Image processing apparatus and method

Abstract

An image processing apparatus according to an embodiment disclosed in this document may include: an image acquisition unit configured to acquire an active material image of a plurality of active materials; an artificial intelligence model training unit configured to train an artificial intelligence model using training data including a plurality of reference active material images and a plurality of reference binary images or a plurality of reference edge images corresponding to the plurality of reference active material images; an image generation unit configured to generate a binary image or an edge image by inputting the active material image to an artificial intelligence model; an object recognition unit configured to recognize a plurality of objects based on the binary image or the edge image; and an image segmentation unit configured to acquire a segmented image by segmenting the plurality of active materials included in the active material image based on the plurality of objects.

Description

Technical Field

[0001] Cross - Reference to Related Applications

[0002] This application claims priority and the benefit of Korean Patent Application Nos. 10 - 2022 - 0175984, 10 - 2022 - 0175985, and 10 - 2022 - 0175986, filed with the Korean Intellectual Property Office on December 15, 2022, the entire contents of which are incorporated herein by reference. Technical Field

[0004] Embodiments disclosed in this document relate to an image processing apparatus and method. Background Art

[0005] With the development of technology and the growing demand for mobile devices and electric vehicles, the demand for lithium - ion batteries as an energy source is increasing rapidly.

[0006] A lithium - ion battery is manufactured by placing an electrode assembly composed of a positive electrode, a separator, and a negative electrode in a battery case and injecting an electrolyte. The positive electrode is manufactured by coating a composition for forming a positive electrode active material layer containing a positive electrode active material, a conductive agent, and a binder on a positive electrode current collector, drying, and rolling. The negative electrode is manufactured by coating a composition for forming a negative electrode active material layer containing a negative electrode active material, a conductive agent, and a binder on a negative electrode current collector, drying, and rolling.

[0007] Meanwhile, the electrochemical performance of a lithium - ion battery is affected not only by the composition of the positive and negative electrode active materials used but also by the shape of the active material particles. This is because electrochemical properties such as electrode density, conductivity, and resistance characteristics can vary depending on the shape or size of the active material particles. Therefore, in order to manufacture a lithium - ion battery with desired performance, the shape of the active material particles must be accurately analyzed.

[0008] Conventionally, an image of positive electrode active material particles obtained by a scanning electron microscope (SEM) has been used to analyze the shape of the positive electrode active material particles. However, in the conventional method of analyzing the shape of the positive electrode active material using SEM images, qualitative analysis is possible, but for quantitative analysis, a person must manually measure them by selecting individual active material particles from the SEM image, and automatic quantitative analysis cannot be performed. Therefore, it is difficult to perform quantitative analysis on a large number of positive electrode active material particles, and the accuracy of the analysis may be low because the measured values may vary depending on the person performing the measurement. Summary of the Invention

[0009] [Technical Problem]

[0010] The present invention has been conceived to solve the above problems, and an object of the present invention is to provide a method for automatically quantitatively analyzing the shape characteristics of a large number of positive electrode active material and / or precursor particles.

[0011] Another object of the present invention is to provide a quantitative analysis method that minimizes user participation, thereby alleviating the problem of deterioration in analysis accuracy caused by measurement variations depending on the user.

[0012] [Technical Solution]

[0013] An image processing apparatus according to an embodiment disclosed in this document may include: an image acquisition unit configured to acquire active material images of a plurality of active materials; an artificial intelligence model training unit configured to train an artificial intelligence model using training data including a plurality of reference active material images and a plurality of reference binary images or a plurality of reference edge images corresponding to the plurality of reference active material images; an image generation unit configured to generate a binary image or an edge image by inputting the active material image into the artificial intelligence model; an object recognition unit configured to recognize a plurality of objects based on the binary image or the edge image; and an image segmentation unit configured to obtain a segmented image by segmenting a plurality of active materials included in the active material image based on the plurality of objects.

[0014] In an embodiment, the artificial intelligence model training unit may train the artificial intelligence model to ensure that the difference between a plurality of output images obtained by inputting the plurality of reference active material images into the artificial intelligence model and the plurality of reference binary images is equal to or less than a predetermined reference value.

[0015] In an embodiment, the plurality of reference active material images may include one or more first reference active material images and one or more second reference active material images, the plurality of reference binary images may include one or more first reference binary images and one or more second reference binary images, the first reference active material image is obtained by capturing an image of active material powder, the second reference active material image is generated by applying a predetermined image processing algorithm to the first reference active material image, the first reference binary image corresponds to the first reference active material image, and the second reference binary image corresponds to the second reference active material image.

[0016] In an embodiment, the artificial intelligence model may include a plurality of layers, and at least one layer included in an encoding region and at least one layer included in a decoding region are connected by a skip connection among the plurality of layers.

[0017] In an embodiment, the layer included in the decoding region among the plurality of layers may have a structure in which a two-dimensional (2D) convolutional layer, a batch normalization layer, an activation layer, and an upsampling layer are sequentially connected.

[0018] In an embodiment, among multiple layers, a layer included in an encoding region may have a structure in which a two-dimensional (2D) convolutional layer, a batch normalization layer, an activation layer, and a max pooling layer are sequentially connected.

[0019] In an embodiment, the active material image may be an image based on a scanning electron microscope (SEM), a transmission electron microscope (TEM), an optical microscope (OM), structured illumination microscopy (SIM), or a focused ion beam (FIB).

[0020] In an embodiment, the image segmentation unit may segment multiple active materials included in the active material image based on a watershed algorithm.

[0021] In an embodiment, the image processing apparatus may further include an information extraction unit configured to extract information about multiple active materials based on the segmented image, where the information about the multiple active materials may include the particle diameter, perimeter, sphericity, aspect ratio, convexity, solidity, distribution, percentile value, or any combination thereof of the active materials.

[0022] In an embodiment, the image processing apparatus may further include an edge removal unit configured to generate an edge-removed image by removing edges in the active material image based on an edge image, where the object recognition unit may recognize multiple objects included in the edge-removed image.

[0023] In an embodiment, the artificial intelligence model training unit may train an artificial intelligence model to ensure that the difference between multiple output images obtained by inputting multiple reference active material images into the artificial intelligence model and multiple reference edge images is equal to or less than a predetermined reference value.

[0024] In an embodiment, the artificial intelligence model training unit may calculate the difference between multiple output images and multiple reference edge images based on an average absolute error function, a root mean square error, a mean square error function, or a binary cross-entropy loss function.

[0025] In an embodiment, the binary cross-entropy loss function may be assigned weights for a predetermined color.

[0026] In an embodiment, the image processing apparatus may further include a distance transformation unit configured to transform a binary image into a distance transformation image based on a distance transformation algorithm, and

[0027] a binary image filtering unit configured to filter the binary image using a threshold set based on the distance transformation image, where the object recognition unit may recognize multiple objects included in the filtered binary image.

[0028] In an embodiment, the distance transformation unit may generate a distance transformation image by calculating the distance from each pixel included in the binary image to the nearest black pixel and normalizing the calculated distance for each pixel.

[0029] In an embodiment, the distance transformation unit may generate a distance transformation image by identifying the maximum distance among the distances calculated for each pixel included in the binary image and performing max - min normalization on the distances calculated for each pixel based on the maximum distance.

[0030] In an embodiment, the binary image filtering unit may filter the binary image by selecting pixels in the distance transformation image whose distance values are equal to or less than a predetermined threshold and setting the color values of the pixels in the binary image corresponding to the selected pixels to a predetermined color value.

[0031] In an embodiment, the predetermined threshold may be a value obtained by multiplying a predetermined ratio by the normalized maximum distance of the distance transformation image.

[0032] [Advantageous Effects]

[0033] According to various embodiments disclosed in this document, the image processing apparatus and method have advantages in achieving quantitative analysis of the shape characteristics of a large number of positive electrode active material particles by: acquiring active material images of a plurality of active materials, generating a binary image by inputting the active material images into an artificial intelligence model, identifying a plurality of objects included in the binary image, and segmenting the plurality of active materials within the active material images based on the identified objects.

[0034] According to various embodiments disclosed in this document, the image processing apparatus and method have advantages in achieving quantitative analysis of the shape characteristics of a large number of positive electrode active material particles by: acquiring active material images of a plurality of active materials, generating an edge image by inputting the active material images into an artificial intelligence model, generating an edge - removed image by removing the boundaries of the active material images based on the edge image, identifying a plurality of objects included in the edge - removed image, and segmenting the plurality of active materials within the active material images.

[0035] According to various embodiments disclosed in this document, the image processing apparatus and method have an advantage in achieving quantitative analysis of the shape characteristics of a large number of positive electrode active material particles by: obtaining active material images of a plurality of active materials, generating a binary image by inputting the active material images into an artificial intelligence model, transforming the binary image into a distance transform image based on a distance transform algorithm, filtering the binary image using a threshold set based on the distance transform image, identifying a plurality of objects included in the filtered binary image, and segmenting a plurality of active material objects included in the active material image based on the plurality of objects.

[0036] According to various embodiments disclosed in this document, the image processing apparatus and method also have an advantage in alleviating the problem of deterioration in analysis accuracy due to measurement variations depending on the user by minimizing user participation.

[0037] The beneficial effects of the image processing apparatus and method disclosed in this document are not limited to the above, and other effects not described herein will be clearly understood by those skilled in the art from the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 is a block diagram illustrating an image processing apparatus according to an embodiment of the present disclosure;

[0039] Figure 2 illustrates training data according to an embodiment of the present disclosure;

[0040] Figure 3 illustrates an artificial intelligence model according to an embodiment of the present disclosure;

[0041] Figure 4 illustrates an image generated by image processing of an image processing apparatus according to an embodiment of the present disclosure;

[0042] Figure 5 is a flowchart illustrating a method for processing an active material image in an image processing apparatus according to an embodiment of the present disclosure;

[0043] Figure 6 is a flowchart illustrating a method for training an artificial intelligence model in an image processing apparatus according to an embodiment of the present disclosure;

[0044] Figure 7 is a block diagram illustrating an image processing apparatus according to another embodiment of the present disclosure;

[0045] Figure 8 illustrates training data according to another embodiment of the present disclosure;

[0046] Figure 9 illustrates an artificial intelligence model according to another embodiment of the present disclosure;

[0047] Figure 10a An image generated by image processing of an image processing apparatus according to another embodiment of the present disclosure;

[0048] Figure 10b Illustrates the differences in the image generated by the image processing of the image processing apparatus according to another embodiment of the present disclosure;

[0049] Figure 11 Is a flowchart illustrating a method for processing an active material image in an image processing apparatus according to another embodiment of the present disclosure;

[0050] Figure 12 Is a flowchart illustrating a method for training an artificial intelligence model in an image processing apparatus according to another embodiment of the present disclosure;

[0051] Figure 13 Is a block diagram illustrating an image processing apparatus according to still another embodiment of the present disclosure;

[0052] Figure 14 Illustrates training data according to still another embodiment of the present disclosure;

[0053] Figure 15 Illustrates an artificial intelligence model according to still another embodiment of the present disclosure;

[0054] Figure 16a An image generated by image processing of an image processing apparatus according to still another embodiment of the present disclosure;

[0055] Figure 16b Illustrates the differences in the image generated by the image processing of the image processing apparatus according to yet another embodiment of the present disclosure; and

[0056] Figure 17 Is a flowchart illustrating a method for processing an active material image in an image processing apparatus according to yet another embodiment of the present disclosure.

[0057] For the explanation of the drawings, the same or similar reference numerals may be used for the same or similar components. Detailed Description

[0058] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. However, the description is not intended to limit the present invention to specific embodiments, but should be understood to include various modifications, equivalents, and / or alternatives of the embodiments described herein.

[0059] The embodiments and terms used in this document are not intended to limit the technical features disclosed in this document to specific embodiments, but should be understood to cover various modifications, equivalents, or alternatives of the disclosed embodiments. In the description with reference to the accompanying drawings, like reference numerals may be used to represent like or related elements. Unless otherwise clearly specified in the relevant context, the singular form of a noun corresponding to an item may include one item or multiple items.

[0060] In this document, each of the phrases such as "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" may include any item listed together in the corresponding phrase or any possible combination thereof. Unless otherwise clearly stated, terms such as "the first", "the second", "first", "second", "A", "B", "(a)", or "(b)" may be used only to distinguish these components from other components and do not otherwise limit the corresponding components (such as in terms of importance or order).

[0061] In this document, when a (for example, first) component is referred to as being "connected", "coupled", or "accessed" to another (for example, second) component, with or without the terms "functionally" or "communicatively", it means that the component can be directly (for example, wired or wirelessly) or indirectly (for example, through a third component) connected to the other component.

[0062] The methods of the various embodiments disclosed in this document may be included and provided 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 device-readable storage medium (for example, a compact disc read-only memory, CD-ROM), or may be distributed directly, online, between two user devices, or through an application store (for example, downloaded or uploaded). In the case of online distribution, at least a part of the computer program product may be temporarily stored or temporarily created in a device-readable storage medium, such as in the memory of a manufacturer's server, an application store server, or a relay server.

