Method, device and program for evaluating quality and trait of pork

Abstract

To provide a method, a device and a program for objectively evaluating quality and a trait of pork.SOLUTION: A method comprises steps of: subjecting a fat emphasis processing area G channel image acquired from a pork cross-sectional RGB image to blur processing; creating a difference image by using images before and after the blur processing; creating a synthetic RGB image by adding the difference image to the pork cross-sectional RGB image; creating a marbling area image by binarizing the synthetic RGB image; and creating a fat emphasis image by adding the marbling area image to the pork cross-sectional RGB image.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The present invention relates to a method for creating a fat-enhanced image of a pork cross-section, a method for evaluating pork color and a method for evaluating pork marbling using the method, and an image analysis device and program that can be used in these methods. [Background technology]

[0002] Meat quality traits such as meat color and marbling (so-called marbling) of beef, pork, and other edible meats are important in the distribution and consumption of meat. In Japan, beef is visually evaluated and graded by trained graders, and the price is determined at auction based on the grading information. In addition, a method for measuring the meat quality traits of beef loin center using image analysis has been developed to objectively evaluate meat quality traits without relying on visual inspection (Patent Documents 1 and 2, Non-Patent Document 1, etc.).

[0003] On the other hand, pork trading standards are determined by weight and backfat thickness, and there are fewer opportunities to evaluate meat quality traits than for beef. The Japan Meat Grading Association, a public interest incorporated association, has established the Pork Marbling Standard as a criterion for determining marbling, and the Pork Color Standard as a criterion for determining pork color. However, due to the limited opportunities for evaluation, there are only a limited number of graders who can accurately evaluate and grade pork quality traits. Furthermore, there are concerns about inconsistencies in evaluation due to graders' inexperience. Therefore, there is a need for the development of an objective method for evaluating pork quality traits.

[0004] As with beef, the use of image analysis in the evaluation of pork meat quality traits is also being considered. For example, Non-Patent Document 2 discloses a method of binarizing pork based on enhanced images of stained tissue from thinly sliced ​​pork loin. Non-Patent Document 3 discloses a method of quantifying the fat content within pork loin using magnetic resonance images. Furthermore, Non-Patent Document 4 discloses a method of binarizing thick slices of longissimus thoracis muscle by minimizing the variations in meat color using a rolling ball algorithm. However, these conventional techniques are image analysis methods targeted at dressed meat and cut meat, and no image analysis method or standard for pork that can be used for carcasses during the distribution process has yet been established.

[0005] Furthermore, unlike beef, pork has characteristics such as a wide range of color values ​​for meat color and low contrast between meat color and fat color, making it important to efficiently separate muscle and fat marbling using image processing. [Prior art documents] [Patent documents]

[0006] [Patent Document 1] Japanese Patent Application Laid-Open No. 2014-2136 [Patent Document 2] Japanese Patent Application Laid-Open No. 2014-71018 [Non-patent literature]

[0007] [Non-Patent Document 1] Jackman P et. al., Meat Science 83, 187-194 (2009). [Non-patent document 2] Faucitanoa L et. al., Meat Science 69, 537-543 (2005). [Non-patent document 3] Avila MM et. al., Pattern Recognition and Image Analysis, Proceedings, Part I, 145-152 (2007). [Non-patent document 4] Uttaro B et. al., Meat Science 179, 108549 (2021). Summary of the Invention [Problem to be solved by the invention]

[0008] An object of the present invention is to provide a new method for objectively evaluating pork quality traits. [Means for solving the problem]

[0009] The present inventors have found that muscle and marbling can be separated by processing the RGB image of a pork cross section to emphasize the marbling region. * value and a * We found that the values ​​tended to correlate with the pork color score (PCS) determined by graders based on the Pork Color Standard, and that the fat area percentage and roughness index calculated from the images tended to correlate with the pork marbling score (PMS) determined by graders based on the Pork Marbling Standard. Therefore, we found that these parameters could be used as new meat quality trait evaluation criteria using image analysis.

