An ore grinding screening monitoring method and system based on image particle size analysis

Abstract

The present application belongs to the technical field of image processing, and particularly relates to a kind of ore grinding screening monitoring method and system based on image particle size analysis, and the method comprises the following steps: collecting ore image and generating distance transform graph, and constructing extreme value topological network;Calculate the relative topological significance of adjacent maximum points, and construct peak-valley independence index in combination with spatial distance;Based on peak-valley independence index, the final marking set is obtained by iterative fusion of maximum points;With the final marking set as the seed, the watershed segmentation is carried out, the cumulative distribution of ore particle size is calculated, and the characteristic particle size index is extracted, and the grinding process parameters are adjusted accordingly.The present application can adapt to multi-scale features and effectively eliminate the interference of ore surface texture through the joint constraint of relative topological significance and peak-valley independence index, solves the over-segmentation problem caused by surface roughness in traditional methods, and realizes real-time optimization and energy saving and consumption reduction in the grinding process.

Description

Technical Field

[0001] This invention relates to the field of image processing technology. More specifically, this invention relates to a method and system for monitoring ore grinding and screening based on image particle size analysis. Background Technology

[0002] Ore particle size distribution is a key indicator for evaluating grinding efficiency and guiding mineral processing technology. It is directly related to the recovery rate of subsequent flotation or magnetic separation. In industrial sites, industrial cameras installed above conveyor belts are usually used to collect ore images, and online monitoring is performed using image processing technology to replace manual screening. However, the ground ore is in a high-density random stacking state during the transmission process, and the surface naturally has rough textures, cracks, and water film reflections. In addition, the dust and uneven lighting at the site make the image background complex, which brings great challenges to the accurate segmentation of ore particles.

[0003] Existing technologies mostly use watershed algorithms based on distance transformation to handle adhesion problems. The core of these algorithms relies on local maxima in the distance transformation graph to mark the particle core. In order to suppress oversegmentation caused by noise, H-minima transformation is often introduced. By setting a fixed depth threshold, local extrema are merged, thereby reducing the number of invalid seeds.

[0004] However, this method faces a serious contradiction between scale sensitivity and texture interference. If a large threshold is used to suppress spurious peaks, the maximum points of small-diameter ores are easily submerged and missed, resulting in undersegmentation. Conversely, if a small threshold is used, the texture undulations on the surface of large-diameter ores cannot be effectively suppressed, causing a single piece of ore to be incorrectly segmented into multiple fragments, i.e., oversegmentation. This leads to large fluctuations in particle size statistics, which cannot accurately reflect the quality of the grinding product. Consequently, the control signals fed back to the grinding mill are inaccurate, making it difficult to achieve refined closed-loop control and energy saving in the grinding process. Summary of the Invention

[0005] To address the technical problem that the existing technology cannot simultaneously achieve texture suppression and multi-scale particle recognition, resulting in over-segmentation or under-segmentation of images and difficulty in accurately guiding grinding production, the present invention provides solutions in the following aspects.

[0006] In a first aspect, the present invention provides a method for monitoring ore grinding and screening based on image particle size analysis, comprising:

[0007] Ore images are acquired, and illumination equalization and morphological reconstruction are performed to generate a distance transform map. Maximum points are extracted from the distance transform map, and an extreme value topology network connecting adjacent maximum points is constructed. The relative topological salience between adjacent maximum points is calculated based on the distances between them in the distance transform map and the distances between saddle points on the paths between them. A peak-valley independence index is calculated based on the relative topological salience, the spatial distance between adjacent maximum points, and the distances between them in the distance transform map. The maximum points in the extreme value topology network are iteratively fused based on the peak-valley independence index to obtain the final label set. Using the final label set as seed points, watershed segmentation is performed based on the distance transform map. The cumulative distribution of ore particle size is determined based on the segmentation results. Characteristic particle size indices are extracted from the cumulative distribution of ore particle size, and the grinding process parameters are adjusted based on these characteristic particle size indices.

