Meat identification device
The meat discrimination device uses image processing and deep learning to accurately identify beef cuts, correcting errors and adapting to imaging changes, enhancing processing efficiency.
Patent Information
- Application Number
- JP2020045158
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- Filing Date
- 2020-03-16
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2040-03-16
AI Technical Summary
Existing methods struggle to accurately identify beef meat parts using X-ray imaging due to variations in thickness and shading, making it difficult for automated systems to distinguish between different cuts with high accuracy.
A meat discrimination device utilizing a camera to capture images, a trained model for deep learning, and a control unit to determine meat parts by excluding previously identified parts, with features to correct errors and adapt to changes in imaging conditions.
The device achieves high-accuracy meat part identification, streamlining operations by automating tasks that were previously difficult for human workers, reducing errors, and improving efficiency in meat processing.
Smart Images

Figure 0007583986000001 
Figure 0007583986000002 
Figure 0007583986000003
Abstract
Description
[Technical field]
[0001] The disclosed technology, for example, cuts a beef carcass into portions and processes each portion using image processing technology. Therefore, this relates to an automated technology that determines which part of meat is being cut and notifies the subsequent process. [Background technology]
[0002] For example, the left and right carcasses cut from a cow are divided into 13 parts in the cut meat processing room: neck, shoulder, shoulder loin, shoulder belly, fillet, rib eye, sirloin, inner thigh, lower thigh, rib eye, outer thigh, and shank. Each part may be further divided according to the client's request (see Patent Document 2). Each cut meat is vacuum packed, cooled in a chiller or similar device, and then transferred to the measurement and printing line.
[0003] On the weighing and printing line, the worker weighs the cut of meat, then enters the name of the cut, The label on which the part of meat, its weight, production date, processor, individual identification number, etc. are printed is A label is then printed on the meat and attached to the meat portion (see Patent Document 1). [Prior art documents] [Patent documents]
[0004] [Patent Document 1] Special Publication No. 6-55617 [Patent Document 2] Patent No. 4020659 [Patent Document 3] JP 2004-45072 A Summary of the Invention [Problem to be solved by the invention]
[0005] However, the cuts of meat conveyed to the weighing and printing line have similar external shapes. Unless you are an experienced worker, you will not be able to immediately tell which part of meat is which. A known example of an automated method is described in Patent Document 3.
[0006] In this patent document 3, meat (chicken meat) is irradiated with radiation, and the reflected light and transmitted light are measured. The system captures images using separate cameras, and comprehensively identifies which part of the meat is contained in each image. This method makes it easier to tell which part of the chicken it is, especially when it is thin. This can be easily distinguished by image analysis.
[0007] However, even a small cut of beef can weigh several kilograms, and a large cut can weigh more than ten kilograms. However, the thickness of the material varies depending on the orientation, and the shade and shape of the X-ray image also change significantly. Therefore, the method of Patent Document 3 has a problem that it is difficult to identify the part of the meat. There is. The present invention is able to distinguish cuts of beef, which are difficult to distinguish using X-ray images, with the same accuracy as humans. To provide a new meat discriminating device capable of automatically discriminating the part of meat with high accuracy. The objective is to: [Means for solving the problem]
[0008] The meat discrimination device according to the present invention comprises: Disassembled down to the individual parts A camera that captures images of meat parts and acquires image information; The image information is input, and the part of the meat is First name A trained model that outputs candidates; The part output from the trained model First name candidate From among them, The part of meat photographed by the camera given name A meat discrimination device comprising: And, A memory unit that stores the part names of each meat cut from the carcass in units of the carcass, The control unit, when determining the part name of the meat, determines the part name from among the part names of the carcass stored in the memory unit, excluding the part names already determined. It is characterized by:
[0009] Here, meat cut from a cow's carcass is used as an example of meat. For example, meat such as pork or chicken may be used. Dell uses image information of meat as training data and a system that indicates which part of meat each part of meat is. This is a deep learning process in which the correct answer data obtained by the training is applied to a multi-layer neural network. Therefore, when image information of a portion of meat is input into this trained model, candidate parts of the portion of meat estimated from that image information are output. For example, the probabilities of part A are output as 75%, part B as 15%, and part C as 10%. If the estimated probability of the output candidate parts exceeds a predetermined threshold, for example 75%, the control unit outputs the part with the probability exceeding this threshold as the discrimination result. If the probability does not exceed this threshold, the control unit outputs the candidate part as it is as the discrimination result. In this case, the operator can simply specify the part that he or she thinks is most suitable from the output candidate parts.
