JP7737795B2 - Gripping position calculation system, Gripping position calculation program, Trained model, and bone-in meat gripping system - Google Patents

Description

This disclosure relates to a gripping position calculation system, a gripping position calculation program, a trained model, and a bone-in meat gripping system.

Gripping devices for gripping bone-in meat are known in the art. For example, if the bone-in meat is a bone-in leg that includes a neck with a cut, the gripping device needs to grip a portion of the neck near the cut position, depending on the size of the bone-in leg. In the gripping device disclosed in Patent Document 1, the length from the cut position to the tip of the neck is collected in advance as data by actual measurement. Based on the collected data, the average value of this length and the standard deviation of this data are calculated. The position displaced from the tip of the neck by a specified amount determined by these values is considered to be the gripping position, and the gripping device grips the portion of the neck located at this position.

Japanese Patent Application Laid-Open No. 2020-178633

In the gripping device disclosed in Patent Document 1, the same specified amount is applied to each of multiple pieces of bone-in leg meat when the gripping operation is performed, but it is preferable that the gripping position be determined by further taking into consideration the size or shape of each piece of bone-in leg meat.

One embodiment of the present disclosure aims to provide a gripping position calculation system, a gripping position calculation program, a trained model, and a bone-in meat gripping system that can identify an appropriate gripping position according to the size or shape of bone-in meat.

A grip position calculation system according to at least one embodiment of the present invention includes:
an image acquisition unit for acquiring image data relating to a captured image of the bone-in meat;
A memory unit that stores a trained model using a neural network that has been trained to output grip position data regarding the grip position of the bone-in meat when the image data is input;
and a grip position acquisition unit for inputting the image data acquired by the image acquisition unit into the trained model to acquire the grip position data.

A gripping position calculation program according to at least one embodiment of the present invention includes:
On the computer,
an image acquisition step for acquiring image data relating to a photographed image of the bone-in meat;
When the image data is input, a trained model using a neural network trained to output gripping position data regarding the gripping position of the bone-in meat is input with the image data acquired by the image acquisition step, and a gripping position acquisition step is executed to acquire the gripping position data.

A learning model according to at least one embodiment of the present invention comprises:
A trained model using a neural network,
an input layer for receiving image data relating to a photographed image of bone-in meat;
an output layer for outputting gripping position data relating to the gripping position of the bone-in meat in the image data input to the input layer;
a plurality of intermediate layers in which parameters including weights and biases of connections between neurons are learned based on teacher data that associates the teacher image data with the teacher grip position data of the bone-in meat included in the image data,
When the image data is input to the input layer, the computer is caused to function so that the grip position data is output from the output layer after being subjected to calculations by the plurality of intermediate layers.

A bone-in meat gripping system according to at least one embodiment of the present invention comprises:
a photographing device for photographing the bone-in meat;
a gripping device for gripping the bone-in meat;
an imaging control unit for controlling the imaging device to photograph the bone-in meat;
The grip position calculation system is configured such that the image acquisition unit acquires the image data of the captured image related to the image captured by the imaging device;
and a gripping control unit for controlling the gripping device to grip the bone-in meat based on the gripping position data acquired by the gripping position acquisition unit.

Some embodiments provide a gripping position calculation system, gripping position calculation program, trained model, and bone-in meat gripping system that can identify an appropriate gripping position according to the size or shape of the bone-in meat.

1 is a plan view of a bone-in meat gripping system according to one embodiment; FIG. FIG. 1 is a front view of a bone-in meat gripping system according to one embodiment. FIG. 2 is a diagram illustrating an image generated by an imaging device according to an embodiment. FIG. 1 illustrates a trained model according to an embodiment. FIG. 10 is a diagram illustrating training data according to an embodiment. FIG. 2 is a block diagram illustrating the functions of a bone-in meat gripping system according to one embodiment. 10 is a flowchart illustrating a process for identifying a grip position according to an embodiment.

