An intelligent counting method for sintered ore particles based on multi-level image feature matching

Abstract

The present invention is applicable to the field of image processing technology, and provides an intelligent counting method for sintered ore particles based on multi-level image feature matching, including: obtaining video data of ore particles in free fall after being crushed by a crusher; extracting the morphological features of each ore particle in the Kth frame ore particle image and the K+1th frame ore particle image in the video data, and based on the dynamic model and the extracted morphological features, tracking the morphological features of each ore particle in the Kth frame ore particle image in the K+1th frame ore particle image through multi-level feature matching, and obtaining the preliminary determination result of the area corresponding to each ore particle in the video data; respectively for each ore particle in the video data, determining the maximum area of ​​the ore particle according to the preliminary determination results of all areas corresponding to the ore particle; according to the preset ore particle area threshold and the maximum area of ​​each ore particle, dividing each ore particle into particle sizes, and counting the number of ore particles contained in different particle sizes. The present invention can improve the accuracy of sintered ore particle counting.

Description

Technical Field

[0001] The invention belongs to the technical field of image processing, and in particular relates to an intelligent counting method for sintered ore particles based on multi-level feature matching of images. Background Art

[0002] The sintering production process is to crush the raw materials such as iron ore and coke powder into powder by crusher, mix the powdered iron-containing raw materials, flux, fuel, return ore, etc. in a certain proportion, add appropriate amount of water to mix and granulate, then spread them on the sintering machine, ignite and sinter under the action of lower exhaust. Under the action of high temperature generated by fuel combustion and a series of physical and chemical reactions, porous large ore particles are generated, and then sintered ore is generated.

[0003] During the sintering process, there are strict requirements on the particle size of various solid raw materials entering the sintering machine. For example, the particle size of concentrate should be less than 10mm, and the particle size of solid fuels such as coke powder and anthracite should be less than 3mm. At present, many steel mills generally obtain the particle size distribution of coke powder through manual sampling and screening, and adjust the crusher based on this to control the particle size of coke powder. In the context of the continuous strengthening of resource and environmental constraints, innovation drives the transformation and upgrading of traditional manufacturing to intelligent manufacturing. Efficient and accurate ore particle tracking and counting are the basis for improving the control accuracy of sintering fuel particle size and the level of intelligent sintering production, which is of great significance for improving the quality, reducing carbon and increasing efficiency of the sintering process.

[0004] With the development of machine vision technology, researchers at home and abroad have conducted research on the problem of mineral particle counting. Among them, the area statistics method is to divide the total area of ​​all mineral particles by the mineral particle area threshold to obtain the number of mineral particles. The area statistics method is suitable for static scenes with fixed mineral particle area, but coke powder mineral particles have various particle sizes, and the counting accuracy of this method is not high; the virtual coil method is to set a virtual frame on the image and count when mineral particles pass through the virtual frame, but when the mineral particles move faster, some mineral particles will not appear in the virtual frame, and the counting error is large; the counting method based on the detection line sets a virtual detection line on the image to count the mineral particles below the detection line, but coke powder mineral particles will undergo spin motion during free fall, and the same mineral particle presents multi-visual dimensional features. This method is difficult to extract multi-visual dimensional features of mineral particles.

[0005] Therefore, how to combine the actual situation of the sintering production process, use machine vision technology to solve the problem of inaccurate sintering particle counting, quickly and accurately perform online intelligent counting of ore particles, effectively control the particle size of coke powder and stabilize sintering production, improve the quality and output of sintered ore, and reduce production energy consumption during the sintering process is an urgent problem to be solved in the current sintering field. Summary of the invention

[0006] The embodiment of the present invention provides a method for intelligently counting sintered ore particles based on multi-level feature matching of images, which can solve the problem of inaccurate counting of sintered ore particles.

[0007] The embodiment of the present invention provides a method for intelligently counting sintered ore particles based on multi-level feature matching of images, comprising:

[0008] Obtain video data of the free-falling motion of ore particles after being crushed by the crusher;

[0009] For the adjacent K-th frame mineral particle image and K+1-th frame mineral particle image in the video data, the morphological features of each mineral particle in the K-th frame mineral particle image and the K+1-th frame mineral particle image are extracted, and based on the dynamic model and the extracted morphological features, the morphological features of each mineral particle in the K-th frame mineral particle image in the K+1-th frame mineral particle image are tracked through multi-level feature matching to obtain a preliminary determination result of the area corresponding to each mineral particle in the video data; K=1,2,…,m-1, m represents the total number of frames of mineral particle images in the video data;

[0010] For each mineral particle in the video data, the maximum area of ​​the mineral particle is determined according to the preliminary determination results of all areas corresponding to the mineral particle;

[0011] According to the preset mineral particle area threshold and the maximum area of ​​each mineral particle, each mineral particle is divided into particle sizes, and the number of mineral particles contained in different particle sizes is counted.

