Cut point determination system, cut point determination method, and cut point determination program for meat

Abstract

To determine a cut point to be used to process meat with high accuracy.SOLUTION: The present invention relates to a cut point determination system for meat. The cut point determination system acquires first image data corresponding to a first area of a picked-up image of the meat. A plurality of pieces of position information is inferred by inputting the acquired first image data to a first inference model. The validity of the inferred position information is determined on the basis of second image data corresponding to a second area of a picked-up image specified by using at least some of the plurality of pieces of position information that are inferred, and a cut point is determined on the basis of the determination result.SELECTED DRAWING: Figure 2

Description

Technical Field

[0001] The present application relates to a cut point determination system for meat, a cut point determination method, and a cut point determination program.

Background Art

[0002] For example, in meat processing targeting the meat carcasses of relatively large livestock such as pigs and cows, the processed meat carcass is cut into a pair of forequarters by splitting it in half left and right at the position of the spine (backbone), and further, each forequarter is cut into the fore part, the middle part, and the hind part. Conventionally, such cutting processes have been performed manually by workers using knives, but from the viewpoints of ensuring quality uniformity, worker safety, and improving processing efficiency, automation using meat processing machines has been desired.

[0003] For example, Patent Document 1 discloses a technique related to a cutting device and method in the large division process of forequarters. In this document, it is described that a pair of forequarters to be cut is suspended and supported, and is supported by a support bar extending substantially horizontally from the front and rear, so that the posture of the forequarter at the time of cutting is fixed.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] In order to cut the meat with good quality in an automated cutting process as in the above Patent Document 1, for example, it is necessary to perform a cutting operation along a predetermined cutting line while stably maintaining the posture of the meat by sandwiching the meat with a rod-shaped gripping member. Such a cutting position (hereinafter referred to as "cut point") can be determined based on position information for specifying the position of a characteristic portion of the meat, which is detected by analyzing an image obtained by imaging the meat to be processed.

[0006] However, since there is no single piece of meat that is the same as the one to be processed, it is difficult to accurately detect the cut point only by image processing. Also, the accuracy of the cut point is greatly involved in the cut success rate and quality of the meat. Therefore, an estimation method using an inference model constructed by machine learning such as deep learning (DL: Deep Learning) that utilizes the information of the accumulated image data has been studied. However, whether the cut point estimated by such a method indicates the correct position cannot be determined from the deep learning model. Therefore, in order to improve the accuracy of the cut point, it is necessary to evaluate the validity of the estimated cut point and set it to the correct position.

[0007] At least one embodiment of the present disclosure has been made in view of the above circumstances, and an object thereof is to provide a cut point determination system for meat, a cut point determination method, and a cut point determination program that can accurately determine a cut point used for processing meat.

Means for Solving the Problems

[0008] The cut point determination system for meat according to at least one embodiment of the present disclosure is configured to solve the above problems by an image data acquisition unit configured to acquire first image data corresponding to a first region of an imaged image of meat, a position information estimation unit that inputs the first image data acquired by the image data acquisition unit into a first inference model and estimates a plurality of pieces of position information, A validity determination unit that determines the validity of the estimation result in the position information estimation unit based on second image data corresponding to a second region of the captured image specified using at least a part of the plurality of position information; A cut point determination unit that determines a cut point based on the determination result of the validity determination unit; It is provided with.

[0009] In order to solve the above problems, a cut point determination method for meat according to at least one embodiment of the present disclosure A step of acquiring first image data corresponding to a first region of a captured image of meat; A step of inputting the first image data into a first inference model to estimate a plurality of position information; A step of determining the validity of the estimation result of the plurality of position information based on second image data corresponding to a second region of the captured image specified using at least a part of the plurality of position information; A step of determining a cut point based on the determination result of the validity; It is provided with.

[0010] In order to solve the above problems, a cut point determination program for meat according to at least one embodiment of the present disclosure To a computer device A step of acquiring first image data corresponding to a first region of a captured image of meat; A step of inputting the first image data into a first inference model to estimate a plurality of position information; A step of determining the validity of the estimation result of the plurality of position information based on second image data corresponding to a second region of the captured image specified using at least a part of the plurality of position information; A step of determining a cut point based on the determination result of the validity; It is executable.

Effect of the Invention

[0011] According to at least one embodiment of the present disclosure, it is possible to provide a cut point determination system for meat, a cut point determination method, and a cut point determination program that can accurately determine cut points used for processing meat.

Brief Description of the Drawings

[0012]

Figure 1

Figure 2

Figure 3

Figure 4A

Figure 4B

Figure 4C

Figure 5

Figure 6A

Figure 6B

Figure 7A

Figure 7B

Figure 7C

Figure 8A

Figure 8B

Figure 9A

Figure 9B

Figure 9C

Figure 10

Figure 11A

Figure 11B

[0013] Hereinafter, some embodiments of the present invention will be described with reference to the accompanying drawings. However, the configurations described as embodiments or shown in the drawings are not intended to limit the scope of the present invention, but are merely illustrative examples.