[0063] According to embodiments disclosed in this document, each of the described components (e.g., modules or programs) may include one or more entities, and some of the multiple entities may be individually arranged in other components. According to embodiments disclosed in this document, one or more components or operations may be omitted, or one or more other components or operations may be added. Alternatively or additionally, multiple components (e.g., modules or programs) may be integrated into a single component. In this case, the integrated component may perform one or more functions of each of the multiple components in the same or similar manner as the corresponding components among the multiple components before integration. According to embodiments disclosed in this document, the operations performed by a module, program, or other component are executed sequentially, in parallel, iteratively, or heuristically, or one or more of the operations may be executed in a different order, one or more of the operations may be omitted, or one or more other operations may be added.

[0064] Figure 1 is a block diagram illustrating an image processing apparatus 101 according to an embodiment of the present disclosure.

[0065] Refer to Figure 1 , the image processing apparatus 101 may be connected to the image acquisition device 103 through a wired and / or wireless connection.

[0066] According to an embodiment, the connection 105 between the image processing apparatus 101 and the image acquisition device 103 may be a communication link established through a wired and / or wireless network. In an embodiment, the wired network may be based on a local area network (LAN) or power line communication. In an embodiment, the wireless network may be based on a short-range communication network (e.g., Bluetooth, Wi-Fi (Wireless Fidelity), Infrared Data Association (IrDA)) or a long-range communication network (e.g., a cellular network including fourth-generation (4G) and 5G networks).

[0067] According to another embodiment, the connection 105 between the image processing apparatus 101 and the image acquisition device 103 may be a connection established through an inter-device communication interface (e.g., a bus, General-Purpose Input / Output (GPIO), Serial Peripheral Interface (SPI), or Mobile Industry Processor Interface (MIPI)).

[0068] In an embodiment, the image acquisition device 103 may be a microscope (e.g., a scanning electron microscope). In an embodiment, the image acquisition device 103 may be a device that scans a focused electron beam onto the surface of a sample and converts secondary electrons generated by the interaction between the electron beam and the sample into an image signal to obtain an image of the sample surface.

[0069] In an embodiment, the image acquisition device 103 may acquire an image of an active material and / or a precursor. Hereinafter, the image of the active material and / or the precursor may be referred to as an active material image. However, referring to the image as an active material image does not exclude an image of a precursor.

[0070] In an embodiment, the image acquisition device 103 may acquire an active image of the active material. For example, the image acquisition device 103 may scan an electron beam onto the positive or negative active material powder to acquire a SEM image. That is, the SEM image may include images of the positive and negative active materials. Depending on the embodiment, the SEM image may be replaced by an image based on a transmission electron microscope (TEM), an optical microscope (OM), a structured illumination microscopy (SIM), or a focused ion beam (FIB).

[0071] In an embodiment, the image acquisition device 103 may transmit the active material image of the active material to the image processing device 101. For example, the image acquisition device 103 may transmit the active material image of the active material to the image processing device 101 through the connection 105.

[0072] In an embodiment, the image processing device 101 may be a mobile device (e.g., a phone, a laptop, a smartphone, and a smart tablet) or a computer (e.g., a general-purpose computer, a special-purpose computer).

[0073] Referring to Figure 1 , the image processing device 101 may include a communication circuit 110, a memory 120, and a processor 130. According to an embodiment, Figure 1 the image processing device 101 shown in Figure 1 may further include at least one additional component (e.g., a display, an input device, or an output device) in addition to the illustrated components.

[0074] According to an embodiment, the communication circuit 110 may establish a wired and / or wireless communication channel between the image processing device 101 and the image acquisition device 103, and exchange data with the image acquisition device 103 through the established communication channel.

[0075] In an embodiment, the memory 120 may include a volatile memory and / or a non-volatile memory.

[0076] In an 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 125 (or related instructions), input data, or output data. In an embodiment, the instructions may be executed by the processor 130 to cause the image processing device 101 to perform operations defined by the instructions.

[0077] In an embodiment, the memory 120 may include programs 125 (e.g., an artificial intelligence model training unit 141, an artificial intelligence model 145, an image acquisition unit 150, an image generation unit 160, an object recognition unit 170, an image segmentation unit 180, and / or an information extraction unit 190).

[0078] In an 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.

[0079] In an embodiment, the processor 130 may execute programs 125 (e.g., an artificial intelligence model training unit 141, an artificial intelligence model 145, an image acquisition unit 150, an image generation unit 160, an object recognition 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) connected to the processor 130 in the image processing apparatus 101 and perform various data processing or operations.

[0080] In an embodiment, the artificial intelligence model training unit 141 may train the artificial intelligence model 145 based on training data. In an embodiment, the artificial intelligence model 145 may be a model trained to convert an active material image into a binary image. In an embodiment, the image acquisition unit 150 may acquire an active material image of the active material from the image acquisition device 103. In an embodiment, the image generation unit 160 may generate a binary image by inputting the active material image into the artificial intelligence model 145. In an embodiment, the object recognition unit 170 may recognize a plurality of objects included in the binary image. In an embodiment, the image segmentation unit 180 may obtain a segmented image by segmenting a plurality of active material objects included in the active material image based on the recognized objects. In an embodiment, the information extraction unit 190 may extract information about the active material based on the active material objects included in the segmented image.

[0081] In the following, reference will be made to Figures 2 to 4 A method for the image processing apparatus 101 to process an image acquired from the image acquisition device 103 via the artificial intelligence model training unit 141, the artificial intelligence model 145, the image acquisition unit 150, the image generation unit 160, the object recognition unit 170, the image segmentation unit 180, and / or the information extraction unit 190 will be described.

[0082] Training data

[0083] Figure 2 Illustrates training data according to an embodiment of the present disclosure.

[0084] Reference Figure 2 The training data may include reference active material images 211, 213, 215, 221, 223, 225, 231, 233, and 235, and reference binary images 251, 253, 255, 261, 263, 265, 271, 273, and 275.

[0085] The reference active material images 211, 213, 215, 221, 223, 225, 231, 233, 235 and the reference binary images 251, 253, 255, 261, 263, 265, 271, 273, 275 included in the training data may be of the same size. In an embodiment, the size of the image may be defined as the number of horizontal pixels × the number of vertical pixels. In the training data, the number of horizontal pixels 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.

[0086] In an embodiment, the multiple reference active material images 211, 213, 215, 221, 223, 225, 231, 233, and 235 may include one or more first reference active material images and one or more second reference active material images.

[0087] In an embodiment, the first reference active material image may be obtained by capturing an image of the active material powder. In an embodiment, the first reference active material image may be a SEM image directly obtained by the image acquisition device 103.

[0088] In an embodiment, the second reference active material image may be an image derived or modified from the first reference active material image. In an embodiment, the second reference active material image may be generated by applying a predetermined first image processing algorithm to the first reference active material image. Here, the first image processing algorithm may include rotation, tilting, shearing, brightness adjustment, contrast adjustment, magnification, reduction, or a combination thereof.

[0089] In an embodiment, the multiple reference binary images 251, 253, 255, 261, 263, 265, 271, 273, and 275 may include one or more first reference binary images and one or more second reference binary images.

[0090] In an embodiment, the first reference binary image 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, an edge detection algorithm, an edge removal algorithm, a binarization algorithm, or any combination thereof.

[0091] In an embodiment, the second reference binary image may be an image derived from or modified from the first reference active material image. In an embodiment, the second reference binary image may be generated by applying a predetermined first image processing algorithm to the first reference binary image. Here, the first image processing algorithm may include rotation, tilting, shearing, brightness adjustment, contrast adjustment, magnification, reduction, or any combination thereof.

[0092] In an embodiment, the first reference binary image may be a binary image of the first reference active material image, and the second reference binary image may be a binary image of the second reference active material image. For example, the reference binary image 251 may be a binary image of the reference active material image 211, the reference binary image 263 may be a binary image of the reference active material image 223, and the reference binary image 275 may be a binary image of the reference active material image 265. Thus, multiple reference active material images 211, 213, 215, 221, 223, 225, 231, 233, and 235 and multiple reference binary images 251, 253, 255, 261, 263, 265, 271, 273, and 275 may be classified into image sets of corresponding images. For example, the reference active material image 211 and the reference binary image 251 may be classified into one image set.

[0093] In an embodiment, training data may be used to train the artificial intelligence model 145. In an embodiment, the training data may be used to train the artificial intelligence model 145 for a predetermined number of epochs. Here, the predetermined number may be determined between 100 and 10,000. For example, the predetermined number may be 3000.

[0094] In an embodiment, the training data may be divided into mini - batches of a predetermined 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 mini - batch may be composed of 3 image sets (i.e., 3 reference active material images and 3 reference binary images). For example, the reference active material images 211, 213, and 215 and the reference binary images 251, 253, and 255 may form the first mini - batch, the reference active material images 221, 223, and 225 and the reference binary images 261, 263, and 265 may form the second mini - batch, and the reference active material images 231, 233, and 235 and the reference binary images 271, 273, and 275 may form the third mini - batch.

[0095] Artificial intelligence model

[0096] Figure 3 FIG. shows the artificial intelligence model 145 according to an embodiment of the present disclosure. Specifically, Figure 3An example of training the artificial intelligence model 145 with a small batch consisting of reference active material images 301, 303, and 305 and reference binary images 391, 393, and 395 is shown.

[0097] In an embodiment, the artificial intelligence model 145 can be a model based on a convolutional neural network (CNN) or a U-Net. In an embodiment, the artificial intelligence model 145 can be a model trained to convert an active material image into a binary image.

[0098] Reference Figure 3 , the artificial intelligence model 145 can include multiple layers 310, 320, 330, 340, 350, 360, and 370. The multiple layers 310, 320, 330, 340, 350, 360, and 370 can be sequentially connected. The input of the sequentially connected layers is the output of the previous layer, and the output of the sequentially connected layers can be the input of the subsequent layer. For example, the output of layer 310 can be the input of layer 320.

[0099] At least two of the multiple layers 310, 320, 330, 340, 350, 360, and 370 (e.g., layers 310 and 370, layers 320 and 360, or layers 330 and 350) can be connected by skip connections 315, 325, and 335. In an embodiment, the skip connections 315, 325, and 335 can connect the layers included in the encoding region among the multiple layers 310, 320, 330, 340, 350, 360, and 370 (e.g., layers 310, 320, and 330) to the layers included in the decoding region (e.g., layers 350, 360, and 370). Here, the skip connections 315, 325, and 335 can be intermediate layer connections for inputting the outputs of layers 310, 320, and 330 into layers 350, 360, and 370. For example, through the skip connection 335, layer 350 can receive the output of layer 340 and the output of layer 330 as inputs.

[0100] In an embodiment, the multiple layers 310, 320, 330, 340, 350, 360, and 370 can include an input layer, a batch normalization layer, a 2D convolutional layer, an activation layer, a max pooling layer, an upsampling layer, a concatenation layer, or any combination thereof. In an embodiment, the layers included in the encoding region (e.g., layers 310, 320, and 330) can have a structure in which a 2D convolutional layer, a batch normalization layer, an activation layer, and a max pooling layer are sequentially connected. Similarly, the layers included in the decoding region (e.g., layers 350, 360, and 370) can have a structure in which a 2D convolutional layer, a batch normalization layer, an activation layer, and an upsampling layer are sequentially connected.

[0101] In an embodiment, the input layer may be a layer that receives bit values included in the reference active material images 301, 303, and 305. In an embodiment, the input layer may obtain an input composed of bit values corresponding 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 the depth.

[0102] In an embodiment, the batch normalization layer may be a layer that normalizes the output values of the previous layer on a batch basis. Depending on the embodiment, the batch normalization layer may be placed immediately before the activation layer. Here, a batch may include the reference active material images input to the artificial intelligence model 145 during one iteration to train the model. 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).

[0103] In an embodiment, the 2D convolutional layer may be a layer that performs a convolution operation on the input and a filter of a predetermined size to obtain an output. In an embodiment, the 2D convolutional layer included in the artificial intelligence model 145 assumes that the output image size (the number of horizontal pixels × the number of vertical pixels) is the same as the input image size (the number of horizontal pixels × the number of vertical pixels). In an embodiment, the number of channels in the output may vary depending on the number of filters (or depth) in the 2D convolutional layer. For example, when the number of filters is 2, the number of channels in the output may increase by two times compared to the number of channels in the input. In an embodiment, the number of channels in the output may vary depending on the stride in the depth direction of the 2D convolutional layer. For example, when the stride in the depth direction is 2, the number of channels in the output may be reduced by half compared to the number of channels in the input.

[0104] In an embodiment, the activation layer may be a layer that applies a predetermined activation function to the input to obtain an output. For example, the predetermined activation function may include step, sigmoid, rectifier linear unit (ReLU), exponential linear unit (ELU), softmax, or any combination thereof.

