A crusher classified material monitoring method and system based on image recognition

Abstract

The application relates to the technical field of image recognition, in particular to a crusher graded material monitoring method and system based on image recognition. The method comprises the following steps: carrying out threshold segmentation on a pretreated gray image to distinguish a gap and a potential ore area. Pixel points are divided into a highlight area and a texture area through a clustering algorithm and density analysis. After the highlight area is subjected to smoothing treatment, superpixel segmentation is carried out, and whether surrounding superpixel blocks are merged is judged by using the gray distribution and gradient features of the texture area, so that an accurate ore particle area is obtained. The application has the effect of improving the calculation accuracy of granularity.

Description

Technical Field

[0001] This application relates to the field of image recognition technology, and in particular to a method and system for monitoring graded materials in a crusher based on image recognition. Background Technology

[0002] In the mining and chemical industries, real-time monitoring of ore particle size after crushing and conveyor belt transport is typically required to optimize crushing processes and improve production efficiency. Common methods for particle size detection after crushing include manual sampling, sieving, or centrifugation. While sieving offers basic particle size distribution acquisition, its process usually involves manual sampling, drying, sieving, and weighing, with detection times generally ranging from several minutes to tens of minutes, limiting it to low-frequency offline monitoring.

[0003] To improve monitoring efficiency, some image processing algorithms are applied to particle size detection. For example, superpixel segmentation algorithms are used to process acquired ore images, aiming to divide the image into several regions with similar features to identify particles.

[0004] However, due to the highly irregular shape and complex surface texture of broken ores, the surface often exhibits reflective edges or near-white dotted or linear textures formed by the material itself. Superpixel segmentation algorithms, which primarily rely on pixel color and spatial distance for clustering, often fail to effectively distinguish between the internal texture of the ore surface and its edges. This makes it easy for the algorithm to misclassify surface texture regions belonging to the same ore particle as different objects, thus incorrectly segmenting a complete particle into multiple superpixel blocks. This segmentation error leads to subsequently calculated particle areas being too small, resulting in insufficient overall particle size detection accuracy. Summary of the Invention

[0005] To improve the accuracy of particle size detection, this application provides a method and system for monitoring crushed material classification based on image recognition.

[0006] Firstly, this application provides a method for monitoring graded materials in a crusher based on image recognition, employing the following technical solution:

[0007] A crusher-graded material monitoring method based on image recognition includes: segmenting a pre-processed grayscale image of the crushed material to obtain multiple particle regions, calculating the area of ​​each particle region to obtain the particle size of the crushed material; and adjusting the grinding intensity of the crusher based on the particle size.

[0008] The step of segmenting the grayscale image of the pre-processed crushed material to obtain multiple particle regions includes: taking pixels with grayscale values ​​less than or equal to a grayscale threshold as first pixels and pixels with grayscale values ​​greater than a grayscale threshold as second pixels.

[0009] Multiple clustering regions are obtained by using a clustering algorithm on each second pixel. For any clustering region, the density of each clustering region is calculated based on the pixel distribution. Clustering regions with a density lower than the first density threshold are recorded as first-class regions, clustering regions with a density greater than the second density threshold are recorded as second-class regions, and the remaining clustering regions are recorded as third-class regions.

[0010] After smoothing the gray values ​​of pixels in the first type of region, superpixel segmentation is performed on the pixels in the first and third types of regions; it is determined whether the second type of region is merged with the surrounding superpixel blocks based on the gray value distribution characteristics; based on the merged superpixel blocks, the region of each mineral particle in the image is obtained.

[0011] For ore particles, the most noticeable feature is the black shadow formed by the gaps between particles. Therefore, the grayscale image is initially segmented based on grayscale values: pixels with grayscale values ​​less than a threshold are designated as first pixels, and those with grayscale values ​​greater than a second threshold are designated as second pixels. Within the second pixels, one type is sparsely distributed, typically representing the highlights at particle edges, while another type is linearly densely distributed, typically representing particle edges. Based on density, clustering regions are divided into first, second, and third types. Pixels in the first type are smoothed to reduce highlights and prevent isolation during subsequent superpixel segmentation. Then, for the edges of the second type, the grayscale changes along the corresponding traversal path are used to determine whether they can be merged with adjacent superpixel blocks. Superpixel blocks are merged based on the region probability of each second type region to obtain precise particle regions. The particle size of the current image is determined based on the area of ​​the particle region, thus enabling the monitoring of the particle size of the crusher grinding process.