[0012] The above scheme of the present invention has the following beneficial effects:

[0013] In an embodiment of the present invention, the mineral particles are tracked by using a multi-level feature matching method, each mineral particle is taken as an object, and a preliminary determination result of an area corresponding to the object is determined by the morphological features of the object in two adjacent frames of mineral particle images, and then the maximum area of ​​the object is screened out based on the preliminary determination results of all areas of the object, and finally each object is divided into particle sizes based on the maximum area of ​​each object, and the number of mineral particles contained in different particle sizes is statistically obtained. Among them, the image used for mineral particle tracking comes from the video data of the mineral particles in free fall. Based on the morphological features presented by the self-spin of the mineral particles in the free fall process, the counting method of the present invention can effectively extract the multi-visual dimensional features presented by each mineral particle in the free fall process, and fully obtain the real particle size information of the mineral particles. Therefore, the counting method of the present invention can greatly reduce the counting error of mineral particles of different particle sizes and improve the accuracy of sintered mineral particle counting.

[0014] Other beneficial effects of the present invention will be described in detail in the subsequent specific implementation section. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0016] Figure 1 A flow chart of a method for intelligently counting sintered ore particles based on multi-level image feature matching provided by an embodiment of the present invention;

[0017] Figure 2 It is a bar chart comparing the counting results of mineral particles with a size less than 3 mm in different experimental samples using different methods;

[0018] Figure 3 It is a bar chart comparing the counting results of mineral particles with a size of 3-5 mm in different experimental samples using different methods;

[0019] Figure 4 It is a bar chart comparing the counting results of mineral particles with a diameter greater than 5 mm in different experimental samples using different methods;

[0020] Figure 5 It is a bar chart comparing the relative errors of the counting results of mineral particles with a diameter less than 3 mm in different experimental samples using different methods;

[0021] Figure 6 It is a bar chart comparing the relative errors of the counting results of mineral particles with a diameter of 3-5 mm in different experimental samples using different methods;

[0022] Figure 7 This is a bar chart comparing the relative errors of the counting results of mineral particles with a diameter greater than 5 mm in different experimental samples using different methods. DETAILED DESCRIPTION

[0023] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present invention. However, it should be clear to those skilled in the art that the present invention may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present invention.

[0024] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof.

[0025] It should also be understood that the term "and / or" used in the present description and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0026] As used in the present specification and the appended claims, the term "if" may be interpreted as "when" or "upon" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" may be interpreted as meaning "upon determination" or "in response to determining" or "upon detection of [described condition or event]" or "in response to detecting [described condition or event]", depending on the context.

[0027] In addition, in the description of the present specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0028] References to "one embodiment" or "some embodiments" etc. described in the present specification mean that one or more embodiments of the present invention include specific features, structures or characteristics described in conjunction with the embodiment. Therefore, the statements "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. that appear in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0029] In view of the current problem of inaccurate counting of sintered ore particles, an embodiment of the present invention provides an intelligent counting method for sintered ore particles based on multi-level feature matching of images. The technical method tracks the ore particles by using multi-level feature matching, takes each ore particle as an object, and determines a preliminary determination result of an area corresponding to the object based on the morphological features of the object in two adjacent frames of ore particle images. Then, based on the preliminary determination results of all areas of the object, the maximum area of ​​the object is screened out. Finally, each object is divided into particle sizes based on the maximum area of ​​each object, and the number of ore particles contained in different particle sizes is statistically obtained. Among them, the image used for ore particle tracking comes from the video data of ore particles in free fall. Based on the morphological features presented by the self-spinning of ore particles in the free fall process, the counting method of the present invention can effectively extract the multi-visual dimensional features presented by each ore particle in the free fall process, and fully obtain the real particle size information of the ore particles. Therefore, the counting method of the present invention can greatly reduce the counting error of ore particles of different particle sizes and improve the accuracy of sintered ore particle counting.

[0030] The following is an exemplary description of the sintered ore particle intelligent counting method based on multi-level image feature matching provided by the present invention in conjunction with specific embodiments.

[0031] The embodiment of the present invention provides a method for intelligent counting of sintered ore particles based on multi-level feature matching of images. The method can be executed by a terminal device or by a device (such as a chip) applied to the terminal device. The following embodiment takes the method executed by a terminal device as an example. As an example, the terminal device can be a tablet, a server, or a laptop computer, etc., which is not limited in the embodiment of the present invention.

[0032] like Figure 1 As shown, the sintered ore particle intelligent counting method based on multi-level image feature matching provided by the embodiment of the present invention comprises the following steps:

[0033] Step 11, obtaining video data of the free-falling motion of the ore particles after being crushed by the crusher.

[0034] In some embodiments of the present invention, a high-speed industrial camera may be used to capture video data of the free-falling motion of the above-mentioned mineral particles at a fixed frame rate F. It should be noted that the mineral particles in the above-mentioned video data are of the same mineral particle material.

[0035] For example, the above video data can be shot against a white background, so that the black mineral particles can be distinguished from the white background. After preprocessing based on machine vision technology, the morphological characteristics of each mineral particle can be effectively extracted, laying the foundation for effectively tracking each mineral particle.

[0036] Step 12, for the adjacent K-th frame mineral particle image and K+1-th frame mineral particle image in the video data, extract the morphological features of each mineral particle in the K-th frame mineral particle image and the K+1-th frame mineral particle image, and based on the dynamic model and the extracted morphological features, track the morphological features of each mineral particle in the K-th frame mineral particle image in the K+1-th frame mineral particle image through multi-level feature matching to obtain a preliminary determination result of the area corresponding to each mineral particle in the video data.

