Material property prediction device, material property prediction method, and material property prediction program

Abstract

To improve the accuracy of predicting material properties of metals.SOLUTION: A material property prediction apparatus includes: a receiving unit 15A which receives an original image obtained by imaging a metal through a microscope; a division unit 15B which divides the original image into a plurality of partial images; and a prediction unit 15F which predicts material properties corresponding to the original image, on the basis of a label for material properties which is output for each partial image by a machine learning model by inputting the partial images to the machine learning model for each partial image.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a material property prediction device, a material property prediction method, and a material property prediction program. [Background technology]

[0002] A material property estimation device has been proposed as one technology for estimating the material properties of metals. For example, the material property estimation device reads a microstructure image of a test specimen taken with a microscope and estimates the material property value category to which the material belongs using a neural network model. For example, if the tensile strength of all test specimens is within the range of 560 MPa to 699 MPa, 11 categories, from category 0 to category 10, are assigned in increments of 10 MPa starting from 560 MPa. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Publication No. 2019-12037 Summary of the Invention [Problem to be solved by the invention]

[0004] However, in the prior art, including the material property estimation device described above, when the microstructure image to be estimated is input to the neural network model created in the prediction model creation process, it is compressed to an image size corresponding to the model. Therefore, in the prior art described above, information essential for predicting material properties is lost, resulting in a decrease in the accuracy of the prediction of material properties.

[0005] In one aspect, the present invention aims to provide a material property prediction device, a material property prediction method, and a material property prediction program that can improve the prediction accuracy of material properties of metals. [Means for solving the problem]

[0006] One aspect of the material property prediction device includes a receiving unit that receives an original image of a metal captured by a microscope, a dividing unit that divides the original image into a plurality of partial images, and a prediction unit that inputs each partial image into a machine learning model and predicts the material properties corresponding to the original image based on the material property labels output by the machine learning model for each partial image. [Effects of the Invention]

[0007] This can improve the accuracy of predicting metal material properties. [Brief explanation of the drawings]

[0008] [Figure 1] FIG. 1 is a block diagram illustrating an example of the functional configuration of a server device. [Figure 2] FIG. 2 is a schematic diagram (1) showing an example of resizing an original image. [Figure 3] FIG. 3 is a schematic diagram (2) showing an example of resizing an original image. [Figure 4] FIG. 4 is a schematic diagram showing an example of a method for dividing an original image. [Figure 5] FIG. 5 is a schematic diagram showing an example of input and output related to a machine learning model. [Figure 6] FIG. 6 is a schematic diagram showing an example of a method for counting output labels. [Figure 7] FIG. 7 is a flowchart showing the procedure of the material property prediction process. [Figure 8] FIG. 8 is a schematic diagram showing an example of a material property map. [Figure 9] FIG. 9 illustrates an example of a hardware configuration. DETAILED DESCRIPTION OF THE INVENTION

[0009] Hereinafter, embodiments of a material property prediction device, a material property prediction method, and a material property prediction program according to the present application will be described with reference to the accompanying drawings. Each embodiment merely illustrates one example or aspect, and the range of values, functions, and usage scenarios are not limited by such examples. Furthermore, each embodiment can be appropriately combined within the scope of not causing any contradiction in the processing content.

[0010] Fig. 1 is a block diagram showing an example of the functional configuration of a server device 10. The server device 10 shown in Fig. 1 provides a material property prediction function that uses a trained machine learning model to predict the material properties of a metal from an image of the metal captured by a microscope or the like.

[0011] Below, we will use the example of predicting metal hardness as just one example of material properties, but we would like to point out that other material properties, such as tensile strength, fatigue resistance, and creep resistance, may also be predicted.

[0012] The server device 10 is an example of a computer that provides the material property prediction function. In one embodiment, the server device 10 can provide the material property prediction function by executing software that realizes the material property prediction function. For example, the server device 10 can be realized as a server that provides the material property prediction function on-premise. Alternatively, the server device 10 can be realized as a PaaS (Platform as a Service) or SaaS (Software as a Service) application, thereby providing the material property prediction function as a cloud service.