[0012] Optionally, the step of calculating the density of each cluster region based on the pixel distribution includes: performing linear fitting based on the coordinates of the pixels in the cluster region to obtain the corresponding fitted line; obtaining the density based on the distance from the pixel to the fitted line and the distance from the pixel to the cluster center of the cluster region, wherein the density is negatively correlated with the average Euclidean distance from all pixels in the cluster region to the cluster center and the average distance from all pixels in the cluster region to the corresponding fitted line.

[0013] Reflective noise is spatially discrete, with a large distance from the center and no obvious linear trend, resulting in low density. In contrast, the surface texture of particles is striped, with a small average distance from the fitted line, indicating higher density. This method models both types of features simultaneously, which can correctly identify textured and bright areas in subsequent superpixel processing, reducing missegmentation and ensuring the integrity of particle region contours, thereby improving the accuracy of final particle size calculation.

[0014] Optionally, the step of smoothing the gray values ​​of pixels in the first type of region includes: for any pixel in the first type of region, obtaining the gray values ​​of all pixels in the neighborhood of the pixel and obtaining the neighborhood gray-level histogram, and obtaining the peak gray value of the neighborhood gray-level histogram in the gray-level histogram.

[0015] The local grayscale difference is obtained by averaging the absolute values ​​of the differences between the grayscale value of a pixel and the grayscale values ​​of all pixels in its neighborhood.

[0016] The smoothed grayscale value is negatively correlated with the local grayscale difference, the difference between the grayscale value of a pixel and the peak grayscale value.

[0017] By combining the peak value of the neighborhood gray-level histogram and the local gray-level difference to smooth the first type of region, this adaptive smoothing method takes into account the local texture characteristics. It can effectively suppress specular noise and stray pixel interference while preserving important structural information in the image, providing a cleaner data foundation for subsequent superpixel segmentation and reducing erroneous segmentation boundaries caused by noise.

[0018] Optionally, the step of determining whether the second type of region is merged with the surrounding superpixel blocks based on the grayscale distribution characteristics of the second type of region includes: taking the edge pixels of the region formed by the first pixel as the first edge point, and taking the edge pixels of the second type of region as the second edge point; and dividing the second edge point into fuzzy boundary points and defined boundary points based on the distance between the second edge pixel and the first edge point.

[0019] For a given boundary point, the traversal path of the boundary point is obtained based on the gradient direction of the boundary point; the difference in grayscale value between each two adjacent pixels on the traversal path is obtained, and the average value of all differences is used to determine whether the second type of region is merged with the surrounding superpixel block.

[0020] The ability to merge texture boundaries with superpixel blocks on both sides is determined by the edge grayscale change rate, enabling the system to distinguish between texture boundaries and real grain boundaries. The second type of region mainly contains linear textures, where the grayscale difference between the pixels adjacent to the grains is usually small and continuous, while the grayscale abrupt change at the true outer edge is large and irregular. Therefore, by traversing the path and calculating the average grayscale difference between adjacent pixels, it is possible to accurately determine whether the texture can merge across both sides.

[0021] Optionally, the steps for obtaining the traversal path of the second edge point based on the gradient direction of the second edge point include:

[0022] Based on the distance between the second edge pixel and the first edge pixel, the second edge pixel is classified into fuzzy boundary points and definite boundary points;

[0023] For a given boundary point, the first edge point closest to the given boundary point is taken as the nearest boundary point, and the line segment extending along the gradient direction of the second edge point and intersecting the line perpendicular to the gradient direction of the first edge point is taken as the traversal path.

[0024] For a fuzzy boundary point, find the nearest known boundary point to the fuzzy boundary point as a replacement boundary point; obtain the nearest boundary point of the replacement boundary point as the nearest replacement point; find the intersection of a straight line perpendicular to the gradient direction of the nearest replacement point along the gradient direction of the fuzzy boundary point, and use the path between the fuzzy boundary point and the intersection point as the traversal path.