Hereinafter, several embodiments of the present invention will be described with reference to the accompanying drawings. However, the dimensions, materials, shapes, relative arrangements, etc. of components described as embodiments or shown in the drawings are merely illustrative examples and are not intended to limit the scope of the present invention.
For example, expressions expressing relative or absolute arrangement such as "in a certain direction,""along a certain direction,""parallel,""orthogonal,""center,""concentric," or "coaxial" not only express such an arrangement exactly, but also express a state in which there is a relative displacement with a tolerance or an angle or distance to the extent that the same function is obtained.
For example, expressions such as "identical,""equal," and "homogeneous" that indicate that something is in an equal state not only indicate a state of strict equality, but also indicate a state in which there is a tolerance or a difference to the extent that the same function is obtained.
For example, expressions representing shapes such as a square shape or a cylindrical shape not only represent shapes such as a square shape or a cylindrical shape in the strict geometric sense, but also represent shapes including uneven portions, chamfered portions, etc., to the extent that the same effect can be obtained.
On the other hand, the expressions "comprises,""includes,""has,""includes," or "has" one element are not exclusive expressions that exclude the presence of other elements.

Fig. 1 is a plan view of a bone-in meat gripping system 1 according to one embodiment, and Fig. 2 is a front view of the bone-in meat gripping system 1 according to one embodiment. The bone-in meat gripping system 1 is configured to grip bone-in meat 5. The bone-in meat 5 is bone-in beef, horse meat, venison, wild boar meat, whale meat, fish meat, or the like.
In one embodiment, the bone-in meat 5 is pork leg meat including a main body portion 5a and a neck portion 5b protruding from the main body portion 5a.

In one embodiment, the bone-in meat gripping system 1 includes a photographing device 40 that photographs the bone-in meat 5, a gripping position calculation system 300 that identifies the gripping position of the photographed bone-in meat 5, and a gripping device 50 that grips the bone-in meat 5 at the identified gripping position.
In one embodiment, the bone-in meat 5 to be photographed is positioned by a conveying device 100 .
The conveying device 100 includes a conveying section 101 that conveys the bone-in meat 5, and a positioning member 104 that positions the bone-in meat 5 being conveyed. The conveying section 101 is a belt conveyor located below the photographing device 40. The positioning member 104 is a pair of plate-like bodies 104a.
In one embodiment, the bone-in meat 5 held by the holding device 50 is transferred to the conveying device 60 and conveyed.
In other embodiments, the bone-in meat gripping system 1 may not include the transport device 100 and the conveying device 60. For example, the photographing device 40 may photograph the bone-in meat 5 that is transported by human power.

The grip position calculation system 300 according to one embodiment includes an image acquisition unit 322 for acquiring data relating to an image captured by the imaging device 40 as image data, a memory unit 310 that stores a trained model 305 that has been trained to output grip position data indicating a target grip position, a grip position acquisition unit 324 that inputs the image data acquired by the image acquisition unit 322 into the trained model 305 to acquire grip position data, and a grip position output unit 327 that outputs the acquired grip position data to the grip device 50.

The gripping device 50 grips the bone-in meat 5 at the position indicated by the gripping position data output from the gripping position output unit 327 .
The gripping device 50 of one embodiment includes an articulated robot arm 52 and a clamp 53 provided at the tip of the articulated robot arm 52. By the operation of the articulated robot arm 52, the clamp 53 grips the bone-in meat 5 at the position indicated by the gripping position data.
In one embodiment, the clamp 53 includes a fixed portion 53a and a movable portion 53b that is displaceable in a direction toward and away from the fixed portion 53a.
In one embodiment, the clamp 53 holds the neck portion 5b of the bone-in meat 5. The held bone-in meat 5 is delivered to the transport device 60 in a hanging position.

