extrapolate the complete size based on the visible area of ​​the obscured ore block according to an empirically visible ratio, where the complete area is calculated according to... It is estimated that among them For the complete area, The visible area, This is the scale visible from experience. This alternative method can also complete the size of the obscured ore block, and therefore also belongs to the implementation method of size completion in S5.

[0082] S6. Map the size of each ore block in the corrected ore block set to the physical size, statistically obtain the block size distribution of the ore on the conveyor belt, and output the block size distribution.

[0083] S6 maps pixel dimensions to physical dimensions. The identified ore block dimensions, expressed in pixels, need to be converted to physical units such as millimeters to obtain a meaningful block size distribution. Specifically, a mapping relationship between pixel dimensions and physical dimensions is established based on the known dimensions of the conveyor belt and the installation angle and height configuration parameters of the image acquisition device. The dimensions of each ore block are then mapped to their physical dimensions according to this mapping relationship. The physical dimension corresponding to a single pixel is... ,in This is the actual bandwidth. This refers to the number of pixels occupied by the bandwidth. For example, using the known bandwidth of a conveyor belt as a size reference, millimeters At pixel level, mm / pixel; when the installation height or angle of the image acquisition device is adjusted, the mapping relationship is updated accordingly to adapt the mapping result to the change in installation configuration.

[0084] The above describes one implementation of pixel-to-physical-size mapping, namely calibration based on a known reference point on the conveyor belt and installation configuration parameters. In other embodiments, a calibration plate of known size can be placed on the conveyor belt, and a mapping relationship can be established between the pixels occupied by the calibration plate in the image and its actual size. This alternative method can also map pixel size to physical size, and therefore also belongs to the implementation of pixel size mapping.

[0085] S6, on the other hand, outputs the block size distribution. In some embodiments, a predetermined length of material flow on the conveyor belt is used as the statistical unit, and a sliding window is used to incrementally update the block size distribution, which includes a cumulative distribution curve and a uniformity index. As the material flow is continuously conveyed on the conveyor belt, the sliding window moves forward accordingly. The size of newly entering ore blocks is included, and the size of ore blocks moving out of the window is removed, thereby incrementally updating the block size distribution. This allows the cumulative distribution curve and uniformity index to be output in real time along with the material flow, adapting to the continuous operation of the conveyor belt. Figure 5 A schematic diagram of the cumulative distribution curve of block size distribution according to some embodiments of this application is shown, where the horizontal axis is the physical size of the ore block and the vertical axis is the cumulative proportion of ore blocks not larger than the corresponding size.

[0086] Once the block size distribution is obtained, it can be output. For example, the block size distribution can be visualized as a granularity distribution map on a web page and exported as a report file in PDF or Excel format for easy on-site viewing and archiving.

[0087] Figure 2 A structural block diagram of a lightweight ore block size identification system for conveyor belts according to some embodiments of this application is shown. Figure 2 As shown, the conveyor belt ore block size lightweight identification system 100 includes a motion acquisition module 110, an image acquisition module 120, a deblurring module 130, an identification association module 140, a deduplication module 150, an occlusion correction module 160, and a block size statistics module 170.

[0088] Motion acquisition module 110 is configured to acquire motion information of the conveyor belt. Image acquisition module 120 is connected to motion acquisition module 110 and configured to acquire images of moving ore on the conveyor belt according to the motion information, obtaining an ore image sequence. Deblurring module 130 is connected to image acquisition module 120 and configured to determine a motion blur kernel based on the motion information, and perform non-blind deblurring on the ore image sequence based on the motion blur kernel, obtaining a sharpened image sequence. Recognition association module 140 is connected to deblurring module 130 and configured to perform ore block recognition on the sharpened image sequence to obtain the recognition results of each frame, and perform cross-frame association on the recognition results of adjacent frames according to the motion information, obtaining the ore block trajectory. Deduplication module 150 is connected to recognition association module 140 and configured to associate and deduplicate the recognition results of the same ore block in multiple frames according to the ore block trajectory, obtaining a deduplicated set of ore blocks. The occlusion correction module 160 is connected to the deduplication module 150 and is configured to complete the dimensions of occluded ore blocks in the deduplicated ore block set and correct the block size statistical weights to obtain a corrected ore block set. The block size statistics module 170 is connected to the occlusion correction module 160 and is configured to map the dimensions of each ore block in the corrected ore block set to physical dimensions, statistically obtain the block size distribution of the ore on the conveyor belt, and output it. The functions of each module correspond to the steps of the above method. The motion acquisition module 110 and the image acquisition module 120 jointly implement S1, while the deblurring module 130, the recognition and association module 140, the deduplication module 150, the occlusion correction module 160, and the block size statistics module 170 respectively implement S2 to S6.

