JP7311455B2 - Scrap grade determination system, scrap grade determination method, estimation device, learning device, learned model generation method, and program - Google Patents

Description

The present invention relates to a scrap grade determination system, a scrap grade determination method, an estimation device, a learning device, a learned model generation method, and a program.

In general, at the scrap receiving site of an electric furnace manufacturer, an inspection inspector observes the scrap iron scrap being lifted from a truck or ship by a lifting device such as a lifting magnet, transported, and unloaded to the scrap yard, and judges the grade ratio of the iron scrap. .

JP-A-7-286969

However, since acceptance inspectors visually check iron scrap from a remote location for safety reasons, there are cases where the accuracy of determining the grade ratio is not sufficient. In addition, the judgment criteria may vary depending on the experience of the inspector.

In particular, it is difficult to determine the grade ratio of iron scrap called heavy scrap, which is graded according to size, because iron scrap of various sizes coexists.

The present invention has been made in view of the above problems, and its main purpose is to provide a scrap grade determination system, a scrap grade determination method, an estimation device, and a learning method capable of improving the determination accuracy of the grade ratio of iron scrap. An object of the present invention is to provide an apparatus, a method for generating a trained model, and a program.

In order to solve the above problems, a scrap grade determination system according to one aspect of the present invention includes a crane that lifts a set of iron scraps in which iron scraps of a plurality of grades according to size are mixed by a hoisting tool, and A camera that captures a set of lifted iron scraps to generate a camera image, a preprocessing unit that divides the camera image into a plurality of image pieces, and a machine using the iron scrap image as input data and the grade as training data. an estimating unit for estimating a grade for each of the plurality of image pieces using a trained model generated in advance by learning; and a ratio calculation unit for calculating the ratio of the grades included in the set of iron scraps.

In another aspect of the present invention, there is provided a method for judging scrap grades, wherein a set of iron scraps in which iron scraps of a plurality of grades according to size are mixed is lifted by a hoisting tool of a crane, and the scraps lifted by the hoisting tool are A set of iron scraps is photographed by a camera to generate a camera image, the camera image is divided into a plurality of image pieces, the image of the iron scrap is input data, and the grade is training data. Using a model, the grade is estimated for each of the plurality of image pieces, and based on the estimated grade for each of the plurality of image pieces, the grade included in the set of iron scraps lifted by the sling. Calculate the percentage.

In addition, the estimating device of another aspect of the present invention is a camera image generated by a camera that captures a set of iron scraps mixed with a plurality of grades of iron scraps according to the size, lifted by a crane sling. , a preprocessing unit that divides the camera image into a plurality of image pieces, and a learned model generated in advance by machine learning using the image of iron scrap as input data and the grade as teacher data, an estimating unit for estimating a grade for each of the plurality of image pieces; and a percentage of the grade included in the set of iron scraps lifted by the sling based on the grade estimated for each of the plurality of image pieces. and a ratio calculation unit that calculates the

Further, a program according to another aspect of the present invention captures a camera image generated by a camera that captures a set of iron scraps mixed with a plurality of grades of iron scraps according to size, lifted by a crane sling. obtaining, dividing the camera image into a plurality of image pieces, using a trained model generated in advance by machine learning with the image of iron scrap as input data and the grade as teacher data, each of the plurality of image pieces estimating the grade of each of the plurality of image pieces, and calculating the ratio of grades included in the set of iron scraps lifted by the lifting device based on the estimated grade of each of the plurality of image pieces. to execute.

A learning device according to another aspect of the present invention includes an image acquisition unit that acquires a learning image generated by a camera that captures a set of iron scraps in which iron scraps of a plurality of grades according to size are mixed, a preprocessing unit that divides the learning image into a plurality of image pieces; a grade acquisition unit that acquires grades of iron scrap appearing in each of the plurality of image pieces; As data, a learning unit for generating a trained model for estimating the grade of iron scrap appearing in the image by machine learning.

In another aspect of the present invention, there is provided a method of generating a trained model, which acquires a learning image generated by a camera that captures a set of iron scraps in which a plurality of grades of iron scraps according to size are mixed, The learning image is divided into a plurality of image pieces, the grade of iron scrap appearing in each of the plurality of image pieces is acquired, the image pieces are used as input data, the grades are used as training data, and the iron appearing in the image is obtained. A trained model for estimating the grade of scrap is generated by machine learning.