In one embodiment, the conveying device 60 includes a receiving section 62 and a conveying section 64 that conveys the bone-in meat 5 received by the receiving section 62 .
The receiving portion 62 according to one embodiment includes a hole 63 that is open in the vertical direction. The hole 63 includes a guide hole 65 for guiding the neck portion 5 b and an engagement hole 66 with which the guided bone-in meat 5 engages. In one embodiment, the engagement hole 66 is configured to engage with the neck portion 5 b of the bone-in meat 5.
In one embodiment, in a plan view, the guide hole 65 narrows toward the engagement hole 66 , and the engagement hole 66 extends linearly in a direction away from the guide hole 65 .
In one embodiment, the bone-in meat 5 engaged with the engagement hole 66 is transported by the transport portion 64 .

3, an image 45 generated by the imaging device 40 includes a portion of the bone-in meat 5. More specifically, as an example, the image 45 includes a downstream portion of the positioned bone-in meat 5 in the conveyance direction.
In one embodiment, in addition to the neck portion 5 b, the image 45 also includes a contact portion 5 c, which is a portion of the main body portion 5 a that comes into contact with the positioning member 104 .
In other embodiments, the image 45 may include all of the positioned bone-in meat 5 .

The image 45 of the embodiment includes the specified area 44. Image data relating to the image of the specified area 44 (hereinafter also referred to as a captured image 48) is output to the image acquisition unit 322 described above.
In one embodiment, the captured image 48 partially captures a portion of the bone-in meat 5 that is to be the target gripping position. In one embodiment, the target gripping position is the neck portion 5b of the bone-in meat 5. More specifically, as an example, the target gripping position is a cut C formed in the neck portion 5b. In this case, the gripping position data indicates the position of the cut C.
As an example, the incision C is formed by an operator cutting the neck portion 5b with a knife so as to cut the Achilles tendon included in the neck portion 5b. The incision C may also be formed automatically by a cutting device.
In other embodiments, the incision C may not be formed in the neck portion 5b included in the specified area 44. Furthermore, the part of the bone-in meat 5 included in the specified area 44 may not be the target gripping position. In other words, the part of the bone-in meat 5 that is not included in the specified area 44 may be the target gripping position.
Furthermore, the captured image 48 may be any image related to the image 45 generated by the image capturing device 40. For example, instead of being an image of the specified area 44, the captured image 48 may be the same image as the image 45.

FIG. 4 is a diagram illustrating a trained model 305 stored in the storage unit 310 according to an embodiment.
The trained model 305 using a neural network is machine-trained to output gripping position data in response to input of image data. The trained model 305 includes an input layer 311, an output layer 313, and multiple intermediate layers 312. Note that in FIG. 4 , the multiple intermediate layers 312 are simplified and illustrated using a single reference numeral 312. The input layer 311 is configured to receive image data. The output layer 313 is configured to output gripping position data for the bone-in meat 5. In the intermediate layer 312, hyperparameters including the weights and biases of connections between neurons are trained based on training data 450 (described below) that associates image data with gripping position data. When image data is input to the input layer 311, the trained model 305 controls the gripping position acquisition unit 324 to output gripping position data from the output layer 313 after calculations by the multiple intermediate layers 312.
The grip position data output from the output layer 313 may be any data as long as it indicates the target grip position. For example, the grip position data may be coordinate data in the captured image 48 (or image 45). As a more specific example, the grip position data may be the horizontal length from the reference line Lb at an arbitrary specified position to the target grip position (dimension La in FIG. 3). Alternatively, the grip position data may be the horizontal length from the tip of the neck portion 5b to the target grip position, where Lc = Lw - La (dimension Lc in FIG. 3). Alternatively, in another embodiment, the grip position data may be coordinate data in real three-dimensional space.

FIG. 5 is a diagram showing training data 450 according to an embodiment for training a training model, which is a model before training of the trained model 305.
In one embodiment, the process of generating the trained model 305 from the learning model includes a process of generating training data 450 and a process of training the learning model.

