Image processing device and operating method thereof

Abstract

The image processing device according to one embodiment disclosed in the present document may comprise: an image acquisition unit for acquiring active material images of a plurality of active materials; an image segmentation unit for segmenting the plurality of active materials included in the active material image to acquire a segmented image; and an image generation unit for inputting the active material image into an artificial intelligence model to generate at least one among a binary image in which the active material and surface cracks of the active material are separated, a pore binary image representing pore characteristics of the active material on the basis of the segmented image, or a combination thereof.

Description

Image processing device and its operating method

[0001] Cross-citation with related applications

[0002] This application claims the benefit of priority to Korean Patent Application No. 10-2023-0169803, filed November 29, 2023, Korean Patent Application No. 10-2023-0169805, filed November 29, 2023, and Korean Patent Application No. 10-2023-0169801, filed November 29, 2023, the entire contents of which are incorporated herein by reference.

[0003] Technology field

[0004] The embodiments disclosed in this document relate to an image processing device and an operating method thereof.

[0005] As technological developments and demand for mobile devices and electric vehicles increase, the demand for lithium secondary batteries as an energy source is rapidly increasing.

[0006] A lithium secondary battery is manufactured by placing an electrode assembly composed of a positive electrode, a separator, and an anode into a battery case and injecting an electrolyte. The positive electrode is manufactured by coating a composition for forming a positive electrode active material layer, which includes a positive electrode active material, a conductive material, and a binder, on a positive electrode current collector, followed by drying and rolling. The negative electrode is manufactured by coating a composition for forming a negative electrode active material layer, which includes a negative electrode active material, a conductive material, and a binder, on a negative electrode current collector, followed by drying and rolling.

[0007] Meanwhile, the electrochemical performance of lithium secondary batteries is influenced not only by the composition of the positive and negative active material particles used, but also by the shape of the active material particles. This is because electrochemical properties such as electrode density, electrical conductivity, and resistance characteristics vary depending on the shape and particle size of the active material particles. Therefore, to manufacture lithium secondary batteries with the desired performance, it is necessary to accurately analyze the shape of the active material particles.

[0008] Traditionally, the shape of active material particles has been analyzed using images obtained through a scanning electron microscope (SEM). However, while this method of analyzing the shape of active materials using SEM images allows for qualitative analysis, quantitative analysis requires manual measurement of individual active material particles selected from the SEM images. This method cannot be performed automatically. Therefore, quantitative analysis of large quantities of active material particles is difficult, and measurements can vary depending on the person performing the measurement, resulting in poor analytical accuracy.

[0009] The present invention is intended to solve the above problems and to provide a method capable of automatically quantitatively analyzing the shape characteristics of a large amount of active material (positive electrode active material or negative electrode active material) and / or precursor particles.

[0010] In addition, the present invention seeks to provide a quantitative analysis method that minimizes user involvement in order to solve the problem of reduced analysis accuracy due to differences in measurement values ​​depending on the user.

[0011] The technical problems of the embodiments disclosed in this document are not limited to the technical problems mentioned above, and other technical problems not mentioned will be clearly understood by those skilled in the art from the descriptions below.

[0012] An image processing device according to an embodiment disclosed in the present document may include an image acquisition unit that acquires an active material image for a plurality of active materials, an image segmentation unit that segments the plurality of active materials included in the active material image to acquire a segmentation image, and an image generation unit that inputs the active material image into an artificial intelligence model to generate at least one of a binary image in which the active material and a surface crack of the active material are separated, a pore binary image representing pore characteristics of the active material based on the segmentation image, or any combination thereof.

[0013] An image processing method according to an embodiment disclosed in the present document may include an operation of obtaining an active material image for a plurality of active materials, an operation of segmenting the plurality of active materials included in the active material image to obtain a segmentation image, and an operation of inputting the active material image into an artificial intelligence model to generate at least one of a binary image in which the active material and a surface crack of the active material are separated, a pore binary image representing pore characteristics of the active material based on the segmentation image, or any combination thereof.

[0014] According to the embodiments disclosed in this document, the shape characteristics of a large quantity of active material particles can be automatically and quantitatively analyzed.

[0015] Additionally, according to the embodiments disclosed in this document, the problem of measurements varying depending on the user can be minimized by utilizing an artificial intelligence model that minimizes user input.

[0016] In addition, various effects may be provided, either directly or indirectly, through this document.

[0017] FIG. 1 is a block diagram of an image processing device according to an embodiment of the present disclosure.

[0018] FIGS. 2A to 2D illustrate learning data according to various embodiments of the present disclosure.

[0019] FIGS. 3A to 3D illustrate artificial intelligence models according to various embodiments of the present disclosure.

[0020] FIGS. 4A to 4D illustrate images processed by an image processing device according to various embodiments of the present disclosure.

[0021] FIG. 5 is a flowchart illustrating an active material image processing method of an image processing device according to an embodiment of the present disclosure.

[0022] FIG. 6 is a flowchart illustrating an artificial intelligence model learning method of an image processing device according to an embodiment of the present disclosure.

[0023] FIG. 7 is a block diagram of an image processing device according to an embodiment of the present disclosure.

[0024] FIG. 8 illustrates images processed by an image processing device according to an embodiment of the present disclosure.

[0025] FIG. 9 is a flowchart illustrating an active material image processing method of an image processing device according to an embodiment of the present disclosure.

[0026] FIG. 10 is a block diagram of an image processing device according to an embodiment of the present disclosure.

[0027] FIG. 11 illustrates images processed by an image processing device according to an embodiment of the present disclosure.

[0028] FIG. 12 is a flowchart illustrating an active material image processing method of an image processing device according to an embodiment of the present disclosure.

[0029] FIG. 13 is a block diagram of an image processing device according to an embodiment of the present disclosure.

[0030] FIG. 14 illustrates images processed by an image processing device according to an embodiment of the present disclosure.

[0031] FIG. 15 is a drawing for explaining a method for setting a core region and a shell region of an active material by an image processing device according to an embodiment of the present disclosure.

[0032] FIG. 16 is a flowchart illustrating an active material image processing method of an image processing device according to an embodiment of the present disclosure.

[0033] FIG. 17 is a flowchart illustrating a method for setting a core region and a shell region of an active material by an image processing device according to an embodiment of the present disclosure.

[0034] Hereinafter, embodiments of the present invention will be described with reference to the attached drawings. However, this is not intended to limit the present invention to specific embodiments, and it should be understood that the present invention encompasses various modifications, equivalents, and / or alternatives of the embodiments.

[0035] The embodiments and terminology used in this document are not intended to limit the technical features described in this document to a specific embodiment, but should be understood to encompass various modifications, equivalents, or alternatives of the embodiment. In connection with the description of the drawings, similar reference numerals may be used to refer to similar or related components. The singular form of a noun corresponding to an item may include one or more of the item, unless the relevant context clearly indicates otherwise.

[0036] In this document, the phrases "A or B", "at least one of A and B", "at least one of A or B", "A, B, or C", "at least one of A, B, and C", and "at least one of A, B, or C" can each include any one of the items listed together in that phrase, or all possible combinations thereof. Terms such as "first", "second", "first", "second", "A", "B", "(a)", or "(b)" may be used merely to distinguish the corresponding component from other corresponding components, and do not limit the corresponding components in any other respect (e.g., importance or order) unless specifically stated otherwise.

[0037] In this document, when a component (e.g., a first component) is referred to as being “connected,” “coupled,” or “connected,” with or without the terms “functionally” or “communicatively,” or “coupled” or “connected,” it means that the component can be connected to the other component directly (e.g., wired or wirelessly), or indirectly (e.g., via a third component).

[0038] The methods according to various embodiments disclosed in this document may be provided as included in a computer program product. The computer program product may be traded as a commodity between a seller and a buyer. The computer program product may be distributed in the form of a machine-readable storage medium (e.g., compact disc read-only memory, CD-ROM), or may be distributed online (e.g., downloaded or uploaded) through an application store or directly between two user devices. In the case of online distribution, at least a portion of the computer program product may be temporarily stored or temporarily generated in a machine-readable storage medium, such as the memory of a manufacturer's server, an application store's server, or an intermediary server.

[0039] According to the embodiments disclosed in this document, each component (e.g., a module or a program) of the above-described components may include one or more entities, and some of the entities may be separated and placed in other components. According to the embodiments disclosed in this document, one or more components or operations of the aforementioned components may be omitted, or one or more other components or operations may be added. Alternatively or additionally, a plurality of components (e.g., a module or a program) may be integrated into a single component. In this case, the integrated component may perform one or more functions of each of the plurality of components identically or similarly to those performed by the corresponding component among the plurality of components prior to the integration. According to the embodiments disclosed in this document, the operations performed by a module, program, or other component may be executed sequentially, in parallel, iteratively, or heuristically, or one or more of the operations may be executed in a different order, omitted, or one or more other operations may be added.

[0040] In this document, the term "active material" refers to an electrode material of a lithium secondary battery, and may refer to a positive electrode active material constituting a positive electrode of the secondary battery or a negative electrode active material constituting a negative electrode of the secondary battery. For example, an active material image processed by an image processing device according to various embodiments of the present disclosure may be a positive electrode active material image or a negative electrode active material image. Accordingly, the various embodiments of the present disclosure are not limited to a subject of either a positive electrode active material image or a negative electrode active material image, but may be applied to each of a positive electrode active material image and a negative electrode active material image.

[0041] Hereinafter, an image processing device and an operating method thereof according to the first embodiment, the second embodiment, the third embodiment, and the fourth embodiment of the present disclosure will be described with reference to FIGS. 1 to 6.

[0042] Fig. 1 is a block diagram of an image processing device (101) according to an embodiment of the present disclosure. The image processing device (101) of Fig. 1 may correspond to the image processing devices according to the first to fourth embodiments of the present disclosure.

[0043] Referring to FIG. 1, the image processing device (101) can be connected to the image acquisition device (103) wired and / or wirelessly.

[0044] In one embodiment, the connection (105) between the image processing device (101) and the image acquisition device (103) may be a communication connection via a wired and / or wireless network. In one embodiment, the wired network may be based on a local area network (LAN) communication or a power line communication. In one embodiment, the wireless network may be based on a short-range communication network (e.g., Bluetooth, wireless fidelity (WiFi), or infrared data association (IrDA)), or a wide-range communication network (cellular network, 4G network, 5G network).

[0045] In another embodiment, the connection (105) between the image processing device (101) and the image acquisition device (103) may be a connection via a device-to-device communication method (e.g., a bus, a general purpose input and output (GPIO), a serial peripheral interface (SPI), or a mobile industry processor interface (MIPI)).

[0046] In one embodiment, the image acquisition device (103) may be a microscope (e.g., a scanning electron microscope). In one embodiment, the image acquisition device (103) may be a device that acquires an image of the sample surface by scanning a focused electron beam onto the sample surface and converting secondary electrons generated by the interaction between the electron beam and the sample into an image signal.

[0047] In one embodiment, the image acquisition device (103) can acquire images of an active material and / or a precursor. Hereinafter, the images of the active material and / or the precursor may be referred to as active material images. However, even if referred to as active material images, the present disclosure does not exclude images of precursors.

[0048] In one embodiment, the image acquisition device (103) can acquire an active material image for the active material. For example, the image acquisition device (103) can acquire a scanning electron microscope (SEM) image by scanning an electron beam on the active material powder. Depending on the embodiment, the SEM image can be replaced with an image based on a transmission electron microscope (TEM), an optical microscope (OM), a structured illumination microscopy (SIM), or a fused ion beam (FIB).

[0049] In one embodiment, the image acquisition device (103) can transmit an active material image of the active material to the image processing device (101). For example, the image acquisition device (103) can transmit an active material image of the active material to the image processing device (101) via a connection (105).

[0050] In one embodiment, the image processing device (101) may be a mobile device (e.g., a mobile phone, a laptop computer, a smart phone, a smart pad), a computer (e.g., a general-purpose computer, a special-purpose computer).

[0051] In one embodiment, the image processing device (101) may include a communication circuit (110), a memory (120), and a processor (130). According to an embodiment, the image processing device (101) illustrated in FIG. 1 may further include at least one component (e.g., a display, an input device, or an output device) other than the components illustrated in FIG. 1.

[0052] In one embodiment, the communication circuit (110) can establish a wired communication channel and / or a wireless communication channel between the image processing device (101) and / or the image acquisition device (103), and transmit and receive data to and from the image acquisition device (103) through the established communication channel.

[0053] In one embodiment, the memory (120) may include volatile memory and / or non-volatile memory.

[0054] In one embodiment, the memory (120) may store data used by at least one component (e.g., the processor (130)) of the image processing device (101). For example, the data may include a program (725) (or instructions related thereto), input data, or output data. In one embodiment, the instructions, when executed by the processor (130), may cause the image processing device (101) to perform operations defined by the instructions.

[0055] In one embodiment, the memory (120) may include a program (725) (e.g., an artificial intelligence model learning unit (141), an artificial intelligence model (145), an image acquisition unit (150), an image generation unit (160), an object identification unit (170), an image segmentation unit (180), and / or an information extraction unit (190)).

[0056] In one embodiment, the processor (130) may include a central processing unit, an application processor, a graphics processing unit, a neural processing unit (NPU), an image signal processor, a sensor hub processor, or a communication processor.

[0057] In one embodiment, the processor (130) may execute a program (725) (e.g., an artificial intelligence model learning unit (141), an artificial intelligence model (145), an image acquisition unit (150), an image generation unit (160), an object identification unit (170), an image segmentation unit (180), and / or an information extraction unit (190)) to control at least one other component (e.g., a hardware or software component) of an image processing device (101) connected to the processor (130) and perform various data processing or operations.

[0058] Hereinafter, with reference to FIGS. 2 to 4, a method for processing an image acquired from an image acquisition device (103) by an image processing device (101) through an artificial intelligence model learning unit (141), an artificial intelligence model (145), an image acquisition unit (150), an image generation unit (160), an object identification unit (170), an image segmentation unit (180), and / or an information extraction unit (190) will be specifically described.

[0059] Training data

[0060] Figures 2a to 2d illustrate learning data according to various embodiments of the present disclosure. Specifically, Figure 2a illustrates learning data according to the first embodiment, Figure 2b illustrates learning data according to the second embodiment, Figure 2c illustrates learning data according to the third embodiment, and Figure 2d illustrates learning data according to the fourth embodiment.

[0061] Referring to FIG. 2a, the learning data according to the first embodiment may include reference active material images (211, 212, 213, 214, 215, 216, 217, 218, 219) and reference binary images (221, 222, 223, 224, 225, 226, 227, 228, 229).

[0062] The sizes of the reference active material images (211, 212, 213, 214, 215, 216, 217, 218, 219) and the reference binary images (221, 222, 223, 224, 225, 226, 227, 228, 229) included in the learning data may be the same. In one embodiment, the size of the image may be defined as the number of horizontal pixels × the number of vertical pixels. The number of horizontal pixels of the images included in the learning data may be an integer between 32 and 4096, and the number of vertical pixels may be an integer between 32 and 4096. For example, the size of the image may be 256 × 256.

[0063] In one embodiment, the plurality of reference active material images (211, 212, 213, 214, 215, 216, 217, 218, 219) may include one or more first reference active material images and one or more second reference active material images.

[0064] In one embodiment, the first reference active material images may be acquired by photographing the active material powder. In one embodiment, the first reference active material images may be SEM images acquired directly through an image acquisition device (103).

[0065] In one embodiment, the second reference active material images may be images transformed from the first reference active material images. In one embodiment, the second reference active material images may be generated by applying a first image processing algorithm specified for the first reference active material images. Here, the first image processing algorithm may include rotation, tilt, shear, brightness adjustment, contrast adjustment, zoom in, zoom out, or a combination thereof.