Further, a program according to another aspect of the present invention acquires a learning image generated by a camera that captures a set of iron scraps in which a plurality of grades of iron scrap according to size are mixed; is divided into a plurality of image pieces, obtaining grades of iron scraps appearing in each of the plurality of image pieces, and using the image pieces as input data and the grades as teacher data, the iron scraps appearing in the image Generating a trained model for estimating the grade of scrap by machine learning.

ADVANTAGE OF THE INVENTION According to this invention, it becomes possible to improve the determination accuracy of the grade ratio of iron scrap.

FIG. 2 is a diagram showing an example of a scrap receiving site; It is a figure which shows the structural example of a scrap grade determination system. FIG. 1 shows heavy scrap grades and requirements. FIG. 4 is a diagram showing an example of a data set; FIG. 10 is a diagram showing an example of a learning image; It is a figure which shows the modification of the image for learning. It is a figure which shows the modification of the image for learning. It is a figure explaining a learning phase. It is a figure which shows the procedure example of a learning phase. It is a figure which shows the procedure example of a machine-learning process. It is a figure explaining an inference phase. FIG. 10 is a diagram illustrating an example of the procedure of an inference phase; It is a figure which shows the example of a camera image. FIG. 4 is a diagram illustrating an example of preprocessing of camera images; It is a figure which shows the calculation example of a weight ratio. It is a figure which shows the modification of a weight index. It is a figure which shows the output example of a determination result.

BEST MODE FOR CARRYING OUT THE INVENTION Hereinafter, embodiments of the present invention will be described with reference to the drawings.

[System Overview]
FIG. 1 is a diagram showing an example of a scrap receiving site where a scrap grade determination system 100 according to an embodiment is installed. FIG. 2 is a diagram showing a configuration example of the scrap grade determination system 100. As shown in FIG. A scrap grade determination system 100 includes an estimation device 1 and a learning device 3 according to the embodiment.

The scrap grade determination system 100 uses a camera 2 to photograph iron scrap S lifted by a sling 91 of a crane 9 at a scrap receiving site of an electric furnace manufacturer or the like, determines the grade ratio from the image, and outputs the determination result to the output unit 6, It is a system for outputting to 7.

At the scrap receiving site, the iron scrap S is lifted from the truck T by the hoist 91 of the crane 9, transported, and lowered to the scrap yard Y. The suspender 91 is, for example, a lifting magnet. The hoisting tool 91 is not limited to this, and may be a grab bucket or the like.

In the iron scrap S, iron scrap S of various sizes are mixed. Iron scrap S is, for example, heavy waste. As shown in FIG. 3, heavy scraps are graded according to size (specifically, thickness, width or height, and length) and weight (for example, H1 to L1).

A camera 2 for photographing the iron scrap S lifted by the lifting tool 91 is installed at the scrap receiving site. The camera 2 may be a video camera that generates moving images, or a still camera that generates still images.

A plurality of cameras 2 may be provided so as to photograph the iron scrap S lifted by the hanging tool 91 from different angles. It is preferable that the camera 2 photographs the iron scrap S lifted by the hanging tool 91 horizontally from the side or upward from the bottom.

The scrap grade determination system 100 may be applied to a scrap shipping site of a scrap processing company. At the scrap shipping site, the iron scrap S is lifted from the scrap yard Y by the hoist 91 of the crane 9, conveyed, and loaded onto the truck T.

As shown in FIG. 2, the scrap grade determination system 100 includes an estimation device 1, a camera 2, a learning device 3, a storage device 5, and output units 6,7. These devices are capable of network communication with each other via a communication network such as a LAN. Note that the estimation device 1 and the learning device 3 may be configured as an integrated device.

The estimation device 1 includes a control unit 10 . The control unit 10 is a computer including a CPU, RAM, ROM, nonvolatile memory, an input/output interface, and the like. The CPU of the control unit 10 executes information processing according to a program loaded from the ROM or nonvolatile memory to the RAM.

The control unit 10 includes an acquisition unit 11 , a detection unit 12 , a preprocessing unit 13 , an estimation unit 14 , a ratio calculation unit 15 and a total calculation unit 16 . These functional units are realized by the CPU of the control unit 10 executing information processing according to programs loaded from the ROM or nonvolatile memory to the RAM.

