Meat composition calculation method and system

Abstract

A meat composition calculation method. The method comprises: executing dual-energy X-ray scanning on a piece of meat, so as to obtain scanned image data of the piece of meat; converting the expression form of the scanned image data, so as to obtain image conversion data of the piece of meat; training a meat composition calculation model on the basis of a plurality of meat samples, so as to obtain a trained meat composition calculation model; and on the basis of the image conversion data, using the trained meat composition calculation model to calculate meat composition information of the piece of meat.

Description

Meat quality component calculation method and system

[0001] This application claims priority to Chinese patent application No. 202311651948.9 filed on December 5, 2023, the entire contents of which are incorporated by reference into this application. Technical Field

[0002] The present disclosure relates to the field of detection technology, and more particularly to a method and system for calculating meat quality components, and more particularly to a method and system for calculating meat quality components based on dual-energy X-ray scanning. Background Art

[0003] In the livestock industry, such as pigs, sheep, and cattle, obtaining meat composition information from individual animals provides valuable feedback. This information can be used to guide individual selling prices and individual value-added farming strategies. Meat composition analysis is used to determine the weight and / or percentage of muscle, fat, and bone, either locally or globally. Dual-energy X-ray imaging is a non-invasive method that has been used in some meat composition analysis studies, achieving promising results.

[0004] Current methods for calculating meat composition using dual-energy X-rays primarily rely on real meat / bone slices with known composition, or meat / bone phantoms made of polymer materials, as calibration elements. These elements are then scanned to create a corresponding lookup table. This method is relatively straightforward, but the challenge lies in producing calibration elements that meet the requirements, especially since real meat / bone slices are difficult to preserve. Summary of the Invention

[0005] According to one aspect of the present disclosure, a method for calculating meat quality composition is provided, the method comprising: performing dual-energy X-ray scanning on a meat piece to obtain scanned image data of the meat piece; converting the expression form of the scanned image data to obtain image conversion data of the meat piece; training a meat quality composition calculation model based on multiple meat samples to obtain a trained meat quality composition calculation model; and calculating meat quality composition information of the meat piece using the trained meat quality composition calculation model based on the image conversion data.

[0006] In an embodiment, the method further comprises: converting the scanned image data of the meat piece into one or more expression forms to obtain one or more image conversion data of the meat piece corresponding to the one or more expression forms.

[0007] In an embodiment, the expression forms include dual-energy form, grayscale form, RGB form, atomic number form and multiple base material form.

[0008] In an embodiment, the method further comprises: calibrating the scanned image data of the meat piece or the image conversion data of the meat piece to unify it to an imaging standard corresponding to the trained meat quality composition calculation model.

[0009] In an embodiment, the method further includes: performing image processing on the scanned image data of the meat piece to divide at least one area to be detected from the scanned image data of the meat piece; and inputting the image conversion data of the meat piece corresponding to the at least one area to be detected into the trained meat quality composition calculation model to calculate the meat quality composition information of the at least one area to be detected.

[0010] In an embodiment, the training includes: performing the dual-energy X-ray scanning on the meat sample to obtain the scanned image data of the meat sample, and converting the expression form of the scanned image data of the meat sample to obtain the image conversion data of the meat sample; performing high-precision measurement on the meat sample to obtain the true value of the meat quality composition of the meat sample; and training the meat quality composition calculation model based on the image conversion data of the meat sample and the true value of the meat quality composition to obtain the trained meat quality composition calculation model.

[0011] In an embodiment, the training further comprises: calibrating the scanned image data of the plurality of meat samples or the image conversion data of the meat samples to unify them to the same reference imaging standard.

[0012] In an embodiment, the training also includes: converting the scanned image data of the meat sample into one or more expression forms to obtain image conversion data of one or more meat samples corresponding to the one or more expression forms; and obtaining a plurality of the trained meat composition calculation models based on the image conversion data of a plurality of the meat samples.

[0013] In an embodiment, the training also includes: establishing a mapping relationship between the scanned image data of the meat sample and the true value of the meat quality component for each of the multiple meat samples; and based on the mapping relationship of the multiple meat samples, training the meat quality component calculation model by calculating the difference between the calculated value of the meat quality component of the same meat sample and the true value of the meat quality component, and updating the model parameters of the meat quality component calculation model based on the difference.

