JP7724042B2 - Information processing device, information processing method, and information processing program - Google Patents

Description

Article 30, paragraph 2 of the Patent Act applies. Date of submission: January 25, 2021. Location of submission: 7-3-1 Hongo, Bunkyo-ku, Tokyo. Submitted to the Hoshino-Daigo Laboratory, Materials Engineering, Graduate School of Engineering, The University of Tokyo.

Article 30, Paragraph 2 of the Patent Act applies. Date of publication on the website: February 1, 2021. Website address: https://zoom.us/j/95081284638

Article 30, Paragraph 2 of the Patent Act applies. Dates: March 17, 2021 - March 21, 2021 Website address: https://online.sxsw.com/event/sxsw-online/exhibitor/RXhoaWJpdG9yXzMwMTA2Mg==

Article 30, Paragraph 2 of the Patent Act applies. Date held: September 3, 2021. Website address: https://soubunprintingcoltd.my.webex.com/soubunprintingcoltd.my/j.php?MTID=m7efc538f0fea3c8e2ea5c0ae218f2f85

Article 30, Paragraph 2 of the Patent Act applies. Dates: July 28, 2021 - September 10, 2021. Website address: https://a00.hm-f.jp/cc.php?t=M1199228&c=135&d=c3b0.

Article 30, paragraph 2 of the Patent Act applies. Date: July 11, 2022 Meeting name and location: July 2022 Toyotsukakai "Lecture (Regular Meeting)" Published online using Zoom hosted by Toyotsuka Material Co., Ltd.

Article 30, paragraph 2 of the Patent Act applies. Date: September 13, 2022 Name of the meeting, location: Steel scrap online seminar "Carbon neutrality will change steel scrap!" Published online using Teams, hosted by Steel Recycling Research Co., Ltd.

The present invention relates to an information processing device, an information processing method, and an information processing program.

When producing steel from iron ore, huge amounts of carbon dioxide are emitted during the steelmaking process using iron ore and coal in a blast furnace. With the current demand for reducing carbon dioxide emissions, recycling scrap steel is becoming more important.

Here, the scrap iron collected by the collector is piled up at a collection site and then transported to an electric furnace manufacturer, etc., where the price is calculated based on the measured weight, grade of the scrap iron, etc.

Conventionally, the grade ratios of iron scrap brought into electric furnace manufacturers' scrap receiving sites were assessed visually by inspectors, which was extremely inefficient. However, a grading system has been proposed in which, each time a lifting magnet is used to lift some of the iron scrap piled up on the loading platform of a scrap-loading vehicle, the iron scrap remaining on the loading platform and the iron scrap lifted by the lifting magnet are photographed; each time the lifting magnet lifts iron scrap, a partial ratio determination unit averages the grade ratio estimated from the image generated by the vehicle's camera just before lifting with the grade ratio estimated from the image generated by the lifting magnet camera at the time of lifting to determine the grade ratio of the iron scrap lifted; and each time the lifting magnet lifts iron scrap, a total ratio determination unit tallying the grade ratios of the iron scrap determined by the partial ratio determination unit and calculating the average value. This determines the grade ratio of all iron scrap unloaded from scrap-loading vehicles to the scrap yard. The partial ratio determination unit estimates the grade ratio for each image showing the iron scrap lifted by the lifting magnet from different viewpoints and determines the grade ratio for each lifted iron scrap as the average of these, and the overall ratio determination unit further averages the grade ratios for each lifted iron scrap to determine the grade ratio for the entire iron scrap (see Patent Document 1).

Japanese Patent Application Laid-Open No. 2020-95709

However, there is room for improvement in the accuracy of identifying iron scrap with conventional technology.

The present invention was made in light of these points, and its objective is to provide an information processing device, information processing method, or information processing program that can more accurately determine the quality of a group of iron scrap containing one or more grades of iron scrap.