[0014] First, referring to FIG. 1, the meat to be determined for the cut point by the cut point determination system 100 according to at least one embodiment of the present disclosure will be described. Hereinafter, as an example of meat, a pair of side meats 1 obtained by bisecting a meat carcass of a relatively large livestock such as a pig or a cow in half left and right at the position of the backbone (vertebra) 7 is shown. FIG. 1 is a schematic diagram showing a pair of side meats 1 according to one embodiment from the front.

[0015] The pair of branch muscles 1 includes a left branch muscle 1A and a right branch muscle 1B. Each of the pair of branch muscles 1 is positioned symmetrically about the central axis C with the anterior trunk 2a side downward (the posterior trunk 2c side upward), the inner side dorsal (the outer side thoracic), and in a posture where the cross-sectional plane 4 where the spine 7 and the sternum (ribs) 8 can be seen is on the front side. Such a pair of branch muscles is suspended and supported by gripping the leg part 5 located above the paper surface in FIG. 1 with a clamp mechanism (not shown).

[0016] The pair of branch muscles 1 suspended and supported in this way is subjected to a cutting process based on the cut point determined by the cut point determination system 100 described later. The cutting process includes, for example, a process of separating the anterior trunk 2a and the middle trunk 2b by cutting between the anterior trunk 2a and the middle trunk 2b based on a first cutting line L1 specified based on the determined cut point, and a process of separating the middle trunk 2b and the posterior trunk 2c by cutting between the middle trunk 2b and the posterior trunk 2c based on a second cutting line L2 specified based on the determined cut point.

[0017] Subsequently, a cut point determination system 100 for determining the cut point for the pair of branch muscles 1 having the above configuration will be described. FIG. 2 is a block diagram showing the schematic configuration of the cut point determination system 100 according to an embodiment, and FIG. 3 is a schematic diagram showing the layout of the imaging device 20 in FIG. 2 from above.

[0018] As shown in FIG. 2, the cut point determination system 100 is a system for determining the cut points of a pair of branch muscles 1 based on a captured image captured by the imaging device 20, and is composed of, for example, a CPU (Central Processing Unit), a RAM (Random Access Memory), a ROM (Read Only Memory), and a computer-readable storage medium, etc. A series of processes for realizing various functions are stored in a storage medium, etc. in the form of a program as an example. The CPU reads this program into the RAM, etc. and executes information processing and arithmetic processing to realize various functions. Note that the program may be in a form pre-installed in the ROM or other storage media, a form provided in a state stored in a computer-readable storage medium, a form distributed via wired or wireless communication means, etc. A computer-readable storage medium is a magnetic disk, a magneto-optical disk, a CD-ROM, a DVD-ROM, a semiconductor memory, etc.

[0019] As shown in FIG. 3, the imaging device 20 is a device such as a camera that can image a pair of branch muscles 1 by being arranged to face the cut surface 4 of the pair of branch muscles 1. The image acquired by the imaging device 20 may be a still image or a moving image. Also, the number of imaging devices 20 may be single or plural.

[0020] In FIG. 2, as a functional configuration of the cut point determination system 100, it includes an imaging image acquisition unit 102, a first image data acquisition unit 104, a position information estimation unit 106, a storage unit 108, a second image data acquisition unit 110, a validity determination unit 112, a position information correction unit 114, and a cut point determination unit 116.

[0021] The imaging image acquisition unit 102 is a configuration for acquiring the imaging image captured by the imaging device 20.

[0022] The first image data acquisition unit 104 is configured to acquire first image data G1 corresponding to a first region specified from the captured image acquired by the captured image acquisition unit 102. The first region is specified as at least a part of the captured image.

[0023] The position information estimation unit 106 is configured to estimate a plurality of pieces of position information based on the first image data G1. In particular, the position information estimation unit 106 estimates a plurality of pieces of position information from the first image data G1 using a first inference model M1 stored in advance in the storage unit 108.

[0024] The first inference model M1 is an inference model constructed by being pre-trained, for example, using a large number of teacher data in which captured images acquired as samples in the past are associated with correct answer data of the position information of the feature portions included in the captured images. The position information estimation unit 106 estimates a plurality of pieces of position information included in the first image data G1 by inputting the first image data G1 acquired by the first image data acquisition unit 104 into such a first inference model M1. Note that the first image data is image data created by cutting out a part of a captured image of any size, and is cut out, for example, as substantially square image data in which the number of pixels included in one side is a predetermined value.

[0025] Here, FIG. 4A is an example of the position information estimated for the entire body of each piece of meat 1 by the position information estimation unit 106 in FIG. 2, FIG. 4B is an example of the position information estimated by the position information estimation unit 106 in FIG. 2 from the front body 2a to the middle body 2b of each piece of meat 1, and FIG. 4C is an example of the position information estimated by the position information estimation unit 106 in FIG. 2 from the middle body 2b to the rear body 2c of each piece of meat 1.

[0026] As shown in FIG. 4A, for the entire body of each piece of meat 1, position information KPa of each point is obtained. In the present embodiment, the position information KPa is estimated as the lower end of the chest cavity (so-called lower end of the spare rib) included in the inner region of the contour R of the cut surface 4.

