A method and system for detecting the particle size and strength grade of phosphate rock pellets

Abstract

A detection method for particle size and strength grade of phosphate concentrate pellets: 1) collecting images of the roasted phosphate concentrate pellets, and inputting the collected images into a deep learning instance segmentation network to obtain Mask information; 2) obtaining the particle size information of any pellet according to the boundary range of the pellet in the Mask information, and similarly obtaining the particle size information of each pellet; 3) cutting the corresponding Mask region in the original image according to the Mask information, classifying the pellets in the Mask region according to color grade, and weighting and determining according to the color grade of the pellets and the corresponding particle size information, and then obtaining the strength of the pellets. The scheme determines the boundary range and color information of each pellet by collecting pellet images, finally obtains the particle size and strength information of the pellets, avoids manual sampling, improves sampling efficiency, can monitor the particle size and strength grade of the pellets in real time, and avoids sample differences in the sampling process.

Description

Technical Field

[0001] This invention relates to a method and system for detecting particle size and strength grade, specifically a method and system for detecting the particle size and strength grade of phosphate rock pellets. Background Technology

[0002] Phosphate rock is the main raw material for the electric arc furnace (EAF) production of yellow phosphorus. Currently, the phosphate rock used in EAFs for yellow phosphorus production is lump ore with a particle size of 5mm or larger. However, given the characteristics of my country's phosphate rock resources, lump ore suitable for yellow phosphorus production accounts for only one-fiftieth of the total phosphate rock output, and high-quality lump ore is becoming increasingly scarce. Furthermore, a large amount of fine ore with a particle size of less than 5mm is generated during the mining, transportation, and drying pretreatment processes before entering the EAF. Typically, the amount of phosphate rock fines generated accounts for approximately 30%-50% of the total phosphate rock mined. In some highly weathered phosphate mining areas, the amount of fines generated during the process from mining to furnace entry can reach over 65%, resulting in large-scale accumulation of unused fines. In recent years, the yellow phosphorus market has remained at a high level, and the use of phosphate rock fines to prepare new specialized furnace feedstock for EAF yellow phosphorus smelting has become an emerging technology for improving the utilization rate of phosphate rock resources.

[0003] In the production of phosphate rock pellets and their application in electric arc furnace smelting, the strength and particle size of the pellets are important indicators for evaluating their performance. Generally speaking, the higher the strength of the phosphate rock pellets, the better their impact resistance, abrasion resistance, and high-temperature resistance. This is crucial for the transportation of phosphate rock pellets and the stable operation of the electric arc furnace smelting process. The particle size distribution of the phosphate rock pellets also has a significant impact on the electric arc furnace smelting process. Particles that are too large or too small can lead to problems such as poor airflow and difficulty in steam rising within the furnace.

[0004] Therefore, timely acquisition of particle size and strength indicators for phosphate rock pellets is of significant practical importance for both phosphate rock pellet production and electric arc furnace smelting of yellow phosphorus. To ensure that the particle size and strength of phosphate rock pellets meet requirements, various online monitoring devices, such as laser particle size analyzers and strength testers, are typically used in the production process to obtain these indicators, allowing for timely adjustments to production process parameters and ensuring stable quality of the phosphate rock pellets.

[0005] However, existing particle size and strength testing methods require a series of operations such as sampling, cooling, sieving, and strength testing of phosphate rock pellets. The test results take a long time to obtain, some operations require manual intervention, and the level of intelligence is insufficient.

[0006] Existing technologies include methods for detecting pellet size. For example, CN114627103A proposes a pellet size classification method based on video image recognition. This method involves preparing a training set of pellet images, constructing a pellet size recognition neural network, and initializing its parameters to obtain a primary pellet size recognition neural network. A loss function is then constructed, with the goal of minimizing the loss function. This primary neural network is trained using the training set until the loss function is less than α, resulting in a trained pellet size recognition neural network. Images of the pellets to be identified are acquired, pre-processed, and input into the trained neural network to obtain statistical results of the pellet size classification. By implementing statistical analysis and percentage analysis of pellet size classification, this method assists production operators or on-site staff in adjusting water application rates to bring the pellet size range back to normal. However, this method uses a neural network to train the image, which can only obtain the particle size classification information of the spheres. It cannot obtain the actual size of each sphere, nor can it obtain the strength information of the spheres. Summary of the Invention

[0007] To address the problems in existing technologies, such as long acquisition cycles for phosphate rock pellet detection results, the need for manual intervention in some operations, insufficient level of intelligence, and the inability to obtain the actual size and strength information of each pellet, this invention proposes a method and system for detecting the particle size and strength grade of phosphate rock pellets. After image acquisition and processing of roasted pellets, the particle size information and strength grade of the pellets are obtained based on the boundary range and color information of the pellets. The detection cycle is short, the results are accurate, and no manual operation is required.