To solve these problems, a first aspect of the present invention is a method for determining the grade proportions of each grade contained in a group of iron scrap to be evaluated, comprising the steps of: receiving a set of iron scrap group data obtained by photographing the group of iron scrap to be evaluated as it is being unloaded from a loading bay; and using a learning model created by learning using a set of training image data obtained by photographing a group of iron scrap containing one or more grades as it is being unloaded from a loading bay, and labels including grade proportions of one or more grades associated with the entire group of iron scrap that is the subject of the set of training data, and determining the grade proportions of each grade contained in the entire group of iron scrap to be evaluated by using the set of iron scrap group data as input to the learning model.

A second aspect of the present invention is the method of the first aspect, wherein the determining step includes a step of obtaining, as intermediate features, correlation values between features extracted from multiple image data included in the input set of iron scrap group data.

A third aspect of the present invention is the method of the first aspect, wherein the determining step includes a step of determining a grade previously associated with a specified steel material when the group of iron scrap to be determined includes the specified steel material.

A fourth aspect of the present invention is a method according to the first or second aspect, further comprising a step of determining the types of steel scrap contained in the steel scrap group to be determined based on the set of steel scrap group data.

A fifth aspect of the present invention is a method according to the first or second aspect, wherein the learning model is created using additional learning sound data obtained by capturing sounds made when each grade of iron scrap is dropped, and the judgment further uses, as input to the learning model, sound data obtained by capturing sounds made when the group of iron scrap to be judged is dropped.

A sixth aspect of the present invention is a method according to the first or second aspect, further comprising the step of detecting foreign matter contained in the group of iron scrap to be evaluated.

A seventh aspect of the present invention is the method of the sixth aspect, wherein the learning model is created using composite data obtained by combining an image of a foreign object with an image based on at least one of the set of learning image data.

An eighth aspect of the present invention is a method according to the first or second aspect, further comprising a step of calculating the value of the group of iron scrap to be evaluated in accordance with the proportion of each grade contained in the group of iron scrap to be evaluated.

A ninth aspect of the present invention is a program for causing a computer to execute a method for determining the grade ratio of each grade contained in a group of iron scrap to be evaluated, the method including the steps of: receiving a set of iron scrap group data obtained by photographing the group of iron scrap to be evaluated as it is being unloaded from a loading bay; and using a learning model created by learning using a set of learning image data obtained by photographing a group of iron scrap containing one or more grades as it is being unloaded from a loading bay, and labels including grade ratios of one or more grades associated with the entire group of iron scrap that is the subject of the set of learning data, and determining the grade ratio of each grade contained in the entire group of iron scrap to be evaluated by using the set of iron scrap group data as input to the learning model.

A tenth aspect of the present invention is a device for determining the grade ratio of each grade contained in a group of iron scrap to be evaluated, which receives a set of iron scrap group data in which the group of iron scrap to be evaluated is imaged as it is being unloaded from a loading platform, and uses a learning model created by learning using a set of training image data in which an iron scrap group containing one or more grades of iron scrap is imaged as it is being unloaded from a loading platform, and labels including grade ratios of one or more grades associated with the entire group of iron scrap that is the subject of the set of training data, as input to the learning model, to determine the grade ratio of each grade contained in the entire group of iron scrap to be evaluated.

One aspect of the present invention provides an information processing device, information processing method, or information processing program that can more accurately determine iron scrap batches.

1 is a schematic configuration diagram of an iron scrap yard according to an embodiment of the present invention. 1 is a diagram illustrating an example of a schematic configuration of an information processing system according to an embodiment of the present invention. FIG. 2 is a diagram illustrating an example of a hardware configuration of a server according to the present embodiment. FIG. 2 is a diagram illustrating an example of a functional configuration of a server according to the present embodiment. FIG. 10 is a diagram showing an example of a display screen displayed on a display device of a user terminal according to the embodiment. FIG. 10 is a diagram showing an example of a display screen displayed on a display device of a user terminal according to the embodiment. 10 is a flowchart illustrating an example of processing executed by the server according to the embodiment. 10 is a graph showing a determination result based on a learning model according to the present embodiment. 10 is a graph showing a comparative example of a determination result using a learning model according to the related art.