[0013] As shown in Fig. 1, the server device 10 can be communicatively connected to a client terminal 30 via a network NW. For example, the network NW may be any type of communication network, whether wired or wireless, such as the Internet or a local area network (LAN). Note that Fig. 1 shows an example in which one client terminal 30 is connected to one server device 10, but any number of client terminals 30 may be connected.

[0014] The client terminal 30 corresponds to an example of a computer that receives the material property prediction function. For example, the client terminal 30 may be realized by a desktop or laptop personal computer. This is merely an example, and the client terminal 30 may be any computer such as a mobile terminal device or a wearable terminal.

[0015] Note that, although FIG. 1 shows an example in which the above-mentioned material property prediction function is provided in a client-server system, this is merely an example, and the above-mentioned material property prediction function may also be provided as a standalone function.

[0016] <One aspect of the issue> As explained in the background art section above, in the conventional technology, when the microstructure image to be estimated is input into the neural network model created in the prediction model creation process, it is compressed to an image size corresponding to the model.

[0017] Figure 2 is a schematic diagram (1) showing an example of resizing an original image. Figure 2 shows how the image size of an original image 21, which is 4 pixels wide by 4 pixels high, is resized to the input size of a machine learning model, which is 2 pixels wide by 2 pixels high.

[0018] 2, an original image 21 is compressed to a compressed image 22 of 2 pixels wide and 2 pixels high by thinning out 12 pixels out of 16 pixels. When resizing is achieved by compressing the image size in this way, 12 pixels are thinned out during compression, so the detailed information observable in the original image 21 is lost from the compressed image 22.

[0019] Here, Fig. 2 shows a schematic example, but in reality, there are cases where more information is lost than in the example shown in Fig. 2. For example, in the example described in the above-mentioned prior art, a microstructure image with an image size of 1252 pixels wide by 990 pixels high is compressed to 120 pixels wide by 94 pixels high, resulting in the loss of 1,228,200 pieces of information.

[0020] When the amount of information lost increases when the image is resized to fit the input size of a machine learning model, information essential for predicting material properties may also be lost. For example, when capturing an image of a tempered martensite structure, such as that found in spring steel, information about fine iron carbides that can be observed in the original image may be lost in the resized, compressed image.

[0021] These fine iron carbides have a significant impact on the material properties of metals. Therefore, even if a machine learning model is used to make predictions using compressed images that lose detailed information such as fine iron carbides as input data, it is difficult for the machine learning model to capture the material properties of metals, and the accuracy of the prediction of material properties may be reduced.

[0022] <One aspect of the problem-solving approach> Therefore, the material property prediction function according to this embodiment differs from the conventional techniques in that it takes an approach of resizing the image size of an original image of a metal structure captured by a microscope such as a scanning electron microscope to the input size of a machine learning model. That is, the material property prediction function according to this embodiment achieves resizing to the input size of a machine learning model by dividing the original image.

[0023] Figure 3 is a schematic diagram (2) showing an example of resizing an original image. As with the example in Figure 2, Figure 3 also shows a schematic diagram of how the image size of an original image 21, which is 4 pixels wide by 4 pixels high, is resized to the input size of a machine learning model, which is 2 pixels wide by 2 pixels high.

[0024] As shown in FIG. 3, the material property prediction function according to this embodiment divides an original image 21 into four partial images 21A to 21D, and resizes them to the input size of a machine learning model, which is 2 pixels wide by 2 pixels high.

[0025] In this way, the partial images 21A to 21D obtained by dividing the original image 21 contain all of the information contained in the original image 21 without any loss of information due to thinning or interpolation, and therefore, loss of detailed information such as fine iron carbides can be prevented.

[0026] Therefore, partial images that retain detailed information such as minute iron carbides can be input as input data into a trained machine learning model, increasing the likelihood that the trained machine learning model will capture the material properties of the metal.

[0027] Therefore, the material property prediction function according to this embodiment can improve the accuracy of predicting the material properties of metals. By realizing the prediction of material properties using a machine learning model in this way, it is possible to reduce the number of material evaluation tests that need to be performed and to support the construction of a materials database in which metallographic images and material properties are associated with each other.

[0028] <Configuration of Server Device 10> Next, an example of the functional configuration of the server device 10 according to this embodiment will be described. FIG. 1 shows a block diagram related to the material property prediction function of the server device 10. As shown in FIG. 1, the server device 10 has a communication control unit 11, a storage unit 13, and a control unit 15. Note that FIG. 1 only shows an excerpt of the functional units related to the material property prediction function, and the server device 10 may also be provided with functional units other than those shown.