[0008] According to a first embodiment of the present invention, a method for detecting the particle size and strength grade of phosphate rock pellets is provided.

[0009] A method for detecting the particle size and strength grade of phosphate rock pellets, the method comprising the following steps:

[0010] 1) Images of roasted phosphate rock pellets are acquired and input into a deep learning instance segmentation network to obtain the mask information corresponding to the images;

[0011] 2) Based on the boundary range of any given pellet in the obtained Mask information, the particle size information of that pellet is obtained, and so on, to obtain the particle size information of each pellet.

[0012] 3) Based on the Mask information, the corresponding Mask region is cut out from the original image. The spheres in the Mask region are color identified and classified by color level. The sphere strength is obtained by weighting the color level of the spheres and their corresponding granularity information.

[0013] Preferably, the deep learning instance segmentation network described in step 1) includes, but is not limited to, FCIS, Mask R-CNN, YOLACT-700, PolarMask, SoLo, PointRend, BlendMask, and SAM.

[0014] Preferably, the deep learning instance segmentation network in step 1) is Mask R-CNN; the specific process of using Mask R-CNN for instance segmentation is as follows:

[0015] (1) Image acquisition is performed on the roasted phosphate rock pellets, and the acquired phosphate rock pellet images are input into MaskR-CNN to obtain feature maps;

[0016] (2) Set a fixed number of regions of interest for each pixel position of the feature map, and feed the regions of interest into the region generation network to distinguish the foreground and background, as well as coordinate regression, to obtain the refined regions of interest;

[0017] (3) Perform region feature aggregation operation on the refined region of interest, that is, match the pixels of the original image and the feature map, and match the feature map with fixed features;

[0018] (4) Perform multi-class classification on the region of interest, namely candidate box regression and full convolution to generate a mask, and complete the instance segmentation task.

[0019] Preferably, in step 2), the boundary range of each ball is the minimum bounding rectangle of each ball;

[0020] Preferably, the boundary range of each pellet is determined by: determining the pellet size information based on the side length of the minimum bounding rectangle of each individual phosphate rock pellet; more preferably, the minimum bounding rectangle is defined by the x-axis of the pellet. min x max y min y max If the bounding rectangle is formed, then the particle size of the pellet is the longest side length of its smallest bounding rectangle, that is:

[0021] L = Max(x) min -x max y min -y max ... (Equation 1).

[0022] Preferably, the method further includes: 4) returning the unqualified pellets to the pelletizing process based on the pellet size information;

[0023] Preferably, when the particle size L of the pellet is >50mm or <8mm, the pellet is crushed and then returned to the pelletizing process; when the particle size L of the pellet satisfies 8mm≤L≤50mm, the pellet is qualified.

[0024] Preferably, the step 3) of determining the color information of each pellet comprises: distinguishing each independent phosphate rock pellet according to color, and dividing the pellets into n grades, wherein n≥3;

[0025] Preferably, according to the pellet strength in the original image, the corresponding colors from high strength to low strength are white (W), yellow (Y), red (R) and black (B) in sequence.

[0026] Preferably, the color classification further comprises color subdivision grades; each color grade is subdivided again, and the corresponding color subdivision grades from high strength to low strength are 1, 2, ..., m in sequence; m is 2 to 20, preferably 3 to 10. For example, white (W) is subdivided into W1, W2, ..., Wm, and other colors are deduced by analogy.

[0027] Preferably, the obtaining the pellet strength through weighted determination described in step 3) is obtained by fitting according to the pellet particle size information and the pellet color grade as follows:

[0028] x=αL+βC+δ......(Formula 2)

[0029] In the formula, x is the strength representative value of a single pellet; L is the particle size of a single pellet, in mm; C is the quantitative value of the pellet color grade; α is the particle size weight coefficient, with a value range of -10 to 10, preferably -5 to 5; β is the color weight coefficient, with a value range of 100 to 500, preferably 200 to 400; δ is the steady-state error coefficient, with a value range of 150 to 450, preferably 250 to 340; wherein the value of C is determined according to the actual situation in the actual production process.

[0030] Preferably, the method further comprises: 5) classifying the pellets into strength grades according to the specific strength of the pellets.