[0037] The above K=1, 2, ..., m-1, m represents the total number of frames of mineral particle images in the video data, that is, for every two adjacent frames of mineral particle images in the above video data, the above step 12 will be executed until all adjacent frames in the video data are processed.

[0038] It is worth mentioning that by tracking mineral particles through multi-level feature matching, each mineral particle is treated as an object, and the morphological features of the object in the previous frame and the current frame are used to jointly decide whether to count it. This can effectively track the appearance and departure of each mineral particle and perform accurate counting.

[0039] Step 13, for each mineral particle in the video data, determine the maximum area of ​​the mineral particle according to the preliminary determination results of all areas corresponding to the mineral particle.

[0040] In an embodiment of the present invention, after all adjacent frames in the video data are processed according to the above step 12, the number of preliminary determination results of the area corresponding to each mineral particle in the video data is one or more. When determining the maximum area of ​​each mineral particle in the video data, for each mineral particle in the video data, the maximum value of all preliminary determination results of the area corresponding to the mineral particle can be used as the maximum area of ​​the mineral particle.

[0041] Step 14, according to the preset mineral particle area threshold and the maximum area of ​​each mineral particle, each mineral particle is divided into particle sizes, and the number of mineral particles contained in different particle sizes is counted.

[0042] Since the same type of mineral particles have different particle sizes, the above-mentioned preset mineral particle area threshold may include multiple values. For example, assuming that the particle sizes of the mineral particles include three types: less than 3 mm, 3-5 mm, and greater than 5 mm, the mineral particle area thresholds corresponding to the three particle sizes are different from each other.

[0043] Based on this, after obtaining the maximum area of ​​the mineral particles in the video, the maximum area can be compared with the preset mineral particle area threshold to determine the mineral particle area threshold corresponding to the maximum area, and the particle size corresponding to the determined mineral particle area threshold is used as the particle size of the mineral particles. It can be understood that after determining the particle size of each mineral particle, the number of mineral particles contained in each particle size can be quickly counted.

[0044] It can be seen that in some embodiments of the present invention, the mineral particles are tracked by using a multi-level feature matching method, each mineral particle is taken as an object, and a preliminary determination result of an area corresponding to the object is determined by the morphological features of the object in two adjacent frames of mineral particle images, and then the maximum area of ​​the object is screened out based on the preliminary determination results of all areas of the object, and finally each object is divided into particle sizes based on the maximum area of ​​each object, and the number of mineral particles contained in different particle sizes is statistically obtained. Among them, the image used for mineral particle tracking comes from the video data of the mineral particles in free fall. Based on the morphological features presented by the spin of the mineral particles in the free fall process, the counting method of the present invention can effectively extract the multi-visual dimensional features presented by each mineral particle in the free fall process, and fully obtain the real particle size information of the mineral particles. Therefore, the counting method of the present invention can greatly reduce the counting error of mineral particles of different particle sizes and improve the accuracy of sintered mineral particle counting.

[0045] The specific implementation of the above step 12 is exemplarily described below in conjunction with a specific embodiment.

[0046] In some embodiments of the present invention, the morphological features in the above step 12 include centroid coordinates and area.

[0047] Correspondingly, the specific implementation method of extracting the morphological features of each mineral particle in the Kth frame mineral particle image and the K+1th frame mineral particle image is: extracting the centroid coordinates and area of ​​each mineral particle in the Kth frame mineral particle image, and the centroid coordinates and area of ​​each mineral particle in the K+1th frame mineral particle image. The centroid coordinates of each mineral particle include the horizontal coordinate and the vertical coordinate of the mineral particle.

[0048] In some scenarios of the present invention, in order to improve the accuracy of the above-mentioned center of mass coordinates and area, the specific implementation method of extracting the morphological characteristics of each mineral particle in the Kth frame of mineral particle image and the K+1th frame of mineral particle image can be: first, the Kth frame of mineral particle image and the K+1th frame of mineral particle image are converted into grayscale images, and the noise in the grayscale images is eliminated, and then the grayscale images after noise elimination are binarized to obtain the Kth frame of mineral particle image and the K+1th frame of mineral particle image after binarization, and finally the morphological characteristics of each mineral particle in the Kth frame of mineral particle image and the K+1th frame of mineral particle image after binarization are extracted.

[0049] In some embodiments of the present invention, the morphological features of the mineral particles can be extracted by using the Regionprops function, which is a function used to measure the properties of the image region and is often used to calculate the properties of the marked region, such as the area and centroid.

[0050] In some embodiments of the present invention, the specific implementation method of determining the preliminary determination result of the area corresponding to each mineral particle in the video data in the above step 12 includes the following steps:

[0051] Step 12.1, for the i-th mineral particle q in the K-th frame mineral particle image i,k , based on the kinetic model, calculate the i-th mineral particle q i,k Free fall from the initial position of free fall motion to position (x i,k ,y i,k ) at a speed v i,k , and calculate the i-th mineral particle q i,k Displacement s during falling time T i,k+1 i=1,2,…,I, where I represents the total number of mineral particles in the Kth frame of mineral particle image.