[0066] The content of the reference active material image described above can be equally applied to the reference active material images of FIGS. 2b to 2d described later.

[0067] In one embodiment, the plurality of reference binary images (221, 222, 223, 224, 225, 226, 227, 228, 229) may include one or more first reference binary images and one or more second reference binary images.

[0068] In one embodiment, the first reference binary images may be generated by applying a second image processing algorithm to the first reference active material image. Here, the second image processing algorithm may include a mean shift filter, a boundary extraction algorithm, a boundary removal algorithm, a binarization algorithm, or a combination thereof.

[0069] In one embodiment, the second reference binary images may be images transformed from the first reference binary images. In one embodiment, the second reference binary images may be generated by applying a first image processing algorithm specified to the first reference binary images. Here, the first image processing algorithm may include rotation, tilt, share, brightness adjustment, contrast adjustment, zoom in, zoom out, or a combination thereof.

[0070] In one embodiment, the first reference binary images may be binary images of the first reference active material images, and the second reference binary images may be binary images of the second reference active material images. For example, the reference binary image (221) may be a binary image for the reference active material image (211), the reference binary image (225) may be a binary image for the reference active material image (215), and the reference binary image (229) may be a binary image for the reference active material image (219). Accordingly, the plurality of reference active material images (211, 212, 213, 214, 215, 216, 217, 218, 219) and the plurality of reference binary images (221, 222, 223, 224, 225, 226, 227, 228, 229) can be classified into an image set with corresponding images. For example, the reference active material image (211) and the reference binary image (221) can be classified into one image set.

[0071] In one embodiment, the training data may be used to train an artificial intelligence model (145). In one embodiment, the training data may be used to train the artificial intelligence model (145) for a specified number of epochs. Here, the specified number of epochs may be determined between 100 and 10,000. For example, the specified number of epochs may be 3,000.

[0072] In one embodiment, the training data may be divided into mini-batches divided by a batch size. Here, the batch size may be determined between 1 and 512. For example, the batch size may be 3. When the batch size is 3, each of three sets of images (i.e., three reference active material images and three reference binary images) may be divided into a mini-batch. For example, reference active material images (211, 212, 213) and reference binary images (221, 222, 223) may constitute a first mini-batch, reference active material images (214, 215, 216) and reference binary images (224, 225, 226) may constitute a second mini-batch, and reference active material images (217, 218, 219) and reference binary images (227, 228, 229) may constitute a third mini-batch.

[0073] Referring to FIG. 2b, the learning data according to the second embodiment includes reference active material images (231, 232, 233, 234, 235, 236, 237, 238, 239), first type reference binary images (231-1, 232-1, 233-1, 234-1, 235-1, 236-1, 237-1, 238-1, 239-1) corresponding to the first particle type, second type reference binary images (231-2, 232-2, 233-2, 234-2, 235-2, 236-2, 237-2, 238-2, 239-2) corresponding to the second particle type, and third type reference binary images (231-2, 232-2, 233-2, 234-2, 235-2, 236-2, 237-2, 238-2, 239-2) corresponding to the third particle type. It may include images (231-3, 232-3, 233-3, 234-3, 235-3, 236-3, 237-3, 238-3, 239-3).

[0074] Here, each particle type (e.g., the first particle type, the second particle type, and the third particle type) may be preset based on the shape (e.g., size, particle size, sphericity, aspect ratio, convexity, and / or solidity) of the active material particles. For example, active material particles larger than a specified size may be set as large particles, and active material particles smaller than a specified size may be set as small particles.

[0075] The reference image corresponding to a specific particle type may be a binary image including only active material particles of the specific particle type classified according to preset shape criteria among the active material particles included in the reference active material image. For example, the first type reference binary images (231-1, 232-1, 233-1, 234-1, 235-1, 236-1, 237-1, 238-1, 239-1) are binary images that include only the first particle type of active material particles among the active material particles included in the reference active material images (231, 232, 233, 234, 235, 236, 237, 238, 239), and the second type reference binary images (231-2, 232-2, 233-2, 234-2, 235-2, 236-2, 237-2, 238-2, 239-2) are binary images that include only the first particle type of active material particles among the active material particles included in the reference active material images (231, 232, 233, 234, 235, 236, 237, 238, 239). The binary images include only the second particle type of active material particles among the active material particles included in the reference images (234, 235, 236, 237, 238, 239), and the third type reference binary images (231-3, 232-3, 233-3, 234-3, 235-3, 236-3, 237-3, 238-3, 239-3) may be binary images including only the third particle type of active material particles among the reference active material images (231, 232, 233, 234, 235, 236, 237, 238, 239).

[0076] In the present disclosure, it is explained assuming that there are three types of particles, but this is not limited to this, and the types of active material particles can be set in various ways.

[0077] Reference active material images included in the training data (231, 232, 233, 234, 235, 236, 237, 238, 239), first type reference binary images (231-1, 232-1, 233-1, 234-1, 235-1, 236-1, 237-1, 238-1, 239-1), second type reference binary images (231-2, 232-2, 233-2, 234-2, 235-2, 236-2, 237-2, 238-2, 239-2), and third type reference binary images (231-3, 232-3, 233-3, 234-3, 235-3, The sizes of 236-3, 237-3, 238-3, 239-3) may be the same.

[0078] In one embodiment, the plurality of reference active material images (231, 232, 233, 234, 235, 236, 237, 238, 239) may include one or more first reference active material images and one or more second reference active material images.

[0079] In one embodiment, the first type reference binary images (231-1, 232-1, 233-1, 234-1, 235-1, 236-1, 237-1, 238-1, 239-1), the second type reference binary images (231-2, 232-2, 233-2, 234-2, 235-2, 236-2, 237-2, 238-2, 239-2), and the third type reference binary images (231-3, 232-3, 233-3, 234-3, 235-3, 236-3, 237-3, 238-3, 239-3) may include one or more first reference binary images and one or more second reference binary images.

[0080] In one embodiment, the first reference binary images may be generated by applying a third image processing algorithm to the first reference active material image. Here, the third image processing algorithm may be an algorithm that combines a mean shift filter, a boundary extraction algorithm, a boundary removal algorithm, a binarization algorithm, or a combination thereof with a particle extraction algorithm that extracts active material particles corresponding to a specific particle type based on shape information of the active material particles. For example, the first reference binary images included in the first type reference binary images (231-1, 232-1, 233-1, 234-1, 235-1, 236-1, 237-1, 238-1, 239-1) may be generated by applying a third image processing algorithm that combines a binarization algorithm and a particle extraction algorithm that extracts active material particles of the first particle type to the first reference active material images.

[0081] In one embodiment, the second reference binary images may be images transformed from the first reference binary images. In one embodiment, the second reference binary images may be generated by applying a first image processing algorithm specified to the first reference binary images. Here, the first image processing algorithm may include rotation, tilt, share, brightness adjustment, contrast adjustment, zoom in, zoom out, or a combination thereof.

[0082] In one embodiment, the first reference binary images may be binary images from which specific particle types are extracted from the first reference active material images, and the second reference binary images may be binary images from which specific particle types are extracted from the second reference active material images.

[0083] For example, the first type reference binary image (231-1) may be a binary image from which the first particle type is extracted from the reference active material image (231), the first type reference binary image (235-1) may be a binary image from which the first particle type is extracted from the reference active material image (235), and the first type reference binary image (239-1) may be a binary image from which the first particle type is extracted from the reference active material image (239). The second type reference binary image (231-2) may be a binary image from which the second particle type is extracted from the reference active material image (231), the second type reference binary image (235-2) may be a binary image from which the second particle type is extracted from the reference active material image (235), and the second type reference binary image (239-2) may be a binary image from which the second particle type is extracted from the reference active material image (239). The third type reference binary image (231-3) may be a binary image in which a third particle type is extracted from the reference active material image (231), the third type reference binary image (235-3) may be a binary image in which a third particle type is extracted from the reference active material image (235), and the third type reference binary image (239-3) may be a binary image in which a third particle type is extracted from the reference active material image (239).

[0084] Therefore, multiple reference active material images (231, 232, 233, 234, 235, 236, 237, 238, 239) and multiple reference binary images (231-1, 232-1, 233-1, 234-1, 235-1, 236-1, 237-1, 238-1, 239-1, 231-2, 232-2, 233-2, 234-2, 235-2, 236-2, 237-2, 238-2, 239-2, 231-3, 232-3, 233-3, 234-3, 235-3, 236-3, 237-3, 238-3, 239-3) can be classified into image sets based on corresponding images. For example, the reference active material image (231) and the first type reference binary image (231-1) can be classified into one image set. As another example, the reference active material image (231) and the second type reference binary image (231-2) can be classified into one image set.

[0085] In one embodiment, the training data may be used to train an artificial intelligence model (145). In one embodiment, the training data may be used to train the artificial intelligence model (145) for a specified number of epochs. Here, the specified number of epochs may be determined between 100 and 10,000. For example, the specified number of epochs may be 3,000.

[0086] In one embodiment, the training data may be divided into mini-batches divided by a batch size. Here, the batch size may be determined between 1 and 512. For example, the batch size may be 3. When the batch size is 3, each of three sets of images (i.e., three reference active material images and three reference binary images) may be divided into a mini-batch.

[0087] For example, reference active material images (231, 232, 233) and first type reference binary images (231-1, 232-1, 233-1) may constitute a first mini-batch, reference active material images (234, 235, 236) and first type reference binary images (234-1, 235-1, 236-1) may constitute a second mini-batch, and reference active material images (237, 238, 239) and first type reference binary images (237-1, 238-1, 239-1) may constitute a third mini-batch. As another example, the reference active material images (231, 232, 233) and the second type reference binary images (231-2, 232-2, 233-2) may constitute a first mini-batch, the reference active material images (234, 235, 236) and the second type reference binary images (234-2, 235-2, 236-2) may constitute a second mini-batch, and the reference active material images (237, 238, 239) and the second type reference binary images (237-2, 238-2, 239-2) may constitute a third mini-batch.

[0088] Referring to FIG. 2c, learning data according to the third embodiment may include reference active material images (251, 252, 253, 254, 255, 256, 257, 258, 259), reference active material binary images corresponding to the active material (251-1, 252-1, 253-1, 254-1, 255-1, 256-1, 257-1, 258-1, 259-1), and reference impurity binary images corresponding to impurities such as excess Li present on the surface of the active material (251-2, 252-2, 253-2, 254-2, 255-2, 256-2, 257-2, 258-2, 259-2).

[0089] Here, the reference active material binary image may be a binary image that includes only active material particles from which impurities have been removed from the active material particles included in the reference active material image. Additionally, the reference impurity binary image may be a binary image that includes only impurities present on the surface of the active material particles included in the reference active material image.

[0090] The sizes of the reference active material images (251, 252, 253, 254, 255, 256, 257, 258, 259), the reference active material binary images (251-1, 252-1, 253-1, 254-1, 255-1, 256-1, 257-1, 258-1, 259-1) and the reference impurity binary images (251-2, 252-2, 253-2, 254-2, 255-2, 256-2, 257-2, 258-2, 259-2) included in the learning data may be the same.

[0091] In one embodiment, the plurality of reference active material images (251, 252, 253, 254, 255, 256, 257, 258, 259) may include one or more first reference active material images and one or more second reference active material images.

[0092] In one embodiment, the reference active material binary images (251-1, 252-1, 253-1, 254-1, 255-1, 256-1, 257-1, 258-1, 259-1) and the reference impurity binary images (251-2, 252-2, 253-2, 254-2, 255-2, 256-2, 257-2, 258-2, 259-2) may include one or more first reference binary images and one or more second reference binary images.

[0093] In one embodiment, the first reference binary images may be generated by applying a fourth image processing algorithm to the first reference active material image. Here, the fourth image processing algorithm may be an algorithm that combines a mean shift filter, a boundary extraction algorithm, a boundary removal algorithm, a binarization algorithm, or a combination thereof with an active material particle extraction algorithm that extracts active material particles from which impurities have been removed, or an impurity extraction algorithm that extracts impurities. For example, the first reference binary images included in the reference active material binary images (251-1, 252-1, 253-1, 254-1, 255-1, 256-1, 257-1, 258-1, 259-1) may be generated by applying the fourth image processing algorithm that combines a binarization algorithm and an active material particle extraction algorithm to the first reference active material images. As another example, the first reference binary images included in the reference impurity binary images (251-2, 252-2, 253-2, 254-2, 255-2, 256-2, 257-2, 258-2, 259-2) can be generated by applying a fourth image processing algorithm that combines a binarization algorithm and an impurity extraction algorithm to the first reference active material images.

[0094] In one embodiment, the second reference binary images may be images transformed from the first reference binary images. In one embodiment, the second reference binary images may be generated by applying a first image processing algorithm specified to the first reference binary images. Here, the first image processing algorithm may include rotation, tilt, share, brightness adjustment, contrast adjustment, zoom in, zoom out, or a combination thereof.

[0095] In one embodiment, the first reference binary images may be binary images from which the active material or impurities are extracted from the first reference active material images, and the second reference binary images may be binary images from which the active material or impurities are extracted from the second reference active material images.

[0096] For example, the reference active material binary image (251-1) may be a binary image from which the active material is extracted from the reference active material image (251), the reference active material binary image (255-1) may be a binary image from which the active material is extracted from the reference active material image (255), and the reference active material binary image (259-1) may be a binary image from which the active material is extracted from the reference active material image (259). The reference impurity binary image (251-2) may be a binary image from which the impurity is extracted from the reference active material image (251), the reference impurity binary image (255-2) may be a binary image from which the impurity is extracted from the reference active material image (255), and the reference impurity binary image (259-2) may be a binary image from which the impurity is extracted from the reference active material image (259).

[0097] Accordingly, the plurality of reference active material images (251, 252, 253, 254, 255, 256, 257, 258, 259) and the plurality of reference binary images (251-1, 252-1, 253-1, 254-1, 255-1, 256-1, 257-1, 258-1, 259-1, 251-2, 252-2, 253-2, 254-2, 255-2, 256-2, 257-2, 258-2, 259-2) can be classified into image sets with corresponding images. For example, the reference active material image (251) and the reference active material binary image (251-1) can be classified into one image set. As another example, the reference active material image (251) and the reference impurity binary image (251-2) can be separated into one image set.

[0098] In one embodiment, the training data may be used to train an artificial intelligence model (145). In one embodiment, the training data may be used to train the artificial intelligence model (145) for a specified number of epochs. Here, the specified number of epochs may be determined between 100 and 10,000. For example, the specified number of epochs may be 3,000.

[0099] In one embodiment, the training data may be divided into mini-batches divided by a batch size. Here, the batch size may be determined between 1 and 512. For example, the batch size may be 3. When the batch size is 3, each of three sets of images (i.e., three reference active material images and three reference binary images) may be divided into a mini-batch.

[0100] For example, reference active material images (251, 252, 253) and reference active material binary images (251-1, 252-1, 253-1) may constitute a first mini-batch, reference active material images (254, 255, 256) and reference active material binary images (254-1, 255-1, 256-1) may constitute a second mini-batch, and reference active material images (257, 258, 259) and reference active material binary images (257-1, 258-1, 259-1) may constitute a third mini-batch. As another example, reference active material images (251, 252, 253) and reference impurity binary images (251-2, 252-2, 253-2) may constitute a first mini-batch, reference active material images (254, 255, 256) and reference impurity binary images (254-2, 255-2, 256-2) may constitute a second mini-batch, and reference active material images (257, 258, 259) and reference impurity binary images (257-2, 258-2, 259-2) may constitute a third mini-batch.