[0008] This invention effectively suppresses image noise and gradient defects caused by environmental interference by performing illumination equalization and morphological reconstruction on the acquired ore images, constructing a distance transformation map and extreme value topology network that accurately reflects the ore's geometric structure. Utilizing the relative topological saliency between adjacent maxima, it achieves adaptive evaluation of particle independence at different scales, overcoming the scale sensitivity of traditional methods. By combining spatial geometric constraints to construct a peak-valley independence index and performing iterative fusion, and eliminating spurious maxima based on physical laws, it effectively solves the over-segmentation problem caused by ore surface texture. Finally, the selected final label set guides watershed segmentation, extracting ore contours and obtaining characteristic particle size indicators. Based on this, grinding process parameters are adjusted in real time, thereby achieving energy consumption optimization and automatic control of the production process while ensuring the qualified particle size of the grinding product.

[0009] Preferably, the step of performing illumination equalization processing and morphological reconstruction on the ore image to generate a distance transformation map includes: performing contrast-limited adaptive histogram equalization on the ore image to output an illumination equalized image; performing morphological opening reconstruction on the illumination equalized image to obtain a reconstructed image; performing adaptive threshold segmentation on the reconstructed image to obtain a binary mask; and performing Euclidean distance transformation on the binary mask to generate a distance transformation map.

[0010] Preferably, the step of extracting maxima points from the distance transformation map and constructing an extreme value topology network connecting adjacent maxima points includes: extracting all pixels in the distance transformation map whose distance value is greater than the distance value of all pixels in their 8-neighborhood as maxima points, forming a set of maxima points; constructing a Delaunay triangulation network based on the spatial coordinates of the maxima point set, generating an extreme value topology network, and defining two maxima points directly connected by Delaunay triangle edges in the extreme value topology network as adjacent maxima points.

[0011] Preferably, the relative topological saliency satisfies the expression: In the formula, Indicates adjacent maxima and The relative topological significance between them; Indicates the first The distance values ​​of the maxima points in the distance transformation graph; Indicates the first The distance values ​​of the maxima points in the distance transformation graph; This represents the distance value of a saddle point on the path connecting two maxima in the distance transformation graph. The minimum value function is used; the method for obtaining the saddle point is as follows: for each edge connecting two adjacent maxima in the extreme value topology network, search for the pixel with the smallest distance value along the straight path between the two adjacent maxima on the distance transformation graph, and define it as the saddle point on the path.

[0012] This invention establishes an adaptive evaluation standard to measure the independence of raised structures by calculating the normalized difference between adjacent maxima and saddle points on the path, and introducing the scale of the maxima themselves as the denominator for comparison. This eliminates the sensitivity to particle size when judging solely based on absolute depth difference, enabling the algorithm to accurately distinguish between deep texture interference on the surface of large-sized ores and shallow adhesion boundaries between small-sized ores without manual intervention or parameter adjustment. This improves the robustness and accuracy of the algorithm in identifying ores in mixed particle size scenarios.

[0013] Preferably, the peak-valley independence index satisfies the expression: In the formula, Indicates adjacent maxima and The peak-valley independence index between them; Indicates adjacent maxima and The relative topological significance between them; Represents the maximum point and Euclidean distance on the image plane; Indicates the first The distance values ​​of the maxima points in the distance transformation graph; Indicates the first The distance values ​​of the maxima points in the distance transformation graph.

[0014] The peak-valley independence index constructed in this invention deeply couples topological depth features with spatial geometric constraints. It uses the ratio of the Euclidean distance between adjacent maxima on the image plane to the sum of their distances to quantify spatial compactness. This effectively identifies and suppresses pseudo-maxima that, although topologically bimodal, are too close in spatial location. As a result, it accurately distinguishes between the texture undulations of the same ore surface and closely adjacent independent particles, significantly reducing the false detection rate caused by irregular ore shapes or rough surfaces.

[0015] Preferably, the iterative fusion of maxima in the extreme value topology network based on the peak-valley independence index to obtain the final label set includes: storing the peak-valley independence index of all edges in the extreme value topology network and their corresponding connection relationships in a min-heap; popping the edge with the smallest peak-valley independence index from the top of the min-heap, and recording the two maxima connected to it as the first maxima and the second maxima; in response to the smallest peak-valley independence index being less than the fusion threshold, determining that the first maxima and the second maxima belong to the same particle, and performing a fusion operation; in response to the smallest peak-valley independence index being not less than the fusion threshold, stopping the iteration, and outputting the set of remaining valid maxima as the final label set.