[0052] Specifically, the formula Calculate the i-th mineral particle q i,k Free fall from the initial position of free fall motion to position (x i，k ,y i，k ) at a speed v i，k , through the formula Calculate the i-th mineral particle q i,k Displacement s during falling time T i,k+1 .

[0053] It should be noted that the formula and formula This is the kinetic model mentioned above.

[0054] Where T represents the time interval between taking the Kth frame of mineral particle image and the K+1th frame of mineral particle image, T=1 / F, (x i,k ,y i,k ) represents the i-th mineral particle q in the K-th mineral particle image i,k The coordinates of the center of mass, x i,k represents the i-th mineral particle q in the K-th mineral particle image i,k The horizontal axis, y i,k represents the i-th mineral particle q in the K-th mineral particle image i,k The vertical coordinate of Represents the mineral particle q i,k The acceleration of the free fall motion, h represents the particle Q i,k It falls from the initial position of the free fall motion to a fixed height at the upper edge of an imaging area (the imaging area is the imaging area of ​​an image acquisition device used to capture the above video data).

[0055] It should be noted that the above It can be obtained in advance. For example, before executing the above step 11, the acceleration of the same mineral particles with different particle sizes in free fall can be obtained in advance and stored in the acceleration query table. Based on this, the mineral particle Q in the Kth frame mineral particle image is i,k The area of ​​the mineral particle q i,k The particle size l in the Kth frame of the mineral particle image i,k , you can query and determine the mineral particles q i,k The acceleration of free fall.

[0056] Step 12.2, based on the initial position and displacement s i,k+1 , determine the i-th mineral particle q i,k Whether to leave the imaging area of ​​the image acquisition device used to collect video data, if the i-th mineral particle q i,k If the i-th mineral particle q i,k Leave the imaging area and proceed to step 12.8.

[0057] Among them, since the imaging area of ​​the above-mentioned image acquisition device and the initial position of the free fall movement are fixed, the initial position and displacement of the mineral particles can be used to determine whether the current position of the mineral particles exceeds the imaging area of ​​the image acquisition device. If so, it is considered that the mineral particles have left the imaging area, otherwise, it is considered that the mineral particles have not left the imaging area.

[0058] Step 12.3, select the set of mineral particles Q that meet the first-level feature matching conditions from the K+1th frame of mineral particle image nl,k+1 .

[0059] Among them, the first-level feature matching conditions are:

[0060] |x i,k -x r,k+1 |≤ε x

[0061] The above x r,k+1 represents the horizontal coordinate of the rth mineral particle in the K+1th mineral particle image, ε x Represents the mineral particle q i,k The upper limit of the horizontal coordinate offset between the Kth frame mineral particle image and the K+1th frame mineral particle image, ε x The specific data can be set according to actual conditions.

[0062] Correspondingly, the set of all mineral particles that meet the first-level feature matching conditions in the K+1th frame of mineral particle image can be taken as the mineral particle set Q nl,k+1 That is, if the horizontal coordinate (x r,k+1 ) meets the above first-level feature matching conditions, then the mineral particle is taken as the mineral particle set Q nl,k+1 An element in .

[0063] Step 12.4, at the i-th mineral particle q i,k The particle size l i,k Next, determine the mineral particle set Q nl,k+1 Is there a mineral particle q that meets the second-level feature matching conditions? nl,k+1 , if the mineral particle set Q nl,k+1 There are mineral particles q nl,k+1 , then execute step 12.5. If the mineral particle set Q nl,k+1 There is no mineral particle q nl,k+1 , then proceed to step 12.6.

[0064] Among them, the second-level feature matching conditions are:

[0065]

[0066] The above x r,k+1 represents the horizontal coordinate of the rth mineral particle in the K+1th frame mineral particle image, y r,k+1 represents the ordinate of the rth mineral particle in the K+1th frame mineral particle image, ε xy Represents the mineral particle q i,k After the falling time T, the upper limit of the deviation between the position predicted by the dynamic model and the actual position, ε xy The specific data can be set according to actual conditions.

[0067] That is, the mineral particles can be aggregated into Q nl,k+1 The median coordinate (x r,k+1 ,y r,k+1 ) The mineral particles that meet the above second-level feature matching conditions are taken as mineral particles q nl,k+1 .

[0068] Step 12.5, the i-th mineral particle q i,k The area and mineral particles q nl,k+1 The larger area of ​​the mineral particle q i,k The corresponding area is preliminarily determined.

[0069] Step 12.6, determine the particle set Q at the first particle size nl,k+1 Is there a mineral particle q that meets the second-level feature matching conditions? bl,k+1 , and at the second particle size, the mineral particle set Q nl,k+1 Is there a mineral particle q that meets the second-level feature matching conditions? al,k+1 , if the mineral particle set Q nl,k+1 There are mineral particles q bl,k+1 and mineral particles al,k+1 , then proceed to step 12.7.

