Classification system that facilitates space group inference and method for recommending zone axes therein

Abstract

An object of the present invention is to provide a classification system that facilitates space group inference and a method for constructing the classification system therein, and to provide a method for inferring space group candidates of a target material and recommending a zone axis for the next imaging. [Solution to the Problem] A classification system that can easily infer space groups and a method for constructing the classification system therein are disclosed. The SADP classification system includes a plurality of labels, which are constructed by clustering SADP (Selected Area Diffraction Pattern) images taken with a Transmission Electron Microscope (TEM) based on specific criteria, and the labels are linked to space groups of the classification system from the crystallography perspective.

Description

[Technical Field]

[0001] The present invention relates to a classification system that facilitates space group inference and a method for constructing a classification system therein. The present invention also relates to a method for inferring potential space groups of target materials and recommending zone axes for subsequent imaging. [Background technology]

[0002] A Selected Area Diffraction Pattern (SADP) is created by irradiating an electron beam onto the material to be analyzed and creating an image of the electron diffraction. Because SADPs view a three-dimensional crystal structure in two-dimensional image form, a single SADP cannot clearly identify the space group of the material. To solve this problem, Ziletti et al. proposed a method to create a single diffraction fingerprint using six SADPs, and then learn / analyze the diffraction fingerprint to identify the space group of the material.

[0003] When photographing SADPs with a TEM, the material was rotated in six directions to obtain the SADPs. The rotation was based on the

[0001] zone axis. The SADPs obtained by rotating ±45 degrees around the x-axis were accumulated in the red channel of RGB, the SADPs obtained by rotating ±45 degrees around the y-axis were accumulated in the green channel, and the SADPs obtained by rotating ±45 degrees around the z-axis were accumulated in the blue channel to generate the diffraction fingerprint. To generate a diffraction fingerprint, the direction of the electron beam must first be aligned with the material's <0001> zone axis. However, it is nearly impossible to align the direction of the electron beam with the material's <0001> zone axis without knowing the material's crystalline structure. Even if we assume that the direction of the electron beam is aligned with the 0001 crystal zone axis of the material, for crystal structures that are not isotropic (cubic system) or hexagonal system, it is difficult for SADPs photographed by rotating the image by ±45 degrees along the x, y, and z axes to contain meaningful information. Summary of the Invention [Problem to be solved by the invention]

[0004] The present invention provides a classification system that facilitates space group inference and a method for constructing a classification system therein. The present invention also provides a method for inferring potential space groups for a target material and recommending zone axes for subsequent imaging. [Means for solving the problem]

[0005] To achieve the above-mentioned object, an SADP classification system according to one embodiment of the present invention includes a plurality of labels, which are constructed by clustering SADP (Selected Area Diffraction) images taken with a Transmission Electron Microscope (TEM) based on specific criteria, and the labels are linked to space groups of a crystallographic classification system.

[0006] The classification system of the present invention includes a diffraction pattern analysis unit that generates an SADP classification system having multiple labels of crystal structures by clustering SADP images taken with a TEM based on specific criteria, and a classification system linking unit that matches the SADP classification system with a classification system from a crystallographic perspective, where the classification system from a crystallographic perspective has space group and zone axis information of the crystal structure, and the labels of the SADP classification system are matched with the space group in a one-to-many or one-to-one manner.

[0007] The classification system of the present invention includes a diffraction pattern analysis unit that clusters SADP images taken with a TEM based on specific criteria to generate an SADP classification system having multiple labels, and a probability-based space group inference unit that applies an algorithm learned by machine learning to the SADP images of a material to analyze a crystallographic classification system linked to the acquired labels, and probabilistically infers a space group of the material, where the crystallographic classification system has space group and zone axis information of a crystal structure.

[0008] The classification system of the present invention includes a learning unit that trains a classification algorithm to classify SADP images taken with a TEM through machine learning, and a probability-based space group inference unit that applies the classification algorithm to SADP images of a material, analyzes a crystallographic classification system linked with labels acquired by applying the classification algorithm to the SADP images of the material, and probabilistically infers a space group of the material. Here, the labels are constructed by clustering the SADP images based on specific criteria and are linked to the space group of the crystallographic classification system.