[0105] In an embodiment, the max pooling layer can be a layer that selects the maximum value in each pooling region of the input to obtain an output. In an embodiment, the size of the pooling region in the max pooling layer may affect the size of the output image (i.e., the number of horizontal and vertical pixels). For example, when the pooling region size is 2×2, the output image size (e.g., 128×128) may be reduced by half compared to the input image size (e.g., 256×256). Depending on the embodiment, the artificial intelligence model 145 may include pooling layers other than the max pooling layer. For example, another type of pooling layer may include an average pooling layer.

[0106] In an embodiment, the upsampling layer can be a layer for increasing the image resolution. In an embodiment, the upsampling layer can be a layer that uses a predetermined interpolation algorithm to increase the size of the output image compared to the input.

[0107] In an embodiment, the concatenation layer can concatenate two or more inputs to produce an output. Here, the concatenation can be depthwise concatenation. For example, when the first input is 128×128×128 and the second input is 128×128×128, the output of the concatenation layer can be 128×128×256. When the first input is 256×256×64 and the second input is 128×128×128, the output of the concatenation layer can be 256×256×128.

[0108] Although Figure 3 seven layers 310, 320, 330, 340, 350, 360, and 370 are depicted, this is only an illustrative example, and the number of layers is not limited to 7. For example, the number of layers included in the artificial intelligence model 145 can be 32.

[0109] When the number of layers included in the artificial intelligence model 145 is 32, the artificial intelligence model 145 can have a structure in which the input layer, the first batch normalization layer, the first 2D convolutional layer, and the second 2D convolutional layer are sequentially connected. The input of the sequentially connected layers is the output of the previous layer, and the output of the sequentially connected layers can be the input of the subsequent layer. Here, the input of the input layer can be 256×256×1, and the output can be 256×256×1. The output of the first batch normalization layer can be 256×256×1. The output of the first 2D convolutional layer can be 256×256×64. The output of the second 2D convolutional layer can be 256×556×64.

[0110] After the second 2D convolutional layer, the artificial intelligence model 145 can 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 can be 256×256×64. The output of the first activation layer can be 256×256×64. The output of the first max pooling layer can be 128×128×64.

[0111] After the first max pooling layer, the artificial intelligence model 145 can have a structure in which a third 2D convolutional layer, a fourth 2D convolutional 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 convolutional layer can be 128×128×128. The output of the fourth 2D convolutional layer can be 128×128×128. The output of the third batch normalization layer can be 128×128×128. The output of the second activation layer can be 128×128×128. The output of the second max pooling layer can be 64×64×128.

[0112] After the second max pooling layer, the artificial intelligence model 145 can have a structure in which a fifth 2D convolutional layer, a sixth 2D convolutional layer, a fourth batch normalization layer, a third activation layer, and a first upsampling layer are sequentially connected. Here, the output of the fifth 2D convolutional layer can be 64×64×256. The output of the sixth 2D convolutional layer can be 64×64×256. The output of the fourth batch normalization layer can be 64×64×256. The output of the third activation layer can be 64×64×256. The output of the first upsampling layer can be 128×128×256.

[0113] After the first upsampling layer, the artificial intelligence model 145 can have a structure in which a seventh 2D convolutional layer, a first concatenation layer, an eighth 2D convolutional layer, a ninth 2D convolutional layer, a fifth batch normalization layer, a fourth activation layer, and a second upsampling layer are sequentially connected. Among the two inputs of the first concatenation layer, the first input can be the output of the seventh 2D convolutional layer, and the second input can be the output of the fourth 2D convolutional layer. Here, the output of the seventh 2D convolutional layer can be 128×128×128. The output of the first concatenation layer can be 128×128×256. The output of the eighth 2D convolutional layer can be 128×128×128. The output of the ninth 2D convolutional layer can 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.

[0114] After the second upsampling layer, the artificial intelligence model 145 may have a structure in which the tenth 2D convolutional layer, the second concatenation layer, the eleventh 2D convolutional layer, the twelfth 2D convolutional layer, the sixth batch normalization layer, the fifth activation layer, the thirteenth 2D convolutional layer, and the fourteenth 2D convolutional layer are sequentially connected. Among the two inputs to the second concatenation layer, the first input may be the output of the tenth 2D convolutional layer, and the second input may be the output of the second 2D convolutional layer. Here, the output of the tenth 2D convolutional layer may be 256×256×64. The output of the second concatenation layer may be 256×256×128. The output of the eleventh 2D convolutional layer may be 256×256×64. The output of the twelfth 2D convolutional layer may be 256×256×64. The output of the sixth batch normalization layer may be 256×256×64. The output of the fifth activation layer may be 256×256×64. The output of the thirteenth 2D convolutional layer may be 256×256×2. The output of the fourteenth 2D convolutional layer may be 256×256×2.

[0115] Artificial intelligence model learning

[0116] Reference Figure 3 , the operations of the artificial intelligence model training unit 141 for training the artificial intelligence model 145 by using the images 301, 303, 305, 391, 393, and 395 included in the mini-batch are described.

[0117] In an embodiment, the artificial intelligence model training unit 141 may sequentially input the reference active material images 301, 303, and 305 into the artificial intelligence model 145.

[0118] In an embodiment, the artificial intelligence model training unit 141 may train the artificial intelligence model 145 to ensure that the difference between the sequentially obtained output images of the artificial intelligence model 145 and the reference binary images 391, 393, and 395 is equal to or less than a predetermined threshold.

[0119] For example, the artificial intelligence model training unit 141 may calculate the difference between the output image and the reference binary images 391, 393, and 395 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. Depending on the embodiment, the binary cross-entropy loss function may be a function with a weight between 1 and 1000 applied to one of two colors (or categories) (e.g., white in black and white).

[0120] Subsequently, the artificial intelligence model training unit 141 may adjust the weights of the artificial intelligence model 145 to ensure that the difference is equal to or less than a predetermined threshold (or reaches a minimum value). In an embodiment, the artificial intelligence model training unit 141 may use an algorithm based on gradient descent (e.g., Adam, SGD, and Momentum) to adjust the weights of the artificial intelligence model 145 to ensure that the difference is equal to or less than a predetermined threshold (or has a minimum value). Here, the adjusted weights may be determined based on a learning rate. The learning rate may be determined within a range of 0.00000001 to 0.1. For example, the learning rate may be 0.0001.

[0121] Subsequently, the artificial intelligence model training unit 141 may use the images included in the next mini - batch to train the artificial intelligence model 145. After training the artificial intelligence model 145 using all the images included in the mini - batch, the artificial intelligence model training unit 141 may retrain the artificial intelligence model 145 based on a predetermined number of epochs using the images included in the mini - batch.

[0122] Image processing by artificial intelligence model

[0123] Figure 4 FIG. 410, 420, and 430 illustrate images generated by the image processing of the image processing apparatus 101 according to an embodiment of the present disclosure.

[0124] In an embodiment, the image acquisition unit 150 may acquire the active material images 410 of the active materials 411, 413, and 415. In an embodiment, the image acquisition unit 150 may acquire the active material image 410 through the image acquisition device 103.

[0125] In an embodiment, the image generation unit 160 may generate the binary image 430. In an embodiment, the image generation unit 160 may generate the binary image 430 by inputting the active material image 410 into the artificial intelligence model 145.

[0126] In an embodiment, the object recognition unit 170 may recognize a plurality of objects 431, 433, and 435 included in the binary image 430. Here, the plurality of objects 431, 433, and 435 included in the binary image 430 may correspond to the active materials 411, 413, and 415 in the active material image 410. In an embodiment, the plurality of objects 431, 433, and 435 may be composed of regions having a predetermined value (e.g., a value representing white). In an embodiment, the plurality of objects 431, 433, and 435 may be distinguished by regions having different predetermined values (e.g., a value representing black). Although only three objects are assigned reference numerals in Figure 4 only three objects are assigned reference numerals, it can be seen that in Figure 4There are more objects in the binary image 430.

[0127] In an embodiment, the image segmentation unit 180 may segment the plurality of active materials 411, 413, and 415 included in the active material image 410 based on the plurality of objects 431, 433, and 435 to obtain a segmented image 450. In an embodiment, the image segmentation unit 180 may segment the plurality of active materials 411, 413, and 415 in the active material image 410 based on a watershed algorithm.

[0128] In an embodiment, the information extraction unit 190 may extract information about the active materials 411, 413, and 415 based on the active material objects 451, 453, and 455 included in the segmented image 450. In an embodiment, the information about the active materials 411, 413, and 415 may include information about the average diameter, average perimeter, average sphericity, average aspect ratio, average convexity, average solidity, or distribution value of these properties. Additionally, the information about the active materials 411, 413, and 415 may include information about the individual properties of each active material - such as diameter, perimeter, sphericity, aspect ratio, convexity, and / or solidity. Furthermore, the information about the active materials 411, 413, and 415 may include information about the percentile values (e.g., 1% to 99%, or D5, D50, D95, etc.) for each active material.

[0129] Figure 5 is a flowchart illustrating a method for processing an active material image of an image processing apparatus 101 according to an embodiment of the present disclosure.

[0130] Refer to Figure 5 , the image acquisition device 101 may acquire the active material image 410 of the active materials 411, 413, and 415 in operation 510. In an embodiment, the image acquisition device 101 may acquire the active material image 410 through the image acquisition device 103.

[0131] In operation 520, the image processing apparatus 101 may generate a binary image 430 by inputting the active material image 410 into the artificial intelligence model 145. Here, the artificial intelligence model 145 may be a model trained by Figure 6 the illustrated artificial intelligence model training method.

[0132] In operation 530, the image processing apparatus 101 may identify the plurality of objects 431, 433, and 435 included in the binary image 430.

[0133] In operation 540, the image processing apparatus 101 may segment the plurality of active materials 411, 413, and 415 included in the active material image 410 based on the plurality of objects 431, 433, and 435 to obtain a segmented image 450. In an embodiment, the image processing apparatus 101 may segment the plurality of active materials 411, 413, and 415 in the active material image 410 based on a watershed algorithm.

[0134] Figure 6 is a flowchart illustrating a method for training an artificial intelligence model of the image processing apparatus 101 according to an embodiment of the present disclosure. Figure 6 The operation of can be applied during the first round of training data.

[0135] Reference Figure 6 , the image processing apparatus 101 may set parameters of the artificial intelligence model 145 in operation 610. In an embodiment, the image processing apparatus 101 may set parameters of the artificial intelligence model 145 based on initial parameter values.

[0136] In operation 615, the image processing apparatus 101 may set i to 1. Here, i may represent an index of an image set included in the training data. The image set may include a reference active material image and a corresponding reference binary image.

[0137] In operation 620, the image processing apparatus 101 may obtain an output image based on the i-th reference active material image. In an embodiment, the image processing apparatus 101 may generate an output image by inputting the i-th reference active material image into the artificial intelligence model 145.

[0138] In operation 630, the image processing apparatus 101 may determine whether the difference between the output image and the i-th reference binary image is equal to or less than a predetermined threshold. In an embodiment, the image processing apparatus 101 may calculate the difference between the output image and the i-th 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. Depending on the embodiment, the binary cross-entropy loss function may be a function having a weight between 1 and 1000 applied to one of two colors (or classes) (e.g., white among black and white).

[0139] Based on the determination result in operation 630, when the difference is equal to or less than the predetermined threshold (yes), the image processing apparatus 101 may proceed to operation 650. Based on the determination result in operation 630, when the difference is greater than the threshold (no), the image processing apparatus 101 may proceed to operation 640.

[0140] In operation 640, the image processing apparatus 101 may adjust the parameters of the artificial intelligence model 145. In an embodiment, the image processing apparatus 101 may adjust the parameters of the artificial intelligence model 145 to ensure that the difference is equal to or less than a predetermined threshold (or has a minimum value). In an embodiment, the image processing apparatus 101 may use an algorithm based on gradient descent (e.g., Adam, SGD, Momentum) to adjust the parameters of the artificial intelligence model 145 to ensure that the difference is equal to or less than a predetermined threshold (or has a minimum value). Here, the adjusted weights may be determined based on a learning rate. The learning rate may be determined within a range of 0.00000001 to 0.1. For example, the learning rate may be 0.0001.

[0141] In operation 650, the image processing apparatus 101 may determine whether the training is completed. In an embodiment, when i represents the end of the image set, it may be determined that the training is completed.

[0142] Based on the determination result in operation 650, when the training is completed (yes), the image processing apparatus 101 may terminate Figure 6 the operations therein. Based on the determination result in operation 650, when the training is not completed (no), the image processing apparatus 101 may proceed to operation 655.

[0143] In operation 655, the image processing apparatus 101 may increment the value of i by 1. After that, operation 620 may be executed again.