[0070] The first particle size and particle size l i,k The second particle size is one of three particle sizes arranged in sequence in the preset particle size type table, wherein the preset particle size type table records a plurality of particle sizes arranged from small to large (such as three particle sizes of less than 3 mm, 3-5 mm, and greater than 5 mm), and the first particle size is smaller than the particle size l i,k , particle size l i,k Based on this, the above step 12.6 can be understood as: Under the lower particle size, determine the mineral particle set Q nl,k+1 Is there a mineral particle q that meets the second-level feature matching conditions? bl,k+1 , and at a higher particle size, the mineral particle set Q nl,k+1 Is there a mineral particle q that meets the second-level feature matching conditions? al,k+1 It is worth mentioning here that by screening and matching the mineral particles at a lower particle size and a higher particle size, the multi-visual dimensional features of each mineral particle during the free fall process can be effectively extracted, the true particle size information of the mineral particles can be fully obtained, and the counting error of mineral particles of different particle sizes can be reduced.

[0071] Step 12.7, from the mineral particle q bl,k+1 and mineral particles al,k+1 The mineral particles that meet the third-level feature matching conditions are screened, and the area and q of the mineral particles that meet the third-level feature matching conditions are calculated. i,k The larger of the areas of i,k The corresponding area is preliminarily determined.

[0072] The third-level feature matching conditions are as follows:

[0073] min(ω1ΔS1+ω2ΔA1,ω1ΔS2+ω2ΔA2)

[0074] Among them, ΔS1 represents the ore particle q bl,k+1 With mineral particles i,k The Euclidean distance between the two particles, ΔA1 indicates the ore particle q bl,k+1 With mineral particles i,k The area difference between the two, ΔS2 represents the ... al,k+1 With mineral particles i,k The Euclidean distance between the two particles, ΔA2 indicates the ore particle q al,k+1 With mineral particles i,k ω1 and ω2 represent weighting coefficients, and both ω1 and ω2 can be set according to actual conditions.

[0075] The above-mentioned Euclidean distance can be calculated by using a commonly used Euclidean distance calculation method. Here, ΔS1 is taken as an example to illustrate the formula.

[0076] Specifically, Among them, x i,k+1 Represents the mineral particle q bl,k+1 The horizontal axis, y i,k+1 Represents the mineral particle q bl,k+1 The vertical coordinate of .

[0077] Step 12.8, the i-th mineral particle q i,k The area of ​​the mineral particle q i,k The corresponding area is preliminarily determined.

[0078] It should be noted that the matching method using the first-level feature matching condition, the second-level feature matching condition and the third-level feature matching condition is the multi-level feature matching mentioned above.

[0079] The effect of the above-mentioned intelligent counting method for sintered ore particles is illustrated below by taking a specific example.

[0080] Step 1: Obtain the acceleration of the same type of mineral particles with different particle sizes in free fall.

[0081] In this example, 100 particles of each of three types of coke powder particles with particle sizes of less than 3mm, 3-5mm and greater than 5mm are taken. A high-speed camera is used to shoot a video of the free fall of 100 particles of the first particle size at a fixed frame rate. The average acceleration g1 of the 100 particles in free fall is calculated as the acceleration of the particles of the first particle size in the falling motion. Similarly, the accelerations of the particles of the second and third particle sizes in the falling motion are g2 and g3.

[0082] Step 2: Take the same mineral particles of different particle sizes and construct mineral particle experimental samples according to the proportions of different mineral particle numbers.

[0083] Specifically, three kinds of coke powder particles with particle sizes of less than 3 mm, 3-5 mm and greater than 5 mm were taken, and the particle experimental samples were constructed according to different particle number ratios, as shown in Table 1.

[0084]

[0085] Table 1

[0086] Step 3, using a high-speed industrial camera to shoot a free-fall motion video of the mineral particle experimental sample at a fixed frame rate.

[0087] Specifically, the experimental mineral particle samples are made to freely fall in a curtain-like manner from a fixed height, and a high-speed camera is used to shoot a video of the free-falling movement of the mineral particles.

[0088] Step 4: extract the morphological features of each mineral particle in the mineral particle image of the Kth frame and the K+1th frame in the free-fall motion video.

[0089] Specifically, the centroid coordinates and area of ​​each mineral particle in the K-th frame mineral particle map, and the centroid coordinates and area of ​​each mineral particle in the K+1-th frame mineral particle map are extracted.

[0090] Step 5, based on the mineral particle dynamics model and multi-level feature matching algorithm, track the position and morphological characteristics of each mineral particle in the Kth frame mineral particle image in the K+1th frame mineral particle image, and determine the preliminary result of the area of ​​each mineral particle.

[0091] Step 6, after processing all adjacent frames of the free-fall motion video, determine the maximum area of ​​each mineral particle based on the preliminary determination result of the area of ​​each mineral particle, and divide each mineral particle into particle sizes based on the mineral particle area threshold, and count the number of mineral particles contained in the three particle sizes.

[0092] Specifically, after processing all adjacent frames of a free-fall motion video of a mineral particle experimental sample, all mineral particles appearing in the video are divided into particle sizes based on the pixel area corresponding to the particle sizes of 3 mm and 5 mm, and the number of mineral particles contained in the three particle sizes is counted.

[0093] The following is an exemplary description of step 4 in conjunction with a specific embodiment.

