JP7410375B2 - Label identification device, label identification method, learning device, identification device and computer program - Google Patents

Description

The present invention relates to a label identification device, a label identification method, a learning device, an identification device, and a computer program.

Generally, steel manufacturers that manufacture steel products display information for managing the identification information of steel products (hereinafter referred to as "management information") on the steel products themselves for the purpose of managing the steel products. There is. For example, management information may be printed or engraved on the surface of a steel product , or a display medium such as a sticker or indicator on which management information is written may be affixed to the surface of packaging paper.

An example of such a steel product is a coil material manufactured by a steel manufacturer. A coil material is a type of steel product manufactured by rolling a steel material formed into a plate shape by rolling into a coil shape. Since the coil material is large in size, it is preferable that the management information displayed on the coil material be displayed in a large and easy-to-understand manner so that a person can easily identify it visually. Therefore, one method of displaying management information on such a coil material is to display one piece of management information by a combination of one or more displays shown on the surface of the coil material.

Japanese Patent Application Publication No. 2018-185552 Japanese Patent Application Publication No. 2018-200531

However, in the past, steel manufacturers had to visually determine whether the management information was correctly displayed on the coil material, which could lead to errors in their judgment.

In view of the above circumstances, the present invention provides a technology that allows management information represented by a combination of one or more displays shown on the surface of a coil material to be identified based on an image taken of the coil material. It is intended to.

One aspect of the present invention is one or more displays that can be displayed at different display positions to be managed, and includes learning image data in which label elements representing one label are captured by a combination thereof, and the learning image data. a learning unit that uses a machine learning method to learn the relationship between the learning image data and the label element based on teacher data indicating the label element shown in the management target captured in the image data of and the management target imaged in the image data for identification based on the image data for identification in which the label element to be identified is imaged, and the learning data indicating the learning result of the relationship by the learning unit. a first identification unit that identifies the displayed label element; and a second identification unit that identifies the label displayed on the management target based on the identification result of one or more label elements shown on the management target. This is a label identification device comprising: an identification section;

One aspect of the present invention is the above-mentioned label identification device, wherein the learning unit learns the relationship between the image data captured at each display position to be managed and the label element displayed at the display position. The first identification unit generates learning data for each display position, and the first identification unit generates learning data for each display position, and the first identification unit performs the management based on the learning data for each display position and identification image data in which each display position to be managed is captured. Identify the label element shown in each display position of the target.

One aspect of the present invention is the above-mentioned label identification device, wherein the learning unit deletes part or all of an area other than the area where the display position was imaged from the image data acquired by the imaging unit. The image data is used as the learning image data.

One aspect of the present invention is the above-mentioned label identification device, wherein the object to be managed is a coil material manufactured by winding a steel material formed into a plate shape by rolling into a coil shape, and the display position is: It is an end surface, an inner diameter surface, or an outer peripheral surface of the coil material.

One aspect of the present invention is the above label identification device, in which the label elements are identified by a combination of their shapes and colors.

One aspect of the present invention includes an imaging step of acquiring image data in which label elements representing one label by a combination of one or more displays that can be displayed at different display positions to be managed; The relationship between the image data and the label element is learned by a machine learning method based on training data indicating the image data for the image data and the label element shown on the management target captured in the image data. identifying the label element shown in the management target imaged in the image data based on a learning step for identifying, image data for identification, and learning data indicating a learning result of the relationship in the learning step; and a second identification step of identifying a label displayed on the management target based on the identification result of one or more label elements shown on the management target. It's a method.

One aspect of the present invention is one or more displays that can be displayed at different display positions to be managed, including image data in which a label element representing one label is captured in combination, and image data in which a label element is captured in the image data. The learning device uses a machine learning method to learn the relationship between the image data and the label element based on teacher data indicating the label element shown in the managed object.

One aspect of the present invention is one or more displays that can be displayed at different display positions to be managed, including image data in which a label element representing one label is captured in combination, and image data in which a label element is captured in the image data. learning the relationship learned by a learning device that learns the relationship between the image data and the label element using a machine learning method based on teacher data indicating the label element shown in the managed target; Identifying the label element shown in the management target imaged in the identification image data based on learning data indicating the result and identification image data in which the identification target label element is imaged, The present invention is an identification device that identifies a label displayed on the management target based on an identification result of one or more label elements shown on the management target.