[0014] In an embodiment, the method for performing the high-precision measurement includes: obtaining the true value of the meat quality composition of the meat sample using a high-precision imaging device; and / or obtaining the true value of the meat quality composition of the meat sample by manually cutting the meat sample and measuring.

[0015] In an embodiment, the true value of the meat composition includes at least one of the muscle weight, fat weight, bone weight, total weight, muscle weight percentage, fat weight percentage, and bone weight percentage of the meat sample.

[0016] The calculated meat composition information includes at least one of muscle weight, fat weight, bone weight, total weight, muscle weight percentage, fat weight percentage and bone weight percentage; and the method also includes: output processing the calculated meat composition information to output at least one of muscle weight, fat weight, bone weight, total weight, muscle weight percentage, fat weight percentage and bone weight percentage to the outside.

[0017] According to another aspect of the present disclosure, a meat quality composition calculation system is provided, which implements the meat quality composition calculation method disclosed herein, and the system includes: an image scanning module, which is configured to perform dual-energy X-ray scanning on a meat piece or a meat sample to obtain scanned image data of the meat piece or the meat sample; a conversion module, which is configured to convert the expression form of the scanned image data to obtain image conversion data of the meat piece or the meat sample; a true value acquisition module, which is configured to perform high-precision measurement on the meat sample to obtain the true value of the meat quality composition of the meat sample; and a meat quality composition calculation model, which is trained based on the image conversion data of a plurality of the meat samples and the meat quality composition true values ​​to obtain a trained meat quality composition calculation model, and uses the trained meat quality composition calculation model to calculate the meat quality composition information of the meat piece based on the image conversion data of the meat piece.

[0018] In an embodiment, the system further comprises: a calibration module configured to calibrate the scanned image data or the image conversion data to unify them to the same reference imaging standard.

[0019] In an embodiment, the system further includes: an image segmentation module, which is configured to perform image processing on the scanned image data of the meat piece to divide at least one area to be detected from the scanned image data of the meat piece, so that the system obtains image conversion data of the meat piece corresponding to the at least one area to be detected.

[0020] In an embodiment, the system further comprises: a mapping establishment module configured to establish, for each of the meat samples, a mapping relationship between the scanned image data of the meat sample and the true value of the meat quality component.

[0021] In an embodiment, the system further includes: an output processing module, which is configured to perform output processing on the calculated meat composition information to output at least one of muscle weight, fat weight, bone weight, total weight, muscle weight percentage, fat weight percentage and bone weight percentage to the outside.

[0022] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present disclosure, nor are they intended to limit the scope of the present disclosure. Other features of the present disclosure will become readily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present disclosure, wherein:

[0024] FIG1 shows a flow chart of a method for calculating meat quality components according to an embodiment of the present disclosure.

[0025] FIG2 shows a flowchart of a method for training a meat quality composition calculation model according to an embodiment of the present disclosure.

[0026] FIG3 shows an example of a training flowchart of a meat quality composition calculation model according to an embodiment of the present disclosure.

[0027] FIG4 shows an example of a meat quality composition calculation system according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0028] To more clearly illustrate the objectives, technical solutions, and advantages of the present disclosure, embodiments of the present disclosure will be described in detail below with reference to the accompanying drawings. It should be understood that the following description of the embodiments is intended to explain and illustrate the overall concept of the present disclosure and should not be construed as limiting the present disclosure. In the specification and drawings, the same or similar reference numerals refer to the same or similar parts or components. For the sake of clarity, the drawings are not necessarily drawn to scale, and some well-known parts and structures may be omitted in the drawings.

[0029] Unless otherwise defined, technical or scientific terms used in this disclosure should have the ordinary meaning understood by a person of ordinary skill in the art to which this disclosure belongs. The terms "first," "second," and similar expressions used in this disclosure do not denote any order, quantity, or importance, but are simply used to distinguish different components. The terms "a" or "an" do not exclude a plurality. "Include" or "comprising" and similar expressions mean that the element or object preceding the word includes the elements or objects listed after the word and their equivalents, but do not exclude other elements or objects. "Connected" or "connected" and similar expressions are not limited to physical or mechanical connections but can include electrical connections, whether direct or indirect. "Up," "down," "left," "right," "top," or "bottom" are used only to indicate relative positional relationships; if the absolute position of the described object changes, the relative positional relationship may also change accordingly. When an element, such as a layer, film, region, or substrate, is referred to as being "on" or "under" another element, the element may be "directly" "on" or "under" the other element, or intervening elements may be present.

