Processing device, processing method, and program

Abstract

To reduce a processing load when determining a type of a dust particle discharged from facilities, and to increase discrimination precision of the type of the dust particle.SOLUTION: A processing device 100 creates a first binarization image 220 by binarizing a pixel value of a binarization object image by using a first binarization threshold Th0 common to all the pixel values of the binarization object image. The processing device 100 creates a second binarization image 230 by binarizing the pixel value of the binarization object image by using second binarization thresholds Th(i) set for each primary object i included in the first binarization image 220 by individually using each of the second binarization thresholds Th(i). The processing device 100 determines a type of a dust particle included in a polarization image on the basis of a result of processing including the image processing.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present disclosure relates to a processing device, a processing method, and a program, which are particularly suitable for use in determining the type of dust particles emitted from equipment. [Background technology]

[0002] In facilities that process solid raw materials to produce products, some of the solid raw materials are discharged as dust particles. An example of such equipment is a blast furnace. In a blast furnace, various dust particles are discharged from the top of the furnace, including dust particles originating from the charges fed into the furnace from the top, such as iron ore, sintered ore, ironmaking pellets, and coke, as well as dust particles originating from the pulverized coal blown into the tuyeres at the bottom. These dust particles reach the top of the furnace accompanied by exhaust gas from the furnace top. Dust particles contained in the exhaust gas are mainly collected by dust catchers and venturi scrubbers. The former collects relatively coarse dust particles, while the latter collects finer dust particles.

[0003] Dust particles discharged from such equipment often contain a mixture of multiple types of dust particles. Identifying the type of dust particles discharged from the equipment makes it possible to determine the operating status of the equipment. For example, by identifying the type of dust particles discharged from the equipment, it is possible to obtain information such as what type of solid raw material and what proportion of the solid raw material remains untreated. Therefore, the type of dust particles discharged from the equipment serves as an indicator for determining whether the solid raw material is being properly treated in the equipment.

[0004] Patent Document 1 describes a technique for identifying the type of dust particles emitted from equipment. Patent Document 1 discloses the use of two types of color images: a bright-field image taken by irradiating white light onto a sample prepared by polishing a resin with dust embedded in it, and a polarized image taken by irradiating polarized light onto the sample. Patent Document 1 also discloses that particle portions are extracted from the observation target areas of these two types of color images by edge detection processing or the like, multiple feature amounts are derived for the particle portions, and the type of dust in the particle portions is identified based on the multiple feature amounts. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] Patent No. 6822283 Summary of the Invention [Problem to be solved by the invention]

[0006] However, the technology described in Patent Document 1 requires capturing both a bright-field image and a polarized image. Furthermore, the technology described in Patent Document 1 is premised on the correct extraction of the contours of the particle portion. Therefore, if the contours of the particle portion are extracted with low accuracy, the accuracy of deriving the feature quantities of the particle portion may decrease. As a result, the accuracy of identifying the type of dust particles may decrease. As described above, the technology described in Patent Document 1 increases the processing load (including the workload) when identifying the type of dust particles emitted from the equipment, and there is a risk that the accuracy of identifying the type of dust may decrease.

[0007] The present disclosure has been made in consideration of the above-mentioned problems, and aims to achieve both reducing the processing load when identifying the type of dust particles emitted from equipment and improving the accuracy of identifying the type of dust particles. [Means for solving the problem]

[0008] The processing device disclosed herein is a processing device that performs processing including identifying the type of dust particles discharged from equipment that processes solid raw materials, and includes an acquisition unit that acquires a polarized image including the dust particles, an image processing unit that performs processing including image processing on the polarized image, and a particle type identification unit that identifies the type of the dust particles included in the polarized image based on a result of the processing including the image processing by the image processing unit, and the image processing unit has a binarization unit that binarizes pixel values ​​of an image to be binarized to create a binarized image, and the binarization target image is generated by performing processing on the polarized image or the polarized image different from the binarization processing. The binarization unit is the polarized image that has been processed, and the binarization unit has: a first binarization unit that creates a first binarized image by processing including binarizing pixel values ​​of the binarization target image using a first binarization threshold that is a binarization threshold common to all pixels of the binarization target image; and a second binarization unit that creates a second binarized image by processing including binarizing pixel values ​​of the binarization target image using second binarization thresholds that are binarization thresholds set for each primary object, which is an object represented as the dust particle in the first binarized image, by individually using each of the second binarization thresholds.

[0009] The processing method of the present disclosure is a processing method for performing processing including identifying the type of dust particles discharged from equipment for processing solid raw materials, and includes an acquisition step of acquiring a polarized image including the dust particles, an image processing step of performing processing including image processing on the polarized image, and a particle type identification step of identifying the type of the dust particles included in the polarized image based on a result of the processing including image processing by the image processing step, wherein the image processing step includes a binarization step of binarizing pixel values ​​of an image to be binarized to create a binarized image, and the binarized image is generated by binarizing the polarized image or a binarized image obtained by binarizing the polarized image. The binarization process includes a first binarization process for creating a first binarized image by processing including binarizing pixel values ​​of the binarization target image using a first binarization threshold that is a common binarization threshold for all pixels of the binarization target image, and a second binarization process for creating a second binarized image by processing including binarizing pixel values ​​of the binarization target image using second binarization thresholds that are binarization thresholds set for each primary object representing the dust particle in the first binarized image, individually using each of the second binarization thresholds.

[0010] The program of the present disclosure causes a computer to function as each part of the processing device. [Effects of the Invention]

[0011] According to the present disclosure, it is possible to reduce the processing load when determining the type of dust particles emitted from equipment and to improve the accuracy of determining the type of dust particles. [Brief explanation of the drawings]

[0012] [Figure 1] FIG. 2 illustrates an example of a functional configuration of a processing device. [Figure 2] FIG. 10 is a diagram showing an example of a processed polarized image (grayscale image, binarized image). [Figure 3] 4A and 4B are diagrams illustrating an example of a first binarization threshold and a second binarization threshold. [Figure 4] 1A and 1B are diagrams showing examples of polarized images of coal, coke, iron ore, and gangue. [Figure 5] FIG. 10 is a diagram showing a specific example of filling holes in a secondary object. [Figure 6] FIG. 10 illustrates an example histogram of hole size ratios for secondary objects. [Figure 7] FIG. 10 illustrates an example of a convex hull of a secondary object. [Figure 8] FIG. 10 is a diagram illustrating an example of a histogram of the ratio of differences between the convex hull of a secondary object. [Figure 9] FIG. 10 is a diagram showing an example of a discrimination result image. [Figure 10] FIG. 10 is a diagram conceptually illustrating an example of changes over time in the feature amount of a dust particle. [Figure 11-1] 10 is a flowchart illustrating an example of a processing method. [Figure 11-2] This is a flowchart following Figure 11-1. [Figure 11-3] This is a flowchart following Figure 11-2. DETAILED DESCRIPTION OF THE INVENTION

[0013] Hereinafter, an embodiment of the present disclosure will be described with reference to the drawings. In addition, the comparison of length, position, size, spacing, etc. being the same includes not only cases where the objects are exactly the same, but also cases where they are different within the scope of the gist of this disclosure (for example, cases where they are different within the tolerance range determined at the time of design).

[0014] <Processing device 100> FIG. 1 is a diagram illustrating an example of the functional configuration of a processing device 100. As shown in FIG. The processing device 100 has, as hardware, one or more hardware processors such as a CPU (Central Processing Unit), and one or more memories such as a RAM (Random Access Memory), a ROM (Read Only Memory), etc. The processing device 100 performs various operations by, for example, executing one or more programs stored in the memory using one or more hardware processors.

[0015] Furthermore, an input device 110 and an output device 120 are connected to the processing device 100 so as to be able to communicate with the processing device 100. Communication between the processing device 100 and the input device 110 and the output device 120 may be wired communication or wireless communication. Furthermore, the processing device 100 may be equipped with the input device 110 and the output device 120.

[0016] Furthermore, the processing device 100 may be realized by dedicated hardware such as an ASIC (Application Specific Integrated Circuit).

[0017] The processing device 100 performs processing that includes identifying the type of dust particles discharged from equipment that processes solid raw materials. In this embodiment, the processing device 100 derives dust particle feature values ​​for each type of dust particle. Deriving dust particle feature values ​​in this manner is preferable because more detailed information about the dust particles can be obtained for each identified dust particle type. However, the processing device 100 does not necessarily need to derive dust particle feature values. In this embodiment, the processing device 100 determines the operational status of the equipment based on at least one of the dust particle type and feature values. This is preferable because it allows, for example, determining whether the equipment operation needs to be changed or stopped. However, the processing device 100 does not necessarily need to determine the operational status of the equipment.

[0018] In addition, in this embodiment, a case where the facility for processing solid raw materials is a blast furnace is exemplified. In this embodiment, a case where dust particles to be identified as types are gangue, iron ore, coal, and coke is exemplified. However, the dust particles to be identified as types are not limited to these. For example, the dust particles to be identified as types may be coal and other types of dust particles (in this case, the number of types of dust particles to be identified is two). Furthermore, coke is produced using coal as a raw material. Therefore, coal and coke may be treated as the same type of dust particle. As will be described later, in this embodiment, a case where coal and coke are identified as the same type of dust particle and then the dust particles of the same type are further identified as either coal or coke is exemplified (see the first particle type identification unit 103a and the second particle type identification unit 103b of the particle type identification unit 103). Furthermore, the facility for processing solid raw materials is not limited to a blast furnace and may be, for example, an industrial furnace such as a shaft furnace.