[0010] The present disclosure provides the following: Section 1. extracting a fat-enhanced processing region set within the region of interest from the G channel image separated from the RGB image of the pork cross section, and blurring the G channel image of the extracted fat-enhanced processing region; creating a difference image using the G channel image of the fat-enhanced processed region after blurring and the G channel image of the fat-enhanced processed region before blurring; adding the difference image to the pork cross-section RGB image to create a composite RGB image; binarizing the composite RGB image to create a marbling region image; and A step of adding the marbling region image to the pork cross-section RGB image to create a fat-enhanced image. A method for producing a fat-enhanced image of a pork cross section, comprising: Section 2. The step of creating a difference image comprises: A step of adding the G channel image of the fat-enhanced processing area after the blurring processing to the G channel image of the fat-enhanced processing area before the blurring processing to create a composite G channel image; and A step of extracting a fat-enhanced processing region from the composite G channel image and subtracting the extracted image from the G channel image of the fat-enhanced processing region before blurring to create a difference image. The method according to item 1, comprising: Section 3. multiplying the RGB image of the pork cross section by the G channel image separated from the image to create a meat color-enhanced RGB image; extracting a fat-enhanced processing region set within the region of interest from the G channel image separated from the flesh-enhanced RGB image, and performing blurring on the G channel image of the extracted fat-enhanced processing region; creating a difference image using the G channel image of the fat-enhanced processed region after blurring and the G channel image of the fat-enhanced processed region before blurring; adding the difference image to the flesh-enhanced RGB image to create a composite RGB image; binarizing the composite RGB image to create a marbling region image; and A step of adding the marbling region image to the pork cross-section RGB image to create a fat-enhanced image. A method for producing a fat-enhanced image of a pork cross section, comprising: Section 4. The step of creating a difference image comprises: A step of adding the G channel image of the fat-enhanced processing area after the blurring processing to the G channel image of the fat-enhanced processing area before the blurring processing to create a composite G channel image; and A step of extracting a fat-enhanced processing region from the composite G channel image and subtracting the extracted image from the G channel image of the fat-enhanced processing region before blurring to create a difference image. Item 3. The method according to Item 3, comprising: Section 5. In addition to the steps of creating a fat-enhanced image from the pork cross-sectional RGB image defined in any one of items 1 to 4, binarizing the fat-weighted image and separating the region of interest into muscle and marbling regions; Obtaining L* and a* values ​​from the muscle region of the pork cross-section RGB image; and determining a flesh color score for the muscle region based on the L* and a* values A method for assessing pork meat color, including: Section 6. In addition to the steps of creating a fat-enhanced image from the pork cross-sectional RGB image defined in any one of items 1 to 4, binarizing the fat-weighted image and separating the region of interest into muscle and marbling regions; calculating the fat area percentage and roughness index of the region of interest; and Determining the marbling score based on the fat area percentage and roughness index The method for evaluating pork marbling further comprises: Section 7. A program that causes a computer to execute the method described in any one of items 1 to 4. Section 8. Item 5. A program that causes a computer to execute the method described in item 5. Section 9. A program that causes a computer to execute the method described in item 6. Section 10. An image analysis device including a fat enhancement processing unit that performs fat enhancement processing on a pork cross-sectional RGB image, the fat enhancement processing comprising: A fat-enhanced processing region set within the region of interest is extracted from the G channel image separated from the RGB image of the pork cross section, and a blurring process is performed on the G channel image of the extracted fat-enhanced processing region. creating a difference image using the G channel image of the fat-enhanced processing area after blurring and the G channel image of the fat-enhanced processing area before blurring; Adding the difference image to the pork cross-section RGB image to create a composite RGB image; Binarizing the composite RGB image to create a marbling region image; and Adding the marbling region image to the pork cross-section RGB image to create a fat-enhanced image. The image analysis device comprising: Section 11. The creation of the difference image is Adding the G channel image of the fat-enhanced processing area after the blurring processing to the G channel image of the fat-enhanced processing area before the blurring processing to create a composite G channel image; and Extracting the fat-enhanced processing region from the composite G channel image, and subtracting the extracted image from the G channel image of the fat-enhanced processing region before blurring. Item 11. The image analysis device according to Item 10, comprising: Section 12. An image analysis device including a fat enhancement processing unit that performs fat enhancement processing on a pork cross-sectional RGB image, the fat enhancement processing comprising: Multiplying the RGB image of the pork cross section by the G channel image separated from the image to create a meat color-enhanced RGB image; extracting a fat-enhanced processing region set within the region of interest from the G channel image separated from the flesh-enhanced RGB image, and performing blurring on the G channel image of the extracted fat-enhanced processing region; creating a difference image using the G channel image of the fat-enhanced processing area after blurring and the G channel image of the fat-enhanced processing area before blurring; adding the difference image to the flesh-enhanced RGB image to create a composite RGB image; Binarizing the composite RGB image to create a marbling region image; and Adding the marbling region image to the pork cross-section RGB image to create a fat-enhanced image. The image analysis device comprising: Section 13. The creation of the difference image is Adding the G channel image of the fat-enhanced processing area after the blurring processing to the G channel image of the fat-enhanced processing area before the blurring processing to create a composite G channel image; and Extracting the fat-enhanced processing region from the composite G channel image, and subtracting the extracted image from the G channel image of the fat-enhanced processing region before blurring. Item 13. The image analysis device according to Item 12, comprising: Section 14. Item 14. The image analysis device according to any one of items 10 to 13, further comprising an image analysis unit that performs flesh color analysis on a fat-enhanced image, wherein the flesh color analysis is binarizing the fat-weighted image and separating the region of interest into muscle and marbling regions; Obtaining L* and a* values ​​from the muscle region of the pork cross-section RGB image; and Determining the flesh color score of the muscle region based on the L* and a* values The device comprising: Section 15. Item 14. The image analysis device according to any one of items 10 to 13, further comprising an image analysis unit that performs marbling analysis on a fat-enhanced image, binarizing the fat-weighted image and separating the region of interest into muscle and marbling regions; Calculating the fat area percentage and roughness index of the area of ​​interest; and Determine the marbling score based on the fat area percentage and roughness index. The device comprising: [Effects of the Invention]

[0011] According to the present invention, pork quality traits can be objectively evaluated by image analysis. [Brief explanation of the drawings]