[0031] Preferably, the strength grade classification is specifically: when 0<x≤P1, the pellet is classified as grade A; when P1<x≤P2, the pellet is classified as grade B; when P2<x≤P3, the pellet is classified as grade C; when P3<x≤P4, the pellet is classified as grade D.

[0032] Wherein, P1, P2, P3 and P4 are grade boundaries, the value range of P1 is 200 to 300; the value range of P2 is 300 to 800; the value range of P3 is 800 to 1200; the value range of P4 is >1200. The specific values of P1, P2, P3 and P4 are determined according to the requirements in the actual production process.

[0033] Preferably, the method further includes: 6) returning the unqualified pellets to the pelletizing process or roasting process according to the strength grade of the pellets.

[0034] Preferably, after crushing the Grade A pellets, they are recycled to the pelletizing process; the Grade B pellets are recycled to the pellet roasting process for re-roasting; and the Grade C and Grade D pellets are roasted and then transported to the finished product warehouse.

[0035] According to a second embodiment of the present invention, a system for monitoring the particle size and strength of phosphate rock pellets is provided.

[0036] A system for monitoring the particle size and strength of phosphate rock pellets includes an image generation device, a transmission device, and an image analysis device. The image generation device is positioned above the roasted phosphate rock. The image generation device is connected to the image analysis device via the transmission device.

[0037] Preferably, the image generation device includes an industrial camera and an electrical control box. The industrial camera is connected to the electrical control box, and the electrical control box is connected to a transmission device.

[0038] Preferably, the transmission device includes a first photoelectric converter, an optical fiber, and a second photoelectric converter connected in series. The first photoelectric converter is connected to the image generation device, and the second photoelectric converter is connected to the image analysis device.

[0039] Preferably, the image analysis device is an image processing server.

[0040] In this invention, based on observations and research during actual production, different colors of pellets after roasting represent different roasting conditions and contain different materials. Therefore, the strength levels corresponding to different colored pellets are different. Generally speaking, the color and strength of the pellets follow the pattern: white > yellow > red > black. Based on this theory, this invention first acquires images of the roasted pellets, converts the images, and inputs them into a deep learning instance segmentation network. Then, it determines the boundary range of each pellet in the image to obtain the particle size information of each pellet. Finally, combining the particle size and color information, a weighted determination of the pellet strength is made.

[0041] In this invention, instance segmentation provides information such as the location and region of each independent cluster, which in turn allows for the calculation of the corresponding cluster's granularity. Specifically, the granularity of each cluster is determined based on its boundary range, i.e., the maximum side length L of the cluster's minimum bounding rectangle. Furthermore, since the graph contains numboxes of targets, their diameters need to be calculated individually to obtain the overall granularity distribution.

[0042] In this invention, by combining mask information and original image information, each individual phosphate rock pellet x can be distinguished.min / y min x max / y max The original image region, enclosed by a rectangle, is classified and identified (e.g., by constructing a convolutional neural network classifier). The original image corresponding to each individual phosphate rock pellet is then classified by color, such as white (W), yellow (Y), red (R), and black (B). Since different colors of phosphate rock pellets after roasting represent different roasting conditions and contain different materials, their corresponding intensity levels differ. The color-intensity relationship is: white > yellow > red > black. Preferably, the color concentration of the pellets is further subdivided according to the required intensity level, proposing intensity levels such as W1, W2, and W3, making the subsequent pellet intensity judgment process more accurate.

[0043] This invention proposes a weighted assessment of pellet strength by combining color grade and particle size information. The color grade of the pellets is quantified based on experience in actual production processes (e.g., W1, W2, W3; Y1, Y2, Y3; R1, R2, R3; B1, B2, B3, etc., which can be quantified as W1=3.6, W2=3.3, W3=3; Y1=2.6, Y2=2.3, Y3=2; R1=1.6, R2=1.3, R3=1; B1=0.6, B2=0.3, B3=0). The pellet strength is then calculated by combining weighting coefficients and steady-state coefficients. Preferably, the pellets are classified according to their strength grade to facilitate subsequent processing and storage. Furthermore, some pellets with substandard strength can be returned to the pelletizing or roasting process for re-pelleting or roasting, improving the overall strength of the phosphate rock pellets and simultaneously increasing the utilization rate of phosphate rock powder. In addition, depending on the particle size of the pattern, pellets that are too large or too small can be broken up and returned to the pelletizing process for re-pelleting and roasting.