[0144] Although the image processing apparatus 101 is exemplified in the present disclosure as executing an image processing method based on an active material image, this is only an exemplary example. Depending on the embodiment, the image processing apparatus 101 may separate and / or analyze not only the particles in the active material image but also the particles in the precursor image by performing the same image processing. In an embodiment, the image acquisition device 103 may acquire an image of the active material and / or the precursor.

[0145] In addition, although the active material image is exemplified as an SEM image in the present disclosure, this is only an exemplary example. Depending on the embodiment, the active material image may be replaced with an image based on a transmission electron microscope (TEM), an optical microscope (OM), a structured illumination microscope (SIM), or a focused ion beam (FIB).

[0146] The above image processing apparatus 101 and image processing method can automatically quantify the shape characteristics of a large number of positive electrode active material particles by obtaining active material images 410 of a plurality of active materials 411, 413, 415, inputting the active material images 410 into an artificial intelligence model 145 to generate a binary image 430, identifying a plurality of objects 431, 433, 435 included in the binary image 430, and segmenting the plurality of active materials 411, 413, 415 included in the active material image 410 based on the plurality of objects 431, 433, 435.

[0147] In addition, the above image processing apparatus 101 and image processing method can alleviate the problem of measurement variations depending on the user by using the artificial intelligence model 145 that minimizes the need for user input.

[0148] Meanwhile, although the specification has provided illustrative examples of positive electrode active material images, the technical principles disclosed in this document can be applied in a substantially similar manner even if the active material image is a negative electrode active material image.

[0149] Figure 7 FIG. is a block diagram illustrating an image processing apparatus 701 according to another embodiment of the present disclosure.

[0150] Reference Figure 7 , the image processing apparatus 701 can be connected to the image acquisition device 703 through a wired and / or wireless connection.

[0151] According to an embodiment, the connection 705 between the image processing apparatus 701 and the image acquisition device 703 can be a communication link established through a wired and / or wireless network. In an embodiment, the wired network can be based on a local area network (LAN) or power line communication. In an embodiment, the wireless network can be based on a short-range communication network (e.g., Bluetooth, Wi-Fi, Infrared Data Association (IrDA)) or a long-range communication network (e.g., a cellular network including fourth-generation (4G) and 5G networks).

[0152] According to another embodiment, the connection 705 between the image processing apparatus 701 and the image acquisition device 703 can be a connection established through an inter-device communication interface (e.g., a bus, General-Purpose Input / Output (GPIO), Serial Peripheral Interface (SPI), or Mobile Industry Processor Interface (MIPI)).

[0153] In an embodiment, the image acquisition device 703 can acquire an image of an active material and / or a precursor. Hereinafter, the image of the active material and / or the precursor may be referred to as an active material image. However, referring to the image as an active material image does not exclude an image of a precursor.

[0154] In an embodiment, the image acquisition device 703 can be a microscope (e.g., a scanning electron microscope). In an embodiment, the image acquisition device 703 can be a device that scans a focused electron beam onto the surface of a sample and converts secondary electrons generated by the interaction between the electron beam and the sample into an image signal to obtain an image of the sample surface.

[0155] In an embodiment, the image acquisition device 703 can acquire an active image of the active material. For example, the image acquisition device 703 can scan an electron beam onto the positive or negative active material powder to obtain a SEM image. That is, the SEM image can include images of the positive and negative active materials. Depending on the embodiment, the SEM image can be replaced with an image based on a transmission electron microscope (TEM), an optical microscope (OM), structured illumination microscopy (SIM), or a focused ion beam (FIB).

[0156] In an embodiment, the image acquisition device 703 can transmit the active material image of the active material to the image processing device 701. For example, the image acquisition device 703 can transmit the active material image of the active material to the image processing device 701 through the connection 705.

[0157] In an embodiment, the image processing device 701 can be a mobile device (e.g., a phone, a laptop, a smartphone, and a smart tablet) or a computer (e.g., a general-purpose computer, a dedicated computer).

[0158] Reference Figure 1 , the image processing device 701 can include a communication circuit 710, a memory 720, and a processor 730. According to an embodiment, Figure 7 The image processing device 701 shown in Figure 7 In addition to the illustrated components, it can also include at least one additional component (e.g., a display, an input device, or an output device).

[0159] According to an embodiment, the communication circuit 710 can establish a wired and / or wireless communication channel between the image processing device 701 and the image acquisition device 703 and exchange data with the image acquisition device 703 through the established communication channel.

[0160] In an embodiment, the memory 720 can include a volatile memory and / or a non-volatile memory.

[0161] In an embodiment, the memory 720 can store data used by at least one component (e.g., the processor 730) of the image processing device 701. For example, the data can include a program 725 (or related instructions), input data, or output data. In an embodiment, the instructions can be executed by the processor 730 to cause the image processing device 701 to perform operations defined by the instructions.

[0162] In an embodiment, the memory 720 may include a program 725 (e.g., an artificial intelligence model training unit 741, an artificial intelligence model 745, an image acquisition unit 750, an image generation unit 760, an edge removal unit 765, an object recognition unit 770, an image segmentation unit 780, and / or an information extraction unit 790).

[0163] In an 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.

[0164] In an embodiment, the processor 730 may execute the program 725 (e.g., an artificial intelligence model training unit 741, an artificial intelligence model 745, an image acquisition unit 750, an image generation unit 760, an edge removal unit 765, an object recognition unit 770, an image segmentation unit 780, and / or an information extraction unit 790) to control at least one other component (e.g., a hardware or software component) connected to the processor 730 in the image processing apparatus 701 and perform various data processing or operations.

[0165] In an embodiment, the artificial intelligence model training unit 741 may train the artificial intelligence model 745 based on training data. In an embodiment, the artificial intelligence model 745 may be a model trained to convert an active material image into a binary image. In an embodiment, the image acquisition unit 750 may acquire an active material image of the active material from the image acquisition device 703. In an embodiment, the image generation unit 760 may generate an edge image by inputting the active material image into the artificial intelligence model 745. In an embodiment, the edge removal unit 765 may generate an edge-removed image by removing a boundary from the active material image based on the edge image. In an embodiment, the object recognition unit 770 may recognize a plurality of objects included in the edge-removed image. In an embodiment, the image segmentation unit 780 may obtain a segmented image by segmenting a plurality of active materials included in the active material image based on the plurality of objects. In an embodiment, the information extraction unit 790 may extract information about the active material based on the active material objects included in the segmented image.

[0166] Reference will be made below to Figure 8 、 Figure 9 、 Figure 10a and Figure 10bA method for an image processing apparatus 701 to process an image acquired from an image acquisition device 703 via an artificial intelligence model training unit 741, an artificial intelligence model 745, an image acquisition unit 750, an image generation unit 760, a boundary removal unit 765, an object recognition unit 770, an image segmentation unit 780, and / or an information extraction unit 790 is described.

[0167] Training data

[0168] Figure 8 The figure shows training data according to another embodiment of the present disclosure.

[0169] Reference Figure 8 , the training data may include reference active material images 811, 812, 813, 814, 815, 818, 819, 820, and 821 and reference edge images 851, 852, 853, 854, 855, 856, 857, 858, 859, 860, and 861.

[0170] The reference active material images 811, 812, 813, 814, 815, 816, 817, 818, 819, 820, 821 and the reference edge images 851, 852, 853, 854, 855, 856, 857, 858, 859, 860, 861 included in the training data may have the same size. In an embodiment, the size of an image may be defined as the number of horizontal pixels × the number of vertical pixels. In the training data, the number of horizontal pixels 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.

[0171] In an embodiment, the multiple reference active material images 811, 812, 813, 814, 815, 816, 817, 818, 819, 820, and 821 may include one or more first reference active material images and one or more second reference active material images.

[0172] In an embodiment, a first reference active material image may be obtained by capturing an image of active material powder. In an embodiment, the first reference active material image may be an SEM image directly obtained by the image acquisition device 703.

[0173] In an embodiment, a second reference active material image may be an image derived or modified from the first reference active material image. In an embodiment, the second reference active material image may be generated by applying a predetermined first image processing algorithm to the first reference active material image. Here, the first image processing algorithm may include rotation, tilting, shearing, brightness adjustment, contrast adjustment, magnification, reduction, or a combination thereof.

[0174] In an embodiment, the multiple reference edge images 851, 852, 853, 854, 855, 856, 857, 858, 859, 860, and 861 may include one or more first reference edge images and one or more second reference edge images.

[0175] In an embodiment, the first reference edge image 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 and / or an edge detection algorithm.

[0176] In an embodiment, the second reference edge image may be an image derived from or modified from the first reference edge material image. In an embodiment, the second reference edge image may be generated by applying a predetermined first image processing algorithm to the first reference edge image. Here, the first image processing algorithm may include rotation, skew, shear, brightness adjustment, contrast adjustment, enlargement, reduction, or any combination thereof.

[0177] In an embodiment, the first reference edge image may be an edge image of the first reference active material image, and the second reference edge image may be an edge image of the second reference active material image. For example, the reference edge image 851 may be an edge image of the reference active material image 811, the reference edge image 856 may be an edge image of the reference active material image 816, and the reference edge image 861 may be an edge image of the reference active material image 821. Thus, the multiple reference active material images 811, 812, 813, 814, 815, 816, 817, 818, 819, 820, and 821 and the multiple reference edge images 851, 852, 853, 854, 855, 856, 857, 858, 859, 860, and 861 may be classified into image sets of corresponding images. For example, the reference active material image 811 and the reference edge image 851 may be classified into one image set.

[0178] In an embodiment, the training data may be used to train the artificial intelligence model 745. In an embodiment, the training data may be used to train the artificial intelligence model 745 for a predetermined number of rounds. Here, the predetermined number may be determined between 100 and 10,000. For example, the predetermined number may be 3,000.

[0179] In an embodiment, the training data can be divided into small batches with a predetermined batch size. Here, the batch size can be determined between 1 and 512. For example, the batch size can be 4. When the batch size is 4, each small batch can be composed of 4 sets of images (i.e., 4 reference active material images and 4 reference edge images). For example, the reference active material images 811, 812, 813, and 814 and the reference edge images 851, 852, 853, and 854 can form the first small batch, and the reference active material images 815, 816, 817, and 818 and the reference edge images 855, 856, 857, and 858 can form the second small batch.

[0180] Artificial intelligence model

[0181] Figure 9 FIG. illustrates an artificial intelligence model 745 according to another embodiment of the present disclosure. Specifically, Figure 9 An example of training the artificial intelligence model 745 with a small batch composed of the reference active material images 901 and 905 and the reference edge images 991 and 995 is shown.

[0182] In an embodiment, the artificial intelligence model 745 can be a model based on a convolutional neural network (CNN) or a U-Net. In an embodiment, the artificial intelligence model 745 can be a model trained to convert an active material image into an edge image.

[0183] Refer to Figure 9 and the artificial intelligence model 745 can include multiple layers 910, 920, 930, 940, 950, 960, and 970. The multiple layers 910, 920, 930, 940, 950, 960, and 970 can be sequentially connected. The input of the sequentially connected layers is the output of the layer before it, and the output of the sequentially connected layers can be the input of the layer after it. For example, the output of layer 910 can be the input of layer 920.

[0184] At least two of the multiple layers 910, 920, 930, 940, 950, 960, and 970 (e.g., layer 910 and 970, layer 920 and 960, or layer 930 and 950) can be connected by skip connections 915, 925, and 935. In an embodiment, the skip connections 915, 925, and 935 can connect the layers included in the encoding region among the multiple layers 910, 920, 930, 940, 950, 960, and 970 (e.g., layers 910, 920, and 930) to the layers included in the decoding region (e.g., layers 950, 960, and 970). Here, the skip connections 915, 925, and 935 can be intermediate layer connections for inputting the outputs of layers 910, 920, and 930 into layers 950, 960, and 970. For example, through the skip connection 935, layer 950 can receive the outputs of layer 940 and layer 930 as inputs.

[0185] In an embodiment, the multiple layers 910, 920, 930, 940, 950, 960, and 970 can include an input layer, a batch normalization layer, a 2D convolutional layer, an activation layer, a max pooling layer, an upsampling layer, a concatenation layer, or any combination thereof. In an embodiment, the layers included in the encoding region (e.g., layers 910, 920, and 930) can have a structure in which a 2D convolutional layer, a batch normalization layer, an activation layer, and a max pooling layer are sequentially connected. Similarly, the layers included in the decoding region (e.g., layers 950, 960, and 970) can have a structure in which a 2D convolutional layer, a batch normalization layer, an activation layer, and an upsampling layer are sequentially connected.

[0186] In an embodiment, the input layer can be a layer that receives the bit values included in the reference active material images 901 and 905. In an embodiment, the input layer can obtain an input composed of bit values corresponding 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 can receive N×N×1 bit values. Here, N can be an integer between 32 and 4096 (e.g., 256). Depending on the embodiment, the number of channels can also be referred to as the depth.