[0030] The present disclosure proposes a meat quality composition calculation method and system. The method obtains meat quality composition information by utilizing a meat quality composition calculation model trained by deep learning, thereby avoiding the production of meat slice / bone slice calibration pieces in the traditional method, and improving information accuracy and economic benefits.

[0031] FIG1 shows a flowchart of a meat quality component calculation method 1000 according to an embodiment of the present disclosure.

[0032] As shown in FIG1 , at step S101, a dual-energy X-ray imaging device may be used to perform a dual-energy X-ray scan on a meat item to be inspected, thereby obtaining scanned image data 10 of the meat item. According to embodiments of the present disclosure, "meat item" herein may include, for example, pork, beef, or lamb, and, depending on practical needs, may refer to the entire meat item or a portion thereof (e.g., a front leg or hind leg). By way of example, scanned image data 10 may be two-dimensional imaging data.

[0033] At step S102, the obtained scanned image data 10 of the meat piece can be calibrated to be unified under a reference imaging standard. In this article, the "reference imaging standard" may refer to an imaging standard specified according to the imaging standard of one or more dual-energy X-ray imaging devices used, so that the obtained scanned image data can be calibrated according to the specified imaging standard, thereby being unified under the imaging standard of the same device, and thereby eliminating the inaccuracy of the calculation results that may be caused by the difference in imaging standards between multiple devices. As an example, the reference imaging standard may be the imaging standard used by one of the multiple dual-energy X-ray imaging devices. Or as another example, in the case of using only one dual-energy X-ray imaging device for detection, and in the case of a high requirement for the accuracy of the results, the reference imaging standard may be a pre-set imaging standard. However, the embodiments of the present disclosure are not limited to this.

[0034] Here, the reference imaging standard can be an imaging standard corresponding to a trained meat quality composition calculation model (described below). That is, since the scanned image data 10 needs to be applied to the trained meat quality composition calculation model 100L to calculate the meat quality composition information 200 in the subsequent process, the trained meat quality composition calculation model 100L and the scanned image data 10 should be based on the same imaging standard. In other words, the trained meat quality composition calculation model 100L can be trained based on scanned image data under one imaging standard, so the model 100L will have a higher calculation accuracy for the scanned image data under the imaging standard. In this case, by calibrating the scanned image data 10 obtained from step S101 to the imaging standard corresponding to the trained meat quality composition calculation model 100L to obtain the calibrated scanned image data 10C, the calibrated scanned image data 10C can be better matched with the trained meat quality composition calculation model 100L, thereby ensuring the calculation accuracy and reliability.

[0035] As an example, calibrating the scanned image data 10 may include, under the condition of having the same parameter settings, calculating the difference between the scanned image data of each dual-energy X-ray imaging device and the scanned image data under the reference imaging standard, and recording the difference as a compensation amount. Furthermore, the compensation amounts can be calculated and recorded for a plurality of different parameter setting schemes, respectively, and respectively corresponded to the corresponding dual-energy X-ray imaging devices, thereby forming a compensation mapping table or a compensation trend curve. Thus, the scanned image data 10 can be calibrated according to the compensation amount corresponding to the parameter setting scheme used. However, this is merely an example, and the embodiments of the present disclosure are not limited thereto. The present disclosure may include other suitable calibration methods, as long as the calibrated scanned image data 10C can be in the same imaging standard as the trained meat composition calculation model 100L to be used.

[0036] It should be noted that the meat quality composition calculation method 1000 according to the embodiments of the present disclosure may not include step S102. That is, step S102 is optional. For example, if the accuracy of meat quality composition calculation is not required to be high, the calibration step can be omitted. This can be selected based on specific usage and requirements.

[0037] Alternatively, as another example, step S102 may be performed after step S103. That is, after the conversion step, the above calibration process may be performed on the converted image data 20, thereby achieving uniformity to the reference imaging standard.

[0038] In step S103, the expression form of the calibrated scanned image data 10C is converted to obtain image conversion data 20. As an example, the image conversion data 20 can be dual-energy form data, grayscale form data, RGB form data, atomic number form data, or multiple base material form data, etc. In this article, "dual-energy form data" can be low-energy and high-energy data acquired by a dual-energy X-ray detection system, reflecting the attenuation of the detected object under different radiation energies; "grayscale form data" can be converted from the dual-energy data and is a comprehensive expression of the attenuation under different radiation energies; "atomic number form data" is also calculated based on the dual-energy data and represents the equivalent atomic number of each part of the current detection object; "RGB form data" is the result of pseudo-coloring processing of the grayscale form data and the atomic number form data; "base material form data" is the base material decomposition coefficient obtained from the dual-energy data based on pre-set base material information.