[0044] In this invention, the weighting coefficients α and β, and the steady-state error coefficient δ in the pellet strength assessment process are determined based on the actual production process. Furthermore, when specifically classifying pellet strength grades, the specific values ​​of P1, P2, P3, and P4 are determined based on the actual production process.

[0045] In this invention, various deep learning-based instance segmentation algorithms can be used to process the obtained phosphate rock pellet images. For example, when using Mask R-CNN, the acquired phosphate rock pellet images are first input into a feature extraction network to obtain feature maps. Then, a fixed number of regions of interest (ROIs) are defined for each pixel position in the feature maps. These ROIs are then fed into a region generation network for binary classification (distinguishing between foreground and background) and coordinate regression to obtain refined ROIs. Next, a region feature aggregation operation is performed on the ROIs obtained in the previous step, i.e., pixel mapping between the original image and the feature map, and then mapping the feature map to fixed features. Finally, multi-class classification is performed on these ROIs, i.e., candidate box regression and fully convolutional mask generation, completing the instance segmentation task.

[0046] In this invention, a camera is used to capture images of the roasted phosphate rock pellets, avoiding damage to the samples caused by manual handling and improving sampling efficiency. Simultaneously, an algorithmic process allows for real-time monitoring of the particle size and strength grade of the phosphate rock pellets, shortening the detection cycle, avoiding sample differences during manual inspection and sampling, and enhancing the overall intelligence of the process.

[0047] In this invention, Figure 2 'a' represents the original image of the input instance segmentation; Figure 2 b is the visualized Mask image, and the original information of the Mask is a (0, 1) matrix; Figure 2 c is a color map of each individual phosphate rock pellet on the original image based on the mask information. It has no actual meaning and is not related to the subsequent pellet color identification. The color on this map only represents the function of distinguishing individuals and is used for visualization.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] 1. The present invention provides a method and system for detecting the particle size and strength grade of phosphate rock pellets. By acquiring images of the pellets, the boundary range and color information of each pellet are determined, and the particle size and strength information of the pellets are finally obtained. This avoids manual sampling, improves sampling efficiency, and can monitor the particle size and strength grade of the pellets in real time, avoiding sample differences during the sampling process.

[0050] 2. The present invention provides a method and system for detecting the particle size and strength grade of phosphate rock pellets. The particle size grade of the pellets is determined by the minimum bounding rectangle of the pellets, and the strength grade of the pellets is determined by the color and particle size of the pellets. Unqualified pellets are returned to the pelletizing or roasting process according to the particle size and strength grade, thereby improving the yield rate. Attached Figure Description

[0051] Figure 1This is a schematic diagram of the structure of a phosphate rock pellet particle size and strength grade detection system provided by the present invention.

[0052] Figure 2 'a' represents the original image segmented from the input instance.

[0053] Figure 2 b is the Mask image visualized after segmenting the input instance.

[0054] Figure 2 c is a colorized image of each individual phosphate pellet on the original image based on the mask information.

[0055] Figure 3 This is a schematic flowchart illustrating the strength grade determination process of a method for detecting the particle size and strength grade of phosphate rock pellets provided by the present invention.

[0056] Reference numerals: 1: Image generating device; 101: Industrial camera; 102: Electrical control box; 2: Transmission device; 201: First photoelectric converter; 202: Optical fiber; 203: Second photoelectric converter; 3: Image analysis device. Detailed Implementation

[0057] The technical solution of the present invention will be illustrated below with examples. The scope of protection sought by the present invention includes, but is not limited to, the following embodiments.

[0058] According to a first embodiment of the present invention, a method for detecting the particle size and strength grade of phosphate rock pellets is provided.

[0059] A method for detecting the particle size and strength grade of phosphate rock pellets, the method comprising the following steps:

[0060] 1) Images of roasted phosphate rock pellets are acquired and input into a deep learning instance segmentation network to obtain the mask information corresponding to the images;

[0061] 2) Based on the boundary range of any given pellet in the obtained Mask information, the particle size information of that pellet is obtained, and so on, to obtain the particle size information of each pellet.

[0062] 3) Based on the Mask information, the corresponding Mask region is cut out from the original image. The spheres in the Mask region are color identified and classified by color level. The sphere strength is obtained by weighting the color level of the spheres and their corresponding granularity information.

[0063] Preferably, the deep learning instance segmentation network described in step 1) includes, but is not limited to, FCIS, Mask R-CNN, YOLACT-700, PolarMask, SoLo, PointRend, BlendMask, and SAM.