[0025] By distinguishing between blurred boundary points and definitive boundary points, and finding the optimal traversal path perpendicular to the gradient direction, the edge feature extraction process is optimized. This method automatically avoids areas with unclear edges, finds the most representative path for grayscale difference comparison, and ensures the accuracy of the merging judgment criteria.

[0026] Optionally, the step of obtaining the region probability of merging the second type region with the superpixel blocks on both sides based on the average of all differences includes: for any determined boundary point, calculating the average of the absolute values ​​of the differences between adjacent gray values ​​on the traversal path to obtain the gray change rate; the gray change rate is negatively correlated with the merging probability of the determined boundary point;

[0027] Based on the merging probability of all edge pixels in the second type region, the region probability of merging the second type region with the superpixel blocks on both sides is calculated; the region probability of merging the second type region with the superpixel blocks on both sides is positively correlated with the mean of the merging probability of all edge pixels and negatively correlated with the variance of the merging probability of all edge pixels.

[0028] In response to the probability that the second type region and the superpixel blocks on both sides are merged is greater than a set threshold, the superpixel blocks on both sides of the second type region are merged.

[0029] The calculation of the overall merging probability statistics of the region, rather than relying solely on single-point judgments, avoids the impact of local outliers on the overall segmentation results.

[0030] Optionally, the preprocessing of the grayscale image of the crushed material includes: processing the original image using adaptive filtering and enhancing the edges of the original image using edge sharpening techniques.

[0031] Adaptive filtering and edge sharpening preprocessing are introduced. Adaptive filtering effectively removes salt-and-pepper noise and Gaussian noise introduced by the industrial environment, while edge sharpening enhances the true physical contour gradient of the ore. This combination of preprocessing improves the signal-to-noise ratio of the original image.

[0032] Optionally, the average area of ​​each particle region can be used as the particle size.

[0033] Optionally, the product of particle size and a preset weighting coefficient can be used as the grinding intensity.

[0034] Secondly, this application provides a crusher grading material monitoring system based on image recognition, which adopts the following technical solution:

[0035] A crusher grading material monitoring system based on image recognition includes a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement the crusher grading material monitoring method based on image recognition described above.

[0036] The above-mentioned image recognition-based crusher grading material monitoring method is generated into a computer program and stored in a memory for loading and execution by a processor. Thus, a system is created based on the memory and processor for convenient use.

[0037] This application has the following technical effects:

[0038] This method distinguishes between reflective noise and linear textures on the ore surface using a clustering algorithm, and determines the connectivity of adjacent superpixel blocks by utilizing the grayscale distribution and gradient features of the textured regions. This effectively differentiates the internal texture of the ore from its physical edges, accurately reconstructing the ore outline while eliminating oversegmentation errors, thus significantly improving particle size detection accuracy and enabling precise automated adjustment of the crusher's grinding intensity. Attached Figure Description

[0039] Figure 1 This is a flowchart of a crusher grading material monitoring method based on image recognition, as described in this application.

[0040] Figure 2 This is a flowchart of step S1 in the image recognition-based crusher grading material monitoring method of this application.

[0041] Figure 3 This is a preliminary coarse segmentation result of the grayscale image of the crushed material in the crusher grading material monitoring method based on image recognition proposed in this application.

[0042] Figure 4 This is an image showing the effect of superpixel segmentation after smoothing the gray values ​​of pixels in the first type of region in the image recognition-based crusher grading material monitoring method of this application.

[0043] Figure 5 This is a flowchart of step S14 in the image recognition-based crusher grading material monitoring method of this application. Detailed Implementation

[0044] This application discloses a crusher grading material monitoring method based on image recognition. The application scenario of this embodiment is set as follows: real-time monitoring of ore that has undergone preliminary crushing on a belt conveyor. During the operation of the conveyor, images of crushed particles are collected and superpixel segmentation is performed. Then, the grayscale of the segmented images is analyzed, and superpixel blocks that can be merged are merged, thereby reducing the situation where the same particle is segmented into two superpixel blocks, resulting in low particle size calculation accuracy.