[0101] Referring to FIG. 2d, learning data according to the third embodiment may include reference active material images (271, 272, 273, 274, 275, 276), reference active material binary images corresponding to the active material (271-1, 272-1, 273-1, 274-1, 275-1, 276-1), and reference crack binary images corresponding to active material surface cracks (271-2, 272-2, 273-2, 274-2, 275-2, 276-2).

[0102] Here, the reference active material binary image may be a binary image that includes only active material particles from which surface cracks of the active material particles have been removed from the active material particles included in the reference active material image. In addition, the reference crack binary image may be a binary image that includes only surface cracks of the active material particles included in the reference active material image.

[0103] The sizes of the reference active material images (271, 272, 273, 274, 275, 276), reference active material binary images (271-1, 272-1, 273-1, 274-1, 275-1, 276-1) and reference crack binary images (271-2, 272-2, 273-2, 274-2, 275-2, 276-2) included in the learning data may be the same.

[0104] In one embodiment, the plurality of reference active material images (271, 272, 273, 274, 275, 276) may include one or more first reference active material images and one or more second reference active material images.

[0105] In one embodiment, the reference active material binary images (271-1, 272-1, 273-1, 274-1, 275-1, 276-1) and the reference crack binary images (271-2, 272-2, 273-2, 274-2, 275-2, 276-2) may include one or more first reference binary images and one or more second reference binary images.

[0106] In one embodiment, the first reference binary images can be generated by applying a fifth image processing algorithm to the first reference active material image. Here, the fifth image processing algorithm can be an algorithm that combines a mean shift filter, a boundary extraction algorithm, a boundary removal algorithm, a binarization algorithm, or a combination thereof with an active material particle extraction algorithm that extracts active material particles from which surface cracks of the active material have been removed, or a crack extraction algorithm that extracts surface cracks of the active material particles. For example, the first reference binary images included in the reference active material binary images (271-1, 272-1, 273-1, 274-1, 275-1, 276-1) can be generated by applying the fifth image processing algorithm that combines a binarization algorithm and an active material particle extraction algorithm to the first reference active material images. As another example, the first reference binary images included in the reference crack binary images (271-2, 272-2, 273-2, 274-2, 275-2, 276-2) can be generated by applying a fifth image processing algorithm that combines a binarization algorithm and a crack extraction algorithm to the first reference active material images.

[0107] In one embodiment, the second reference binary images may be images transformed from the first reference binary images. In one embodiment, the second reference binary images may be generated by applying a first image processing algorithm specified to the first reference binary images. Here, the first image processing algorithm may include rotation, tilt, share, brightness adjustment, contrast adjustment, zoom in, zoom out, or a combination thereof.

[0108] In one embodiment, the first reference binary images may be binary images from which active materials or active material surface cracks are extracted from the first reference active material images, and the second reference binary images may be binary images from which active materials or active material surface cracks are extracted from the second reference active material images.

[0109] For example, the reference active material binary image (271-1) may be a binary image in which the active material is extracted from the reference active material image (271), and the reference active material binary image (275-1) may be a binary image in which the active material is extracted from the reference active material image (275). The reference crack binary image (271-2) may be a binary image in which the active material surface crack is extracted from the reference active material image (271), and the reference crack binary image (275-2) may be a binary image in which the active material surface crack is extracted from the reference active material image (275).

[0110] Accordingly, the plurality of reference active material images (271, 272, 273, 274, 275, 276) and the plurality of reference binary images (271-1, 272-1, 273-1, 274-1, 275-1, 276-1, 271-2, 272-2, 273-2, 274-2, 275-2, 276-2) can be classified into image sets with corresponding images. For example, the reference active material image (271) and the reference active material binary image (271-1) can be classified into one image set. As another example, the reference active material image (271) and the reference crack binary image (271-2) can be classified into one image set.

[0111] In one embodiment, the training data may be used to train an artificial intelligence model (145). In one embodiment, the training data may be used to train the artificial intelligence model (145) for a specified number of epochs. Here, the specified number of epochs may be determined between 100 and 10,000. For example, the specified number of epochs may be 3,000.

[0112] In one embodiment, the training data may be divided into mini-batches divided by a batch size. Here, the batch size may be determined between 1 and 512. For example, the batch size may be 3. When the batch size is 3, each of three sets of images (i.e., three reference active material images and three reference binary images) may be divided into a mini-batch.

[0113] For example, reference active material images (271, 272, 273) and reference active material binary images (271-1, 272-1, 273-1) may constitute a first mini-batch, and reference active material images (274, 275, 276) and reference active material binary images (274-1, 275-1, 276-1) may constitute a second mini-batch. As another example, the reference active material images (271, 272, 273) and the reference crack binary images (271-2, 272-2, 273-2) may constitute a first mini-batch, and the reference active material images (274, 275, 276) and the reference crack binary images (274-2, 275-2, 276-2) may constitute a second mini-batch.

[0114] artificial intelligence model

[0115] FIGS. 3A to 3D illustrate artificial intelligence models (145) according to various embodiments of the present disclosure. Specifically, FIG. 3A illustrates an artificial intelligence model according to a first embodiment, FIG. 3B illustrates an artificial intelligence model according to a second embodiment, FIG. 3C illustrates an artificial intelligence model according to a third embodiment, and FIG. 3D illustrates an artificial intelligence model according to a fourth embodiment.

[0116] In one embodiment, the artificial intelligence model (145) may be a model based on a convolutional neural network (CNN) or U-NET. In one embodiment, the artificial intelligence model (145) may be a model trained to convert an active material image into a binary image.

[0117] Figure 3a shows an example of training an artificial intelligence model (145) through a mini-batch consisting of reference active material images (381, 382, ​​383) and reference binary images (381-1, 382-1, 383-1).

[0118] Referring to FIG. 3A, the artificial intelligence model (145) may include a plurality of layers (310, 320, 330, 340, 350, 360, 370). The plurality of layers (310, 320, 330, 340, 350, 360, 370) may be sequentially connected. The input of the sequentially connected layers may be the output of the layer that is connected immediately before, and the output of the sequentially connected layers may be the input of the layer that is connected immediately after. For example, the output of the layer (310) may be the input of the layer (320).

[0119] At least two layers (e.g., layers (310, 370), layers (320, 360), or layers (330, 350)) among the plurality of layers (310, 320, 330, 340, 350, 360, 370) may be connected by a skip connection (315, 325, 335). In one embodiment, the skip connection (315, 325, 335) may connect a layer included in an encoding area (e.g., layers (310, 320, 330)) and a layer included in a decoding area (e.g., layers (350, 360, 370)) among the plurality of layers (310, 320, 330, 340, 350, 360, 370). Here, skip connections (315, 325, 335) may mean connections between layers for inputting the outputs of layers (310, 320, 330) to layers (350, 360, 370). For example, the outputs of layer (340) and layer (330) may be input to layer (350) through skip connections (335).

[0120] In one embodiment, the plurality of layers (310, 320, 330, 340, 350, 360, 370) may include an input layer, a batch normalization layer, a convolution layer, an activation layer, a max pooling layer, an upsampling layer, a concatenation layer, or a combination thereof. For example, a layer included in an encoding region (e.g., layers (310, 320, 330)) may have a structure in which a convolution layer, a batch normalization layer, an activation layer, and a max pooling layer are sequentially connected. As another example, a layer included in a decoding region (e.g., layers (350, 360, 370)) may have a structure in which a convolution layer, a batch normalization layer, an activation layer, and an upsampling layer are sequentially connected.

[0121] In one embodiment, the input layer may be a layer that receives bit values ​​included in the reference active material images (381, 383, 385). In one embodiment, the input layer may obtain an input composed of bit values ​​equal to the number of horizontal pixels × the number of vertical pixels × the number of channels (or depth). For example, when the reference active material image has a size of N × N, the input layer may receive N × N × 1 bit values. Here, N may be an integer between 32 and 4096 (e.g., 256). Depending on the embodiment, the number of channels may also be referred to as depth.

[0122] In one embodiment, the batch normalization layer may be a layer for normalizing the output values ​​of the previous layer on a batch basis. According to an embodiment, the batch normalization layer may be positioned immediately before the activation layer. Here, the batch may include reference active material images input during one iteration to train the artificial intelligence model (145). For example, when the batch size is 3, the number of reference active material images input to the artificial intelligence model (145) during one iteration may be 3. Here, the batch size may be an integer between 1 and 512 (e.g., 3).

[0123] In one embodiment, the convolution layer may be a layer for obtaining an output by performing a convolution operation on an input and a filter of a specified size. In one embodiment, the convolution layer included in the artificial intelligence model (145) assumes that the image size of the output (the number of horizontal pixels × the number of vertical pixels) is the same as the image size of the input (the number of horizontal pixels × the number of vertical pixels). In one embodiment, the number of channels of the output may change depending on the number of filters (or the depth of the filters) of the convolution layer. For example, when the number of filters is 2, the number of channels of the output may increase by twice compared to the number of channels of the input. In one embodiment, the number of channels of the output may change depending on the stride in the depth direction of the convolution layer. For example, when the stride in the depth direction is 2, the number of channels of the output may decrease by half compared to the number of channels of the input.

[0124] In one embodiment, an activation layer may be a layer that obtains an output by applying a specified activation function to an input. For example, the specified activation function may include a step, a sigmoid, a rectifier linear unit (ReLU), an exponential linear unit (ELU), a softmax, or a combination thereof.

[0125] In one embodiment, the max pooling layer may be a layer for obtaining an output by selecting the largest value in each pooling region of the input. In one embodiment, the image size of the output (i.e., the number of horizontal pixels and the number of vertical pixels) may change depending on the size of the pooling region of the max pooling layer. For example, when the size of the pooling region is 2 × 2, the image size of the output (e.g., 128 × 128) may be reduced by half compared to the image size of the input (e.g., 256 × 256). According to an embodiment, the artificial intelligence model (145) may include a pooling layer of a type other than the max pooling layer. For example, the pooling layer of a type other than the max pooling layer may include an average pooling layer.

[0126] In one embodiment, the upsampling layer may be a layer for increasing the resolution of an image. In one embodiment, the upsampling layer may be a layer for increasing the size of an output image compared to the size of an input image using a specified interpolation algorithm.

[0127] In one embodiment, a concatenation layer can concatenate two or more inputs and output an output. Here, the concatenation can be a connection in the depth direction. For example, if a first input is 128 × 128 × 128 and a second input is 128 × 128 × 128, the output of the concatenation layer can be 128 × 128 × 256. If another first input is 256 × 256 × 64 and a second input is 128 × 128 × 128, the output of the concatenation layer can be 256 × 256 × 128.

[0128] In one embodiment, the artificial intelligence model learning unit (141) can train the artificial intelligence model (145) through images (381, 382, ​​383, 381-1, 382-1, 383-1) included in the mini-batch.

[0129] In one embodiment, the artificial intelligence model learning unit (141) can sequentially input reference activation images (381, 382, ​​383) into the artificial intelligence model (145).

[0130] In one embodiment, the artificial intelligence model learning unit (141) can train the artificial intelligence model (145) so that the difference between the sequentially acquired output images of the artificial intelligence model (145) and the reference binary images (381-1, 382-1, 383-1) is less than or equal to a specified reference value.

[0131] For example, the artificial intelligence model learning unit (141) can calculate the difference between the output images and the reference binary images (381-1, 382-1, 383-1) based on a loss function (or cost function). Here, the loss function may include a mean absolute error function, a root mean square error, a mean square error function, or a binary cross entropy loss function. According to an embodiment, the binary cross entropy loss function may be a function to which a weight between 1 and 1000 is applied for one of two colors (or classes) (e.g., white among black and white).

[0132] Thereafter, the artificial intelligence model learning unit (141) can adjust the weights of the artificial intelligence model (145) so that the difference is less than or equal to a specified reference value (or has a minimum value). In one embodiment, the artificial intelligence model learning unit (141) can adjust the weights of the artificial intelligence model (145) so that the difference is less than or equal to a specified reference value (or has a minimum value) through an algorithm based on gradient descent (e.g., Adam, SGD, Momentum). Here, the adjusted weights can be determined according to a learning rate. The learning rate can be determined between 0.00000001 and 0.1. For example, the learning rate can be 0.0001.

[0133] Thereafter, the artificial intelligence model learning unit (141) can train the artificial intelligence model (145) through the images included in the next mini-batch. Afterwards, when the artificial intelligence model (145) is trained through the images included in all mini-batches, the artificial intelligence model learning unit (141) can train the artificial intelligence model (145) again through the images included in the mini-batches according to the specified number of epochs.

[0134] FIG. 3b shows an example of training an artificial intelligence model (145) through mini-batches consisting of reference active material images (384, 385, 386) and reference binary images (384-1, 385-1, 386-1, 384-2, 385-2, 386-2, 384-3, 385-3, 386-3). Here, the reference binary images (384-1, 385-1, 386-1, 384-2, 385-2, 386-2, 384-3, 385-3, 386-3) may include first type reference binary images (384-1, 385-1, 386-1) corresponding to the first particle type, second type reference binary images (384-2, 385-2, 386-2) corresponding to the second particle type, and third type reference binary images (384-3, 385-3, 386-3) corresponding to the third particle type.

[0135] The contents of the artificial intelligence model (145) of FIG. 3a described above can be equally applied to the artificial intelligence model (145) of FIG. 3b, and therefore, descriptions of overlapping contents may be omitted. In the following, only the embodiments added to the artificial intelligence model (145) of FIG. 3b will be described.

[0136] In one embodiment, the artificial intelligence model (145) can generate and output a number of images equal to the number of particle types from the input image. Here, the output images can each correspond one-to-one to each of the particle types. To this end, the number of output channels of the artificial intelligence model (145) can be N1 × N2, which is the product of N1, the number of input channels of the artificial intelligence model (145), (where N1 is a natural number), and N2, the number of particle types, (where N2 is a natural number).

[0137] For example, when one active material image is input to the artificial intelligence model (145), the artificial intelligence model (145) can generate and output three binary images, which are the number of particle types, and the output binary images can each correspond one-to-one to each of the three particle types.

[0138] In one embodiment, the artificial intelligence model (145) may include a convolution layer such that the number of output channels is N1 × N2. In this case, the convolution layer may be included in the decoding area of ​​the artificial intelligence model (145), and more specifically, may be configured as the last layer of the decoding area. For example, the convolution layer may be positioned at the rear of the layer (370) to receive the output image of the layer (370) as input. However, the location of the convolution layer is not limited thereto, and may be positioned at various locations.

[0139] In one embodiment, the convolution layer may include a plurality of filters that filter the input image according to particle types based on shape information of active material particles included in the input image. Here, the number of the plurality of filters may be N2, which is the number of particle types, and each of the plurality of filters may correspond one-to-one to each of the particle types.

[0140] Multiple filters included in the convolution layer can generate output images corresponding to each particle type on a one-to-one basis based on the input image. Here, the number of output images can be N2, which is the number of particle types.

[0141] For example, a first filter included in a convolution layer corresponds to a first particle type and can generate an output image that includes only active material particles corresponding to the first particle type among the active material particles included in the input image. In other words, the first filter can filter out active material particles corresponding to at least one particle type other than the first particle type in the input image.

[0142] In one embodiment, the artificial intelligence model learning unit (141) can train the artificial intelligence model (145) using images (384, 385, 386, 384-1, 385-1, 386-1, 384-2, 385-2, 386-2, 384-3, 385-3, 386-3) that are learning data included in a mini-batch.

[0143] In one embodiment, the artificial intelligence model learning unit (141) may train the artificial intelligence model (145) to generate binary images classified according to particle type based on shape information of active material particles included in an active material image input to the artificial intelligence model (145). For example, the binary images may include a first type binary image corresponding to a first particle type, a second type binary image corresponding to a second particle type, and a third type binary image corresponding to a third particle type.

[0144] In one embodiment, the artificial intelligence model learning unit (141) can sequentially input reference activation images (384, 385, 386) into the artificial intelligence model (145).