[0009] In a computer-readable recording medium having recorded thereon program code of the present invention, the program code is used to carry out a method including the steps of clustering SADP images taken with a TEM based on specific criteria to generate an SADP classification system having multiple labels, and linking the new classification system to a classification system from a crystallographic perspective, wherein the classification system from a crystallographic perspective has space group and zone axis information of a crystal structure, and the labels of the SADP classification system are linked to the space group in a one-to-many or one-to-one manner.

[0010] The classification system of the present invention includes a classification system linking unit that matches a Selected Area Diffraction Pattern (SADP) classification system having multiple labels for 2D patterns with a crystallographic classification system, and a zone axis recommending unit that recommends a zone axis along which an electron beam is scanned for subsequent imaging of a material. Here, the crystallographic classification system has information on the space group and zone axis of a crystal structure, and the SADP classification system labels are matched one-to-many or one-to-one with the space group.

[0011] The classification system of the present invention includes a table including labels having information about space groups and zone axes of crystalline structures, and a zone axis recommendation unit for recommending a zone axis along which an electron beam is scanned for subsequent imaging of a material, wherein the recommended zone axis is one of the zone axes included in the labels.

[0012] The classification system of the present invention includes a zone axis recommender that recommends a zone axis along which an electron beam is scanned for a next image of a material from a label having information about a space group of a crystal structure and a zone axis that matches the space group, wherein the zone axis recommender includes a calculation unit that calculates a degree of disorder for a zone axis belonging to the label, and a recommendation unit that recommends a zone axis having the lowest degree of disorder as a result of the calculation as the zone axis for the next image. [Effects of the Invention]

[0013] The classification system and the classification system construction method of the present invention generate a new classification system by clustering similar SADP images according to a specific criterion, and space groups can be easily inferred using this classification system. Furthermore, the classification system can infer the space group of a material without the aid of a crystal structure expert by utilizing one or more SADP images taken with a TEM. Furthermore, the classification system recommends zone axes that will unambiguously identify the space group for the next image, allowing the analysis to proceed with minimal stress on the material. [Brief explanation of the drawings]

[0014] [Figure 1] 1 is a diagram illustrating a classification system of the present invention. [Figure 2] 1 is a diagram illustrating the classification system structure and classification system linkage of the present invention. [Figure 3] 1 is a diagram illustrating the SADP image classification criteria of the present invention. [Figure 4] 1 is a diagram illustrating the process of inferring a space group according to the present invention. [Figure 5] 1 is a diagram illustrating an example of SADP images belonging to the same decision class; [Figure 6] 1 is a diagram illustrating a classification system of the present invention. [Figure 7] This is a diagram illustrating a table that links the SADP classification system from a crystallographic perspective with the new SADP classification system that is easy for computers to distinguish. [Figure 8] 2 is a diagram illustrating the structure of a zone axis recommendation unit according to the present invention; DETAILED DESCRIPTION OF THE INVENTION

[0015] As used herein, singular expressions include plural expressions unless the context clearly dictates otherwise. In this specification, terms such as "comprise" or "include" should not be interpreted as including all of the components or steps described in the specification, but should be interpreted as meaning that some components or steps may not be included, or that additional components or steps may be included. Furthermore, terms such as "unit," "module," etc. used in the specification refer to a unit that processes at least one function or operation, and may be implemented in hardware or software, or a combination of hardware and software.

[0016] The present invention relates to a classification system and a method for constructing a classification system therein, which can reconstruct a classification system from a crystallographic viewpoint into a new classification system that is easy for a computer to classify. Here, the new classification system can include multiple labels generated by classifying SADP (Selected Area Diffraction Pattern) images based on specific criteria, i.e., SADP images with common features or similar structures can belong to the same label.

[0017] In addition, the classification system can link (match) a classification system from a crystallographic perspective with a classification system that is easy for a computer to distinguish, and can infer the space group of a material by analyzing one or more SADP images taken with a transmission electron microscope (TEM) using artificial intelligence, particularly machine learning (e.g., deep learning) technology.

[0018] Various embodiments of the present invention will now be described in detail with reference to the accompanying drawings. Figure 1 illustrates a classification system according to an embodiment of the present invention, Figure 2 illustrates a classification system structure and classification system connections according to an embodiment of the present invention, Figure 3 illustrates an SADP image classification standard according to an embodiment of the present invention, Figure 4 illustrates a process for inferring space groups according to an embodiment of the present invention, and Figure 5 illustrates an example of an SADP image belonging to the same determined classification.