[0019] 1, in this embodiment, a case is illustrated in which the processing apparatus 100 includes an acquisition unit 101, an image processing unit 102, a particle type discrimination unit 103, a particle feature derivation unit 104, an operation state discrimination unit 105, and an output unit 106. As described above, the processing apparatus 100 does not necessarily have to include the particle feature derivation unit 104. Moreover, the processing apparatus 100 does not necessarily have to include the operation state discrimination unit 105. An example of each unit of the processing apparatus 100 of this embodiment will be described below.

[0020] <<Acquisition part 101>> The acquisition unit 101 acquires advance acquisition information, which is information that the processing device 100 needs to acquire in advance in order to perform processing. In this embodiment, a case where the pre-acquired information includes a polarized image containing dust particles (i.e., a polarized image in which dust particles are captured) will be exemplified. In the following description, the polarized image containing dust particles will be referred to as a polarized image as necessary.

[0021] A polarized image is an image obtained by capturing an image of a sample with a camera while irradiated with polarized light. In this embodiment, a case where the polarized image is a color image is illustrated. However, the polarized image captured by the acquisition unit 101 is not limited to a color image. The polarized image captured by the acquisition unit 101 may be, for example, a grayscale image (in this case, the grayscale conversion unit 102a of the image processing unit 102, which will be described later, is unnecessary). A polarized image is obtained, for example, by capturing a micrograph of a sample observed under a polarized microscope with a camera provided in the polarized microscope.

[0022] In this embodiment, an example of preparing a sample for obtaining a polarized image is described below. First, dust particles contained in exhaust gas discharged from the top of a blast furnace are captured by a dust catcher and embedded in resin. Note that the dust catcher captures relatively coarse dust particles with particle sizes of several tens of micrometers or less. Then, the surface of the resin in which the dust particles are embedded is polished. The resin with the polished surface in this manner is used as the sample. In this case, the polished surface of the sample is the observation surface of the polarized microscope. Note that the location where the dust particles are captured is not limited to the dust catcher.

[0023] The method for creating a polarized image and the method for preparing a sample can be realized by known techniques as described in Patent Document 1. Therefore, detailed explanations thereof will be omitted here.

[0024] In this embodiment, the acquisition unit 101 acquires the polarization image obtained as described above from the input device 110. The input device 110 may include, for example, a communication device connected to a communication device connected to the polarizing microscope via a communication cable. In this case, the input device 110 receives, for example, a polarization image transmitted from the communication device connected to the polarizing microscope via the communication cable. In this case, the acquisition unit 101 acquires the polarization image received by the input device 110. Note that the input device 110 and the communication device connected to the polarizing microscope may communicate wirelessly or via a network. The input device 110 may also include a storage medium. In this case, the input device 110 stores the polarization image. In this case, the acquisition unit 101 acquires the polarization image by reading it out from the input device 110.

[0025] In addition, this embodiment illustrates a case where the acquisition unit 101 also acquires pre-acquired information other than the polarization image from the input device 110. The input device 110 may include a user interface. In this case, an operator may input information such as a numerical value indicating the pre-acquired information by operating the user interface. In this case, the acquisition unit 101 acquires the pre-acquired information input in this manner. The input device 110 may also include a communication device. In this case, the input device 110 may receive pre-acquired information other than the polarization image from an external device. In this case, the acquisition unit 101 acquires the pre-acquired information received by the input device 110 in this manner. The input device 110 may also include a storage medium in which pre-acquired information other than the polarization image is stored. In this case, the acquisition unit 101 acquires the pre-acquired information by reading it from the storage medium.

[0026] <<Image processing unit 102>> The image processing unit 102 performs processing, including image processing, on the polarization image. The image processing performed by the image processing unit 102 includes binarization processing. As described above, this embodiment illustrates a case where the polarization image is a color image. Therefore, this embodiment illustrates a case where the image processing unit 102 converts the polarization image into a grayscale image to reduce the processing load. However, the image processing unit 102 does not necessarily need to create a grayscale image. In this case, the image processing unit 102 may perform binarization processing on the polarization image, which is a color image. Note that a color image is an image in which information for each pixel is expressed as 256 gradation values ​​(8-bit values) for each of R (red), G (green), and B (blue). In this case, a grayscale image is an image in which information for each pixel is expressed as 256 gradation values ​​(8-bit values). The method of expressing information for each pixel is not limited to R, G, and B. For example, the information for each pixel may be expressed using R (red), G (green), B (blue), and A (alpha value (transparency)), or may be expressed using H (hue), S (saturation), and V (brightness), or may be expressed using other methods (for example, information expressed in HSV converted to brightness). As described above, the image that is the target of the binarization process by image processing unit 102 (image to be binarized) may be a polarization image (itself), or may be a polarization image that has been subjected to a process other than the binarization process, such as grayscaling, on the polarization image.

[0027] In this embodiment, the image processing unit 102 includes a grayscale conversion unit 102a and a binarization unit 102b. <<<Grayscaler 102a>>> The grayscale conversion unit 102a converts into a grayscale image the polarized image acquired by the acquisition unit 101. FIG.

[0028] <<<Binarization Unit 102b>>> The binarization unit 102b binarizes the pixel values ​​of the image to be binarized to create a binarized image. In this embodiment, the image to be binarized is a polarization image or a grayscale image 210. In this embodiment, the binarization unit 102b includes a first binarization unit 102b1, a second binarization unit 102b2, and a noise reduction unit 102b3.

[0029] <<<<First Binarization Unit 102b1>>>> The first binarization unit 102b1 creates a first binarized image by processing that includes binarizing pixel values ​​of the binarization target image using a first binarization threshold that is a binarization threshold common to all pixels of the binarization target image. In this embodiment, a case where the first binarization threshold is a binarization threshold that is common to all pixels of the grayscale image 210 is illustrated as an example.

[0030] The first binarization threshold may be included in the pre-acquired information. However, in order to improve the accuracy of separating the regions corresponding to dust particles and the regions corresponding to resin in the first binarized image, this embodiment illustrates a case in which the first binarization unit 102b1 derives the first binarization threshold using a physical quantity representing the brightness of each pixel in the grayscale image 210. Furthermore, in this embodiment, the physical quantity representing the brightness of each pixel in the grayscale image 210 used to derive the first binarization threshold may be lightness or the like, but this embodiment illustrates a case in which the physical quantity representing the brightness is luminance. In the following description, dust particles shown in the image will be referred to as objects as necessary. Furthermore, resin (other than dust particles) shown in the image will be referred to as non-objects as necessary. Note that dust particles (objects) shown in the binarized image are referred to as blobs.

[0031] For example, the first binarization unit 102b1 derives a histogram of the brightness values ​​of each pixel in the grayscale image 210, then derives the most frequent value Ga in the histogram, and derives the most frequent value Ga or a value based on the most frequent value Ga as the first binarization threshold Th0.

[0032] As described above, in this embodiment, the dust particles to be identified are assumed to be gangue, iron ore, coal, and coke. The brightness of the objects (gangue, iron ore, coal, and coke) is considered to be brighter than the mode Ga (the representative brightness of non-objects (resin)). Therefore, in this embodiment, the first binarization unit 102b1 derives the first binarization threshold Th0 using the following equation (1): Th0 = Ga + dG (1) dG is a positive integer. In this case, the pre-acquired information includes the value of dG.

[0033] In this embodiment, an example is shown in which the first binarization unit 102b1 creates a binarized image as the first binarized image by setting, for each pixel of the grayscale image 210, the pixel value of pixels having a brightness greater than the first binarization threshold Th0 to 1 and the pixel value of pixels having a brightness equal to or less than the first binarization threshold Th0 to 0. In this case, in this embodiment, areas with a pixel value of 1 indicate areas corresponding to objects (gangue, iron ore, coal, and coke), and areas with a pixel value of 0 indicate areas corresponding to non-objects (resin).

[0034] Fig. 2(b) is a diagram showing an example of a first binarized image 220. Fig. 2(b) illustrates an example in which objects are represented in white and non-objects in black (this type of representation is the same for other binarized images). Note that the first binarization unit 102b1 may derive a first binarization threshold value using a method such as Otsu's binarization to create the first binarized image.

[0035] <<<<<Second Binarization Unit 102b2>>>> The second binarization unit 102b2 creates a second binarized image by processing that includes binarizing pixel values ​​of the image to be binarized (grayscale image 210) individually using second binarization thresholds, which are binarization thresholds set for each primary object, which is an object represented as a dust particle in the first binarized image 220. In order to avoid confusion between objects in the second binarized image and objects in the first binarized image 220, in the following description, objects in the first binarized image 220 and objects in the second binarized image will be referred to as primary objects and secondary objects, respectively, as necessary.

[0036] Here, the reason for creating the second binarized image will be explained. In a polarized image, brightness changes continuously near the boundary between objects and non-objects. Furthermore, there are multiple types of dust particles to be identified, and the brightness of dust particles in a polarized image varies depending on the type of dust particle. Therefore, if a common first binarization threshold Th0 is used for binarization of all pixels in a polarized image, the accuracy of separating primary objects (gangue, iron ore, coal, and coke) from non-objects (resin) in the first binarized image 220 may not be sufficiently high. For example, if the first binarization threshold Th0 is close to the brightness of relatively dark objects among the primary objects, relatively bright primary objects may appear larger than their actual size in the first binarized image 220. Conversely, if the first binarization threshold Th0 is close to the brightness of relatively bright objects among the primary objects, relatively dark primary objects may appear smaller than their actual size in the first binarized image 220.

[0037] Therefore, in this embodiment, a case will be illustrated in which the second binarization unit 102b2 binarizes the pixel values ​​of the grayscale image 210 by individually using binarization thresholds (second binarization thresholds) set for each primary object. A specific example of such processing by the second binarization unit 102b2 will be described below.