[0145] In one embodiment, the artificial intelligence model learning unit (141) can train the artificial intelligence model (145) so that the difference between sequentially acquired output images and reference binary images (384-1, 385-1, 386-1, 384-2, 385-2, 386-2, 384-3, 385-3, 386-3) is less than or equal to a specified reference value. Specifically, the artificial intelligence model learning unit (141) can train the artificial intelligence model (145) so that the difference between output images corresponding to the same particle type and reference binary images is less than or equal to a specified reference value.

[0146] For example, the artificial intelligence model learning unit (141) can train the artificial intelligence model (145) so that the difference between the output images corresponding to the first particle type among the output images and the first type reference binary images (384-1, 385-1, 386-1) is less than or equal to a specified reference value. Here, the output images corresponding to the first particle type may be images output from an output channel corresponding to the first particle type among the output channels of the artificial intelligence model (145).

[0147] As another example, the artificial intelligence model learning unit (141) can train the artificial intelligence model (145) so that the difference between the output images corresponding to the second particle type among the output images and the second type reference binary images (384-2, 385-2, 386-2) is less than or equal to a specified reference value. Here, the output images corresponding to the second particle type may be images output from an output channel corresponding to the second particle type among the output channels of the artificial intelligence model (145).

[0148] As another example, the artificial intelligence model learning unit (141) can train the artificial intelligence model (145) so that the difference between the output images corresponding to the third particle type among the output images and the third type reference binary images (384-3, 385-3, 386-3) is less than or equal to a specified reference value. Here, the output images corresponding to the third particle type may be images output from an output channel corresponding to the third particle type among the output channels of the artificial intelligence model (145).

[0149] FIG. 3c shows an example of training an artificial intelligence model (145) through a mini-batch consisting of reference active material images (387, 388, 389) and reference binary images (387-1, 388-1, 389-1, 387-2, 388-2, 389-2). Here, the reference binary images (387-1, 388-1, 389-1, 387-2, 388-2, 389-2) may include reference active material binary images (387-1, 388-1, 389-1) corresponding to the active material and reference impurity binary images (387-2, 388-2, 389-2) corresponding to impurities present on the surface of the active material.

[0150] The description of the artificial intelligence model (145) of FIG. 3a described above can be equally applied to the artificial intelligence model (145) of FIG. 3c, and therefore, description of overlapping content may be omitted. In the following, only the additional embodiments of the artificial intelligence model (145) of FIG. 3c will be described.

[0151] In one embodiment, the artificial intelligence model (145) can generate and output two images from the input image. Here, the two output images can correspond one-to-one to the active material from which the impurities have been removed and the impurities, respectively. To this end, the number of output channels of the artificial intelligence model (145) can be N × 2, which is twice the number of input channels of the artificial intelligence model (145), N (where N is a natural number).

[0152] In one embodiment, the artificial intelligence model (145) may include a convolution layer that allows the number of output channels to be N × 2. In this case, the convolution layer may be included in the decoding area of ​​the artificial intelligence model (145), and more specifically, may be configured as the last layer of the decoding area. For example, the convolution layer may be positioned at the rear of the layer (370) to receive the output image of the layer (370) as input. However, the location of the convolution layer is not limited thereto, and may be positioned at various locations.

[0153] In one embodiment, the convolution layer may include a plurality of filters that filter the input image by distinguishing between active materials and impurities based on shape information of objects in the input image. Here, the number of filters may be 2, and each of the filters may correspond one-to-one to the active material from which the impurities have been removed and to the impurities, respectively.

[0154] A plurality of filters included in the convolution layer can generate output images corresponding one-to-one to each of the active material and impurities from which the impurities have been removed based on the input image. Here, the number of output images can be 2.

[0155] For example, a first filter included in a convolution layer corresponds to an active material from which impurities have been removed, and can generate an output image that includes only active material particles from which impurities have been removed among objects included in an input image. In other words, the first filter can filter out impurities from an input image.

[0156] As another example, a second filter included in a convolutional layer corresponds to impurities and can generate an output image that contains only impurities among the objects included in the input image. In other words, the second filter can filter out active materials from the input image.

[0157] In one embodiment, the artificial intelligence model learning unit (141) can train the artificial intelligence model (145) through images (387, 388, 389, 387-1, 388-1, 389-1, 387-2, 388-2, 389-2) included in the mini-batch.

[0158] In one embodiment, the artificial intelligence model learning unit (141) can sequentially input reference activation images (387, 388, 389) into the artificial intelligence model (145).

[0159] In one embodiment, the artificial intelligence model learning unit (141) may train the artificial intelligence model (145) to generate binary images in which the active material and impurities are separated based on shape information of objects included in the active material image input to the artificial intelligence model (145). For example, the binary images may include an active material binary image corresponding to the active material from which the impurities have been removed and an impurity binary image corresponding to the impurities.

[0160] In one embodiment, the artificial intelligence model learning unit (141) can train the artificial intelligence model (145) so that the difference between sequentially acquired output images and reference binary images (387-1, 388-1, 389-1, 387-2, 388-2, 389-2) is less than or equal to a specified reference value. Specifically, the artificial intelligence model learning unit (141) can train the artificial intelligence model (145) so that the difference between output images corresponding to the same object type (e.g., active material or impurities from which impurities have been removed) and reference binary images is less than or equal to a specified reference value.

[0161] For example, the artificial intelligence model learning unit (141) can train the artificial intelligence model (145) so that the difference between the output images corresponding to the active material among the output images and the reference active material binary images (387-1, 388-1, 389-1) is less than or equal to a specified reference value. Here, the output images corresponding to the active material may be images output from an output channel corresponding to the active material among the output channels of the artificial intelligence model (145).

[0162] As another example, the artificial intelligence model learning unit (141) can train the artificial intelligence model (145) so that the difference between the output images corresponding to the impurity among the output images and the reference impurity binary images (387-2, 388-2, 389-2) is less than or equal to a specified reference value. Here, the output images corresponding to the impurity may be images output from the output channel corresponding to the impurity among the output channels of the artificial intelligence model (145).

[0163] FIG. 3d shows an example of training an artificial intelligence model (145) through mini-batches consisting of reference active material images (390, 391, 392) and reference binary images (390-1, 391-1, 392-1, 390-2, 391-2, 392-2). Here, the reference binary images (390-1, 391-1, 392-1, 390-2, 391-2, 392-2) may include reference active material binary images (390-1, 391-1, 392-1) corresponding to active materials and reference crack binary images (390-2, 391-2, 392-2) corresponding to active material surface cracks.

[0164] The contents of the artificial intelligence model (145) of FIG. 3a described above can be equally applied to the artificial intelligence model (145) of FIG. 3d, and therefore, descriptions of overlapping contents may be omitted. In the following, only the embodiments added to the artificial intelligence model (145) of FIG. 3d will be described.

[0165] In one embodiment, the artificial intelligence model (145) can generate and output two images from the input image. Here, the two output images can correspond one-to-one to the active material from which the surface cracks of the active material have been removed and the active material surface cracks, respectively. To this end, the number of output channels of the artificial intelligence model (145) can be N χ 2, which is twice the number of input channels of the artificial intelligence model (145), N (where N is a natural number).

[0166] In one embodiment, the artificial intelligence model (145) may include a convolution layer such that the number of output channels is N X 2. In this case, the convolution layer may be included in the decoding area of ​​the artificial intelligence model (145), and more specifically, may be configured as the last layer of the decoding area. For example, the convolution layer may be positioned at the rear of the layer (370) to receive the output image of the layer (370) as input. However, the position of the convolution layer is not limited thereto, and may be positioned at various positions.

[0167] In one embodiment, the convolution layer may include a plurality of filters that filter the input image by distinguishing between active material and surface cracks of the active material based on shape information of objects in the input image. Here, the number of the plurality of filters may be 2, and each of the plurality of filters may correspond one-to-one to the active material from which the surface cracks of the active material have been removed and to each of the surface cracks of the active material.

[0168] A plurality of filters included in the convolution layer can generate output images corresponding one-to-one to each of the active material with cracks removed from the surface of the active material and the active material surface cracks, based on the input image. Here, the number of output images can be 2.

[0169] For example, a first filter included in a convolution layer corresponds to an active material with cracks removed from the active material surface, and can generate an output image that includes only active material particles with cracks removed from the active material surface among objects included in the input image. In other words, the first filter can filter out cracks from the active material surface from the input image.

[0170] As another example, a second filter included in a convolutional layer may correspond to cracks on the surface of an active material and may generate an output image that includes only cracks on the surface of an active material among objects included in an input image. In other words, the second filter may filter out active materials from the input image.

[0171] In one embodiment, the artificial intelligence model learning unit (141) can train the artificial intelligence model (145) through images (390, 391, 392, 390-1, 391-1, 392-1, 390-2, 391-2, 392-2) included in the mini-batch.

[0172] In one embodiment, the artificial intelligence model learning unit (141) can sequentially input reference activation images (390, 391, 392) into the artificial intelligence model (145).

[0173] In one embodiment, the artificial intelligence model learning unit (141) may train the artificial intelligence model (145) to generate binary images in which the active material and the active material surface cracks are separated based on shape information of objects included in the active material image input to the artificial intelligence model (145). For example, the binary images may include an active material binary image corresponding to the active material from which the active material surface cracks have been removed and a crack binary image corresponding to the active material surface cracks.

[0174] In one embodiment, the artificial intelligence model learning unit (141) can train the artificial intelligence model (145) so that the difference between sequentially acquired output images and reference binary images (390-1, 391-1, 392-1, 390-2, 391-2, 392-2) is less than or equal to a specified reference value. Specifically, the artificial intelligence model learning unit (141) can train the artificial intelligence model (145) so that the difference between output images corresponding to the same object type (e.g., active material with cracks removed from the surface of the active material or active material surface cracks) and reference binary images is less than or equal to a specified reference value.

[0175] For example, the artificial intelligence model learning unit (141) can train the artificial intelligence model (145) so that the difference between the output images corresponding to the active material among the output images and the reference active material binary images (390-1, 391-1, 392-1) is less than or equal to a specified reference value. Here, the output images corresponding to the active material may be images output from an output channel corresponding to the active material among the output channels of the artificial intelligence model (145).

[0176] As another example, the artificial intelligence model learning unit (141) can train the artificial intelligence model (145) so that the difference between the output images corresponding to the active material surface cracks among the output images and the reference crack binary images (390-2, 391-2, 392-2) is less than or equal to a specified reference value. Here, the output images corresponding to the active material surface cracks may be images output from an output channel corresponding to the active material surface cracks among the output channels of the artificial intelligence model (145).

[0177] In FIGS. 3A to 3D, the number of layers (310, 320, 330, 340, 350, 360, 370) is illustrated as being 7, but this is merely an example, and the number of layers is not limited to 7. For example, the number of layers included in the artificial intelligence model (145) may be 32.

[0178] When the number of layers included in the artificial intelligence model (145) is 32, the artificial intelligence model (145) may have a structure in which an input layer, a first batch normalization layer, a first 2D convolution layer, and a second 2D convolution layer are sequentially connected. The input of the sequentially connected layers may be the output of the layer connected immediately before, and the output of the sequentially connected layers may be the input of the layer connected immediately after. Here, the input of the input layer may be 256 × 256 × 1, and the output may be 256 × 256 × 1. The output of the first batch normalization layer may be 256 × 256 × 1. The output of the first 2D convolution layer may be 256 × 256 × 64. The output of the second 2D convolution layer may be 256 × 256 × 64.

[0179] Following the second 2D convolution layer, the artificial intelligence model (145) may have a structure in which a second batch normalization layer, a first activation layer, and a first max pooling layer are sequentially connected. Here, the output of the second batch normalization layer may be 256 × 256 × 64. The output of the first activation layer may be 256 × 256 × 64. The output of the first max pooling layer may be 128 × 128 × 64.

[0180] Following the first max pooling layer, the artificial intelligence model (145) may have a structure in which a third 2D convolution layer, a fourth 2D convolution layer, a third batch normalization layer, a second activation layer, and a second max pooling layer are sequentially connected. Here, the output of the third 2D convolution layer may be 128 × 128 × 128. The output of the fourth 2D convolution layer may be 128 × 128 × 128. The output of the third batch normalization layer may be 128 × 128 × 128. The output of the second activation layer may be 128 × 128 × 128. The output of the second max pooling layer may be 64 × 64 × 128.

[0181] Following the second max pooling layer, the artificial intelligence model (145) may have a structure in which a fifth 2D convolution layer, a sixth 2D convolution layer, a fourth batch normalization layer, a third activation layer, and a first up-sampling layer are sequentially connected. Here, the output of the fifth 2D convolution layer may be 64 × 64 × 256. The output of the sixth 2D convolution layer may be 64 × 64 × 256. The output of the fourth batch normalization layer may be 64 × 64 × 256. The output of the third activation layer may be 64 × 64 × 256. The output of the first up-sampling layer may be 128 × 128 × 256.

[0182] Following the first up-sampling layer, the artificial intelligence model (145) may have a structure in which a seventh 2D convolution layer, a first connection layer, an eighth 2D convolution layer, a ninth 2D convolution layer, a fifth batch normalization layer, a fourth activation layer, and a second up-sampling layer are sequentially connected. Among the two inputs of the first connection layer, the first input may be the output of the seventh 2D convolution layer, and the second input may be the output of the fourth 2D convolution layer. Here, the output of the seventh 2D convolution layer may be 128 × 128 × 128. The output of the first connection layer may be 128 × 128 × 256. The output of the eighth 2D convolution layer may be 128 × 128 × 128. The output of the ninth 2D convolution layer may be 128 × 128 × 128. The output of the fifth batch normalization layer can be 128 × 128 × 128. The output of the fourth activation layer can be 128 × 128 × 128. The output of the second upsampling layer can be 256 × 256 × 128.

[0183] Following the second up-sampling layer, the artificial intelligence model (145) may have a structure in which a tenth 2D convolution layer, a second connection layer, an eleventh 2D convolution layer, a twelfth 2D convolution layer, a sixth batch normalization layer, a fifth activation layer, a thirteenth 2D convolution layer, and a fourteenth 2D convolution layer are sequentially connected. Among the two inputs of the second connection layer, the first input may be the output of the tenth 2D convolution layer, and the second input may be the output of the second 2D convolution layer. Here, the output of the tenth 2D convolution layer may be 256 × 256 × 64. The output of the second connection layer may be 256 × 256 × 128. The output of the eleventh 2D convolution layer may be 256 × 256 × 64. The output of the 12th 2D convolution layer may be 256 × 256 × 64. The output of the 6th batch normalization layer may be 256 × 256 × 64. The output of the 5th activation layer may be 256 × 256 × 64. The output of the 13th 2D convolution layer may be 256 × 256 × 2. The output of the 14th 2D convolution layer may be 256 × 256 × 2.

[0184] Image processing

[0185] FIGS. 4A to 4D illustrate images processed by an image processing device (101) according to various embodiments of the present disclosure.

[0186] FIG. 4a illustrates images (410, 412, 420) obtained by image processing of an image processing device (101) according to a first embodiment of the present disclosure.

[0187] In one embodiment, the image acquisition unit (150) can acquire an active material image (410) for the active material (411). In one embodiment, the image acquisition unit (150) can acquire the active material image (410) through the image acquisition device (103). In FIG. 4A, only one active material is indicated by a symbol, but referring to FIG. 4A, it can be seen that other active materials are present in the active material image (410).

[0188] In one embodiment, the image generation unit (160) can generate a binary image (412). In one embodiment, the image generation unit (160) can generate a binary image (412) by inputting an active material image (410) into an artificial intelligence model (145).