As shown in the schematic diagram of FIG. 1, an iron scrap yard according to one embodiment of the present invention includes a lift magnet 30 that controls magnetic force to attract and release iron scrap loads 10 containing one or more grades loaded on the loading platform of a truck 40 parked at a predetermined location, and a crane rail 50 that moves a portion of the iron scrap loads 10 attracted to the lift magnet 30 to an iron scrap storage area 20. The iron scrap yard may also include a measuring device (not shown) that measures the weight of the truck 40 carrying the iron scrap loads 10 on its loading platform. The weight measured by this measuring device minus the weight of the truck 40 can be input as the weight of the iron scrap loads 10 to the server 3 shown in FIG. 2.

The information processing system 1 according to this embodiment has one or more imaging devices 2 that capture images of a group of iron scrap 10 loaded on the bed of a truck 40. As shown in FIG. 2, each imaging device 2 is connected to a server or information processing device 3 via a network 5 so as to enable data communication. The server 3 is also connected to a user terminal 4 via the network 5 so as to enable data communication. Note that the number of servers 3, user terminals 4 connected to the server 3 via the network 5, and imaging devices 2 are each arbitrary.

(Imaging device 2)
The imaging device 2 is preferably installed in a position where it can photograph the pile of iron scraps 10 from directly above, images the pile of iron scraps 10, and transmits the photographed pile of iron scraps data (image data) to the server 3. As an example, the imaging device 2 photographs the surface of the pile of iron scraps 10 remaining on the bed of the truck 40 each time the lift magnet 30 lifts up some of the iron scrap from the pile of iron scraps 10 loaded on the bed of the truck 40 parked at a predetermined position. The imaging device 2 repeats this photographing of new surfaces that appear until all the iron scrap is removed from the bed of the truck 40. In this way, a plurality of pile of iron scraps data is obtained, photographed as the lift magnet 30 unloads the pile of iron scraps 10 from the bed.

Furthermore, the imaging device 2 is preferably installed in a position that overlooks directly below the pile of iron scrap 10 loaded on the bed of a truck 40 parked at a predetermined location. However, if certain objects in the iron scrap yard, such as a lift magnet 30 or crane rail 50, are within the imaging range during imaging, the imaging device 2 may be installed in a position that keeps the lift magnet 30 or crane rail 50 out of the imaging range or its center.

(Server 3)
Fig. 3 is a diagram showing an example of the hardware configuration and functional configuration of the server 3. As shown in Fig. 3, the server 3 includes a communication IF 300A, a storage device 300B, a CPU 300C, etc. The server 3 may also include an input device (e.g., a keyboard, a touch panel, etc.) and a display device (e.g., a liquid crystal monitor, an organic EL monitor, etc.).

Communication IF300A is an interface for communicating with an external terminal (such as user terminal 4 in this embodiment).

Storage device 300B is, for example, a HDD or semiconductor storage device. Storage device 300B stores information processing programs and various databases used by server 3. For example, storage device 300B stores table data for calculating the value of iron scrap groups 10. For example, storage device 300B stores unit prices (prices per specified weight) according to the grade of iron scrap for each type. In this embodiment, the information processing programs and various databases are stored in storage device 300B of server 3, but they may also be stored in an external storage device such as a USB memory or an external server connected via a network, and configured to be referenced or downloaded as needed.

The CPU 300C controls the server 3 and includes, for example, a ROM (Read Only Memory) and a RAM (Random Access Memory), not shown.

As shown in FIG. 4, server 3 has functions such as a receiving unit 301, a transmitting unit 302, a storage device control unit 303, a variety determination unit 304, a grade determination unit 305, a foreign object detection unit 306, and a value calculation unit 307. Note that each process or operation shown in FIG. 4 is realized by CPU 300C executing instructions contained in an information processing program stored in storage device 300B or the like of server 3. The program may include one or more programs, and can be recorded on a computer-readable storage medium to form a non-transitory program product.

The receiving unit 301 receives, for example, imaging data (image data) of the iron scrap group 10 transmitted from the imaging device 2. The receiving unit 301 also receives, for example, inspection information (see the left side of Figure 5) from a supplier or the like that delivers the iron scrap group 10. The receiving unit 301 also receives, for example, information on the weight of the iron scrap group 10.