[0038] First, the second binarization unit 102b2 selects one primary object in the first binarized image 220. The second binarization unit 102b2 binarizes the pixel values ​​of the grayscale image 210 using a second binarization threshold for the primary object. The second binarization unit 102b2 then derives the object (dust particle) in the binarized image that is closest to the primary object as the object (secondary object) in the second binarized image. The object closest to the primary object may be, for example, an object that includes a pixel at a representative position (e.g., the centroid) of the primary object. The second binarization unit 102b2 derives the object (secondary object) in the second binarized image individually for each primary object included in the first binarized image 220. The second binarization unit 102b2 also defines regions (pixels) not derived as secondary objects as non-object regions in the second binarized image.

[0039] The second binarization threshold for each primary object is set, for example, using a physical quantity representing the brightness of pixels in a region of the binarization target image that corresponds to the primary object. In this embodiment, in order to more accurately extract the boundary between objects and non-objects, the second binarization unit 102b2 derives the second binarization threshold for the primary object based on a physical quantity representing the brightness of a region of the binarization target image that corresponds to the primary object and a physical quantity representing the brightness of a region of the binarization target image that corresponds to a non-object (a region that does not correspond to a primary object). In the following description, a region of the binarization target image that corresponds to a primary object will be referred to as an image region corresponding to the primary object, as necessary. Furthermore, a region of the binarization target image that does not correspond to a primary object will be referred to as an image region that does not correspond to a primary object, as necessary.

[0040] Here, the physical quantity representing the brightness of an image region corresponding to a primary object and the physical quantity representing the brightness of an image region not corresponding to a primary object may each be represented by a basic statistical quantity. While the physical quantity representing the brightness may be lightness or the like, this embodiment illustrates a case in which the physical quantity representing the brightness is luminance. Furthermore, this embodiment illustrates a case in which the basic statistical quantity of the physical quantity (luminance) representing the brightness of an image region corresponding to a primary object is an arithmetic mean value. In the following description, this arithmetic mean value will be referred to as the "primary object luminance mean value" as needed. Furthermore, this embodiment illustrates a case in which the basic statistical quantity of the physical quantity (luminance) representing the brightness of an image region not corresponding to a primary object is a mode. In the following description, this mode will be referred to as the "non-object luminance mode" as needed.

[0041] Let i be a variable that identifies a primary object included in the first binarized image 220. In this embodiment, a case will be illustrated in which the second binarization unit 102b2 derives the second binarization threshold Th(i) for the primary object i using the following equation (2). Th(i)={Ga+Gm(i)}÷2 (2) Here, Ga is the most frequent luminance value within a non-object, and is the same as Ga in the first term on the right-hand side of equation (1). Gm(i) is the average luminance value within a primary object for primary object i. The average luminance value within a primary object Gm(i) is derived for each primary object. The average luminance value within a primary object Gm(i) will be a different value for each primary object (however, the average luminance values ​​within multiple primary objects may be the same value).

[0042] Fig. 3 is a diagram illustrating an example of the first binarization threshold Th0 and the second binarization threshold Th(i). Fig. 3(a) is a diagram illustrating one object 311 extracted from the grayscale image 210. Fig. 3(b) is a diagram illustrating the luminance 321 of a pixel that is located at a position overlapping a line 312 among the pixels that make up the grayscale image 210 shown in Fig. 3(a).

[0043] When the grayscale image 210 is binarized using the first binarization threshold Th0, the object 311 and non-object are separated at a pixel near the start (or end) of the brightness change. This is susceptible to the influence of noise, etc. Therefore, at the boundary between the object 311 and the non-object, even though they are actually close to each other, the pixels at which the object 311 and non-object are separated may be far apart. Therefore, in the binarized image, the boundary between the object 311 and non-object may become unclear. In contrast, when the grayscale image 210 is binarized using the second binarization threshold Th(i), the object 311 and non-object are separated at a pixel in the middle of the brightness change. Therefore, it is possible to prevent the boundary between the object 311 and non-object from becoming unclear in the binarized image.

[0044] As described above, in this embodiment, the second binarization unit 102b2 first selects one primary object i. Then, for each pixel in the grayscale image 210, the second binarization unit 102b2 sets the pixel value of pixels having a luminance greater than the second binarization threshold Th(i) for the primary object i to 1, and sets the pixel value of pixels having a luminance equal to or less than the second binarization threshold Th(i) to 0. Then, the second binarization unit 102b2 derives, from among the objects (dust particles) in the binarized image thus created, the object closest to the primary object i as the object (secondary object) in the second binarized image. The second binarization unit 102b2 derives such a secondary object individually for each primary object i. Then, the second binarization unit 102b2 creates a binarized image as a second binarized image in which the pixel values ​​of the areas (pixels) derived as secondary objects are set to 1 and the pixel values ​​of the areas not derived as secondary objects are set to 0.

[0045] FIG. 2( c ) is a diagram showing an example of the second binarized image 230 . 2(b), dust particles that appear relatively bright (high brightness) and dust particles that appear relatively dark (low brightness) are extracted as objects (primary objects) in the first binarized image 220. However, in the example shown in FIG. 2(b), the primary objects corresponding to the dust particles that appear relatively bright (high brightness) are wide, so there is a risk that multiple objects in the grayscale image 210 will be extracted as a single primary object (see, for example, primary object 221).

[0046] In contrast, as shown in FIG. 2(c), in the second binarized image 230, it is possible to prevent multiple objects in the grayscale image 210 from becoming one secondary object (see, for example, secondary objects 231 and 232).

[0047] <<<<Noise reduction unit 102b3>>>> The noise reduction unit 102b3 performs processing that includes reducing noise contained in the second binarized image 230. Secondary objects that are clearly different in size from the expected size of the dust particle whose type is being determined may be considered not to be objects representing dust particles. Therefore, in this embodiment, the noise reduction unit 102b3 considers secondary objects that have a predetermined number of pixels or less to be noise and changes them to non-objects.

[0048] Fig. 2(d) is a diagram showing an example of a noise-reduced second binarized image 240. As shown in Fig. 2(d), in the noise-reduced second binarized image 240, small secondary objects included in the second binarized image 230 (before noise reduction) can be removed.

[0049] The noise reduction unit 102b3 performs labeling on the secondary objects included in the thus-obtained noise-reduced second binarized image 240. The labeling is a process that includes setting identification information for each secondary object.

[0050] The binarization unit 102b may further include a noise reduction unit that reduces noise contained in the first binarized image 220. For example, the noise reduction unit may consider primary objects having a predetermined number of pixels or less among the primary objects as noise and change them to non-objects. In this way, the second binarization unit 102b2 can reduce the number of primary objects from which the second binarization threshold Th(i) is derived.

[0051] It is preferable to reduce noise contained in the binarized images (first binarized image 220 and second binarized image 230) as in the present embodiment. However, this is not necessarily required. For example, the image processing unit 102 may perform noise reduction processing on the grayscale image 210 before binarization processing. Examples of such noise reduction processing include processing using a smoothing filter such as a median filter. If the noise contained in the binarized image is reduced by performing such noise reduction processing, it is not necessary to perform processing to reduce the noise contained in the binarized image (however, even in such cases, processing to reduce the noise contained in the binarized image may be performed). Furthermore, other processing may be performed on the binarized image in addition to or instead of the processing to reduce the noise contained in the binarized image. For example, in addition to noise reduction, at least one of dilation processing, erosion processing, closing processing, and opening processing may be performed to adjust the shape of an object. Furthermore, thinning processing may be performed to clarify the shape of an object.

[0052] <<Particle type discrimination unit 103>> The particle type discrimination unit 103 discriminates the type of dust particles contained in the polarized image based on the results of image processing by the image processing unit 102. For example, the particle type discrimination unit 103 may discriminate the type of dust particles based on the area or shape of at least one of secondary objects and non-objects in the noise-reduced second binarized image 240. However, a binarized image does not contain information indicating the brightness of the image. Therefore, it is not easy to discriminate the type of dust particles with high accuracy from the binarized image alone. Note that if no processing is performed on the second binarized image 230, such as processing to reduce noise contained in the binarized image, the particle type discrimination unit 103 may use the second binarized image 230 instead of the noise-reduced second binarized image 240, for example.

[0053] Therefore, in this embodiment, an example will be given in which the particle type discrimination unit 103 discriminates the type of dust particle based on feature amounts of a region corresponding to a secondary object in the second binarized image 230 or a binarized image obtained by processing the second binarized image 230, among the regions of the binarization target image, the second binarized image 230, or a binarized image obtained by processing the second binarized image 230. Note that the processing of the second binarized image 230 includes, for example, the noise reduction process described above. In the following description, a region corresponding to a secondary object in the binarized image created by the binarization unit 102b among the regions of the binarization target image, the second binarized image 230, or a binarized image obtained by processing the second binarized image 230 will be referred to as an image region corresponding to a secondary object, as necessary.

[0054] When a binarization target image is used to derive the feature amount of an image region corresponding to a secondary object, the binarization target image may be a polarized image (color image) or a grayscale image 210.

[0055] The feature quantity of an image region corresponding to a secondary object may be any feature quantity derived using information obtained from the image (pixels) of the image region. In this embodiment, however, the feature quantity of an image region corresponding to a secondary object includes at least one of a physical quantity representing brightness and a shape. The physical quantity representing brightness may be lightness or the like. In this embodiment, however, the physical quantity representing brightness is luminance. Furthermore, for example, when the feature quantity of an image region corresponding to a secondary object is a physical quantity representing brightness, it is preferable to express the feature quantity of the image region corresponding to the secondary object as a basic statistical quantity. Therefore, in this embodiment, the particle type discrimination unit 103 discriminates the type of dust particle based on the basic statistical quantity of brightness of the image region corresponding to the secondary object.