[0045] Reference Figure 1 This includes steps S1-S2.

[0046] S1: Segment the grayscale image of the pre-processed crushed material to obtain multiple particle regions.

[0047] Reference Figure 2 Step S1 includes steps S11-S14.

[0048] S11: Pixels with gray values ​​less than or equal to the gray threshold in the grayscale image are designated as first pixels, and pixels with gray values ​​greater than the gray threshold are designated as second pixels.

[0049] First, construct the image acquisition environment. Install a high-definition industrial camera directly above the conveyor belt. Preferably, select the middle area of ​​the conveyor belt (e.g., 0.5 meters long and 0.3 meters wide) as the monitoring range. The camera is preferably installed at a height of 1 meter above the conveyor belt surface to ensure that the field of view covers the entire monitoring area and minimizes image distortion. Simultaneously, configure a uniform lighting source to ensure consistent illumination on the surface of the crushed ore, avoiding interference from high reflectivity or shadows not caused by the ore itself due to the angle of the light source.

[0050] The camera continuously captures surface images of the crushed ore over extended periods. To eliminate noise interference from the industrial environment and improve image quality, the following preprocessing steps are performed on the acquired raw images:

[0051] Denoising and sharpening: Adaptive filters are used to remove salt-and-pepper noise and Gaussian noise from the image; then edge sharpening algorithms (such as Laplacian sharpening) are used to enhance the gradient features of the ore edges, making the ore outline clearer.

[0052] Contrast Enhancement: The image is processed using the Limiting Contrast Adaptive Histogram Equalization (CLAHE) algorithm. This algorithm effectively reduces the brightness of highlight areas and enhances the local contrast of shadow areas, allowing the texture details of the mineral surface to be significantly presented under different lighting conditions.

[0053] Grayscale conversion: Converting a color image to a grayscale image reduces the amount of data required for subsequent calculations and improves system processing speed.

[0054] Because the stacking of ore creates natural physical gaps, these gaps appear as deep black shadows in the image. This physical property is used to perform preliminary coarse segmentation of the image.

[0055] Specifically, in this embodiment, the grayscale value of each pixel in the preprocessed grayscale image is traversed. A grayscale threshold is set; in this embodiment, the grayscale threshold is set to 25. This grayscale threshold is determined by those skilled in the art based on the grayscale of the gaps between particles in the grayscale image and their own experience, and is mainly used to filter out the black shadows between particles. Combined with... Figure 3 In the image, the left side is the original image of the ore particles, and the right side is the image after thresholding. Pixels with a gray value less than or equal to 25 are marked as the first pixel, corresponding to the black area in the right image; pixels with a gray value greater than 25 are marked as the second pixel, corresponding to the gray area in the image.

[0056] S12: Use a clustering algorithm to obtain multiple cluster regions for each second pixel; for any cluster region, calculate the density of each cluster region based on the pixel distribution; the cluster regions with a density lower than the first density threshold are recorded as the first type of region, the cluster regions with a density greater than the second density threshold are recorded as the second type of region, and the remaining cluster regions are recorded as the third type of region.

[0057] The second pixel has a higher grayscale value and is not a shadow between particles, but rather a surface area of ​​hardened particles. This type of area requires refined analysis. The features of a mineral surface typically fall into two categories: one is the distribution of bright, reflective pixels at edges in a patchy pattern, and the other is the surface texture, which is distributed in a linear pattern. To avoid the superpixel segmentation algorithm misclassifying textures on the same mineral surface as boundaries, these white or bright pixels need to be classified.

[0058] In this step, a clustering algorithm is first used to cluster the second pixels in the grayscale image, obtaining multiple cluster regions. In this embodiment, the K-means clustering algorithm is used. The number of clusters during the clustering process can be determined by methods such as the elbow method or contour coefficients. For example, in this embodiment, 20 second pixels are selected as initial cluster centers, and all second pixels are clustered into multiple cluster regions.