In the process of generating the teacher data 450, teacher image data is prepared.
In one embodiment, a plurality of images 45 of bone-in meat 5 are prepared for use as a teacher, and a photographed image 48 for use as a teacher is generated from these images 45 .
In one embodiment, the captured image 48 is generated by a cropping process, trimming process, or the like.
For each of the generated teacher's photographed images 48, teacher's gripping position data that serves as correct answer data is identified.
In one embodiment, the teacher gripping position data is determined based on the teacher image 45. More specifically, as an example, the operator of the bone-in meat gripping system 1 specifies the incision C in each image 45 using an image analysis device (not shown), and determines the length in the lateral direction from the reference line Lb to the incision C (the dimension La in FIG. 3 ).
The identified teacher grip position data is associated with each teacher photographed image 48, thereby generating teacher data 450.
The reference line Lb is an arbitrary predetermined position. In one embodiment, the reference line Lb is, for example, at a predetermined position on the transport path of the bone-in meat 5 transported by the transport device 100. More specifically, the reference line Lb in one embodiment is set at a position included in the image 45.
In other embodiments, the teacher's grip position data may be specified by the length from one end (e.g., the left end) of the image 45. Alternatively, the teacher's grip position data may be specified by the length from one end (e.g., the left end) of the captured image 48. Alternatively, the teacher's grip position data may be specified by actual measurement instead of using an image analysis device.
In still another embodiment, the cut C may not be visible in the bone-in meat 5 shown in the teacher image 45. Even in this case, the operator can specify the teacher's grip position data.

In the learning model learning process that follows the process of generating the training data 450, the training data 450 is divided into a training data set and a validation set. When the training data set is input into the learning model, hyperparameters such as the weights and biases of the connections between neurons are set. The hyperparameters include the learning coefficient and batch size of the learning model. The batch size is the amount of image data read by the learning model in one learning session (i.e., the number of training images 48). After tuning is complete, the above training data 450 is input again into the learning model, and learning begins. This generates a trained model 305, and the learning process of the learning model is completed.

FIG. 6 is a block diagram showing the functions of the bone-in meat gripping system 1 according to one embodiment.
The bone-in meat gripping system 1 of one embodiment includes a control unit 90 , an image processing control unit 200 , and a gripping position calculation system 300 .
In one embodiment, each of these components is provided with at least one processor unit. The processor included in the processor unit is configured to load the read program into memory and execute instructions included in the loaded program (the grip position calculation program 10). The processor may be, for example, a CPU, a GPU, an MPU, a DSP, or any other type of computing device, or a combination of these. The processor may also be implemented by an integrated circuit such as a PLD, an ASIC, an FPGA, or an MCU.
In one embodiment, these components transmit and receive data (signals) via wires or wirelessly.

In one embodiment, the control unit 90 is a programmable logic controller (PLC) including an MPU. In one embodiment, the control unit 90 includes a transport control unit 103, an imaging command unit 106, a grip control unit 109, and a transportation control unit 112.

The transport control unit 103 is configured to control the transport device 100 .
In one embodiment, the transport control unit 103 controls a transport driving unit 102 for driving the transport unit 101 .
In one embodiment, the conveyance control unit 103 controls the conveyance drive unit 102 based on at least one of the timer 35 and the sensor 105 that detects the conveyed bone-in meat 5. For example, the conveyance control unit 103 drives the conveyance drive unit 102 from when the detection result of the sensor 105 changes until the timer 35 counts a specified time. As a result, the bone-in meat 5 comes into contact with the positioning member 104 and is positioned, regardless of its size or shape.
In one embodiment, the sensor 105 is a photoelectric sensor that detects bone-in meat 5 that has passed the downstream end of the conveying section 101.

The photographing command unit 106 is configured to output a photographing instruction signal to the image processing control unit 200 when controlled by the transport control unit 103 .
In some embodiments, the imaging command unit 106 may output an illumination instruction signal to the illumination unit 107 in addition to the imaging instruction signal. The illumination unit 107 is configured to irradiate the imaging area of the imaging device 40 with light.