[0056] As described above, in this embodiment, the dust particles to be identified are gangue, iron ore, coal, and coke. FIG. 4 shows examples of polarized images of coal ( FIG. 4(a)), coke ( FIG. 4(b)), iron ore ( FIG. 4(c)), and gangue ( FIG. 4(d)). For ease of illustration, each polarized image in FIG. 4 is grayscaled. While this is not entirely clear in FIG. 4, the present inventors have discovered that, by adjusting the imaging conditions of a polarized microscope, it is possible to determine the order of brightness (intensity) of the polarized images of coal and coke ( FIG. 4(a) and FIG. 4(b)), iron ore ( FIG. 4(c)), and gangue ( FIG. 4(c)). In the example shown in FIG. 4, the order of brightness (intensity) is as follows: iron ore ( FIG. 4(c)), coal and coke ( FIG. 4(a) and FIG. 4(b)), and gangue ( FIG. 4(c)).

[0057] On the other hand, coke is produced using coal as a raw material. Therefore, in a polarized image, the brightness distribution of coal (FIG. 4(a)) and coke (FIG. 4(b)) is similar. Specifically, in a polarized image, there is a large color difference between coal and coke, such as yellow areas being observed, but the brightness values ​​themselves do not change significantly. In this case, the area (number of pixels) can be used as a feature of the image area corresponding to the secondary object (coal and coke). However, the coal and coke represented as secondary objects also include objects with approximately the same area (number of pixels).

[0058] Coal contains a large amount of volatile components, which causes it to foam in a blast furnace. Therefore, coal that has turned into dust particles (pulverized coal) has large holes inside or an extremely hollowed-out shape (for example, a crescent-shaped cross section). On the other hand, coke is produced by carbonizing coal. Therefore, coke contains fewer volatile components. Therefore, coke foams less in a blast furnace. For this reason, the shape of coke (dust particles) is not distorted compared to coal.

[0059] For the above reasons, in this embodiment, the particle type discrimination unit 103 first determines whether each image region corresponding to a secondary object is gangue, iron ore, coke, or coal based on basic statistics of a physical quantity (in this embodiment, luminance) that represents the brightness of the image region corresponding to the secondary object, and then determines whether each image region corresponding to the secondary object representing the coke and coal is coke or coal based on the shape of the image region corresponding to the secondary object representing the coke and coal. For this reason, this embodiment illustrates a case in which the particle type discrimination unit 103 has a first particle type discrimination unit 103a that performs the former discrimination and a second particle type discrimination unit 103b that performs the latter discrimination. In the following description, when coke and coal are treated as the same type (i.e., when coke and coal are not distinguished), they are referred to as coke / coal.

[0060] <<<First particle type discrimination unit 103a>>> The first particle type discrimination unit 103a discriminates the type of dust particle based on a physical quantity representing the brightness of an image region corresponding to a secondary object. As described above, in this embodiment, the first particle type discrimination unit 103a discriminates whether a dust particle included in an image to be binarized is gangue, iron ore, or coke / coal based on basic statistics of the brightness of an image region corresponding to a secondary object.

[0061] Furthermore, in this embodiment, a case is illustrated in which the basic statistics of the luminance of an image region corresponding to a secondary object include the arithmetic mean value of the luminance of each of RGB and the standard deviation of the luminance of each of RGB. However, the basic statistics of the luminance of an image region corresponding to a secondary object are not limited to these. For example, the basic statistics of the luminance of an image region corresponding to a secondary object may include only one of the arithmetic mean value of the luminance of each of RGB and the standard deviation of the luminance of each of RGB. Furthermore, the basic statistics of the luminance of an image region corresponding to a secondary object may include a median in addition to or instead of the arithmetic mean value. Furthermore, the basic statistics of the luminance of an image region corresponding to a secondary object may include a variance in addition to or instead of the standard deviation.

[0062] Although a rule-based classification method may be used to identify the type of dust particles as described above, the present embodiment illustrates a case in which the first particle type identification unit 103a identifies the type of dust particles corresponding to the luminance of an image region corresponding to a secondary object using a learning model that has learned the relationship between the feature amounts of the image region corresponding to the secondary object and the type of dust particles. Furthermore, in order to reduce the amount of calculation, the present embodiment illustrates a case in which the first particle type identification unit 103a uses a learning model including a classifier that performs clustering by machine learning. Furthermore, in order to reduce the burden of creating training data due to the creation of correct labels, the present embodiment illustrates a case in which the first particle type identification unit 103a uses a learning model that performs unsupervised learning. Specifically, the present embodiment illustrates a case in which the first particle type identification unit 103a performs clustering using the k-means method. However, the clustering method is not limited to the k-means method. For example, the first particle type identification unit 103a may perform clustering using a Gaussian mixture model. The first particle type determination unit 103a may also determine the type of dust particle using a convolutional neural network or the like.

[0063] In the k-means method, data (secondary objects in this embodiment) can be clustered into multiple clusters. However, in the k-means method, it is not easy to identify the nature of each cluster. However, as described above, it is possible to determine in advance the order of brightness (luminance) of coke / coal, iron ore, and gangue in a polarization image. Therefore, for example, the cluster with the highest brightness, the cluster with the middle brightness, and the cluster with the lowest brightness can be set as the iron ore cluster, the coke / coal cluster, and the gangue cluster, respectively. By setting the clusters in advance in this manner, the k-means method can learn the relationship between the feature amounts of image regions corresponding to secondary objects and the types of dust particles in an unsupervised manner. Note that the k-means method itself can be implemented using known technology. Therefore, a detailed description thereof will be omitted here.

[0064] <<<Second particle type discrimination unit 103b>>> The second particle type discrimination unit 103b classifies the types of dust particles determined to be of the same type based on the shape of the image region corresponding to the secondary object. The particle type discrimination unit 103 may include at least one of the first particle type discrimination unit 103a and the second particle type discrimination unit 103b. However, in the present embodiment, when dust particles whose types are difficult to discriminate without evaluating both the physical quantity representing brightness and the shape are the dust particles to be discriminated, the particle type discrimination unit 103 preferably includes both the first particle type discrimination unit 103a and the second particle type discrimination unit 103b. Therefore, as described above, this embodiment illustrates a case in which the second particle type discrimination unit 103b discriminates the types of dust particles determined to be of the same type by the first particle type discrimination unit 103a. Specifically, in this embodiment, an example is given in which the second particle type discrimination unit 103b discriminates whether a dust particle is coke or coal based on the shape of an image area corresponding to a secondary object representing the dust particle that has been discriminated to be coke / coal by the first particle type discrimination unit 103a.

[0065] In the present embodiment, an example is shown in which the second particle type discrimination unit 103b uses the noise-reduced second binarized image 240 to determine whether an image region corresponding to a secondary object is an image region of coke or coal. Note that instead of using the noise-reduced second binarized image 240 itself, for example, a mask image obtained by masking the grayscale image 210 with the noise-reduced second binarized image 240 may be used. Masking the grayscale image 210 with the noise-reduced second binarized image 240 refers to, for example, setting the pixel values ​​of regions of the grayscale image 210 that do not correspond to objects in the noise-reduced second binarized image 240 to 0 and leaving the pixel values ​​of other regions unchanged from the values ​​in the grayscale image 210.

[0066] As described above, among coke and coal, dust particles having large holes formed therein can be determined to be coal (pulverized coal). Therefore, in this embodiment, a case is illustrated in which the second particle type determination unit 103b determines, as an image region of coal, an image region corresponding to a secondary object whose internal hole size is evaluated as being larger than a predetermined condition based on an evaluation index for evaluating the internal hole size of the secondary object. This determination may be performed only on secondary objects corresponding to dust particles determined to be coke / coal, or may be performed on all secondary objects.

[0067] For example, the second particle type discrimination unit 103b may perform hole filling on secondary objects included in the second binarized image 240 after noise reduction, and calculate the ratio of the number of filled pixels to the number of pixels of the secondary object after hole filling for each secondary object labeled by the noise reduction unit 102b3. In the following description, this ratio will be referred to as the ratio of hole size in the secondary object as necessary. Filling holes in a secondary object refers to changing a non-object surrounded by secondary objects inside the secondary object into a secondary object. The number of pixels in the secondary object after hole filling is the number of pixels in the secondary object after changing the non-object into the secondary object in this way. The number of filled pixels is the number of pixels changed from a non-object to a secondary object.

[0068] Specifically, in this embodiment, the case where the second particle type determination unit 103b derives the ratio Rh(j) of the hole sizes of the secondary object by the following equation (3) is illustrated. Rh(j) = H(j) ÷ P(j) (3) Here, j is a variable that identifies a secondary object included in the second binarized image 240 after noise reduction (identification information of the labeled secondary object). H(j) is the number of pixels filled in secondary object j. P(j) is the number of pixels in secondary object j after changing a non-object into a secondary object.

[0069] In this embodiment, an example is given in which the second particle type discrimination unit 103b discriminates, among the regions of the image to be binarized, an image region corresponding to a secondary object j in which the ratio Rh(j) of the hole sizes of the secondary object exceeds a threshold, as an image region of coal.

[0070] Fig. 5 is a diagram showing a specific example of hole filling for secondary objects. Fig. 5(b) shows a second binarized image 520 after noise reduction created using the grayscale image 510 shown in Fig. 5(a). Fig. 5(c) shows a second binarized image 530 after hole filling has been performed on secondary objects included in the second binarized image 520 after noise reduction shown in Fig. 5(b). In the following description, the second binarized image after hole filling has been performed on secondary objects included in the second binarized image after noise reduction will be referred to as the second binarized image after hole filling, as necessary.

[0071] FIG. 6 is a diagram showing a histogram of the hole size ratio Rh(j) of secondary objects derived from the second binarized image 530 after hole filling shown in FIG. 5(c). In the example shown in Figure 6, the hole size ratio Rh(j) of most secondary objects j is 0.2 or less, but there are some secondary objects j whose hole size ratio Rh(j) is extremely large (see secondary object 531 shown in Figure 5(c)). Therefore, as described above, the second particle type discrimination unit 103b can discriminate, among the regions of the image to be binarized, image regions corresponding to secondary objects j whose hole size ratio Rh(j) exceeds a threshold, as image regions of coal.