[0012] [Figure 1] 1 is a block diagram showing the overall configuration of an example of an image analysis device. [Figure 2] 1 is a flowchart showing steps in a fat-enhanced image creation method that does not include flesh-color enhancement processing. [Figure 3] 1 is a flowchart showing steps in a fat-enhanced image creating method including flesh color enhancement processing. [Figure 4] 1 is a flowchart showing steps in a meat color evaluation method. [Figure 5] 1 is a flowchart showing each step in a marbling evaluation method. [Figure 6] These are photographs showing the process of fat enhancement processing on a pork cross-sectional RGB image. Figure 6a is an RGB cross-sectional image of pork with a region of interest set using a contour line, Figure 6b is a grayscale image of the G channel separated from Figure 6a, Figure 6c is a meat-color-enhanced image created by multiplying Figures 6a and 6b, and Figure 6d is an image in Figure 6c where the fat-enhanced region is shown in black. [Figure 7] These are photographs showing the process of fat enhancement processing on an RGB image of a pork cross section. Figure 7a is a grayscale image of the G channel separated from the meat-color-enhanced image in Figure 6c, Figure 7b is a G channel image of the fat-enhanced region extracted from Figure 7a, Figure 7c is a blurred image after applying a Gaussian filter to Figure 7b, and Figure 7d is a composite G channel image created by adding Figures 7b and 7c. [Figure 8] These are photographs showing the process of fat enhancement processing on an RGB image of a pork cross section. Figure 8a is an image in which the fat-enhanced processing region is extracted from Figure 7d, Figure 8b is a difference image obtained by subtracting the image in Figure 8a from the G channel image of the fat-enhanced processing region in Figure 7b, Figure 8c is an image brightened by adjusting the luminosity of the difference image, and Figure 8d is a composite RGB image obtained by adding Figure 8c three times to the meat-color-enhanced image in Figure 6c. [Figure 9] These are photographs showing the process of fat enhancement processing on a pork cross-section RGB image. Figure 9a is an image obtained by binarizing the composite RGB image of Figure 8d using a threshold value of G250, Figure 9b is the image obtained by removing particles smaller than 5 pixels from Figure 8a and performing a closing process, and Figure 9c is a fat-enhanced image obtained by adding Figure 9b to the pork cross-section RGB image of Figure 6a. [Figure 10] These photographs show the effect of fat enhancement on the discrimination of muscle and marbling by binarization. Figure 10a is an RGB image of a pork cross section without fat enhancement, Figure 10b is a binarized image of Figure 10a, Figure 10c is an RGB image of a pork cross section with fat enhancement, and Figure 10d is a binarized image of Figure 10c. [Figure 11]FIG. 1 is a scatter plot showing the relationship between visually graded PCS and average loin center muscle L* value. [Figure 12] FIG. 1 is a scatter plot showing the relationship between visually graded PCS and average a* value of loin center muscle. [Figure 13] 1 is a three-dimensional scatter plot of muscle mean L* value, a* value and PCS. [Figure 14] Figure 14a is an L-type matrix showing the relationship between L* and a* values ​​and PCS, and Figure 14b is an L-type matrix (Image Pork Color Score (iPCS) judgment criteria table) showing the meat color judgment criteria based on image analysis created based on Figure 14a. [Figure 15] 1 is a graph showing the difference between PCS and iPCS and its frequency. [Figure 16] FIG. 1 is a scatter plot showing the relationship between visually graded PMS and loin eye fat area percentage. [Figure 17] Scatter plot showing the relationship between roughness index and PMS, grouped according to fat area percentage. [Figure 18] These are fat-enhanced images of three individuals (a, b, c), each with a fat area ratio of 7%. [Figure 19] 3D scatter plot of fat area percentage, roughness index and PMS. [Figure 20] Figure 20a is an L-type matrix showing the relationship between fat area percentage and roughness index and PMS, and Figure 20b is an L-type matrix (Image Pork Marbling Score (iPMS) Judgment Criteria Table) showing the marbling judgment criteria by image analysis created based on Figure 20a. [Figure 21] This is a graph showing the difference between PMS and iPMS and their frequency. Note that the RGB images included in all drawings have been converted to grayscale images due to restrictions imposed by the specifications of the internet application software. DETAILED DESCRIPTION OF THE INVENTION

[0013] The following description may be based on representative embodiments or specific examples, but the present invention is not limited to such embodiments or specific examples. The upper and lower limit values ​​of each numerical range shown in this disclosure can be combined arbitrarily. In this disclosure, numerical ranges expressed using "~" or "-" mean ranges that include the numerical values ​​at both ends as upper and lower limits, unless otherwise specified.

[0014] The present disclosure provides a method for creating a fat-enhanced image from a pork cross-sectional RGB image (hereinafter also referred to as a fat-enhanced image creation method), a method for evaluating pork color using the method (hereinafter also referred to as a meat color evaluation method), and a method for evaluating pork marbling (hereinafter also referred to as a marbling evaluation method), as well as an image analysis device and a program that can be used in these methods.

[0015] An RGB image consists of three color channels: R (Red), G (Green), and B (Blue), and each channel generally has color information in 256 levels ranging from 0 to 255. Separating the color channels from an RGB image results in three single-channel images: an R channel image, a G channel image, and a B channel image. Each of these single-channel images is generally a grayscale image with brightness information (luminosity) in 256 levels ranging from 0 to 255.

[0016] The pork cross-section RGB image in the present disclosure is an RGB color digital image that can be obtained by photographing a pork cross-section. The pork to be photographed may be any part of pork for which evaluation of meat quality traits is desired. Current visual evaluation of pork meat quality traits is performed on the loin eye, which is a cross section of the longissimus thoracis muscle (loin) that appears when a half-round pork carcass is incised between the fourth and fifth ribs. Therefore, it is preferable to use a pork cross-section RGB image that includes the loin eye in order to enable evaluation of marbling and meat color that correlates with the grading by a grader.

[0017] RGB images can be captured using a general digital camera, and it is particularly preferable to use a known carcass imaging device that can capture images in focus across the entire cross section, such as MIJ mobile with BEAK50 (MIJ labo Co., Ltd.) or a mirror-type imaging device (Hayasaka Riko Co., Ltd.).

[0018] [Image analysis equipment] An exemplary embodiment of an image analysis device that can be used in the fat-enhanced image creation method, marbling evaluation method, and meat color evaluation method will be described with reference to FIG.