The gripping control unit 109 is configured to control the gripping device 50 based on the gripping position data.
In one embodiment, the grip control unit 109 controls the grip drive unit 55 included in the grip device 50 .
In one embodiment, the gripping drive unit 55 drives the articulated robot arm 52 and the clamp 53. For example, the gripping drive unit 55 includes at least one motor that drives the articulated robot arm 52 and an air cylinder that drives the clamp 53.
In another embodiment, the gripping drive unit 55 may include a motor for driving the clamp 53 instead of the air cylinder.

The transport control unit 112 is configured to control the transport unit 64 after being controlled by the grip control unit 109.

In one embodiment, the image processing control unit 200 is realized by a processor unit including a CPU.
In one embodiment, the image processing control unit 200 includes an imaging control unit 203 and an image processing unit 205 .
The imaging control unit 203 is configured to cause the imaging device 40 to perform imaging. For example, upon receiving an imaging instruction signal from the imaging command unit 106, the imaging control unit 203 causes the imaging device 40 to perform imaging.
In one embodiment, the image processing unit 205 is configured to perform image processing on the image 45 generated by the image capturing device 40 to extract the specified region 44 of the image 45. The image processing for extracting the specified region 44 is, for example, cropping or trimming.
In one embodiment, the image data generated by the image processing unit 205 is output to the grip position calculation system 300 .
In another embodiment, the image processing control unit 200 may not include the image processing unit 205. In this case, the image data of the image 45 may be output to the image acquisition unit 322 of the gripping position calculation system 300 without being subjected to image processing.

In one embodiment, the grip position calculation system 300 is implemented by a processor unit including a GPU. An example of the configuration of the grip position calculation system 300 is as described above.
In one embodiment, the image acquisition unit 322 of the grip position calculation system 300 acquires image data output from the image processing unit 205. In one embodiment, the image data includes RGB brightness values for each pixel. The grip position output unit 327 outputs the grip position data acquired by the grip position acquisition unit 324 to the grip control unit 109 via the image processing unit 205.
In other embodiments, the grip position output unit 327 may output the grip position data directly to the grip control unit 109 or the grip device 50 .
In still another embodiment, the gripping position calculation system 300 may not include the gripping position output unit 327. In this case, the gripping position calculation system 300 may be provided in a remote location away from the control unit 90 or the image processing unit 205. The gripping position calculation system 300 may transmit and receive data (signals) to and from the control unit 90 or the image processing unit 205 via the Internet, for example.

Referring to Figure 7, the control process executed by the bone-in meat gripping system 1 when gripping bone-in meat 5 will be described.

The transport control unit 103 causes the transport unit 101 to position the bone-in meat 5 (S11). In one embodiment, positioning is performed by the transport control unit 103 stopping the transport drive unit 102 after a specified time has elapsed since the detection result of the sensor 105 changed.

When controlled by the transport control unit 103, the photography control unit 203 causes the photography device 40 to photograph the bone-in meat 5 (S13). In one embodiment, the photography control unit 203 controls the photography device 40 upon receiving a photography instruction signal from the photography command unit 106. An image 45 of the positioned bone-in meat 5 is generated and output to the image processing unit 205.

After control by the shooting control unit 203, the image processing unit 205 processes the image 45 (S15). In one embodiment, the image processing unit 205 performs crop processing on the image 45 to generate image data for the captured image 48 (S15). The image acquisition unit 322 acquires the image data from the image processing unit 205 (S17). Note that in other embodiments, the captured image 48 may be the same image as the image 45. For example, in one embodiment in which the image processing unit 205 is not provided, the image acquisition unit 322 acquires the image data for the image 45 (S17).

The grip position acquisition unit 324 acquires grip position data (S19). In one embodiment, the grip position acquisition unit 324 inputs image data to the input layer 311 of the trained model 305, thereby acquiring grip position data from the output layer 313. In one embodiment, the grip position output unit 327 acquires this grip position data and outputs it to the grip control unit 109 via the image processing unit 205.