[0016] This invention employs a dynamic iteration strategy based on a min-heap, ensuring that the fusion process always proceeds in an orderly manner from the smallest to the largest index, that is, prioritizing the processing of regions that most resemble texture noise. This mechanism, which combines global optimization with local updates, avoids the false deletions or omissions that may be caused by traditional one-time global threshold filtering, and ensures that the final retained label set has a high degree of independence in both statistical and physical senses, thereby achieving high-fidelity extraction of ore particle cores in complex stacking scenarios.

[0017] Preferably, the fusion operation includes: comparing the distance values ​​of the first maximum point and the second maximum point, and retaining the point with the largest distance value as the retained point; if the distance values ​​of the first maximum point and the second maximum point are equal, retaining the point with the smallest index number as the retained point, and marking the other vertex as an invalid point; reconnecting all neighboring nodes originally connected to the invalid point to the retained point; recalculating the relative topological salience and peak-valley independence index between the retained point and the new neighboring nodes, and updating the min-heap.

[0018] This invention preferentially retains points with larger distance values ​​during fusion, ensuring that the merged maximum point is always located at the center of the ore particle with the largest geometric thickness, preventing the marker point position from drifting. At the same time, this invention dynamically updates the connection relationship of neighboring nodes and the corresponding evaluation index, ensuring the real-time consistency of the extreme value topology network after each fusion, so that subsequent fusion judgments are always based on the latest network structure, thereby restoring the true physical form of the ore and eliminating the fragmentation segmentation problem caused by local texture.

[0019] Preferably, the step of using the final label set as seed points, performing watershed segmentation based on the distance transform map, and determining the cumulative distribution of ore particle size according to the segmentation result includes: using the final label set as forced seed points, performing a watershed transform algorithm on the inverted image of the distance transform map to generate a segmentation label map containing multiple independent connected components; traversing each connected component in the segmentation label map, calculating its pixel area, and converting it into the equivalent circle diameter. In the formula, Indicates the first The equivalent circle diameter of the ore particles corresponding to each connected domain; Indicates the first The number of pixels contained in each connected component; Pi; The calibration coefficients for pixels and physical dimensions are used; the equivalent circle diameters of all particles in the current frame of the ore image are calculated, and the cumulative distribution curve of ore particle size is constructed with the equivalent circle diameter as the horizontal axis and the cumulative distribution percentage as the vertical axis.

[0020] This invention utilizes a final set of selected markers as forced seed points to guide watershed transformation, converting traditional unsupervised segmentation into controlled segmentation. This solves the oversegmentation problem caused by the sensitivity of traditional watershed algorithms to noise, enabling precise segmentation of adherent ores along real physical boundaries. By converting the physical dimensions from pixel area to equivalent circle diameter and constructing cumulative distribution curves, this invention transforms image segmentation results into statistical data that conforms to industrial standards. This data can intuitively reflect the particle size distribution characteristics of the current ore population, providing reliable data support for accurate quantitative evaluation and closed-loop control of grinding processes.

[0021] Preferably, the step of extracting characteristic particle size index from the cumulative distribution of ore particle size and adjusting grinding process parameters according to the characteristic particle size index includes: extracting the equivalent circle diameter corresponding to the cumulative distribution percentage reaching 80% as the characteristic particle size index; responding to the characteristic particle size index being greater than the preset target upper limit, outputting a control signal to increase the grinding mill power or decrease the feed rate; responding to the characteristic particle size index being less than the preset target lower limit, outputting a control signal to decrease the grinding mill power or increase the feed rate; and responding to the characteristic particle size index being between the preset target upper limit and target lower limit, keeping the current grinding mill power and feed rate unchanged.

[0022] Secondly, the present invention provides an ore grinding and screening monitoring system based on image particle size analysis, including a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned ore grinding and screening monitoring method based on image particle size analysis is implemented.