The program may be supplied via an information storage medium such as an optical disk or memory card, or may be supplied via a communication network such as the Internet or LAN.

The learning device 3 also includes a controller 30 similar to that of the estimation device 1 . The control unit 30 includes an acquisition unit 31 , a preprocessing unit 32 and a learning unit 33 . The acquisition unit 31 is an example of an image acquisition unit and a grade acquisition unit.

The estimation device 1 and the learning device 3 can access the storage device 5 . The learned model generated by the learning device 3 is stored in the storage device 5 so as to be readable by the estimation device 1 .

The camera 2 photographs the iron scrap S hoisted by the hoisting tool 91 and inputs the generated camera image to the estimation device 1 . The estimation device 1 determines the grade ratio of the iron scrap S from the camera image generated by the camera 2 and transmits the determination result to the output units 6 and 7 .

Output units 6 and 7 output determination results from the estimation device 1 . The output unit 6 is, for example, a terminal such as a tablet computer carried by the inspector A. FIG. The output unit 7 is, for example, a printer for issuing a receipt.

[Learning phase]
FIG. 4 is a diagram showing an example of a data set used in the learning phase as a method of generating a trained model according to the embodiment. The dataset used in the training phase contains training images and grade labels.

The learning images and grade labels are managed in association with each other in a table as shown in FIG. The training image is used as input data to the learning model. Grade labels are used as teacher data.

The grade label represents the grade of iron scrap that appears in the training images. If multiple scrap iron appears in the training image, the grade label represents the grade of the representative scrap iron among them.

In the illustrated example, a grade of H1, H2, H3, or L1 is associated with each of the plurality of training images. The grade label is determined and input by, for example, an acceptance inspector who has seen the learning image.

FIG. 5 is a diagram showing an example of a learning image. In this embodiment, the learning image is an image piece (mesh image) obtained by dividing a camera image of a set of iron scraps S lifted by the hoisting tool 91 by the camera 2 . A part of the iron scrap S appears in the image piece.

The preprocessing for obtaining image fragments from camera images is the same as the preprocessing in the inference phase, which will be described later. That is, the function of the preprocessing unit 32 of the learning device 3 is the same as the function of the preprocessing unit 13 of the estimation device 1 . Details of the preprocessing will be described later.

As the learning image, not only the above-mentioned image pieces, but also camera images before being divided into image pieces may be used, or images in which one or more iron scraps photographed at another location appear may be used. may

As shown in FIGS. 6 and 7, in the pretreatment, representative iron scrap MS and other iron scrap NS may be identified and displayed. As a result, iron scraps to which attention should be focused are specified, so that learning can be made more efficient.

Specifically, as shown in FIG. 6, only the representative iron scrap MS may be colored, or as shown in FIG. 7, scraps other than the representative iron scrap MS may be excluded by masking.

FIG. 8 is a diagram for explaining the learning phase realized by the learning device 3. As shown in FIG. FIG. 9 is a diagram showing an example procedure of the learning phase. FIG. 10 is a diagram showing a procedure example of the machine learning process S12. The control unit 30 of the learning device 3 functions as an acquisition unit 31, a preprocessing unit 32, and a learning unit 33 by executing the processes shown in FIGS. 9 and 10 according to programs.

A learning model is, for example, a convolutional neural network, and includes a convolutional layer, a pooling layer, a fully connected layer, and an output layer. In particular, a deep neural network in which neurons are combined in multiple stages is suitable.

The output layer is provided with elements corresponding to each grade. An element corresponding to each grade is composed of, for example, a softmax function, and an output value represented by a real number between 0 and 1 can be treated as a probability of corresponding to the grade.

The layer structure is not limited to the illustrated example, and the number of convolution layers, pooling layers, and fully connected layers may be different. Machine learning other than neural networks such as support vector machines, Gaussian processes, or decision trees may also be used.

As shown in FIG. 9, the control unit 30 of the learning device 3 first extracts some data sets from a large number of data sets as training data (S11), and performs machine learning processing using the extracted training data. (S12: processing by the learning unit 33).

Next, the control unit 30 extracts some data sets other than the training data from among the large number of data sets as test data (S13), and evaluates the trained model using the extracted test data ( S14). A trained model with a predetermined evaluation or higher is stored in the storage device 5 (S15).