One aspect of the present invention is one or more displays that can be displayed at different display positions to be managed, including image data in which a label element representing one label is captured in combination, and image data in which a label element is captured in the image data. A computer program for causing a computer to execute a step of learning a relationship between the image data and the label element using a machine learning method based on training data indicating the label element shown in a management target. It is.

One aspect of the present invention is one or more displays that can be displayed at different display positions to be managed, including image data in which a label element representing one label is captured in combination, and image data in which a label element is captured in the image data. learning the relationship learned by a learning device that learns the relationship between the image data and the label element using a machine learning method based on teacher data indicating the label element shown in the managed target; a step of identifying the label element shown in the management target imaged in the identification image data based on learning data indicating the result and identification image data in which the identification target label element is imaged; and identifying a label displayed on the management target based on an identification result of one or more label elements shown on the management target.

According to the present invention, management information represented by a combination of one or more displays shown on the surface of a managed object can be identified with higher accuracy based on an image of the managed object.

FIG. 2 is a system configuration diagram showing a specific example of the system configuration of the management information identification system according to the present embodiment. FIG. 2 is a functional block diagram showing a specific example of the functional configuration of the learning device according to the present embodiment. 2 is a flowchart showing the flow of processing in which the learning device of this embodiment generates learning data based on learning image data. It is a figure showing an example of the appearance of coil material. It is a figure showing an example of the appearance of coil material. FIG. 3 is a diagram illustrating a specific example of a learning data generation method according to the present embodiment. FIG. 3 is a diagram showing a specific example of a label element that can be displayed on a coil material. It is a figure which shows the specific example of the identification information of a label element. FIG. 2 is a functional block diagram showing a specific example of the functional configuration of the identification device according to the present embodiment. It is a flowchart which shows the specific example of the flow of the process by which the identification device of this embodiment identifies the management information displayed on the coil material. FIG. 7 is a diagram showing the correspondence between label element identification results and labels on each display screen.

Hereinafter, a label identification device, a label identification method, a learning device, an identification device, and a computer program according to embodiments will be described with reference to the drawings.

FIG. 1 is a system configuration diagram showing a specific example of the system configuration of a management information identification system 100 in this embodiment. The management information identification system 100 includes an imaging device 2, a learning device 3, and an identification device 4. The imaging device 2 is a device that images the coil material to be identified. A coil material is a type of steel product manufactured by rolling a steel material formed into a plate shape by rolling into a coil shape. The imaging device 2 is communicably connected to the learning device 3 and the identification device 4, and transmits image data obtained by imaging the coil material to the learning device 3 and the identification device 4. Specifically, the imaging device 2 transmits image data captured for learning management information (hereinafter referred to as "learning image data") to the learning device 3, and transmits image data (hereinafter referred to as "learning image data") captured for the purpose of learning management information to the learning device 3. (hereinafter referred to as "identification image data") is transmitted to the identification device 4. Note that, hereinafter, the image represented by the learning image data will be referred to as a "learning image", and the image represented by the identification image data will be referred to as an "identification image".

The learning device 3 is a device that uses a machine learning method to learn the relationship between learning image data and management information displayed on the coil material captured in the learning image. Specifically, the learning device 3 has a function of generating teacher data for learning the above relationships based on the learning image data, and a function of learning the above relationships using the generated teacher data. The learning device 3 supplies the identification device 4 with data (hereinafter referred to as “learning data”) indicating the learning results of the relationships.

The identification device 4 is a device that identifies management information of the coil material captured in the identification image based on the learning data generated by the learning device 3. Specifically, based on the learning data and the identification image data, the identification device 4 uses identification information (hereinafter referred to as "label") of the management information displayed on the coil material captured in the identification image. identify. The identification device 4 outputs information associated with the obtained label among the management information stored in advance as the management information of the coil material. It is assumed that information indicating the correspondence between management information and labels (hereinafter referred to as "correspondence information") is stored in advance in the identification device 4.

FIG. 2 is a functional block diagram showing a specific example of the functional configuration of the learning device 3 of this embodiment. The learning device 3 includes a CPU (Central Processing Unit), a memory, an auxiliary storage device, etc. connected via a bus, and executes a program. The learning device 3 is configured as a device including an image data storage section 31, a teacher data storage section 32, a learning image data input section 33, a teacher data generation section 34, a learning data generation section 35, and a learning data output section 36 by executing the program. Function. Note that all or part of each function of the learning device 3 may be realized using hardware such as an ASIC (Application Specific Integrated Circuit), a PLD (Programmable Logic Device), or an FPGA (Field Programmable Gate Array). The program may be recorded on a computer-readable recording medium. The computer-readable recording medium is, for example, a portable medium such as a flexible disk, magneto-optical disk, ROM, or CD-ROM, or a storage device such as a hard disk built into a computer system. The program may be transmitted via a telecommunications line.