[0027] Also, as shown in FIG. 4B, position information KPb1 to KPb8 at eight points are estimated from the front trunk 2a to the middle trunk 2b of each branch muscle 1. In the present embodiment, the position information KPb1 is estimated as a position corresponding to the backbone 7 between the second rib and the third rib, the position information KPb2 is estimated as a position corresponding to the backbone 7 between the third rib and the fourth rib, the position information KPb3 is estimated as a position corresponding to the backbone 7 between the fourth rib and the fifth rib, and the position information KPb4 is estimated as a position corresponding to the backbone 7 between the fifth rib and the sixth rib. Also, the position information KPb5 is estimated as a position corresponding to the lower end of the spare rib, the position information KPb6 is estimated as a position corresponding to the sternum 8 between the third rib and the fourth rib, the position information KPb7 is estimated as a position corresponding to the sternum 8 between the fourth rib and the fifth rib, and the position information KPb8 is estimated as a position corresponding to the sternum 8 between the fifth rib and the sixth rib.

[0028] Also, as shown in FIG. 4C, position information KPc1 to KPc3 at three points are estimated from the middle trunk 2b to the rear trunk 2c of each branch muscle 1. In the present embodiment, the position information KPc1 is estimated as a position corresponding to the intervertebral disc at the upper end of the last lumbar vertebra t, the position information KPc2 is estimated as a position corresponding to the intervertebral disc at the lower end of the last lumbar vertebra t, and the position information KPc3 is estimated as a position corresponding to the intervertebral disc at the lower end of one vertebra below the last lumbar vertebra t.

[0029] Note that the first inference model M1 for obtaining the position information KPa shown in FIG. 4A, the first inference model M1 for obtaining the position information KPb1 to KPb8 shown in FIG. 4B, and the first inference model M1 for obtaining the position information KPc1 to KPc3 shown in FIG. 4C may be constructed as a common inference model or as separate inference models.

[0030] Returning to FIG. 2, the second image data acquisition unit 110 is configured to acquire second image data G2 corresponding to a second region specified as at least a part of the captured image acquired by the captured image acquisition unit 102. The second region is specified using at least a part of the plurality of pieces of position information estimated by the position information estimation unit 106.

[0031] The validity determination unit 112 is configured to determine the validity of a plurality of pieces of position information estimated by the position information estimation unit 106 based on the second image data G2 acquired by the second image data acquisition unit 110. The validity of the position information KPb3 to KPb7 (see FIG. 4B) used to identify the first cutting line L1 between the fourth rib and the fifth rib, and the position information KPc2 (see FIG. 4C) used to identify the second cutting line L2 is determined.

[0032] The validity determination of the position information in the validity determination unit 112 is performed using the second inference model M2 stored in advance in the storage unit 108. The second inference model M2 is an inference model constructed by, for example, being pre-trained using teacher data having an image indicating the validity of the estimation result as correct data. The validity determination unit 112 determines the validity of the position information by inputting the second image data G2 acquired by the second image data acquisition unit 110 into the second inference model M2 constructed in this way. The specific content of the validity determination of each piece of position information by the validity determination unit 112 will be described later.

[0033] The position information correction unit 114 is configured to perform correction processing on the position information estimated by the position information estimation unit 106 when it is determined by the validity determination unit 112 that the validity is not present. Thereby, when the position information is not estimated correctly, the position information can be corrected correctly by performing the correction processing. The specific content of the correction processing by the position information correction unit 114 will be described later.

[0034] The cut point determination unit 116 is a configuration for determining a cut point based on the determination result of the validity determination unit 112. When the validity determination unit 112 determines that there is validity, the position information estimated by the position information estimation unit 106 is directly determined as the final cut point. On the other hand, when the validity determination unit 112 determines that there is no validity, the position information to which the correction process is applied by the position information correction unit 114 is determined as the final cut point.

[0035] Subsequently, a cut point determination method implemented by the cut point determination system 100 having the above configuration will be described. FIG. 5 is a flowchart showing a cut point determination method according to an embodiment.

[0036] The imaging image acquisition unit 102 acquires an imaging image by imaging a pair of branch muscles 1 with the imaging device 20 (step S100). Subsequently, the first image data acquisition unit 104 acquires first image data G1 from the imaging image acquired in step S100 (step S101). The position information estimation unit 106 acquires the first inference model M1 from the storage unit 108 (step S102), and inputs the first image data G1 acquired in step S101 into the first inference model M1 acquired in step S102 to estimate a plurality of position information (step S103).

[0037] Subsequently, the second image data acquisition unit 110 acquires second image data G2 specified based on the position information estimated in step S103 (step S104). Subsequently, the validity determination unit 112 acquires the second inference model M2 from the storage unit 108 (step S105), and inputs the second image data G2 acquired in step S104 into the second inference model M2 acquired in step S105 to determine the validity of a plurality of position information (step S106).

[0038] If it is determined in the determination of step S106 that the position information is not valid (step S107: YES), correction processing is performed on the position information estimated in step S103 by the position information estimation unit 106 (step S108), and the position information on which the correction processing has been performed is determined as the final cut point (step S109). On the other hand, if it is determined that the position information is valid (step S108: NO), the position information estimated by the position information estimation unit 106 is directly determined as the final cut point (step S109).