[0187] In an embodiment, the batch normalization layer can be a layer that normalizes the output values of the previous layer on a batch basis. Depending on the embodiment, the batch normalization layer can be placed only before the activation layer. Here, a batch can include the reference active material images input to the artificial intelligence model 745 during one iteration to train the model. For example, when the batch size is 4, the number of reference active material images input to the artificial intelligence model 745 during one iteration can be 4. Here, the batch size can be an integer between 1 and 512 (e.g., 4).

[0188] In an embodiment, the 2D convolutional layer may be a layer that performs a convolution operation on an input and a filter of a predetermined size to obtain an output. In an embodiment, the 2D convolutional layer included in the artificial intelligence model 745 assumes that the output image size (number of horizontal pixels × number of vertical pixels) is the same as the input image size (number of horizontal pixels × number of vertical pixels). In an embodiment, the number of channels in the output may vary depending on the number (or depth) of filters in the 2D convolutional layer. For example, when the number of filters is 2, the number of channels in the output may be doubled compared to the number of channels in the input. In an embodiment, the number of channels in the output may vary depending on the stride in the depth direction of the 2D convolutional layer. For example, when the stride in the depth direction is 2, the number of channels in the output may be reduced by half compared to the number of channels in the input.

[0189] In an embodiment, the activation layer may be a layer that applies a predetermined activation function to an input to obtain an output. For example, the predetermined activation function may include step, sigmoid, rectified linear unit (ReLU), exponential linear unit (ELU), softmax, or any combination thereof.

[0190] In an embodiment, the max pooling layer may be a layer that selects the maximum value in each pooling region of an input to obtain an output. In an embodiment, the size of the pooling region in the max pooling layer may affect the output image size (i.e., the number of horizontal and vertical pixels). For example, when the pooling region size is 2×2, the output image size (e.g., 128×128) may be reduced by half compared to the input image size (e.g., 256×256). Depending on the embodiment, the artificial intelligence model 745 may include a pooling layer other than the max pooling layer. For example, another type of pooling layer may include an average pooling layer.

[0191] In an embodiment, the upsampling layer may be a layer for increasing the image resolution. In an embodiment, the upsampling layer may be a layer that uses a predetermined interpolation algorithm to increase the image size of the output compared to the input.

[0192] In an embodiment, the concatenation layer may concatenate two or more inputs to produce an output. Here, the concatenation may be depth-wise concatenation. For example, when the first input is 128×128×128 and the second input is 128×128×128, the output of the concatenation layer may be 128×128×256. When the first input is 256×256×64 and the second input is 128×128×128, the output of the concatenation layer may be 256×256×128.

[0193] Although Figure 9Seven layers 910, 920, 930, 940, 950, 960, and 970 are depicted, but this is only an illustrative example, and the number of layers is not limited to seven. For example, the number of layers included in the artificial intelligence model 745 can be 32.

[0194] When the number of layers included in the artificial intelligence model 745 is 32, the artificial intelligence model 745 can have a structure in which an input layer, a first batch normalization layer, a first 2D convolutional layer, and a second 2D convolutional layer are sequentially connected. The input of the sequentially connected layers is the output of the previous layer, and the output of the sequentially connected layers can be the input of the subsequent layer. Here, the input of the input layer can be 256×256×1, and the output can be 256×256×1. The output of the first batch normalization layer can be 256×256×1. The output of the first 2D convolutional layer can be 256×256×64. The output of the second 2D convolutional layer can be 256×256×64.

[0195] After the second 2D convolutional layer, the artificial intelligence model 745 can 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 can be 256×256×64. The output of the first activation layer can be 256×256×64. The output of the first max pooling layer can be 128×128×64.

[0196] After the first max pooling layer, the artificial intelligence model 745 can have a structure in which a third 2D convolutional layer, a fourth 2D convolutional 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 convolutional layer can be 128×128×128. The output of the fourth 2D convolutional layer can be 128×128×128. The output of the third batch normalization layer can be 128×128×128. The output of the second activation layer can be 128×128×128. The output of the second max pooling layer can be 64×64×128.

[0197] After the second max pooling layer, the artificial intelligence model 745 can have a structure in which a fifth 2D convolutional layer, a sixth 2D convolutional layer, a fourth batch normalization layer, a third activation layer, and a first upsampling layer are sequentially connected. Here, the output of the fifth 2D convolutional layer can be 64×64×256. The output of the sixth 2D convolutional 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 can be 64×64×256. The output of the first upsampling layer can be 128×128×256.

[0198] After the first upsampling layer, the artificial intelligence model 745 may have a structure in which a seventh 2D convolutional layer, a first concatenation layer, an eighth 2D convolutional layer, a ninth 2D convolutional layer, a fifth batch normalization layer, a fourth activation layer, and a second upsampling layer are sequentially connected. Among the two inputs of the first concatenation layer, the first input may be the output of the seventh 2D convolutional layer, and the second input may be the output of the fourth 2D convolutional layer. Here, the output of the seventh 2D convolutional layer may be 128×128×128. The output of the first concatenation layer may be 128×128×256. The output of the eighth 2D convolutional layer may be 128×128×128. The output of the ninth 2D convolutional layer may be 128×128×128. The output of the fifth batch normalization layer may be 128×228×128. The output of the fourth activation layer may be 128×128×128. The output of the second upsampling layer may be 256×256×128.

[0199] After the second upsampling layer, the artificial intelligence model 745 may have a structure in which a tenth 2D convolutional layer, a second concatenation layer, an eleventh 2D convolutional layer, a twelfth 2D convolutional layer, a sixth batch normalization layer, a fifth activation layer, a thirteenth 2D convolutional layer, and a fourteenth 2D convolutional layer are sequentially connected. Among the two inputs of the second concatenation layer, the first input may be the output of the tenth 2D convolutional layer, and the second input may be the output of the second 2D convolutional layer. Here, the output of the tenth 2D convolutional layer may be 128×128×64. The output of the second concatenation layer may be 256×256×128. The output of the eleventh 2D convolutional layer may be 256×256×64. The output of the twelfth 2D convolutional layer may be 256×256×64. The output of the sixth batch normalization layer may be 256×256×64. The output of the fifth activation layer may be 256×256×64. The output of the thirteenth 2D convolutional layer may be 256×256×2. The output of the fourteenth 2D convolutional layer may be 256×256×2.

[0200] Artificial intelligence model learning

[0201] Reference Figure 9 is used to describe the operation of the artificial intelligence model training unit 741 for training the artificial intelligence model 745 by using the images 901, 905, 991, and 995 included in the mini-batch.

[0202] In an embodiment, the artificial intelligence model training unit 741 may sequentially input the reference active material images 901 and 905 into the artificial intelligence model 745.

[0203] In an embodiment, the artificial intelligence model training unit 741 may train the artificial intelligence model 745 to ensure that the difference between the output images sequentially obtained by the artificial intelligence model 745 and the reference edge images 991 and 995 is equal to or less than a predetermined threshold.

[0204] For example, the artificial intelligence model training unit 741 may calculate the difference between the output image and the reference edge images 991 and 995 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. Depending on the embodiment, the binary cross-entropy loss function may be a function having a weight between 1 and 1000 applied to one of two colors (or classes) (e.g., white in black and white).

[0205] Subsequently, the artificial intelligence model training unit 741 may adjust the weights of the artificial intelligence model 745 to ensure that the difference is equal to or less than a predetermined threshold (or reaches a minimum value). In an embodiment, the artificial intelligence model training unit 741 may use an algorithm based on gradient descent (e.g., Adam, SGD, and Momentum) to adjust the weights of the artificial intelligence model 745 to ensure that the difference is equal to or less than a predetermined threshold (or has a minimum value). Here, the adjusted weights may be determined based on a learning rate. The learning rate may be determined within a range of 0.00000001 to 0.1. For example, the learning rate may be 0.0001.

[0206] Subsequently, the artificial intelligence model training unit 741 may train the artificial intelligence model 745 using the images included in the next mini-batch. After training the artificial intelligence model 745 using all the images included in the mini-batch, the artificial intelligence model training unit 741 may re-train the artificial intelligence model 745 using the images included in the mini-batch based on a predetermined number of epochs.

[0207] Image processing by artificial intelligence model

[0208] Figure 10a Images 1010, 1020, 1030, and 1040 resulting from the image processing of an image processing apparatus according to another embodiment of the present disclosure are shown.

[0209] In an embodiment, the image acquisition unit 750 may acquire the active material images 1010 of the active materials 1011, 1013, and 1015. In an embodiment, the image acquisition unit 750 may acquire the active material images 1010 through the image acquisition device 703.

[0210] In an embodiment, the image generation unit 760 may generate an edge image 1020. In an embodiment, the image generation unit 760 may generate the edge image 1020 by inputting the active material image 1010 into the artificial intelligence model 745.

[0211] In an embodiment, the edge removal unit 765 may generate an edge-removed image 1030 by removing edges from the active material image 1010 based on the edge image 1020. In an embodiment, the edge removal unit 765 may generate the edge-removed image 1030 by subtracting the edge image 1020 from the active material image 1010. In an embodiment, the edge removal unit 765 may generate the edge-removed image 1030 by subtracting the edge image 1020 from the active material image 1010 and then binarizing the result. Depending on the embodiment, the edge-removed image may also be referred to as a binary image.

[0212] In an embodiment, the object recognition unit 770 may recognize a plurality of objects 1031, 1033, and 1035 included in the edge-removed image 1030. Here, the plurality of objects 1031, 1033, and 1035 included in the edge-removed image 1030 may correspond to the active materials 1011, 1013, and 1015 in the active material image 1010. In an embodiment, the plurality of objects 1031, 1033, and 1035 may be constituted by regions having a predetermined value (e.g., a value representing white). In an embodiment, the plurality of objects 1031, 1033, and 1035 may be distinguished by regions having different predetermined values (e.g., a value representing black). Although only three objects are assigned reference numerals in Figure 7 it can be seen that there are more objects in the edge-removed image 1030 in Figure 10a than that.

[0213] In an embodiment, the image segmentation unit 780 may segment the plurality of active materials 1011, 1013, and 1015 included in the active material image 1010 based on the plurality of objects 1031, 1033, and 1035 to obtain a segmented image 1040. In an embodiment, the image segmentation unit 780 may segment the plurality of active materials 1011, 1013, and 1015 included in the active material image 1010 based on the watershed algorithm. Here, the plurality of active materials 1011, 1013, and 1015 included in the active material image 1010 may respectively correspond to the plurality of active material objects 1041, 1043, and 1045 included in the segmented image 1040.

[0214] In an embodiment, the information extraction unit 790 may extract information about the active materials 1011, 1013, and 1015 based on the active material objects 1041, 1043, and 1045 included in the segmented image 1040. In an embodiment, the information about the active materials 1011, 1013, and 1015 may include information about the average diameter, average perimeter, average sphericity, average aspect ratio, average convexity, average solidity, or distribution value of these properties. Additionally, the information about the active materials 1011, 1013, and 1015 may include information about the individual properties of each active material, such as diameter, perimeter, sphericity, aspect ratio, convexity, and / or solidity. Further, the information about the active materials 1011, 1013, and 1015 may include information about the percentile values (e.g., 1% to 99%, or D5, D50, D95, etc.) of each active material.

[0215] Figure 10b FIG. illustrates the difference between the images generated by the image processing device 701 performing image processing according to another embodiment of the present disclosure.

[0216] Reference Figure 10b , the user can visually perceive the difference between the two active material images 1051 and 1053, but it is difficult to quantify them. However, by segmenting the active materials as shown in the two segmented images 1061 and 1063, the image processing device 701 can obtain information about the active materials, thereby allowing quantitative measurement of the active materials.

[0217] Figure 11 is a flowchart illustrating a method for processing active materials of the image processing device 101 according to another embodiment of the present disclosure.

[0218] Reference Figure 11 , the image acquisition device 701 may acquire the active material images 1010 of the active materials 1011, 1013, and 1015 in operation 1110. In an embodiment, the image acquisition device 701 may acquire the active material images 1010 through the image acquisition device 703.

[0219] In operation 1120, the image processing device 701 may generate an edge image 1020 by inputting the active material image 1010 into the artificial intelligence model 745. Here, the artificial intelligence model 745 may be a model trained by the Figure 12 illustrated artificial intelligence model training method.

[0220] In operation 1130, the image processing device 701 may generate an edge-removed image 1030 by removing edges from the active material image 1010 based on the edge image 1020. In an embodiment, the image processing device 701 may generate the edge-removed image 1030 by subtracting the edge image 1020 from the active material image 1010. In an embodiment, the image processing device 701 may generate the edge-removed image 1030 by subtracting the edge image 1020 from the active material image 1010 and binarizing the result.

[0221] In operation 1140, the image processing device 701 may identify a plurality of objects 1031, 1033, and 1035 included in the edge-removed image 1030.