[0189] In one embodiment, the object identification unit (170) can identify multiple objects (e.g., 413) included in the binary image (412). Here, the object (413) included in the binary image (412) can correspond to the active material (411) in the active material image (410). In one embodiment, the object (413) can be composed of an area having a designated pixel value (e.g., a value representing white color). In one embodiment, the object (413) can be distinguished by an area having another designated value (e.g., a value representing black color). In FIG. 4A, only one object is indicated by a symbol, but referring to FIG. 4A, it can be seen that other objects exist in the binary image (412).

[0190] In one embodiment, the image segmentation unit (180) may segment a plurality of active materials (e.g., 411) included in the active material image (410) based on a plurality of objects (e.g., 413) identified by the object identification unit (170) to obtain a segmentation image (420). In one embodiment, the image segmentation unit (180) may segment a plurality of active materials (e.g., 411) included in the active material image (410) based on a watershed algorithm.

[0191] FIG. 4b illustrates images (430, 431, 433, 435, 440) processed by an image processing device (101) according to a second embodiment of the present disclosure.

[0192] In one embodiment, the image acquisition unit (150) can acquire active material images (430) for a plurality of active materials. In one embodiment, the image acquisition unit (150) can acquire the active material images (430) through the image acquisition device (103).

[0193] In one embodiment, the image generating unit (160) can generate binary images (431, 433, and / or 435) classified according to particle type. In one embodiment, the image generating unit (160) can generate binary images (431, 433, and / or 435) by inputting the active material image (430) into the artificial intelligence model (145). Here, each binary image (431, 433, or 435) can correspond one-to-one to each particle type. For example, the binary image (431) can correspond to a first particle type, the binary image (433) can correspond to a second particle type, and the binary image (435) can correspond to a third particle type.

[0194] In one embodiment, the object identification unit (170) can identify a plurality of objects included in a binary image (431, 433, and / or 435). Here, the plurality of objects included in a specific binary image (431, 433, or 435) can correspond to a plurality of active materials corresponding to a specific particle type in the active material image (430). In one embodiment, the plurality of objects can be composed of areas having a specified pixel value (e.g., a value representing a white color). In one embodiment, the plurality of objects can be distinguished by areas having another specified value (e.g., a value representing a black color).

[0195] In one embodiment, the object identification unit (170) can identify particle types of a plurality of objects included in the binary image (431, 433, and / or 435) based on an output channel of the artificial intelligence model (145) from which the binary image (431, 433, and / or 435) is output. For example, the object identification unit (170) can identify particle types of a plurality of objects included in the binary image (431) as a first particle type based on a first output channel of the artificial intelligence model (145) from which the binary image (431) is output. As another example, the object identification unit (170) can identify particle types of a plurality of objects included in the binary image (433) as a second particle type based on a second output channel of the artificial intelligence model (145) from which the binary image (433) is output.

[0196] In one embodiment, the image segmentation unit (180) can segment a plurality of active materials included in the active material image (430) based on a plurality of objects identified by the object identification unit (170) to obtain a segmentation image (440). In one embodiment, the image segmentation unit (180) can segment a plurality of active materials included in the active material image (430) based on a watershed algorithm.

[0197] In one embodiment, the image segmentation unit (180) may obtain a segmentation image (440) including label information that enables distinguishing the particle types of each of the plurality of active materials based on the particle types identified by the object identification unit (170). For example, the segmentation image (440) may display the plurality of active materials in different ways (e.g., colors) so as to enable distinguishing the active materials into an active material (441) corresponding to a first particle type, an active material (443) corresponding to a second particle type, and an active material (445) corresponding to a third particle type.

[0198] FIG. 4c illustrates images (450, 451, 453, 460) obtained by image processing of an image processing device (101) according to a third embodiment of the present disclosure.

[0199] In one embodiment, the image acquisition unit (150) can acquire active material images (450) for a plurality of active materials. In one embodiment, the image acquisition unit (150) can acquire the active material images (450) through the image acquisition device (103).

[0200] In one embodiment, the image generation unit (160) can generate binary images (451, 453) in which the active material and impurities are separated. In one embodiment, the image generation unit (160) can generate binary images (451, 453) in which the active material and impurities are separated by inputting the active material image (450) into the artificial intelligence model (145). Here, each binary image (451 or 453) can correspond one-to-one to the active material or the impurities, respectively. For example, the binary image (451) can correspond to the active material, and the binary image (453) can correspond to the impurities.

[0201] In one embodiment, the object identification unit (170) can identify a plurality of objects included in a binary image (451, 453). Here, the plurality of objects included in a specific binary image (451 or 453) can correspond to an active material or an impurity in the active material image (450). In one embodiment, the plurality of objects can be composed of areas having a specified pixel value (e.g., a value representing a white color). In one embodiment, the plurality of objects can be distinguished by areas having another specified value (e.g., a value representing a black color).

[0202] In one embodiment, the object identification unit (170) can identify the object types of the plurality of objects included in the binary images (451, 453) as active materials or impurities based on the output channels of the artificial intelligence model (145) from which the binary images (451, 453) are output. For example, the object identification unit (170) can identify the object types of the plurality of objects included in the binary image (451) as active materials based on the output channels of the artificial intelligence model (145) from which the binary image (451) is output. As another example, the object identification unit (170) can identify the object types of the plurality of objects included in the binary image (453) as impurities based on the output channels of the artificial intelligence model (145) from which the binary image (453) is output.

[0203] In one embodiment, the image segmentation unit (180) can segment a plurality of active materials included in the active material image (450) based on a plurality of objects identified by the object identification unit (170) to obtain a segmentation image (460). In one embodiment, the image segmentation unit (180) can segment a plurality of active materials included in the active material image (450) based on a watershed algorithm.

[0204] In one embodiment, the image segmentation unit (180) may obtain a segmentation image (460) including label information that enables each of a plurality of objects to be distinguished as an active material or an impurity based on the object type identified by the object identification unit (170). For example, the segmentation image (460) may display the plurality of objects in different ways (e.g., colors) so that the objects (461) corresponding to the active material and the objects (463) corresponding to the impurity can be distinguished.

[0205] FIG. 4d illustrates images (470, 471, 473, 480) obtained by image processing of an image processing device (101) according to a fourth embodiment of the present disclosure.

[0206] In one embodiment, the image acquisition unit (150) can acquire active material images (470) for a plurality of active materials. In one embodiment, the image acquisition unit (150) can acquire the active material images (470) through the image acquisition device (103).

[0207] In one embodiment, the image generation unit (160) can generate binary images (471, 473) in which the active material and the active material surface cracks are separated. In one embodiment, the image generation unit (160) can generate binary images (471, 473) in which the active material and the active material surface cracks are separated by inputting the active material image (470) into the artificial intelligence model (147). Here, each binary image (471 or 473) can correspond one-to-one to the active material or the active material surface cracks, respectively. For example, the binary image (471) can correspond to the active material, and the binary image (473) can correspond to the active material surface cracks.

[0208] In one embodiment, the object identification unit (170) can identify a plurality of objects included in a binary image (471, 473). Here, the plurality of objects included in a specific binary image (471 or 473) can correspond to an active material or a crack on the surface of the active material in the active material image (470). In one embodiment, the plurality of objects can be composed of areas having a specified pixel value (e.g., a value representing a white color). In one embodiment, the plurality of objects can be distinguished by areas having another specified value (e.g., a value representing a black color).

[0209] In one embodiment, the object identification unit (170) can identify the object types of the plurality of objects included in the binary images (471, 473) as active materials or active material surface cracks based on the output channels of the artificial intelligence model (147) from which the binary images (471, 473) are output. For example, the object identification unit (170) can identify the object types of the plurality of objects included in the binary image (471) as active materials based on the output channels of the artificial intelligence model (147) from which the binary image (471) is output. As another example, the object identification unit (170) can identify the object types of the plurality of objects included in the binary image (473) as active material surface cracks based on the output channels of the artificial intelligence model (147) from which the binary image (473) is output.

[0210] In one embodiment, the image segmentation unit (180) can segment a plurality of active materials included in the active material image (470) based on a plurality of objects identified by the object identification unit (170) to obtain a segmentation image (480). In one embodiment, the image segmentation unit (180) can segment a plurality of active materials included in the active material image (470) based on a watershed algorithm.

[0211] In one embodiment, the image segmentation unit (180) may obtain a segmentation image (480) including label information that enables each of a plurality of objects to be distinguished as an active material or an active material surface crack based on the object type identified by the object identification unit (170). For example, the segmentation image (480) may display the plurality of objects in different ways (e.g., colors) so that the objects (481) corresponding to the active material and the objects (483) corresponding to the active material surface crack can be distinguished.

[0212] Information extraction

[0213] In one embodiment, the information extraction unit (190) can extract information about objects included in a segmentation image (420, 440, 460, or 480).

[0214] In one embodiment, the information about the objects may include information about at least one of an average particle diameter, an average perimeter, an average sphericity, an average aspect ratio, an average convexity, an average solidity, or a distribution thereof of the active material (e.g., the active material (421) of FIG. 4A). Furthermore, the information about the objects may also include information about the individual particle diameter, perimeter, sphericity, aspect ratio, convexity, and / or solidity of each active material. Furthermore, the information about the objects may also include information about a quantile value (e.g., 1 to 99%, or D5, D50, D95, etc.) of each active material.

[0215] In one embodiment, information about objects may include information about particle types of the active material. In one embodiment, the information extraction unit (190) may classify the particle types of each object based on shape information of the objects (e.g., particle size, sphericity, aspect ratio, convexity, and / or hardness). For example, the information extraction unit (190) may input the shape information of the objects into a machine learning-based classification model to classify the particle types of each object. Here, the classification model may include a Gaussian mixture model, a K-menas clustering model, a support vector machine, a Gaussian naive Bayes model, or a combination thereof.

[0216] In one embodiment, the information about the objects may include information about at least one of a number ratio by particle type, an area ratio by particle type, or a volume ratio by particle type. For example, the information about the objects may include information about at least one of a number ratio, an area ratio, or a volume ratio of the active material (e.g., 441) corresponding to the first particle type relative to all active material particles (e.g., active material particles (441, 443, 445) of FIG. 4B).

[0217] In one embodiment, information about objects may include information about at least one of an area ratio of an active material (e.g., an active material (461) in FIG. 4c), a volume ratio of an active material, an area ratio of an impurity (e.g., an impurity (463) in FIG. 4c), or a volume ratio of an impurity.

[0218] In one embodiment, information about objects may include information about at least one of a number ratio, an area ratio, or a volume ratio of active materials having surface cracks (e.g., active materials (483) of FIG. 4d).

[0219] FIG. 5 is a flowchart illustrating an active material image processing method of an image processing device according to an embodiment of the present disclosure. FIG. 5 can be explained using the configurations of FIG. 1.

[0220] The embodiment illustrated in FIG. 5 is merely an example, and the order of steps according to various embodiments of the present invention may differ from that illustrated in FIG. 5, and some of the steps illustrated in FIG. 5 may be omitted, the order between steps may be changed, or steps may be merged. For example, operation 525 in FIG. 5 may be omitted.

[0221] Referring to FIG. 5, in operation 505, the image processing device (101) may acquire an active material image for the active material. In one embodiment, the image processing device (101) may acquire the active material image through the image acquisition device (103). In one embodiment, the image processing device (101) may generate, based on the active material image, at least one of a binary image in which the active material and impurities are separated, a binary image classified according to particle type, a binary image in which impurities or active material surface cracks are separated, or any combination thereof.

[0222] In operation 510, the image processing device (101) can generate a binary image by inputting the active material image acquired in operation 505 into the artificial intelligence model (145). Here, the artificial intelligence model (145) may be a model learned by the artificial intelligence model learning method according to FIG. 6.

[0223] In one embodiment, the image processing device (101) can generate binary images classified according to particle types. For example, the image processing device (101) can generate and output a number of images equal to the number of particle types from an input active material image using an artificial intelligence model (145). To this end, the number of output channels of the artificial intelligence model (145) may be N1 × N2, which is the product of N1, which is the number of input channels of the artificial intelligence model (145), (where N1 is a natural number), and N2, which is the number of particle types, (where N2 is a natural number).

[0224] In one embodiment, the image processing device (101) can generate a binary image in which the active material and impurities are separated. For example, the image processing device (101) can generate and output two images from an input active material image using an artificial intelligence model (145). Here, the two output images can correspond one-to-one to the active material from which the impurities have been removed and the impurities, respectively. To this end, the number of output channels of the artificial intelligence model (145) can be N × 2, which is twice the number of input channels of the artificial intelligence model (145), N (where N is a natural number).

[0225] In one embodiment, the image processing device (101) can generate a binary image in which the active material and the active material surface cracks are separated. For example, the image processing device (101) can generate and output two images from an input active material image using an artificial intelligence model (145). Here, the two output images can correspond one-to-one to the active material from which the active material surface cracks have been removed and the active material surface cracks, respectively. To this end, the number of output channels of the artificial intelligence model (145) can be N × 2, which is twice the number of input channels of the artificial intelligence model (145), N (where N is a natural number).

[0226] In operation 515, the image processing device (101) can identify a plurality of objects included in the binary image generated in operation 510.

[0227] In operation 520, the image processing device (101) may segment a plurality of active materials included in the active material image based on the plurality of objects identified in operation 515 to obtain a segmentation image. In one embodiment, the image processing device (101) may segment a plurality of active materials included in the active material image based on a watershed algorithm.

[0228] In operation 525, the image processing device (101) can extract information about a plurality of active substances or a plurality of objects included in the segmentation image acquired in operation 520.

[0229] FIG. 6 is a flowchart illustrating an artificial intelligence model learning method of an image processing device according to an embodiment of the present disclosure. FIG. 6 can be explained using the configurations of FIG. 1.

[0230] The embodiment illustrated in FIG. 6 is only one embodiment, and the order of steps according to various embodiments of the present invention may be different from that illustrated in FIG. 6, and some of the steps illustrated in FIG. 6 may be omitted, the order between steps may be changed, or steps may be merged.

[0231] Referring to FIG. 6, in operation 610, the image processing device (101) may set parameters of the artificial intelligence model (145). In one embodiment, the image processing device (101) may set parameters of the artificial intelligence model (145) based on initial parameter values.

[0232] In operation 615, the image processing device (101) may set i to 1. i may represent an index of an image set included in the learning data. Here, the image set may include a reference active material image and a reference binary image therefor.

[0233] In operation 620, the image processing device (101) can obtain an output image based on the i-th reference active material image. In one embodiment, the image processing device (101) can generate an output image by inputting the i-th reference active material image into the artificial intelligence model (145).

[0234] In operation 630, the image processing device (101) may determine whether the difference between the output image and the ith reference binary image is less than or equal to a reference. In one embodiment, the image processing device (101) may calculate the difference between the output image and the ith reference binary image based on a loss function (or cost function). Here, the loss function may include a mean absolute error function, a root mean square error, a mean square error function, or a binary cross entropy loss function. According to an embodiment, the binary cross entropy loss function may be a function to which a weight between 1 and 1000 is applied for one of two colors (or classes) (e.g., white among black and white).

[0235] As a result of the judgment in operation 630, if the difference is below the standard (yes judgment), the image processing device (101) can proceed to operation 650. As a result of the judgment in operation 630, if the difference exceeds the standard (no judgment), the image processing device (101) can proceed to operation 640.

[0236] In operation 640, the image processing device (101) may adjust the parameters of the artificial intelligence model (145). In one embodiment, the image processing device (101) may adjust the parameters of the artificial intelligence model (145) so that the difference is less than or equal to a specified reference value (or has a minimum value). In one embodiment, the image processing device (101) may adjust the parameters of the artificial intelligence model (145) so that the difference is less than or equal to a specified reference value (or has a minimum value) through an algorithm based on gradient descent (e.g., Adam, SGD, Momentum). Here, the adjusted weights may be determined according to a learning rate. The learning rate may be determined between 0.00000001 and 0.1. For example, the learning rate may be 0.0001.