[0039] In one embodiment, the calibrated scanned image data 10C can be converted into one of dual-energy form data, grayscale form data, RGB form data, atomic number form data, and multiple base material form data as image conversion data 20. In this way, by converting the expression form of the scanned image data and providing multiple expression forms for selection, the scanned image data can have multiple different expression types, thereby increasing data diversity. However, for the calculation characteristics of different types of meat pieces, the accuracy and fit of the scanned image data presented in different expression forms may be different. Therefore, when there are multiple expression forms to choose from, one or several expression forms can be selected for conversion according to actual needs (for example, according to the type of meat piece, the characteristics of the part, etc.), so as to improve the calculation accuracy and enable the meat composition calculation method 1000 to have the flexibility to adapt and adjust according to the characteristics of the meat piece.

[0040] Alternatively, the converted representation can be selected based on a trained meat quality composition calculation model 100L. For example, the meat quality composition calculation model 100 can be trained for each data representation (e.g., dual-energy, grayscale, RGB, atomic number, and multiple base material representations) to obtain multiple trained meat quality composition calculation models 100L corresponding to each data representation. After a verification operation, the trained meat quality composition calculation model 100L with the best computational performance can be determined as the model to be used, and the data representation corresponding to the model can be determined. In this case, the calibrated scanned image data 10C can be converted into the determined data representation.

[0041] In another embodiment, the calibrated scanned image data 10C can be converted into multiple representations. In this case, multiple types of image conversion data 20 can be obtained. For example, the multiple types of image conversion data 20 can include at least two of dual-energy data, grayscale data, RGB data, atomic number data, or multiple base material data. In this manner, different meat composition information can be calculated for each piece of meat based on different representations. Specifically, the meat composition information can be calculated multiple times using different data representations for the same piece of meat, resulting in multiple meat composition information results. This allows, for example, the multiple meat composition information results to be averaged to obtain the final meat composition information, ensuring accuracy and reliability of the calculation. Furthermore, if the meat piece to be tested is a whole meat sample, subsequent calculations may require separate calculations for multiple local regions within the sample. As mentioned above, different regions may require different data representations. Therefore, by converting the data into multiple different representations, the appropriate data representation can be selected for each local region, thereby accelerating the calculation process and improving accuracy.

[0042] In step S104, image processing may be performed on the calibrated scanned image data 10C of the meat piece to segment at least one area to be detected from the calibrated scanned image data 10C and obtain image conversion data 20D corresponding to the segmented area to be detected.

[0043] In the case where the meat piece is a whole piece of meat (for example, a whole pork carcass), the calibrated scanned image data 10C of the meat piece obtained is also scanned image data about the meat as a whole. However, in actual meat quality component detection, what may be needed is the meat quality component information of the local part of the meat. Taking pork as an example, it may be necessary to separately detect the meat quality component information of local parts such as the pig's front legs, pig's hind legs, pig's head, pig's buttocks, etc. In this case, the required local areas can be divided from the scanned image data of the meat piece as a whole, that is, the areas to be detected are divided. Then, according to the divided areas to be detected, the image conversion data 20D corresponding to these areas are extracted from the image conversion data 20 to be used as input to the meat quality component calculation model in the subsequent process, so that the meat quality component information can be calculated for each meat part.

[0044] In one embodiment, automated image processing of image data can be performed by using image segmentation and positioning processing technology, and optionally in combination with image noise reduction processing technology and image enhancement processing technology. As an example, the image data can first be subjected to preliminary sharpening processing using an image noise reduction processing method and an image enhancement processing method. Then, the image segmentation and positioning processing method is used to extract multiple feature points from the target area of ​​the image data according to pre-set feature point extraction information, thereby performing region segmentation along the extracted multiple feature points, thereby dividing the area to be detected, and using this area as a mask to obtain the area to be detected 20D of the image conversion data 20. As an example, the feature point extraction information of the corresponding part area can be set based on the geometric structure information of the part to be divided of the meat piece. In addition, the set feature point extraction information can be pre-inputted into the relevant image processing software for retrieval and use when performing image processing. The image data described in this embodiment can be scanned image data 10, calibrated scanned image data 10C, or image conversion data 20.