[0059] For any clustered region, the density of each clustered region is calculated based on the distribution of pixels within the clustered region. The steps include: performing linear fitting based on the coordinates of pixels in the clustered region to obtain the corresponding fitted line; obtaining the density based on the distance from the pixel to the fitted line and the distance from the pixel to the cluster center of the clustered region. The density is negatively correlated with the average Euclidean distance from all pixels in the clustered region to the cluster center and the average distance from all pixels in the clustered region to the corresponding fitted line.

[0060] The density of any cluster region can be calculated using the following formula:

[0061] In the formula, Indicates the first The density of each cluster region; Indicates the first The number of pixels in each cluster region excluding the initial cluster center; Represents pixels with the initial cluster centers The Euclidean distance between them; This represents the average distance between the line fitted by the least squares method to the coordinates of all pixels in the region and each pixel (i.e., the average fitted distance), which is used to characterize the linearity of the distribution; As the first weighting coefficient, As the second weighting coefficient, in this embodiment The weighting coefficients can be adjusted based on the experience of those skilled in the art. The weighting coefficients are set here mainly to improve the flexibility of the system. Indicated by An exponential function with base 0.

[0062] For the The smaller the average Euclidean distance between all pixels in a given cluster region and the cluster center, the better. The smaller the average fitting distance of each cluster region, the greater the density. Conversely, the larger the average Euclidean distance between all pixels within that region and the cluster center, the denser the cluster. The larger the average fitting distance of a cluster region, the smaller the density.

[0063] when , The first density threshold is set to [value] in this embodiment. The region is determined to be non-dense, meaning the pixels are distributed in a discrete point-like manner. It is marked as the first type of region, and the pixels in it are defined as the first target pixels, corresponding to the area of ​​edge reflection, for the red area in the figure.

[0064] when , The second density threshold is set to [value] in this embodiment. The region is determined to be dense, meaning that the pixels are densely distributed in lines or bands, and it is marked as a second type of region. The pixels in it are defined as second target pixels, corresponding to the region of the surface texture.

[0065] Based on the above analysis, a lower density indicates that the pixels in the clustered region are distributed as discrete points, while a higher density indicates that the pixels are distributed in a linear or band-like pattern. In this embodiment, the first density threshold is set to filter out regions where the pixels are distributed as discrete points, and the second density threshold is set to filter out regions where the pixels are distributed in a linear or band-like pattern. Therefore, the first density threshold is lower than the second density threshold.

[0066] The remaining areas are marked as the third type of area, and the pixels within the third type of area are the third target pixels.

[0067] S13: After smoothing the gray values ​​of the pixels in the first type region, perform superpixel segmentation on the pixels in the first type region and the third type region.

[0068] The first type of region typically contains noise or isolated highlights, which can easily lead to oversegmentation. Grayscale smoothing is applied to the first target pixel within this region.

[0069] The steps for smoothing the gray values ​​of pixels in the first type of region include: for any pixel in the first type of region, obtaining the gray values ​​of all pixels in the neighborhood of the pixel and obtaining the neighborhood gray-level histogram, and obtaining the peak gray value of the neighborhood gray-level histogram in the gray-level histogram.

[0070] The local grayscale difference is obtained by averaging the absolute values ​​of the differences between the grayscale value of a pixel and the grayscale values ​​of all pixels in its neighborhood.

[0071] The smoothed grayscale value is negatively correlated with the local grayscale difference, the difference between the grayscale value of a pixel and the peak grayscale value.

[0072] For any pixel within the first type of region, the formula for calculating its smoothed grayscale value can be expressed as: In the formula, It is the first type of region The grayscale value of each pixel after smoothing It is the first type of region The original grayscale value of each pixel; Indicates the first type of region The grayscale value of each pixel in the neighborhood of the i-th pixel is related to the grayscale value of the i-th pixel. The average value of the difference in grayscale values ​​of each pixel; Indicated by An exponential function with base 0; This represents the gray value of the pixel corresponding to the peak value in the grayscale histogram.

[0073] Indicates the first type of region The difference between the unsmoothed gray value of a pixel and the gray value of the pixel corresponding to the peak in the gray-level histogram. The larger this value is, the greater the difference between the pixel and its neighboring pixels, and thus the greater the degree of smoothing.