[0023] By adopting the above technical solution, a computer program for monitoring ore grinding and screening based on image particle size analysis is generated and stored in a memory so that it can be loaded and executed by a processor. A terminal device can then be made based on the memory and processor for convenient use.

[0024] The beneficial effects of this invention are as follows: This invention acquires ore images under controlled light fields, generates distance transformation maps and constructs extreme value topology networks after illumination equalization and morphological reconstruction, effectively suppressing image noise and gradient defects caused by environmental interference. It utilizes the relative topological saliency between adjacent maxima points for normalized difference calculation, achieving adaptive evaluation of particle independence at different scales and overcoming the scale sensitivity of traditional methods. Furthermore, it combines spatial geometric constraints to construct a peak-valley independence index and dynamically eliminates pseudo-maxima points through a minimum heap-driven iterative fusion strategy. Based on physical laws, it effectively identifies and fuses texture pseudo-peaks on the ore surface, solving the over-segmentation problem caused by surface roughness. Finally, it uses the selected final label set to guide controlled watershed segmentation, extracts the ore physical contour and obtains characteristic particle size indicators, and converts image data into control signals to provide real-time feedback adjustment of grinding process parameters. This achieves energy consumption optimization and automatic control of the production process while ensuring the qualified particle size of the grinding product. Attached Figure Description

[0025] Figure 1 This is a flowchart illustrating an image particle size analysis-based method for monitoring ore grinding and screening according to the present invention.

[0026] Figure 2 This is a schematic diagram of an ore image in an embodiment of the present invention;

[0027] Figure 3 This is a distance transformation diagram in an embodiment of the present invention;

[0028] Figure 4 This is a schematic diagram showing the comparison between the initial maximum points and the final label set after iterative fusion in an embodiment of the present invention;

[0029] Figure 5 The segmentation result is obtained using the traditional watershed algorithm based on existing technology;

[0030] Figure 6 This is the result of watershed segmentation based on the final tag set in this invention;

[0031] Figure 7 This is a schematic diagram of the cumulative particle size distribution curve of ore in an embodiment of the present invention. Detailed Implementation

[0032] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0033] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0034] This invention discloses a method for monitoring ore grinding and screening based on image particle size analysis, referring to... Figure 1 This includes steps S1-S4:

[0035] S1. Acquire ore images, perform illumination equalization and morphological reconstruction on the ore images, generate a distance transformation map, extract the maxima points in the distance transformation map, and construct an extremum topology network connecting adjacent maxima points.

[0036] It should be noted that, due to the high concentration of dust and water mist in the ore grinding site, and the fact that lighting conditions are easily affected by equipment obstruction, the directly acquired images contain uneven illumination components and high-gloss reflections from the ore surface. If distance transformation is performed directly, a large number of false holes and breaks will appear in the distance transformation image, leading to distortion in the extraction of maximum points. Therefore, this invention enhances image details through illumination equalization and repairs gradient defects using morphological reconstruction, thereby constructing a distance topography and topology network that can truly reflect the geometric uplift structure of the ore.

[0037] Specifically, this invention utilizes an industrial camera equipped with a diffuse dome light source to acquire real-time grayscale images of the ore conveyor belt, which are denoted as ore images. For example, Figure 2 This is a schematic diagram of an ore image in an embodiment of the present invention.

[0038] Contrast-limited adaptive histogram equalization is performed on the ore image to output an illumination-equalized image, suppressing local highlights and enhancing shadow details. To suppress texture noise within the ore while preserving edge gradients, morphological opening reconstruction is performed on the illumination-equalized image to obtain a reconstructed image. Adaptive threshold segmentation is performed on the reconstructed image to obtain a binary mask that can distinguish between foreground ore and background seams. Euclidean distance transformation is performed on the binary mask to generate a distance transformation map, where the pixel value of each foreground pixel represents its Euclidean distance to the nearest background pixel, and the pixel value of the background pixel is 0. For example, Figure 3 This is a distance transformation diagram in an embodiment of the present invention.