[0045] In another embodiment, an image segmentation and localization model can be used to perform image processing on the image conversion data 20. For example, the image conversion data of multiple samples can be used to train the image segmentation and localization model, enabling the trained image segmentation and localization model to accurately segment the image based on the input image conversion data, thereby demarcating the desired area. In this embodiment, in step S104, the image conversion data 20 of the meat piece, the scanned image data 10, or the calibrated scanned image data 10C can be input into the trained image segmentation and localization model, which outputs an image segmentation result (i.e., at least one area to be detected) from the model, thereby extracting image conversion data 20D corresponding to the area to be detected.

[0046] It should be noted that step S104 is optional. That is, step S104 can be omitted. For example, when calculating meat quality information for a specific part of the meat, or when the image conversion data 20 itself relates to the local area to be detected, it is more appropriate to use the entire image conversion data 20 as the model input, rather than partitioning the image conversion data 20 to extract partial data. In this case, step S104 is unnecessary and can be omitted.

[0047] In addition, it should be noted that step S104 may be placed between S101 and S102, or between S102 and S103, or between S103 and S105, which can be selected according to actual needs.

[0048] In step S105, the meat quality composition calculation model 100 may be trained based on a plurality of meat samples to obtain a trained meat quality composition calculation model 100L. The training method of the meat quality composition calculation model 100 will be described in detail below with reference to FIG.

[0049] In step S106, the meat quality composition information 200 of the meat piece can be calculated using the trained meat quality composition calculation model 100L based on the image conversion data 20D. As an example, the image conversion data 20D can be input into the trained meat quality composition calculation model 100L, so that the model 100L can solve the image conversion data 20D to calculate the meat quality composition information 200 of the to-be-detected area of ​​the meat piece.

[0050] According to an embodiment of the present disclosure, "meat composition" may refer to muscle composition, fat composition, and bone composition in a piece of meat. As an example, the meat composition information 200 may include at least one of muscle weight, fat weight, bone weight, total weight, muscle weight percentage, fat weight percentage, and bone weight percentage.

[0051] In step S107, the calculated meat quality composition information 200 can be output processed. As an example, "output processing" can mean calculating and converting the meat quality composition information 200, and outputting the result in the meat quality composition representation form required by the user, so as to facilitate the user to compare, measure and grade, etc. For example, the output processing can include converting the information of muscle weight, fat weight, bone weight and total weight into muscle proportion information (i.e., muscle weight percentage), fat proportion information (i.e., fat percentage) and bone proportion information (i.e., bone weight percentage). Thus, after the output processing, at least one of the muscle weight, fat weight, bone weight, total weight, muscle weight percentage, fat weight percentage and bone weight percentage can be output to the outside. As another example, "output processing" can mean converting the meat quality composition information 200 into a format so that it can be received by an external device in a suitable format and displayed, analyzed, etc.

[0052] FIG2 shows a flowchart of a method 2000 for training the meat quality composition calculation model 100 according to an embodiment of the present disclosure.

[0053] In step S201, a plurality of meat samples are obtained, and model training is performed based on the obtained plurality of meat samples. In this article, "meat sample" may refer to meat as a whole or meat parts (such as, forelegs, hind legs, etc.). In one embodiment, in the training of the model, there may be no requirements for meat samples. That is, various types of samples can be selected as meat samples for training, such as whole meat samples, local meat samples of various parts. In another embodiment, when there are higher requirements for the calculation results, meat samples can be obtained for a certain part. As an example, if it is desired to obtain a model for calculating the meat composition of a pig's forelegs, a plurality of pig's forelegs samples can be obtained for training. For example, a pig's foreleg sample can be obtained by decomposing the pork pieces in advance. However, the present disclosure is not limited to the above embodiments, and the required meat samples can be selected according to actual needs, and the plurality of meat samples used for training may include only one sample type, or may include more than two sample types.

[0054] Furthermore, the prepared meat samples need to be as diverse as possible, covering all possible scenarios. For example, the weight of the meat samples should range from the smallest to the largest, the size should range from the longest to the shortest, and the percentage of each component of the meat sample (such as muscle, fat, and bone) should range from the smallest to the largest, etc. Furthermore, if conditions permit, scanned image data of the same meat sample should be collected in as many poses as possible.

[0055] In step S202 , a dual-energy X-ray scan may be performed on the meat sample to obtain scan image data 50 of the meat sample.