[0237] In operation 650, the image processing device (101) may determine whether learning is complete. In one embodiment, if i represents the last of the image set, learning may be determined to be complete.

[0238] If the judgment result in operation 650 indicates that learning is complete (yes judgment), the image processing device (101) can terminate the operations according to FIG. 6. If the judgment result in operation 650 indicates that learning is not complete (no judgment), the image processing device (101) can proceed to operation 655.

[0239] In operation 655, the image processing device (101) may increase the value of i by 1. Thereafter, operation 620 may be performed again.

[0240] Hereinafter, an image processing device and its operation method according to a fifth embodiment of the present disclosure will be described with reference to FIGS. 7 to 9.

[0241] FIG. 7 is a block diagram of an image processing device (701) according to an embodiment of the present disclosure. Specifically, the image processing device (701) of FIG. 7 may correspond to the image processing device according to the fifth embodiment of the present disclosure.

[0242] In one embodiment, the connection (705) between the image processing device (701) and the image acquisition device (703) may be a communication connection via a wired and / or wireless network. In one embodiment, the wired network may be based on LAN communication or power line communication. In one embodiment, the wireless network may be based on a short-range communication network (e.g., Bluetooth, WiFi, or IrDA) or a long-range communication network (e.g., a cellular network, a 4G network, a 5G network).

[0243] In another embodiment, the connection (705) between the image processing device (701) and the image acquisition device (703) may be a connection via a device-to-device communication method (e.g., bus, GPIO, SPI, or MIPI).

[0244] In one embodiment, the image acquisition device (703) may be a microscope (e.g., a scanning electron microscope). In one embodiment, the image acquisition device (703) may be a device that acquires an image of the sample surface by scanning a focused electron beam on the sample surface and converting secondary electrons generated by the interaction between the electron beam and the sample into an image signal.

[0245] In one embodiment, the image acquisition device (703) can acquire images of an active material and / or a precursor. Hereinafter, the images of the active material and / or the precursor may be referred to as active material images. However, even if referred to as active material images, the present disclosure does not exclude images of precursors.

[0246] In one embodiment, the image acquisition device (703) can acquire an active material image for the active material. For example, the image acquisition device (703) can acquire an SEM image by scanning an electron beam onto the active material powder. In some embodiments, the SEM image can be replaced with an image based on TEM, OM, SIM, or FIB.

[0247] In one embodiment, the image acquisition device (703) can transmit an active material image of the active material to the image processing device (701). For example, the image acquisition device (703) can transmit an active material image of the active material to the image processing device (701) via a connection (705).

[0248] In one embodiment, the image processing device (701) may be a mobile device (e.g., a mobile phone, a laptop computer, a smart phone, a smart pad), a computer (e.g., a general-purpose computer, a special-purpose computer).

[0249] In one embodiment, the image processing device (701) may include a communication circuit (710), a memory (720), and a processor (730). According to an embodiment, the image processing device (701) illustrated in FIG. 7 may further include at least one component (e.g., a display, an input device, or an output device) other than the components illustrated in FIG. 7.

[0250] In one embodiment, the communication circuit (710) can establish a wired communication channel and / or a wireless communication channel between the image processing device (701) and / or the image acquisition device (703), and transmit and receive data to and from the image acquisition device (703) through the established communication channel.

[0251] In one embodiment, the memory (720) may include volatile memory and / or non-volatile memory.

[0252] In one embodiment, the memory (720) may store data used by at least one component (e.g., the processor (730)) of the image processing device (701). For example, the data may include a program (725) (or instructions related thereto), input data, or output data. In one embodiment, the instructions, when executed by the processor (730), may cause the image processing device (701) to perform operations defined by the instructions.

[0253] In one embodiment, the memory (720) may include a program (725) (e.g., an artificial intelligence model learning unit (741), an artificial intelligence model (745), an image acquisition unit (750), a first image segmentation unit (760), and / or a second image segmentation unit (770)).

[0254] In one embodiment, the processor (730) may include a central processing unit, an application processor, a graphics processing unit, a neural processing unit (NPU), an image signal processor, a sensor hub processor, or a communication processor.

[0255] In one embodiment, the processor (730) may execute a program (725) (e.g., an artificial intelligence model learning unit (741), an artificial intelligence model (745), an image acquisition unit (750), a first image segmentation unit (760), and / or a second image segmentation unit (770)) to control at least one other component (e.g., a hardware or software component) of an image processing device (701) connected to the processor (730) and perform various data processing or calculations.

[0256] Hereinafter, with reference to FIG. 8, a method for processing an image acquired from an image acquisition device (703) by an image processing device (701) through an artificial intelligence model learning unit (741), an artificial intelligence model (745), an image acquisition unit (750), a first image segmentation unit (760), and / or a second image segmentation unit (770) will be specifically described.

[0257] FIG. 8 illustrates images (810, 820, 830, 840) obtained by image processing of an image processing device (101) according to an embodiment of the present disclosure.

[0258] In the fifth embodiment, the artificial intelligence model learning unit (741), the artificial intelligence model (745), and the image acquisition unit (750) may have the same configuration as the artificial intelligence model learning unit (141), the artificial intelligence model (145), and the image acquisition unit (150) according to any one of the first to fourth embodiments described in FIGS. 1 to 6. In addition, the first image segmentation unit (760) may have a configuration including the image generation unit (160), the object identification unit (170), and the image segmentation unit (180) according to any one of the first to fourth embodiments described in FIGS. 1 to 6. That is, the first segmentation unit (760) may obtain a first segmentation image through binary image generation, object identification, and active material segmentation from an active material image.

[0259] Accordingly, in the following, descriptions of contents overlapping with those described in FIGS. 1 to 6 may be omitted.

[0260] In one embodiment, the image acquisition unit (750) can acquire active material images for a plurality of active materials. In one embodiment, the image acquisition unit (750) can acquire active material images through the image acquisition device (703).

[0261] In one embodiment, the first image segmentation unit (760) can segment a plurality of active materials included in the active material image acquired by the image acquisition unit (750) to acquire a first segmentation image (810). Here, the first segmentation image (810) can include an active material to be analyzed (811) and an active material not to be analyzed (813).

[0262] In one embodiment, the second image segmentation unit (770) can segment the target active material (811) to be analyzed in the first segmentation image (810) to obtain a second segmentation image (840).

[0263] In one embodiment, the second image segmentation unit (770) can convert the first segmentation image (810) into a binary image (820). Here, the binary image (820) can include active material regions including the analysis target region (821) and the remaining region (823).

[0264] In another embodiment, the second image segmentation unit (770) may convert the first segmentation image (810) into a binary image, and then segment active material regions corresponding to a plurality of active materials included in the converted binary image based on a watershed algorithm to obtain an auto-mask image (820). Here, the auto-mask image (820) may be a binary image including active material regions including an analysis target region (821) and the remaining region (823). In addition, the active material regions (821, 823) of the auto-mask image (820) may be configured as regions having one pixel value (e.g., a value representing a white color).

[0265] That is, the second image segmentation unit (770) can simply convert the first segmentation image (810) into a binary image (820), or can obtain an automask image (820) from the converted image after converting it into a binary image. Hereinafter, all images (binary images or automask images) (820) may be referred to as binary images (820).

[0266] In one embodiment, the second image segmentation unit (770) can extract an analysis target area (821) based on a binary image (820). In one embodiment, the second image segmentation unit (770) can extract an analysis target area (821) from a binary image (820) according to specified criteria.

[0267] In one embodiment, the second image segmentation unit (770) can extract an analysis target region (821) from the binary image (820) based on the areas of active material regions (821, 823) included in the binary image (820). For example, the second image segmentation unit (770) can extract the region (821) with the largest area in the binary image (820) as the analysis target region. As another example, the second image segmentation unit (770) can also extract the region (821) with an area greater than a reference area in the binary image (820) as the analysis target region. Here, the reference area can be set by the second image segmentation unit (770) based on the region (821) with the largest area in the binary image (820). Specifically, the second image segmentation unit (770) can set the reference area as a value obtained by multiplying the area of ​​the region (821) by a specified value (e.g., 0.3).

[0268] In one embodiment, the second image segmentation unit (770) can extract the analysis target region (821) from the binary image (820) based on shape information of the active material regions (821, 823) included in the binary image (820). For example, the shape information can include information on at least one of circularity, convexity, hardness, or aspect ratio of the active material regions (821, 833).

[0269] In one embodiment, the second image segmentation unit (770) can extract an analysis target area (821) from a binary image (820) and then generate a binary image (830) that includes only the analysis target area (821).

[0270] In one embodiment, the second image segmentation unit (770) may segment the analysis target active material (811) corresponding to the analysis target area (821) from the first segmentation image (810) based on the binary image (830) to obtain a second segmentation image (840). In one embodiment, the second image segmentation unit (770) may segment the analysis target active material (811) corresponding to the analysis target area (821) based on a watershed algorithm.

[0271] Fig. 9 is a flowchart illustrating an active material image processing method of an image processing device according to an embodiment of the present disclosure. Fig. 9 can be explained using the configurations of Fig. 7.

[0272] The embodiment illustrated in FIG. 9 is only one embodiment, and the order of steps according to various embodiments of the present invention may be different from that illustrated in FIG. 9, and some of the steps illustrated in FIG. 9 may be omitted, the order between steps may be changed, or steps may be merged.

[0273] Referring to FIG. 9, in operation 905, the image processing device (101) can acquire active material images for a plurality of active materials. In one embodiment, the image processing device (101) can acquire active material images through the image acquisition device (703).

[0274] In operation 910, the image processing device (101) can segment a plurality of active materials included in the active material image acquired in operation 905 to acquire a first segmentation image. Here, the first segmentation image can include an active material to be analyzed and an active material that is not a target of analysis.

[0275] In operation 915, the image processing device (101) can segment the active material to be analyzed from the first segmentation image acquired in operation 910 to acquire a second segmentation image.

[0276] In one embodiment, the image processing device (101) can convert the first segmentation image into a binary image. Here, the binary image can include active material regions including the analysis target region and the remaining regions.

[0277] In another embodiment, the image processing device (101) may convert the first segmentation image into a binary image, and then segment active material regions corresponding to a plurality of active materials included in the converted binary image based on a watershed algorithm to obtain an automask image. Here, the automask image may be a binary image including active material regions including an analysis target region and the remaining regions. In addition, the active material regions of the automask image may be composed of regions having one pixel value (e.g., a value representing white color).

[0278] That is, the image processing device (101) can simply convert the first segmentation image into a binary image, or can obtain an automask image from the converted image after converting it into a binary image. Hereinafter, both the binary image and the automask image may be referred to as a binary image.

[0279] In one embodiment, the image processing device (101) can extract an analysis target area based on a binary image. In one embodiment, the image processing device (101) can extract an analysis target area from a binary image according to specified criteria.

[0280] In one embodiment, the image processing device (101) can extract an analysis target area from a binary image based on the area of ​​active material areas included in the binary image. For example, the image processing device (101) can extract an area with the largest area in the binary image as the analysis target area. As another example, the image processing device (101) can also extract an area with an area greater than a reference area in the binary image as the analysis target area. Here, the reference area can be set based on the area with the largest area in the binary image. Specifically, the image processing device (101) can set a value obtained by multiplying the area of ​​the largest area in the binary image by a specified value (e.g., 0.3) as the reference area.

[0281] In one embodiment, the image processing device (101) can extract an analysis target region from a binary image based on shape information of active material regions included in the binary image. For example, the shape information can include information on at least one of circularity, convexity, hardness, or aspect ratio of the active material regions.

[0282] In one embodiment, the image processing device (101) can extract an analysis target area from a binary image and then generate a binary image that includes only the analysis target area.

[0283] In one embodiment, the image processing device (101) may segment the analysis target active material corresponding to the analysis target area from the first segmentation image based on a binary image containing only the analysis target area, thereby obtaining a second segmentation image. In one embodiment, the image processing device (101) may segment the analysis target active material corresponding to the analysis target area based on a watershed algorithm.

[0284] Hereinafter, an image processing device and its operation method according to a sixth embodiment of the present disclosure will be described with reference to FIGS. 10 to 12.

[0285] FIG. 10 is a block diagram of an image processing device (1001) according to an embodiment of the present disclosure. Specifically, the image processing device (1001) of FIG. 10 may correspond to an image processing device according to a sixth embodiment of the present disclosure.

[0286] In one embodiment, the connection (1005) between the image processing device (1001) and the image acquisition device (1003) may be a communication connection via a wired and / or wireless network. In one embodiment, the wired network may be based on LAN communication or power line communication. In one embodiment, the wireless network may be based on a short-range communication network (e.g., Bluetooth, WiFi, or IrDA) or a wide-range communication network (e.g., a cellular network, a 4G network, a 5G network).

[0287] In another embodiment, the connection (1005) between the image processing device (1001) and the image acquisition device (1003) may be a connection via a device-to-device communication method (e.g., bus, GPIO, SPI, or MIPI).

[0288] In one embodiment, the image acquisition device (1003) may be a microscope (e.g., a scanning electron microscope). In one embodiment, the image acquisition device (1003) may be a device that acquires an image of the sample surface by scanning a focused electron beam across the sample surface and converting secondary electrons generated by the interaction between the electron beam and the sample into image signals.

[0289] In one embodiment, the image acquisition device (1003) can acquire images of an active material and / or a precursor. Hereinafter, the images of the active material and / or the precursor may be referred to as active material images. However, even if referred to as active material images, the present disclosure does not exclude images of precursors.

[0290] In one embodiment, the image acquisition device (1003) can acquire an active material image of the active material. For example, the image acquisition device (1003) can acquire an SEM image by scanning an electron beam through the active material powder. In some embodiments, the SEM image can be replaced with an image based on TEM, OM, SIM, or FIB.

[0291] In one embodiment, the image acquisition device (1003) can transmit an active material image of the active material to the image processing device (1001). For example, the image acquisition device (1003) can transmit an active material image of the active material to the image processing device (1001) via a connection (1005).

[0292] In one embodiment, the image processing device (1001) may be a mobile device (e.g., a mobile phone, a laptop computer, a smart phone, a smart pad), a computer (e.g., a general-purpose computer, a special-purpose computer).

[0293] In one embodiment, the image processing device (1001) may include a communication circuit (1010), a memory (1020), and a processor (1030). According to an embodiment, the image processing device (1001) illustrated in FIG. 10 may further include at least one component (e.g., a display, an input device, or an output device) other than the components illustrated in FIG. 10.

[0294] In one embodiment, the communication circuit (1010) can establish a wired communication channel and / or a wireless communication channel between the image processing device (1001) and / or the image acquisition device (1003), and transmit and receive data to and from the image acquisition device (1003) through the established communication channel.

[0295] In one embodiment, the memory (1020) may include volatile memory and / or non-volatile memory.

[0296] In one embodiment, the memory (1020) may store data used by at least one component (e.g., the processor (1030)) of the image processing device (1001). For example, the data may include a program (1025) (or instructions related thereto), input data, or output data. In one embodiment, the instructions, when executed by the processor (1030), may cause the image processing device (1001) to perform operations defined by the instructions.

[0297] In one embodiment, the memory (1020) may include a program (1025) (e.g., an artificial intelligence model learning unit (1041), an artificial intelligence model (1045), an image acquisition unit (1050), an image segmentation unit (1060), an image generation unit (1070), and / or an information extraction unit (1080)).

[0298] In one embodiment, the processor (1030) may include a central processing unit, an application processor, a graphics processing unit, a neural processing unit (NPU), an image signal processor, a sensor hub processor, or a communication processor.