The image data storage unit 31 stores learning image data. For example, the image data storage section 31 is configured using a storage device such as a magnetic hard disk device or a semiconductor storage device.
The teacher data storage unit 32 stores teacher data. For example, like the image data storage section 31, the teacher data storage section 32 is configured using a storage device such as a magnetic hard disk device or a semiconductor storage device.

The learning image data input unit 33 has a function of inputting learning image data to the learning device 3. For example, the learning image data input unit 33 is configured using a communication interface, and inputs the learning image data by receiving the learning image data transmitted from the imaging device 2 . Note that the learning image data input unit 33 may input the learning image data by a method other than the communication input method. For example, the learning image data input unit 33 may input the learning image data by reading the learning image data from a recording medium that stores the learning image data. The learning image data input unit 33 records the input learning image data in the image data storage unit 31.

The teacher data generation unit 34 has a function of generating teacher data based on the learning image data. For example, the teacher data generation unit 34 is configured using input devices such as a touch panel, a mouse, and a keyboard, and performs editing processing on the learning image data according to an input operation to generate the learning image data and the learning image data. Training data for learning the relationship with the management information displayed on the imaged coil material is generated. The teacher data generation unit 34 records the generated teacher data in the teacher data storage unit 32. Note that, instead of the learning image data input section 33 and the teacher data generation section 34, the learning device 3 may include a functional section that inputs pre-generated teacher data. Furthermore, part or all of the process of generating teacher data may be automated based on predetermined rules.

The learning data generation unit 35 has a function of generating learning data for identifying the label of management information displayed on the coil material by learning the teacher data. Specifically, the learning data generation unit 35 generates parameters (hereinafter referred to as " The discriminator parameters (referred to as "discriminator parameters") are determined based on the training data.

For example, the learning data generation unit 35 can determine classifier parameters using a neural network that inputs each pixel value of the identification image data and outputs information necessary to identify one label. Here, the information necessary to identify one label is the identification information for each constituent element (hereinafter referred to as "label element") that constitutes one label, and the information necessary to identify one label is This is the information needed to integrate and identify the results. Note that the relationships between labels and label elements and the relationships between label elements and learning data will be explained in detail below, so further explanation will be omitted here.

Note that input to the neural network is not limited to pixel values of image data, and any value acquired based on image data may be input. For example, instead of each pixel value of the image data, a feature extracted from the image data may be input to the neural network, or the same feature may be input in addition to each pixel value of the image data. . Further, the relationship between image data and label elements does not necessarily need to be described by a neural network. For example, the learning data generation unit 35 may describe the above-mentioned relationship using a method such as deep learning or deep learning based on a neural network, or may describe the above-mentioned relationship using a method such as deep learning or deep learning based on a neural network, or may use a regression method such as linear regression, nonlinear regression, simple regression, or multiple regression. The above relationship may be described using an analytical method. Furthermore, when classifying image data into one of a plurality of corresponding label elements is considered to be learning of the above relationships, the learning data generation unit 35 uses SVM (support vector machine), random forest, k-means method, etc. The relationship between image data and label elements may be described by a classification method determined using a clustering method.

The learning data generation unit 35 generates the values of the discriminator parameters determined in this manner as learning data, and outputs the generated learning data to the learning data output unit 36.

The learning data output unit 36 has a function of outputting learning data to the identification device 4. For example, the learning data output unit 36 is configured using a communication interface, and outputs the learning data by transmitting the learning data generated by the learning data generation unit 35 to the identification device 4. Note that, similarly to the input of the learning image data, the learning data output unit 36 may output the learning data by a method other than the communication method. For example, the learning data output unit 36 may output the learning data by writing the learning data on a recording medium that is removably attached to the own device and the identification device 4. In this case, the recording medium storing the learning data is removed from the learning device 3 and then connected to the identification device 4, and the learning data is input to the identification device 4 by reading the learning data from the recording medium. .