[0039] Subsequently, a specific method for the validity determination unit 112 to determine the validity of the position information estimated by the position information estimation unit 106 will be described. First, a method for determining the validity of the position information KPb3 and KPb7 (see FIG. 4B) used to identify the first cut line L1 among the position information estimated by the position information estimation unit 106 will be described. The method for determining the validity of the position information KPb3 and KPb7 generally includes "bone bridging determination" and "interval detection determination". The bone bridging determination is a process of determining whether the first cut line L1 specified by the position information KPb3 and KPb7 crosses a rib. The interval detection determination is a process of determining whether the first cut line L1 specified by the position information KPb3 and KPb7 is between the fourth rib and the fifth rib as intended, and is further applied to the position information KPb3 and KPb7 whose validity has been confirmed in the bone bridging determination.

[0040] FIG. 6A is a schematic diagram showing the second image data G2 input to the second inference model M2 in the bone bridging determination of the validity determination unit 112 in FIG. 2, and FIG. 6B is a schematic diagram showing the classification process by the second inference model M2 in the bone bridging determination of the validity determination unit 112 in FIG. 2.

[0041] In the rib crossing determination, the second image data G2 input to the second inference model M2 is specified using the position information KPb3 and KPb7 estimated by the first inference model M1. Specifically, as shown in FIG. 6A, the second image data G2 is specified as image data having a frame FL passing through the position information KPb3 and KPb7 as a contour.

[0042] The second inference model M2 is constructed by machine learning using teacher data so as to classify the input second image data G2 into any of classes 0 to 2. The second inference model M2 is machine-learned using teacher data sufficiently including image data belonging to class 0 (no rib crossing), image data belonging to class 1 (rib crossing: upward movement of the sternum), and image data belonging to class 2 (rib crossing: downward movement of the sternum), respectively. As shown in FIG. 6B, when the second image data G2 specified using the position information KPb3 and KPb7 estimated by the position information estimation unit 106 is input to the second inference model M2 constructed in this way, the second image data G2 is classified into any of classes 0 to 2.

[0043] When classified into class 0 by the second inference model M2, the validity determination unit 112 determines that the first cutting line L1 specified by the position information KPb3 and KPb7 has no rib crossing. In this case, as shown in FIG. 7A, it is confirmed that the position information KPb3 and KPb7 are estimated without crossing the ribs.

[0044] When classified into class 1 by the second inference model M2, the validity determination unit 112 determines that the first cutting line L1 specified by the position information KPb3 and KPb7 has a rib crossing (upward movement of the sternum). In this case, as shown in FIG. 7B, the position information correction unit 114 performs correction processing so as to move the position of the position information KPb7 on the sternum side upward by a fixed amount Δx. Thereby, the first cutting line L1 specified by the position information KPb3 and the corrected position information KPb7 is corrected so as not to cross the ribs.

[0045] On the other hand, when classified into Class 2 by the second inference model M2, the validity determination unit 112 determines that the first cutting line L1 specified by the position information KPb3 and KPb7 has a rib crossing (substernal depression). In this case, as shown in FIG. 7C, the position information correction unit 114 performs correction processing to move the position of the position information KPb7 on the sternum side downward by a fixed amount Δx. As a result, the first cutting line L1 specified by the position information KPb3 and the corrected position information KPb7 is corrected so as not to cross the rib.

[0046] In addition, in the present embodiment, the position information correction unit 114 has exemplified the case of correcting the position information on the sternum side among the plurality of position information estimated by the position information estimation unit 106 based on the determination result of the validity determination unit 112. However, based on the same idea, the position information on the backbone side may be corrected.

[0047] Subsequently, FIG. 8A is a schematic diagram showing the second image data G2 input to the second inference model M2 in the intermediate detection determination of the validity determination unit 112 in FIG. 2, and FIG. 8B is a schematic diagram showing the classification process by the second inference model M2 in the intermediate detection determination of the validity determination unit 112 in FIG. 2.

[0048] In the intermediate detection determination, the second image data G2 input to the second inference model M2 is specified using the position information KPb3 and KPb7 for defining the first cutting line L1 among the position information estimated by the first inference model M1. Specifically, as shown in FIG. 8A, the second image data G2 is specified as image data having a frame FL2 passing through three points including the position information KPb3 and KPb7 and the position information KPb5 corresponding to the lower end of the thoracic cavity (so-called spare rib lower end) specified by machine learning in the same manner.

[0049] The second inference model M2 is constructed to classify the input second image data G2 into any one of classes 0 to 3 by being machine - learned using teacher data. The second inference model M2 is machine - learned using teacher data that sufficiently includes image data belonging to class 0, image data belonging to class 1, image data belonging to class 2, and image data belonging to class 3, respectively.

[0050] The training data (image data) corresponding to class 0 is prepared to pass through three points of position information KPb2, KPb5, and KPb6 as shown by frame FL1 in FIG. 8A (specifically, frame FL1 is prepared as a rectangle having a straight line connecting position information KPb2 and KPb6 as one side and including position information KPb2, KPb5, and KPb6). The training data (image data) corresponding to class 1 is prepared to pass through three points of position information KPb3, KPb5, and KPb7 as shown by frame FL2 in FIG. 8A (specifically, frame FL2 is prepared as a rectangle having a straight line connecting position information KPb3 and KPb7 as one side and including position information KPb3, KPb5, and KPb7). The training data (image data) corresponding to class 2 is prepared to pass through three points of position information KPb4, KPb5, and KPb8 as shown by frame FL3 in FIG. 8A (specifically, frame FL3 is prepared as a rectangle having a straight line connecting position information KPb4 and KPb8 as one side and including position information KPb4, KPb5, and KPb8). The training data (image data) corresponding to class 3 is prepared as an image obtained, for example, when imaging fails.