[0074] Combination Figure 4 After smoothing the pixels in the first type of region, the bright areas at the edges of the mineral particles are smoothed, thereby reducing the influence of the position of the mineral particle edges when performing superpixel segmentation. This allows for superpixel segmentation of the pixels in the first and third types of regions, resulting in multiple superpixel blocks.

[0075] After the pixels in the first cluster region are smoothed, the first cluster region will not affect the segmentation result when superpixel segmentation is performed on the surface of the ore particles, thus playing the role of merging superpixel blocks.

[0076] S14: Determine whether the second type of region is merged with the surrounding superpixel blocks based on the grayscale distribution characteristics of the second type of region; obtain the region of each mineral particle in the image based on the merged superpixel blocks.

[0077] For the second type of region within the surface of the ore particles, if the edge pixels of the second type of region exhibit a relatively uniform grayscale value change along its gradient direction, it indicates that the superpixel blocks on both sides of the second cluster region can be merged. If the edge pixels of the second cluster region exhibit a relatively non-uniform grayscale value change along its gradient direction, such as the appearance of black edge regions, it indicates that the superpixel blocks on both sides of the second cluster region cannot be merged.

[0078] Reference Figure 5 Step S14 includes steps S141-S142.

[0079] S141: Take the edge pixels of the region formed by the first pixel as the first edge point, and take the edge pixels of the second type of region as the second edge point; based on the gradient direction of the second edge point, obtain the boundary point traversal path.

[0080] In this embodiment, the second edge point is divided into fuzzy boundary points and defined boundary points based on the distance between the second edge pixel and the first edge point.

[0081] Specifically, for any second edge point, the nearest first edge point is obtained, and the Euclidean distance between them is calculated. For any second-class region, the average of the multiple Euclidean distances corresponding to each second edge point is calculated and denoted as the average distance of the second-class region. For example, second edge points in second-class regions with an average distance greater than 10 are classified as fuzzy boundary points, and second edge points in second-class regions with an average Euclidean distance less than or equal to 10 are classified as defined boundary points.

[0082] For a defined boundary point, the gradient direction of the defined boundary point can be obtained by using the Sobel operator to obtain the traversal path of the defined boundary point; for the traversal path of the second edge point, if the second edge point is a defined boundary point, the first edge point closest to the second edge point is obtained as the nearest boundary point, and the line segment extending along the gradient direction of the second edge point and intersecting the line perpendicular to the gradient direction of the first edge point is taken as the traversal path.

[0083] For fuzzy boundary points, the nearest definite boundary point is found. For ease of description, this nearest definite boundary point is defined as the alternative boundary point. The nearest boundary point of the alternative boundary point is obtained as the alternative nearest point. The intersection point of a straight line perpendicular to the gradient direction of the alternative nearest point is found along the gradient direction of the fuzzy boundary point. The probability of no intersection point is extremely low, so this case is not considered in this application. The path between the fuzzy boundary point and the intersection point is used as the traversal path.

[0084] S142: Obtain the difference in grayscale values ​​between every two adjacent pixels on the traversal path, and obtain the probability of merging the second type region with the superpixel blocks on both sides based on the average of all differences.

[0085] For any given boundary point, the average of the absolute values ​​of the differences between adjacent gray values ​​on the traversal path is calculated to obtain the gray change rate; the gray change rate is negatively correlated with the merging probability of the given boundary point.

[0086] The merging probability of any given boundary point can be calculated using the following formula:

[0087] In the formula, Indicates the first The merging probability that a given boundary point can be merged with a pixel on the traversal path. Indicates the first The traversal path of the nth defined boundary point 1 pixel Indicates the first The traversal path of the nth defined boundary point 1 pixel; This represents the difference in grayscale values ​​between two adjacent pixels along the traversal path.

[0088] This represents the average of the sums of the differences in grayscale values ​​between every two adjacent pixels along the traversal path. The larger this average value, the better the grayscale value along the traversal path. If the gradient direction of an edge pixel traverses pixels with significantly different grayscale values, then the... The probability of edge pixels merging with pixels along the traversal direction is relatively small. Indicated by An exponential function with base 0.