[0010] The control unit extracts a region of meat from the image information and learns the image information in the extracted region. The part of the meat is identified by inputting the data into the model. do.
[0011] The image information captured by the camera contains various background information. The meat region is extracted from the image data, and the image information of the extracted region is input to the trained model. do.
[0012] The control unit extracts The meat If there are multiple regions, input the image information of each region into the trained model, The above Meat parts by region given name Output candidates for do.
[0013] The cuts of meat are transported to the weighing and printing line in sequence, and the camera records them. Multiple cuts of meat (meat) Shoot If the control unit detects one or more meats in the output image information, it extracts the area of the detected meat, inputs the image information of the extracted area to the trained model, and calculates the part of each meat. First name Output the candidates.
[0014] child Such a meat discrimination device further comprises a memory unit which stores each part of meat cut from one carcass on a carcass unit basis, and the control unit, when discriminating the part of meat, takes into consideration whether or not the part is a part that has already been discriminated within the carcass.
[0015] Since beef cuts are cut into carcass units, once the cut processing of one carcass is completed, The same parts are not cut from the same carcass. Therefore, the downstream meat discriminator The parts of each meat (cut meat) cut from the carcass are stored in the memory unit on a carcass unit basis, and the control When the department determines the part of meat, it will consider whether the part has already been determined and will select the remaining parts. Select the body part candidate with the highest probability from the top 10 candidates. In this case, the body part candidates output from the trained model are However, if it is not among the remaining parts, the part that was identified earlier may have been incorrect, or the part that was estimated later may have been incorrect. Either the part candidate is incorrect or the part candidate is incorrect. When this happens, the system notifies the user. Mistakes can be corrected before the item is packed. Also, the trained model can be used to correct mistakes made by workers. By retraining the data as additional training data, the discrimination accuracy can be improved. In addition, if a part is divided into smaller parts at the request of the customer, This can be handled by registering in advance that the digits have been divided into multiple parts.
[0016] The control unit is a trained model When acquiring the photographic images used to generate theThe photographing conditions are stored, and the stored photographing conditions are used by the camera. Capture the image again If the photographing conditions differ from those at the time of photographing, the fact is notified.
[0017] In this case, the shooting conditions are, for example, the brightness, contrast, and color temperature of the subject photographed by the camera. If these parameters change, such as the degree of convergence, the results of the trained model may change. For example, changing the color temperature of a lighting fixture may cause the camera to The shooting conditions of the initial image taken by the camera and the shooting conditions stored when the trained model was obtained are If they are different, the control unit notifies the user that the photographing conditions have changed.
[0018] The control unit stores an initial image taken by the camera under normal conditions, and the stored initial image When the image differs from the initial image when the camera is started, the fact is notified. do.
[0019] The characteristics of the cut meat appear on the cross section and on the ventral side, but the orientation of the cut meat being transported Since the meat is not uniform, it is arranged so that multiple cameras can photograph the meat from multiple angles. In such a case, for example, a camera that takes pictures of meat from a horizontal direction should be placed on the operator. If the camera comes into contact with the food inadvertently, it may change its orientation, making it difficult to identify the part of the meat. In such cases, it can take a lot of time to discover the cause of the problem. Therefore, the image including the background captured by the camera under normal conditions is stored, and the stored initial image is the initial image when the camera is started. Background included in If the difference is not the same as the above, it is determined that the camera has been displaced and a notification is issued to that effect. This prevents the device from becoming unable to make any distinctions.