FIG. 3 is a flowchart showing the flow of processing in which the learning device 3 of this embodiment generates learning data based on learning image data. First, the learning image data input unit 33 inputs the learning image data acquired by the imaging device 2 to the learning device 3 (step S101). The learning image data input unit 33 records the input learning image data in the image data storage unit 31 (step S102).

Subsequently, the teacher data generation unit 34 applies identification information of each label element displayed on the coil material captured in each learning image to the learning image data stored in the image data storage unit 31. Teacher data is generated by making the correspondence (step S103). More specifically, the teacher data generation unit 34 determines the presence/absence and type of label elements for each area in the learning image where each label element can be displayed (corresponding to the end face, outer peripheral surface, and inner diameter surface of the coil material described later). Generate training data for identification. Here, a method for generating teacher data will be described in detail, with appropriate reference to various specific examples shown in FIGS. 4 to 8.

First, specific examples of the appearance of the coil material to be identified are shown in FIGS. 4 and 5. Coil materials may be stored in different packaging depending on the shipping destination, use, whether or not they are coated, etc. For example, FIG. 4 shows the appearance of a coil material wrapped in a thin product sheet. The example in FIG. 4 is an example in which one label is represented by a combination of a blue tape B1 affixed to the inner diameter surface of the coil and a blue tape B2 affixed to the end surface of the coil. On the other hand, the example in FIG. 5 is an example in which one label is represented by a red tape R1 attached to the outer peripheral surface of the coil.

Here, as in the example in FIG. 4, if no tape other than the tapes constituting the label (for example, B1 and B2) is attached to the coil material, the coil material is displayed on the coil material for the image data taken. What is necessary is to learn what is associated with the identification information of the label itself as training data. However, as in the example of FIG. 5, if there is a possibility that a tape (for example, R2) other than the tape (for example, R1) that constitutes the label is attached to the coil material, such image data may include the identification information of the label itself. When learning the association of , as training data, tapes that do not constitute a label may be learned as part of the label, so it may not always be possible to identify the label with high accuracy. In addition, coil materials may be marked in various ways other than tape, or objects that have no particular meaning as information may be attached (for example, packing strings or tape), and images of these may be attached. This can be a factor that reduces label identification accuracy.

Therefore, in the management information identification system 100 of this embodiment, the presence or absence and type of each of one or more constituent elements (namely, label elements) constituting one label are identified, and the identification results are integrated to form one label. This identification makes it possible to more accurately identify the management information associated with the label captured in the image. Specifically, the label identification method in this embodiment includes a step in which the learning device 3 generates learning data for identifying individual label elements based on learning image data, and a step in which the identifying device 4 generates learning data for identifying individual label elements. This is realized by a step of identifying each label element based on learning data, and a step of the identification device 4 integrating the identification results of each label element to identify one label.

FIG. 6 is a diagram showing a specific example of the learning data generation method in this embodiment. For example, FIG. 6 shows the presence or absence of a label element for each display surface in a case where one label element can be displayed on each of the end surface, inner diameter surface, and outer circumferential surface (an example of a display position; hereinafter referred to as the "display surface") of the coil material. and a method for generating learning data for identifying the type. In this case, first, the imaging device 2 acquires learning image data in which each display surface is the main subject.

For example, FIG. 6(A) shows a specific example of the learning image data with the end face of the coil material as the main subject, and FIG. 6(B) shows a specific example of the learning image data with the outer peripheral surface of the coil material as the main subject. It shows. Moreover, FIG. 6(C) shows a specific example of the learning image data in which the inner diameter surface of the coil material is the main subject. The learning image data should be larger and more clearly show each display surface that is the main subject, but in cases where it is difficult to obtain image data for each display surface, or the accuracy of identification may be affected. In cases where there is no influence, one piece of learning image data obtained by capturing images of a plurality of display surfaces may be acquired as learning image data of a plurality of display surfaces. For example, there is no big difference in the imaging conditions between the image data of FIG. 6(A) (end surface) and the image data of FIG. 6(B) (outer surface), so in such a case, the learning image data The image data may be aggregated into either one of the image data.

Next, in the learning device 3, the teacher data generation unit 34 edits the learning image data acquired for each display surface, and deletes part or all of the information other than the display surface from the learning image data. This is a process for reducing information that may be a factor in reducing the identification accuracy of label elements, and is performed as pre-processing for generating learning data. Through this editing process, for example, the learning image data G1 of the end face shown in FIG. 6(A) is edited to become the image data G1' within the rectangular area, and the learning image data of the outer peripheral face shown in FIG. 6(B) is edited. G2 is edited so that it becomes image data G2' within a rectangular area, and the learning image data G3 on the inner diameter surface shown in FIG. 6(C) is edited so that it becomes image data G3' within a rectangular area.