[0019] As shown in FIG. 1, the classification system system of this embodiment constructs a new classification system that is easy for a computer to distinguish from a classification system based on crystallography, and includes a diffraction pattern analysis unit 100, a classification system linking unit 102, a rotation pattern classification algorithm learning unit 104, a probability-based space group inference unit 106, and a control unit (not shown) that controls the overall operations of these units. The diffraction pattern analyzer 100 uses information on the interior angles of a triangle formed by one diffraction point and its two nearest diffraction points and forbidden reflection information that can be determined through structure factor calculations to cluster SADP images in a form that is easy for a computer to distinguish, thereby generating a new classification system (SADP classification system).Since it generates a classification system, the rotation pattern analyzer 100 can be called a classification system generator.

[0020] For example, the diffraction pattern analyzer 100 can create a new classification system by clustering externally similar SADP images using the interior angle information (∠AOB, ∠OAB) of a triangle formed by one diffraction point (O) and its two closest diffraction points (A, B) shown on the left side of Fig. 3 and the forbidden reflection information shown on the right side of Fig. 3. Here, the new classification system can be divided into square (square primitive, A), rectangle (rectangle primitive, B), regular hexagon (hexagonal primitive, C), and oblique primitive (parallelogram, D) based on the basic shapes that a two-dimensional lattice can have.

[0021] According to another embodiment, the diffraction pattern analysis unit 100 can also generate a new classification system by clustering SADP images based on the internal angle information between one diffraction point or three or more diffraction points, rather than between one diffraction point and its two nearest diffraction points. According to yet another embodiment, the diffraction pattern analyzer 100 may cluster the SADP images using only one of the interior angle information and the forbidden reflection information to generate a new classification system. According to yet another embodiment, the diffraction pattern analyzer 100 may cluster the SADP images using the average angle between the central diffraction spot and its surrounding diffraction spots to generate a new classification system.

[0022] As can be seen from the classification system structure in Figure 2, the classification system from the perspective of crystallography includes space groups, although space groups with similar structures can exist. Considering the process of constructing a new classification system, the classification system first uses a TEM to obtain an SADP image for each space group. In Figure 2, a designation such as "001" can refer to the direction of the electron beam scanning the material. For example, "001" can indicate the direction of the electron beam when the electron beam is emitted parallel to the Z axis. In this case, when classified according to a specific criterion, multiple SADP images with similar or identical structures can exist.

[0023] The diffraction pattern analyzer 100 can cluster SADP images and generate a new classification system using information on the interior angles (∠AOB, ∠OAB) of the triangles formed by diffraction points and forbidden reflection information. In this case, classifying SADP images into the new classification system can be performed by a rotation pattern classification algorithm (model) as shown in FIG. For example, the diffraction pattern analysis unit 100 can define 60 labels by clustering the SADP images acquired based on 16 zone axes (0001, 0101, 0102, 0103, 0104, 0111, 0112, 0113, 0114, 0203, 0212, 0213, 0214, 0223, 0313, 0323) of five space groups 213, 221, 225, 227, and 229. In this case, the SADP images can be classified into 112 labels based on the crystallography classification system. Due to this clustering, the number of labels is smaller than the number of labels based on the crystallography classification system.

[0024] Figure 5 shows SADP images that belong to different space groups but the same label, i.e., SADP images corresponding to multiple space groups can belong to one label. Meanwhile, the criteria for clustering SADP images can be varied in many ways, as long as new classification systems are created using diffraction points.

[0025] The classification system linking unit 102 can link a classification system from a crystallographic perspective, which can be expressed by a space group and a zone axis as shown in Figure 2, with an SADP classification system that is easy for a computer to distinguish. That is, the classification system linking unit 102 can link a space group of a 3D structure with a label of a 2D structure. The label and the space group are basically connected one-to-many, but can also be connected one-to-one depending on the form of the SADP image. That is, an SADP image belonging to a specific space group and zone axis can belong to one label. In this case, accurate space group and zone axis information can be found by inputting only one SADP image.