[0020] The apparatus is equipped with the above meat discrimination device and a measuring and printing device that inputs the part name from the meat discrimination device. The measurement and printing system includes a measurement and printing device that displays the meat photographed by the camera. a display unit that displays the meat image determined by the control unit on the displayed meat image, It is characterized by the display of the names of meat parts in layers.
[0021] When multiple meats are lined up on the weighing and printing line, the meats are weighed in order starting from the top. In some cases, meat from specified parts is extracted from the line in response to requests from clients. In the latter case, meat was selected from the meat piled up on the line. However, this invention displays the information on the display. The names of each part of the meat are displayed over the image of the meat. By looking at the label, you can instantly decide which meat to pick.
[0022] The measurement and printing device includes a measurement unit that measures the weight of the loaded meat and a control unit. is the meat displayed on the display unit The name of the meat part in question is displayed over the image. And, new meat is loaded onto the weighing section. When the meat displayed on the display unit disappears from the display unit, The part of the meat given name The part of meat loaded on the weighing section given name The present invention is characterized in that the above-mentioned is set as follows.
[0023] The meat lined up on the weighing and printing line is photographed by a camera and the image is displayed on the display. In that state, one of the meats displayed was 、 It is removed from the line and transferred to the weighing section. Can And the control unit , take Parts of meat that have been removed and therefore disappeared from the display screen given name The part of meat loaded on the weighing section given nameThis eliminates the need to set and register the part of the meat to be loaded each time it is weighed. As a pick-up target If there is a mismatch between the pre-specified part of the meat and the part of the meat loaded on the weighing section, the control section can sound an alarm to prompt a re-inspection.
[0024] The measurement and printing device further includes a printer for printing the attribute information of the meat. It is characterized by:
[0025] The attribute information here includes, for example, the name of the cut of meat (cut name), the name of the delivery destination, the individual identification number, Branch number, processing date, net weight, etc. Also, if necessary, left / right branch classification, storage temperature, The name of the processor, etc. may be included. Effect of the Invention
[0026] According to the present invention, even if the part of meat is difficult to distinguish from the part of meat in an X-ray radiographic image, the external characteristics of the meat can be easily distinguished. Therefore, until now, experienced workers have been By mechanizing tasks that would otherwise be difficult to distinguish, shipping operations can be streamlined. [Brief description of the drawings]
[0027] [Figure 1] FIG. 2 is an explanatory diagram showing an overview of the work process from cutting meat portions from a carcass to storing them. [Diagram 2] FIG. 1 is an explanatory diagram showing a schematic configuration of a measurement printing system. [Diagram 3] FIG. 4 is an explanatory diagram showing an example of a display screen of the display unit in a standby mode. [Figure 4] FIG. 13 is a diagram showing an example of a display screen of the display unit in a measurement printing mode. [Diagram 5] FIG. 2 is a diagram showing an example of a label attached to meat. [Figure 6] FIG. 13 is an explanatory diagram showing an example of the stored contents of part names cut from one carcass. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0028] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. The above is an example for illustrating the present invention, and is not intended to limit the technical scope of the present invention.
[0029] First, the outline of the process from cutting the beef from the carcass to storing it is based on the process diagram in Figure 1. In Figure 1, a slaughtered cow is split into left and right carcasses. Then, each carcass is transported from the refrigerator to the meat section while hanging on a hanging rail. In the processing room, a single carcass is divided into the front, loin, front and back, thigh, After being cut into large pieces, it is further divided, deboned and cut into 13 pieces of meat. The cuts are vacuum packed, inspected for metals with a metal detector, and then shrink-wrapped. The cooled cuts of meat are sent to the weighing and printing line. The cut of meat is then identified and its part name and weight are printed on the cut of meat. The processed meat is then packaged according to the customer's requirements. It is packed, packaged and stored in a freezer.