[0029] The communication control unit 11 is a functional unit that controls communication with other devices such as the client terminal 30. As just one example, the communication control unit 11 can be realized by a network interface card such as a LAN card. In one aspect, the communication control unit 11 receives requests such as a material property prediction from the client terminal 30, or outputs the material property prediction results to the client terminal 30.

[0030] The storage unit 13 is a functional unit that stores various types of data. As just one example, the storage unit 13 is realized by internal, external, or auxiliary storage of the server device 10. For example, the storage unit 13 stores model data 13M.

[0031] The model data 13M is data related to a machine learning model used to predict the material properties of metals. The "machine learning model" referred to here refers to a trained machine learning model, and an example is a machine learning model that executes a classification task in which an image is input and outputs a Vickers hardness class. This is merely one example, and the model may also output a class related to a hardness index other than Vickers hardness, or a class related to a material property other than hardness, or may output a confidence level for each class.

[0032] For example, the machine learning model can be realized by a neural network, etc. One example is a convolved neural network (CNN) used for image recognition tasks.

[0033] Training samples for such machine learning models can be partial images obtained by dividing an image of a metal taken with a microscope, and the correct Vickers hardness labels assigned to the images.

[0034] That is, the partial images included in the training samples are used as explanatory variables of the machine learning model, and the correct labels are used as objective variables of the machine learning model, and the machine learning model is trained according to an arbitrary machine learning algorithm, such as deep learning, etc. This results in a trained machine learning model.

[0035] For example, the model data 13M may include hyperparameters related to the layer structure such as neurons and synapses of the input layer, hidden layer, and output layer that form a machine learning model, such as a CNN, as well as parameters related to the objective function such as the weights and biases of each layer.

[0036] In addition to the model data 13M, other data such as the original image itself, its shooting magnification, division size settings, and data used for filtering partial images may also be stored in the storage unit 13.

[0037] The control unit 15 is a functional unit that performs overall control of the server device 10. For example, the control unit 15 can be realized by a hardware processor. As shown in FIG. 1, the control unit 15 has a reception unit 15A, a division unit 15B, a filtering unit 15C, a model execution unit 15D, an aggregation unit 15E, and a prediction unit 15F. The control unit 15 may also be realized by hardwired logic or the like.

[0038] The reception unit 15A is a processing unit that receives various requests from the client terminal 30. As just one example, the reception unit 15A can receive from the client terminal 30 a prediction request to execute prediction of material properties.

[0039] When receiving such a request, the reception unit 15A can also receive a designation of an image for which material properties are to be predicted. In one aspect, the reception unit 15A can receive an image for which material properties are to be predicted from the client terminal 30 via the network NW. In another aspect, the reception unit 15A can also receive a designation from among images stored in the storage unit 13. In a further aspect, the reception unit 15A can also receive a designation from among images stored in a database server, file system, or the like (not shown).

[0040] Here, examples of images for which material properties are to be predicted include scanning electron microscope (SEM) images of metal structures photographed by a microscope such as a scanning electron microscope (SEM).

[0041] For example, when photographing a metal structure, the magnification of the SEM can be set to a value within the range of 5,000x to 50,000x. Setting the magnification within this range has the advantage of preventing situations where information on fine iron carbides, etc., cannot be observed due to photographing at a magnification lower than the above-mentioned magnification. Furthermore, it has the advantage of preventing bias in the amount of iron carbides appearing in the photographed area due to photographing at a magnification higher than the above-mentioned magnification.

[0042] In the following, in order to distinguish between the image obtained at the stage of capturing an image using a microscope and the partial images into which the image is divided, the former may be referred to as the "original image" and the latter as the "partial image."

[0043] The dividing unit 15B is a processing unit that divides the original image. FIG. 4 is a schematic diagram showing an example of a method for dividing an original image. FIG. 4 shows an example in which an SEM image of a metal structure is received by the receiving unit 15A, as just one example of an original image 40. As shown in FIG. 4, the dividing unit 15B divides the original image 40 into a mesh. By performing the mesh division in this manner, 15 partial images 40A to 40P are obtained from the original image 40.