[0056] In step S203, the meat sample scanned image data 50 can be calibrated to conform to the reference imaging standard, thereby obtaining calibrated meat sample scanned image data 50C. As described above with reference to step S102, calibration to the same reference imaging standard is performed to eliminate differences in imaging standards between different dual-energy X-ray imaging devices, allowing the calibrated scanned image data to be considered as scanned data from the same dual-energy X-ray imaging device, i.e., based on the same imaging standard. In some cases, calibration may be performed even when only a single dual-energy X-ray imaging device is used, in order to obtain more accurate and reliable inspection results.

[0057] In this way, model training inaccuracies caused by differences between different dual-energy X-ray imaging devices can be reduced, thereby reducing subsequent calculation inaccuracies. In step S202, since multiple meat samples are involved, multiple different dual-energy X-ray imaging devices may be involved. In order to ensure that the meat quality composition calculation model 100 is trained using scan data under a unified imaging standard, it is necessary to set a reference imaging standard and calibrate the scan image data 50 of multiple meat samples according to the set reference imaging standard to obtain calibrated scan image data 50C. As a result, the trained meat quality composition calculation model 100L is trained based on data under the same imaging standard and therefore corresponds to this imaging standard.

[0058] It should be noted that, similar to step S102, in some cases, step S203 may be performed after step S204. That is, after the conversion step, the above-mentioned calibration process may be performed on the converted image data 60 of the meat sample to achieve uniformity to the reference imaging standard.

[0059] In step S204, the representation of the calibrated scanned image data 50C can be converted to obtain image conversion data 60 of the meat sample. As described with reference to step S103, by converting the representation of the scanned image data and obtaining image conversion data of one or more representations, the type of data used and its corresponding characteristics can be enriched. As an example, the calibrated scanned image data 50C can be converted into one or more representations, thereby obtaining one or more image conversion data 60 corresponding to the one or more representations. This allows, in subsequent processes, the meat quality composition calculation model 100 to be trained for each type of image conversion data 60, thereby obtaining one or more trained meat quality composition calculation models 100L. Each trained meat quality composition calculation model 100L corresponds to a respective type of image conversion data 60.

[0060] For example, the image conversion data 60 can be dual-energy data, grayscale data, RGB data, atomic number data, or data in multiple base material formats. Taking grayscale data and RGB data as examples, if models are trained using these two types of image conversion data 60, two models can be obtained: a meat quality composition calculation model 100L trained based on grayscale data and a meat quality composition calculation model 100L trained based on RGB data. In other words, when using these two models, the corresponding data format is required as input for calculation. For example, using the meat quality composition calculation model 100L trained based on grayscale data requires grayscale image conversion data 60.

[0061] In this way, models suitable for different data types can be obtained. Because the structural characteristics of different meat parts may result in different calculation accuracy and compatibility of models for different data types, the best performing model can be selected from multiple trained models for subsequent meat composition calculations based on the specific part being tested, ensuring accuracy and reliability. Thus, by converting data into multiple formats, the flexibility of model usage is increased.

[0062] In step S205, the meat sample can be measured with high precision to obtain the meat quality composition true value 300 of the meat sample. In this article, the "meat quality composition true value" can refer to data that enables the meat quality composition calculation model to calculate the meat quality composition with the required accuracy. That is, when the meat quality composition true value is used, the expected calculation result can be obtained. Therefore, according to an embodiment of the present disclosure, the "meat quality composition true value" can be used as reference data for training the meat quality composition calculation model 100. In other words, the meat quality composition calculation model 100 can be trained with the meat quality composition true value 300, so that the meat quality composition information 200 calculated by the trained meat quality composition calculation model 100L has the same accuracy and reliability as the meat quality composition true value 300.

[0063] As an example, high-precision measurement may include using manual cutting measurement. Manual cutting measurement may include cutting and dividing a meat sample to obtain the true value 300 of the meat quality composition.

[0064] As another example, a high-precision imaging device can be used to obtain the true meat composition value 300. The high-precision imaging device can be a device with a higher resolution than a dual-energy X-ray imaging device. For example, the high-precision imaging device can have higher performance indicators, such as spatial resolution and material resolution, than a dual-energy X-ray imaging device. As an example, the high-precision imaging device can include a high-performance medical CT device.

[0065] As an example, the meat quality composition true value 300 may include at least one of muscle weight, fat weight, bone weight, total weight, muscle weight percentage, fat weight percentage, and bone weight percentage of the meat sample. However, the embodiments of the present disclosure are not limited thereto.