[0039] Scan the distance transformation map and extract all pixels whose distance value is greater than the distance value of all pixels in their 8-neighborhood. These pixels are designated as maxima, forming a set of maxima. Based on the spatial coordinates of the maxima set, a Delaunay triangulation network is constructed to generate an extremum topology network. The maxima set is the node set of the extremum topology network, and two maxima directly connected by Delaunay triangle edges are defined as adjacent maxima. The connections between adjacent maxima constitute the edge set of the extremum topology network. For each edge connecting two maxima in the extremum topology network, search for the pixel with the smallest distance value along the straight path between the two maxima on the distance transformation map, and define this pixel as a saddle point on that path.

[0040] Thus, a complete data structure containing node and edge information has been obtained.

[0041] S2. Calculate the relative topological salience between adjacent maxima points based on the distance values ​​of adjacent maxima points in the distance transformation graph and the distance values ​​of saddle points on the path between adjacent maxima points in the distance transformation graph.

[0042] It should be noted that, due to the extremely wide range of particle sizes in the ground ore, including both coarse particles with large radii and fine powder with small radii, and the fact that the depth of the texture depressions on the surface of large particles is often greater than the radius of small particles, it is impossible to distinguish the independent boundaries of the texture on the surface of large particles from those of small particles simply by relying on the absolute depth difference. Therefore, this invention introduces relative topological saliency, which uses the particle's own scale to normalize the depth difference, thereby accurately evaluating the independence of peaks at different scales under a unified standard.

[0043] Specifically, for each edge in the extreme value topology network, the distance values ​​corresponding to the maxima of the two endpoints of the edge and the distance values ​​of the saddle points corresponding to the edge are obtained from the distance transformation graph, and the relative topological salience of the edge is calculated:

[0044]

[0045] In the formula, Indicates adjacent maxima and The relative topological significance between them; Indicates the first The distance value of a maximum point in the distance transformation map represents the approximate radius of the potential particle corresponding to that maximum point; Indicates the first The distance values ​​of the maxima points in the distance transformation graph; This represents the distance value of a saddle point on the path connecting two maxima in the distance transform graph. Since a pixel value in the distance transform graph is geometrically equivalent to the radius of the largest inscribed circle centered at that pixel, therefore... It intuitively represents the narrowest geometric width of the channel at the adhesion point of two potential particles, or the maximum inscribed circle radius of the texture depression inside the same particle; This is a minimum value function. The distance value from the saddle point... The smaller the molecular size, the deeper the adhesion point, the more molecules... The larger the value, the deeper the absolute depth, making... A larger value, and one that approaches 1, indicates a significant physical separation between the two maxima; when the distance value of the saddle point... The larger the size, that is, the shallower the indentation, the more molecules... The smaller the value, the better. The smaller the value and the closer it is to 0, the more likely the two maxima belong to the texture fluctuations of the same particle surface. This invention utilizes... Normalization is applied to the denominator, so that for deep textures on large-particle surfaces, although the absolute depth is large, the calculated value is lower due to the large denominator. The value is still relatively small; however, for shallow adhesion of small particles, although the absolute depth is small, the calculated value is relatively small because the denominator is small. The value is still relatively large, thus enabling adaptive evaluation of particles of different scales.

[0046] S3. Calculate the peak-valley independence index based on the relative topological salience, the spatial distance between adjacent maxima, and the distance between adjacent maxima in the distance transformation graph. Iteratively fuse the maxima in the extreme value topology network based on the peak-valley independence index to obtain the final label set.

[0047] It should be noted that, since the geometric center of some elongated or plate-shaped ores may appear as two distant peaks with a shallow indentation in the distance transformation diagram, it is easy to misjudge them as two independent particles based solely on depth information. However, the actual independent particles should meet specific spatial distance constraints. Therefore, this invention constructs a peak-valley independence index, which combines topological depth features with spatial geometric features, and dynamically merges pseudo-maximum points belonging to the same particle through an iterative fusion strategy, ensuring that the seed points retained in the end correspond one-to-one with the actual ore particles.