[0030] FIG. 2 is an explanatory diagram showing an overview of the measurement and printing line. In this figure, In the first step, the meat (hereinafter referred to as "cut meat M") sent out from the cooling device is conveyed to the measuring and printing device 1. The conveyor C conveying the meat and the cut meat M on the conveyor C are photographed, and the cut meat M is identified from the obtained cut meat image. A meat discriminating device 2 is provided to discriminate the part of the meat M and output the result to the measuring and printing device 1. It is being done.
[0031] Conveyor C is composed of an achromatic conveyor belt with little reflected light. A roller conveyor RC is used for the conveying path that is not visible to the camera 20 installed above. do.
[0032] The measuring and printing device 1 includes a measuring unit 10 for measuring the cut of meat M, a display unit 11, and a printing unit 12. and a control unit 13 for controlling them.
[0033] The weighing unit 10 detects the mass of the cut meat M loaded on the weighing plate and controls it as a digital signal. The weight is output to the weighing unit 13. Also, a weighing conveyor equipped with a roller conveyor may be used instead of the weighing pan. The display unit 11 is configured with a touch panel, and the control unit 13 is a microcomputer. By running the installed programs, the following functions can be realized: do.
[0034] That is, in the standby mode, the control unit 13 controls the part of the conveyor C photographed by the camera 20. The meat M is displayed on the display unit 11 as it is. When the part name of the meat M is input, the control unit 13 displays the name of the part M on the display unit 11 as shown in FIG. The candidate part of each portion of meat m is displayed superimposed on the image of each portion of meat m. The name of the part to be weighed and printed is displayed on the top of the display screen. Look at the screen and touch the specified part name on the display screen, or The part of meat M is picked up and placed on the weighing plate. When it is determined that the portion m of meat has been loaded onto the weighing pan, the display on the display unit 11 is changed as shown in FIG. Switch the screen to the measurement print mode screen.
[0035] In the measurement print mode, at least the part name of the meat M and the part meat measured by the measuring unit 10 are printed. The weight of M, the individual identification number, and other printed items are displayed on the display unit 11. It is.
[0036] The printing unit 12 is composed of a label printer and a journal printer. , and issues labels to be affixed to the shrink-wrapped meat portion M. Meanwhile, the journal print The printer prints the contents printed on the label onto the receipt paper. An example of a label is shown below. This label includes the name of the cut of meat M (part name), individual identification number, processing The date, branch number, left / right classification, storage temperature, name of the processor, etc. are printed on it.
[0037] The meat discriminating device 2 is composed of a camera 20, a control unit 24, and a storage unit 25. The control unit 24 has an image processing unit 21, a trained model 22, and a discrimination unit 23. Each of the above will be described in detail below.
[0038] The camera 20 photographs the portion of meat M and outputs image information. The camera 20 is, for example, a high-definition digital camera, and captures a plurality of images of the objects arranged on the conveyor C. The camera 20 captures the image of the cut of meat M on the conveyor C and outputs the image information. The camera is installed at a position and height that allows it to take pictures of a number of pieces of meat M. Although only the camera 20 that takes a picture from the side is shown in the figure, the cross section of the partial meat M can also be taken from the side. A camera (not shown) is also installed to take a picture of the meat part M from above. The image of the camera 20 is input as is to the display unit 11 of the weighing and printing device 1, and as shown in FIG. The arrangement of the cuts of meat M on the conveyor C is displayed as it is.
[0039] The camera 20 further captures the conveyance of the cut meat M on the conveying surface of the conveyor C. The photographing direction is adjusted so that the entire width in the direction perpendicular to the conveyor can be photographed. No matter where the meat part M is located in C, it can be photographed. The camera 20 is oriented in such a way that it can capture the cut of meat M on the roller conveyor RC. In this way, the image information obtained by the camera 20 is used to Therefore, the timing when the cut meat M is delivered to the end of the conveyor C can be output to the control unit 24. By the time the part is reached, the part candidates can be reliably displayed on the display unit 11, improving the work efficiency of the worker. You can do it.