[0066] In step S206, a mapping relationship between the scanned image data 50 and the true meat quality component value 300 can be established for each of the multiple meat samples. In this way, a correspondence can be achieved between the scanned image data 50 of each meat sample and its true meat quality component value 300, thereby training the model based on the scanned image data 50 and the true meat quality component value 300 of the same meat sample, and training can be performed on multiple samples in this manner.

[0067] As an example, the mapping relationships between the scanned image data 50 of multiple meat samples and the true meat quality composition values ​​300 can be stored in the form of a mapping table or a mapping database. During training, the corresponding scanned image data 50 and true meat quality composition values ​​300 can be selected according to the mapping relationships recorded in the mapping table. Furthermore, the mapping relationships for multiple meat samples can also include mappings between the true meat quality composition values ​​300 and the calibrated scanned image data 50C and the converted image data 60. However, the embodiments of the present disclosure are not limited to this, as long as the calculated value of the meat quality composition calculation model 100 based on the scanned image data 50 of a meat sample can correspond to the true meat quality composition value 300 of the meat sample.

[0068] In step S207, the meat quality composition calculation value 400 can be calculated based on the image conversion data 60 of the meat sample, and the meat quality composition calculation model 100 can be trained based on the meat quality composition calculation value 400 and the meat quality composition true value 300 to obtain the trained meat quality composition calculation model 100L.

[0069] FIG3 shows an example of a training flowchart of the meat quality composition calculation model 100 according to an embodiment of the present disclosure.

[0070] As shown in Figure 3, the meat quality composition calculation model 100 can calculate the meat quality composition calculation value 400 for the first time based on the image conversion data 60. The meat quality composition calculation model 100 at this time is an untrained model, so the meat quality composition calculation value 400 calculated for the first time cannot meet the expected effect (such as accuracy). In this case, it is necessary to iteratively train the meat quality composition calculation model 100 based on the true value data, and by modifying the model parameters of the meat quality composition calculation model 100 during each iteration, the calculated result can be gradually approached to the true value data. Ultimately, the model parameters of the meat quality composition calculation model 100 can be corrected to be able to calculate the desired meat quality composition calculation value, that is, the meat quality composition information 200, based on the image conversion data 60. It must be pointed out that the judgment conditions in the current example are merely examples, and the present disclosure may not be limited to this.

[0071] Therefore, as shown in FIG3 , the meat quality composition calculation model 100 can calculate the difference between the first calculated meat quality composition value 400 and the meat quality composition true value 300 (represented as |400-300| in FIG3 ), and then compare the difference with the preset reference data FD to determine whether the difference is less than or equal to the reference data FD (represented as |400-300|≤FD? in FIG3 ). If the difference is not less than or equal to (i.e., greater than) the reference data FD ("No" in FIG3 ), the meat quality composition calculation model 100 can update the model parameters and use the updated meat quality composition calculation model 100 to calculate the meat quality composition calculated value 400 again (i.e., a second time). Then, the difference between the second calculated meat quality composition calculated value 400 and the meat quality composition true value 300 is calculated, and the above-described comparison process is repeated, i.e., the second difference is compared with the reference data FD to determine whether the difference is less than or equal to the reference data FD. If the difference is still not less than or equal to (i.e., greater than) the reference data FD ("No" in FIG3 ), the above parameter updating and calculation process is repeated until the difference is less than or equal to the reference data FD. If the difference is less than or equal to the reference data FD for the second time ("Yes" in FIG3 ), the training of the meat quality composition calculation model 100 is determined to be complete, and the meat quality composition calculation model 100 at this time is determined to be the trained meat quality composition calculation model 100L.

[0072] Optionally, the training process may also include a verification process. As an example, a portion of multiple meat samples, such as a small number of samples, can be selected as meat samples for verification. At the end of training the meat quality composition calculation model 100, as shown in FIG3 , the image conversion data 60 of the meat sample used for verification can be input into the meat quality composition calculation model 100, and a meat quality composition verification value can be calculated. The meat quality composition verification value is then compared with the meat quality composition true value 300. For example, the difference between the two is compared with a preset threshold range. If it is within the threshold range, the meat quality composition calculation model 100 at this time has passed verification, and the meat quality composition calculation model 100 is output as the trained meat quality composition calculation model 100L. If it is outside the threshold range, the meat quality composition calculation model 100 at this time has failed verification, and the training process shown in FIG3 is re-executed. In some cases, a pass rate can be set for the multiple meat samples used for verification to ultimately determine whether the meat quality composition calculation model 100 has been successfully trained.