The gripping control unit 109 causes the gripping device 50 to perform the gripping operation of the positioned bone-in meat 5 (S21). In one embodiment, the gripping control unit 109 controls the gripping drive unit 55 to grip the part of the neck portion 5b located at the position indicated by the gripping position data.
In one embodiment, the gripping device 50 grips the incision C with the clamp 53, lifts the bone-in meat 5, and delivers it to the transporting device 60. After delivery, the clamp 53 releases its grip on the bone-in meat 5. Note that the gripping control unit 109 may control the gripping device 50 based on the result of some kind of processing performed on the gripping position data.

The following provides an overview of the gripping position calculation system 300, gripping position calculation program 10, trained model 305, and bone-in meat gripping system 1 according to several embodiments.

(1) The grip position calculation system 300 according to at least one embodiment of the present invention includes:
an image acquisition unit 322 for acquiring image data relating to the captured image 48 of the bone-in meat 5;
a memory unit (310) that stores a trained model (305) using a neural network that has been trained to output gripping position data regarding the gripping position of the bone-in meat (5) when the image data is input;
and a gripping position acquisition unit 324 for inputting the image data acquired by the image acquisition unit 322 into the trained model 305 to acquire the gripping position data.

According to the configuration (1) above, the photographed image 48 of the bone-in meat 5 is input to the trained model 305, and the gripping position acquisition unit 324 can acquire gripping position data corresponding to the photographed bone-in meat 5. Therefore, an appropriate gripping position according to the size or shape of the bone-in meat 5 can be identified.
In one embodiment, if the grip position is not properly identified, four possible problems may occur:
First, the gripping device 50 may attempt to grip a thick portion of the meat that has shifted too far toward the main body 5a from the appropriate gripping position, resulting in the gripping device 50 failing to grip the bone-in meat 5. In this case, the production line for the bone-in meat 5 may stop.
Secondly, even if the gripping device 50 can grip the thick portion, in one embodiment in which the bone-in meat 5 is transferred to the conveying device 60, the gripped portion cannot enter the guide hole 65 and becomes stuck. This case may also cause the line to stop.
Thirdly, even if the thick portion can enter the guide hole 65, this portion may come into excessive contact with the guide hole 65. For example, in one embodiment in which the gripping device 50 grips the bone-in meat 5 in a hanging manner, the bone-in meat 5 may come into excessive contact with the guide hole 65 and rotate around its axial direction in the up-and-down direction. In this case, the bone-in meat 5 may be transported to the transport unit 64 in an inappropriate posture.
Fourth, there is a problem that the gripping device 50 grips a part of the neck portion 5b that is displaced too far toward the tip side from the appropriate gripping position, and as a result, the bone-in meat 5 being gripped is displaced downward due to its own weight, etc. In this case, in one embodiment in which the bone-in meat 5 is transferred from the gripping device 50 to the transport device 60, the bone-in meat 5 may fall off.
In this regard, in one embodiment, the gripping position calculation system 300 identifies an appropriate gripping position, allowing the gripping device 50 to properly grip the neck portion 5b, thereby suppressing the above four problems.

(2) In some embodiments, in the configuration of (1),
The image acquisition unit 322 is configured to acquire the image data regarding the captured image 48, which partially captures the neck portion 5b, which is an example of an end portion of the bone-in meat 5 positioned by the positioning member 104 that is away from the contact portion 5c with the positioning member 104.

According to the configuration (2) above, in the photographed image 48 that partially captures the neck portion 5b, which is an example of an end portion that is separated from the contact portion 5c with the positioning member 104, the appearance of the photographed image 48 changes significantly overall depending on the size of the bone-in meat 5. By inputting the image data of the photographed image 48, which is prone to change in appearance, into the trained model 305, the gripping position acquisition unit 324 can identify a more appropriate gripping position.
The appearance of the photographed image 48 refers to, for example, the position of the tip of the neck 5b in the horizontal direction of the photographed image 48, the position of the tip of the neck 5b in the vertical direction, or the area of the neck 5b shown in the photographed image 48.
Furthermore, in one embodiment, the photographed image 48 does not include the conveying unit 101, which appears the same regardless of the size of the bone-in meat 5. Therefore, the appearance of the photographed image 48 changes further depending on the size of the bone-in meat 5, and a more appropriate gripping position can be identified.