[0094] In some embodiments of the present invention, the specific implementation method before extracting the morphological features of each mineral particle in the Kth frame and the K+1th frame mineral particle image in the free-fall motion video further includes the following steps:

[0095] Step 401, converting the Kth frame and the K+1th frame mineral particle images into grayscale images.

[0096] Specifically, the mean method is introduced to convert the K-th frame and the K+1-th frame mineral particle images into grayscale images.

[0097] Step 402, eliminating the noise of the grayscale images of the Kth frame and the K+1th frame.

[0098] Specifically, a Gaussian filter is performed on the grayscale images of the Kth frame and the K+1th frame using a Gaussian kernel with a size of 5×5 and a standard deviation of 0 to obtain a filtered mineral particle map.

[0099] Step 403: Binarization processing is performed on the grayscale images of the Kth frame and the K+1th frame after noise elimination.

[0100] Specifically, an adaptive threshold algorithm is introduced to perform binarization processing on the grayscale images of the Kth frame and the K+1th frame after noise elimination to obtain the preprocessed mineral particle image.

[0101] The above-mentioned introduction of the adaptive threshold algorithm to perform binarization processing on the grayscale images of the Kth frame and the K+1th frame after noise elimination includes the following steps:

[0102] Step 1, record the input grayscale image as:

[0103] I(x,y)

[0104] where I(x,y)∈[0,255].

[0105] Step 2: For a certain pixel src(x,y)∈I(x,y) of the grayscale image in step 1, calculate the average value Mean_C of all pixels in a neighborhood of size Block_size centered on the pixel, where Block_size is the height and width of the area, which is generally an odd number.

[0106] Step 3: Subtract the offset adjustment amount C from the mean value Mean_C obtained in step 2 to obtain the local threshold T(x, y), that is, T(x, y) = Mean_C - C.

[0107] Step 4: When src(x,y)>T(x,y), the pixel dst(x,y) of the output image is 255, otherwise, the pixel dst(x,y) of the output image is 0.

[0108] Step 5, looping steps 2 to 4 until every pixel of the grayscale image in step 1 is processed, thereby obtaining an output image, that is, a binarized image corresponding to the input grayscale image.

[0109] Based on the dynamic model and multi-level feature matching algorithm, the position and morphological characteristics of each mineral particle in the Kth frame mineral particle image in the K+1th frame mineral particle image are tracked, and the area corresponding to each mineral particle is determined. The specific implementation method of the preliminary determination result includes the following steps:

[0110] Step 501: for each mineral particle in the Kth frame of mineral particle image, based on its centroid coordinate (x i,k ,y i,k ), calculate the time it takes for each particle to fall freely from the initial position of the free fall motion to the position (x i,k ,y i,k )

[0111] Among them, the above l i,k represents the particle size of the i-th mineral particle in the K-th frame mineral particle image, represents the acceleration of the i-th mineral particle in the K-th frame mineral particle image when it is in free fall, and h is the fixed height of the mineral particle falling from the initial position to the upper edge of the imaging area.

[0112] Step 502: for each mineral particle in the Kth frame of the mineral particle image, calculate the displacement of the mineral particle during the falling time T based on the dynamic model. And determine whether the mineral particle leaves the imaging area. If the mineral particle leaves the imaging area, output the preliminary determination result of the area corresponding to the mineral particle, otherwise proceed to step 503.

[0113] Among them, the above s i,k+1 is the displacement of the i-th mineral particle falling time T in the K-th frame mineral particle image, T is the shooting time interval between the K-th frame mineral particle image and the K+1-th frame mineral particle image, T=1 / F, F is the shooting frame rate of the industrial camera.

[0114] Step 503: for each mineral particle q that has not left the imaging area in step 502, i,k , perform the first-level feature matching, and match the mineral particle set Q that meets the first-level feature matching conditions in the K+1th frame mineral particle map. nl,k+1 .

[0115] In some embodiments of the present invention, the first-level feature matching condition can be expressed by the following formula:

[0116] |x i,k -x r,k+1 |≤ε x

[0117] Among them, ε x It represents the upper limit of the horizontal coordinate offset of a mineral particle in the Kth frame and the K+1th frame mineral particle image.

[0118] Step 504, the mineral particles q i,k In step 503, the mineral particle set Q nl,k+1 The second level feature matching is performed in the ore particle q i,k The particle size l i,k Next, determine whether there is a mineral particle q that meets the second-level feature matching conditions. nl,k+1, if the mineral particle q nl,k+1 If exists, the mineral particle q i,k With q nl,k+1 The maximum area of ​​the mineral particle q j,k The corresponding area is initially determined and the process proceeds to step 506 , otherwise the process proceeds to step 505 .

[0119] In some embodiments of the present invention, the second-level feature matching condition can be expressed by the following formula:

[0120]

[0121] Among them, ε xy It represents the upper limit of the deviation between the position predicted by the dynamic model and the actual position after a mineral particle falls for a period of time T.

[0122] Step 505, the mineral particles q i,k In step 503, the mineral particle set Q nl,k+1 The third level feature matching is performed in the i,k Low particle size (l i,k -1), determine whether there is a mineral particle q that meets the second-level feature matching conditions bl,k+1 , at the same time, compared to q i,k High particle size (l i,k +1), determine whether there is a mineral particle q that meets the second-level feature matching conditions al,k+1 From the mineral particles q bl,k+1 and q al,k+1 In the process, the mineral particles that meet the third-level feature matching conditions are selected and matched with the mineral particles q i,k The maximum area of ​​the mineral particle q i,k The area is preliminarily determined.