[0040] The control unit 24 is composed of a microcomputer and executes a program stored in the memory unit 25. By executing the above, the functions of the image processing unit 21 and the discrimination unit 23 are realized. In FIG. 2, for convenience of explanation, the trained model 22, the control unit 24, and the storage unit 25 are shown in the weighing and printing device 1. Although these are shown separately, they are programmed into the computer in the weighing and printing device 1. The storage unit 25 also stores the shooting conditions described below and the time when the camera 20 is started. The initial image of the image is stored.
[0041] The control unit 24 further stores the photographing conditions when the trained model 22 is obtained in the storage unit 25. If the photographing conditions stored in the memory are different from those when the camera 20 was started, The device is programmed to notify you of this.
[0042] In this case, the photographing conditions are, for example, the brightness, contrast, and If these parameters change, such as color temperature, the results of the estimation using the trained model 22 may change. The shooting conditions of the initial image captured by the camera 20 and the learned model are If the photographing conditions are different from those stored when the filter 22 is obtained, the control unit 24 The device is programmed to output any changes on the display unit 11 to alert the user.
[0043] In addition, the control unit 24 stores the initial image taken by the camera 20 in a normal state in the storage unit 25. If the stored initial image is different from the initial image when the camera 20 was started, The camera is programmed to output the information to the display unit 11 and notify the user. Even if the orientation of 20 changes, the worker can be notified.
[0044] The image processing unit 21 included in the control unit 24 receives image information output from the camera 20. , image information of the partial meat is extracted from the input image information and output. That is, image information captured by the camera 20 is sequentially input to the image processing unit 21. The image processing unit 21 performs image processing on the input image and inputs it to the trained model. Generates image information for the purpose of
[0045] Specifically, the image processing unit 21 extracts only the image information of the portion m from which the background in the image has been removed. For example, the image processor 21 detects a change in shading as an edge and extracts the region of the partial meat m. Alternatively, instead of this extraction method, the image processing unit 21 may binarize the image using a predetermined threshold value. In addition, the area of the partial meat may be extracted.
[0046] As another process, the image processing unit 21 removes image information in which the partial meat m is not reflected. When there are a plurality of regions of the portion meat m, the image processing unit 21 further performs labeling processing. Therefore, a number is assigned to each region, and the image information of the portion m of each region is linked to the assigned number. Then, for each linked number, image information of the meat portion m is input to the trained model 22.
[0047] The trained model 22 receives image information of a portion of meat output from the image processing unit 21, A candidate part of the cut of meat is output.
[0048] Specifically, the trained model 22 includes a plurality of image information pieces of each meat part m that constitutes training data, Applying the correct answer data indicating the location of each meat part m to a multi-layer neural network for deep learning Therefore, in advance, in order to obtain the above training data, In order to do this, the meat portions M of each part are lined up on a conveyor C and photographed, and a large number of images of each meat portion M are obtained. The image information is stored in the storage unit 25 to create teacher data. You can flip the stored image information, shift the position, change the contrast, You can also increase the training data by zooming and transforming the image. When the image information of the portion m processed by the image processing unit 21 is input to the model 22, a trained model 22 outputs a candidate part estimated from an image of the meat part m. When image information is input, the trained model 22 calculates the following for each part m of each region: For example, as shown in Figure 3, the part of meat in the upper left area of the screen is estimated and output. m has an 80% chance of being "Shintama", a 10% chance of being "Ranichi", and a 10% chance of being "Nekku". Output as 10% etc.
[0049] This trained model 22 is configured as a program, and is therefore transmitted via a communication means (not shown). Upload it to an external cloud computing service and download it when you want to use it. The software may be installed in the control unit 24. Alternatively, an FPGA may be used instead of the software. It can also be configured with logic circuits such as Field Programmable Gate Array (FPGA). When using a multi-layer neural network, CNN (Convolutional Neural Network) is an open source product that offers "caffe", "TensorFlow", You can also use "Chainer", "CNTK", etc.