[0044] The image size of the grid cells during such mesh division can be set to a width and height that is equal to or smaller than the size of the input layer of a machine learning model, such as a CNN, used to predict the material properties of metals.

[0045] For example, if the image size of the grid cells is made too small, in other words, if the mesh density is made too fine, the iron carbides that appear within the range of the grid cells may be unevenly distributed, which may adversely affect the prediction accuracy.

[0046] Such uneven distribution of iron carbides is more frequently observed when the width and height of the image size of the grid cell are 2 μm or less, so it is preferable to set the width and height of the image size of the grid cell to values ​​exceeding 2 μm.

[0047] The filtering unit 15C is a processing unit that filters partial images. The term "filtering" here refers to selection, such as extracting partial images that satisfy certain conditions listed below from the partial images divided from the original image, or excluding partial images that satisfy certain conditions.

[0048] In one aspect, the filtering unit 15C extracts, from the M partial images into which the original image is divided by the division unit 15B, partial images that contain detailed information that contributes to predicting material properties, for example, partial images in which the number of pixels in the frequency band corresponding to fine iron carbides is equal to or greater than a threshold value.

[0049] More specifically, the filtering unit 15C applies a band-pass filter, the center frequency of which corresponds to iron carbide, to each of the M partial images. Based on the result of applying this band-pass filter, the filtering unit 15C calculates the number of pixels in the partial images that correspond to iron carbide. At this time, if the number of pixels corresponding to iron carbide in the partial image is equal to or greater than a threshold value Th1, the filtering unit 15C extracts the partial image as a sample to be predicted. On the other hand, if the number of pixels corresponding to iron carbide in the partial image is less than the threshold value Th1, the filtering unit 15C prohibits the extraction of the partial image as a sample to be predicted. This enables filtering to extract partial images that contribute to the prediction of material properties as samples to be predicted.

[0050] In another aspect, the filtering unit 15C excludes, from the M partial images into which the original image is divided by the division unit 15B, partial images that contain information that becomes noise when predicting material properties, such as peculiar parts that are similar to contamination such as organic matter.

[0051] More specifically, the filtering unit 15C applies the following template matching to each of the M partial images. For example, in template matching, images of one or more patterns of unique parts observed in an SEM image are used as templates. These templates are then compared with the input partial image, which is the input image, while being enlarged, reduced, and rotated. This searches for areas where the similarity between the unique part template and the partial image is equal to or greater than threshold Th2, or areas where the distance between the unique part template and the partial image is equal to or less than threshold Th3. If the search results in an area similar to the unique part template, the filtering unit 15C excludes the partial image from the sample to be predicted. On the other hand, if no area similar to the unique part template is obtained, the filtering unit 15C prohibits the partial image from being excluded from the sample to be predicted. This enables filtering to exclude partial images containing unique parts such as contamination from the sample to be predicted.

[0052] The filtering unit 15C can perform one or both of filtering to extract partial images that contribute to the prediction of material properties and filtering to remove partial images that include peculiar parts such as contamination. Although an example in which partial image filtering is performed has been given here, all partial images may be used as samples to be predicted, or some partial images may be randomly extracted or excluded.

[0053] The model execution unit 15D is a processing unit that executes a machine learning model. In one embodiment, the model execution unit 15D deploys a trained machine learning model on a work area (not shown) in accordance with the model data 13M stored in the storage unit 13. Then, for each partial image extracted by the filtering unit 15C as a sample to be predicted, the model execution unit 15D inputs the partial image to the trained machine learning model. This allows the model execution unit 15D to obtain a label related to the class of material properties output by the machine learning model to which the partial image is input.

[0054] Fig. 5 is a schematic diagram showing an example of input and output related to a machine learning model. Fig. 5 shows an example in which the 15 partial images 40A to 40P shown in Fig. 4 are input to the machine learning model M. Furthermore, Fig. 5 shows an example in which the machine learning model M executes a multi-class classification task in which the input data is classified into four classes: "500HV," "600HV," "700HV," and "800HV."