Note that this method of extracting an image of a part of a rectangular area is an example of an editing method that deletes information other than the display surface from the learning image data, and the learning image data may not be edited using other methods. Good too. For example, if it is known that each display surface is located within a predetermined range within the image, the learning device 3 may delete information other than the display surface by masking the image outside the range.

Next, the teacher data generation unit 34 generates teacher data for each display surface by associating each edited learning image data with information indicating the presence or absence and type of a label element on the captured display surface. do. For example, in the example of FIG. 6, a label element represented by "one straight line" is displayed on the end surface of FIG. 6(A) and the inner diameter surface of FIG. 6(C), and the label element shown in FIG. 6(B) is displayed. No label element is displayed on the outer circumferential surface. In such a case, the teacher data generation unit 34 associates the learning image data of FIG. 6(A) and the learning image data of FIG. 6(C) with the identification information of the displayed label element, Teacher data is generated by associating information indicating that no label element is displayed with the learning image data of FIG. 6(B).

FIG. 7 is a diagram showing a specific example of a label element that can be displayed on a coil material. For example, FIG. 7(A), FIG. 7(B), and FIG. 7(C) show a label element represented by one straight line as an example of a label element that can be displayed on the end face, inner diameter face, and outer circumferential face. Further, for example, FIG. 7(D) shows a label element represented by two straight lines as an example of a label element that can be displayed on the end surface. Further, for example, FIG. 7E shows a label element represented by a cross as an example of a label element that can be displayed on the inner diameter surface and the outer circumference surface. 7(A) to 7(E) can each represent one different label depending on the combination of one or more label elements displayed on the coil material.

Note that the label element patterns shown in FIG. 7 and the label patterns represented by the combinations thereof are merely examples, and patterns different from these may be used for the label elements and labels. The more patterns there are for label elements, the more label patterns can be represented by their combinations, making it possible to identify more types of management information. It is also conceivable that the identification accuracy of the label element itself will decrease because there is a higher possibility that the label element itself will be required. Therefore, it is preferable that the label elements be set in a pattern as simple as possible depending on the type of management information to be identified. Note that the label elements are not limited to being patterned by shape as in the example of FIG. 7, but may be patterned by display color.

FIG. 8 is a diagram showing a specific example of identification information of a label element (hereinafter referred to as "element ID"). The first column from the left of the table shown in FIG. 8 represents the color of the label element, and the second column represents the shape of the label element. The third column represents the element ID of the label element determined by the corresponding combination of color and shape. In this example, if a label element represented by a single red straight line is displayed on the display screen, the teacher data generation unit 34 generates an element ID for the learning image data indicating the display screen. What is associated with “R1” is generated as teacher data. For example, if a label element represented by a yellow cross is displayed on the display screen, the teacher data generation unit 34 associates the element ID "Y+" with the learning image data showing the display surface. The attached data is generated as training data. On the other hand, if the label element is not displayed on the display surface, the teacher data generation unit 34 associates the element ID "Z" with the learning image data indicating the display surface as the teacher data. generate.

By learning the training data generated in this way using a machine learning method, the training data generation unit 34 can determine the presence or absence of a label element on each display surface, the type of label element displayed on each display surface, It is possible to generate training data that makes it possible to identify.

Returning to the explanation of FIG. 3. The teacher data generating section 34 records the thus generated teacher data for each display screen in the teacher data storage section 32 (step S104). Subsequently, the learning data generation unit 35 learns the teacher data for each display screen stored in the teacher data storage unit 32 to perform learning indicating the relationship between the learning image data and label elements for each display screen. Data is generated (step S105). The learning data generation unit 35 outputs the learning data for each display screen generated in this way to the learning data output unit 36 (step S106). Then, the learning data output unit 36 outputs the learning data generated for each display surface to the identification device 4.