[0051] As shown in FIG. 8B, when the second image data G2 specified using the position information KPb3, KPb7, and KPb5 estimated by the position information estimation unit 106 is input to the thus - constructed second inference model M2, the second image data G2 is classified into any one of classes 0 to 3.

[0052] When classified into class 0 by the second inference model M2, the validity determination unit 112 determines that the first cutting line L1 specified by the position information KPb3 and KPb7 is between the third rib and the fourth rib. In this case, as shown in FIG. 9A, the position information correction unit 114 performs correction processing to move the positions of the position information KPb3 and KPb7 upward by a fixed amount Δx. Thereby, the first cutting line L1 specified by the position information KPb3 and the corrected position information KPb7 is corrected to be between the fourth rib and the fifth rib as it should be originally.

[0053] When classified into class 1 by the second inference model M2, the validity determination unit 112 determines that the first cutting line L1 specified by the position information KPb3 and KPb7 is between the fourth rib and the fifth rib. In this case, as shown in FIG. 9B, it is confirmed that the position information KPb3 and KPb7 are in the correct positions as they should be originally.

[0054] When classified into class 2 by the second inference model M2, the validity determination unit 112 determines that the first cutting line L1 specified by the position information KPb3 and KPb7 is between the fifth rib and the sixth rib. In this case, as shown in FIG. 9C, the position information correction unit 114 performs correction processing to move the positions of the position information KPb3 and KPb7 downward by a fixed amount Δx. Thereby, the first cutting line L1 specified by the position information KPb3 and the corrected position information KPb7 is corrected to be between the fourth rib and the fifth rib as it should be originally.

[0055] When classified into class 3 by the second inference model M2, the validity determination unit 112 makes a determination of unknown validity on the grounds that the second image data G2 capable of determining the validity of the position information has not been obtained.

[0056] Further, in the present embodiment, the position information correction unit 114 is exemplified as a case where the position information is corrected based on the determination result of the validity determination unit 112 so that the cutting line specified based on the plurality of pieces of position information estimated by the position information estimation unit 106 is between the fourth rib and the fifth rib. However, the position information may be corrected so that the cutting line specified based on the plurality of pieces of position information estimated by the position information estimation unit 106 is between the third rib and the fourth rib, or the position information may be corrected so that the cutting line specified based on the plurality of pieces of position information estimated by the position information estimation unit 106 is between the fifth rib and the sixth rib.

[0057] The determination result by such a validity determination unit 112 and the flow of the correction process performed by the position information correction unit 114 according to the determination result will be described again with reference to FIG. 10. FIG. 10 is a flowchart showing the validity determination by the validity determination unit 112 in FIG. 2 and the correction process for the position information by the position information correction unit 114.

[0058] First, the validity determination unit 112 determines whether or not the first cutting line L1 specified by the position information KPb3 and KPb7 straddles a rib (step S200). As a result, when it is determined that there is no rib crossing of the first cutting line L1 (step S200: YES), the validity determination unit 112 further performs an intermediate detection determination (step S201). In the intermediate detection determination, it is determined whether or not the first cutting line L1 specified by the position information KPb3 and KPb7 is located between the fourth rib and the fifth rib. As a result, when the first cutting line L1 is located between the fourth rib and the fifth rib (step S201: YES), since the position information KPb3 and KPb7 are correctly estimated by the position information estimation unit 106, the correction process by the position information correction unit 114 is not performed. On the other hand, when the first cutting line L1 is not located between the fourth rib and the fifth rib (step S201: NO), as described above with reference to FIG. 9A or FIG. 9C, the position information correction unit 114 performs a correction process so as to move the position information KPb3 and KPb7 estimated by the position information estimation unit 106 by a fixed amount Δx so as to be between the fourth rib and the fifth rib (step S202).

[0059] In addition, in rib crossing determination, when it is determined that the first cutting line L1 specified by the position information KPb3 and KPb7 crosses a rib (step S200: NO), the validity determination unit 112 further determines whether or not the rib crossing is a sternal elevation (step S203). When the rib crossing is a sternal elevation (step S203: YES), as described above with reference to FIG. 7B, the position information correction unit 114 performs correction processing to move the position information KPb7 on the sternum side of the position information KPb3 and KPb7 upward by a fixed amount Δx (step S204). Then, the validity determination unit 112 performs an inter-detection determination on the first cutting line L1 specified by the position information KPb3 and the corrected position information KPb7 (step S205). In the inter-detection determination, it is determined whether or not the first cutting line L1 specified by the position information KPb3 and the corrected position information KPb7 is located between the fourth rib and the fifth rib. As a result, when the first cutting line L1 is located between the fourth rib and the fifth rib (step S205: YES), it is confirmed that the position information KPb3 and KPb7 have been correctly corrected by the correction processing in step S204. On the other hand, when the first cutting line L1 is not located between the fourth rib and the fifth rib (step S205: NO), the position information correction unit 114 returns the position information KPb7 corrected in the correction processing of step S204 to its original position (step S206), that is, cancels the correction processing of step S204, and performs correction processing again to move the position information KPb3 on the backbone side downward by a fixed amount Δx (step S207).