[0089] In some embodiments, the recognition model used by the recognition association module 140, the acquisition device corresponding to the image acquisition module 120, and the user interface used to display the block size distribution are configured to be upgraded independently, so as to maintain and update each part separately without affecting the overall operation of the system.

[0090] Figure 4 A schematic diagram of the field deployment of a lightweight ore block size identification system for conveyor belts according to some embodiments of this application is shown. Figure 4 As shown, the image acquisition device 210 is mounted above the conveyor belt 200 via a modular mounting bracket 220, facing the ore being transported on the conveyor belt 200.

[0091] The modular mounting bracket 220 can adjust the mounting angle and height of the image acquisition device 210 to adapt the field of view of the image acquisition device 210 to conveyor belts 200 with different bandwidths. For example, the modular mounting bracket 220 has elongated holes. The mounting height is adjusted by allowing the image acquisition device 210 to slide along the elongated holes through the engagement of bolts. The mounting angle is adjusted by rotating the hinge structure at the bracket connection. In addition to the method of using bolts with elongated holes, the mounting height can also be adjusted by snap-fit, threaded telescopic, etc., which are not limited here.

[0092] The image acquisition device 210 is mounted on a vibration isolation base 230, which absorbs the vibrations generated by the conveyor belt 200, reducing imaging shift caused by vibration. The image acquisition device 210 is housed in a dustproof and waterproof enclosure 240 with constant temperature and heat dissipation, adapting to the high temperature, high humidity, and dusty environment of the initial demolition site. The images acquired by the image acquisition device 210 are transmitted to a computer device 250, which performs the aforementioned deblurring, identification and correlation, deduplication, occlusion correction, and block size statistical processing.

[0093] In some embodiments, the feed rate or discharge port size of the crusher upstream of the conveyor belt is adjusted based on the block size distribution feedback, forming a closed loop between block size measurement and crushing control. For example, when the block size distribution shows that the proportion of larger blocks is too high, the discharge port size is reduced or the feed rate is lowered so that the block size distribution of the crushed ore tends to the target distribution, thereby dynamically optimizing the crushing effect based on the block size measurement results.

[0094] Based on the same inventive concept, this application also provides a computer device, which includes one or more processors and a memory. The memory stores one or more application programs, and when the one or more application programs are executed by the one or more processors, they implement the above-described conveyor belt ore block lightweight identification method.

[0095] This application also provides a computer-readable storage medium storing a computer program thereon. When the computer program is loaded and executed by a processor, it implements the aforementioned method for lightweight identification of ore blocks on a conveyor belt. The computer-readable storage medium may be a read-only memory, a random access memory, a magnetic disk, or an optical disk, etc.

[0096] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for lightweight identification of ore block size on a conveyor belt, characterized in that, Includes the following steps: S1. Obtain the motion information of the conveyor belt, and collect images of the moving ore on the conveyor belt according to the motion information to obtain an ore image sequence; S2. Determine the motion blur kernel based on the motion information, and perform non-blind deblurring on the ore image sequence based on the motion blur kernel to obtain a sharpened image sequence; S3. Perform ore block recognition on the sharpened image sequence to obtain the recognition results of each frame, and perform cross-frame association on the recognition results of adjacent frames based on the motion information to obtain the trajectory of the ore block; S4. Based on the trajectory of the ore block, the identification results of the same ore block in multiple frames are associated and deduplicated, so that each ore block is counted in the block size statistics only once, and a set of deduplicated ore blocks is obtained. S5. For the obscured ore blocks in the deduplicated ore block set, complete the full size based on the visible outline and convex prior of the obscured ore blocks, and correct the block size statistical weight based on the position of the ore blocks at the top or bottom of the material flow on the conveyor belt, to obtain the corrected ore block set. S6. Map the size of each ore block in the corrected ore block set to physical size, statistically obtain the block size distribution of the ore on the conveyor belt, and output the block size distribution.