(3) In some embodiments, in the configuration of (2),
The image acquisition unit 322 is configured to acquire the image data relating to the captured image 48, which partially captures the neck portion 5b, which is an example of one end portion in the transport direction by the transport device 100, of the bone-in meat 5 that has been positioned and prevented from being transported by the transport device 100 due to contact between the contact area 5c and the positioning member 104.

According to the configuration (3) above, when the bone-in meat 5 is positioned by the positioning member 104, the position of the neck portion 5b, which is an example of one end of the bone-in meat 5, in the conveying direction changes depending on the size of the bone-in meat 5. By inputting image data relating to this captured image 48 into the trained model 305, the gripping position acquisition unit 324 can identify a more appropriate gripping position.

(4) In some embodiments, in any of the configurations (1) to (3) above,
The gripping position calculation system 300 includes a gripping position output unit 327 for outputting the gripping position data acquired by the gripping position acquisition unit 324 to a gripping device 50 configured to grip the photographed bone-in meat 5.

With the configuration (4) above, the gripping device 50 can properly grip the bone-in meat 5 based on the gripping position data acquired by the gripping position acquisition unit 324.

(5) The grip position calculation program 10 according to at least one embodiment of the present invention
On the computer,
an image acquisition step (S17) for acquiring image data relating to the photographed image 48 of the bone-in meat 5;
When the image data is input, the image data acquired in the image acquisition step (S17) is input into a trained model 305 using a neural network trained to output gripping position data regarding the gripping position of the bone-in meat 5, and a gripping position acquisition step (S19) is executed to acquire the gripping position data.

The configuration (5) above allows for the same reason as (1) above to identify an appropriate gripping position depending on the size or shape of the bone-in meat 5.

(6) The trained model 305 according to at least one embodiment of the present invention is
A trained model 305 using a neural network,
an input layer 311 for inputting image data relating to the photographed image 48 of the bone-in meat 5;
an output layer 313 for outputting gripping position data relating to the gripping position of the bone-in meat 5 of the image data input to the input layer 311;
a plurality of intermediate layers 312 in which parameters including weights and biases of connections between neurons are learned based on teacher data 450 that associates the image data for teacher use with the grip position data for teacher use of the bone-in meat included in the image data;
When the image data is input to the input layer 311, the computer is caused to function so that the image data is calculated by the plurality of intermediate layers 312 and then output from the intermediate layers 312 the grip position data.

The configuration (6) above allows for the same reason as (1) above to identify an appropriate gripping position depending on the size or shape of the bone-in meat 5.

(7) The bone-in meat gripping system 1 according to at least one embodiment of the present invention includes:
a photographing device 40 for photographing the bone-in meat 5;
A gripping device 50 for gripping the bone-in meat 5;
an imaging control unit 203 for controlling the imaging device 40 to photograph the bone-in meat;
The grip position calculation system 300 according to any one of (1) to (4), wherein the image acquisition unit 322 is configured to acquire the image data of the captured image 48 relating to the image 45 captured by the imaging device 40;
and a gripping control unit 109 for controlling the gripping device 50 to grip the bone-in meat 5 based on the gripping position data acquired by the gripping position acquisition unit 324.

The configuration (7) above allows for the same reason as (1) above to identify an appropriate gripping position depending on the size or shape of the bone-in meat 5.

Various modifications can be applied to the bone-in meat gripping system 1 according to one embodiment.
The gripping device 50 may grip, for example, the end of the main body 5 a on the opposite side to the neck 5 b instead of gripping the neck 5 b. In this case, the gripping position data indicates the gripping position of the end of the main body 5 a. This gripping position data may be determined based on the above-mentioned captured image 48 that partially captures the neck 5 b, or may be determined based on an image that partially captures the end of the main body 5 a.

At least two of the control unit 90, image processing control unit 200, and grip position calculation system 300 may be realized by a single processor unit.

