Pork color accurate scoring model based on pork color standard scoring board image and rapid construction method thereof
Through image processing and feature parameter extraction based on the pork meat color standard score board, the ridge regression model was constructed, which solved the problems of accurate and robustness of meat color grade prediction in the existing technology, and achieved efficient and fast meat color grade scores, replacing manual labor Rating improves the accuracy and rate of ratings.
Patent Information
- Application Number
- CN202411291476.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-14
- Publication Date
- 2024-12-24
AI Technical Summary
The prior art lacks an accurate and fast pork meat color rating evaluation model that is full-scale, equal-sample, easy to calibrate and does not rely on the population sample number, resulting in limited accuracy and robustness of meat color rating prediction.
Using a rapid construction method based on the pork meat color standard scoring board image, images were collected through a closed scanner, image processing and feature parameter extraction were used to use MATLAB to construct a meat color scoring model based on ridge regression, and a one-leave cross-validation and constant term calibration were performed. An independent and highly generalized flesh color scoring model was obtained.
It realizes high accuracy and high efficiency of meat color rating prediction, with small errors and easy calibration of model parameters. It can quickly build an accurate meat color scoring model, replace artificial sensory scoring, and improve the accuracy and rate of meat color scoring.
Smart Images

Figure CN119180873A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of meat quality evaluation in livestock breeding, and specifically relates to pork color standard scoring plate and actual meat sample image acquisition, image processing, image feature parameter extraction, and pork color scoring model construction and application based on image feature parameters. Background Art
[0002] The pig farming industry occupies a dominant position in the global meat production. According to statistics, the pig farming industry can currently provide about 33% of the world's consumption meat. With the improvement of people's living standards and the upgrading of meat consumption, the demand for high-quality pork has become an inevitable trend of people's consumption. Pork quality is usually evaluated by a series of indicators, such as meat color, marbling, pH value, drip loss, water holding capacity, tenderness, flavor, etc. Among these evaluation indicators, meat color is the most important indicator. The difference in meat color will directly affect consumers' perception and purchasing decisions. Unsatisfactory meat color will also cause huge economic losses to meat consumption. Therefore, pork color has been a long-term concern of farms, slaughterhouses, retailers and consumers. It is also expected to develop accurate, fast and objective pork color evaluation methods and apply pork color evaluation to provide valuable information for all links of pig production, including breeding, commercial pig production, slaughtering and sales, and consumption, and provide guidance for production decisions in all links.
[0003] Meat color measurement methods are mainly divided into two categories: visual inspection and instrumental measurement. Colorimeter measurement is a traditional instrumental measurement method widely used for meat color evaluation. It is an objective meat color characterization method that can collect indicators including L* (indicating brightness), a* (indicating redness) and b* (indicating yellowness), hue angle H0 (evaluating color change) and saturation index or chromaticity C* (evaluating red intensity). The colorimeter method of measuring meat color requires that the measurement surface of the meat must be uniform and the measurement area must be very small (~2-5cm 2 ), and the color grade of meat cannot be directly obtained. It is worth noting that studies have shown that there are certain differences in the meat color values measured by different colorimeters, which indicates that the final output results of meat color grade may be inconsistent in different instrument measurement systems.
[0004] Visual evaluation of meat color is the "basic standard" of color measurement because they are closely related to consumers' perceptual evaluation and serve as the benchmark for comparison of instrumental measurements. Although the visual evaluation of meat color is affected by many conditions, such as lighting type and angle, environmental differences, and human factors, well-trained expert members can obtain objective and reliable visual evaluation results of meat color under standardized conditions. In addition, the visual evaluation of meat color usually uses a standard meat color scoring plate as a reference material, such as the pork color and marbling standards developed by the National Pork Producers Council (NPPC) in 1999, to improve the reliability of the visual evaluation of meat color through subjective evaluation. Nevertheless, there are still some inevitable limitations in visual sensory evaluation, such as the low efficiency of evaluation by trained experts or consumers, easy visual fatigue, and the accuracy is often affected by the sensitivity of individual eyes.