[0019] 1 shows the overall configuration of an image analysis device according to an embodiment of the present disclosure. The image analysis device 1 is a computer having an interface unit 11, an input unit 12, a storage unit 13, a calculation unit 14, and an output unit 15. The image analysis device 1 is connected to an external device 20, and data is read from the external device 20.

[0020] The interface unit 11 controls input and output of data such as images and programs exchanged with the external device 20. The interface unit 11 stores data received from the external device 20 in the storage unit 13.

[0021] The input unit 12 is a device such as a keyboard, a mouse, or a touch panel, and receives signals in response to operations by an operator.

[0022] The storage unit 13 includes a storage medium such as a hard disk, flash memory, RAM, or ROM, and a reading device for reading information stored in the storage medium. The storage unit 13 stores pork cross-section RGB image data used in the fat-enhanced image production method, the marbling evaluation method, and the meat color evaluation method, various processed image data generated from the image data, image analysis data obtained from the processed image data, reference data for determining marbling and meat color, programs for image processing and analysis, programs for controlling each functional unit of the image processing device 1, and various setting information.

[0023] The calculation unit 14 is configured with hardware such as a CPU. The calculation unit 14 has a fat enhancement processing unit 141 that reads a program stored in the storage unit 13 and performs calculations for fat enhancement processing up to creating a fat-enhanced image from the pork cross-sectional RGB image or meat color-enhanced RGB image stored in the storage unit 13. When settings required for calculations such as setting a region of interest and a fat enhancement processing region are manually input, the calculation unit 14 accepts setting operations by the operator via the input unit 12.

[0024] In one embodiment, the calculation unit 14 further includes an image analysis unit 142 that reads a program stored in the storage unit 13 and performs at least one of a calculation for determining a meat color score from the fat-enhanced image stored in the storage unit 13 and a calculation for determining a marbling score.

[0025] The output unit 15 outputs the pork cross-sectional RGB image stored in the memory unit 13, the fat-enhanced image obtained by the calculations of the calculation unit 14, and various information such as the marbling score and meat color score to an external device of the image analysis device 1, typically to a display connected to the output unit 15.

[0026] [Fat enhancement processing] The function of the fat-enhancement processing unit 141 included in the calculation unit 14 according to an embodiment of the present disclosure will be described together with each step of the fat-enhancement image creation method with reference to the flowchart shown in FIG.

[0027] The fat emphasis processing unit 141 sets a region of interest on the pork cross-section RGB image (hereinafter also referred to as the original image) (step S10). The region of interest is an area where evaluation of meat quality traits by image analysis is desired, and is preferably an area corresponding to the loin center. The region of interest may be set manually or automatically based on the functions of image analysis software.

[0028] The fat enhancement processing unit 141 separates color components from the pork cross-sectional RGB image in which the region of interest has been set to obtain a G channel image (step S11), extracts a fat enhancement processing region from the G channel image, and creates a G channel image of the fat enhancement processing region (step S12). Note that the order of steps S10 and S11 can be reversed; if step S11 is performed first, the region of interest can be set at the corresponding positions in the pork cross-sectional RGB image and the G channel image separated from that image.

[0029] The fat-enhancement processing region is set within the region of interest so as to include a region where the contrast between meat color and fat color is low or where the color value of meat color is wide. The fat-enhancement processing region can be set at any location within the region of interest with any size, and may be set so as to completely coincide with the region of interest. When the region of interest is a region surrounded by adipose tissue such as a loin center, the fat-enhancement processing region can be set close to the boundary of the region of interest so as to be slightly smaller than the region of interest, for example, 0.1 to 5 mm, preferably 0.2 to 3 mm, more preferably 0.5 to 2 mm inward from the boundary of the region of interest when converted into the actual size of a pork cross-section.

[0030] The fat enhancement processing unit 141 performs blurring on the G channel image of the fat enhancement processing region (step S13). Bluring is an image processing method that blends the value of each pixel in an image with the values ​​of neighboring pixels to smooth the edges and details of the image and achieve a blurring effect, and is also called a smoothing filter. Known filters such as a Gaussian filter, averaging filter, median filter, and box filter can be used for the blurring, and processing using a Gaussian filter is preferred.

[0031] The fat enhancement processing unit 141 creates a difference image using the G channel image of the fat enhancement processed region after blurring and the G channel image of the fat enhancement processed region before blurring (step S14). The difference image can be created by subtracting the G channel image of the fat enhancement processed region after blurring from the G channel image of the fat enhancement processed region before blurring. Alternatively, the difference image can be created by adding the G channel image of the fat enhancement processed region after blurring to the G channel image of the fat enhancement processed region before blurring to create a composite G channel image, extracting the fat enhancement processed region from the composite G channel image, and subtracting the extracted image from the G channel image of the fat enhancement processed region before blurring.

[0032] The difference image created in step S14 can be used in the next step as is or after adjustment to increase the luminosity.

[0033] The fat enhancement processing unit 141 generates a composite RGB image by adding the difference image to the corresponding position of the original pork cross-sectional RGB image (step S15). The addition of the difference image can be performed one or more times, for example, 1 to 5 times, and preferably 1 to 2 times.

[0034] The fat-enhancing processor 141 creates a marbling region image by binarizing the composite RGB image (step S16). The composite RGB image can be binarized by setting the threshold value of the G channel to a maximum value or a value close to that maximum value. When the maximum value of the G channel is 255, the threshold value can be set, for example, within the range of 240 to 255, preferably within the range of 245 to 253, and more preferably within the range of 248 to 252.

[0035] The marbling region image created in step S16 can be used in the next step as is, or after undergoing fine particle removal and closing processing.