1: Bone-in meat gripping system 5: Bone-in meat 5c: Contact area 10: Grip position calculation program 40: Photography device 45: Image 48: Photographed image 50: Grip device 90: Control unit 100: Conveying device 104: Positioning member 109: Grip control unit 203: Photography control unit 300: Grip position calculation system 305: Trained model 310: Memory unit 311: Input layer 312: Intermediate layer 313: Output layer 322: Image acquisition unit 324: Grip position acquisition unit 327: Grip position output unit 450: Training data

Claims (7)

A calculation system for a gripping position of bone-in meat by an articulated robot,
an image acquisition unit for acquiring image data relating to a photographed image of the bone-in meat;
A memory unit that stores a trained model using a neural network that has been trained to output grip position data regarding the grip position of the bone-in meat when the image data is input;
a gripping position acquisition unit for inputting the image data acquired by the image acquisition unit into the trained model to acquire the gripping position data ,
The image acquisition unit is configured to acquire the image data relating to the photographed image, which partially captures an end portion of the bone-in meat that has been positioned and prevented from being conveyed by the conveying device due to contact with the positioning member, the end portion being away from the contact portion with the positioning member, and which does not capture the conveying device.
Grip position calculation system. 2. The gripping position calculation system according to claim 1 , wherein the image acquisition unit is configured to acquire the image data relating to the captured image, which partially captures only the portion of the bone-in meat positioned by the positioning member that protrudes downstream in the conveying direction from the conveying unit of the conveying device. 3. The gripping position calculation system according to claim 1 , wherein the image acquisition unit is configured to acquire the image data relating to the captured image, which partially captures the downstream end of the bone-in meat in the conveying direction by the conveying device, of the bone-in meat that has been positioned and prevented from being conveyed by the conveying device due to contact between the contact portion and the positioning member. A gripping position calculation system according to any one of claims 1 to 3, comprising a gripping device configured to grip the photographed bone-in meat, the gripping device comprising a gripping position output unit for outputting the gripping position data acquired by the gripping position acquisition unit. A program for calculating a gripping position of bone-in meat by an articulated robot,
On the computer,
an image acquisition step for acquiring image data relating to a photographed image of the bone-in meat;
When the image data is input, the image data acquired by the image acquisition step is input to a trained model using a neural network that has been trained to output gripping position data regarding the gripping position of the bone-in meat, and a gripping position acquisition step is executed to acquire the gripping position data;
The image acquisition step is configured to acquire the image data relating to the photographed image that partially captures an end portion of the bone-in meat that has been positioned and prevented from being conveyed by the conveying device due to contact with the positioning member, the end portion being away from the contact portion with the positioning member, and the conveying device is not captured in the photographed image.
Gripping position calculation program. A trained model using a neural network for calculating a gripping position of bone-in meat by an articulated robot,
an input layer for receiving image data relating to a photographed image of bone-in meat;
an output layer for outputting gripping position data relating to the gripping position of the bone-in meat of the image data input to the input layer;
a plurality of intermediate layers in which parameters including weights and biases of connections between neurons are learned based on teacher data that associates the teacher image data with the teacher grip position data of the bone-in meat included in the image data,
the input layer is configured to receive the image data relating to the photographed image, which partially captures an end portion of the bone-in meat that has been positioned by contact with a positioning member and prevented from being conveyed by the conveying device, that is not in contact with the positioning member, and which does not capture the conveying device;
A trained model that causes a computer to function such that, when the image data is input to the input layer, the grip position data is output from the output layer after undergoing calculations by the multiple intermediate layers. a photographing device for photographing the bone-in meat;
a gripping device for gripping the bone-in meat;
an imaging control unit for controlling the imaging device to photograph the bone-in meat;
The gripping position calculation system according to claim 1 , wherein the image acquisition unit is configured to acquire the image data of the captured image related to the image captured by the imaging device;
a gripping control unit for controlling the gripping device to grip the bone-in meat based on the gripping position data acquired by the gripping position acquisition unit.