[0072] Furthermore, as described above, among coke and coal, dust particles with extremely hollowed shapes can be identified as coal (pulverized coal). For example, in the second binarized image 240 after noise reduction shown in FIG. 4(d), a secondary object 241 is considered to represent coal. However, since the hollowed-out region (non-object region) of the secondary object 241 is not surrounded by the secondary object 241, even if the hole filling described above is performed, the hole size ratio Rh(j) of the secondary object does not increase.

[0073] Therefore, in this embodiment, a case will be illustrated in which the second particle type discrimination unit 103b discriminates, as an image region of coal, an image region corresponding to a secondary object whose degree of carving is evaluated to be greater than a predetermined condition based on an evaluation index for evaluating the degree of carving. Note that this discrimination may be performed only on secondary objects corresponding to dust particles discriminated as coke / coal, or may be performed on all secondary objects.

[0074] For example, the second particle type determination unit 103b may derive the convex hull of a secondary object included in the noise-reduced second binarized image 240, and calculate the ratio of the difference between the area (number of pixels) of the convex hull and the area (number of pixels) of the secondary object to the area (number of pixels) of the secondary object for each secondary object labeled by the noise reduction unit 102b3. In the following description, this ratio will be referred to as the ratio of the difference from the convex hull of the secondary object, as necessary.

[0075] Specifically, in this embodiment, the case where the second particle type determination unit 103b derives the ratio Rc(j) of the difference from the convex hull of the secondary object by the following equation (4) is illustrated. Rc(j)={C(j)-Q(j)}÷Q(j) ···(4) Here, j is a variable that identifies a secondary object included in the second binarized image 240 after noise reduction (identification information of the labeled secondary object), and is the same as j in equation (3). C(j) is the number of pixels of secondary object j. Q(j) is the number of pixels of the convex hull of secondary object j.

[0076] In this embodiment, an example is shown in which the second particle type discrimination unit 103b discriminates, among the regions of the image to be binarized, an image region corresponding to a secondary object j where the ratio Rc(j) of the difference from the convex hull of the secondary object exceeds a threshold, as an image region of coal.

[0077] Fig. 7 shows a specific example of a convex hull of a secondary object. Fig. 7(a) is an enlarged view of the secondary object 241 shown in Fig. 2(d). Fig. 7(b) is a view showing the convex hull of the secondary object 241.

[0078] FIG. 8 is a diagram showing a histogram of the ratio Rc(j) of the difference between the convex hull of the secondary object derived from the second binarized image 240 after noise reduction shown in FIG. 2(d). In the example shown in Fig. 8, the ratio Rc(j) of the difference from the convex hull of the secondary object is 1.0 or less for many secondary objects j, but there are some secondary objects j whose ratio Rc(j) of the difference from the convex hull of the secondary object is extremely large (see secondary object 241 shown in Fig. 2(d)). Therefore, as described above, the second particle type discrimination unit 103b can discriminate, from among the regions of the image to be binarized, image regions corresponding to secondary objects j whose ratio Rc(j) of the difference from the convex hull of the secondary object exceeds a threshold, as image regions of coal.

[0079] In this embodiment, a case where dust particles contained in the binarization target image are classified into four types, namely, gangue, iron ore, coke, and coal, is illustrated. The particle type discrimination unit 103 may set information identifying the type determined as described above to the image on which the image processing unit 102 has performed processing including image processing. For example, the particle type discrimination unit 103 may color the pixels in the region of each object contained in the second binarized image 240 after noise reduction in a different color for each type determined as described above. In the following description, an image colored in this manner will be referred to as a discrimination result image as necessary. An image on which the image processing unit 102 has performed processing including image processing is, for example, the second binarized image 240 after noise reduction. However, the image on which the image processing unit 102 has performed processing including image processing is not limited to the second binarized image 240 after noise reduction. For example, the image on which the image processing unit 102 has performed processing including image processing may be the mask image described above. FIG. 9 is a diagram showing a discrimination result image 910 based on the second binarized image 240 after noise reduction.

[0080] <<Particle feature amount derivation unit 104>> The particle feature derivation unit 104 derives dust particle feature values ​​for each type of dust particle identified by the particle type discrimination unit 103 based on the polarized image that has been processed, including image processing, by the image processing unit 102. The dust particle feature values ​​represent, for example, at least one of the number, size, and shape of the dust particles. In this embodiment, the dust particle feature values ​​include at least one of the number fraction, phase fraction, and aspect ratio.

[0081] In this embodiment, a case will be illustrated in which the polarized image that has been subjected to processing including image processing by the image processing unit 102 is the noise-reduced second binarized image 240. However, as described above in the section <<<Second particle type discrimination unit 103b>>>, the image that has been subjected to processing including image processing by the image processing unit 102 is not limited to the noise-reduced second binarized image 240.

[0082] The particle feature derivation unit 104 may derive, for example, the ratio of the total number of secondary objects representing a certain type of dust particle to the total number of secondary objects included in the noise-reduced second binarized image 240 as the number rate of the dust particles of that type. Note that the number rate may be derived by, for example, deriving the number of dust particles included in the noise-reduced second binarized image 240 for each type of dust particle based on the results of counting the number of lattice point centers occupied by secondary objects (inclusions) for each type of dust particle using the point counting method specified in JIS G0555:2020.

[0083] Furthermore, the particle feature derivation unit 104 may derive, as the phase fraction of a type of dust particle, the ratio (area ratio) of the total area (total number of pixels) of secondary objects representing a certain type of dust particle to the total area (total number of pixels) of secondary objects included in the noise-reduced second binarized image 240. Note that the phase fraction may be derived by, for example, deriving the area of ​​dust particles included in the noise-reduced second binarized image 240 for each type of dust particle based on the results of counting the number of lattice point centers occupied by secondary objects (inclusions) for each type of dust particle using the point counting method specified in JIS G0555:2020.

[0084] Furthermore, the particle feature derivation unit 104 may, for example, derive a circumscribing rectangle for each secondary object included in the noise-reduced second binarized image 240, and derive the aspect ratio of the circumscribing rectangle as the aspect ratio of the type of dust particle identified for that secondary object by the particle type identification unit 103. Furthermore, the particle feature derivation unit 104 may derive a representative value (e.g., arithmetic mean, median, mode) for each type of dust particle as the aspect ratio of the dust particle.

[0085] <<Operational state determination unit 105>> The operational state determination unit 105 determines the operational state of the equipment based on at least one of the type of dust particle determined by the particle type determination unit 103 and the feature of the dust particle derived by the particle feature derivation unit 104.

[0086] For example, when the processing apparatus 100 of this embodiment is applied to equipment that can determine that the operating state is not a normal operating state when a certain type of dust particle is contained, the operating state determination unit 105 may determine that the operating state is not a normal operating state when the type of dust particle determined by the particle type determination unit 103 is the specific type. For example, the operating state determination unit 105 may determine that the operating state of a blast furnace is not a normal operating state when at least one of coal and iron ore is contained as the type of dust particle.

[0087] Furthermore, in order to derive the operation state of the equipment more accurately, the operation state determination unit 105 may determine the operation state of the equipment based on the feature amounts of dust particles derived by the particle feature amount derivation unit 104. For example, the operation state determination unit 105 may determine that the operation state of the blast furnace is not a normal operation state when at least one of the number rate of coal, the phase fraction of coal, the number rate of iron ore, and the phase fraction of iron ore exceeds a predetermined threshold value for each.

[0088] The operational state determination unit 105 may determine the operational state of the equipment based on the time change of the feature amount of dust particles derived by the particle feature amount derivation unit 104. Fig. 10 is a diagram conceptually showing an example of the time change of the feature amount C of dust particles. In Fig. 10, C max is the upper limit of the dust particle feature value C. max is the upper limit of the (absolute value) change in the dust particle feature C per unit time (note that in FIG. 10, the unit time is t k+1 -t k (k is 1 to 5). In the graph 1010 showing the change over time of the feature quantity C of the dust particle, the feature quantity C of the dust particle reaches the upper limit value C at the time t6. max In this case, the operation state determination unit 105 may determine that the operation state of the equipment is not a normal operation state. In addition, in the graph 1020 showing the time variation of the characteristic amount C of the dust particle, the characteristic amount C of the dust particle exceeds the upper limit value C at the time t6. maxAlthough the change ΔC per unit time does not exceed the upper limit value ΔC max exceeding the upper limit value. In such a case, the operational state determination unit 105 may determine that the operational state of the equipment is not in a normal operational state. In addition to the upper limit value, a control value that is smaller than the upper limit value may be set. In this case, the operational state determination unit 105 may determine that the equipment is showing signs of not being in a normal operational state when, for example, the feature amount of dust particles or the amount of change per unit time of the feature amount of dust particles exceeds the control value and is smaller than the upper limit value.

[0089] In recent years, special operating methods, such as the injection of hydrogen gas and coke oven gas (COG), have been explored for blast furnaces. The purpose of these operating methods is to reduce CO2 emissions and carbon consumption from the blast furnace. The injection of hydrogen gas and COG is primarily achieved by replacing the carbon-containing coke or pulverized coal introduced into the furnace with these gases. This means that the amount of air and oxygen introduced into the blast furnace tuyeres is partially replaced with hydrogen or COG. Under these circumstances, the amount of carbon and oxygen introduced into the furnace decreases. This reduces the temperature of the bosh gas (tuyere combustion temperature) supplied from the blast furnace tuyeres, potentially resulting in unburned pulverized coal and coke. Unburned pulverized coal, in particular, is not consumed in the furnace but is discharged as dust particles from the furnace top, preventing it from being consumed as a heat source or reducing agent. Therefore, early detection and correction of the generation and increase of unburned pulverized coal is highly necessary in order to stably maintain the above-mentioned operation mode or to enjoy a sufficient carbon reduction effect. As described above, the application of the treatment device 100 of this embodiment is not limited to blast furnaces, and the operation of blast furnaces is not limited to the above-mentioned operation in which hydrogen gas or COG (coke oven gas) is injected. However, from the above-mentioned perspective, a blast furnace in which hydrogen gas or COG (coke oven gas) is injected is an example of a facility suitable for application of the treatment device 100 of this embodiment.