[0222] In operation 1150, the image processing device 701 may segment a plurality of active materials 1011, 1013, and 1015 included in the active material image 1010 based on the plurality of objects 1031, 1033, and 1035 to obtain a segmented image 1040. In an embodiment, the image processing device 701 may segment the plurality of active materials 1011, 1013, and 1015 included in the active material image 1010 based on the watershed algorithm. Here, the plurality of objects 1011, 1013, and 1015 included in the active material image 1010 may respectively correspond to the active material objects 1041, 1043, and 1045 included in the segmented image 1040.

[0223] Figure 12 is a flowchart illustrating a method for training an intelligent model of the image processing device 701 according to another embodiment of the present disclosure. Figure 12 The operations of can be applied during the first round of training data.

[0224] Reference Figure 12 In operation 1210, the image processing device 701 may set parameters of the artificial intelligence model 745. In an embodiment, the image processing device 701 may set the parameters of the artificial intelligence model 745 based on initial parameter values.

[0225] In operation 1215, the image processing device 701 may set i to 1. Here, i may represent an index of a set of images included in the training data. The set of images may include a reference active material image and a corresponding reference edge image.

[0226] In step 120, the image processing device 701 may obtain an output image based on the i-th reference active material image. In an embodiment, the image processing device 701 may generate the output image by inputting the i-th reference active material image into the artificial intelligence model 745.

[0227] In operation 1230, the image processing apparatus 701 may determine whether the difference between the output image and the i-th reference edge image is equal to or less than a predetermined threshold. In an embodiment, the image processing apparatus 701 may calculate the difference between the output image and the i-th reference edge 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. Depending on the embodiment, the binary cross-entropy loss function may be a function having a weight between 1 and 1000 applied to one of two colors (or categories) (e.g., white among black and white).

[0228] Based on the determination result in operation 1230, when the difference is equal to or less than the predetermined threshold (yes), the image processing apparatus 701 may proceed to operation 1250. Based on the determination result in operation 1230, when the difference is greater than the threshold (no), the image processing apparatus 701 may proceed to operation 1240.

[0229] In operation 1240, the image processing apparatus 701 may adjust the parameters of the artificial intelligence model 745. In an embodiment, the image processing apparatus 701 may adjust the parameters of the artificial intelligence model 745 to ensure that the difference is equal to or less than the predetermined threshold (or has a minimum value). In an embodiment, the image processing apparatus 701 may use an algorithm based on gradient descent (e.g., Adam, SGD, Momentum) to adjust the parameters of the artificial intelligence model 745 to ensure that the difference is equal to or less than the predetermined threshold (or has a minimum value). Here, the adjusted weights may be determined based on a learning rate. The learning rate may be determined in the range of 0.00000001 to 0.1. For example, the learning rate may be 0.0001.

[0230] In operation 1250, the image processing apparatus 701 may determine that the training is completed. In an embodiment, when i represents the end of the image set, it may be determined that the training is completed.

[0231] According to the determination result in operation 1250, when the training is completed (yes), the image processing apparatus 701 may terminate Figure 12 the operations therein. According to the determination result in operation 1250, when the training is not completed (no), the image processing apparatus 701 may proceed to operation 655.

[0232] In operation 1255, the image processing apparatus 701 may increment the value of i by 1. Thereafter, operation 1220 may be executed again.

[0233] Although the image processing apparatus 701 is exemplified in the present disclosure as performing an image processing method based on an active material image, this is merely an exemplary example. Depending on the embodiment, the image processing apparatus 701 may separate and / or analyze not only particles in the active material image but also particles in the precursor image by performing the same image processing. In an embodiment, the image acquisition device 703 may acquire an image of the active material and / or the precursor.

[0234] In addition, although the active material image is exemplified as an SEM image in the present disclosure, this is merely an exemplary example. Depending on the embodiment, the active material image may be replaced with an image based on a transmission electron microscope (TEM), an optical microscope (OM), a structured illumination microscope (SIM), or a focused ion beam (FIB).

[0235] The above-described image processing apparatus 701 and image processing method can automatically quantify the shape characteristics of a large number of positive electrode active material particles by the following: acquiring active material images 1010 of a plurality of active materials 1011, 1013, 1015, inputting the active material images 1010 into an artificial intelligence model 745 to generate an edge image 1020, removing the edges from the active material images 1010 based on the edge image 1020 to create an edge-removed image 1030, identifying a plurality of objects 1031, 1033, 1035 included in the edge-removed image 1030, and segmenting the plurality of active materials 1011, 1013, 1015 included in the active material images 1010 based on the plurality of objects 1031, 1033, 1035.

[0236] In addition, the above-described image processing apparatus 701 and image processing method can alleviate the problem of user-dependent measurement variations by using the artificial intelligence model 745 that minimizes the need for user input.

[0237] Meanwhile, although the specification has provided illustrative examples of positive electrode active material images, the technical principles disclosed in this document can be applied in substantially the same manner even when the active material image is a negative electrode active material image.

[0238] Figure 13 is a block diagram showing an image processing apparatus 1301 according to another embodiment of the present disclosure.

[0239] Referring Figure 13 , the image processing apparatus 1301 may be connected to the image acquisition device 1303 through a wired and / or wireless connection.

[0240] According to an embodiment, the connection 1305 between the image processing device 1301 and the image acquisition device 1303 can be a communication link established through a wired and / or wireless network. In an embodiment, the wired network can be based on a local area network (LAN) or power line communication. In an embodiment, the wireless network can be based on a short-range communication network (e.g., Bluetooth, Wi-Fi, Infrared Data Association (IrDA)) or a long-range communication network (e.g., a cellular network including fourth-generation (4G) and 5G networks).

[0241] According to another embodiment, the connection 1305 between the image processing device 1301 and the image acquisition device 1303 can be a connection established through an inter-device communication interface (e.g., a bus, General-Purpose Input / Output (GPIO), Serial Peripheral Interface (SPI), or Mobile Industry Processor Interface (MIPI)).

[0242] In an embodiment, the image acquisition device 1303 can acquire an image of an active material and / or a precursor. Hereinafter, an image of an active material and / or a precursor may be referred to as an active material image. However, referring to an image as an active material image does not exclude an image of a precursor.

[0243] In an embodiment, the image acquisition device 1303 can be a microscope (e.g., a scanning electron microscope). In an embodiment, the image acquisition device 1303 can be a device that scans a focused electron beam onto the surface of a sample and converts secondary electrons generated by the interaction between the electron beam and the sample into an image signal to acquire an image of the sample surface.

[0244] In an embodiment, the image acquisition device 1303 can acquire an active image of an active material. For example, the image acquisition device 1303 can scan an electron beam onto the positive or negative electrode active material powder to acquire a SEM image. That is, the SEM image can include images of the positive and negative electrode active materials. Depending on the embodiment, the SEM image can be replaced by an image based on a transmission electron microscope (TEM), an optical microscope (OM), structured illumination microscopy (SIM), or a focused ion beam (FIB).

[0245] In an embodiment, the image acquisition device 1303 can transmit the active material image of the active material to the image processing device 1301. For example, the image acquisition device 1303 can transmit the active material image of the active material to the image processing device 1301 through the connection 1305.

[0246] In an embodiment, the image processing device 1301 can be a mobile device (e.g., a phone, a laptop, a smartphone, and a smart tablet) or a computer (e.g., a general-purpose computer, a dedicated computer).

[0247] In an embodiment, the image processing apparatus 1301 may include a communication circuit 1310, a memory 1320, and a processor 1330. According to an embodiment, Figure 13 the image processing apparatus 1301 shown in Figure 13 may further include at least one additional component (e.g., a display, an input device, or an output device) in addition to the components shown.

[0248] According to an embodiment, the communication circuit 1310 may establish a wired and / or wireless communication channel between the image processing apparatus 1301 and the image acquisition device 1303, and exchange data with the image acquisition device 1303 through the established communication channel.

[0249] In an embodiment, the memory 1320 may include a volatile memory and / or a non-volatile memory.

[0250] In an embodiment, the memory 1320 may store data used by at least one component of the image processing apparatus 1301 (e.g., the processor 1330). For example, the data may include a program 1325 (or related instructions), input data, or output data. In an embodiment, the instructions may be executed by the processor 1330 to cause the image processing apparatus 1301 to perform operations defined by the instructions.

[0251] In an embodiment, the memory 1320 may include a program 1325 (e.g., an artificial intelligence model training unit 1341, an artificial intelligence model 1345, an image acquisition unit 1350, an image generation unit 1360, a distance transformation unit 1361, a binary image filtering unit 1365, an object recognition unit 1370, an image segmentation unit 1380, and / or an information extraction unit 1390).

[0252] In an 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.

[0253] In an embodiment, the processor 1330 may execute a program 1325 (e.g., an artificial intelligence model training unit 1341, an artificial intelligence model 1345, an image acquisition unit 1350, an image generation unit 1360, a distance transformation unit 1361, a binary image filtering unit 1365, an object recognition unit 1370, an image segmentation unit 1380, and / or an information extraction unit 1390) to control at least one other component (e.g., a hardware or software component) connected to the processor 1330 in the image processing apparatus 1301 and perform various data processing or operations.

[0254] In an embodiment, the artificial intelligence model training unit 1341 may train the artificial intelligence model 1345 based on training data. In an embodiment, the artificial intelligence model 1345 may be a model trained to convert an active material image into a binary image. In an embodiment, the image acquisition unit 1350 may acquire an active material image of the active material from the image acquisition device 1303. In an embodiment, the image generation unit 1360 may generate a binary image by inputting the active material image into the artificial intelligence model 1345. In an embodiment, the distance transformation unit 1361 may convert the binary image into a distance transformation image based on a distance transformation algorithm. In an embodiment, the binary image filtering unit 1365 may filter the binary image using a threshold set based on the distance transformation image. In an embodiment, the object recognition unit 1370 may recognize a plurality of objects included in the filtered image. In an embodiment, the image segmentation unit 1380 may obtain a segmented image by segmenting a plurality of active materials included in the active material image based on the plurality of objects. In an embodiment, the information extraction unit 1390 may extract information of the active material based on the active material object included in the segmented image.

[0255] A method for processing an image acquired from the image acquisition device 1303 by the image processing device 1301 via the artificial intelligence model training unit 1341, the artificial intelligence model 1345, the image acquisition unit 1350, the image generation unit 1360, the distance transformation unit 1361, the binary image filtering unit 1365, the object recognition unit 1370, the image segmentation unit 1380, and / or the information extraction unit 1390 will be described below with reference to Figure 14 , Figure 15 , Figure 16a and Figure 16b .

[0256] Training data

[0257] Figure 14 FIG. shows training data according to another embodiment of the present disclosure;

[0258] Referring to Figure 14 , the training data may include reference active material images 1411, 1421, and 1431 and reference binary images 1451, 1461, and 1471.

[0259] The sizes of the reference active material images 1411, 1421, 1431 and the reference binary images 1451, 1461, 1471 included in the training data may be the same. In an embodiment, the size of the image may be defined as the number of horizontal pixels × the number of vertical pixels. In the training data, the number of horizontal pixels 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.

[0260] In an embodiment, the multiple reference active material images 1411, 1421 and 1431 may include one or more first reference active material images and one or more second reference active material images.

[0261] In an embodiment, the first reference active material image may be obtained by capturing an image of the active material powder. In an embodiment, the first reference active material image may be an SEM image directly obtained by the image acquisition device 1303.

[0262] In an embodiment, the second reference active material image may be an image derived or modified from the first reference active material image. In an embodiment, the second reference active material image may be generated by applying a predetermined first image processing algorithm to the first reference active material image. Here, the first image processing algorithm may include rotation, tilting, shearing, brightness adjustment, contrast adjustment, magnification, reduction, or a combination thereof.

[0263] In an embodiment, the multiple reference binary images 1451, 1461 and 1471 may include one or more first reference binary images and one or more second reference binary images.

[0264] In an embodiment, the first reference binary image 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, an edge detection algorithm, an edge removal algorithm, a binarization algorithm, or any combination thereof.

[0265] In an embodiment, the second reference binary image may be an image derived or modified from the first reference active material image. In an embodiment, the second reference binary image may be generated by applying a predetermined first image processing algorithm to the first reference binary image. Here, the first image processing algorithm may include rotation, tilting, shearing, brightness adjustment, contrast adjustment, magnification, reduction, or any combination thereof.

[0266] In an embodiment, the first reference binary image may be a binary image of the first reference active material image, and the second reference binary image may be a binary image of the second reference active material image. For example, the reference binary image 1451 may be a binary image of the reference active material image 1411, the reference binary image 1461 may be a binary image of the reference active material image 1421, and the reference binary image 1471 may be a binary image of the reference active material image 1431. Accordingly, a plurality of reference active material images 1411, 1421, and 1431 and a plurality of reference binary images 1451, 1461, and 1471 may be classified into an image set of corresponding images. For example, the reference active material image 1411 and the reference binary image 1451 may be classified into one image set.

[0267] In an embodiment, the training data may be used to train the artificial intelligence model 1345. In an embodiment, the training data may be used to train the artificial intelligence model 1345 for a predetermined number of rounds. Here, the predetermined number may be determined between 100 and 10,000. For example, the predetermined number may be 1500.