As shown in FIG. 10, in the machine learning process S12, the control unit 30 inputs learning images (image fragments in this embodiment) included in the data set to the learning model as input data (S21). Calculation is performed (S22), and the probability corresponding to each grade is output as output data (S23).

Next, the control unit 30 calculates the difference between each grade label as teacher data included in the data set and the probability corresponding to each grade as output data (S24), and performs error backpropagation calculation for learning. The parameters of the model are adjusted (S25). The grade label represents a binary value of 0 or 1 as to whether or not it corresponds to the grade.

As described above, a trained model for estimating the grade of iron scrap appearing in the image is generated.

[Inference phase]
FIG. 11 is a diagram illustrating an inference phase as a scrap grade determination method according to an embodiment, which is implemented in the estimation device 1. FIG. FIG. 12 is a diagram illustrating a procedure example of the inference phase. The control unit 10 of the estimating device 1 executes the processing shown in FIG. Function.

First, the control unit 10 acquires a camera image generated by the camera 2 that captures a collection of iron scraps S lifted by the hoisting tool 91 (S31: processing performed by the acquisition unit 11). FIG. 13 is a diagram showing an example of a camera image.

Next, the control unit 10 detects the hanging tool 91 in the camera image (S32: function as the detection unit 12).

Specifically, the control unit 10 uses an object detection model such as YOLO (You Only Look Once) or SSD (Single Shot MultiBox Detector) to detect the range M of the sling 91 in the camera image (Fig. 13).

Next, the control unit 10 performs preprocessing on the acquired camera image (S33 to S37: processing as the preprocessing unit 13). FIG. 14 is a diagram for explaining an example of a preprocessing procedure for a camera image.

In S33, the control section 10 cuts out a rectangular area of a predetermined size defined on the basis of the sling 91 from the camera image to obtain a cutout image HC (see FIG. 14A). The size of the rectangular area to be cut out is determined in advance so as to include all the set of iron scraps S lifted by the lifting tool 91 .

Specifically, the left and right edges of the clipped image HC are located slightly outside the left and right edges of the hanger 91 . The width of the clipped image HC is slightly wider than the width of the hanger 91 . The upper end of the clipped image HC is positioned at the upper end of the hanger 91 . The lower end of the clipped image HC is located below the lower end of the hanger 91 by a predetermined length.

In S34, the control unit 10 divides the clipped image HC into a plurality of image pieces (mesh images) MG (see FIG. 14(b)).

In S35, the control unit 10 sorts the plurality of image pieces MG into image pieces MG1 used for class estimation and image pieces MG0 not used for class estimation accuracy.

In the present embodiment, the control unit 10 selects the image piece 91 at a predetermined position with respect to the hanger 91 from among the plurality of image pieces MG as the image piece MG1 used for estimating the grade. Specifically, since the set of iron scraps S lifted by the hoisting tool 91 often has a substantially inverted cone shape (substantially an inverted triangular shape when viewed from the side) whose width gradually narrows as it goes downward, the hoisting tool The image piece MG belonging to a range that gradually narrows downward from 91 is selected as the image piece MG1 used for grade estimation (see FIG. 14(c)).

Therefore, by using the image piece MG1 belonging to a range that gradually narrows downward from the hanging tool 91 for estimating the grade and not using the image piece MG0 belonging to the other range for estimating the grade, the iron scrap S is Since the image pieces MG that do not appear are excluded, it is possible to improve the accuracy of the grade determination.

Not limited to this, the control unit 10 compares an image of a set of iron scraps S lifted by the hoisting tool 91 with a background image of an image taken when the hoisting tool 91 and the set of iron scraps S are not present. , the image piece MG with the difference may be selected as the image piece MG1 used for class estimation.

In S36, the control unit 10 divides the clipped image HC to obtain a plurality of image pieces MG1 used for grade estimation.

In S37, the control unit 10 changes the size of each of the plurality of image pieces MG1 used for grade estimation to a size that can be input to the trained model. Specifically, the size of each of the plurality of image pieces MG1 (eg, 450×500 pixels) is compressed to a size that can be input to the trained model (eg, 224×224 pixels).

In this way, by compressing the size of each of the plurality of image pieces MG1 obtained by dividing the clipped image HC to a size that can be input to the trained model, the clipped image HC can be input to the trained model. Since the image compression ratio can be reduced compared to the case where the image is directly compressed to , it is possible to improve the determination accuracy.