[0123] In some embodiments of the present invention, the third-level feature matching condition can be expressed by the following formula:

[0124] min(ω1ΔS1+ω2ΔA1,ω1ΔS2+ω2ΔA2)

[0125] Among them, ΔS1 represents the ore particle q bl,k+1 With mineral particles i,k The Euclidean distance between the two particles, ΔA1 indicates the ore particle q bl,k+1 With mineral particles i,k The area difference between the two, ΔS2 represents the ... al,k+1 With mineral particles i,k The Euclidean distance between the two particles, ΔA2 indicates the ore particle q al,k+1 With mineral particles i,k The area difference between them, ω1 and ω2 both represent weighting coefficients.

[0126] Step 506, repeating steps 503 to 505 until all the mineral particles that have not left the imaging area in step 502 are matched.

[0127] The counting accuracy of the intelligent method for counting mineral particles based on multi-level image feature matching of the present invention is exemplified below in combination with specific experimental data.

[0128] In the experiment, Table 2, Table 3 and Table 4 show the counting results and relative errors of mineral particles of different experimental samples using different methods.

[0129]

[0130] Table 2

[0131]

[0132] Table 3

[0133]

[0134] Table 4

[0135] Figures 2 to 7 The counting results and relative errors of mineral particles of different experimental samples by different methods are shown. By comparing the method proposed in the present invention with the counting method based on virtual coils, it can be seen that when counting mineral particles of different particle sizes in each experimental sample, the counting result of the method proposed in the present invention is closer to the actual number of mineral particles, and the relative error is generally lower. At the same time, the applicability and robustness of the method proposed in the present invention are better than the counting method based on virtual coils. When the camera frame rate is unstable or the mineral particles do not fall into the virtual coil, the method proposed in the present invention can still accurately track and count each mineral particle; by comparing the method proposed in this article with the counting method based on detection lines, it can be seen that the method proposed in the present invention fully considers the spin problem of mineral particles in the falling process, and performs multi-visual dimension tracking on each mineral particle in free fall, fully extracts the most real particle size of the mineral particles, and has higher counting accuracy and smaller relative error.

[0136] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for intelligent counting of sintered ore particles based on multi-level feature matching of images, characterized in that: include: Obtain video data of the free-falling motion of ore particles after being crushed by the crusher; For the adjacent K-th frame mineral particle image and K+1-th frame mineral particle image in the video data, the morphological features of each mineral particle in the K-th frame mineral particle image and the K+1-th frame mineral particle image are extracted, and based on the dynamic model and the extracted morphological features, the morphological features of each mineral particle in the K-th frame mineral particle image in the K+1-th frame mineral particle image are tracked through multi-level feature matching to obtain a preliminary determination result of the area corresponding to each mineral particle in the video data; K=1,2,…,m-1, m represents the total number of frames of mineral particle images in the video data; For each mineral particle in the video data, determine the maximum area of ​​the mineral particle according to the preliminary determination results of all areas corresponding to the mineral particle; According to the preset mineral particle area threshold and the maximum area of ​​each mineral particle, each mineral particle is divided into particle sizes, and the number of mineral particles contained in different particle sizes is counted; The step of extracting the morphological features of each mineral particle in the Kth mineral particle image frame and the K+1th mineral particle image frame includes: Extracting the centroid coordinates and area of ​​each mineral particle in the K-th frame of mineral particle image, and the centroid coordinates and area of ​​each mineral particle in the K+1-th frame of mineral particle image; Based on the dynamic model and the extracted morphological features, the morphological features of each mineral particle in the Kth frame mineral particle image in the K+1th frame mineral particle image are tracked through multi-level feature matching to obtain a preliminary determination result of the area corresponding to each mineral particle in the video data, including: For the i-th mineral particle q in the K-th mineral particle image i,k , based on the kinetic model, calculate the i-th mineral particle q i,k Freely fall from the initial position of the free fall motion to the position (x i,k ,y i,k ) at a speed v i,k , and calculate the i-th mineral particle q i,k Displacement s during falling time T i,k+1 ; T represents the time interval between taking the Kth frame of mineral particle image and the K+1th frame of mineral particle image, (x i,k ,y i,k ) represents the i-th mineral particle q in the K-th mineral particle image i,k The coordinates of the center of mass, x i,k represents the i-th mineral particle q in the K-th mineral particle image i,k The horizontal axis, y i,k represents the i-th mineral particle q in the K-th mineral particle image i,k , i=1, 2, ..., I, where I represents the total number of mineral particles in the K-th frame of mineral particle image; According to the initial position and the displacement s i,k+1 , determine the i-th mineral particle q i,k Whether to leave the imaging area of ​​the image acquisition device used to acquire the video data; If the i-th mineral particle q i,k Without leaving the imaging area, a set of mineral particles Q that meet the first-level feature matching conditions are screened out from the K+1th frame of mineral particle image. nl,k+1 ; In the i-th mineral particle q i,k The particle size l i,k Next, determine the mineral particle set Q nl,k+1 Is there a mineral particle q that meets the second-level feature matching conditions? nl,k+1 ; If the mineral particle set Q nl,k+1 There are mineral particles q nl,k+1 , then the i-th mineral particle q i,k The area and the mineral particle q nl,k+1 The larger of the areas is taken as the mineral particle q i,k The corresponding area is initially determined; If the mineral particle set Q nl,k+1 There is no mineral particle q nl,k+1 , then it is determined that under the first particle size, the mineral particle set Q nl,k+1 Is there a mineral particle q that meets the second-level feature matching condition? bl,k+1 , and at the second particle size, the mineral particle set Q nl,k+1 Is there a mineral particle q that meets the second-level feature matching condition? al,k+1 The first particle size, the particle size l i,k and the second particle size is one of three particle sizes arranged in sequence in a preset particle size type table, wherein the preset particle size type table records a plurality of particle sizes arranged from small to large; and the first particle size is smaller than the particle size l i,k , the particle size l i,k smaller than the second particle size; If the mineral particle set Q nl,k+1 There are mineral particles q bl,k+1 and mineral particles al,k+1 , then from the mineral particle q bl,k+1 and mineral particles al,k+1 The mineral particles that meet the third-level feature matching conditions are screened, and the area of ​​the mineral particles that meet the third-level feature matching conditions and the mineral particles q i,k The larger of the areas of i,k The corresponding area is preliminarily determined.