[0048] Specifically, for topology networks For each edge in the array, combined with its endpoints and Calculate the peak-valley independence index based on the pixel coordinates:

[0049]

[0050] In the formula, Indicates adjacent maxima and The peak-valley independence index is used to comprehensively determine whether two maximum points belong to different independent particles; Indicates adjacent maxima and The relative topological significance between them; Represents the maximum point and Euclidean distance on the image plane; Indicates the first The distance values ​​of the maxima points in the distance transformation graph; Indicates the first The distance values ​​of two maxima in the distance transformation map. When two maxima belong to the same grain of texture, although their distance values... and It may be large, but its spatial distance Typically very small, resulting in a ratio The peak-valley independence index approaches 0, thus making the peak-valley independence index... Minimal; when the two maxima represent two tangent independent particles, their spatial distance is... Approximately equal to the sum of their radii, the ratio approaches 1, combined with a relatively high topological saliency. , making The value is relatively large.

[0051] Furthermore, a min-heap data structure is used to manage the fusion process:

[0052] Store the peak-valley independence indices of all edges in the extreme value topology network and their corresponding connectivity relationships in a min-heap, and set a dimensionless fusion threshold. Pop the edge with the smallest peak-valley independence index from the top of the min-heap, and denote the two maxima connected by it as the first maxima and the second maxima.

[0053] The response is that the minimum peak-valley independence index is less than the fusion threshold. If the first and second maxima points belong to the same particle, perform a fusion operation: compare the distance values ​​of the first and second maxima points, retain the point with the larger distance value as the retained point, if the two distance values ​​are equal, retain the point with the smaller index number as the retained point, mark the other vertex as an invalid point, reconnect all neighbor nodes that were originally connected to the invalid point to the retained point, recalculate the relative topological salience and peak-valley independence index between the retained point and the new neighbor nodes, and update the min-heap;

[0054] The minimum peak-valley independence index is not less than the fusion threshold. This indicates that the two most closely connected points in the current extreme value topology network are also independent. Stop the iteration and output the set of remaining valid maximum points as the final label set.

[0055] For example, Figure 4 This is a schematic diagram comparing the initial maximum points and the final label set after iterative fusion in an embodiment of the present invention. The cross points represent the final label set, which successfully eliminated and fused multiple pseudo-maximum points within the same particle.

[0056] It should be noted that the fusion threshold The setup method is as follows: A set of sample images containing only individual ore particles are pre-acquired. The K-values ​​between each pair of pseudo-maximum points generated by texture within each individual ore particle are calculated. The distribution of these K-values ​​is statistically analyzed. T is set to be slightly larger than the statistical upper bound of the K-values ​​of pseudo-maximum points in the sample, such as the upper limit of the 95% confidence interval, thereby ensuring that most texture pseudo-peaks can be effectively fused in actual detection. In this embodiment, the fusion threshold is... The empirical value is 0.25. In other embodiments, implementers may set it according to the actual implementation situation.

[0057] S4. Using the final set of markers as seed points, perform watershed segmentation based on the distance transformation map, determine the cumulative distribution of ore particle size based on the segmentation results, extract characteristic particle size indicators from the cumulative distribution of ore particle size, and adjust the grinding process parameters based on the characteristic particle size indicators.

[0058] It should be noted that since the final label set has eliminated texture interference and retained the core of multi-scale particles, the segmentation based on this can accurately extract the physical contour of the ore. However, the simple image segmentation results cannot directly guide production and need to be converted into statistical indicators to form closed-loop control. Therefore, this invention calculates the cumulative distribution of particle size and extracts key feature parameters to convert image data into control signals for the grinding machine, thereby realizing the automated adjustment of the production process.

[0059] Specifically, using the final label set as the forced seed points, a watershed transform algorithm is performed on the inverted image of the distance transform map to generate a segmentation label map containing multiple independent connected components. For example, Figure 5 The results of the traditional watershed algorithm using existing technology show that it incorrectly divides a single piece of ore into multiple fragments due to interference from the surface texture of the ore, and cannot truly reflect the particle size of the ore. Figure 6 The result of watershed segmentation based on the final tag set in this invention shows that it accurately reconstructs the physical contour of the ore.