2. The method for identifying lightweight ore blocks on a conveyor belt according to claim 1, characterized in that, The step S4 involves associating and deduplicating the identification results of the same ore block in multiple frames based on the trajectory of the ore block, including: establishing an inter-frame displacement model based on the motion information; determining that the identification results in adjacent frames whose positions conform to the inter-frame displacement model belong to the same ore block; and using only one size count for the block size statistics of multiple frame identification results that belong to the same ore block.

3. The method for lightweight identification of ore blocks in a conveyor belt according to claim 1, characterized in that, The motion information includes the speed of the conveyor belt, and step S1 includes the following sub-steps: S11. Obtain the speed of the conveyor belt; S12. Adjust the exposure time and frame rate of image acquisition according to the movement speed so that the displacement of the ore block in a single frame image does not exceed a preset displacement threshold. S13. Acquire images of the moving ore on the conveyor belt according to the adjusted exposure time and frame rate to obtain the ore image sequence.

4. The method for lightweight identification of ore blocks in a conveyor belt according to claim 3, characterized in that, In step S2, determining the motion blur kernel based on the motion information includes: determining the direction of the motion blur kernel as the running direction of the conveyor belt, and determining the length of the motion blur kernel as the displacement length obtained based on the motion speed and the exposure time.

5. The method for identifying lightweight ore blocks on a conveyor belt according to claim 1, characterized in that, S3 includes the following sub-steps: S31. A lightweight recognition model incorporating an attention mechanism is used to identify mineral blocks in the sharpened image sequence, and the recognition results for each frame are obtained. S32. Based on the motion information, predict the position of each ore block in adjacent frames, perform temporal consistency verification on the recognition results of adjacent frames, eliminate recognition results that do not meet temporal consistency, and associate the recognition results that pass the verification across frames to obtain the trajectory of the ore block.

6. The method for identifying lightweight ore blocks on a conveyor belt according to claim 1, characterized in that, In step S6, mapping the size of each ore block in the corrected ore block set to its physical size includes: establishing a mapping relationship from pixel size to physical size based on the known size reference of the conveyor belt and the installation angle and installation height configuration parameters of the image acquisition device, and mapping the size of each ore block to its physical size according to the mapping relationship.

7. The method for lightweight identification of ore blocks in a conveyor belt according to claim 1, characterized in that, The block size distribution obtained in S6 includes: using a preset length of material flow on the conveyor belt as the statistical unit, incrementally updating the block size distribution using a sliding window, wherein the block size distribution includes a cumulative distribution curve and a uniformity index.

8. A lightweight identification system for ore block size on a conveyor belt, characterized in that, include: The motion acquisition module is configured to acquire motion information of the conveyor belt; An image acquisition module, connected to the motion acquisition module, is configured to acquire images of moving ore on the conveyor belt according to the motion information, and obtain an ore image sequence. The deblurring module, connected to the image acquisition module, is configured to determine a motion blur kernel based on the motion information, and to perform non-blind deblurring on the ore image sequence based on the motion blur kernel to obtain a sharpened image sequence. The identification and association module, connected to the deblurring module, is configured to perform ore block identification on the sharpened image sequence to obtain the identification results of each frame, and to perform cross-frame association on the identification results of adjacent frames based on the motion information to obtain the trajectory of the ore block. The deduplication module, connected to the identification association module, is configured to associate and deduplicate the identification results of the same ore block in multiple frames according to the trajectory of the ore block, so that each ore block is counted in the block size statistics only once, and a set of deduplicated ore blocks is obtained. The occlusion correction module, connected to the deduplication module, is configured to correct the block size statistical weight of the occluded ore blocks in the deduplicated ore block set based on the visible outline and convex prior of the occluded ore blocks, and based on the position of the ore blocks at the top or bottom of the material flow on the conveyor belt, to obtain the corrected ore block set. The block size statistics module, connected to the occlusion correction module, is configured to map the size of each ore block in the corrected ore block set to physical size, statistically obtain the block size distribution of the ore on the conveyor belt, and output the block size distribution.

9. An electronic device, characterized in that, It includes a processor and a memory, the memory storing a computer program that, when executed by the processor, implements the conveyor belt ore block size lightweight identification method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the conveyor belt ore block size lightweight identification method according to any one of claims 1 to 7.

Images