2. The method according to claim 1, characterized in that In the case of the initial position and the displacement s i,k+1 , determine the i-th mineral particle q i,k After leaving the imaging area of ​​the image acquisition device used to acquire the video data, the method further includes: If the i-th mineral particle q i,k Leave the imaging area, then the i-th mineral particle q i,k The area of ​​the mineral particle q i,k The corresponding area is preliminarily determined.

3. The method according to claim 1, characterized in that The first-level feature matching conditions are: |x i,k -x r,k+1 |≤ε x Among them, x r,k+1 represents the horizontal coordinate of the rth mineral particle in the K+1th mineral particle image, ε x Represents the mineral particle q i,k An upper limit value of the horizontal coordinate offset between the Kth frame of mineral particle image and the K+1th frame of mineral particle image; A set of mineral particles Q meeting the first-level feature matching condition is selected from the K+1th frame of mineral particle image. nl,k+1 ,include: The set of all mineral particles in the K+1th frame of mineral particle image that meet the first-level feature matching condition is taken as the mineral particle set Q nl,k+1 .

4. The method according to claim 1, characterized in that: The second-level feature matching condition is: Among them, x r,k+1 represents the horizontal coordinate of the rth mineral particle in the K+1th frame mineral particle image, y r,k+1 represents the ordinate of the rth mineral particle in the K+1th mineral particle image, ε xy Represents the mineral particle q i,k After the falling time T, the upper limit of the deviation between the position predicted by the dynamic model and the actual position.

5. The method according to claim 1, characterized in that The third-level feature matching condition is: min(ω1ΔS1+ω2ΔA1,ω1ΔS2+ω2ΔA2) Among them, ΔS1 represents the ore particle q bl,k+1 With mineral particles i,k The Euclidean distance between the two particles, ΔA1 indicates the ore particle q bl,k+1 With mineral particles i,k The area difference between the two, ΔS2 represents the ... al,k+1 With mineral particles i,k The Euclidean distance between the two particles, ΔA2 indicates the ore particle q al,k+1 With mineral particles i,k The area difference between them, ω1 and ω2 both represent weighting coefficients.

6. The method according to claim 1, characterized in that The calculation of the i-th mineral particle q i,k Freely fall from the initial position of the free fall motion to the position (x i,k ,y i,k ) at a speed v i,k , and calculate the i-th mineral particle q i,k Displacement s during falling time T i,k+1 ,include: By formula Calculate the i-th mineral particle q i,k Freely fall from the initial position of the free fall motion to the position (x i,k ,y i,k ) at a speed v i,k ; By formula Calculate the i-th mineral particle q i,k Displacement s during falling time T i,k+1 ; in, Represents the mineral particle q i,k The acceleration of the free fall motion, h represents the ore particle q i,k The device falls from an initial position to a fixed height at the upper edge of the imaging area.

7. The method according to claim 1, characterized in that The method of determining the maximum area of ​​the mineral particle according to the preliminary determination results of all areas corresponding to the mineral particle includes: The maximum value among all the preliminary determination results of the areas corresponding to the mineral particle is taken as the maximum area of ​​the mineral particle.

8. The method according to claim 1, characterized in that The step of extracting the morphological features of each mineral particle in the Kth mineral particle image frame and the K+1th mineral particle image frame includes: Converting the Kth frame mineral particle image and the K+1th frame mineral particle image into grayscale images, and eliminating noise in the grayscale images; Binarization is performed on the grayscale image after noise elimination to obtain the Kth frame of mineral particle image and the K+1th frame of mineral particle image after binary processing; The morphological features of each mineral particle in the Kth frame mineral particle image and the K+1th frame mineral particle image after binarization are extracted.

Images