[0055] As shown in FIG. 5, when partial image 40A is input to machine learning model M, machine learning model M outputs a label of the Vickers hardness class "800HV." When partial image 40B is input to machine learning model M, machine learning model M outputs a label of the Vickers hardness class "800HV." When partial image 40H is input to machine learning model M, machine learning model M outputs a label of the Vickers hardness class "600HV." Similarly, when partial image 40P is input to machine learning model M, machine learning model M outputs a label of the Vickers hardness class "800HV."

[0056] Hereinafter, among the labels related to the classes of material properties, the latter may be referred to as the "output label" in order to distinguish it from the labels related to the classes of material properties output by the machine learning model.

[0057] The counting unit 15E is a processing unit that counts the output labels of the machine learning model by material property class. Fig. 6 is a schematic diagram showing an example of a method for counting the output labels. Fig. 6 shows the classification result 50 shown in Fig. 5 in a state in which each of the output labels of the machine learning model M is labeled to each of the partial images 40A to 40P.

[0058] Here, if machine learning model M performs a multi-class classification task to classify input data into four classes, "500HV," "600HV," "700HV," and "800HV," the output labels of the machine learning model are aggregated for each of the four classes.

[0059] As shown in FIG. 6, a total of 10 output labels corresponding to each of partial images 40A, 40B, 40C, 40E, 40F, 40G, 40I, 40M, 40O, and 40P are tallied to obtain a total value of "10" for the output label of class "800HV." Furthermore, a total of four output labels corresponding to each of partial images 40D, 40K, 40L, and 40N are tallied to obtain a total value of "4" for the output label of class "700HV." Furthermore, a total of one output label corresponding to partial image 40H is tallied to obtain a total value of "1" for the output label of class "600HV." Since there is no output label for class "500HV," the total value of the output label of class "500HV" is "0."

[0060] In this way, the aggregated values ​​of the output labels for the material property classes "800HV", "700HV", "600HV" and "500HV", "10", "4", "1" and "0", are obtained as the aggregated result 60.

[0061] The prediction unit 15F is a processing unit that predicts material properties corresponding to the original image based on the labels of material properties output by the machine learning model for each partial image. As just one example, the prediction unit 15F predicts material properties corresponding to the original image based on the aggregated values ​​of the output labels aggregated for each class by the aggregation unit 15E.

[0062] For example, in the example shown in Figure 6, the class "800HV" with the largest aggregated output label value is selected from the aggregated values ​​"10", "4", "1" and "0" for the classes "800HV", "700HV", "600HV" and "500HV" obtained as the aggregation result 60.

[0063] The class "800HV" selected in this way is output as the predicted result of the material properties to any output destination. As just one example, the predicted result of the material properties can be output to an external device such as the client terminal 30, or to software or a service that uses the predicted result of the material properties.

[0064] Predicting material properties by such majority voting allows for global prediction based on features that appear frequently in partial images, resulting in prediction of material properties that are robust to the influence of localized anomalous areas, such as contamination, etc. For example, in the example shown in Figure 6, it is possible to prevent the prediction results from being affected by localized features of partial images 40D, 40K, 40L, and 40N, or partial image 40H.

[0065] <Processing flow> 7 is a flowchart showing the procedure of the material property prediction process. This process can be executed, by way of example only, when a prediction request for executing a prediction of material properties is received from the client terminal 30.

[0066] 7, the receiving unit 15A receives a designation of an original image for which material properties are to be predicted (step S101). Subsequently, the dividing unit 15B divides the original image received in step S101 into a plurality of partial images (step S102).

[0067] Next, the filtering unit 15C performs filtering such as extracting partial images that satisfy a specific condition from the partial images obtained by the division in step S102, or excluding partial images that satisfy a specific condition (step S103).

[0068] Thereafter, loop processing 1 is executed in which the processing of step S104 described below is repeated a number of times corresponding to the M partial images extracted as samples to be predicted by filtering in step S103. Note that, although Fig. 7 shows an example in which the processing of step S104 is executed as loop processing 1, the processing of step S104 does not necessarily have to be executed serially, and may be executed in parallel for each of the M partial images.

[0069] That is, the model execution unit 15D inputs the partial image extracted as a sample to be predicted by filtering in step S103 into the machine learning model, thereby obtaining a label for the class of material properties output by the machine learning model (step S104).

[0070] By repeating this loop process 1, output labels of the machine learning model are obtained for each of the M partial images.