The transmission unit 302 transmits to the user terminal 4 information such as the results of the server 3's determination of the type (e.g., dry, heavy, shredded, new scrap, old pig iron, etc.) and grade of the iron scrap group 10, the results of detection of foreign matter contained in the iron scrap group 10, and the calculated value of the iron scrap group 10. The transmission unit 302 can also transmit information that displays the detected foreign matter in a recognizable manner.

The storage device control unit 303 controls the storage device 300B. For example, the storage device control unit 303 writes and reads information to and from the storage device 300B.

The type determination unit 304 can determine the type of iron scrap contained in the iron scrap group 10 to be determined by using, for example, a learning model created by machine learning using learning image data of an image of an iron scrap group and labels including the type associated with the learning image data, and inputting the iron scrap group data received by the receiving unit 301 into the learning model. The "label" is the type of iron scrap, such as hopper, heavy, shredded, new scrap, or old iron. The "label" may be assigned a type determined by, for example, an inspector. Note that the type may not be determined by the type determination unit 304, but may instead be input by a user, such as a worker, from the user terminal 4.

The grade determination unit 305 may use a learning model corresponding to the variety determined by the variety determination unit 304 or the variety input (for example, by a user) from the user terminal 4, or may use a single learning model rather than preparing a learning model for each variety.

The grade determination unit 305 may use, for example, a learning model created by machine learning using a set of training image data obtained by capturing images of a group of iron scrap containing one or more grades of iron scrap as it is being unloaded from a loading platform, and labels corresponding to the set of training image data, including grade proportions of one or more grades associated with the entire group of iron scrap. The learning model can also determine the proportion of each grade contained in the entire group of iron scrap 10, using a set of iron scrap group data received by the receiving unit 301, captured by capturing images of the group of iron scrap 10 as it is being unloaded from a loading platform. By using the grade proportions associated with the entire group of iron scrap that is the subject of the set of training image data as labels for the set of training image data captured by capturing images of the group of iron scrap as it is being unloaded from a loading platform, label annotation for each set of training image data is unnecessary. In this case, it is not necessary to use all of the set of learning image data when learning for type determination, foreign object detection, or other processing or operations, and it is not necessary to use all of the set of iron scrap group data when learning for type determination, foreign object detection, or other processing or operations.

As in the prior art, determining the grade ratio of the iron scrap 10 identified by the learning image data for each piece of learning image data captured during the process of the lift magnet 30 unloading the cargo from the bed of the truck 40, and using the determination results as annotations for each piece of learning image data to determine the grade ratio of the entire iron scrap 10, would require the use of more annotations and at first glance appear appropriate for using machine learning. However, this is work that does not exist in the visual appraisal of each vehicle that has traditionally been performed by inspectors, and therefore creating the many sets of learning image data required for machine learning is a significant burden, and is not a method that allows for practical improvement of accuracy through the accumulation of data. Furthermore, the inventors noticed that there was a discrepancy between the accuracy of trainees' assessments of the proportions of each grade contained in a collection of iron scrap and that in the new task of assessing each training image data, high accuracy could not be expected, even from skilled workers. Therefore, the reliability of annotations for each training image data was low, and therefore the reliability of the learning model obtained using such annotations was also low. Based on this, they developed the present invention, which does not necessarily require label annotation for each training image data. As described below, the present invention achieves higher judgment accuracy than conventional technology and is expected to achieve further improvement. It objectively verifies or replaces the subjective assessment work performed by inspection personnel, contributing to the ever-increasing demand for iron scrap recycling.

For example, it is preferable that the number of sets of learning image data be between 3 and 15. If there are fewer than 3, there will be a lack of information necessary for grading, increasing the risk of a decrease in the inference accuracy of the learning model. If there are more than 15, the analysis time may increase, but the analysis itself can still be performed.

The "label" includes, for example, the grade percentage, which indicates the percentage of each grade of heavy scrap contained in the iron scrap group 10. The grades are indicated, for example, by HS, H1, H2, H3, and H4. The "label" may be assigned a value determined by an inspector, for example, or an actually measured value. The percentage may be, for example, a weight percentage, but is not limited to this and may also be an area percentage or a volume percentage.