[0064] Preferably, the deep learning instance segmentation network in step 1) is Mask R-CNN; the specific process of using Mask R-CNN for instance segmentation is as follows:

[0065] (1) Image acquisition is performed on the roasted phosphate rock pellets, and the acquired phosphate rock pellet images are input into Mask R-CNN to obtain feature maps;

[0066] (2) Set a fixed number of regions of interest for each pixel position of the feature map, and feed the regions of interest into the region generation network to distinguish the foreground and background, as well as coordinate regression, to obtain the refined regions of interest;

[0067] (3) Perform region feature aggregation operation on the refined region of interest, that is, match the pixels of the original image and the feature map, and match the feature map with fixed features;

[0068] (4) Perform multi-class classification on the region of interest, namely candidate box regression and full convolution to generate a mask, and complete the instance segmentation task.

[0069] Preferably, in step 2), the boundary range of each ball is the minimum bounding rectangle of each ball;

[0070] Preferably, the boundary range of each pellet is determined by: determining the pellet size information based on the side length of the minimum bounding rectangle of each individual phosphate rock pellet; more preferably, the minimum bounding rectangle is defined by the x-axis of the pellet. min x max y min y max If the bounding rectangle is formed, then the particle size of the pellet is the longest side length of its smallest bounding rectangle, that is:

[0071] L = Max(x) min -x max y min -y max ... (Equation 1).

[0072] Preferably, the method further includes: 4) returning the unqualified pellets to the pelletizing process based on the pellet size information;

[0073] Preferably, when the particle size L of the pellets is less than 8 mm or greater than 50 mm, the pellets are crushed and returned to the pelletizing process; when the particle size of the pellets is 8 mm ≤ L ≤ 50 mm, the pellets are qualified.

[0074] Preferably, step 3) involves determining the color information of each pellet by classifying each individual phosphate rock pellet according to its color and dividing it into n grades, where n ≥ 3.

[0075] Preferably, according to the pellet strength of the original image, the corresponding colors from high to low are white (W), yellow (Y), red (R) and black (B) in sequence.

[0076] Preferably, the color classification further comprises color subdivision grades; each color level is subdivided again, and the color subdivision grades corresponding to strength from high to low are 1, 2, ..., m in sequence; m is 2 to 20, preferably 3 to 10. For example, white (W) is subdivided into W1, W2, ..., Wm, and other colors follow the same logic.

[0077] Preferably, the pellet strength obtained by the weighted determination in step 3) is obtained by fitting according to the pellet size information and the pellet color grade as follows:

[0078] x=αL+βC+δ......(Formula 2)

[0079] Wherein, x is the strength representative value of a single pellet; L is the particle size of a single pellet, in mm; C is the quantitative value of the pellet color grade; α is the particle size weight coefficient, with a value range of -10 to 10, preferably -5 to 5; β is the color weight coefficient, with a value range of 100 to 500, preferably 200 to 400; δ is the steady-state error coefficient, with a value range of 150 to 450, preferably 250 to 340; wherein the value of C is determined according to the actual production situation.

[0080] Preferably, the method further comprises: 5) classifying the pellets into strength grades according to the specific strength of the pellets.

[0081] Preferably, the strength grade classification is specifically: when 0 < x ≤ P1, the pellet is classified as grade A; when P1 < x ≤ P2, the pellet is classified as grade B; when P2 < x ≤ P3, the pellet is classified as grade C; when P3 < x ≤ P4, the pellet is classified as grade D.

[0082] Wherein, P1, P2, P3, P4 are grade boundaries, the value range of P1 is 200 to 300; the value range of P2 is 300 to 800; the value range of P3 is 800 to 1200; the value range of P4 is > 1200. The specific values of P1, P2, P3 and P4 are determined according to the requirements in the actual production process.

[0083] Preferably, the method further comprises: 6) returning the unqualified partial pellets to the pelletizing process or roasting process according to the strength grade of the pellets.

[0084] Preferably, the grade A pellets are crushed and then recycled to the pelletizing process; the grade B pellets are recycled to the pellet roasting process for re-roasting; the grade C and grade D pellets finish roasting and are conveyed to the finished product bin.

[0085] According to a second embodiment of the present invention, a system for monitoring the particle size and strength of phosphate rock pellets is provided.

[0086] A system for monitoring the particle size and strength of phosphate rock pellets includes an image generating device 1, a transmission device 2, and an image analysis device 3. The image generating device 1 is positioned above the roasted phosphate rock. The image generating device 1 is connected to the image analysis device 3 via the transmission device 2.