[0060] On the other hand, when the bone bridging is a sub-sternal depression (step S203: NO), as described above with reference to FIG. 7C, the position information correction unit 114 performs correction processing to move the sternum-side position information KPb7 of the position information KPb3 and KPb7 downward by a fixed amount Δx (step S208). Then, the validity determination unit 112 performs an inter-detection determination on the first cutting line L1 specified by the position information KPb3 and the corrected position information KPb7 (step S209). In the inter-detection determination, it is determined whether or not the first cutting line L1 specified by the position information KPb3 and the corrected position information KPb7 is located between the fourth rib and the fifth rib. As a result, when the first cutting line L1 is located between the fourth rib and the fifth rib (step S209: YES), it is confirmed by the correction processing in step S204 that the position information KPb3 and KPb7 have been correctly corrected. On the other hand, when the first cutting line L1 is not located between the fourth rib and the fifth rib (step S209: NO), the position information correction unit 114 returns the position information KPb7 corrected by the correction processing in step S208 to its original position (step S210), that is, cancels the correction processing in step S208, and performs correction processing again to move the vertebral column-side position information KPb3 upward by a fixed amount Δx (step S211).

[0061] Subsequently, a method for determining the validity of the position information KPc2 (see FIG. 4C) used to specify the second cutting line L2 among the position information estimated by the position information estimation unit 106 will be described. FIG. 11A is a schematic diagram showing the second image data G2 input to the second inference model M2 in the validity determination of the validity determination unit 112 in FIG. 2, and FIG. 11B is a schematic diagram showing the classification process by the second inference model M2 in the validity determination of the validity determination unit 112 in FIG. 2.

[0062] In the validity determination of the position information KPc2, as shown in FIG. 11A, the second image data G2 input to the second inference model M2 is specified as image data having a frame FL2 passing through the position information KPc2 for specifying the second cutting line L2 among the position information estimated by the first inference model M1 and a reference point Pref set in advance as a fixed value as a contour.

[0063] The second inference model M2 is constructed to classify the input second image data G2 into any one of classes 0 to 3 by being machine-learned using teacher data. The second inference model M2 is machine-learned using teacher data that sufficiently includes image data belonging to class 0, image data belonging to class 1, image data belonging to class 2, and image data belonging to class 3, respectively. The learning data (image data) corresponding to class 0 is prepared to pass through the reference point Pref and the position information KPc1 as shown by the frame FL1 in FIG. 11A.

[0064] The learning data (image data) corresponding to class 1 is prepared to pass through the reference point Pref and the position information KPc2 as shown by the frame FL2 in FIG. 11A. The learning data (image data) corresponding to class 2 is prepared to pass through the reference point Pref and the position information KPc3 as shown by the frame FL3 in FIG. 11A. The learning data (image data) corresponding to class 3 is prepared as, for example, a noise image obtained when imaging fails.

[0065] As shown in FIG. 11B, when the second image data G2 specified using the position information KPc2 estimated by the position information estimation unit 106 and the reference point Pref is input to the second inference model M2 constructed in this way, the second image data G2 is classified into any one of classes 0 to 3.

[0066] When classified into class 1 by the second inference model M2, the validity determination unit 112 determines that the estimation result of the position information KPc2 for specifying the second cutting line L2 is valid. On the other hand, when classified into class 0 or 2 by the second inference model M2, the validity determination unit 112 determines that the estimation result of the position information KPc2 for specifying the second cutting line L2 is not valid. Further, when classified into class 3 by the second inference model M2, the validity determination unit 112 makes a determination of unknown validity, assuming that the second image data G2 capable of determining the validity of the position information KPc2 has not been obtained.

[0067] As described above, according to each of the above embodiments, by inputting the first image data corresponding to the first region of the captured image of meat into the first inference model, a plurality of position information included in the first image data is estimated. The position information estimated in this way is determined to be valid based on the second image data corresponding to the second region of the captured image, so that the cut point based on the position information can be accurately determined.

[0068] In addition, within the scope not departing from the gist of the present disclosure, it is possible to appropriately replace the components in the above-described embodiments with well-known components, and the above-described embodiments may also be appropriately combined.

[0069] The content described in each of the above embodiments is understood as follows, for example.

[0070] (1) A cut point determination system for meat according to one aspect includes an image data acquisition unit configured to acquire first image data corresponding to a first region of a captured image of meat, a position information estimation unit that inputs the first image data acquired by the image data acquisition unit into a first inference model to estimate a plurality of position information, a validity determination unit that determines the validity of the estimation result of the position information estimation unit based on second image data corresponding to a second region of the captured image specified using at least a part of the plurality of position information, a cut point determination unit that determines a cut point based on the determination result of the validity determination unit, and is provided with.