The grade determination unit 305 may use a learning model created by learning using data that associates a specific steel material with the grade of the specific steel material as training data to determine the grade of the specific steel material contained in the iron scrap mass 10 based on the image data. In this case, the specific steel material is detected using the learning model that has been trained on the specific steel material, and the grade is determined by referring to separate information that pre-links the steel material to its grade. In scrap yards, specific steel materials may have a set grade assigned to them, which is one of the important criteria for determining the grade. Therefore, not only learning the grade but also learning information about the specific steel material enables more accurate determination.

The foreign object detection unit 306 can detect foreign objects contained in the iron scrap group 10 using, for example, learning image data of images of iron scrap groups containing foreign objects (e.g., wire harnesses and motors containing large amounts of copper, tin, lead, zinc, etc.) and a learning model created by machine learning using labels indicating the type of foreign object (e.g., motor) associated with the learning image data. The foreign object detection unit 306 can then use the iron scrap group data received by the receiving unit 301 as input to the learning model. Here, the learning image data can include image data of images of iron scrap groups containing foreign objects, as well as synthetic data obtained by combining images of foreign objects with images of the iron scrap group. While it is typically difficult to obtain images of various patterns containing foreign objects, it is possible to generate synthetic learning data by combining images of various patterns containing foreign objects, thereby improving the efficiency of the learning model.

The value calculation unit 307 calculates, for example, the value of the group of iron scrap 10. Here, the value calculation unit 307 calculates the value of the group of iron scrap 10 based on the determination results of the type determination unit 304 and the grade determination unit 305. Here, the value calculation unit 307 calculates the value of the group of iron scrap 10 based on the proportion or weight of materials other than iron scrap (e.g., foreign matter, dust, moisture, etc.) contained in the group of iron scrap 10. Specifically, the value calculation unit 307 subtracts the weight of materials other than iron scrap (foreign matter, dust, moisture, etc.) contained in the group of iron scrap 10 from the weight of the group of iron scrap 10 ("weight subtraction"), and then multiplies the unit price by the weight of the type determined by the type determination unit 304 (the type may be input by the user) and the weight of each grade determined by the grade determination unit 305. The value calculation unit 307 may automatically determine the proportion or weight of items other than iron scrap contained in the iron scrap group 10 from the image data received by the receiving unit 301, or may use the proportion or weight of items other than iron scrap input by the user via the user terminal 4.

(Display screen example)
5 and 6 are diagrams showing examples of screens displayed on the display device of the user terminal 4 according to this embodiment. Below, the screens displayed on the display device of the user terminal 4 will be described with reference to Fig. 5 and 6. In the following description, the same components will be assigned the same reference numerals, and duplicate descriptions will be omitted.

Figure 5 is a diagram showing an example of an inspection information input screen 51. As shown in Figure 5, inspection information, such as the supplier's name, trading company name, product type, and unloading location, can be entered in input boxes 52 to 55 on the left side of the screen. Also, as shown in Figure 5, an image 56 captured by the imaging device 2 (a top-down image of the iron scrap load 10 loaded on the bed of the truck 40) is displayed on the right side of the screen. Furthermore, when the user operates the input device of the user terminal 4 to select the inspection start button 57, analysis of the iron scrap load (e.g., determining the grade, calculating the value, etc.) begins.

Figure 6 is a diagram showing an example of an inspection result display screen. As shown in Figure 6, the left side of the inspection result screen 71 displays information such as the inspection start time, inspection end time, inspection duration, supplier, responsible trading company, responsible inspector, unloading location, weight of the iron scrap group 10 (before weight reduction), and weight of the iron scrap group 10 (after weight reduction). As shown in Figure 6, the right side of the inspection result screen displays information such as the type, grade, percentage of each grade, unit price, price, discount reason, and purchase price (value) of the inspected iron scrap group 10.

(Value calculation process)
7 is a flowchart showing an example of the value calculation process executed by the server 3. The value calculation process executed by the server 3 will be described below with reference to FIG.