[0087] Preferably, the image generation device 1 includes an industrial camera 101 and an electrical control box 102. The industrial camera 101 is connected to the electrical control box 102, and the electrical control box 102 is connected to the transmission device 2.

[0088] Preferably, the transmission device 2 includes a first photoelectric converter 201, an optical fiber 202, and a second photoelectric converter 203 connected in series. The first photoelectric converter 201 is connected to the image generation device 1, and the second photoelectric converter 203 is connected to the image analysis device 3.

[0089] Preferably, the image analysis device 3 is an image processing server.

[0090] Example 1

[0091] A system for monitoring the particle size and strength of phosphate rock pellets includes an image generating device 1, a transmission device 2, and an image analysis device 3. The image generating device 1 is positioned above the roasted phosphate rock. The image generating device 1 is connected to the image analysis device 3 via the transmission device 2.

[0092] Example 2

[0093] The embodiment 1 is repeated, except that the image generation device 1 includes an industrial camera 101 and an electrical control box 102. The industrial camera 101 is connected to the electrical control box 102, and the electrical control box 102 is connected to the transmission device 2.

[0094] The transmission device 2 includes a first photoelectric converter 201, an optical fiber 202, and a second photoelectric converter 203 connected in series. The first photoelectric converter 201 is connected to the image generation device 1, and the second photoelectric converter 203 is connected to the image analysis device 3.

[0095] The image analysis device 3 is an image processing server.

[0096] Application Example 1

[0097] A method for detecting the particle size and strength grade of phosphate rock pellets includes the following steps: 1) Image acquisition of roasted phosphate rock pellets; inputting the acquired phosphate rock pellets into a feature extraction network to obtain a feature map; then setting 20×20 regions of interest for each pixel position of the feature map; sending the regions of interest into a region generation network to distinguish foreground and background, and performing coordinate regression to obtain refined regions of interest; performing region feature aggregation operation on the refined regions of interest, that is, mapping the pixels of the original image and the feature map, and mapping the feature map to fixed features; finally, performing multi-class classification on the regions of interest, namely candidate box regression and fully convolutional mask generation, to complete the instance segmentation task.

[0098] 2) Determine the particle size information of any pellet based on the side length of the minimum bounding rectangle of the obtained Mask information; the minimum bounding rectangle is defined by the x-axis of the pellet. min x max y min y max If the bounding rectangle is formed, then the particle size of the pellet is the longest side length of its smallest bounding rectangle, that is:

[0099] L = Max(x) min -x max y min -y max )......(Equation 1);

[0100] 3) Based on the mask information, the corresponding mask regions are segmented from the original image. The spheres within these mask regions are then color-coded into four levels, with the colors corresponding to the intensity from highest to lowest being white (W), yellow (Y), red (R), and black (B). These four levels are further subdivided, with the colors corresponding to the intensity from highest to lowest being W1, W2, W3, Y1, Y2, Y3, R1, R2, R3, B1, B2, and B3.

[0101] Based on the pellet size information and pellet color grade, the following can be obtained through fitting:

[0102] x=αL+βC+δ......(Equation 2)

[0103] In the formula, x represents the intensity of a single pellet; L represents the particle size of a single pellet, in mm; C represents the quantified value of the pellet color grade (where W1 = 3.6, W2 = 3.3, W3 = 3; Y1 = 2.6, Y2 = 2.3, Y3 = 2; R1 = 1.6, R2 = 1.3, R3 = 1; B1 = 0.6, B2 = 0.3, B3 = 0); α represents the particle size weighting coefficient, with a value of -2.25; β represents the color weighting coefficient, with a value of 309; and δ represents the steady-state error coefficient, with a value of 299.

[0104] 4) According to the pellet particle size information obtained in step 2), pellets with a particle size satisfying 8mm≤L≤50mm are qualified pellets; the remaining pellets are crushed and returned to the pelletizing process. The proportion of qualified pellets in this batch of phosphate ore pellets is 97%.

[0105] 5) According to the pellet strength information obtained in step 3), the pellets are classified into strength grades: when 0<x≤P1, the pellet is classified as Grade A; when P1<x≤P2, the pellet is classified as Grade B; when P2<x≤P3, the pellet is classified as Grade C; when P3<x≤P4, the pellet is classified as Grade D; wherein, P1 is 220, P2 is 550, P3 is 960, and P4 is 1500.