[0005] In recent years, some new meat quality evaluation technologies have emerged, the typical one is the computer vision system (CVS), which has obvious advantages such as fast, consistent, objective, non-invasive and economical. In terms of meat color measurement, the advantage of CVS is that the measurement results obtained from the image are closer to the real color. This argument has been confirmed in the study of meat color measurement of multiple species. Therefore, CVS is a very promising measurement method for obtaining the real meat color value. It is particularly noteworthy that although various methods can measure meat color value indicators such as L*, a* and b*, there is currently a lack of automatic and accurate prediction models for meat color grade evaluation. In terms of predictive model construction, traditional predictive model construction methods rely on large-scale population samples, and the performance of the model will be affected by the source of the animal population. For example, in 2018, Sun et al. used 1,400 pigs from 7 different programs to develop a model for meat color grade. The accuracy of the developed model in different pig populations ranged from 75.0% to 92.5%, with an average accuracy of 78.9%, indicating that the accuracy and robustness of the model constructed by the traditional method are still limited.
[0006] CN1751812A discloses a method for grading pork color quality. The scheme collects fresh meat materials, uses a digital camera to collect photos, and uses a computer to measure the L*, a* and b* values of the photos to perform computer intelligent grading of pork. However, since the scheme has a large number of modeling samples, the range is concentrated between 3.0 and 5.0 points, the sample size distribution is uneven at each scoring level, and the model parameters are greatly dependent on the meat sample group, the scoring model has a narrow range, depends on the characteristics of the meat sample group, and the model parameters are difficult to calibrate, which cannot meet the actual application requirements of accuracy and speed.
[0007] Therefore, there is an urgent need to develop a new pork color scoring model that is accurate, fast, and universal, with a full range, equal samples, easy to calibrate, and independent of the number of group samples. Summary of the invention
[0008] In view of the above background, an object of the present invention is to provide an accurate meat color scoring model and a rapid construction method based on the pork meat color standard scoring plate image. Another object of the present invention is to apply the constructed accurate meat color scoring model to the evaluation of pork meat color grade.
[0009] In order to achieve the above technical objectives, the present invention provides the following technical solutions:
[0010] The present invention first discloses a method for quickly constructing a meat color accurate scoring model based on a pork meat color standard scoring plate image, which specifically comprises the following steps:
[0011] (1) Image acquisition of pork color standard scoring plate
[0012] The images of the meat color standard scoring plate developed by the National Pork Producers Council (NPPC) of the United States were collected using a scanner with a closed scanning space.
[0013] (2) Actual meat sample image acquisition
[0014] A cross-sectional muscle block of the longissimus dorsi muscle with a thickness of 2.0-3.0 cm was obtained from the 3rd to 4th rib from the last left half of the test pig, and a small number of real meat sample images were collected using a scanner with a closed scanning space.
[0015] (3) Flesh color sensory scoring
[0016] Using the meat color standard scoring board developed by the National Pork Producers Council (NPPC) of the United States as a reference, trained experts scored the meat color grade of the longissimus dorsi muscle cross-section with a thickness of 2.0-3.0 cm between the 3rd and 4th ribs from the last of the left half of the test pigs.
[0017] (4) Image processing
[0018] Using MATLAB2020, the collected standard scoring plate image and the actual meat sample image were binarized based on the OTSU algorithm, and the background and target of the binary image were distinguished by the maximum area characteristics of the pixels in the connected area of the image, and a meat sample color target image with a black background was obtained; then, the edge white fat tissue was removed according to the color characteristics and area characteristics of the color target image, the image was binarized again, and the binary image was modified by the morphological algorithm, and the binary image obtained after the final processing was restored to a color image, so as to obtain a color image with only the remaining muscle part.
[0019] (5) Image feature parameter extraction
[0020] MATLAB2020 was used to extract the three characteristic parameters commonly used to describe color in the Lab color space, namely L*, a* and b*, from the processed standard scoring plate images and actual meat sample images.