(Step S701)
The receiving unit 301 of the server 3 receives the proportion or weight of materials other than iron scrap (for example, dust, moisture, etc.) contained in the iron scrap group 10. The proportion or weight of dust, moisture, etc. contained in the iron scrap group 10 may be automatically determined from the imaging data received by the receiving unit 301.

(Step S702)
The value calculation unit 307 calculates the value of the group of iron scraps 10. Specifically, the value calculation unit 307 subtracts the weight of the non-iron scrap (foreign matter, dust, moisture, etc.) contained in the group of iron scraps 10 from the weight of the group of iron scraps 10 ("weight subtraction"). Next, the value calculation unit 307 references price data for the corresponding product type in the value calculation table data stored in the storage device 300B. Next, the value calculation unit 307 calculates the weight of each grade contained in the group of iron scraps 10 based on the weight of the group of iron scraps 10 after weight subtraction and the proportion of each grade contained in the group of iron scraps 10 determined by the grade determination unit 305. Next, the value calculation unit 307 references the price data and calculates a value (total value) by adding values obtained by multiplying the weight by the unit price for each grade.

(Step S703)
The transmitting unit 302 transmits the calculated value of the group of iron scraps 10 to the user terminal 4. As a result, the value of the group of iron scraps is displayed on the display device of the user terminal 4, as illustrated in FIG.

In a comparative experiment , 530 sets of results were prepared, each of which involved an inspector assessing the proportion of each grade of scrap iron contained in the total amount of scrap iron loaded on the truck bed. Also, 100 sets of results were prepared, each of which involved an inspector assessing the surface of the scrap iron pile on five images taken by the inspector as the scrap iron pile was unloaded from the truck bed. Six grades, HS, H1, H2, H3, L1, and L2, were selected as candidates for assessment. The former data was used to generate a learning model according to the present embodiment (hereinafter also referred to as a "multiple-input model"), and the latter data was used to generate a learning model according to the prior art (hereinafter also referred to as a "single-input model"). More specifically, in this embodiment, a model predicting the grade of a single truck was constructed using ResNet50, a convolutional neural network, as a backbone, and learning using correlation values between features extracted from multiple image data as intermediate features.In the prior art, a model predicting the grade from a single image data was constructed using ResNet50 as a backbone as well, and the grade of a single truck was predicted by averaging predictions for multiple image data.In both methods, data was divided using k-fold cross-validation (k=5).

The evaluation method was to use the inspector's assessment results for each truck as the answer (grand truth (GT)), and if the maximum error for each grade was within 10 points, the answer was evaluated as correct (true). Table 1 shows an example of an evaluation.

Figure 8 shows the results of the multiple-input model, and Figure 9 shows the results of the single-input model. The former had an accuracy rate of 57.7%, while the latter had an accuracy rate of 45.0%, with the former being approximately 12.7% more accurate.

In addition to the above embodiment, the type or grade of the iron scrap group 10 may be determined using a learning model created by further learning using sound data generated when iron scrap of each type or grade is dropped. When using sound data generated when iron scrap is dropped, it is preferable to create learning sound data for each iron scrap yard.

In the case of a learning model created by further learning using sound data generated when the iron scrap is dropped, a microphone is provided to pick up the sound generated when the iron scrap pile 10 is dropped into the iron scrap yard, and the server 3 receives sound data obtained by capturing the sound generated when the iron scrap pile 10 is dropped in addition to the imaging data captured by the imaging device 2.

In addition, in the above embodiments, unless the word "only" is used, such as "based only on XX," "depending only on XX," or "in the case of XX only," please note that this specification assumes that additional information may also be taken into consideration. Also, as an example, please note that the phrase "do b when a" does not necessarily mean "do b always when a" or "do b immediately after a" unless explicitly stated otherwise. Furthermore, the phrase "each a constituting A" does not necessarily mean that A is composed of multiple components, but includes the case where the component is singular.

Furthermore, just to be clear, even if there is an aspect of a method, program, terminal, device, server, or system (hereinafter referred to as "method, etc.") that performs an operation different from that described in this specification, each aspect of the present invention is directed to an operation identical to any of the operations described in this specification, and the existence of an operation different from that described in this specification does not mean that the method, etc. falls outside the scope of each aspect of the present invention.