[0299] In one embodiment, the processor (1030) may execute a program (1025) (e.g., an artificial intelligence model learning unit (1041), an artificial intelligence model (1045), an image acquisition unit (1050), an image segmentation unit (1060), an image generation unit (1070), and / or an information extraction unit (1080)) to control at least one other component (e.g., a hardware or software component) of an image processing device (1001) connected to the processor (1030) and perform various data processing or calculations.

[0300] Hereinafter, with reference to FIG. 11, a method for processing an image acquired from an image acquisition device (1003) by an image processing device (1001) through an artificial intelligence model learning unit (1041), an artificial intelligence model (1045), an image acquisition unit (1050), an image segmentation unit (1060), an image generation unit (1070), and / or an information extraction unit (1080) is specifically described.

[0301] FIG. 11 illustrates images (1110, 1120, 1130) processed by an image processing device (1001) according to an embodiment of the present disclosure.

[0302] In the sixth embodiment, the artificial intelligence model learning unit (1041), the artificial intelligence model (1045), and the image acquisition unit (1050) may have the same configuration as the artificial intelligence model learning unit (141), the artificial intelligence model (145), and the image acquisition unit (150) according to any one of the first to fourth embodiments described in FIGS. 1 to 6. In addition, the image segmentation unit (1060) may have a configuration including the image generation unit (160), the object identification unit (170), and the image segmentation unit (180) according to any one of the first to fourth embodiments described in FIGS. 1 to 6. That is, the image segmentation unit (1060) may obtain a segmentation image through binary image generation, object identification, and active material segmentation from an active material image.

[0303] Accordingly, in the following, descriptions of contents overlapping with those described in FIGS. 1 to 6 may be omitted.

[0304] In one embodiment, the image acquisition unit (1050) can acquire active material images for a plurality of active materials. In one embodiment, the image acquisition unit (1050) can acquire active material images through the image acquisition device (1003).

[0305] In one embodiment, the image segmentation unit (1060) can segment a plurality of active materials included in the active material image acquired by the image acquisition unit (1050) to acquire a segmentation image (1110). Here, the segmentation image (1110) can include the active material and impurities or cracks on the surface of the active material.

[0306] In one embodiment, the image generation unit (1070) can generate a binary image (1120) in which impurities or active material surface cracks are separated based on the segmentation image (1110).

[0307] In one embodiment, the image generation unit (1070) can generate a binary image (1120) through a thresholding technique based on a pixel threshold value.

[0308] In one embodiment, the image generation unit (1070) may set the pixel threshold value (e.g., 100 to 200) based on the entire pixel values ​​of the segmentation image (1110). For example, the image generation unit (1070) may set the pixel threshold value through the Otsu thresholding technique.

[0309] Otsu thresholding uses Otsu's algorithm to convert an input image into a binary image. Otsu's algorithm randomly sets a threshold value when converting the input image to a binary image, divides pixels into two groups, and repeatedly calculates the intensity distributions of the two groups. Then, among all possible cases, the threshold value that produces the most uniform intensity distributions between the two groups is selected. This Otsu algorithm has the advantage of automatically setting the optimal threshold value.

[0310] In one embodiment, the image generation unit (1070) may set a preliminary threshold value based on the entire pixel values ​​of the segmentation image (1110). The image generation unit (1070) may set the pixel threshold value by multiplying the preliminary threshold value by a preset value (e.g., 0.5 to 1.5). For example, the image generation unit (1070) may set the preliminary threshold value through the Otsu thresholding technique.

[0311] In one embodiment, the image generation unit (1070) may obtain a label image (1130) including label information that enables the distinction between a plurality of active materials and impurities or active material surface cracks based on the segmentation image (1110) and the binary image (1120). For example, the label image (1130) may display the active material (1131) and the impurity (1133) (or the active material surface cracks) in different ways (e.g., colors) so as to enable the distinction between them.

[0312] In one embodiment, the information extraction unit (1080) may extract information on impurities or active material surface cracks based on a binary image (1120) or a label image (1130). For example, the information may include information on the area ratio of impurities or active material surface cracks.

[0313] In one embodiment, the information extraction unit (1080) may calculate an area ratio of a first region having a first pixel value (e.g., a value representing a white color) or a second region having a second pixel value (e.g., a value representing a black color) different from the first pixel value in the binary image (1120). The information extraction unit (1080) may calculate an area ratio of an impurity or an active material surface crack based on the area ratio of the first region or the area ratio of the second region.

[0314] Fig. 12 is a flowchart illustrating an active material image processing method of an image processing device according to an embodiment of the present disclosure. Fig. 12 can be explained using the configurations of Fig. 10.

[0315] The embodiment illustrated in FIG. 12 is only one embodiment, and the order of steps according to various embodiments of the present invention may be different from that illustrated in FIG. 12, and some of the steps illustrated in FIG. 12 may be omitted, the order between steps may be changed, or steps may be merged.

[0316] Referring to FIG. 12, in operation 1205, the image processing device (101) can acquire active material images for a plurality of active materials. In one embodiment, the image processing device (101) can acquire active material images through the image acquisition device (1003).

[0317] In operation 1210, the image processing device (101) can segment a plurality of active materials included in the active material image acquired in operation 1205 to acquire a segmentation image. Here, the segmentation image can include the active material and impurities or cracks on the surface of the active material.

[0318] In operation 1215, the image processing device (101) can generate a binary image in which impurities or active material surface cracks are separated based on the segmentation image acquired in operation 1210.

[0319] In one embodiment, the image processing device (101) can generate a binary image through a thresholding technique based on a pixel threshold value.

[0320] In one embodiment, the image processing device (101) can set the pixel threshold value (e.g., 100 to 200) based on the entire pixel values ​​of the segmentation image. For example, the image processing device (101) can set the pixel threshold value through the Otsu thresholding technique.

[0321] In one embodiment, the image processing device (101) may set a preliminary threshold value based on the entire pixel values ​​of the segmentation image. The image processing device (101) may set the pixel threshold value by multiplying the preliminary threshold value by a preset value (e.g., 0.5 to 1.5). For example, the image processing device (101) may set the preliminary threshold value through the Otsu thresholding technique.

[0322] In one embodiment, the image processing device (101) may obtain a label image including label information that enables the distinction between a plurality of active materials and impurities or cracks on the surface of the active materials based on the segmentation image and the binary image. For example, the label image may be displayed in different ways (e.g., colors) to enable the distinction between the active materials and impurities (or cracks on the surface of the active materials).

[0323] In operation 1220, the image processing device (101) can extract information about impurities or active material surface cracks based on the binary image or label image acquired in operation 1215. For example, the information can include information about the area ratio of impurities or active material surface cracks.

[0324] In one embodiment, the image processing device (101) can calculate an area ratio of a first region having a first pixel value (e.g., a value representing a white color) in a binary image or a second region having a second pixel value (e.g., a value representing a black color) different from the first pixel value. The image processing device (101) can calculate an area ratio of an impurity or an active material surface crack based on the area ratio of the first region or the area ratio of the second region.

[0325] Hereinafter, an image processing device and its operation method according to a seventh embodiment of the present disclosure will be described with reference to FIGS. 13 to 17.

[0326] FIG. 13 is a block diagram of an image processing device (1301) according to an embodiment of the present disclosure. Specifically, the image processing device (1301) of FIG. 13 may correspond to the image processing device according to the seventh embodiment of the present disclosure.

[0327] In one embodiment, the connection (1305) between the image processing device (1301) and the image acquisition device (1303) may be a communication connection via a wired and / or wireless network. In one embodiment, the wired network may be based on LAN communication or power line communication. In one embodiment, the wireless network may be based on a short-range communication network (e.g., Bluetooth, WiFi, or IrDA) or a wide-range communication network (e.g., a cellular network, a 4G network, a 5G network).

[0328] In another embodiment, the connection (1305) between the image processing device (1301) and the image acquisition device (1303) may be a connection via a device-to-device communication method (e.g., bus, GPIO, SPI, or MIPI).

[0329] In one embodiment, the image acquisition device (1303) may be a microscope (e.g., a scanning electron microscope). In one embodiment, the image acquisition device (1303) may be a device that acquires an image of the sample surface by scanning a focused electron beam across the sample surface and converting secondary electrons generated by the interaction between the electron beam and the sample into image signals.

[0330] In one embodiment, the image acquisition device (1303) can acquire images of an active material and / or a precursor. Hereinafter, the images of the active material and / or the precursor may be referred to as active material images. However, even if referred to as active material images, the present disclosure does not exclude images of precursors.

[0331] In one embodiment, the image acquisition device (1303) can acquire an active material image of the active material. For example, the image acquisition device (1303) can acquire an SEM image by scanning an electron beam through the active material powder. In some embodiments, the SEM image can be replaced with an image based on TEM, OM, SIM, or FIB.

[0332] In one embodiment, the image acquisition device (1303) can transmit an active material image of the active material to the image processing device (1301). For example, the image acquisition device (1303) can transmit an active material image of the active material to the image processing device (1301) via a connection (1305).

[0333] In one embodiment, the image processing device (1301) may be a mobile device (e.g., a mobile phone, a laptop computer, a smart phone, a smart pad), a computer (e.g., a general-purpose computer, a special-purpose computer).

[0334] In one embodiment, the image processing device (1301) may include a communication circuit (1310), a memory (1320), and a processor (1330). According to an embodiment, the image processing device (1301) illustrated in FIG. 13 may further include at least one component (e.g., a display, an input device, or an output device) other than the components illustrated in FIG. 13.

[0335] In one embodiment, the communication circuit (1310) can establish a wired communication channel and / or a wireless communication channel between the image processing device (1301) and / or the image acquisition device (1303), and transmit and receive data to and from the image acquisition device (1303) through the established communication channel.

[0336] In one embodiment, the memory (1320) may include volatile memory and / or non-volatile memory.

[0337] In one embodiment, the memory (1320) may store data used by at least one component (e.g., the processor (1330)) of the image processing device (1301). For example, the data may include a program (1325) (or instructions related thereto), input data, or output data. In one embodiment, the instructions, when executed by the processor (1330), may cause the image processing device (1301) to perform operations defined by the instructions.

[0338] In one embodiment, the memory (1320) may include a program (1325) (e.g., a first artificial intelligence model learning unit (1341), a first artificial intelligence model (1345), a second artificial intelligence model learning unit (1351), a second artificial intelligence model (1355), an image acquisition unit (1360), an image segmentation unit (1370), an image generation unit (1380), and / or an information extraction unit (1390)).

[0339] In one embodiment, the processor (1330) may include a central processing unit, an application processor, a graphics processing unit, a neural processing unit (NPU), an image signal processor, a sensor hub processor, or a communication processor.

[0340] In one embodiment, the processor (1330) may execute a program (1325) (e.g., a first artificial intelligence model learning unit (1341), a first artificial intelligence model (1345), a second artificial intelligence model learning unit (1351), a second artificial intelligence model (1355), an image acquisition unit (1360), an image segmentation unit (1370), an image generation unit (1380), and / or an information extraction unit (1390)) to control at least one other component (e.g., a hardware or software component) of an image processing device (1301) connected to the processor (1330) and perform various data processing or calculations.

[0341] Hereinafter, with reference to FIGS. 14 and 15, a method for processing an image acquired from an image acquisition device (1303) by an image processing device (1301) through a first artificial intelligence model learning unit (1341), a first artificial intelligence model (1345), a second artificial intelligence model learning unit (1351), a second artificial intelligence model (1355), an image acquisition unit (1360), an image segmentation unit (1370), an image generation unit (1380), and / or an information extraction unit (1390) will be specifically described.

[0342] FIG. 14 illustrates images (1410, 1411, 1412, 1420, 1430, 1440) obtained by image processing of an image processing device (1301) according to an embodiment of the present disclosure.

[0343] In the seventh embodiment, the first artificial intelligence model learning unit (1341), the first artificial intelligence model (1345), and the image acquisition unit (1360) may have the same configuration as the artificial intelligence model learning unit (141), the artificial intelligence model (145), and the image acquisition unit (150) according to the fourth embodiment described in FIGS. 1 to 6. In addition, the image segmentation unit (1370) may have a configuration including the image generation unit (160), the object identification unit (170), and the image segmentation unit (180) according to the fourth embodiment described in FIGS.

[0344] Accordingly, in the following, descriptions of contents overlapping with those of the fourth embodiment described in FIGS. 1 to 6 may be omitted.

[0345] In one embodiment, the image acquisition unit (1360) can acquire active material images (1410) for a plurality of active materials. In one embodiment, the image acquisition unit (1360) can acquire the active material images (1410) through the image acquisition device (1303).

[0346] In one embodiment, the image segmentation unit (1370) can segment a plurality of active materials included in the active material image (1410) to obtain a segmentation image (1420).

[0347] In one embodiment, the image segmentation unit (1370) can generate binary images (1411, 1412) in which the active material and impurities are separated. Here, the binary images (1411, 1412) can include a first binary image (1411) corresponding to the active material and a second binary image (1412) corresponding to a crack on the surface of the active material.

[0348] In one embodiment, the image segmentation unit (1370) can generate a first binary image (1411) and a second binary image (1412) by inputting an active material image (1410) into the first artificial intelligence model (1345).

[0349] In one embodiment, the image segmentation unit (1370) can identify a plurality of objects included in the first binary image (1411) and / or the second binary image (1412). Here, the plurality of objects included in a specific binary image (1411 or 1412) can correspond to an active material or a crack on the surface of the active material in the active material image (1410). In one embodiment, the plurality of objects can be composed of areas having a specified pixel value (e.g., a value representing a white color). In one embodiment, the plurality of objects can be distinguished by areas having another specified value (e.g., a value representing a black color).

[0350] In one embodiment, the image segmentation unit (1370) may segment a plurality of active materials included in the active material image (1410) based on a plurality of identified objects to obtain a segmentation image (1420). In one embodiment, the image segmentation unit (1370) may segment a plurality of active materials included in the active material image (1410) based on a watershed algorithm.

[0351] In one embodiment, the image generation unit (1380) can generate a pore binary image (1440) representing pore characteristics of the active material based on the segmentation image (1420).

[0352] In one embodiment, the image generation unit (1380) can input the segmentation image (1420) to the second artificial intelligence model (1355) to generate a third binary image (1430) including pore features and surface crack features of the active material. Here, the second artificial intelligence model (1355) is a model based on CNN or U-NET, and can be a model trained by the second artificial intelligence model training unit (1351) to convert the image into a binary image. Specifically, the second artificial intelligence model (1355) can generate a binary image that displays all of the shape features, pore features, and surface crack features of the active material from the input image.

[0353] In one embodiment, the second artificial intelligence model learning unit (1351) and the second artificial intelligence model (1355) may be the same as the artificial intelligence model learning unit (141) and the artificial intelligence model (145) according to the first embodiment described in FIGS. 1 to 6, or may have the same configuration but with the addition of a predetermined feature (e.g., a feature that generates a binary image that displays all of the shape features, pore features, and surface crack features of the active material).

[0354] In one embodiment, the image generation unit (1380) may perform mask normalization on the segmentation image (1420) to obtain a normalized image (not shown). Specifically, the image generation unit (1380) may perform mask normalization on an active material region included in the segmentation image (1420) to obtain a normalized image. The mask normalization may be, for example, max-min normalization based on a minimum pixel value (e.g., 0) and a maximum pixel value (e.g., 255).

[0355] In one embodiment, the image generation unit (1380) can input a normalized image into the second artificial intelligence model (1355) to generate a third binary image (1430). The image processing device (1301) can minimize errors caused by the analysis environment by performing mask normalization on the segmentation image (1420) before generating the third binary image (1430) as described above.