[0071] According to the aspect of (1) above, by inputting the first image data corresponding to the first region of the captured image of meat into the first inference model, a plurality of position information included in the first image data is estimated. The position information estimated in this way is determined to be valid based on the second image data corresponding to the second region of the captured image, so that the cut point based on the position information can be accurately determined. The first inference model is, for example, an inference model constructed by machine learning using an imaging image and correct data indicating the position information of the characteristic part of the meat contained in the imaging image as teacher data.

[0072] (2) In another aspect, in the aspect of (1) above, The validity determination unit By inputting the second image data specified using the position information estimated by the position information estimation unit into a second inference model, the validity of the estimation result by the first inference model is determined.

[0073] According to the aspect of (2) above, the validity of the position information estimated by the first inference model can be suitably determined by being input into the second inference model. The second inference model is, for example, an inference model constructed by machine learning using correct image data indicating the validity of the inference result by the first inference model as teacher data.

[0074] (3) In another aspect, in the aspect of (2) above, The second inference model is constructed to classify the input second image data into any one of a plurality of classes prepared corresponding to the determination criterion of the validity, The validity determination unit determines the validity based on whether the classification result of the second image data based on the second inference model belongs to a class specified in advance.

[0075] According to the aspect of (3) above, the validity of the position information estimated by the first inference model can be suitably determined based on the classification result by the second inference model.

[0076] (4) In another aspect, in the aspect of (3) above, The plurality of classes are specified by frames including at least a part of the position information in the second region, and are learned based on a plurality of image data prepared to overlap with each other.

[0077] According to the aspect (4) above, by machine learning using such a plurality of image data as teacher data, a second inference model that can be classified into a plurality of classes can be suitably constructed to determine the validity of the position information.

[0078] (5) In another aspect, in any one of the aspects (1) to (4) above, based on the determination result of the validity determination unit, it further includes a position information correction unit for correcting the position information estimated by the position information estimation unit.

[0079] According to the aspect (5) above, based on the determination result of the validity, by correcting the position information estimated by the first inference model, it is possible to determine a cut point with higher accuracy.

[0080] (6) In another aspect, in the aspect (5) above, when the validity determination unit determines that the validity exists, the cut point determination unit determines the cut point based on the position information estimated by the position information estimation unit.

[0081] According to the aspect (6) above, when it is determined that the position information estimated by the position information estimation unit is valid, the cut point is determined based on the position information.

[0082] (7) In another aspect, in the aspect (5) or (6) above, when the validity determination unit determines that the validity does not exist, the cut point determination unit determines the cut point based on the position information corrected by the position information correction unit.

[0083] According to the aspect (7) above, when it is determined that the position information estimated by the position information estimation unit is not valid, the cut point is determined based on the position information corrected by the position information correction unit.

[0084] (8) In another aspect, in any one of the aspects (1) to (7) above, the validity determination unit determines the validity based on a bone bridging determination of determining whether a line specified using the position information estimated by the position information estimation unit crosses a rib included in the meat.

[0085] According to the aspect (8) above, the validity of the position information estimated by the first inference model can be determined from the viewpoint of whether a line (for example, a cutting line or the like) specified using this position information crosses a bone.

[0086] (9) In another aspect, in the aspect (8) above, further includes a position information correction unit for correcting the position information estimated by the position information estimation unit based on the determination result of the validity determination unit, when the bone bridging determination determines that the line crosses the rib, the position information correction unit corrects the position information so as to move the position information on either the sternum side or the backbone side of the meat among the position information estimated by the position information estimation unit.

[0087] According to the aspect (9) above, in the above-mentioned bone bridging determination, when it is determined that a line (for example, a cutting line or the like) specified using the position information estimated by the first inference model crosses a bone, by performing a correction process on the position information so as to move the position information on either the sternum side or the backbone side of the meat, the estimation accuracy of the cut point can be effectively improved.

[0088] (10) In another aspect, in any one of the aspects (1) to (9) above, the validity determination unit determines the validity based on an intercostal detection determination of determining whether a line specified using the position information estimated by the position information estimation unit is between pre-specified ribs included in the meat.

[0089] According to the aspect (10) above, the determination of the validity of the position information estimated by the first inference model can be made from the perspective of whether a line (such as a cutting line etc.) specified using this position information is between the ribs specified in advance.

[0090] (11) In another aspect, in the aspect (10) above, based on the determination result of the validity determination unit, it further includes a position information correction unit for correcting the position information estimated by the position information estimation unit, when the line is determined not to be between the ribs by the inter-inspection determination, the position information correction unit corrects the position information among the position information estimated by the position information estimation unit so that the line approaches between the ribs.

[0091] According to the aspect (11) above, in the aforementioned inter-inspection determination, when it is determined that a line (such as a cutting line etc.) specified using the position information estimated by the first inference model is not between the ribs specified in advance, by performing correction processing on the position information so that the line approaches between the specified ribs, the estimation accuracy of the cut point can be effectively improved.

[0092] (12) A cut point determination method for meat according to one aspect is a step of obtaining first image data corresponding to a first region of a captured image of meat, a step of inputting the first image data into a first inference model to estimate a plurality of position information, a step of determining the validity of the estimation results of the plurality of position information based on second image data corresponding to a second region of the captured image specified using at least a part of the plurality of position information, a step of determining a cut point based on the determination result of the validity, and includes.