[0268] In an embodiment, the training data may be partitioned into mini - batches having a predetermined batch size. Here, the batch size may be determined between 1 and 512. For example, the batch size may be 16. When the batch size is 4, each mini - batch may be composed of 16 image sets (i.e., 16 reference active material images and 16 reference binary images).

[0269] Artificial intelligence model

[0270] Figure 15 FIG. illustrates the artificial intelligence model 1345 according to another embodiment of the present disclosure. Specifically, Figure 15 An example of training the artificial intelligence model 1345 with a mini - batch composed of the reference active material images 1501, 1503, and 1505 and the reference binary images 1591, 1593, and 1595 is shown.

[0271] In an embodiment, the artificial intelligence model 1345 may be a model based on a convolutional neural network (CNN) or a U - Net. In an embodiment, the artificial intelligence model 1345 may be a model trained to convert an active material image into a binary image.

[0272] Reference Figure 15, the artificial intelligence model 1345 may include multiple layers 1510, 1520, 1530, 1540, 1550, 1560, and 1570. The multiple layers 1510, 1520, 1530, 1540, 1550, 1560, and 1570 may be sequentially connected. The input of the sequentially connected layers is the output of the layer before it, and the output of the sequentially connected layers may be the input of the layer after it. For example, the output of layer 1510 may be the input of layer 1520.

[0273] At least two of the multiple layers 1510, 1520, 1530, 1540, 1550, 1560, and 1570 (e.g., layer 1510 and 1570, layer 1520 and 1560, or layer 1530 and 1550) may be connected by skip connections 1515, 1525, and 1535. In an embodiment, the skip connections 1515, 1525, and 1535 may connect the layers included in the encoding region among the multiple layers 1510, 1520, 1530, 1540, 1550, 1560, and 1570 (e.g., layer 1510, 1520, and 1530) with the layers included in the decoding region (e.g., layer 1550, 1560, and 1570). Here, the skip connections 1515, 1525, and 1535 may be intermediate layer connections for inputting the outputs of layers 1510, 1520, and 1530 into layers 1550, 1560, and 1570. For example, through the skip connection 1535, layer 1550 may receive the output of layer 1540 and the output of layer 1530 as inputs.

[0274] In an embodiment, the multiple layers 1510, 1520, 1530, 1540, 1550, 1560, and 1570 may include an input layer, a batch normalization layer, a 2D convolutional layer, an activation layer, a max pooling layer, an upsampling layer, a concatenation layer, or any combination thereof. In an embodiment, the layers included in the encoding region (e.g., layer 1510, 1520, and 1530) may have a structure in which a 2D convolutional layer, a batch normalization layer, an activation layer, and a max pooling layer are sequentially connected. Similarly, the layers included in the decoding region (e.g., layer 1550, 1560, and 1570) may have a structure in which a 2D convolutional layer, a batch normalization layer, an activation layer, and an upsampling layer are sequentially connected.

[0275] In an embodiment, the input layer may be a layer that receives bit values included in the reference active material images 1501, 1503, and 1505. In an embodiment, the input layer may obtain an input composed of bit values corresponding 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.

[0276] In an embodiment, the batch normalization layer may be a layer that normalizes the output values of the previous layer on a batch basis. Depending on the embodiment, the batch normalization layer may be placed immediately before the activation layer. Here, a batch may include the reference active material images input to the artificial intelligence model 1345 during one iteration to train the model. For example, when the batch size is 16, the number of reference active material images input to the artificial intelligence model 1345 during one iteration may be 16. Here, the batch size may be an integer between 1 and 512 (e.g., 16).

[0277] In an embodiment, the 2D convolutional layer may be a layer that performs a convolution operation on the input and a filter of a predetermined size to obtain an output. In an embodiment, the 2D convolutional layer included in the artificial intelligence model 1345 assumes that the output image size (number of horizontal pixels × number of vertical pixels) is the same as the input image size (number of horizontal pixels × number of vertical pixels). In an embodiment, the number of channels in the output may vary depending on the number of filters (or depth) in the 2D convolutional layer. For example, when the number of filters is 2, the number of channels in the output may increase twofold compared to the number of channels in the input. In an embodiment, the number of channels in the output may vary according to the stride in the depth direction of the 2D convolutional layer. For example, when the stride in the depth direction is 2, the number of channels in the output may be reduced by half compared to the number of channels in the input.

[0278] In an embodiment, the activation layer may be a layer that applies a predetermined activation function to the input to obtain an output. For example, the predetermined activation function may include step, sigmoid, rectified linear unit (ReLU), exponential linear unit (ELU), softmax, or any combination thereof.

[0279] In an embodiment, the max pooling layer can be a layer that selects the maximum value in each pooling region of the input to obtain an output. In an embodiment, the size of the pooling region in the max pooling layer may affect the size of the output image (i.e., the number of horizontal and vertical pixels). For example, when the pooling region size is 2×2, the output image size (e.g., 128×128) may be reduced by half compared to the input image size (e.g., 256×256). Depending on the embodiment, the artificial intelligence model 1345 may include pooling layers other than the max pooling layer. For example, another type of pooling layer may include an average pooling layer.

[0280] In an embodiment, the upsampling layer can be a layer for increasing the image resolution. In an embodiment, the upsampling layer can be a layer that uses a predetermined interpolation algorithm to increase the size of the output image compared to the input.

[0281] In an embodiment, the concatenation layer can concatenate two or more inputs to produce an output. Here, the concatenation can be depth-wise concatenation. For example, when the first input is 128×128×128 and the second input is 128×128×128, the output of the concatenation layer can be 128×128×256. When the first input is 256×256×64 and the second input is 128×228×128, the output of the concatenation layer can be 256×256×128.

[0282] Although Figure 15 seven layers 1510, 1520, 1530, 1540, 1550, 1560, and 1570 are depicted, this is only an illustrative example, and the number of layers is not limited to 7. For example, the number of layers included in the artificial intelligence model 145 can be 32.

[0283] When the number of layers included in the artificial intelligence model 1345 is 32, the artificial intelligence model 1345 can have a structure in which an input layer, a first batch normalization layer, a first 2D convolutional layer, and a second 2D convolutional layer are sequentially connected. The input of the sequentially connected layers is the output of the layer before it, and the output of the sequentially connected layers can be the input of the layer after it. Here, the input of the input layer can be 256×256×1, and the output can be 256×256×1. The output of the first batch normalization layer can be 256×256×1. The output of the first 2D convolutional layer can be 256×256×64. The output of the second 2D convolutional layer can be 256×256×64.

[0284] After the second 2D convolutional layer, the artificial intelligence model 1345 can 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 can be 256×256×64. The output of the first activation layer can be 256×256×64. The output of the first max pooling layer can be 128×128×64.

[0285] After the first max pooling layer, the artificial intelligence model 1345 can have a structure in which a third 2D convolutional layer, a fourth 2D convolutional 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 convolutional layer can be 128×128×128. The output of the fourth 2D convolutional layer can be 128×128×128. The output of the third batch normalization layer can be 128×128×228. The output of the second activation layer can be 128×128×128. The output of the second max pooling layer can be 64×64×128.

[0286] After the second max pooling layer, the artificial intelligence model 1345 can have a structure in which a fifth 2D convolutional layer, a sixth 2D convolutional layer, a fourth batch normalization layer, a third activation layer, and a first upsampling layer are sequentially connected. Here, the output of the fifth 2D convolutional layer can be 64×64×256. The output of the sixth 2D convolutional layer can be 64×64×256. The output of the fourth batch normalization layer can be 64×64×256. The output of the third activation layer can be 64×64×256. The output of the first upsampling layer can be 128×128×256.

[0287] After the first upsampling layer, the artificial intelligence model 1345 can have a structure in which a seventh 2D convolutional layer, a first concatenation layer, an eighth 2D convolutional layer, a ninth 2D convolutional layer, a fifth batch normalization layer, a fourth activation layer, and a second upsampling layer are sequentially connected. Among the two inputs of the first concatenation layer, the first input can be the output of the seventh 2D convolutional layer, and the second input can be the output of the fourth 2D convolutional layer. Here, the output of the seventh 2D convolutional layer can be 128×128×128. The output of the first concatenation layer can be 128×128×256. The output of the eighth 2D convolutional layer can be 128×128×128. The output of the ninth 2D convolutional layer can 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.

[0288] After the second upsampling layer, the artificial intelligence model 1345 may have a structure in which the tenth 2D convolutional layer, the second concatenation layer, the eleventh 2D convolutional layer, the twelfth 2D convolutional layer, the sixth batch normalization layer, the fifth activation layer, the thirteenth 2D convolutional layer, and the fourteenth 2D convolutional layer are sequentially connected. Among the two inputs of the second concatenation layer, the first input may be the output of the tenth 2D convolutional layer, and the second input may be the output of the second 2D convolutional layer. Here, the output of the tenth 2D convolutional layer may be 128×128×64. The output of the second concatenation layer may be 256×256×128. The output of the eleventh 2D convolutional layer may be 256×256×64. The output of the twelfth 2D convolutional layer may be 256×256×64. The output of the sixth batch normalization layer may be 256×256×64. The output of the fifth activation layer may be 256×256×64. The output of the thirteenth 2D convolutional layer may be 256×256×2. The output of the fourteenth 2D convolutional layer may be 256×256×2.

[0289] Artificial intelligence model learning

[0290] Reference Figure 15 , the operations of the artificial intelligence model training unit 1341 for training the artificial intelligence model 1345 by using the images 1501, 1503, 1505, 1591, 1593, and 1595 included in the mini-batch are described.

[0291] In an embodiment, the artificial intelligence model training unit 1341 may sequentially input the reference active material images 1501, 1503, and 1505 into the artificial intelligence model 1345.

[0292] In an embodiment, the artificial intelligence model training unit 1341 may train the artificial intelligence model 1345 to ensure that the difference between the sequentially obtained output images of the artificial intelligence model 1345 and the reference binary images 1591, 1,593, 1595 is less than or equal to a predetermined threshold.

[0293] For example, the artificial intelligence model training unit 1341 may calculate the difference between the output image and the reference binary images 1591, 1593, and 1595 based on a loss function (or cost function). Here, the loss function may include an average absolute error function, a root mean square error, a mean square error function, or a binary cross-entropy loss function. Depending on the embodiment, the binary cross-entropy loss function may be a function having a weight between 1 and 1000 applied to one of two colors (or categories) (for example, white in black and white).

[0294] Subsequently, the artificial intelligence model training unit 1341 may adjust the weights of the artificial intelligence model 1345 to ensure that the difference is equal to or less than a predetermined threshold (or reaches a minimum value). In an embodiment, the artificial intelligence model training unit 1341 may use an algorithm based on gradient descent (e.g., Adam, SGD, and Momentum) to adjust the weights of the artificial intelligence model 1345 to ensure that the difference is equal to or less than a predetermined threshold (or reaches a minimum value). Here, the adjusted weights may be determined based on a learning rate. The learning rate may be determined within a range of 0.00000001 to 0.1. For example, the learning rate may be 0.00001.

[0295] Subsequently, the artificial intelligence model training unit 1341 may use the images included in the next mini-batch to train the artificial intelligence model 1345. After training the artificial intelligence model 1345 using all the images included in the mini-batch, the artificial intelligence model training unit 1341 may retrain the artificial intelligence model 1345 using the images included in the mini-batch based on a predetermined number of epochs.

[0296] Image processing by artificial intelligence model

[0297] Figure 16a Illustrate images 1610, 1620, 1630, 1640, 1650, and 1660 generated by the image processing of the image processing apparatus 1301 according to an embodiment of the present disclosure.

[0298] In an embodiment, the image acquisition unit 1350 may acquire an active material image 1610 of the active material. In an embodiment, the image acquisition unit 1350 may acquire the active material image 1610 through the image acquisition device 1303.

[0299] In an embodiment, the image generation unit 1360 may generate a binary image 1620. In an embodiment, the image generation unit 1360 may generate the binary image 1620 by inputting the active material image 1610 into the artificial intelligence model 1345.

[0300] In an embodiment, the distance transformation unit 1361 may transform the binary image 1620 into a distance transformation image 1640 based on a distance transformation algorithm.

[0301] In an embodiment, the distance transformation unit 1361 may generate a distance transformation image 1640 by calculating the distance from each pixel in the binary image 1620 to the nearest black pixel (or edge pixel) and normalizing the calculated distance for each pixel. Here, the image 1630 may represent the distance from each pixel in the binary image 1620 to the nearest black pixel (or edge pixel). Thus, the value of the pixels included in the image 1630 may be an unnormalized distance value, and the value of the pixels included in the distance transformation image 1640 may be a normalized distance value. Depending on the embodiment, the distance between pixels may be calculated based on the L1 distance and the L2 distance.