FIG. 9 is a functional block diagram showing a specific example of the functional configuration of the identification device 4 of this embodiment. The identification device 4 includes a CPU, memory, auxiliary storage device, etc. connected via a bus, and executes a program. The identification device 4 stores an image data storage section 41, a learning data storage section 42, a correspondence information storage section 43, an identification image data input section 44, a learning data input section 45, an identification section 46, and a management information output section 47 by executing the program. It functions as a device equipped with Note that all or part of each function of the identification device 4 may be realized using hardware such as an ASIC, a PLD, or an FPGA. The program may be recorded on a computer-readable recording medium. The computer-readable recording medium is, for example, a portable medium such as a flexible disk, magneto-optical disk, ROM, or CD-ROM, or a storage device such as a hard disk built into a computer system. The program may be transmitted via a telecommunications line.

The image data storage unit 41 stores learning image data. For example, the image data storage section 41 is configured using a storage device such as a magnetic hard disk device or a semiconductor storage device.
The learning data storage unit 42 stores learning data. For example, like the image data storage section 41, the learning data storage section 42 is configured using a storage device such as a magnetic hard disk device or a semiconductor storage device.
The correspondence information storage unit 43 stores correspondence information. For example, like the image data storage section 41 and the learning data storage section 42, the correspondence information storage section 43 is configured using a storage device such as a magnetic hard disk device or a semiconductor storage device.

The identification image data input section 44 has a function of inputting identification image data to the identification device 4. For example, the identification image data input unit 44 is configured using a communication interface, and inputs the identification image data by receiving the identification image data transmitted from the imaging device 2 . Note that the identification image data input unit 44 may input the identification image data by a method other than the communication input method. For example, the identification image data input unit 44 may input the identification image data by reading the identification image data from a recording medium that stores the identification image data. The identification image data input section 44 records the input identification image data in the image data storage section 41 .

The learning data input section 45 has a function of inputting learning data to the identification device 4. For example, the learning data input unit 45 is configured using a communication interface, and inputs the learning data by receiving the learning data transmitted from the learning device 3. Note that the learning data input unit 45 may input the learning data by a method other than the communication input method. For example, the learning data input unit 45 may input the learning data by reading the learning data from a recording medium that stores the learning data. The learning data input section 45 records the input learning data in the learning data storage section 42 .

The identification unit 46 has a function of identifying the label of management information displayed on the coil material based on the identification image data obtained by capturing the image of the coil material to be identified and the learning data generated by the learning device 3. have Specifically, the identification unit 46 identifies the label elements displayed on each display surface of the coil material based on the identification image data and the learning data generated for each display surface, and identifies the label elements displayed on each display surface of the coil material. One label is identified based on the identification result of the label element. The identification unit 46 outputs the label thus identified to the management information output unit 47.

The management information output unit 47 acquires management information associated with the identified label, based on the label identified by the identification unit 46 and the correspondence information stored in its own device. The management information output unit 47 outputs the management information acquired in this manner in a predetermined manner. For example, the management information output unit 47 may display the management information on a display device such as a CRT (Cathode Ray Tube) display, a liquid crystal display, or an organic EL (Electro-Luminescence) display, or may display the management information on an audio output device such as a speaker. Sound indicating management information may be output. Further, the management information output unit 47 may transmit the management information to another device that can communicate with the own device.

FIG. 10 is a flowchart showing a specific example of a process in which the identification device 4 of this embodiment identifies management information displayed on a coil material based on identification image data. First, the identification image data input unit 44 inputs the identification image data acquired by the imaging device 2 to the identification device 4 (step S201). The identification image data input unit 44 records the input identification image data in the image data storage unit 41 (step S202).

On the other hand, the learning data input unit 45 inputs the learning data for each display surface output from the learning device 3 to the identification device 4 (step S203). The learning data input unit 45 records the input learning data in the learning data storage unit 42 (step S204).

Subsequently, the identification unit 46 identifies the coil imaged in the identification image based on the identification image data stored in the image data storage unit 41 and the learning data stored in the learning data storage unit 42. The label element displayed on the material is identified (step S205). Specifically, the identification unit 46 configures a discriminator that identifies the presence or absence and type of label elements on each display surface using learning data for each display surface, and inputs identification image data to this discriminator. As the output, identification information (ie, element ID) of the label element displayed on each display surface of the coil material captured in the identification image is obtained. The identification unit 46 outputs the element ID thus obtained for each display screen to the management information output unit 47.

Subsequently, the management information output unit 47 performs management that is associated with the label displayed on the coil material captured in the identification image based on the element ID and correspondence information for each display surface output from the identification unit 46. Information is acquired (step S206). Here, details of the method for identifying management information will be explained using FIG. 11.