[0026] The rotation pattern classification algorithm learning unit 104 can learn a model (rotation pattern classification algorithm) so that SADP images can be classified through machine learning. According to one embodiment, the rotation pattern classification algorithm training unit 104 can configure and train a diffraction pattern classification algorithm using off-the-shelf deep learning models such as AlexNet, Inception v3, ResNet, and DenseNet.

[0027] The probability-based space group inference unit 106 can analyze the classification system of crystallographic viewpoints linked to the labels obtained by applying a diffraction pattern classification algorithm to the SADP image of the material, and can probabilistically infer the space group of the crystalline structure of the material. For example, when two SADP images belonging to one material are given as shown in FIG. 4, the probability-based space group inference unit 106 can apply a diffraction pattern classification algorithm to each of the SADP images to analyze the crystallographic classification system associated with the labels obtained, and probabilistically infer the space group of the crystal structure.

[0028] According to one embodiment, the probability-based space group inference unit 106 may derive a final probability by ensembling probabilities inferred from multiple SADP images. In FIG. 4, the final probability is shown by averaging the inferred probabilities, but it is also possible to accumulate the inferred probabilities and then derive only the space group that shows the maximum value.

[0029] Referring to Figure 4, when two SADP images are input to the diffraction pattern classification algorithm, the image located at the top is classified as A1-0 through the diffraction pattern classification algorithm, and this label is inferred to have a 25% probability of belonging to each space group because it is one-to-many matched with space groups 221, 225, 227, and 229. On the other hand, the image located at the bottom is classified as D8-0 through the diffraction pattern classification algorithm, and this label is inferred to have a 100% probability of belonging to the 225 space group because it is one-to-one matched with space group 225. Finally, by averaging these two probabilities, it is inferred that the probability of belonging to the 225 space group is 62.5% and the probability of belonging to the 221, 227, or 229 space groups is 12.5%.

[0030] Previously, the space group of a material was inferred using this crystallographic classification system. Specifically, six SADP images were acquired using a TEM, and the acquired SADP images were analyzed to determine which crystal class the material belonged to. However, this method required complex steps, such as rotating the material in six directions while scanning it with an electron beam, and therefore only experts were able to perform the analysis. Furthermore, because it was difficult to accurately rotate the material in six directions while scanning it with an electron beam, it was not possible to accurately infer the space group.

[0031] Meanwhile, the classification system of this embodiment clusters SADP images to generate multiple labels, and analyzes the space groups associated with the labels to probabilistically infer the space group of a material. As a result, the space group of the material can be inferred using one or more SADP images taken with a TEM without the assistance of a crystal structure expert.

[0032] In summary, the classification system of this embodiment links a 3D classification system from a crystallographic perspective with a new 2D classification system formed by clustering SADP images, and can infer the space group of a material by analyzing the space group linked to the label to which the SADP image of the material obtained via TEM belongs.

[0033] Introducing such a classification system has the following advantages: 1. Even with general-purpose machine learning techniques, it is possible to infer the space group of materials with high accuracy. 2. It is possible to perform probabilistic inference of the space group of the material for a single SADP image, and when multiple SADP images are given, it is possible to perform more accurate inference of the space group of the material by ensembling the inference probabilities of each. 3. When a TEM machine is combined with this invention, it is possible to analyze the currently taken SADP image and suggest which angle to take the next SADP image at to increase the probability of space group inference. When attempting to understand the crystalline structure of a material using TEM, since SADP images are 2D images, analyzing at least three SADP images is recommended to determine the 3D structure. These SADP images are acquired while changing the scanning direction of the electron beam. If the scanning direction of the electron beam is not determined accurately, multiple electron beam scans will be performed. However, if the electron beam is scanned too many times, the material may be damaged due to its inability to withstand the electron beam energy. Therefore, multiple electron beam scans should be avoided, and the scanning direction of the electron beam must be determined accurately. The classification system of the present invention enables accurate recommendations for the scanning direction of the electron beam, i.e., the crystal zone axis.