[0071] Thereafter, the aggregation unit 15E aggregates the output labels of the machine learning model by class of material properties (step S105). Then, the prediction unit 15F selects the class with the largest aggregated value of the output label from the aggregated values ​​by class obtained in step S105 (step S106). Finally, the prediction unit 15F outputs the class selected in step S106 to the client terminal 30 or the like as a prediction result of the material properties (step S107), and ends the process.

[0072] <One aspect of the effect> As described above, the server device 10 according to this embodiment resizes the original image to the input size of the machine learning model by dividing the original image. Therefore, the server device 10 according to this embodiment can improve the prediction accuracy of the material properties of metals. By realizing prediction of material properties using a machine learning model in this way, it is possible to reduce the number of material evaluation tests performed and to support the construction of a materials database in which metallographic images and material properties are associated with each other.

[0073] Furthermore, the server device 10 according to this embodiment predicts material properties by majority voting of the output labels of the machine learning models for each class. Therefore, the server device 10 according to this embodiment performs global prediction based on features that appear frequently in partial images, thereby realizing prediction of material properties that are robust to the influence of localized peculiarities, such as contamination.

[0074] <Application example> The above-described embodiment is merely an example, and various applications are possible.

[0075] For example, the server device 10 may further include a generation unit that generates a map of the material properties of the original image by mapping labels related to the class of material properties output by a machine learning model to which the partial image is input for each area corresponding to the partial image into which the original image is divided.

[0076] FIG. 8 is a schematic diagram showing an example of a material property map. As shown in FIG. 8, a hardness heat map 70 is provided with a gradation of shades corresponding to the output labels of the machine learning model for each area corresponding to each of the partial images 40A to 40P into which the original image 40 is divided. By generating such a hardness heat map 70, it is possible to visualize the distribution of Vickers hardness. Note that FIG. 8 shows an example in which a map is generated using changes in gradation of shades, but it goes without saying that a map can also be generated using other elements such as color.

[0077] <Modification> The above-described embodiment is merely an example, and various modifications are possible.

[0078] For example, in the above embodiment, an example was given in which material properties were predicted by majority vote of the output labels of the machine learning model for each class. However, the material properties do not necessarily have to be predicted by majority vote of the output labels. As just one example, the prediction unit 15F can output the statistical value of the output labels of the machine learning model obtained for each partial image, such as the average value, as the predicted result of the material properties. In the example shown in FIG. 6, the statistical value of the output label can be calculated as 760HV by calculating (800HV × 10 + 700HV × 4 + 600HV × 1 + 500HV × 0) ÷ 15. This makes it possible to predict material properties that are smoothed between partial images.

[0079] Furthermore, among the processes described in the above embodiments, all or part of the processes described as being performed automatically can be performed manually, or all or part of the processes described as being performed manually can be performed automatically using a known method. In addition, the information including the processing procedures, specific names, various data, and parameters shown in the above documents and drawings can be changed as desired unless otherwise specified. For example, the various information shown in each drawing is not limited to the information shown in the drawings.

[0080] Furthermore, the components of each device shown in the figure are conceptual functional components and do not necessarily have to be physically configured as shown in the figure. In other words, the specific form of distribution and integration of each device is not limited to that shown in the figure, and all or part of them can be functionally or physically distributed and integrated in any unit depending on various loads, usage conditions, etc.

[0081] Furthermore, the effects of each embodiment described in this specification are merely examples and are not intended to be limiting, and other effects may also be provided.

[0082] <Hardware configuration> The various processes described in the above embodiments can be realized by executing a prepared program on a computer such as a personal computer or a workstation. Hereinafter, an example of a computer that executes a material property prediction program having the same functions as those in the above embodiments will be described with reference to FIG.

[0083] Fig. 9 is a diagram showing an example of a hardware configuration. As shown in Fig. 9, a computer 100 has an operation unit 110a, a speaker 110b, a camera 110c, a display 120, and a communication unit 130. The computer 100 also has a CPU 150, a ROM 160, an HDD 170, and a RAM 180. These units 110 to 180 are connected via a bus 140.