FIG. 11 is a diagram showing the correspondence between the identification results of label elements and labels on each display surface. The A-side row represents the identification results of label elements on one end surface of the coil material (hereinafter referred to as "A-side"), and the B-side row represents the identification results on an end surface different from A-side (hereinafter referred to as "B-side"). It represents the identification result of the label element. Further, the row of outer circumferential surfaces represents the identification results of label elements on the outer peripheral surface of the coil material, and the row of inner diameter surfaces represents the identification results of label elements on the inner diameter surface of the coil material. The column of labels represents labels of management information specified by a combination of identification results of label elements on the A side, the B side, the outer peripheral surface, and the inner diameter surface.

For example, in this example, the label element identified by the element ID "R1" (that is, the label element represented by a single red straight line) is displayed on the A side and B side of the coil material, and the outer peripheral surface and inner diameter If the identification unit 46 identifies that there is no label element displayed on the surface, the management information output unit 47 identifies that the label displayed on the coil material is “P01”. In this case, the management information output unit 47 acquires information associated with "P01" in the correspondence information as management information.

For example, a label element identified by the element ID "W+" (that is, a label element represented by a white cross) is displayed on the outer peripheral surface and inner diameter surface of the coil material, and label elements are displayed on the A side and B side. When the identification unit 46 identifies that there is no indication of , the management information output unit 47 identifies that the label displayed on the coil material is “P12”. In this case, the management information output unit 47 acquires information associated with "P12" in the correspondence information as management information.

For example, if the identification unit 46 identifies that there is no label element displayed on any of the A side, B side, outer peripheral surface, and inner diameter surface of the coil material (that is, the label is "PZZ"), the management information The output unit 47 determines that no label is displayed on the coil material based on the correspondence information.

For example, the management information output unit 47 may output any combination of identification results of label elements on the A side, B side, outer peripheral surface, and inner diameter surface of the coil material corresponding to "P01" to "P13" and "PZZ". If it is determined based on the correspondence information that there is no label (that is, the label is "PXX"), the management information output unit 47 determines that the label displayed on the coil material is an unknown label.

The management information identification system 100 of the embodiment configured in this manner includes the learning device 3 and the identification device 4 described above, so that one or more indications shown on the surface of the coil material (an example of a managed object) can be identified. It becomes possible to identify the management information represented by the combination with higher accuracy based on the image taken of the coil material.

Specifically, the learning device 3 of the embodiment is based on training data indicating learning image data and label elements shown on a coil material (an example of a management target) imaged in the learning image data. A learning data generation unit 35 (an example of a learning unit) is provided for learning the relationship between image data and label elements using a machine learning method.

Moreover, the identification device 4 of the embodiment is based on the identification image data and the learning data indicating the learning result of the above-mentioned relationship by the learning data generation unit 35. Management for identifying a label displayed on a coil material based on an identification section 46 (an example of a first identification section) that identifies a label element and an identification result of one or more label elements shown on the coil material. An information output section 47 (an example of a second identification section) is provided.

In addition, although the said embodiment demonstrated the case where the learning device 3 and the identification device 4 were comprised separately, the learning device 3 and the identification device 4 do not necessarily need to be comprised separately. The learning device 3 and the identification device 4 may be integrally configured as a label identification device. Further, the imaging device 2 (an example of an imaging section) may be configured separately from the learning device 3 and the identification device 4, or may be configured integrally with either one of them. Moreover, when the learning device 3 and the identification device 4 are integrally configured as a label identification device, the imaging device 2 may be configured integrally with the label identification device.

Although the embodiments of the present invention have been described above in detail with reference to the drawings, the specific configuration is not limited to these embodiments, and includes designs within the scope of the gist of the present invention.

INDUSTRIAL APPLICATION This invention is applicable to the technique which identifies the management information represented by the combination of one or more displays shown on the surface of a management object based on the image which the management object was imaged.