[0089] Through the above steps, the merging probability of all second edge points can be obtained. At this time, based on the merging probability of all second edge points in the second type region, the region probability of merging the second type region with the superpixel blocks on both sides can be calculated. The region probability of merging the second type region with the superpixel blocks on both sides is positively correlated with the mean of the merging probability of all edge pixels and negatively correlated with the variance of the merging probability of all edge pixels.

[0090] For any second-class region, the formula for calculating the region probability of merging with the superpixel blocks on both sides can be expressed as: In the formula, It is the first The probability that a second cluster region can merge with the superpixel blocks on both sides; Indicates the first The variance of the merging probability corresponding to each second edge point of each second cluster region; Indicates the first The number of second edge pixels in each of the second cluster regions; where the first There are a total of 2 second cluster regions One edge pixel here It is not equal to 0, so there is no need to consider the case where the denominator is 0 during the calculation; Indicates the second cluster region. The merging probability corresponding to each second edge point; This represents a linear normalization function, which can be normalized using maximum and minimum values.

[0091] When the If the variance of the merging probability corresponding to each edge pixel of a second cluster region is small and its average value is large, then it can be considered that the second cluster region has a high probability of merging with the surrounding superpixel blocks.

[0092] Based on the region probability, the second type of region is merged with the superpixel blocks on both sides. In this embodiment, a merging threshold is set. When the region probability is greater than the merging threshold, the second cluster region is determined to be the internal texture of the ore and should be merged with the surrounding superpixel blocks; otherwise, it is regarded as an independent region.

[0093] S2: Calculate the area of ​​each particle region to obtain the particle size of the crushed material; adjust the grinding intensity of the crusher based on the particle size.

[0094] After the above segmentation and merging process, the precise outline of each ore particle is obtained. The number of pixels in each independent region is counted, and this number is taken as the area of ​​the ore particle.

[0095] Calculate the average area of ​​all ore particles in the entire image, and record it as the current ore particle size. Adjust the grinding intensity of the crusher in real time based on this particle size. In the formula, This indicates the grinding particle size of the crusher. The preset standard particle size, This is the benchmark grinding efficiency coefficient, which is set based on the experience of those skilled in the art; This indicates the particle size of the material currently being crushed. As can be seen from the formula, when the monitored ore particle size... When the grinding capacity is large, the system automatically increases the grinding intensity to enhance the crushing effect; conversely, it decreases the intensity to achieve an optimal balance between energy consumption and efficiency.

[0096] This application also discloses an image recognition-based crusher grading material monitoring system, including a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement an image recognition-based crusher grading material monitoring method according to this application.

[0097] 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.

[0098] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A method of monitoring the classification of material by a crusher based on image recognition, characterised by, The method comprises the following steps: segmenting a gray image of the pretreated crushed material to obtain a plurality of particle regions, and calculating the area of each particle region to obtain the particle size of the crushed material; adjusting the grinding intensity of the crusher based on the particle size; wherein the step of segmenting the gray image of the pretreated crushed material to obtain a plurality of particle regions comprises: regarding the pixel points with a gray value less than or equal to a gray threshold value as first pixel points, and regarding the pixel points with a gray value greater than the gray threshold value as second pixel points; using a clustering algorithm on each second pixel point to obtain a plurality of clustering regions; for any clustering region, calculating the density of each clustering region based on the distribution of pixel points; regarding the clustering regions with a density lower than a first density threshold value as first-type regions, regarding the clustering regions with a density greater than a second density threshold value as second-type regions, and regarding the remaining clustering regions as third-type regions; performing smoothing processing on the gray values of the pixel points in the first-type regions, and then performing super-pixel segmentation on the pixel points in the first-type regions and the third-type regions; judging whether the second-type regions are merged with the surrounding super-pixel blocks according to the gray distribution characteristics of the second-type regions; and obtaining the region of each ore particle in the image according to the merged super-pixel blocks.