[0356] In one embodiment, the image generating unit (1380) can generate a void binary image (1440) further based on the second binary image (1412). In one embodiment, the image generating unit (1380) can generate a void binary image (1440) by filtering surface crack features of the third binary image (1430) based on the second binary image (1412). Specifically, the image generating unit (1380) can generate the void binary image (1440) by adding or subtracting pixel values ​​of the second binary image (1412) to pixel values ​​of the third binary image (1430). For example, if the pixel values ​​of objects corresponding to the cracks on the surface of the active material in the second binary image (1412) are values ​​representing white, and the pixel values ​​of objects corresponding to the cracks on the surface of the active material in the third binary image (1430) are values ​​representing black, the surface crack features can be filtered by adding the pixel values ​​of the second binary image (1412) to the pixel values ​​of the third binary image (1430).

[0357] In one embodiment, the information extraction unit (1390) may extract information about active material pores based on the pore binary image (1440). For example, the information about active material pores may include information about the porosity of the active material cross-section.

[0358] FIG. 15 is a drawing for explaining a method for setting a core region and a shell region of an active material by an image processing device (1301) according to one embodiment of the present disclosure.

[0359] In one embodiment, the information extraction unit (1390) can extract information about the active material pores (e.g., the porosity of the active material cross-section) based on the pore binary image (1510) generated by the image generation unit (1380).

[0360] In one embodiment, the information extraction unit (1390) can set a core area and a shell area of ​​the active material based on the extracted information.

[0361] Referring to image (1520), the information extraction unit (1390) can calculate the porosity of the central region (1521) of the cross-section of the active material. Here, the central region (1521) can be set by reducing the cross-section of the active material by a specified multiple (e.g., 0.1 times).

[0362] Referring to image (1530), the information extraction unit (1390) can calculate the porosity in a designated area of ​​the cross-section of the active material.

[0363] For example, the designated region may be an area within the boundary (1531) that includes the center point of the cross-section of the active material.

[0364] As another example, the designated region may be a boundary region excluding a second region from a first region. Here, the first region may be a region including the center point of the cross-section of the active material, and the second region may be a region including the center point of the cross-section of the active material and included in the first region.

[0365] The information extraction unit (1390) can calculate the porosity for each region by expanding the specified region from a small region near the center of the cross-section of the active material to the entire cross-section of the active material using a distance transform (DT) and calculating the area ratio occupied by the pores in the region.

[0366] Referring to image (1540), the information extraction unit (1390) can set the core region and shell region of the active material based on the porosity in the designated region.

[0367] In one embodiment, the information extraction unit (1390) may set the core region and shell region of the active material based on the boundary (1531) of the specified region if the porosity in the specified region is less than or equal to a threshold value. Here, the threshold value may be set based on the porosity of the central region (1521) of the cross-section of the active material. For example, the threshold value may be set by multiplying the porosity of the central region (1521) of the cross-section of the active material by a specified multiple (e.g., 0.3 times).

[0368] For example, if the designated area is an area inside a boundary (1531) that includes the center point of the cross-section of the active material, the information extraction unit (1390) can set the area (1541) inside the boundary (1531) of the designated area as a core area and the area (1542) outside the boundary (1531) as a shell area if the porosity in the designated area is less than or equal to a threshold value.

[0369] As another example, if the designated area is a boundary area excluding the second area from the first area, the information extraction unit (1390) can set the inside of the outer boundary of the designated area as a core area and the outside of the outer boundary of the designated area as a shell area if the porosity in the designated area is below a threshold value.

[0370] In one embodiment, the information extraction unit (1390) can calculate at least one of a length ratio, an area ratio, or a porosity of the set core region and shell region.

[0371] Fig. 16 is a flowchart illustrating an active material image processing method of an image processing device according to an embodiment of the present disclosure. Fig. 16 can be explained using the configurations of Fig. 13.

[0372] The embodiment illustrated in FIG. 16 is only one embodiment, and the order of steps according to various embodiments of the present invention may be different from that illustrated in FIG. 16, and some of the steps illustrated in FIG. 16 may be omitted, the order between steps may be changed, or steps may be merged.

[0373] Referring to FIG. 16, in operation 1605, the image processing device (1301) can acquire active material images for a plurality of active materials. In one embodiment, the image processing device (1301) can acquire active material images through the image acquisition device (1303).

[0374] In operation 1610, the image processing device (1301) can segment a plurality of active materials included in the active material image acquired in operation 1605 to acquire a segmentation image.

[0375] In one embodiment, the image processing device (1301) can generate binary images in which the active material and impurities are separated. Here, the binary images can include a first binary image corresponding to the active material and a second binary image corresponding to a crack on the surface of the active material.

[0376] In one embodiment, the image processing device (1301) can generate a first binary image and a second binary image by inputting an active material image into the first artificial intelligence model (1345).

[0377] In one embodiment, the image processing device (1301) can identify a plurality of objects included in a first binary image and / or a second binary image. Here, the plurality of objects included in a specific binary image can correspond to an active material or a crack on the surface of the active material in the active material image. In one embodiment, the plurality of objects can be composed of areas having a designated pixel value (e.g., a value representing a white color). In one embodiment, the plurality of objects can be distinguished by areas having another designated value (e.g., a value representing a black color).

[0378] In one embodiment, the image processing device (1301) can segment a plurality of active materials included in the active material image based on a plurality of identified objects to obtain a segmentation image. In one embodiment, the image processing device (1301) can segment a plurality of active materials included in the active material image based on a watershed algorithm.

[0379] In operation 1615, the image processing device (1301) can generate a pore binary image representing pore characteristics of the active material based on the segmentation image acquired in operation 1610.

[0380] In one embodiment, the image processing device (1301) can input the segmentation image into the second artificial intelligence model (1355) to generate a third binary image including pore features and surface crack features of the active material.

[0381] In one embodiment, the image processing device (1301) may perform mask normalization on the segmentation image to obtain a normalized image. Specifically, the image processing device (1301) may perform mask normalization on an active material region included in the segmentation image to obtain a normalized image. The mask normalization may be, for example, max-min normalization based on a minimum pixel value (e.g., 0) and a maximum pixel value (e.g., 255).

[0382] In one embodiment, the image processing device (1301) can input a normalized image into the second artificial intelligence model (1355) to generate a third binary image. The image processing device (1301) can minimize errors caused by the analysis environment by performing mask normalization on the segmentation image before generating the third binary image as described above.

[0383] In one embodiment, the image processing device (1301) can generate a void binary image further based on the second binary image. In one embodiment, the image processing device (1301) can generate a void binary image by filtering surface crack features of a third binary image based on the second binary image. Specifically, the image processing device (1301) can generate the void binary image by adding or subtracting pixel values ​​of the second binary image to pixel values ​​of the third binary image. For example, when pixel values ​​of objects corresponding to surface cracks of an active material in the second binary image are values ​​representing white, and pixel values ​​of objects corresponding to surface cracks of an active material in the third binary image are values ​​representing black, the surface crack features can be filtered by adding pixel values ​​of the second binary image to the pixel values ​​of the third binary image.

[0384] In operation 1620, the image processing device (1301) can extract information about active material pores based on the pore binary image (1440). For example, the information about active material pores can include information about the porosity of the active material cross-section.

[0385] FIG. 17 is a flowchart illustrating a method for setting a core region and a shell region of an active material by an image processing device according to an embodiment of the present disclosure. FIG. 17 can be explained using the configurations of FIG. 13.

[0386] The embodiment illustrated in FIG. 17 is only one embodiment, and the order of steps according to various embodiments of the present invention may be different from that illustrated in FIG. 17, and some of the steps illustrated in FIG. 17 may be omitted, the order between steps may be changed, or steps may be merged.

[0387] Referring to FIG. 17, in operation 1705, the image processing device (1301) can calculate the porosity in a designated area of ​​the cross-section of the active material based on the porosity binary image generated in operation 1615 of FIG. 16.

[0388] For example, the designated area may be an area that includes the center point of the cross-section of the active material.

[0389] As another example, the designated region may be a boundary region excluding a second region from a first region. Here, the first region may be a region including the center point of the cross-section of the active material, and the second region may be a region including the center point of the cross-section of the active material and included in the first region.

[0390] In operation 1710, the image processing device (1301) can identify whether the porosity calculated in operation 1705 is less than or equal to a threshold value.

[0391] In one embodiment, the threshold value may be set based on the porosity of a central region of the active material cross-section. Here, the central region may be set by reducing the active material cross-section by a specified multiple (e.g., 0.1 times). For example, the threshold value may be set by multiplying the porosity of the central region of the active material cross-section by a specified multiple (e.g., 0.3 times).

[0392] If the porosity is identified as exceeding the threshold value in operation 1710 ('NO'), the image processing device (1301) may re-perform operation 1705. At this time, the image processing device (1301) may calculate the porosity in a wider area than the area designated in the previous operation 1705. That is, the image processing device (1301) may calculate the porosity by expanding the designated area from a small area near the center of the cross-section of the active material to the entire cross-section of the active material until the porosity of the designated area becomes lower than the threshold value.

[0393] If the porosity is identified as being below a threshold value in operation 1710 ('YES'), in operation 1715, the image processing device (1301) can set a core region and a shell region of the active material based on the boundary of the designated region.

[0394] For example, if the designated area is an area including the center point of the cross-section of the active material, the image processing device (1301) can set the area inside the boundary of the designated area as a core area and the area outside the boundary as a shell area if the porosity in the designated area is less than or equal to a threshold value.

[0395] As another example, if the designated area is a boundary area excluding the second area from the first area, the image processing device (1301) can set the inside of the outer boundary of the designated area as a core area and the outside of the outer boundary of the designated area as a shell area if the porosity in the designated area is less than or equal to a threshold value.

[0396] In one embodiment, the image processing device (1301) can calculate at least one of a length ratio, an area ratio, or a porosity of a set core region and a shell region.

[0397] In the present disclosure, the image processing device (101, 701, 1001, or 1301) is exemplified as performing an image processing method based on an image of an active material, but this is merely an example. According to an embodiment, the image processing device (101, 701, 1001, or 1301) can perform the same image processing not only on an image of an active material but also on an image of a precursor, thereby separating and / or analyzing particles. The image acquisition device (103, 703, 1003, or 1303) can acquire images of the active material and / or the precursor.

[0398] The terms "include," "comprise," or "have" used herein, unless otherwise specifically stated, imply that the corresponding component may be included, and therefore should be interpreted to include other components rather than to exclude other components. All terms, including technical or scientific terms, have the same meaning as commonly understood by a person of ordinary skill in the art to which the embodiments disclosed in this document pertain, unless otherwise defined. Commonly used terms, such as terms defined in a dictionary, should be interpreted to be consistent with the contextual meaning of the relevant technology, and shall not be interpreted in an idealized or overly formal sense, unless explicitly defined in this document.

Claims

1. An image acquisition unit for acquiring active material images for multiple active materials; An image segmentation unit that segments the plurality of active materials included in the above active material image to obtain a segmentation image; and An image processing device including an image generating unit that inputs the active material image into an artificial intelligence model to generate at least one of a binary image in which the active material and the active material surface crack are separated, a pore binary image representing the pore characteristics of the active material based on the segmentation image, or a combination thereof.

2. In claim 1, An image processing device further comprising an information extraction unit that extracts information about active material pores based on the above-mentioned pore binary image.

3. In claim 2, The above image segmentation part, By inputting the image of the active material into the first artificial intelligence model, a first binary image corresponding to the active material and a second binary image corresponding to a crack on the surface of the active material are generated, Identifying multiple objects included in the first binary image, Segmenting the plurality of active materials included in the active material image based on the plurality of objects to obtain the segmentation image, The above image generating unit generates the gap binary image further based on the second binary image.

4. In claim 3, The above image generating unit, By inputting the above segmentation image into the second artificial intelligence model, a third binary image including the pore features and surface crack features of the active material is generated, An image processing device that generates the gap binary image by filtering the surface crack features of the third binary image based on the second binary image.

5. In claim 3, The above image generating unit, Perform mask normalization on the above segmentation image to obtain a normalized image, By inputting the normalized image into the second artificial intelligence model, a third binary image including the pore features and surface crack features of the active material is generated, An image processing device that filters surface crack features of the third binary image based on the second binary image to generate the gap binary image.

6. In claim 2, The above information extraction unit, An image processing device that sets a core area and a shell area of ​​an active material based on the extracted information.

7. In claim 6, The above information extraction unit, Based on the above pore binary image, the porosity of the cross-section of the active material is calculated, An image processing device that sets a core region and a shell region of an active material based on the above porosity.

8. In claim 7, The above information extraction unit, Calculate the porosity in a specified area of ​​the cross-section of the active material, An image processing device that sets a core region and a shell region of an active material based on the porosity in the above-mentioned designated region.

9. In claim 1, Further comprising an object identification unit for identifying a plurality of objects included in the binary image; The above image generating unit, An image processing device that segments the plurality of active materials included in the active material image based on the plurality of objects to obtain the segmentation image.

10. In claim 9, An image processing device further comprising an artificial intelligence model learning unit that learns the artificial intelligence model using learning data including a plurality of reference active material images and a plurality of reference binary images corresponding to the plurality of reference active material images.

11. The action of obtaining active material images for multiple active materials; An operation of segmenting the plurality of active materials included in the above active material image to obtain a segmentation image; and An image processing method comprising an operation of inputting the active material image into an artificial intelligence model to generate at least one of a binary image in which the active material and a surface crack of the active material are separated, a pore binary image representing pore characteristics of the active material based on the segmentation image, or any combination thereof.

12. In claim 11, An image processing method further comprising an operation of extracting information about active material pores based on the above-mentioned pore binary image.

13. In claim 12, The operation of obtaining the above segmentation image is as follows: An operation of inputting the image of the active material into the first artificial intelligence model to generate a first binary image corresponding to the active material and a second binary image corresponding to a crack on the surface of the active material; An operation for identifying a plurality of objects included in the first binary image, and An operation of segmenting the plurality of active materials included in the active material image based on the plurality of objects to obtain the segmentation image, An image processing method, wherein the operation of generating the gap binary image includes an operation of generating the gap binary image further based on the second binary image.

14. In claim 13, The operation of generating the above-mentioned pore binary image comprises: an operation of inputting the segmentation image into a second artificial intelligence model to generate a third binary image including pore characteristics and surface crack characteristics of the active material; and An image processing method, comprising an operation of filtering surface crack features of the third binary image based on the second binary image to generate the void binary image.

15. In claim 13, The operation of generating the above gap binary image is: An operation of performing mask normalization on the above segmentation image to obtain a normalized image. An operation of inputting the normalized image into a second artificial intelligence model to generate a third binary image including pore features and surface crack features of the active material; An image processing method, comprising an operation of filtering surface crack features of the third binary image based on the second binary image to generate the void binary image.

16. In claim 12, The action of extracting the above information is: An image processing method, comprising an operation of setting a core area and a shell area of ​​an active material based on the extracted information.

17. In claim 16, The operation of setting the core region and shell region of the above active material is: An operation for calculating the porosity of the cross-section of the active material based on the above-mentioned pore binary image, and An image processing method, comprising an operation of setting a core region and a shell region of an active material based on the above porosity.

18. In claim 17, The operation of setting the core region and shell region of the above active material is: An operation for calculating the porosity in a specified area of ​​the cross-section of an active material, and An image processing method, comprising an operation of setting a core region and a shell region of an active material based on the porosity in the designated region.

19. In claim 11, Further comprising an operation of identifying multiple objects included in the binary image, The operation of obtaining the above segmentation image is as follows: An image processing method, comprising an operation of segmenting the plurality of active materials included in the active material image based on the plurality of objects to obtain the segmentation image.

20. In claim 19, An image processing method further comprising an action of training the artificial intelligence model using training data including a plurality of reference active material images and a plurality of reference binary images corresponding to the plurality of reference active material images.

Images