The flowchart shown in Figure 7 is merely an example and does not necessarily mean that the process will necessarily start or end in the illustrated sequence.

In addition, the above-described embodiments and modifications are merely examples of specific ways of implementing the present invention, and should not be interpreted as limiting the technical scope of the present invention. In other words, the present invention can be implemented in various forms without departing from its gist or main features.

REFERENCE SIGNS LIST 1 Information processing system 2 Imaging device 3 Server 4 User terminal 5 Network 10 Iron scrap pile 20 Iron scrap storage area 30 Lift magnet 40 Truck 50 Crane rail 51 Information input screen 52, 53, 54, 55 Input box 56 Captured image 57 Start button 71 Analysis result screen 300A Communication IF
300B Storage device 300C CPU
301 Receiving unit 302 Transmitting unit 303 Storage device control unit 304 Type determination unit 305 Grade determination unit 306 Foreign object detection unit 307 Value calculation unit

Claims (9)

A method for determining the grade ratio of each grade contained in a group of iron scrap to be determined, comprising:
receiving a set of iron scrap group data obtained by taking a plurality of images of the iron scrap group to be determined along the progress of unloading from a loading platform;
The method includes a step of using a learning model created by learning using a set of learning image data obtained by taking multiple images of a group of iron scrap containing one or more grades of iron scrap as it is being unloaded from a loading bay, and labels including grade proportions of one or more grades associated with the entire group of iron scrap that is the subject of the set of learning data, and using the set of iron scrap group data as an input to the learning model to determine the grade proportions of each grade contained in the entire group of iron scrap that is the subject of evaluation. A method according to claim 1, wherein the determining step includes a step of obtaining, as intermediate features, correlation values between features extracted from multiple image data included in the input set of iron scrap group data. 10. The method of claim 1,
The method further includes a step of determining the type of the iron scrap group to be determined based on the set of iron scrap group data,
The learning model is a model corresponding to the determined variety. 10. The method of claim 1,
The method further includes a step of detecting foreign matter contained in the group of iron scraps to be judged. 5. The method of claim 4,
The learning model is created using composite data obtained by combining an image of a foreign object with an image based on at least one of the set of learning image data. 5. A method according to any one of claims 1 to 4, comprising:
The method further includes a step of calculating a value of the group of iron scraps to be evaluated according to the proportion or weight of each grade of iron scrap contained in the group of iron scraps to be evaluated and the proportion or weight of materials other than the iron scrap contained in the group of iron scraps to be evaluated. 5. A method according to any one of claims 1 to 4, comprising:
The number of sets of learning image data is three or more. A program for causing a computer to execute a method for determining the grade ratio of each grade contained in a group of iron scrap to be determined, the method comprising:
receiving a set of iron scrap group data obtained by taking a plurality of images of the iron scrap group to be determined along the progress of unloading from a loading platform;
The method includes a step of using a learning model created by learning using a set of learning image data obtained by taking multiple images of a group of iron scrap containing one or more grades of iron scrap as it is being unloaded from a loading bay, and labels including grade proportions of one or more grades associated with the entire group of iron scrap that is the subject of the set of learning data, and using the set of iron scrap group data as an input to the learning model to determine the grade proportions of each grade contained in the entire group of iron scrap that is the subject of evaluation. An apparatus for determining the grade ratio of each grade contained in a group of iron scrap to be evaluated,
a receiving unit for receiving a set of data on the group of iron scraps, the data being obtained by taking a plurality of images of the group of iron scraps to be judged along the progress of unloading from a loading platform;
a determination unit that uses a learning model created by learning using a set of learning image data obtained by capturing multiple images of a group of iron scraps including one or more grades of iron scrap as the group is being unloaded from a loading platform and labels including grade ratios of one or more grades associated with the entire group of iron scraps that is the subject of the set of learning data, and that determines the grade ratios of each grade included in the entire group of iron scraps that is the subject of the determination unit by using the set of iron scrap group data as an input to the learning model;
Equipped with .