[0302] In an embodiment, the distance transformation unit 1361 may generate the distance transformation image 1640 by identifying the maximum distance among the calculated distances of each pixel included in the binary image 1620, and perform max - min normalization on the calculated distance for each pixel based on the maximum distance. Depending on the embodiment, the distance transformation unit 1361 may generate the distance transformation image 1640 based on Z - score normalization, L1 normalization, or L2 normalization instead of max - min normalization.

[0303] In an embodiment, the binary image filtering unit 1365 may filter the binary image 1620 using a threshold set based on the distance transformation image 1640. Here, the threshold may be a value obtained by multiplying a predetermined ratio by the normalized maximum distance of the distance transformation image 1640. For example, the predetermined ratio may be a value between 0 and 0.1.

[0304] In an embodiment, the binary image filtering unit 1365 may filter the binary image 1620 based on the distance transformation image 1640 to generate a filtered binary image 1650.

[0305] In an embodiment, the binary image filtering unit 1365 may filter the binary image 1620 by selecting pixels in the distance transformation image 1640 whose distance values are equal to or less than a predetermined threshold, and setting the color value of the pixels among those included in the binary image 1620 that correspond to the selected pixels in the distance transformation image 1640 to a predetermined color value (i.e., the color value corresponding to black).

[0306] In an embodiment, the object recognition unit 1370 may recognize a plurality of objects included in the filtered binary image 1650. Here, the plurality of objects included in the filtered binary image 1650 may correspond to the active materials in the active material image 1610. In an embodiment, the plurality of objects may be composed of regions having a predetermined value (e.g., a value representing white). In an embodiment, the plurality of objects may be distinguished by regions having different predetermined values (e.g., a value representing black).

[0307] In an embodiment, the image segmentation unit 1380 may obtain a segmentation image 1660 by segmenting a plurality of active material objects included in the active material image 1610 based on a plurality of objects. In an embodiment, the image segmentation unit 1380 may segment the plurality of active materials in the active material image 1610 based on a watershed algorithm. Here, the plurality of active materials included in the active material image 1610 may respectively correspond to the plurality of active material objects included in the segmentation image 1660.

[0308] In an embodiment, the information extraction unit 1390 may extract information about the active material based on the active material objects included in the segmentation image. In an embodiment, the information about the active material may include information about an average diameter, an average perimeter, an average sphericity, an average aspect ratio, an average convexity, an average solidity, or distribution values of these characteristics. In addition, the information about the active material may include information about individual properties of each active material, such as a diameter, a perimeter, a sphericity, an aspect ratio, a convexity, and / or a solidity. In addition, the information about the active materials 1611, 1613, and 1615 may include information about percentile values (e.g., 1 to 99%, or D5, D50, D95, etc.) of each active material.

[0309] Figure 16b Illustrates differences in images generated by image processing of the image processing apparatus 1301 according to another embodiment of the present disclosure.

[0310] Reference Figure 16b , the three filtered binary images (1651, 1653, and 1655) are images generated by applying different thresholds to the binary image. For example, the filtered binary image 1651 may be filtered using a threshold value set to a value obtained by multiplying the normalized maximum distance by 0, the filtered binary image 1653 may be filtered using a threshold value set to a value obtained by multiplying the normalized maximum distance by 0.05, and the filtered binary image 1655 may be filtered using a threshold value set to a value obtained by multiplying the normalized maximum distance by 0.1.

[0311] By referring to the regions 1652, 1654, and 1656 of the filtered binary images 1651, 1653, and 1655 and the regions 1662, 1664, and 1666 of the segmentation images 1661, 1662, and 1664, it can be seen that although the threshold is increased, the loss of small particles is prevented, and the separation between adjacent particles becomes clearer.

[0312] Figure 17It is a flowchart showing a method for processing an active material image of an image processing apparatus 1301 according to another embodiment of the present disclosure.

[0313] Referring Figure 17 , in operation 1710, the image processing apparatus 1301 may acquire an active material image 1610 of the active material. In an embodiment, the image acquisition device 1301 may acquire the active material image 1610 through the image acquisition device 1303.

[0314] In operation 1720, the image processing apparatus 1301 may generate a binary image 1620 by inputting the active material image 1610 into the artificial intelligence model 1345. Here, the artificial intelligence model 1345 may be a model trained by the Figure 6 illustrated artificial intelligence model training method.

[0315] In operation 1730, the image processing device 1301 may transform the binary image 1620 into a distance transform image 1640 based on a distance transform algorithm.

[0316] In an embodiment, the image processing apparatus 1301 may generate the distance transform image 1640 by calculating the distance from each pixel in the binary image 1620 to the nearest black pixel (or edge pixel) and normalizing the calculated distance for each pixel. Here, the image 1630 may represent the distance from each pixel in the binary image 1620 to the nearest black pixel (or edge pixel). Therefore, the value of the pixel included in the image 1630 may be an unnormalized distance value, and the value of the pixel included in the distance transform image 1640 may be a normalized distance value. Depending on the embodiment, the distance between pixels may be calculated based on the L1 distance and the L2 distance.

[0317] In an embodiment, the image processing apparatus 1301 may generate the distance transform image 1640 by identifying the maximum distance among the distances calculated for each pixel included in the binary image 1620, and perform max - min normalization on the distances calculated for each pixel based on the maximum distance. Depending on the embodiment, the distance transform unit 1361 may generate the distance transform image 1640 based on Z - score normalization, L1 normalization, or L2 normalization instead of max - min normalization.

[0318] In operation 1740, the image processing apparatus 1301 may filter the binary image 1620 using a threshold set based on the distance transform image 1640. Here, the threshold may be a value obtained by multiplying a predetermined ratio by the normalized maximum distance of the distance transform image 1640. For example, the predetermined ratio may be a value between 0 and 0.1.

[0319] In an embodiment, the image processing device 1301 may generate a filtered binary image 1650 by filtering the binary image based on the distance transform image 1640.

[0320] In an embodiment, the image processing device 1301 may filter the binary image 1620 by selecting pixels in the distance transform image 1640 whose distance values are equal to or less than a predetermined threshold, and setting the color values of the pixels in the binary image 1620 that correspond to the selected pixels in the distance transform image 1640 to a predetermined color value (i.e., a color value corresponding to black).

[0321] In operation 1750, the image processing device 1301 may identify a plurality of objects included in the filtered binary image 1650.

[0322] In operation 1760, the image processing device 1301 may obtain a segmented image 1660 by segmenting a plurality of active materials included in the active material image 1610 based on the plurality of objects. In an embodiment, the image processing device 1301 may segment a plurality of active materials included in the active material image 1610 based on a watershed algorithm. Here, the plurality of active materials included in the active material image 1610 may respectively correspond to a plurality of active material objects included in the segmented image 1660.

[0323] Although the image processing device 1301 is exemplified in the present disclosure as performing an image processing method based on an active material image, this is merely an exemplary example. Depending on the embodiment, the image processing device 1301 may separate and / or analyze not only the particles in the active material image but also the particles in the precursor image by performing the same image processing. In an embodiment, the image acquisition device 303 may acquire an image of the active material and / or the precursor.

[0324] In addition, although the active material image is exemplified as an SEM image in the present disclosure, this is merely an exemplary example. Depending on the embodiment, the active material image may be replaced with an image based on a transmission electron microscope (TEM), an optical microscope (OM), a structured illumination microscope (SIM), or a focused ion beam (FIB).

[0325] The above-described image processing apparatus 1301 and image processing method can automatically quantify and analyze the shape characteristics of a large number of positive electrode active material particles by: obtaining active material images 1610 of a plurality of active materials, generating a binary image 1620 by inputting the active material images 1610 into an artificial intelligence model 1345, transforming the binary image 1620 into a distance transform image 1640 based on a distance transform algorithm, filtering the binary image 1620 using a threshold set based on the distance transform image 1640, identifying a plurality of objects included in the filtered binary image 1650, and segmenting a plurality of active material objects included in the active material image based on the plurality of objects.

[0326] In addition, the above-described image processing apparatus 1301 and image processing method can alleviate the problem of user-dependent measurement variations by using an artificial intelligence model 1345 that minimizes the need for user input.

[0327] Meanwhile, although the specification has provided illustrative examples of positive electrode active material images, the technical principles disclosed in this document can be applied in a substantially similar manner even if the active material image is a negative electrode active material image.

Claims

1. An image processing apparatus, comprising: an image acquisition unit configured to acquire active material images of a plurality of active materials; an artificial intelligence model training unit configured to train an artificial intelligence model using training data including a plurality of reference active material images and a plurality of reference binary images or a plurality of reference edge images corresponding to the plurality of reference active material images; an image generation unit configured to generate a binary image or an edge image by inputting the active material image into the artificial intelligence model; an object recognition unit configured to recognize a plurality of objects based on the binary image or the edge image; and an image segmentation unit configured to obtain a segmented image by segmenting the plurality of active materials included in the active material image based on the plurality of objects.

2. The image processing apparatus according to claim 1, wherein, The artificial intelligence model training unit trains the artificial intelligence model to ensure that the difference between a plurality of output images obtained by inputting the plurality of reference active material images into the artificial intelligence model and the plurality of reference binary images is equal to or less than a predetermined reference value.

3. The image processing apparatus according to claim 2, wherein The plurality of reference active material images include one or more first reference active material images and one or more second reference active material images, the plurality of reference binary images include one or more first reference binary images and one or more second reference binary images, the first reference active material image is obtained by capturing an image of active material powder, the second reference active material image is generated by applying a predetermined image processing algorithm to the first reference active material image, the first reference binary image corresponds to the first reference active material image, and the second reference binary image corresponds to the second reference active material image.

4. The image processing apparatus according to claim 1, wherein, The artificial intelligence model includes a plurality of layers, and at least one layer included in an encoding region among the plurality of layers is connected to at least one layer included in a decoding region by a skip connection.

5. The image processing apparatus according to claim 4, wherein, Among the plurality of layers, the layer included in the decoding region has a structure in which a two-dimensional (2D) convolutional layer, a batch normalization layer, an activation layer, and an upsampling layer are sequentially connected.

6. The image processing apparatus according to claim 4, wherein, Among the plurality of layers, the layer included in the encoding region has a structure in which a two-dimensional (2D) convolutional layer, a batch normalization layer, an activation layer, and a max pooling layer are sequentially connected.

7. The image processing apparatus according to claim 1, wherein: The active material image is an image based on a scanning electron microscope (SEM), a transmission electron microscope (TEM), an optical microscope (OM), a structured illumination microscopy (SIM), or a focused ion beam (FIB).

8. The image processing apparatus according to claim 1, wherein, The image segmentation unit segments the plurality of active materials included in the active material image based on a watershed algorithm.

9. The image processing apparatus according to claim 1, further comprising an information extraction unit configured to extract information about the plurality of active materials based on the segmented image, in, The information about the plurality of active materials includes particle diameter, perimeter, sphericity, aspect ratio, convexity, solidity, distribution, percentile value, or any combination thereof of the active materials.

10. The image processing apparatus according to claim 1, further comprising an edge removal unit configured to generate an edge-removed image by removing edges in the active material image based on the edge image, wherein, The object recognition unit recognizes the plurality of objects included in the edge-removed image.

11. The image processing apparatus according to claim 1, wherein: The artificial intelligence model training unit trains the artificial intelligence model to ensure that the difference between multiple output images obtained by inputting the multiple reference active material images into the artificial intelligence model and the multiple reference edge images is equal to or less than a predetermined reference value.

12. The image processing apparatus according to claim 11, wherein, The artificial intelligence model training unit calculates the differences between the multiple output images and the multiple reference edge images based on a mean absolute error function, a root mean square error, a mean square error function, or a binary cross entropy loss function.

13. The image processing apparatus according to claim 12, wherein: The binary cross entropy loss function is assigned weights for predetermined colors.

14. The image processing apparatus according to claim 1, further comprising: a distance transformation unit configured to transform the binary image into a distance transformation image based on a distance transformation algorithm; as well as a binary image filtering unit configured to filter the binary image using a threshold set based on the distance transform image, The object recognition unit recognizes a plurality of objects included in the filtered binary image.

15. The image processing apparatus according to claim 14, wherein: The distance transform unit generates the distance transformed image by calculating a distance from each pixel included in the binary image to a nearest black pixel and normalizing the calculated distance for each pixel.

16. The image processing apparatus according to claim 15, wherein, The distance transform unit generates the distance transform image by identifying a maximum distance among distances calculated for each pixel included in the binary image and performing maximum-minimum normalization on the distance calculated for each pixel based on the maximum distance.

17. The image processing apparatus according to claim 14, wherein, The binary image filtering unit filters the binary image by selecting pixels whose distance values are equal to or smaller than a predetermined threshold value among the pixels included in the distance transformation image, and setting color values of pixels corresponding to the selected pixels among the pixels included in the binary image to a predetermined color value.

18. The image processing apparatus according to claim 17, wherein, The predetermined threshold value is a value obtained by multiplying a predetermined ratio by the normalized maximum distance of the distance transformation image.

Images