[0060] Furthermore, traverse each connected component in the segmentation label graph, calculate its pixel area, and convert it into the equivalent circle diameter:

[0061]

[0062] In the formula, Indicates the first The equivalent circle diameter of the ore particles corresponding to each connected domain; Indicates the first The number of pixels contained in each connected component; Pi; The calibration coefficient C is a calibration factor for pixels and physical dimensions. The calibration coefficient C is obtained by using a known physical side length... A standard grid calibration plate is placed on a stationary conveyor belt. An industrial camera is used to capture images of the calibration plate, and the pixel side length of the grid in the image is measured. Calculations yielded .

[0063] The equivalent circle diameter of all particles in the current frame of the ore image is statistically analyzed. A cumulative distribution curve of ore particle size is constructed with the equivalent circle diameter as the horizontal axis and the cumulative distribution percentage as the vertical axis. Feature particle size indices are extracted from this curve, specifically the equivalent circle diameter corresponding to a cumulative distribution percentage of 80%, which serves as the core evaluation indicator for the current grinding effect. For example, Figure 7 This is a schematic diagram of the cumulative particle size distribution curve of ore in an embodiment of the present invention.

[0064] The real-time calculated characteristic particle size index is input into the grinder's controller. If the characteristic particle size index exceeds the preset upper target limit, it indicates insufficient grinding and excessively coarse particles; the controller then outputs a control signal to increase the grinder's power or decrease the feed rate. If the characteristic particle size index is below the preset lower target limit, it indicates over-grinding and excessively fine particles; the controller then outputs a control signal to decrease the grinder's power or increase the feed rate to optimize energy consumption. If the characteristic particle size index is between the preset upper and lower target limits, it indicates that the current grinding effect meets the process requirements; the current grinder power and feed rate are maintained unchanged. The upper and lower target limits are set by the implementers based on the required grinding accuracy.

[0065] This invention also discloses an ore grinding and screening monitoring system based on image particle size analysis, including a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement an ore grinding and screening monitoring method based on image particle size analysis according to the present invention.

[0066] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

Claims

1. A method for monitoring ore grinding and screening based on image particle size analysis, characterized in that, include: Acquire ore images, perform illumination equalization and morphological reconstruction on the ore images, generate distance transformation maps, extract the maxima points in the distance transformation maps, and construct an extremum topology network connecting adjacent maxima points. Calculate the relative topological salience between adjacent maxima points based on the distance values ​​of adjacent maxima points in the distance transformation graph and the distance values ​​of saddle points on the path between adjacent maxima points in the distance transformation graph. The peak-valley independence index is calculated based on the relative topological saliency, the spatial distance between adjacent maxima, and the distance between adjacent maxima in the distance transformation graph. The maxima in the extreme value topology network are iteratively fused based on the peak-valley independence index to obtain the final label set. This includes: storing the peak-valley independence index of all edges in the extreme value topology network and their corresponding connectivity in a min-heap; popping the edge with the smallest peak-valley independence index from the top of the min-heap, and labeling the two maxima connected to it as the first maxima and the second maxima; determining that the first maxima and the second maxima belong to the same particle if the smallest peak-valley independence index is less than the fusion threshold, and performing a fusion operation; stopping the iteration if the smallest peak-valley independence index is not less than the fusion threshold, and outputting the set of remaining valid maxima as the final label set. Using the final set of markers as seed points, watershed segmentation is performed based on the distance transformation map. The cumulative distribution of ore particle size is determined based on the segmentation results. Characteristic particle size indexes are extracted from the cumulative distribution of ore particle size, and the grinding process parameters are adjusted based on the characteristic particle size indexes.

2. The ore grinding and screening monitoring method based on image particle size analysis according to claim 1, characterized in that, The process of performing illumination equalization and morphological reconstruction on the ore image to generate a distance transformation map includes: Perform contrast-limited adaptive histogram equalization on the ore image to output an illumination-equalized image; perform morphological opening reconstruction on the illumination-equalized image to obtain a reconstructed image; perform adaptive threshold segmentation on the reconstructed image to obtain a binary mask; perform Euclidean distance transformation on the binary mask to generate a distance transformation map.