[0036] The fat-enhancing processing unit 141 creates a fat-enhanced image by adding the marbling region image to the corresponding position of the original pork cross-sectional RGB image (step S17). The fat-enhanced image is an RGB image in which only the marbling region is enhanced compared to the original pork cross-sectional RGB image.

[0037] The function of the fat-enhancement processing unit 141 included in the calculation unit 14 in another embodiment of the present disclosure will be described together with each step of a fat-enhancement image creation method with reference to the flowchart shown in FIG.

[0038] The fat enhancement processing unit 141 sets a region of interest on the pork cross-sectional RGB image (step S20). Step S20 can be performed in the same manner as step S10.

[0039] The fat enhancement processing unit 141 separates color components from the pork cross-sectional RGB image to obtain a G channel image (step S21), and multiplies the G channel image by a corresponding position in the original pork cross-sectional RGB image to create a meat color enhanced RGB image (step S22). Steps S21 and S22 are also referred to as meat color enhancement processing.

[0040] The fat-enhancement processing unit 141 performs a series of processes from steps S23 to S29 on the meat-color-enhanced RGB image to create a fat-enhanced image. Steps S23 to S27 can be performed by replacing the pork cross-section RGB image from steps S11 to S15 with the meat-color-enhanced RGB image. Steps S28 and S29 can be performed in the same way as steps S16 and S17.

[0041] It should be noted that step S20 can be performed between steps S21 and S22 or after steps S21 and S22. For example, if step S20 is performed after steps S21 to S23, the regions of interest can be set at corresponding positions in the pork cross-section RGB image and the G channel image separated from the meat-color-enhanced RGB image.

[0042] The series of processes from creating a fat-enhanced image from a pork cross-section RGB image is called fat-enhancement processing.

[0043] [Image analysis (flesh color analysis)] One function of the image analysis unit 142 included in the calculation unit 14 will be described together with each step of the meat color evaluation method with reference to the flowchart shown in FIG.

[0044] The image analysis unit 142 separates the region of interest into muscle and marbling regions by binarizing the fat-enhanced image (step S30). The binarization of the fat-enhanced image can be performed using a method similar to that reported in image analysis for evaluating beef meat quality traits. Binarization can be performed, for example, by setting a threshold value for the G channel of the fat-enhanced image so that the muscle and marbling regions can be separated. The threshold value can be set manually or automatically using a value calculated by Otsu's discriminant analysis binarization method or the like.

[0045] In one embodiment, the image analysis unit 142 extracts L from the muscle region of the original pork cross-section RGB image. * value and a * The muscle region of the pork cross-sectional RGB image can be determined based on the position of the muscle region in the binarized fat-enhanced image. * value and a * The values ​​are calculated by calculating the average values ​​(R average, G average, B average) from the R channel, G channel, and B channel values ​​of each pixel in the muscle area, and then using the following formula to calculate these average values.

number

number

[0046] L of muscle area from binarization of fat-weighted image * value and a * The series of processes leading up to the calculation of the value is called flesh color analysis.

[0047] In one embodiment, the image analysis unit 142 calculates L * value and a * Based on the value, the meat color score (also called iPCS) is determined by referring to the meat color evaluation criteria stored in the storage unit 13 (step S32). * value and a * It is an L-shaped matrix with values ​​arranged in rows and columns, with the iPCS at the intersections. The iPCS of the marbling criteria can be set to correlate with the PCS assigned by the grader.

[0048] [Image analysis (marbling analysis)] One function of the image analysis unit 142 included in the calculation unit 14 will be described together with each step of the marbling evaluation method with reference to the flowchart shown in FIG.

[0049] The image analysis unit 142 binarizes the fat-enhanced image to separate the region of interest into a muscle region and a fat marbling region (step S40). Step S40 can be performed in the same manner as step S30.

[0050] In one embodiment, the image analysis unit 142 calculates the fat area percentage and the roughness index using the binarized fat-enhanced image (step S41). The fat area percentage is the area percentage of the marbling region in the region of interest and can be calculated from the respective areas of the region of interest and the marbling region. The roughness index can be calculated using the following formula from the image after thinning processing has been performed X times on the binarized fat-enhanced image.

number

[0051] The series of processes from binarization of fat-enhanced images to calculation of fat area percentage and roughness index is called marbling analysis.

[0052] In one embodiment, the image analysis unit 142 determines the marbling score (also referred to as iPMS) based on the calculated fat area percentage and roughness index, with reference to the marbling assessment criteria stored in the memory unit 13 (step S42). The marbling assessment criteria are an L-shaped matrix in which the fat area percentage and roughness index are arranged in rows and columns, with the iPMS indicated at the intersections. The iPMS of the marbling assessment criteria can be set to correlate with the PMS assigned by the grader.

[0053] The programs that can be used in the fat-enhanced image creation method, meat color evaluation method, and marbling evaluation method provided in the present disclosure are programs for causing a computer to execute each step of the methods, and the details of the programs are as described above. The present disclosure also provides a computer-readable storage medium, for example, a non-transitory storage medium, that stores the programs.

[0054] The present invention will be described in more detail below with reference to examples, but the present invention is not limited to these examples. [Example]

[0055] Materials and Methods (1) Taking images of carcasses The test pigs used were 115 Tokyo X pigs (67 castrated, 46 females, and 2 males) shipped to Meat Companion Co., Ltd. from three producers in October 2022 and January 2023. One to two days after slaughter, RGB color images of the left semicircular 4th-5th intercostal cross section were taken using MIJ mobile with BEAK50 (MIJ labo, Inc.). The image resolution was 100 pixels per cm of actual size.