[0090] <<Output unit 106>> The output unit 106 outputs at least one of information indicating the type of dust particle identified by the particle type identification unit 103, information indicating the feature amount of the dust particle derived by the particle feature amount derivation unit 104, and information indicating the operational state of the equipment identified by the operational state identification unit 105. In the following description, the information output by the output unit 106 will be referred to as processing result information.

[0091] In this embodiment, a case will be exemplified in which the output unit 106 outputs the processing result information to the output device 120. The output device 120 may include, for example, a computer display. In this case, the output device 120 displays the processing result information. The output device 120 may also include a communication device. In this case, the output device 120 transmits the processing result information to an external device. The output device 120 may also include a storage medium. In this case, for example, the processing result information stored in the storage medium included in the output device 120 may be used by the processing device 100 and the external device.

[0092] <Flowchart> Next, an example of a processing method performed using the processing device 100 of this embodiment will be described with reference to the flowcharts shown in Figures 11-1 to 11-3. The flowcharts shown in Figures 11-1 to 11-3 are executed, for example, by the processor included in the processing device 100 reading and developing a computer program stored in a memory.

[0093] In step S1101 of FIG. 11A, the acquisition unit 101 acquires pre-acquired information including a polarization image. Next, in step S1102, the grayscale conversion unit 102a converts the polarization image acquired by the acquisition unit 101 into a grayscale image 210.

[0094] Next, in step S1103, the first binarization unit 102b1 derives a histogram of the luminance values ​​of each pixel in the grayscale image 210, and derives the mode Ga in the histogram. Then, the first binarization unit 102b derives a first binarization threshold Th0 using equation (1).

[0095] Next, in step S1104, the first binarization unit 102b1 creates a binarized image 220 in which, for each pixel of the grayscale image 210, the pixel value of pixels with a brightness greater than the first binarization threshold Th0 is set to 1, and the pixel value of pixels with a brightness equal to or less than the first binarization threshold Th0 is set to 0.

[0096] Next, in step S1105, the second binarization unit 102b2 derives an average luminance value Gm(i) within a primary object. The average luminance value within a primary object is the luminance of an image region corresponding to a primary object. The primary object is a dust particle (object) represented in the first binarized image 220. Then, the second binarization unit 102b2 derives a second binarization threshold Th(i) for primary object i using equation (2) for each primary object.

[0097] Next, in step S1106, the second binarization unit 102b2 creates a binarized image in which, for each pixel of the first binarized image 220, the pixel value of pixels having a brightness greater than a second binarization threshold Th(i) for primary object i is set to 1 and the pixel value of pixels having a brightness equal to or less than the second binarization threshold Th(i) is set to 0, and derives, for each primary object i, the object (dust particle) in the binarized image that is closest to the primary object i as the object (secondary object) in the second binarized image.The second binarization unit 102b2 then creates, as the second binarized image 230, a binarized image in which the pixel value of regions (pixels) derived as secondary objects is set to 1 and the pixel value of regions not derived as secondary objects is set to 0.

[0098] Next, in step S1107, the noise reduction unit 102b3 performs processing including reducing noise included in the second binarized image 230, thereby creating a noise-reduced second binarized image 240. The noise reduction unit 102b3 also performs labeling on secondary objects included in the noise-reduced second binarized image 240.

[0099] Next, in step S1108, the first particle type discrimination unit 103a calculates the arithmetic mean value of the luminance of each of RGB and the standard deviation of the luminance of each of RGB as basic statistics of the luminance of the image region corresponding to the secondary object. The image region corresponding to the secondary object is a region of the image to be binarized or the binarized image created by the binarization unit 102b that corresponds to the secondary object in the binarized image created by the binarization unit 102b.

[0100] Next, in step S1109, the first particle type discrimination unit 103a clusters the secondary object into one of three clusters using the k-means method based on basic statistics of the brightness of the image region corresponding to the secondary object, thereby discriminating whether the image region corresponding to the secondary object is gangue, iron ore, or coke / coal.

[0101] Next, in step S1110 of Fig. 11-2, the second particle type discrimination unit 103b selects one secondary object j included in the noise-reduced second binarized image 240. Then, the second particle type discrimination unit 103b fills holes in the selected secondary object j and derives the hole size ratio Rh(j) of the secondary object using equation (3).

[0102] Next, in step S1111, the second particle type determination unit 103b determines whether the hole size ratio Rh(j) of the secondary object exceeds the threshold value. If the result of this determination is that the hole size ratio Rh(j) of the secondary object exceeds the threshold value (YES in step S1111), the processing of step S1112 is performed. In step S1112, the second particle type determination unit 103b determines that the image region corresponding to the secondary object j selected in step S1110 is an image region of coal.

[0103] Next, in step S1113, the second particle type discrimination unit 103b determines whether or not all secondary objects j included in the noise-reduced second binarized image 240 have been selected. If the result of this determination is that all secondary objects j included in the noise-reduced second binarized image 240 have not been selected, the processing of the above-mentioned step S1110 is performed again. Then, in step S1110 and thereafter, the second particle type discrimination unit 103b performs processing on the secondary objects j included in the noise-reduced second binarized image 240 that have not been selected in step S1110.

[0104] If the result of the determination in step S1111 described above is that the ratio Rh(j) of the hole sizes of the secondary object does not exceed the threshold (NO in step S1111), the process of step S1114 is performed. In step S1114, the second particle type determination unit 103b derives the convex hull of the secondary object j included in the second binarized image 240 after noise reduction, and derives the ratio Rc(j) of the difference from the convex hull of the secondary object using equation (4).

[0105] Next, in step S1115, the second particle type discrimination unit 103b determines whether the ratio Rc(j) of the difference from the convex hull of the secondary object exceeds a threshold value. If the result of this determination is that the ratio Rc(j) of the difference from the convex hull of the secondary object exceeds the threshold value (YES in step S1115), the processing of step S1116 is performed. In step S1116, the second particle type discrimination unit 103b determines that the image region corresponding to the secondary object j selected in step S1110 is an image region of coal. When the processing of step S1116 ends, the processing of step S1113 described above is performed.

[0106] On the other hand, if the ratio Rc(j) of the difference from the convex hull of the secondary object does not exceed the threshold (NO in step S1115), the process of step S1117 is performed. In step S1117, the second particle type determination unit 103b determines that the image region corresponding to the secondary object j selected in step S1110 is an image region of coke. When the process of step S1117 ends, the process of step S1113 described above is performed.

[0107] The processing of steps S1110 to S1117 is repeated until it is determined in step S1113 that all secondary objects j included in noise-reduced second binarized image 240 have been selected. Then, when it is determined in step S1113 that all secondary objects j included in noise-reduced second binarized image 240 have been selected (YES in step S1113), the processing of step S1118 in FIG. 11-3 is performed.

[0108] In step S1118, the particle feature amount deriving unit 104 selects one type of dust particle from the types (gangue, iron ore, coke, coal) determined in steps S1109, S1112, S1116, and S1117. Next, in step S1119, the particle feature amount deriving unit 104 derives the feature amounts (number fraction, phase fraction, and aspect ratio) of the dust particles of the type selected in step S1118.

[0109] Next, in step S1120, the particle feature derivation unit 104 determines whether all of the dust particle types identified in steps S1109, S1112, S1116, and S1117 have been selected. If this determination shows that all of the dust particle types have not been selected (NO in step S1120), the process of step S1118 described above is performed again. Then, in step S1118 and thereafter, the particle feature derivation unit 104 performs processing on any dust particle types identified in steps S1109, S1112, S1116, and S1117 that have not been selected in step S1118. The processes of steps S1118 to S1120 are repeated until it is determined in step S1116 that all of the dust particle types identified in steps S1109, S1112, S1116, and S1117 have been selected.

[0110] Then, in step S1120, if it is determined that all of the types of dust particles determined in steps S1109, S1112, S1116, and S1117 have been selected (YES in step S1120), the process proceeds to step S1121. In step S1121, the operational state determination unit 105 determines the operational state of the equipment based on at least one of the types of dust particles determined in steps S1109, S1112, S1116, and S1117 and the feature amounts of the dust particles derived in step S1119.

[0111] Next, in step S1122, the output unit 106 outputs information indicating the operation status of the equipment determined in step S1121. In addition to or instead of the information indicating the operation status of the equipment, the output unit 106 may output at least one of information indicating the type of dust particle determined in steps S1109, S1112, S1116, and S1117 and information indicating the feature amount of the dust particle derived in step S1119. When the process of step S1117 ends, the process according to the flowcharts of FIGS. 11-1 to 11-3 ends.

[0112] <Summary> As described above, in this embodiment, the processing device 100 binarizes the pixel values ​​of the binarization target image using a first binarization threshold Th0 that is common to all pixels of the binarization target image, thereby creating a first binarized image 220. The processing device 100 binarizes the pixel values ​​of the binarization target image using a second binarization threshold Th(i) that is set for each primary object i included in the first binarized image 220, individually using each second binarization threshold Th(i), thereby creating a second binarized image 230. The processing device 100 identifies the type of dust particles included in the polarized image based on the results of processing including these image processing steps. Therefore, it is not necessary to capture both a bright-field image and a polarized image, as in the technology described in Patent Document 1, for example. Furthermore, by creating the first binarized image 220 and the second binarized image 230, the accuracy of extracting the contours of dust particles in the binarized image (second binarized image 230) can be improved. Therefore, it is possible to reduce the processing load when determining the type of dust particles emitted from the equipment and to improve the accuracy of determining the type of dust particles.