DESCRIPTION OF SYMBOLS 100... Management information identification system, 2... Imaging device, 3... Learning device, 31... Image data storage part, 32... Teacher data storage part, 33... Image data input part for learning, 34... Teacher data generation part, 35... Learning Data generation section, 36...Learning data output section, 4...Identification device, 41...Image data storage section, 42...Learning data storage section, 43...Correspondence information storage section, 44...Identification image data input section, 45...Learning data Input section, 46...Identification section, 47...Management information output section

Claims (9)

Training image data that is a display that can be displayed at different display positions to be managed and that represents one label by a combination of displays displayed at different display positions ; a learning unit that learns a relationship between the learning image data and the label element using a machine learning method based on teacher data indicating the label element shown in the management target captured in the image data; ,
Based on the identification image data in which the label element to be identified is imaged and the learning data indicating the learning result of the relationship by the learning section, the label element shown in the management object imaged in the identification image data is a first identification part that identifies the label element that is
a second identification unit that identifies a label displayed on the management target based on an identification result of the label element shown on the management target;
Equipped with
The object to be managed is a coil material manufactured by winding a steel material formed into a plate shape by rolling into a coil shape,
The display position is an end surface, an inner diameter surface, or an outer peripheral surface of the coil material,
Label identification device. The learning unit generates learning data for each display position by learning a relationship between image data captured at each display position to be managed and a label element displayed at the display position,
The first identification unit identifies a label shown at each display position of the management target based on learning data for each display position and identification image data in which each display position of the management target is captured. identify the element,
The label identification device according to claim 1. The learning unit uses , as the learning image data, image data obtained by removing part or all of an area other than the area where the display position was imaged from the image data acquired by the imaging unit.
The label identification device according to claim 1 or 2. the label element is identified by its shape and color combination;
A label identification device according to any one of claims 1 to 3 . an imaging step of acquiring image data of a label element representing one label by a combination of displays displayed at different display positions, which can be displayed at different display positions to be managed;
Based on image data for learning and training data indicating the label element shown on the management target captured in the image data, the relationship between the image data and the label element is determined by a machine learning method. learning steps to learn;
a first identification that identifies the label element shown in the management target imaged in the image data, based on image data for identification and learning data indicating a learning result of the relationship in the learning step; step and
a second identification step of identifying a label displayed on the management target based on an identification result of the label element shown on the management target;
has
The object to be managed is a coil material manufactured by winding a steel material formed into a plate shape by rolling into a coil shape,
The display position is an end surface, an inner diameter surface, or an outer peripheral surface of the coil material,
Label identification method. Displays that can be displayed at different display positions to be managed, and image data in which a label element representing one label by a combination of displays displayed at different display positions is captured, and image data captured in the image data. A learning device that uses a machine learning method to learn a relationship between the image data and the label element based on teacher data indicating the label element shown in a management target,
The object to be managed is a coil material manufactured by winding a steel material formed into a plate shape by rolling into a coil shape,
The display position is an end surface, an inner diameter surface, or an outer peripheral surface of the coil material,
learning device. Displays that can be displayed at different display positions to be managed, and image data in which a label element representing one label by a combination of displays displayed at different display positions is captured, and image data captured in the image data. A learning result of the relationship learned by a learning device that learns the relationship between the image data and the label element using a machine learning method based on teacher data indicating the label element shown in the management target. and identifying the label element shown in the management target imaged in the identification image data based on learning data indicating the identification target label element and identification image data in which the identification target label element is imaged,
An identification device that identifies a label displayed on the managed object based on an identification result of the label element shown on the managed object,
The object to be managed is a coil material manufactured by winding a steel material formed into a plate shape by rolling into a coil shape,
The display position is an end surface, an inner diameter surface, or an outer peripheral surface of the coil material,
Identification device. Displays that can be displayed at different display positions to be managed, and image data in which a label element representing one label by a combination of displays displayed at different display positions is captured, and image data captured in the image data. a step of learning the relationship between the image data and the label element by a machine learning method based on training data indicating the label element shown in the management target;
A computer program to be executed by a computer ,
The object to be managed is a coil material manufactured by winding a steel material formed into a plate shape by rolling into a coil shape,
The display position is an end surface, an inner diameter surface, or an outer peripheral surface of the coil material,
computer program. Displays that can be displayed at different display positions to be managed, and image data in which a label element representing one label by a combination of displays displayed at different display positions is captured, and image data captured in the image data. A learning result of the relationship learned by a learning device that learns the relationship between the image data and the label element using a machine learning method based on teacher data indicating the label element shown in the management target. and a step of identifying the label element shown in the management target imaged in the identification image data based on learning data indicating the image data and identification image data in which the identification target label element is imaged. ,
identifying a label displayed on the managed object based on an identification result of the label element shown on the managed object;
A computer program for causing a computer to execute
The object to be managed is a coil material manufactured by winding a steel material formed into a plate shape by rolling into a coil shape,
The computer program wherein the display position is an end surface, an inner diameter surface, or an outer circumferential surface of the coil material.

Images