[0021] (6) Rapid construction of an accurate meat color scoring model based on pork meat color standard scoring plate images
[0022] When the pork color standard scoring board model is constructed using the pork color standard scoring board, the sample size is small, but there is a certain collinearity problem between the characteristic parameters. Therefore, the present invention uses the three characteristic parameters L*, a*, and b* extracted from 6 pork color standard scoring boards (NPPC international pork color standard scoring board, 6-point system), adopts the ridge regression method with regularization function to construct a three-variable regression model, and performs leave-one-out cross validation, selects the regularization parameter λ corresponding to the minimum mean square error, and uses the model coefficient corresponding to λ as the initial model coefficient, thereby obtaining a meat color scoring model with independent characteristic parameters and stronger generalization ability.
[0023] In addition, the absolute values of the characteristic parameters of the actual meat sample and the meat color standard scoring plate on the Lab color coordinate axis will show certain differences. Therefore, the present invention further uses a small amount of characteristic parameters of the actual meat sample image to quickly calibrate the constant term in the above-constructed meat color scoring model. The specific calibration method is as follows: a small amount of sample data is selected from the actual meat sample with a large amount of meat color grade samples to calibrate the model constant term, and finally a corrected meat color grade scoring model based on 6 standard scoring plate images is obtained. The present invention adopts the determination coefficient (R 2 ) and root mean square error (RMSE) were used to evaluate the performance of the developed model. The developed meat color accurate scoring model R 2 >0.95, RMSE<0.40.
[0024] The present invention also discloses a meat color accurate scoring model based on the pork meat color standard scoring plate image, which is realized by the construction method proposed in this application.
[0025] As a preferred meat color accurate scoring model, the model is:
[0026] CS=-0.086L*+0.165a*-0.273b*+7.79
[0027] Where CS is the meat color standard score, which corresponds to the meat color standard score board developed by the National Pork Producers Council (NPPC) of the United States; L*, a*, and b* are the characteristic parameters of the meat sample image to be tested in the Lab color space.
[0028] Preferably, the meat sample to be tested is: a cross-sectional muscle block of the longissimus dorsi muscle with a thickness of 2.0-3.0 cm taken from the third to fourth ribs from the left half carcass of the pig after slaughter;
[0029] The image of the meat sample to be tested is based on the real image of the meat sample to be tested acquired by the scanner, and is obtained through the following steps:
[0030] 1) Based on the OTSU algorithm, the collected real image of the meat sample to be tested is binarized, and the background and target of the binary image are distinguished by the maximum area feature of the pixels in the connected area of the image, so as to obtain a colored target image of the meat sample with a black background;
[0031] 2) Remove the edge white fat tissue according to the color characteristics and area characteristics of the color target image, binarize the image again, and modify the binary image using a morphological algorithm;
[0032] 3) The binary image obtained after the final processing is restored to a color image, and a color image of only the remaining muscle part is obtained as the meat sample image to be tested.
[0033] Beneficial Effects of the Invention
[0034] (1) The pork color scoring model constructed by the present invention using the pork color standard scoring board has high prediction accuracy and small error. A certain number of test pork samples were collected and scored using the developed meat color scoring model; at the same time, well-trained experts performed artificial meat color sensory scoring on the test pigs, and the Pearson correlation coefficient (R P ), Spearman correlation coefficient (R S ), root mean square error (RMSE), residual value distribution, slope of the fitting equation and significance test of the intercept are used to evaluate the prediction ability of the model. The accuracy Rp and Rs are both greater than 0.85, and the RMSE is less than 0.30. Based on the manual scoring scale of 0.5 points, the absolute error between the predicted value and the true value of the model developed by the present invention is less than 0.5 points, accounting for more than 95%, and the accuracy of meat color grade classification is very high.
[0035] (2) The present invention is based on 6 pork meat color standard scoring plate images, and only needs to collect a small amount of actual meat samples for calibration to obtain an accurate meat color scoring model. The model construction method is fast, the model parameters are easy to calibrate and do not rely on large population samples.