[0056] (2) Visual grading Using the photographed carcass images, experienced staff from the Ome Livestock Center of the Tokyo Metropolitan Agriculture, Forestry and Fisheries Promotion Foundation, conducted meat quality grading of the longissimus thoracis muscle of the carcass. Grading was performed in accordance with the pork carcass trading standards of the Japan Meat Grading Association, using two evaluation items: pork color score (PCS) and pork marbling score (PMS). Grading was performed by multiple people (3-5 people), and the most frequent score for each meat quality item was calculated for each individual, which was used as the final score for each trait. When it was not possible to narrow down the most frequent score to a single one due to the number of people evaluating, the multiple most frequent scores were averaged. In the data from this study, all individuals with multiple most frequent scores had consecutive scores.

[0057] (3) Image processing The image of the area containing the loin eye (hereafter referred to as the loin eye image) cut out from the photographed carcass image was subjected to fat enhancement processing using the image processing software Photoshop (Adobe Systems Inc.), ImageMagick (ImageMagick Studio LLC), and PopImaging (Digital Being Kids, Ltd.) in the following procedure. a) Determining the image analysis area Using the beef carcass image analysis software BeefAnalyzer-II (Hayasaka Riko Co., Ltd.), the outline of the loin eye was manually drawn in green (RGB = 0,255,0) on the loin eye image, and this was designated as the loin eye region (also called the image analysis region or region of interest) to be analyzed (Figure 6a). b) Flesh color enhancement processing The grayscale image of the G channel separated from the loin eye image (Fig. 6b) was multiplied by the original loin eye image to enhance the meat color (Fig. 6c). c) Determination of fat enhancement processing area The area 1.0 mm in actual size and 10 pixels in size inside the outline of the image analysis area (Fig. 6d) was defined as the fat enhancement processing area. d) Acquiring the G channel and blurring The loin eye image with the meat color enhanced was again decomposed into RGB channels to obtain the G channel (Figure 7a). The fat-enhanced areas of the G channel were then removed (Figure 7b), and a Gaussian blur was applied with (radius) × (sigma) set to 40 × 40 (Figure 7c). Figure 7b and Figure 7c were then combined (Figure 7d), and the fat-enhanced areas were removed again (Figure 8a). e) Subtraction and brightness adjustment The blurred image (Figure 8a) was subtracted from the G channel image (Figure 7a) to extract marbling information (Figure 8b).Then, the brightness was adjusted to 200% to clarify the marbling (Figure 8c). f) Binarization The marbling image (Figure 8c) was added three times to the loin eye image (Figure 8d). Afterwards, binarization was performed with a G channel threshold of 250 (Figure 9a). After removing particles smaller than 5 pixels, a closing process was performed to remove noise and thin lines (Figure 9b). g) Addition The binarized marbling image (Fig. 9b) was added to the loin eye image to obtain a loin eye image in which only the marbling was emphasized (Fig. 9c).

[0058] (4) Image analysis The fat-enhanced image of the loin eye region created in (3) was analyzed using the beef carcass image analysis software BeefAnalyzer-II (Hayasaka Riko Co., Ltd.) according to the following procedure, and the fat area percentage, roughness index, muscle average R value, muscle average G value, and muscle average B value were calculated. The fat area percentage indicates the area percentage of marbling particles in the entire loin eye. The roughness index is an index of the overall roughness of the marbling particles in the loin eye, with higher values ​​indicating the presence of coarse marbling particles within the muscle (Kuchida K, Suzuki M, Miyoshi S., 2002. Image analysis of the evaluation method for the roughness of marbling particles in the longissimus thoracis muscle. Bulletin of the Japanese Society of Animal Science 73, 9-17.). Because the shape of the loin eye and fat particles in pork is smaller than that in beef, the number of thinning passes for the roughness index was set to 5 (equivalent to 0.5 mm). The muscle average R value, muscle average G value, and muscle average B value are the average color values ​​of each channel grayscale of the muscle part excluding the fat marbling of the loin eye. In addition, the color values ​​are calculated from the RGB values ​​using L, which is designed to approximate human vision. * a * b * Convert to color space and measure muscle average L * Value, muscle average a * Value and muscle average b * The value was calculated.

[0059] (5) Statistical analysis Statistical analysis was performed using SAS Enterprise Guide 8.2 (SAS Institute Inc.). Mean differences between groups were compared using one-way analysis of variance followed by Tukey's multiple comparison test, with p<0.05 considered significant.

[0060] [Example 1] A cross-sectional RGB image of pork (Fig. 6a) and a fat-enhanced RGB image (Fig. 9c) obtained by subjecting the same image to the image processing described above (3) were binarized using Otsu's discriminant analysis binarization method. Without image processing, it was difficult to distinguish between muscle and marbling by binarization (Fig. 10a, b), but by employing fat-enhancement processing, it became possible to distinguish between muscle and marbling by binarization (Fig. 10c, d).

[0061] [Example 2] Image meat quality traits calculated from fat-enhanced RGB images (muscle average L * Value, muscle average a * A correlation analysis was performed with visually graded PCS using the L * There was a moderate negative correlation between the value and PCS, and L * There was a strong tendency for the lower the value, the higher the evaluation of PCS (Fig. 11). * There was a moderate positive correlation between the value and PCS, and * There was a strong tendency for PCS to be rated higher as the value increased (Figure 12). * value and a * When the values ​​were plotted three-dimensionally, taking into account the PCS at the same time (Fig. 13), a linear relationship was observed between the three. * value and a * It was thought that by using two image meat quality traits of different values ​​simultaneously, it would be possible to create more accurate meat color judgment standards.