[0113] Furthermore, in this embodiment, the processing device 100 derives the second binarization threshold Th(i) for the primary object based on a physical quantity (e.g., luminance) representing the brightness of the image region corresponding to the primary object and a physical quantity (e.g., luminance) representing the brightness of the image region not corresponding to the primary object. Therefore, the second binarization threshold Th(i) can be derived taking into consideration both the brightness of the image region corresponding to the primary object and the brightness of the image region not corresponding to the primary object. This further improves the accuracy of extracting the contours of dust particles in the second binarized image 230.

[0114] Furthermore, in this embodiment, the processing device 100 expresses a physical quantity (e.g., luminance) representing the brightness of an image region corresponding to a primary object and a physical quantity (e.g., luminance) representing the brightness of an image region not corresponding to a primary object, respectively, as basic statistics. Therefore, when deriving the second binarization threshold Th(i), it is possible to take into account the size and brightness distribution of an image region corresponding to a primary object. Similarly, when deriving the second binarization threshold Th(i), it is possible to take into account the size and brightness distribution of an image region not corresponding to a primary object. Therefore, it is possible to further improve the accuracy of extracting the contours of dust particles in the second binarized image 230.

[0115] In this embodiment, the processing device 100 identifies the type of dust particle based on the feature values ​​of image regions corresponding to objects (secondary objects) included in the second binarized image 230. As described above, by creating the first binarized image 220 and the second binarized image 230, the accuracy of extracting the outlines of dust particles in the binarized image (second binarized image 230) can be improved. Therefore, the accuracy of identifying the type of dust particle based on the feature values ​​of image regions corresponding to objects (secondary objects) included in the second binarized image 230 can be improved. For example, when identifying types of dust particles that differ in brightness in an image, the accuracy of identifying the type of dust particle can be improved by including a physical quantity representing brightness in the feature values ​​of image regions corresponding to secondary objects. Furthermore, when identifying types of dust particles that differ in shape in an image, the accuracy of identifying the type of dust particle can be improved by including shape in the feature values ​​of image regions corresponding to secondary objects.

[0116] Furthermore, in this embodiment, the processing device 100 identifies the type of dust particle based on basic statistics of physical quantities that represent the brightness of the image region corresponding to the secondary object. Therefore, the type of dust particle can be identified taking into account the size and brightness distribution of the image region corresponding to the secondary object. This further improves the accuracy of identifying the type of dust particle.

[0117] Furthermore, in this embodiment, the processing device 100 discriminates the type of dust particle using a learning model that has learned the relationship between the physical quantity representing the brightness of an image region corresponding to a secondary object and the type of dust particle. Therefore, for example, the type of dust particle can be discriminated without setting a rule relating the physical quantity representing the brightness of an image region corresponding to a secondary object to the type of dust particle. Here, the computational load can be reduced by using a model including a classifier that performs clustering by machine learning as the learning model. Furthermore, the burden of creating training data by creating correct labels can be reduced by using unsupervised learning as the machine learning method.

[0118] In this embodiment, the processing device 100 distinguishes dust particles that have been determined to be of the same type based on the physical quantity representing the brightness of the image area corresponding to the secondary object, based on the shape of the image area corresponding to the secondary object, thereby improving the accuracy of distinguishing types of dust particles that are difficult to distinguish based on the brightness of the image.

[0119] Furthermore, in this embodiment, the processing device 100 reduces noise contained in the second binarized image 230 to create a noise-reduced second binarized image 240. Therefore, it is possible to further improve the accuracy of extracting the contours of dust particles in the binarized image (noise-reduced second binarized image 240).

[0120] Furthermore, in this embodiment, the processing device 100 converts the polarization image into a grayscale image 210. Therefore, it is possible to reduce the calculation load during processing including image processing.

[0121] Furthermore, in this embodiment, the processing device 100 derives dust particle feature amounts for each type of dust particle identified by the particle type identification unit 103 based on the polarized image or the polarized image that has been processed, including image processing by the image processing unit 102. This improves the accuracy of deriving feature amounts for each type of dust particle.

[0122] In this embodiment, the processing apparatus 100 determines the operating state of the equipment based on at least one of the type of dust particles and the feature amount of the dust particles, and can grasp the operating state of the equipment from the state of the dust particles.

[0123] (Other embodiments) The above-described embodiments of the present disclosure can be realized by a computer executing a program. A computer-readable recording medium having the program recorded thereon and a computer program product such as the program can also be applied as embodiments of the present disclosure. Examples of recording media that can be used include flexible disks, hard disks, optical disks, magneto-optical disks, CD-ROMs, magnetic tapes, non-volatile memory cards, and ROMs. The embodiments of the present disclosure can also be realized by a programmable logic controller (PLC) or dedicated hardware such as an application-specific integrated circuit (ASIC). Furthermore, the above-described embodiments of the present disclosure are merely examples of specific embodiments for carrying out the present disclosure, and the technical scope of the present disclosure should not be interpreted as being limited by these. In other words, the present disclosure can be embodied in various forms without departing from its technical concept or main features.

[0124] The present disclosure can be summarized as follows, for example. [Disclosure 1] A processing device that performs processing including identifying the type of dust particles discharged from equipment that processes solid raw materials, an acquisition unit that acquires a polarized image including the dust particles; an image processing unit that performs processing including image processing on the polarization image; a particle type discriminator that discriminates the type of the dust particles included in the polarized image based on a result of processing including the image processing by the image processor; and Equipped with The image processing unit a binarization unit that binarizes pixel values ​​of an image to be binarized to create a binarized image; the binarization target image is the polarized image or the polarized image obtained by performing a process different from the binarization process on the polarized image, The binarization unit a first binarization unit that creates a first binarized image by processing that includes binarizing pixel values ​​of the binarization target image using a first binarization threshold that is a common binarization threshold for all pixels of the binarization target image; a second binarization unit that creates a second binarized image by processing including binarizing pixel values ​​of the binarization target image using second binarization thresholds that are binarization thresholds set for each primary object that is an object represented as the dust particle in the first binarized image, by individually using each of the second binarization thresholds; A processing device comprising: [Disclosure 2] The processing device described in Disclosure 1, wherein the second binarization unit derives the second binarization threshold for the primary object based on a physical quantity representing the brightness of a region of the binarization target image that corresponds to the primary object and a physical quantity representing the brightness of a region of the binarization target image that does not correspond to the primary object. [Disclosure 3] The processing device according to Disclosure 2, wherein a physical quantity representing the brightness of an area corresponding to the primary object and a physical quantity representing the brightness of an area not corresponding to the primary object are each represented by a basic statistical quantity. [Disclosure 4] 4. The processing device according to claim 2 or 3, wherein the physical quantity representing brightness includes luminance. [Disclosure 5] The processing device according to any one of Disclosures 1 to 4, wherein the particle type discriminator discriminates the type of the dust particle based on a feature amount of a region of the binarization target image, the second binarized image, or a binarized image obtained by processing the second binarized image, which corresponds to a secondary object that is an object represented as the dust particle in the second binarized image or the binarized image obtained by processing the second binarized image. [Disclosure 6] The processing device according to Disclosure 5, wherein the feature amount of the region corresponding to the secondary object includes at least one of a physical amount representing brightness and a shape. [Disclosure 7] The particle type discrimination unit The processing device according to Disclosure 6, further comprising a first particle type discriminator that discriminates the type of the dust particle based on a physical quantity that represents the brightness of a region of the binarization target image, the second binarized image, or a binarized image obtained by processing the second binarized image, that corresponds to a secondary object that is an object represented as the dust particle in the second binarized image or the binarized image obtained by processing the second binarized image. [Disclosure 8] The processing device according to Disclosure 7, wherein the physical quantity representing the brightness of the region corresponding to the secondary object is represented by a basic statistical quantity. [Disclosure 9] The processing device according to Disclosure 7 or 8, wherein the first particle type discrimination unit discriminates the type of the dust particle using a learning model that has learned a relationship between a physical quantity representing the brightness of the region corresponding to the secondary object and the type of the dust particle. [Disclosure 10] The processing device described in Disclosure 9, wherein the learning model includes a classifier that performs clustering using machine learning. [Disclosure 11] The processing device of Disclosure 10, wherein the machine learning is unsupervised learning. [Disclosure 12] The processing device according to any one of Disclosures 6 to 11, wherein the physical quantity representing brightness includes luminance. [Disclosure 13] The particle type discrimination unit The processing device according to any one of Disclosures 6 to 12, further comprising a second particle type discriminator that further classifies the types of the dust particles based on the shape of a region of the binarization target image, the second binarized image, or a binarized image obtained by processing the second binarized image, that corresponds to a secondary object, which is an object represented as the dust particle in the second binarized image or the binarized image obtained by processing the second binarized image. [Disclosure 14] The particle type discrimination unit a first particle type discriminator that discriminates the type of the dust particle based on a physical quantity that represents the brightness of a region of the binarization target image, the second binarized image, or a binarized image obtained by processing the second binarized image, corresponding to a secondary object that is an object represented as the dust particle in the second binarized image or the binarized image obtained by processing the second binarized image; The processing apparatus according to Disclosure 13, wherein the second particle type discrimination unit further classifies the types of the dust particles that have been discriminated to be of the same type by the first particle type discrimination unit. [Disclosure 15] The second binarization unit The processing device according to any one of Disclosures 1 to 14, further comprising a noise reduction unit that performs processing including reducing noise contained in the second binarized image. [Disclosure 16] The image processing unit a grayscale conversion unit that converts the polarized image into a grayscale image; The processing device according to any one of Disclosures 1 to 15, wherein the processing different from the binarization processing includes a processing for converting the polarized image into a gray-scale image. [Disclosure 17] The processing device according to any one of Disclosures 1 to 16, further comprising a particle feature derivation unit that derives feature values ​​of the dust particles for each type of dust particle identified by the particle type discrimination unit, based on the polarized image or a polarized image that has been processed by the image processing unit, including the image processing. [Disclosure 18] 18. The processing apparatus according to claim 17, wherein the characteristic amount of the dust particles includes at least one of a number ratio, a phase fraction, and an aspect ratio of the dust particles. [Disclosure 19] 19. The processing apparatus according to claim 17 or 18, further comprising an operational state determination unit that determines an operational state of the equipment based on the feature amounts of the dust particles derived by the particle feature amount derivation unit. [Disclosure 20] 20. The processing apparatus according to any one of Disclosures 1 to 19, further comprising an operational state determination unit that determines the operational state of the facility based on the type of dust particles determined by the particle type determination unit. [Disclosure 21] The processing apparatus according to any one of Disclosures 1 to 20, wherein the facility is a blast furnace. [Disclosure 22] 22. The processing apparatus according to claim 21, wherein the types of the dust particles identified by the particle type identification unit include iron ore, gangue, and coke or coal. [Disclosure 23] A processing method for performing processing including identifying the type of dust particles discharged from equipment for processing solid raw materials, comprising: an acquisition step of acquiring a polarized image including the dust particles; an image processing step of performing processing including image processing on the polarization image; a particle type determination step of determining the type of the dust particles included in the polarized image based on a result of processing including the image processing by the image processing step; and The image processing step includes: a binarization step of binarizing pixel values ​​of a target image to be binarized to create a binarized image, the binarization target image is the polarized image or the polarized image obtained by performing a process different from the binarization process on the polarized image, The binarization step includes: a first binarization step of creating a first binarized image by processing including binarizing pixel values ​​of the binarization target image using a first binarization threshold that is a common binarization threshold for all pixels of the binarization target image; a second binarization step of binarizing pixel values ​​of the binarization target image using second binarization thresholds, which are binarization thresholds set for each primary object representing the dust particle in the first binarized image, by individually using each of the second binarization thresholds to create a second binarized image; A processing method comprising: [Disclosure 24] A program for causing a computer to function as each part of the processing device according to any one of Disclosures 1 to 22. [Explanation of symbols]