[0106] 6) According to the strength grade of the pellets, after crushing Grade A pellets, they are recycled to the pelletizing process; Grade B pellets are recycled to the pellet roasting process for re-roasting; Grade C and Grade D pellets finish roasting and are conveyed to the finished product warehouse. In this batch of phosphate ore pellets, Grade A pellets account for 8%; Grade B pellets account for 14%; Grade C pellets account for 50%; Grade D pellets account for 25%.

[0107] Application Example 2

[0108] A detection method for particle size and strength grade of phosphate ore pellets, comprising the following steps: 1) acquiring images of roasted phosphate ore pellets, inputting the acquired phosphate ore pellets into a feature extraction network to obtain a feature map; then setting 15×15 regions of interest for each pixel position of the feature map, sending the regions of interest to a region proposal network to distinguish foreground from background and perform coordinate regression to obtain refined regions of interest; performing a region feature aggregation operation on the refined regions of interest, that is, corresponding pixels of the original image and the feature map, and corresponding the feature map to fixed features; finally performing multi-class classification on the regions of interest, that is, bounding box regression and fully convolution to generate a Mask, completing the instance segmentation task.

[0109] 2) determining the particle size information of a pellet according to the side length of the minimum bounding rectangle of any pellet in the obtained Mask information; the minimum bounding rectangle is formed by x min , x max , y min , y max in the rectangle frame, the particle size of the pellet is the maximum side length of its minimum bounding rectangle, that is:

[0110] L=Max(x min -x max y min -y max )......(Formula 1);

[0111] 3) Cutting the corresponding Mask region from the original image according to the Mask information, performing color recognition on the pellets in the Mask region and dividing them into 4 grades. From high strength to low strength, the corresponding colors are white (W), yellow (Y), red (R) and black (B) in sequence. Subdivision is performed on the basis of the 4 grades, and from high strength to low strength, the corresponding classifications are W1, W2, W3, Y1, Y2, Y3, R1, R2, R3, B1, B2, B3 in sequence;

[0112] According to the pellet size information and the pellet color grade, fitting can obtain:

[0113] x=αL+βC+δ......(Equation 2)

[0114] In the formula, x is the strength representation of a single pellet; L is the particle size of a single pellet, in mm; C is the quantitative value of the pellet color grade (wherein, W1=3.6, W2=3.3, W3=3; Y1=2.6, Y2=2.3, Y3=2; R1=1.6, R2=1.3, R3=1; B1=0.6, B2=0.3, B3=0); α is the particle size weight coefficient, with a value of -2.5; β is the color weight coefficient, with a value of 296; δ is the steady-state error coefficient, with a value of 310.

[0115] 4) According to the pellet size information obtained in step 2), pellets with a particle size of 8mm≤L≤50mm are qualified pellets; the remaining pellets are crushed and returned to the pelletizing process. In this batch of phosphate ore pellets, the proportion of qualified pellets is 95%.

[0116] 5) According to the pellet strength information obtained in step 3), the pellets are classified into strength grades: when 0<x≤P1, the pellets are classified as Grade A; when P1<x≤P2, the pellets are classified as Grade B; when P2<x≤P3, the pellets are classified as Grade C; when P3<x≤P4, the pellets are classified as Grade D; wherein P1 takes a value of 240, P2 takes a value of 610, P3 takes a value of 930, and P4 takes a value of 1600.

[0117] 6) According to the strength grade of the pellets, after crushing Grade A pellets, they are recycled to the pelletizing process; Grade B pellets are recycled to the pellet roasting process for re-roasting; Grade C and Grade D pellets after roasting are completed are conveyed to the finished product warehouse. In this batch of phosphate ore pellets, Grade A pellets account for 6%; Grade B pellets account for 17%; Grade C pellets account for 48%; Grade D pellets account for 24%.

Claims

1. A method for detecting the particle size and strength grade of phosphate rock pellets, characterized in that: The method includes the following steps: 1) Images of roasted phosphate rock pellets are acquired and input into a deep learning instance segmentation network to obtain the mask information corresponding to the images; 2) Based on the boundary range of any given pellet in the obtained Mask information, obtain the particle size information of that pellet, and so on, to obtain the particle size information of each pellet. 3) Based on the mask information, the corresponding mask region is cut out from the original image. Color recognition and color level classification are performed on the pellets within the mask region. Each individual phosphate pellet is distinguished by color and divided into n levels; where n≥3; color differentiation also includes subdivision levels; each color level is further subdivided, with the subdivision levels corresponding to the intensity from high to low being 1, 2, ..., m; where m is 2~20; based on the pellet size information and the pellet color level, the pellet intensity can be obtained by fitting: ... (Equation 2); where x is the representative strength value of a single pellet; L is the particle size of a single pellet, mm; C is the quantified value of the pellet color sub-level; α is the particle size weight coefficient, ranging from -10 to 10; β is the color weight coefficient, ranging from 100 to 500; δ is the steady-state error coefficient, ranging from 150 to 450.