[0034] The following will discuss in detail how to recommend the scanning direction, i.e., the zone axis. FIG. 6 is a diagram illustrating a classification system according to another embodiment of the present invention, FIG. 7 is a diagram illustrating a table linking the SADP classification system from a crystallography perspective with a new SADP classification system that is easy for a computer to distinguish, and FIG. 8 is a diagram illustrating the structure of a zone axis recommendation unit according to one embodiment of the present invention. 6, the classification system of this embodiment may include a diffraction pattern analysis unit 100, a classification system linking unit 102, a diffraction pattern classification algorithm learning unit 104, a probability-based space group inference unit 106, and a zone axis recommendation unit 600. The operations of the remaining components except for the zone axis recommendation unit 600 are the same as or similar to those of the previous embodiment, and therefore will not be described below.

[0035] First, the concept of the method for recommending a zone axis will be explained with reference to Fig. 7, and then the function of the zone axis recommendation unit 600 will be explained. The table in Fig. 7 is composed of labels having information about the space group and zone axis of a crystal structure. As shown in the table of Figure 7, when the 3D SADP classification system from a crystallographic perspective and the new 2D SADP classification system are linked, assume that an input SADP image is analyzed to obtain an A1-0 label. In this case, the A1-0 label has a one-to-many match with the zone axis

[0001] of space group 225, the zone axis

[0001] of space group 227, and the zone axis

[0001] of space group 229. Therefore, the input SADP image has a probability of belonging to all three space groups, 225, 227, and 229. As a result, it is not possible to identify the space group of the material corresponding to the input SADP image. However, information regarding the zone axis (

[0001] ) can be accurately determined.

[0036] In this situation, it is necessary to recommend a zone axis that can clearly identify the space group during the next TEM SADP imaging. Therefore, the present invention proposes a method for determining which zone axis should be aligned by rotating the material to clearly identify the space group. If the material is aligned with the zone axis <0111> during the next TEM SADP image, it will acquire the C1-0 label when the next SADP image is analyzed. The C1-0 label can belong to space group 225, space group 227, and space group 229, and therefore the space group of the material cannot be clearly identified. Therefore, the classification system may not recommend the zone axis <0111>.

[0037] If the material is aligned with the zone axis <0102> during the next TEM SADP image, it can acquire a B2-0 or D11-0 label when analyzing the next SADP image. Obtaining the B2-0 label ensures that the material belongs to space group 225, whereas obtaining the D11-0 label makes it impossible to clearly identify which space group, 227 or 229, it belongs to. While this is preferable to aligning with the zone axis <0111>, it still does not clearly identify the space group. Therefore, the classification system may not recommend the zone axis <0102>.

[0038] If the material is aligned with the zone axis <0101> during the next TEM SADP image, the B1-0 label, D7-0 label, or D7-1 label can be obtained when the next SADP image is analyzed. Obtaining the B1-0 label clearly indicates that the material belongs to space group 229, obtaining the D7-0 label clearly indicates that the material belongs to space group 225, and obtaining the D7-1 label clearly indicates that the material belongs to space group 227. Therefore, the space group can be clearly identified in any case, and as a result, the classification system can recommend the zone axis <0101>.

[0039] On the other hand, if the SADP image analysis result is labeled B5-0, accurate zone axis information cannot be obtained because of the existence of zone axes

[0104] and

[0223] . In this case, the zone axis that can more effectively distinguish the space group can be selected from the zone axis that will be aligned at the next imaging acquired through the above process assuming the zone axis is

[0104] , and the zone axis that will be aligned at the next imaging acquired through the above process assuming the zone axis is

[0223] . This series of steps, when coupled with TEM hardware, can significantly reduce the number and time of scanning the electron beam over the material, thereby enabling the material to be analyzed effectively.

[0040] In summary, the zone axis recommendation unit 600 of the classification system of the present invention sequentially selects zone axes that can be used in the next TEM SADP imaging, determines whether the space group can be clearly identified when the corresponding zone axis is selected, and can recommend a zone axis that clearly identifies the space group. According to one embodiment, the zone axis recommendation unit 600 can recommend a zone axis for which only one label matching one space group exists as the next zone axis. That is, the zone axis recommendation unit 600 can recommend a zone axis with a high discrimination power for the space group. Preferably, the zone axis recommendation unit 600 can calculate discrimination power for the zone axes and recommend the zone axis with the highest discrimination power for the next TEM SADP imaging. Of course, if there is no zone axis with only a label that matches one space group, the zone axis recommendation unit 600 can recommend a zone axis with only labels that match two space groups as a second best option. However, if there is no zone axis with only labels that match two space groups, the zone axis recommendation unit 600 may not recommend a zone axis with only labels that match three space groups, because in this case, the probability of identifying the space group becomes too low.