[0036] (3) By integrating the meat color scoring model developed in the present invention with the automatic extraction function of characteristic parameters of images collected by a scanner, an accurate, fast and intelligent software for pork meat color grading can be developed. This software can replace manual meat color sensory scoring, improve the accuracy and speed of meat color scoring, and achieve objective evaluation. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] Figure 1 Processing process of pork color standard scoring plate image and actual meat sample image
[0038] Figure 2 Fitting equation diagram and residual distribution diagram of model prediction value and artificial meat color sensory score value DETAILED DESCRIPTION
[0039] In order to enable personnel in the relevant technical field of the present invention to better understand the technical solutions in the present invention, the present invention is further described below in conjunction with preferred embodiments. The described embodiments are only a part of the embodiments of the present invention, not all of them, and the present invention is not limited to the following preferred embodiments.
[0040] Example 1: A method for rapidly constructing an accurate meat color scoring model based on pork meat color standard scoring plate images
[0041] 1.1 Test materials
[0042] (1) Image of pork color standard scoring board
[0043] A scanner with a closed scanning space (model: EPSON V370) was used to collect images of the six pork color standard scoring plates developed by the National Pork Producers Council (NPPC). The scanner light source is an LED light source with a color temperature of 6500K. To avoid reflection of the cover, the inside of the cover is a black photographic cloth, and the image acquisition resolution is 400dpi. Before collecting pork color images, the scanner needs to be calibrated using an IT8 color card to ensure that the collected images have a high degree of color restoration compared to the real objects.
[0044] (2) Actual meat sample images
[0045] A cross-sectional muscle block of the longissimus dorsi muscle with a thickness of about 2.0-3.0 cm was taken from the 3rd to 4th ribs from the left half of the test pig's carcass. The same method as the pork color standard scoring board image was used to collect real meat sample images of 5 test pigs, and the artificial meat color sensory score was 3.5 points.
[0046] 1.2 Image processing and image feature parameter extraction
[0047] (1) Image processing
[0048] Using MATLAB2020, the collected original standard scoring plate image and the actual meat sample image are binarized. The background in the binary image is removed according to the OTSU method and the area characteristics of the connected area in the image, and the color is restored to obtain a color image with only the meat sample remaining. Then, the edge white fat tissue is removed according to the color characteristics and area characteristics of the meat sample, the image is binarized again, and the binary image is modified using a morphological algorithm. The binary image obtained after the final processing is restored to a color image, and a color image with only the muscle part remaining is obtained. The processing process of the pork meat color standard scoring plate image and the actual meat sample image is as follows: Figure 1 shown.
[0049] (2) Image feature parameter extraction
[0050] MATLAB2020 was used to extract three characteristic parameters commonly used to describe color in the L*, a* and b*Lab color space from the processed standard scoring plate images and actual meat sample images.
[0051] 1.3 Construction of accurate meat color scoring model based on pork meat color standard scoring plate image and rapid calibration of model parameters
[0052] The model was constructed by using the three characteristic parameters extracted from six pork color standard scoring boards (NPPC international pork color standard scoring board, 6-point system) and the ridge regression method with regularization function. The leave-one-out cross-validation was performed and the regularization parameter λ corresponding to the minimum mean square error was selected. The model coefficient corresponding to λ was used as the initial model coefficient to obtain a meat color scoring model with stronger generalization ability.
[0053] Furthermore, the present invention uses the characteristic parameters of actual meat sample images of 5 test pigs with an artificial meat color sensory score of 3.5 points to quickly calibrate the constant term in the meat color scoring model constructed above, and finally obtains a calibrated meat color grade scoring model based on 6 standard scoring plate images. The determination coefficient (R) between the meat color score value predicted by the meat color scoring model and the artificial sensory meat color score is 2 ) and root mean square error (RMSE) were used to evaluate the model performance.
[0054]
[0055] The initial model and calibrated model and performance evaluation based on the pork color standard scoring plate developed by the present invention are shown in Table 1:
[0056] Table 1 Initial model and calibrated model and performance evaluation based on pork color standard scoring plate
[0057]
[0058] Note: Model_C is the model with the constant term corrected.