[0050] The trained model 22 may be configured to be created for each type of cut meat M. When creating a model for each type of meat part M, a trained model is constructed for all meat parts M. For example, if meat from the same part of a carcass is processed, the probability of the estimation is improved. Prepare multiple Ms and feed them to the downstream measurement and printing line to learn about cuts of meat from the same part. The trained model 22 for each part thus obtained is stored and The trained model 22 may be switched. In addition, the trained model 22 is configured as a model capable of discriminating the types of all meat parts M. It's okay.
[0051] The discrimination unit 23 discriminates the part captured by the camera 20 based on the part candidates output from the trained model 22. The part of the shaded meat M is identified. Specifically, the discrimination unit 23 determines whether the probability of the part candidate output from the trained model 22 is a predetermined If the threshold value, for example 75%, is exceeded, the part is output as the discrimination result. If the limit is not exceeded, or if it is preferable to let the operator select the limit, the following is used as shown in Figure 3. As shown in FIG. 3, the region candidates are output to the display unit 11 as the discrimination results. From the displayed cuts of meat m, the worker touches the cut of meat m that he or she thinks is the right one to select. Then, the information is fed back to the discrimination unit 23, which then determines: A check is made to see whether a site with a high estimated probability has been selected.
[0052] In the memory unit 25, as shown in FIG. 6, the part names of each cut of meat cut from one carcass are stored. The meat is stored in units of carcasses. The part of the cut of meat identified by the identification unit 23 or the surface The part of the meat m designated on the display unit 11 is marked as having been identified. When a new cut meat part candidate is output from the finished model 22, the discrimination unit 23 The estimated probability of the area is stored in the discrimination area of FIG. 6, and the area that has already been discriminated is selected from the area. "Rib Loin" is the most likely choice out of "Rib Loin" and "Sirloin" excluding "Chuck Loin" Select and output.
[0053] Next, a method of using the measurement and printing system according to an embodiment and its operation will be described. The measurement printing device 1 is configured to be switchable between a learning mode and an operation mode. First, the learning mode will be described.
[0054] When the measurement and printing device 1 is set to the learning mode, the camera 20 is activated. The image information captured by the camera is stored in the memory unit 25 as an initial image. The shooting conditions of the subject detected from the captured image, such as the brightness, contrast, and color temperature of the subject etc. are stored.
[0055] Next, multiple cuts of meat M are arranged on the conveyor C in a specified order. For example, While being arranged in the order and conveyed, image information of each portion of meat M is taken by a plurality of cameras 20. The image information from multiple directions is input in sequence to the image processing unit 21. The part name of the meat M is “kata” in FIG. 6, which is read from the memory unit 25. The part name of each part of meat M is "shoulder loin", and the part name is assigned to the image information of each part of meat M. The image information is input to the image processing unit 21. Then, the image processing unit 21 extracts each part from the image information. The area of the meat m is extracted, and the extracted image information is stored in the storage unit 25 as training data. The extracted image information and the part name associated with it are stored in the memory unit 2 as correct answer data. This process is repeated several times until a sufficient amount of training data is stored. Then, the training data and the correct answer data are applied to a multi-layered neural network to perform deep learning. By performing layer learning, a trained model 22 is obtained. If the accuracy rate of the trained model 22 increases, The parameters of the neural network of the layer are memorized to complete the trained model 22.
[0056] Next, the operation mode will be described. When the weighing and printing device 1 is set to the operation mode, it first switches to the standby mode. Even when the camera 20 is set to the standby mode, the image captured by the camera 20 is displayed on the display unit 11. Then, when the processed cut of meat M flows from the upstream, the camera 20 The image of the portion of meat m on the display unit 11 is captured and output. moves on the display unit 11 in accordance with the actual movement of the portion of meat M.
[0057] On the other hand, image information of the portion of meat M photographed by the camera 20 is input to an image processing unit 21. The image processing unit 21 extracts image information of only the partial meat m from the image information, excluding background information. The extracted image information is input from the image processing unit 21 to the trained model 22. The part name of the meat part m is estimated from the input image information of the meat part m, and the part name is selected. The complement is output to the discriminator 23.