2. The image recognition-based crusher product monitoring method according to claim 1, characterized in that, The step of calculating the density of each clustering region based on the distribution of pixel points comprises: performing linear fitting based on the coordinates of the pixel points in the clustering region to obtain a corresponding fitting straight line, and obtaining the density based on the distance of the pixel points to the fitting straight line and the distance of the pixel points to the clustering center of the clustering region, wherein the density is negatively correlated with the average Euclidean distance of all pixel points in the clustering region to the clustering center and the average distance of all pixel points in the clustering region to the corresponding fitting straight line.

3. The image recognition-based crusher product monitoring method according to claim 1, characterized in that, The step of performing smoothing processing on the gray values of the pixel points in the first-type regions comprises: for any pixel point in the first-type regions, obtaining the gray values of all pixel points in the neighborhood of the pixel point to obtain a neighborhood gray histogram, and obtaining the peak gray value of the neighborhood gray histogram in the gray histogram; calculating the average value of the absolute values of the differences between the gray value of the pixel point and the gray values of the pixel points in the neighborhood to obtain a local gray difference; the smoothed gray value is negatively correlated with the local gray difference, the difference between the gray value of the pixel point and the peak gray value.

4. The image recognition-based crusher product monitoring method according to claim 1, characterized in that, The step of judging whether the second-type regions are merged with the surrounding super-pixel blocks according to the gray distribution characteristics of the second-type regions comprises: regarding the edge pixel points of the region formed by the first pixel points as first edge points, and regarding the edge pixel points of the second-type regions as second edge points; obtaining a traversal path of the determined boundary points based on the gradient direction of the second edge points; obtaining the difference in gray values between every two adjacent pixel points on the traversal path, and obtaining the region probability of the merging of the second-type regions and the super-pixel blocks on both sides according to the average value of all differences.

5. A method of monitoring the classification of material in a crusher based on image recognition as claimed in claim 4, characterised in that, The step of obtaining the traversal path of the second edge points based on the gradient direction of the second edge points comprises: classifying the second edge pixel points into fuzzy boundary points and determined boundary points based on the distance between the second edge pixel points and the first edge pixel points; For the certain boundary point, the first edge point closest to the certain boundary point is obtained as a nearest boundary point, and a line segment extending along the gradient direction of the second edge point and intersecting a straight line perpendicular to the gradient direction of the first edge point is taken as a traversal path; For the fuzzy boundary point, a certain boundary point closest to the fuzzy boundary point is taken as a substitute boundary point, a nearest boundary point of the substitute boundary point is obtained as a substitute nearest point, and an intersection point of a straight line perpendicular to the gradient direction of the substitute nearest point is found along the gradient direction of the fuzzy boundary point, and a path between the fuzzy boundary point and the intersection point is taken as the traversal path.

6. The image recognition-based crusher product monitoring method according to claim 4, wherein, The step of obtaining the region probability of the second type region and the superpixel blocks on both sides based on the average of all differences comprises: calculating the average of absolute values of differences of adjacent gray values on the traversal path for any certain boundary point to obtain a gray change rate; the gray change rate is negatively correlated with the merging probability of the certain boundary point; The region probability of the second type region and the superpixel blocks on both sides is calculated based on the merging probability of all edge pixel points in the second type region; the region probability of the second type region and the superpixel blocks on both sides is positively correlated with the average of all edge pixel point merging probabilities and negatively correlated with the variance of all edge pixel point merging probabilities; In response to the region probability of the second type region and the superpixel blocks on both sides being greater than a set threshold, the superpixel blocks on both sides of the second type region are merged.

7. The image recognition-based crusher product monitoring method according to claim 1, wherein, The preprocessing of the gray image of the crushed material comprises: processing the original image using adaptive filtering and enhancing the edges of the original image using edge sharpening technology.

8. The image recognition-based crusher product monitoring method according to claim 1, wherein, The average of the areas of the particle regions is taken as the granularity.

9. The image recognition-based crusher product monitoring method according to claim 1, wherein, The product of the granularity and a preset weight coefficient is taken as the grinding intensity.

10. An image recognition based crusher product monitoring system, characterized in that, The method comprises: A processor and a memory, the memory storing computer program instructions which, when executed by the processor, implement an image recognition-based crusher classified material monitoring method according to any one of claims 1-9.

Images