[0062] Figure 14a shows the L * value and a * The table shows the combinations of values ​​and the corresponding average PCS. * The lower the value, the higher the PCS tends to be evaluated. * By taking into account the tendency that the higher the value, the higher the PCS is evaluated, and making adjustments, we created the Image Pork Color Score (iPCS) judgment criteria table, which is a meat color judgment standard based on image analysis in Figure 14b. When creating the iPCS judgment criteria table, L * If are equal, a * The higher the value, the higher the PCS was evaluated. * The iPCS value was adjusted so that it would not decrease as you move to the right where a becomes higher. * When L* is equal, the lower the L*, the higher the PCS tends to be evaluated. * The iPCS was adjusted so that it would not drop as the temperature gets lower towards the top.

[0063] The accuracy of the iPCS criteria table was calculated based on the frequency of discrepancies between visually graded PCS and iPCS. The results showed agreement for 78.3% of animals (Figure 15). The final PCS rating was calculated using the most frequent rating from multiple staff members. However, of the 115 animals, only 28 (23.4%) had unanimous agreement on PCS ratings, while the remaining animals had differing opinions. Furthermore, of the 28 animals for which all staff members agreed on PCS ratings, only 27 had agreement on PCS and iPCS, suggesting that the common evaluation of the staff's PCS standards and the iPCS standards are highly accurate. Furthermore, 28 of the 115 animals (24.3%) had ratings with a difference of two or more points between staff members, suggesting that even experienced staff members have difficulty in making a clear assessment for more than 20% of animals.

[0064] [Example 3] A correlation analysis was conducted with the PMS visually graded using the image meat quality traits (fat area percentage, roughness index) calculated from fat-enhanced RGB images. A significant, strong positive correlation was observed between the PMS visually graded by staff and the fat area percentage determined by image analysis, and the PMS tended to be rated higher as the fat area percentage increased (Figure 16).

[0065] Furthermore, after dividing the individuals into seven groups based on the degree of fat area percentage, the relationship between roughness index and PMS was plotted (Figure 17). In the five groups other than the 2-3% and 4-5% fat area percentage groups, there was a tendency for the lower the roughness index to be rated higher for PMS. The reason this tendency was not observed in the two low fat area percentage groups is thought to be because the low fat area percentage made it difficult for coarse fat particles to appear, and there were few individuals with a high roughness index. Similarly, when comparing the loin eye of three individuals, all with a fat area percentage of 7%, there was a tendency for individuals with coarse fat particles and a higher roughness index to be rated lower for PMS (Figure 18).

[0066] When fat area percentage and roughness index were plotted in three dimensions considering PMS simultaneously (Figure 19), a linear relationship was observed. Therefore, it was thought that by using the two image meat quality traits of fat area percentage and roughness simultaneously, it would be possible to create more accurate image judgment criteria for PMS.

[0067] Figure 20a shows the combinations of fat area percentage and roughness index and the corresponding average PMS. The image pork marbling score (iPMS) criterion table (Figure 20b) was created by taking into account the aforementioned tendency for higher fat area percentages to be rated higher in PCS, and for the same fat area percentages to be rated lower in roughness indexes, making adjustments. When creating the iPMS criterion table, we observed a tendency for the PMS to be rated higher as the roughness index decreased for the same fat area percentages. Therefore, we adjusted the PMS to not decrease as we moved up the row, where the roughness index decreased. Similarly, we observed a tendency for the PMS to be rated higher as the fat area percentage increased for the same roughness index. Therefore, we adjusted the PMS to not decrease as we moved to the right, where the fat area percentage increased.

[0068] Regarding the accuracy of the iPMS criteria table, we calculated the frequency of discrepancies between visually graded PMS and iPMS (Figure 21), and found that the ratings were consistent for 70.4% of animals. The final PMS rating was calculated using the most frequent rating from multiple staff members. However, of the 115 animals, only 27 (20.9%) had PMS ratings that were consistent across all staff members, while the remaining animals showed variation in ratings among staff members. Furthermore, even among the 27 animals for which PMS ratings were consistent across all staff members, only 17 had PMS and iPMS ratings that were consistent across staff members. Furthermore, 13 of the 115 animals (11.3%) had ratings that differed by two or more points, suggesting that even when rated by staff, a clear assessment is difficult for more than 10% of animals.

[0069] Tokyo X is a Tokyo brand of pork characterized by rich marbling, and has established itself as a high-quality brand of pork with added value. By evaluating and grading meat quality traits for high-quality pork like Tokyo X using image analysis, it will be possible to stably monitor meat quality within a lineage. Objectively evaluated meat quality information is expected to be a tool for maintaining and improving the quality of branded pork through feedback guidance to producers and pig breeding and selection through progeny testing. Furthermore, by establishing grades based on meat quality determined by image analysis even within the same lineage, it will be possible to contribute to diversifying consumer demographics and improving unit prices. [Explanation of symbols]

[0070] 1. Image analysis device 11 Interface section 12 Input section 13 Storage section 14 Arithmetic section 141 Fat enhancement processing unit 142 Image Analysis Unit 15 Output section 20 External device

Claims

1. extracting a fat-enhanced processing region set within the region of interest from the G channel image separated from the RGB pork cross-section image, and performing blurring on the G channel image of the extracted fat-enhanced processing region; creating a difference image using the G channel image of the fat-enhanced processed region after blurring and the G channel image of the fat-enhanced processed region before blurring; adding the difference image to the pork cross-section RGB image to create a composite RGB image; Binarizing the composite RGB image to create a marbling region image; and A step of adding the marbling region image to the pork cross-sectional RGB image to create a fat-enhanced image. A method for producing a fat-enhanced image of a pork cross section, comprising:

2. The step of creating a difference image comprises: a step of adding the G channel image of the fat-enhanced processing region after the blurring processing to the G channel image of the fat-enhanced processing region before the blurring processing to generate a composite G channel image; A step of extracting a fat-enhanced processing region from the composite G channel image and subtracting the extracted image from the G channel image of the fat-enhanced processing region before blurring to create a difference image. The method of claim 1 , comprising:

3. a step of multiplying the RGB image of the pork cross section by the G channel image separated from the RGB image to create a meat color enhanced RGB image; extracting a fat-enhanced processing region set within the region of interest from the G channel image separated from the flesh-color-enhanced RGB image, and performing blurring on the G channel image of the extracted fat-enhanced processing region; creating a difference image using the G channel image of the fat-enhanced processed region after blurring and the G channel image of the fat-enhanced processed region before blurring; adding the difference image to the flesh-enhanced RGB image to create a composite RGB image; Binarizing the composite RGB image to create a marbling region image; and A step of adding the marbling region image to the pork cross-sectional RGB image to create a fat-enhanced image. A method for producing a fat-enhanced image of a pork cross section, comprising:

4. The step of creating a difference image comprises: a step of adding the G channel image of the fat-enhanced processing region after the blurring processing to the G channel image of the fat-enhanced processing region before the blurring processing to generate a composite G channel image; A step of extracting a fat-enhanced processing region from the composite G channel image and subtracting the extracted image from the G channel image of the fat-enhanced processing region before blurring to create a difference image. The method of claim 3, comprising:

5. In addition to the steps of creating a fat-enhanced image from the pork cross-sectional RGB image defined in any one of claims 1 to 4, binarizing the fat-weighted image and separating the region of interest into muscle and marbling regions; L from the muscle area of ​​the pork cross-section RGB image ＊ value and a ＊ obtaining the value; and L ＊ value and a ＊ determining a flesh color score for the muscle region based on the value A method for assessing pork meat color, including:

6. In addition to the steps of creating a fat-enhanced image from the pork cross-sectional RGB image defined in any one of claims 1 to 4, binarizing the fat-weighted image and separating the region of interest into muscle and marbling regions; calculating the fat area percentage and roughness index of the region of interest; and Determining the marbling score based on the fat area percentage and roughness index The method for evaluating pork marbling further comprises:

7. A program that causes a computer to execute the method according to any one of claims 1 to 4.

8. A program that causes a computer to execute the method according to claim 5.

9. A program that causes a computer to execute the method according to claim 6.

10. An image analysis device including a fat enhancement processing unit that performs fat enhancement processing on a pork cross-sectional RGB image, the fat enhancement processing comprising: extracting a fat-enhanced processing region set within the region of interest from the G channel image separated from the pork cross-section RGB image, and performing blurring processing on the extracted fat-enhanced processing region G channel image; creating a difference image using the G channel image of the fat-enhanced processed region after blurring and the G channel image of the fat-enhanced processed region before blurring; adding the difference image to the pork cross-section RGB image to create a composite RGB image; Binarizing the composite RGB image to create a marbling region image; and Adding the marbling region image to the pork cross-section RGB image to create a fat-enhanced image The image analysis device comprising:

11. The creation of the difference image is Adding the G channel image of the fat-enhanced processing area after the blurring processing to the G channel image of the fat-enhanced processing area before the blurring processing to create a composite G channel image; and A fat-enhanced processing region is extracted from the composite G channel image, and the extracted image is subtracted from the G channel image of the fat-enhanced processing region before blurring. The image analysis device according to claim 10, comprising:

12. An image analysis device including a fat enhancement processing unit that performs fat enhancement processing on a pork cross-sectional RGB image, the fat enhancement processing comprising: Multiplying the RGB image of the pork cross section by a G channel image separated from the image to create a meat color-enhanced RGB image; extracting a fat-enhanced processing region set within the region of interest from the G channel image separated from the flesh-color-enhanced RGB image, and performing blurring processing on the G channel image of the extracted fat-enhanced processing region; creating a difference image using the G channel image of the fat-enhanced processed region after blurring and the G channel image of the fat-enhanced processed region before blurring; adding the difference image to the flesh-enhanced RGB image to create a composite RGB image; Binarizing the composite RGB image to create a marbling region image; and Adding the marbling region image to the pork cross-section RGB image to create a fat-enhanced image The image analysis device comprising:

13. The creation of the difference image is Adding the G channel image of the fat-enhanced processing area after the blurring processing to the G channel image of the fat-enhanced processing area before the blurring processing to create a composite G channel image; and A fat-enhanced processing region is extracted from the composite G channel image, and the extracted image is subtracted from the G channel image of the fat-enhanced processing region before blurring. The image analysis device according to claim 12, comprising:

14. The image analysis device according to any one of claims 10 to 13, further comprising an image analysis unit that performs a flesh color analysis on the fat-enhanced image, wherein the flesh color analysis binarizing the fat-weighted image and separating the region of interest into muscle and marbling regions; L from the muscle area of ​​the pork cross-section RGB image ＊ value and a ＊ Obtaining the value, and L ＊ value and a ＊ determining a flesh color score for the muscle region based on the value The device comprising:

15. The image analysis device according to any one of claims 10 to 13, further comprising an image analysis unit that performs marbling analysis on a fat-enhanced image, binarizing the fat-weighted image and separating the region of interest into muscle and marbling regions; Calculating the fat area percentage and roughness index of the area of ​​interest; and Determine the marbling score based on the fat area percentage and roughness index. The device comprising:

Images