[0125] 100 Processing equipment 101 Acquisition Department 102 Image processing section 102a Grayscale conversion section 102b Binarization section 102b1 First binarization unit 102b2 Second binarization unit 102b3 Noise reduction section 103 Particle type discrimination section 103a 1st particle type discrimination section 103b 2nd particle type discrimination section 104 Particle feature extraction unit 105 Operation status determination unit 106 Output section 110 Input Device 120 Output Device 210 grayscale images 220 First binarized image 221 Primary Objects 230 Second binarized image 231~232 Secondary Objects 240 Second binarized image after noise reduction 241 Secondary Objects 311 Objects 312 Lines indicating pixels from which luminance is extracted 321 Brightness 510 grayscale images 520 Second binarized image after noise reduction 530 Second binarized image after hole filling 531 Secondary Objects 910 Classification result image Graph showing the time variation of the feature values ​​of 1010 and 1020 dust particles

Claims

1. A processing device that performs processing including identifying the type of dust particles discharged from equipment that processes solid raw materials, an acquisition unit that acquires a polarized image including the dust particles; an image processing unit that performs processing including image processing on the polarization image; a particle type discriminator that discriminates the type of the dust particles included in the polarized image based on a result of processing including the image processing by the image processor; and Equipped with The image processing unit a binarization unit that binarizes pixel values ​​of an image to be binarized to create a binarized image; the image to be binarized is the polarized image or the polarized image obtained by performing a process different from the binarization process on the polarized image, The binarization unit a first binarization unit that creates a first binarized image by processing that includes binarizing pixel values ​​of the binarization target image using a first binarization threshold that is a common binarization threshold for all pixels of the binarization target image; a second binarization unit that creates a second binarized image by processing including binarizing pixel values ​​of the binarization target image using second binarization thresholds that are binarization thresholds set for each primary object that is an object represented as the dust particle in the first binarized image, by individually using each of the second binarization thresholds; A processing device comprising:

2. 2. The processing device according to claim 1, wherein the second binarization unit derives the second binarization threshold for the primary object based on a physical quantity representing the brightness of a region of the binarization target image that corresponds to the primary object and a physical quantity representing the brightness of a region of the binarization target image that does not correspond to the primary object.

3. The processing device according to claim 2 , wherein a physical quantity representing the brightness of the region corresponding to the primary object and a physical quantity representing the brightness of the region not corresponding to the primary object are each represented by a basic statistical quantity.

4. The processing device according to claim 2 , wherein the physical quantity representing brightness includes luminance.

5. 4. The processing device according to claim 1, wherein the particle type determination unit determines the type of the dust particle based on a feature amount of a region of the binarization target image, the second binarized image, or a binarized image obtained by processing the second binarized image, that corresponds to a secondary object, which is an object represented as the dust particle in the second binarized image or the binarized image obtained by processing the second binarized image.

6. The processing device according to claim 5 , wherein the feature amount of the region corresponding to the secondary object includes at least one of a physical amount representing brightness and a shape.

7. The particle type discrimination unit 7. The processing device according to claim 6, further comprising a first particle type discriminator that discriminates the type of the dust particle based on a physical quantity that represents the brightness of a region of the binarization target image, the second binarized image, or a binarized image obtained by processing the second binarized image, that corresponds to a secondary object that is an object represented as the dust particle in the second binarized image or the binarized image obtained by processing the second binarized image.

8. The processing device according to claim 7 , wherein the physical quantity representing the brightness of the region corresponding to the secondary object is represented by a basic statistical quantity.

9. 8. The processing device according to claim 7, wherein the first particle type determination unit determines the type of the dust particle using a learning model that has learned a relationship between the physical quantity representing the brightness of the region corresponding to the secondary object and the type of the dust particle.

10. The processing device according to claim 9 , wherein the learning model includes a classifier that performs clustering by machine learning.

11. The processing device of claim 10 , wherein the machine learning is unsupervised learning.

12. The processing device according to claim 6 , wherein the physical quantity representing brightness includes luminance.

13. The particle type discrimination unit 7. The processing device according to claim 6, further comprising a second particle type discriminator that further classifies the types of the dust particles based on a shape of a region of the binarization target image, the second binarized image, or a binarized image obtained by processing the second binarized image, that corresponds to a secondary object that is an object represented as the dust particle in the second binarized image or the binarized image obtained by processing the second binarized image.

14. The particle type discrimination unit a first particle type discriminator that discriminates the type of the dust particle based on a physical quantity that represents the brightness of a region of the binarization target image, the second binarized image, or a binarized image obtained by processing the second binarized image, corresponding to a secondary object that is an object represented as the dust particle in the second binarized image or the binarized image obtained by processing the second binarized image; The processing apparatus according to claim 13 , wherein the second particle type discriminator further classifies the types of the dust particles that have been discriminated as being of the same type by the first particle type discriminator.

15. The second binarization unit 4. The processing device according to claim 1, further comprising a noise reduction unit that performs processing including reducing noise contained in the second binarized image.

16. The image processing unit a grayscale conversion unit that converts the polarized image into a grayscale image; 4. The processing device according to claim 1, wherein the processing different from the binarization processing includes a processing for converting the polarization image into a gray-scale image.

17. 4. The processing device according to claim 1, further comprising a particle feature amount deriving unit that derives feature amounts of the dust particles for each type of dust particle identified by the particle type discriminator, based on the polarized image or a polarized image that has been processed by the image processing unit, including the image processing.

18. The processing apparatus according to claim 17 , wherein the characteristic amount of the dust particles includes at least one of a number ratio, a phase fraction, and an aspect ratio of the dust particles.

19. The processing apparatus according to claim 17 , further comprising an operation state determination unit that determines an operation state of the facility based on the feature amounts of the dust particles derived by the particle feature amount derivation unit.

20. 4. The processing apparatus according to claim 1, further comprising an operational state determination unit that determines an operational state of the facility based on the type of dust particles determined by the particle type determination unit.

21. The processing apparatus according to any one of claims 1 to 3, wherein the facility is a blast furnace.

22. The processing apparatus according to claim 21 , wherein the types of the dust particles identified by the particle type identifying unit include iron ore, gangue, and coke or coal.

23. A processing method for performing processing including identifying the type of dust particles discharged from equipment for processing solid raw materials, comprising: an acquisition step of acquiring a polarized image including the dust particles; an image processing step of performing processing including image processing on the polarization image; a particle type determination step of determining the type of the dust particles included in the polarized image based on a result of processing including the image processing by the image processing step; Equipped with The image processing step includes: a binarization step of binarizing pixel values ​​of a target image to be binarized to create a binarized image, the image to be binarized is the polarized image or the polarized image obtained by performing a process different from the binarization process on the polarized image, The binarization step includes: a first binarization step of creating a first binarized image by processing including binarizing pixel values ​​of the binarization target image using a first binarization threshold that is a common binarization threshold for all pixels of the binarization target image; a second binarization step of binarizing pixel values ​​of the binarization target image using second binarization thresholds, which are binarization thresholds set for each primary object representing the dust particle in the first binarized image, by individually using each of the second binarization thresholds to create a second binarized image; A processing method comprising:

24. A program for causing a computer to function as each part of the processing device according to any one of claims 1 to 3.

Images