2. The method according to claim 1, characterized in that: The deep learning instance segmentation networks mentioned in step 1) include FCIS, Mask R-CNN, YOLACT-700, PolarMask, SoLo, PointRend, BlendMask, and SAM.

3. The method according to claim 2, characterized in that: Step 1) describes a deep learning instance segmentation network called Mask R-CNN; the specific process for instance segmentation using Mask R-CNN is as follows: (1) Image acquisition of roasted phosphate rock pellets, inputting the acquired phosphate rock pellet images into Mask R-CNN to obtain feature maps; (2) Set a fixed number of regions of interest for each pixel position of the feature map, and feed the regions of interest into the region generation network to distinguish the foreground and background, as well as coordinate regression, to obtain the refined regions of interest; (3) Perform region feature aggregation operation on the refined region of interest, that is, match the pixels of the original image and the feature map, and match the feature map with fixed features; (4) Perform multi-class classification on the region of interest, namely candidate box regression and full convolution to generate a mask, and complete the instance segmentation task.

4. The method according to claim 1, characterized in that: Step 2) The boundary range of each ball is the minimum bounding rectangle of each ball.

5. The method according to claim 4, characterized in that: The boundary range of each pellet is determined by: determining the particle size information of the pellet based on the side length of the smallest bounding rectangle of each individual phosphate rock pellet.

6. The method according to claim 5, characterized in that: The minimum bounding rectangle is formed by the x-axis of the sphere. min x max y min y max If the bounding rectangle is formed, then the particle size of the pellet is the longest side length of its smallest bounding rectangle, that is: ... (Equation 1).

7. The method according to claim 6, characterized in that: The method also includes: 4) Returning unqualified pellets to the pelletizing process based on the pellet size information.

8. The method according to claim 7, characterized in that: When the pellet size L < 8mm or > 50mm, the pellet is crushed and returned to the pelletizing process; when the pellet size 8mm ≤ L ≤ 50mm, the pellet is qualified.

9. The method according to claim 1, characterized in that: Based on the original image, the colors corresponding to the intensity of the spheres from highest to lowest are white (W), yellow (Y), red (R), and black (B).

10. The method according to claim 1, characterized in that: m is 3~10.

11. The method according to claim 1, characterized in that: The value of α ranges from -5 to 5; the value of β ranges from 200 to 400; and the value of δ ranges from 250 to 340.

12. The method according to any one of claims 1-11, characterized in that: The method also includes: 5) classifying the strength grade of the pellets according to their specific strength.

13. The method according to claim 12, characterized in that: The intensity level classification is specifically as follows: when At that time, the team was classified as Grade A; when At that time, the team was classified as Grade B; when At that time, the team was classified as level C; when At that time, the ball was classified as level D; Among them, P1, P2, P3, and P4 are the level limits. The value range of P1 is 200~300; the value range of P2 is 300~800; the value range of P3 is 800~1200; and the value range of P4 is >1200.

14. The method according to claim 13, characterized in that: The method also includes: 6) Returning unqualified pellets to the pelletizing or roasting process according to the strength grade of the pellets.

15. The method according to claim 14, characterized in that: After the Grade A pellets are crushed, they are recycled to the pelletizing process; the Grade B pellets are recycled to the pellet roasting process for further roasting; the Grade C and Grade D pellets are roasted and then transported to the finished product warehouse.

16. A detection system applied to the method of any one of claims 1-15, characterized in that: The system includes an image generation device (1), a transmission device (2), and an image analysis device (3); the image generation device (1) is positioned above the roasted phosphate rock; the image generation device (1) is connected to the image analysis device (3) through the transmission device (2).

17. The detection system according to claim 16, characterized in that: The image generation device (1) includes an industrial camera (101) and an electrical control box (102); the industrial camera (101) is connected to the electrical control box (102), and the electrical control box (102) is connected to the transmission device (2).

18. The detection system according to claim 16, characterized in that: The transmission device (2) includes a first photoelectric converter (201), an optical fiber (202), and a second photoelectric converter (203) connected in series; the first photoelectric converter (201) is connected to the image generation device (1), and the second photoelectric converter (203) is connected to the image analysis device (3).

19. The detection system according to claim 16, characterized in that: The image analysis device (3) is an image processing server.

Images