[0041] For such zone axis recommendation, the zone axis recommendation unit 600 can determine how high the discrimination ability of the aligned zone axes for the space group during the next imaging is through the entropy (H(X)) calculation of the following Equation 1:

[0042]

number

[0043] where p(x j ) means the probability of the jth label in a particular zone axis, k is the total number of labels, and p(x i ) denotes the probability of belonging to the ith space group in the corresponding label, and n denotes the total number of space groups in the corresponding label.

[0044] Since the entropy indicates the degree of disorder, a high entropy can mean a low discrimination ability for space groups, and a low entropy can mean a high discrimination ability for space groups. Therefore, the zone axis recommendation unit 600 can calculate the entropy for all zone axes or for the already set zone axes, and recommend the zone axis with the smallest entropy as the zone axis for the next TEM SADP imaging.

[0045] When the input SADP is analyzed and an A1-0 label is obtained, the probability of belonging to space group 225, space group 227, and space group 229 is 1 / 3, respectively, so the entropy calculation result for the A1-0 label is as shown in Equation 2 below. The maximum value of the entropy is 1, and since the entropy is 1, it can be determined that there is no discrimination ability for space group 225, space group 227, and space group 229.

[0046]

number

[0047] After obtaining the A1-0 label, if the A1-0 label is aligned to the zone axis

[0102] , a B2-0 label or a D11-0 label can be obtained. The B2-0 label is a label with a 100% probability of being in space group 225, while the D11-0 label is a label with a 50% probability of being in space group 227 and a 50% probability of being in space group 229. When aligned to the zone axis

[0102] , it is not known whether the B2-0 label or the D11-0 label will be obtained. If this probability is assumed to be 50%, the entropy calculation result for the zone axis

[0102] is as shown in Equation 3 below. Since the entropy is lower than that of the A1-0 label, aligning to the zone axis

[0102] can have relatively higher discrimination power.

[0048]

number

[0049] When aligned along the zone axis

[0101] , a B1-0 label, a D7-0 label, or a D7-1 label can be obtained. The B1-0 label is a label with a 100% probability of belonging to space group 229, the D7-0 label is a label with a 100% probability of belonging to space group 225, and the D7-1 label is a label with a 100% probability of belonging to space group 227. When aligned along the zone axis

[0101] , it is unknown whether the B1-0 label, the D7-0 label, or the D7-01 label will appear. Assuming this probability is 33.3%, the entropy calculation result for the zone axis

[0101] is as shown in Equation 4 below.

[0050]

number

[0051] Comparing the entropy when the zone axis is aligned with the

[0102] axis and the

[0101] axis, the entropy when the zone axis is aligned with the

[0101] axis is lower. Therefore, the zone axis recommendation unit 600 can recommend the

[0101] axis as the scanning beam direction of the material. As a result, the electron beam can be scanned along the

[0101] axis.

[0052] In summary, the zone axis recommendation unit 600 calculates the entropy for all or a plurality of zone axes and recommends the zone axis with the lowest entropy as the zone axis for the next TEM SADP imaging. Therefore, the space group of the corresponding material can be inferred using the SADP imaged by the TEM without the help of a crystal structure expert, and the zone axis direction that can most effectively classify the space group during the next imaging can be notified, thereby minimizing stress on the material and allowing analysis to proceed. On the other hand, if there are a plurality of zone axes having the lowest entropy, the zone axis recommending unit 600 can arbitrarily recommend one of the zone axes.

[0053] Although the degree of disorder is calculated using entropy in the above, the method of calculating the degree of disorder can be modified in various ways as long as the degree of disorder is calculated and a zone axis is recommended. When the structure of the zone axis recommendation unit 600 is examined in detail with reference to FIG. 8, the zone axis recommendation unit 600 may include a table unit 800, a calculation unit 802, and a recommendation unit 804. The table portion 800 may be configured with labels containing information about the space group of the crystal structure and the corresponding zone axes. The calculation unit 802 can calculate the degree of disorder for the zone axes belonging to the label, for example, the entropy. The recommendation unit 804 can recommend the zone axis with the lowest degree of disorder as the zone axis for the next image capture.