Next, the control unit 10 uses the learned model to estimate the grade for each of the plurality of image pieces MG1 (S38 to S41: processing by the estimation unit 14). Specifically, the control unit 10 inputs the image piece MG1 to the learned model (S38), performs calculations using the learned model (S39), and outputs the probability of corresponding to each grade (S40).

Next, the control unit 10 calculates the weight ratio of the grade included in the set of iron scraps S lifted by the hoisting tool 91 based on the grade estimated for each of the plurality of image pieces MG1 (S42: Ratio processing as the calculation unit 15).

Specifically, as shown in FIG. 15, the number of image pieces MG1 (the number of meshes) is counted for each grade, and the number of image pieces MG1 for each grade and the weight index (a to d ) are respectively multiplied, the weight ratio of the grade contained in the set of iron scraps S lifted by the lifting tool 91 is calculated.

Without being limited to this, as shown in FIG. 16, the weight index may be subdivided according to not only the grade but also the type of steel plate. That is, even if the weight ratio of each grade is calculated based on the number of image pieces MG1 of each grade, the type of iron scrap S, and the weight index (a-1 to 6) determined for each grade and type good.

The type of iron scrap S may be input by a user, or may be determined using a trained model such as a convolutional neural network.

The control unit 10 repeats the processes of S31 to S42 each time the hoisting device 91 lifts a set of iron scraps S until all the iron scraps S are unloaded from the bed of the truck T (S43). Whether or not all iron scrap S has been unloaded is determined by, for example, inspection staff A, and is input.

When all the iron scraps S have been unloaded from the bed of the truck T (S43: YES), the control unit 10 controls the grade weight ratio calculated each time the sling 91 lifts a set of iron scraps S. , the weight ratio of the grade is calculated for all the iron scraps S (S44: processing as the total calculation unit 16).

Specifically, by calculating the average value of the grade weight percentages calculated each time the hoisting tool 91 lifts a set of iron scraps S, the grade weight percentages for all the iron scraps S are determined. An arithmetic average may be used to calculate the average value, or a weighted average according to weight may be used.

After that, the control unit 10 transmits the weight ratio of the grade calculated for all the iron scrap S unloaded from the truck T to the output units 6 and 7 as the determination result (S45). The output units 6 and 7 output the received determination results by displaying, printing, or the like. FIG. 17 is a diagram illustrating an output example of determination results.

According to the embodiment described above, by determining the grade for each of the plurality of divided image pieces, each iron scrap becomes more likely to be noticed, so that it is possible to improve the determination accuracy. .

In particular, the size of each of the divided image pieces is compressed to a size that can be input to the trained model, and the grade is determined, so it is possible to perform the determination while ensuring the characteristics of the iron scrap. be.

Although the embodiments of the present invention have been described above, the present invention is not limited to the embodiments described above, and it goes without saying that various modifications can be made by those skilled in the art.

1 estimation device 2 camera 3 learning device 5 storage device 6, 7 output unit 10 control unit 11 acquisition unit 12 detection unit 13 preprocessing unit 14 estimation unit 15 ratio calculation unit 16 overall calculation Part, 9 Crane, 91 Lifting tool, 100 Scrap grade determination system, A Inspection staff, Y Scrap yard, S Steel scrap, T Scrap loading vehicle, HC Clipped image, MG Image fragment, C Camera image, B Background image, CB Background Removed image

Claims (16)