[0059] Example 2: Performance evaluation of the meat color accurate scoring model based on the pork meat color standard scoring plate image and evaluation of the actual application effect
[0060] 2.1 Test materials
[0061] A 2.0-3.0 cm thick longissimus dorsi muscle cross section was taken from the third to fourth ribs from the left half of the carcass of 108 experimental pigs, and the image was collected using a scanner with a closed scanning space (model: EPSON V370). The scanner light source is an LED light source with a color temperature of 6500K. To avoid reflection from the cover, the inside of the cover is a black photographic cloth, and the image acquisition resolution is 400dpi. Before collecting pork meat color images, the scanner needs to be calibrated using an IT8 color card to ensure that the collected image has a high degree of color restoration compared to the actual object.
[0062] 2.2 Image processing and image feature parameter extraction
[0063] (1) Image processing
[0064] Using MATLAB2020, the collected original standard scoring plate image and the actual meat sample image were binarized. The background in the binary image was removed according to the OTSU method and the area characteristics of the connected area in the image, and the color was restored to obtain a color image with only the meat sample remaining. Then, the white fat tissue at the edge was removed according to the color characteristics and area characteristics of the meat sample, the image was binarized again, and the binary image was modified using a morphological algorithm. The binary image obtained after final processing was restored to a color image to obtain a color image with only the remaining muscle part.
[0065] (2) Image feature parameter extraction
[0066] MATLAB2020 was used to extract three characteristic parameters commonly used to describe color in the L*, a* and b*Lab color space from the processed actual meat sample images.
[0067] (3) Flesh color sensory scoring
[0068] Using the standard meat color scoring board developed by the National Pork Producers Council (NPPC) of the United States as a reference, well-trained experts scored the meat color grade of the longissimus dorsi muscle cross-section with a thickness of about 2.0-3.0 cm between the 3rd and 4th ribs from the left half of the carcass of 108 test pigs.
[0069] (4) Performance evaluation of meat color scoring model
[0070] The L*, a* and b* meat color values extracted from 108 test pigs were substituted into the corrected model Model_C (CS = -0.086L* + 0.165a* - 0.273b* + 7.79) developed in Example 1 to obtain the predicted meat color score of each test pig. The Pearson correlation coefficient (R) between the meat color score predicted by the model and the artificial sensory meat color score was analyzed. P ), Spearman correlation coefficient (R S), root mean square error (RMSE), residual value distribution, slope and intercept significance test of the fitting relationship were used to evaluate the performance of the meat color scoring model. The results showed that the R value between the meat color score predicted by the model and the manual sensory meat color score was P and R S Both reached 0.86, and the RMSE was 0.25; and the meat color score classification accuracy evaluation results combined with Table 2 show that, based on the 0.5-point scale of artificial sensory meat color scoring, the absolute error between the predicted value and the true value of the model developed by the present invention is less than 0.5 points, which can achieve 100% accurate classification.
[0071] Table 2 Performance evaluation of the meat color scoring model based on pork meat color standard plate image and evaluation of actual application effect
[0072]
[0073] .
[0074] In addition, the slope and intercept of the fitting equation between the constructed accurate meat color scoring model prediction value and the artificial meat color sensory score value were tested for significance with 1 and 0, respectively. The results showed that the slope of the fitting equation was 0.950, which was not significantly different from 1 (P=0.328), and the intercept was -0.026, which was not significantly different from 0 (P=0.898). The results of the fitting equation and the scatter plot of the meat color scoring model prediction value and the artificial meat color sensory score value are as shown in the figure. Figure 2 The residual value distribution results show that the absolute difference between the model prediction value and the artificial meat color sensory score value is within 1.96 standard deviations ( Figure 2 b), which further demonstrates that the model developed by the present invention has very good predictive ability.
[0075] Although the present invention has been described in detail, it should be understood by those skilled in the art that modifications within the spirit and scope of the present invention will be obvious. For example, the subject matter, spirit and scope of the present invention may be modified in various ways to adapt to specific situations, materials, material combinations or method steps. All of these changes are included within the scope of the claims of the present invention. And making slight changes or modifications using the above disclosed technical content is equivalent to equivalent implementation cases and falls within the scope of the technical solution.