[0058] If there is a part that exceeds, for example, 75% of the part candidates, the discrimination unit 23 discriminates the part. Output as a separate result. Otherwise, display the estimated part candidates as they are. In this case, the estimated part candidates are used as the discrimination result as they are, as shown in Fig. 3. The part m is displayed superimposed on the part m.
[0059] The worker looks at the designated part name displayed in the upper column of Figure 3 and selects the part of meat M , i.e., cuts of meat with sirloin 80% indicated as cut candidates When the worker picks up the portion of meat M from the conveyor C and places it on the weighing tray, the display screen of the display unit 11 changes to the screen shown in Fig. 4. When the worker operates a print key (not shown) while looking at the screen, a label as shown in Fig. 5 is printed, and the worker picks up the label and affixes it to the portion of meat M that has been weighed.
[0060] The meat portion M with the label attached is transported to the downstream packaging line, where it is They are packed into boxes and stored in a refrigerated storage facility.
[0061] Although one embodiment of the present invention has been described above, the present invention is not limited thereto. Other embodiments are possible within the scope of the claims. In this example, supervised learning was performed, but instead, unsupervised learning was performed. Then, the correct answer data is learned by assigning the part name to each classified cut of meat. It is also possible to do so.
[0062] The control unit 24 of the meat discriminator 2 is configured with a personal computer. A trained model22 may be downloaded from the cloud in advance. The trained model 22 may be installed in the control unit 13 of the measurement and printing device 1. In addition, a plurality of cameras 20 are arranged along the conveying direction of the conveyor C, and the part coming out of the cooling device is captured. The image of each portion of meat M is displayed on the display unit 11 until the portion of meat M reaches the measuring unit 10. That's fine. [Explanation of symbols]
[0063] 1 Weighing printing device 2 Meat discrimination device 10 Measuring part 11 Display section 12 Printers 20 Camera 22 Pre-trained models 24 Control Unit 25 Memory section
Claims
1. A camera that photographs and acquires image information of cuts of meat that have been butchered down to the cut name; A trained model that inputs the image information and outputs candidates for the part names of the meat; A meat discrimination device comprising: a control unit that discriminates a part name of the meat photographed by the camera from candidates of the part name output from the trained model, A memory unit that stores the part names of each meat cut from the carcass in units of the carcass, When determining the part name of the meat, the control unit determines the part name from among the part names of the carcass stored in the memory unit, excluding the part names already determined. Meat identification device.
2. The control unit stores the shooting conditions that affected the shooting when acquiring the captured image used to generate the learned model, and when the stored shooting conditions are different from the shooting conditions when the captured image is acquired again by the camera, notifies the same. The meat discriminating device according to claim 1.
3. The control unit stores a background image when the camera is installed as an initial image, and notifies the user when the background image when the camera is started is different from the initial image. The meat discriminating device according to claim 1.
4. A meat discrimination device according to any one of claims 1 to 3, a measuring and printing device for inputting the part name from the meat discriminator; A measurement and printing system comprising: The weighing and printing device is provided with a display unit that displays the meat photographed by the camera, and the display unit displays candidate names of the meat parts determined by the control unit superimposed on the displayed image of the meat. This is a weighing and printing system.
5. The measuring and printing device includes a measuring unit that measures the loaded meat and a control unit, The control unit displays a part name of the meat superimposed on the image of the meat displayed on the display unit, and when the meat is loaded onto the weighing unit and the meat displayed on the display unit disappears from the display unit, the control unit sets the part name of the meat superimposed on the meat as the part name of the meat loaded on the weighing unit.
5. The measurement and printing system according to claim 4.
6. A weighing and printing system as described in claim 4 or 5, wherein the weighing and printing device is equipped with a printer that prints information including the name of the part of the meat.
Citation Information
Patent Citations
Method for molding opening of preform
JP1994055617A
Meat identification method and its meat identification device
JP2004045072A
Identification device of bread, and program for the same
JP2011170745A
Sales processing system and commodity popularity analysis method
JP2016110585A
Determination device, determination method, and object determination program
JP2019192169A