a crane that lifts a set of iron scraps, in which iron scraps of multiple grades according to size are mixed, by a hoisting tool;
a camera for capturing a set of iron scraps lifted by the sling to generate a camera image;
a preprocessing unit that divides the camera image into a plurality of image pieces;
an estimating unit for estimating the grade for each of the plurality of image pieces using a trained model generated in advance by machine learning with the image of iron scrap as input data and the grade as teacher data;
a ratio calculation unit that calculates, based on the grades estimated for each of the plurality of image pieces, the ratio of the grades included in the collection of the iron scraps lifted by the lifting tool;
a scrap grading system. Further comprising a detection unit that detects the hanging tool in the camera image,
The preprocessing unit divides a predetermined range defined based on the hanging tool into the plurality of image pieces.
The scrap grading system of claim 1. The preprocessing unit selects an image piece to be input to the trained model from among the plurality of divided image pieces.
3. The scrap grading system of claim 2. The preprocessing unit selects an image piece located at a predetermined position with respect to the hanging tool from among the plurality of divided image pieces as an image piece to be input to the trained model.
4. The scrap grading system of claim 3. The preprocessing unit selects, from among the plurality of divided image pieces, image pieces in a range that gradually narrows downward from the hanger as image pieces to be input to the trained model.
The scrap grade determination system according to claim 3 or 4. The preprocessing unit changes the size of each of the plurality of image pieces to a size that can be input to the trained model.
A scrap grade determination system according to any one of claims 1 to 5. The ratio calculation unit calculates the number of grades included in the set of iron scraps lifted by the lifting device based on the grade estimated for each of the plurality of image pieces and the weight index determined for each grade. calculate the weight percentage,
A scrap grade determination system according to any one of claims 1 to 6. The ratio calculation unit is configured to calculate, based on the grade estimated for each of the plurality of image pieces, the grade of the iron scrap, and the weight index determined for each grade and grade, the scrap lifted by the lifting tool. Calculating the weight percentage of the grades in the scrap iron collection,
A scrap grade determination system according to any one of claims 1 to 7. The preprocessing unit compares the camera image with a background image taken when there is no collection of the sling and the iron scrap, and uses image pieces with differences as image pieces to be input to the learned model. sort out,
A scrap grade determination system according to any one of claims 1 to 8. A set of iron scraps mixed with multiple grades of iron scraps according to size is lifted by a crane hoist,
A camera image is generated by photographing the set of iron scraps lifted by the sling with a camera,
dividing the camera image into a plurality of image pieces;
Estimate the grade for each of the plurality of image pieces using a trained model generated in advance by machine learning with the image of iron scrap as input data and the grade as teacher data,
Based on the grades estimated for each of the plurality of image pieces, calculating the percentage of grades included in the set of iron scraps lifted by the lifting tool;
Scrap grade determination method. an image acquisition unit that acquires a camera image generated by a camera that captures a set of iron scraps that are mixed with multiple grades of iron scraps according to size, lifted by a crane sling;
a preprocessing unit that divides the camera image into a plurality of image pieces;
an estimating unit for estimating the grade for each of the plurality of image pieces using a trained model generated in advance by machine learning with the image of iron scrap as input data and the grade as teacher data;
a ratio calculation unit that calculates, based on the grades estimated for each of the plurality of image pieces, the ratio of the grades included in the collection of the iron scraps lifted by the lifting tool;
An estimator, comprising: Acquiring a camera image generated by a camera photographing a collection of steel scrap mixed with multiple grades of steel scrap according to size, lifted by a crane hoist;
dividing the camera image into a plurality of image pieces;
estimating the grade for each of the plurality of image pieces using a trained model generated in advance by machine learning with the image of iron scrap as input data and the grade as teacher data;
calculating, based on the grades estimated for each of the plurality of image pieces, the percentage of grades included in the set of iron scraps lifted by the hoisting tool;
A program that makes a computer run an image acquisition unit that acquires learning images generated by a camera that captures a set of iron scraps in which multiple grades of iron scraps according to size are mixed;
a preprocessing unit that divides the learning image into a plurality of image pieces;
a grade obtaining unit that obtains the grade of the iron scrap appearing in each of the plurality of image pieces;
a learning unit that uses machine learning to generate a trained model for estimating the grade of iron scrap appearing in the image, using the image piece as input data and the grade as teacher data;
A learning device comprising: The preprocessing unit identifies and displays representative iron scraps and other iron scraps in each of the plurality of image pieces,
14. A learning device according to claim 13. Acquire learning images generated by a camera that captures a collection of iron scraps in which multiple grades of iron scraps according to size are mixed,
dividing the training image into a plurality of image pieces;
Obtaining grades of iron scrap appearing in each of the plurality of image pieces;
Using the image piece as input data and the grade as training data, generating a trained model for estimating the grade of iron scrap appearing in the image by machine learning.
How to generate the trained model. Acquiring learning images generated by a camera that captures a set of iron scraps in which multiple grades of iron scraps according to size are mixed,
dividing the training image into a plurality of image pieces;
obtaining a scrap iron grade that appeared in each of the plurality of image pieces; and
Using the image piece as input data and the grade as teacher data, generating a trained model for estimating the grade of iron scrap appearing in the image by machine learning;
A program that makes a computer run

Images