[0054] Meanwhile, the components of the above-described embodiments can be easily understood from a process perspective, i.e., each component can be understood as its own process. Also, the processes of the above-described embodiments can be easily understood from the perspective of the components of the device. Furthermore, the above-described technical content may be embodied in the form of program instructions that can be executed by various computer means and recorded on a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, and the like, alone or in combination. The program instructions recorded on the medium may be specially designed and configured for the embodiments, or may be known and available to those skilled in the art of computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs and DVDs, magneto-optical media such as floptical disks, and hardware devices specially configured to store and execute program instructions, such as ROMs, RAMs, and flash memories. Examples of program instructions include not only machine language code, such as that produced by a compiler, but also high-level language code that can be executed by a computer using an interpreter, etc. A hardware device may be configured to operate as one or more software modules to perform the operations of the embodiments, or vice versa.

[0055] The above-described embodiments of the present invention have been disclosed for illustrative purposes, and those skilled in the art having ordinary skill in the art may make various modifications, changes, and additions within the spirit and scope of the present invention, and such modifications, changes, and additions should be considered to fall within the scope of the following claims.

Claims

1. a diffraction pattern analyzer that clusters the SADP images taken by the TEM based on specific criteria to generate a SADP classification system with multiple labels for the 2D patterns; a classification system linker that matches the SADP classification system with a crystallography classification system; A classification system characterized in that the crystallographic classification system has space group and zone axis information of a crystal structure, and the labels of the SADP classification system are matched one-to-many or one-to-one with the space group.

2. The specific criterion is information on the interior angles of a triangle formed by one diffraction point and two diffraction points located closest to it in the SADP image and forbidden reflection information, The classification system of claim 1, wherein SADP images whose interior angle information of the triangle and forbidden reflection information are determined to be similar belong to the same label.

3. a learning unit that learns a rotation pattern classification algorithm so that the SADP image can be classified through machine learning; The classification system of claim 1, further comprising a probability-based space group inference unit that applies the rotation pattern classification algorithm to the SADP image of the material to analyze a classification system of a crystallographic perspective matched with the label acquired by the application of the rotation pattern classification algorithm to the SADP image of the material, and probabilistically infers a space group of the material.

4. The classification system of claim 3, wherein the probability-based space group inference unit derives a final probability by ensembling the probabilities inferred from the SADP image, or accumulates the inferred probabilities and then derives only the space group that shows the maximum value.

5. a classification system linker that matches a Selected Area Diffraction Pattern (SADP) classification system with multiple labels for 2D patterns with a crystallography-based classification system; a zone axis recommending unit for recommending a zone axis along which the electron beam is scanned for the next photographing of the material; The classification system according to the present invention is characterized in that the classification system according to the present invention has information on the space group and zone axes of the crystal structure, and the labels of the SADP classification system are matched one-to-many or one-to-one with the space group.

6. 6. The classification system of claim 5, wherein the zone axis recommender recommends a zone axis having only a label matching one space group as a zone axis for the next image capture.

7. 6. The classification system of claim 5, wherein the zone axis recommendation unit calculates a disorder degree for the zone axis belonging to the label, and recommends a zone axis having a relatively low disorder degree as a zone axis for the next photographing, while not recommending a zone axis having a relatively high disorder degree as a result of the calculation.

8. 8. The classification system according to claim 7, wherein the zone axis with the lowest degree of disorder in the calculation result is recommended as the zone axis for the next photograph.

9. 9. The classification system of claim 8, wherein the zone axis recommendation unit calculates an entropy of the following equation (1) as the disorder of the calculation result, and recommends a zone axis with the smallest entropy as a zone axis for the next photograph. [Equation 1] (Equation 1) Here, p(x ｊ ) means the probability of the j-th label in a particular zone axis, k indicates the total number of labels, and p(x ｉ ) means the probability of belonging to the i-th space group in the corresponding label, and n indicates the total number of space groups in the corresponding label.

10. The classification system of claim 9 , wherein the zone axis recommender arbitrarily recommends one of the zone axes when there are a plurality of zone axes having the smallest entropy.

Images