3. The ore grinding and screening monitoring method based on image particle size analysis according to claim 1, characterized in that, The step of extracting maxima points from the distance transformation graph and constructing an extremum topology network connecting adjacent maxima points includes: Pixels whose distance value is greater than the distance value of all pixels in their 8-neighborhood in the distance transformation map are extracted as maxima points, forming a set of maxima points; a Delaunay triangulation network is constructed based on the spatial coordinates of the maxima point set to generate an extremum topology network, and two maxima points directly connected by the sides of the Delaunay triangle in the extremum topology network are defined as adjacent maxima points.

4. The ore grinding and screening monitoring method based on image particle size analysis according to claim 1, characterized in that, The relative topological significance satisfies the expression: ； In the formula, Indicates adjacent maxima and The relative topological significance between them; Indicates the first The distance values ​​of the maxima points in the distance transformation graph; Indicates the first The distance values ​​of the maxima points in the distance transformation graph; This represents the distance value of a saddle point on the path connecting two maxima in the distance transformation graph. It is a minimum value function; The method for obtaining the saddle point is as follows: for each edge connecting two adjacent maxima in the extreme value topology network, search for the pixel with the smallest distance value along the straight path between the two adjacent maxima on the distance transformation graph, and define it as the saddle point on the path.

5. The ore grinding and screening monitoring method based on image particle size analysis according to claim 1, characterized in that, The peak-valley independence index satisfies the expression: ； In the formula, Indicates adjacent maxima and The peak-valley independence index between them; Indicates adjacent maxima and The relative topological significance between them; Represents the maximum point and Euclidean distance on the image plane; Indicates the first The distance values ​​of the maxima points in the distance transformation graph; Indicates the first The distance values ​​of the maxima points in the distance transformation graph.

6. The ore grinding and screening monitoring method based on image particle size analysis according to claim 1, characterized in that, The fusion operation includes: Compare the distances between the first and second maxima, and retain the point with the largest distance as the retained point; if the distances between the first and second maxima are equal, retain the point with the smallest index number as the retained point, and mark the other vertex as an invalid point; reconnect all neighboring nodes that were originally connected to the invalid point to the retained point; recalculate the relative topological salience and peak-valley independence index between the retained point and the new neighboring nodes, and update the min-heap.

7. The ore grinding and screening monitoring method based on image particle size analysis according to claim 1, characterized in that, The step of using the final marked set as seed points, performing watershed segmentation based on the distance transformation map, and determining the cumulative distribution of ore particle size based on the segmentation results includes: Using the final label set as the forced seed points, the watershed transform algorithm is applied to the inverted image of the distance transform map to generate a segmentation label map containing multiple independent connected components. Each connected component in the segmentation label map is traversed, its pixel area is calculated, and converted to the equivalent circle diameter. In the formula, Indicates the first The equivalent circle diameter of the ore particles corresponding to each connected domain; Indicates the first The number of pixels contained in each connected component; Pi; The calibration coefficients for pixels and physical dimensions are used; the equivalent circle diameters of all particles in the current frame of the ore image are calculated, and the cumulative distribution curve of ore particle size is constructed with the equivalent circle diameter as the horizontal axis and the cumulative distribution percentage as the vertical axis.

8. The ore grinding and screening monitoring method based on image particle size analysis according to claim 1, characterized in that, The step of extracting characteristic particle size indices from the cumulative distribution of ore particle size and adjusting grinding process parameters based on these indices includes: The equivalent circle diameter corresponding to the cumulative distribution percentage reaching 80% is extracted from the cumulative distribution of ore particle size as the characteristic particle size index; in response to the characteristic particle size index being greater than the preset target upper limit, a control signal is output to increase the grinding mill power or decrease the feed rate; in response to the characteristic particle size index being less than the preset target lower limit, a control signal is output to decrease the grinding mill power or increase the feed rate; in response to the characteristic particle size index being between the preset target upper limit and target lower limit, the current grinding mill power and feed rate remain unchanged.

9. A monitoring system for ore grinding and screening based on image particle size analysis, characterized in that, include: A processor and a memory, the memory storing computer program instructions that, when executed by the processor, implement the ore grinding and screening monitoring method based on image particle size analysis according to any one of claims 1-8.

Images