Claims
1. A method for quickly constructing a meat color accurate scoring model based on a pork meat color standard scoring plate image, the basic feature of which is that it includes the following steps: S1, meat color standard scoring plate image acquisition; S2, under the same image acquisition conditions as S1, perform image acquisition of actual meat samples; S3, meat color sensory scoring, using the meat color standard scoring plate image as a reference, the actual meat sample is scored for meat color; S4, image processing; S5, extracting image feature parameters, extracting three feature parameters of L*, a* and b* from the image obtained in S4; S6. Based on the three feature parameters of L*, a* and b*, a meat color accurate scoring model based on the pork meat color standard scoring plate image is constructed.
2. The method according to claim 1, characterized in that In S1, a scanner with a closed scanning space was used to capture the image of the meat color standard scoring plate developed by the National Pork Producers Council (NPPC).
3. The method according to claim 2, characterized in that The scanner light source is an LED light source with a color temperature of 6500K, and the inner side of the cover is a black photographic cloth. Before image acquisition, the scanner is color calibrated using an IT8 color card to ensure that the acquired image has a high degree of color restoration compared to the actual object.
4. The method according to claim 1, characterized in that In S2, the actual meat sample was obtained from: After the pig was slaughtered, a cross-sectional muscle block of the longissimus dorsi muscle with a thickness of 2.0-3.0 cm was obtained from the 3rd to 4th ribs from the left half of the pig carcass.
5. The method according to claim 1, characterized in that In S4, the specific steps of image processing are: 1) Based on the OTSU algorithm, the collected standard scoring plate image and the actual meat sample image are binarized, and the background and target of the binary image are distinguished by the maximum area feature of the pixels in the connected area in the image, so as to obtain a meat sample color target image with a black background; 2) Remove the edge white fat tissue according to the color characteristics and area characteristics of the color target image, binarize the image again, and modify the binary image using a morphological algorithm; 3) The binary image obtained after the final processing is restored to a color image to obtain a color image of only the remaining muscle part.
6. The method according to claim 1, characterized in that In S6, the ridge regression method with regularization function is used to construct a three-variable regression model, and leave-one-out cross-validation is performed to select the regularization parameter λ corresponding to the minimum mean square error, and the model coefficient corresponding to λ is used as the initial model coefficient, thereby obtaining a meat color accuracy scoring model with stronger generalization ability.
7. The method according to claim 6, characterized in that In S6, a small amount of sample data is selected from actual meat samples with a large sample size of meat color grades to calibrate the model constant term, and finally a corrected meat color accurate scoring model is obtained.
8. A meat color accurate scoring model based on pork meat color standard scoring plate image, characterized in that It is constructed by the method according to any one of claims 1-6.
9. A meat color accurate scoring model based on pork meat color standard scoring plate image, characterized in that The model is: CS=-0.086L*+0.165a*-0.273b*+7.79 Where CS is the meat color standard score, which corresponds to the meat color standard score board developed by the National Pork Producers Council (NPPC) of the United States; L*, a*, and b* are the characteristic parameters of the meat sample image to be tested in the Lab color space.
10. The model according to claim 9, characterized in that The meat sample to be tested is: a cross-sectional muscle block of the longissimus dorsi muscle with a thickness of 2.0-3.0 cm is taken from the third to fourth ribs from the left half carcass of the pig after slaughter; The image of the meat sample to be tested is based on the real image of the meat sample to be tested acquired by the scanner, and is obtained through the following steps: 1) Based on the OTSU algorithm, the collected real image of the meat sample to be tested is binarized, and the background and target of the binary image are distinguished by the maximum area feature of the pixels in the connected area of the image, so as to obtain a colored target image of the meat sample with a black background; 2) Remove the edge white fat tissue according to the color characteristics and area characteristics of the color target image, binarize the image again, and modify the binary image using a morphological algorithm; 3) The binary image obtained after the final processing is restored to a color image, and a color image of only the remaining muscle part is obtained as the meat sample image to be tested.
Citation Information
Cited By
Morphological evaluation method for cattle frozen embryos
CN119991667A