[0093] According to the aspect of (12) above, by inputting the first image data corresponding to the first region of the captured image of meat into the first inference model, a plurality of position information included in the first image data is estimated. The position information estimated in this way is determined to be valid based on the second image data corresponding to the second region of the captured image, so that the cut point can be accurately determined based on the position information.

[0094] (13) A cut point determination program for meat according to one aspect is to a computer device, a step of acquiring first image data corresponding to a first region of a captured image of meat; a step of inputting the first image data into the first inference model to estimate a plurality of position information; a step of determining the validity of the estimation result of the plurality of position information based on the second image data corresponding to the second region of the captured image specified by using at least a part of the plurality of position information; a step of determining a cut point based on the determination result of the validity; and is executable.

[0095] According to the aspect of (13) above, by inputting the first image data corresponding to the first region of the captured image of meat into the first inference model, a plurality of position information included in the first image data is estimated. The position information estimated in this way is determined to be valid based on the second image data corresponding to the second region of the captured image, so that the cut point can be accurately determined based on the position information.

Explanation of Reference Numerals

[0096] 1 Loin 1A Left loin 1B Right loin 2a Forequarter 2b Midquarter 2c Hindquarter 4 Cutting surface 5 Leg 7 Backbone 8 Sternum (ribs) 20 Imaging device 100 Cut point determination system 102 Imaging image acquisition unit 104 First image data acquisition unit 106 Position information estimation unit 108 Memory unit 110 Second image data acquisition unit 112 Validity determination unit 114 Position information correction unit 116 Cut point determination unit C Central axis G1 First image data G2 Second image data

Claims

1. An image data acquisition unit configured to acquire first image data corresponding to a first region of a captured image of meat; A position information estimation unit that inputs the first image data acquired by the image data acquisition unit into a first inference model to estimate a plurality of position information; A validity determination unit that determines the validity of the estimation result in the position information estimation unit based on second image data corresponding to a second region of the captured image specified using at least a part of the plurality of position information; A cut point determination unit that determines a cut point based on the determination result of the validity determination unit; A cut point determination system for meat, comprising:

2. The validity determination unit: The cut point determination system for meat according to claim 1, wherein the validity of the estimation result by the first inference model is determined by inputting the second image data specified using the position information estimated by the position information estimation unit into a second inference model.

3. The second inference model is constructed to classify the input second image data into any one of a plurality of classes prepared corresponding to the determination criteria of the validity, The validity determination unit determines the validity based on whether or not the classification result of the second image data based on the second inference model belongs to a class specified in advance. The cut point determination system for meat according to claim 2.

4. The plurality of classes are specified by a frame including at least a part of the position information in the second region, and are learned based on a plurality of image data prepared to overlap each other. The cut point determination system for meat according to claim 3.

5. The cut point determination system for meat according to claim 1 or 2, further comprising a position information correction unit for correcting the position information estimated by the position information estimation unit based on the determination result of the validity determination unit.

6. The cut point determination unit determines the cut point based on the position information estimated by the position information estimation unit when the validity determination unit determines that the validity is present. The cut point determination system for meat according to claim 5.

7. The cut point determination unit determines the cut point based on the position information corrected by the position information correction unit when the validity determination unit determines that the validity is not present, according to the cut point determination system for meat as described in claim 5.

8. The validity determination unit determines the validity based on a rib crossing determination for determining whether a line specified using the position information estimated by the position information estimation unit crosses a rib included in the meat, according to the cut point determination system for meat as described in claim 1 or 2.

9. Based on the determination result of the validity determination unit, it further includes a position information correction unit for correcting the position information estimated by the position information estimation unit. When the rib crossing determination determines that the line crosses the rib, the position information correction unit corrects the position information so as to move the position information on either the sternum side or the backbone side of the meat among the position information estimated by the position information estimation unit, according to the cut point determination system for meat as described in claim 8.

10. The validity determination unit determines the validity based on an intercostal detection determination for determining whether a line specified using the position information estimated by the position information estimation unit is between pre-specified ribs included in the meat, according to the cut point determination system for meat as described in claim 1 or 2.

11. Based on the determination result of the validity determination unit, it further includes a position information correction unit for correcting the position information estimated by the position information estimation unit. When the intercostal detection determination determines that the line is not between the ribs, the position information correction unit corrects the position information so that the line approaches between the ribs among the position information estimated by the position information estimation unit, according to the cut point determination system for meat as described in claim 10.

12. A step of acquiring first image data corresponding to a first region of an imaging image of meat. A step of inputting the first image data into a first inference model to estimate a plurality of pieces of position information. A step of determining the validity of the estimation results of the plurality of pieces of position information based on second image data corresponding to a second region of the imaging image specified using at least a part of the plurality of pieces of position information. A step of determining a cut point based on the determination result of the validity. A method for determining a cutting point for meat, comprising...

13. In a computer device, a step of obtaining first image data corresponding to a first region of a captured image of meat; a step of inputting the first image data into a first inference model to estimate a plurality of position information; a step of determining the validity of the estimation result of the plurality of position information based on second image data corresponding to a second region of the captured image specified using at least a part of the plurality of position information; a step of determining a cutting point based on the determination result of the validity; A cutting point determination program for meat that can execute the above steps.

Images