[0084] 9, the HDD 170 stores a material property prediction program 170a that performs the same functions as the receiving unit 15A, the dividing unit 15B, the filtering unit 15C, the model executing unit 15D, the tallying unit 15E, and the predicting unit 15F shown in the above embodiment. This material property prediction program 170a may be integrated or separated, similar to the respective components of the receiving unit 15A, the dividing unit 15B, the filtering unit 15C, the model executing unit 15D, the tallying unit 15E, and the predicting unit 15F shown in FIG. 1. In other words, the HDD 170 does not necessarily have to store all of the data shown in the above embodiment, as long as the data used for processing is stored in the HDD 170.

[0085] Under such an environment, the CPU 150 reads the material property prediction program 170a from the HDD 170 and loads it into the RAM 180. As a result, the material property prediction program 170a functions as a material property prediction process 180a, as shown in FIG. 9. The material property prediction process 180a loads various data read from the HDD 170 into an area of ​​the storage area of ​​the RAM 180 allocated to the material property prediction process 180a, and executes various processes using the loaded data. For example, an example of the process executed by the material property prediction process 180a may include the process shown in FIG. 7. Note that the CPU 150 does not necessarily need to operate all of the processing units shown in the first embodiment above; it is sufficient that the processing units corresponding to the processes to be executed are virtually implemented.

[0086] The material property prediction program 170a does not necessarily have to be stored in the HDD 170 or the ROM 160 from the beginning. For example, the material property prediction program 170a may be stored in a "portable physical medium" such as a flexible disk, a so-called FD, a CD-ROM, a DVD disk, a magneto-optical disk, or an IC card that is inserted into the computer 100. The computer 100 may then acquire and execute the material property prediction program 170a from such a portable physical medium. Alternatively, the material property prediction program 170a may be stored in another computer or server device connected to the computer 100 via a public line, the Internet, a LAN, a WAN, or the like. The material property prediction program 170a thus stored may be downloaded to the computer 100 and then executed. [Explanation of symbols]

[0087] 10 Server device 11 Communication control section 13 Storage section 13M model data 15 Control Unit 15A Reception 15B Split section 15C Filtering section 15D Model Execution Department 15E Counting Unit 15F Forecasting Department 30 client terminals

Claims

1. a receiving unit that receives an original image of a metal captured by a microscope; a division unit that divides the original image into a plurality of partial images; a prediction unit that predicts material properties corresponding to the original image based on labels of material properties output by a machine learning model for each of the partial images by inputting the partial images to the machine learning model for each of the partial images; and The material property prediction device is characterized in that the division unit divides the original image into image sizes that are equal to or smaller than the size of the input layer of the machine learning model.

2. The material property prediction device described in Claim 1, characterized in that the prediction unit aggregates the labels output by the machine learning model by class of the material properties and selects the class with the largest aggregated value among the aggregated values.

3. A material property prediction device as described in claim 1 or 2, characterized in that it further has a generation unit that generates a map of material properties corresponding to the original image by mapping labels related to the classes of material properties output by the machine learning model for each area corresponding to each of the multiple partial images.

4. The method further comprises a filtering unit that extracts a partial image from the plurality of partial images in which the number of pixels in a frequency band corresponding to a specific subject is equal to or greater than a threshold value; 4. The material property prediction device according to claim 1, wherein the prediction unit inputs the partial image extracted by the filtering unit to the machine learning model.

5. The method further comprises a filtering unit that excludes partial images that include peculiar parts similar to contamination from among the plurality of partial images; The material property prediction device according to any one of claims 1 to 3, characterized in that the prediction unit prohibits partial images excluded by the filtering unit from being input into the machine learning model.

6. Accepting an original image of a metal taken by a microscope; Dividing the original image into a plurality of sub-images; predicting material properties corresponding to the original image based on labels of material properties output by the machine learning model for each of the partial images by inputting the partial images to a machine learning model for each of the partial images; The computer executes the processing, The material property prediction method is characterized in that the dividing process divides the original image into image sizes that are equal to or smaller than the size of the input layer of the machine learning model.

7. Accepting an original image of a metal taken by a microscope; Dividing the original image into a plurality of sub-images; predicting material properties corresponding to the original image based on labels of material properties output by the machine learning model for each of the partial images by inputting the partial images to a machine learning model for each of the partial images; Have the computer execute the process, The material property prediction program is characterized in that the division process divides the original image into image sizes that are equal to or smaller than the size of the input layer of the machine learning model.

Images