[0073] In this way, by training the model, a meat quality composition calculation model with output close to the true value level can be obtained. When applied to the network model solving process described in reference step S106, the meat quality composition information 200 can be calculated quickly and accurately in an automated manner.

[0074] The meat quality component calculation model 100 may be a regression network model.

[0075] FIG4 shows an example of a meat quality composition calculation system 3000 according to an embodiment of the present disclosure.

[0076] As shown in Figure 4, the meat quality composition calculation system 3000 can include an image scanning module 1, a calibration module 2, a conversion module 3, an image segmentation module 4, a true value acquisition module 5, a mapping establishment module 6, a meat quality composition calculation model 7 and an output processing module 8.

[0077] According to an embodiment of the present disclosure, the image scanning module 1 can be configured to perform dual-energy X-ray scanning on a piece of meat or a meat sample to obtain scanned image data of the meat piece or the meat sample. The conversion module 2 can be configured to convert the representation of the scanned image data to obtain converted image data of the meat piece or the meat sample. The true value acquisition module 5 can be configured to perform high-precision measurement on the meat sample to obtain the true value of the meat quality composition of the meat sample.

[0078] Furthermore, the meat quality composition calculation model 7 can be trained based on the image conversion data and meat quality composition true values ​​of multiple meat samples to obtain a trained meat quality composition calculation model, and use the trained meat quality composition calculation model to calculate the meat quality composition information of the meat piece based on the image conversion data of the meat piece.

[0079] In one embodiment, the calibration module 2 may be configured to calibrate the scanned image data of the meat piece or meat sample before converting the scanned image data into image conversion data so as to unify the scanned image data to the same reference imaging standard.

[0080] In one embodiment, the image segmentation module 4 can be configured to perform image processing on the scanned image data of the meat piece to segment at least one area to be detected from the scanned image data of the meat piece, so that the system 300 obtains image conversion data of the meat piece corresponding to the at least one area to be detected.

[0081] In one embodiment, mapping establishment module 6 can be configured to establish, for each meat sample, a mapping relationship between the scanned image data of the meat sample and the true value of the meat quality component. This allows meat quality composition calculation model 7 to be trained based on the mapping relationship for multiple meat samples by calculating the difference between the calculated meat quality component value and the true value of the meat quality component for the same meat sample and updating the model parameters of meat quality composition calculation model 7 based on the difference. In some cases, mapping establishment module 6 can be integrated with meat quality composition calculation model 7 into the same module, or can be a separate module from meat quality composition calculation model 7.

[0082] In one embodiment, the output processing module 8 can be configured to perform output processing on the calculated meat composition information to output at least one of muscle weight, fat weight, bone weight, total weight, muscle weight percentage, fat weight percentage and bone weight percentage to the outside.

[0083] The meat quality composition calculation method and system according to the embodiment of the present disclosure obtains the relationship between the scanned image data and the meat quality composition information by referring to a deep learning method based on a large amount of sample data, thereby obtaining a trained meat quality composition calculation model. Thus, the meat quality composition information is directly calculated using the trained meat quality composition calculation model. Therefore, the method according to the embodiment of the present disclosure can avoid the tedious problems of preparing and preserving calibration parts in traditional technologies, and has reliable calculation accuracy. In addition, the meat quality composition calculation method and system according to the embodiment of the present disclosure can be directly used in a dual-energy X-ray-based scanning system, thereby quickly and conveniently upgrading the functional modules of the current system.

[0084] It should be understood that the various forms of the processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this disclosure can be performed in parallel, sequentially, or in a different order, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved. This is not limited herein.

[0085] The above specific embodiments do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions may be made based on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure shall be included within the scope of protection of this disclosure.

Claims

1. A method for calculating meat quality components, the method comprising: performing a dual-energy X-ray scan on the meat piece to obtain scanned image data of the meat piece; Converting the expression form of the scanned image data to obtain image conversion data of the meat piece; Training a meat quality composition calculation model based on multiple meat samples to obtain a trained meat quality composition calculation model; as well as Based on the image conversion data, the meat quality composition information of the meat piece is calculated using the trained meat quality composition calculation model.

2. The method according to claim 1, wherein: The method further comprises: The scanned image data of the meat pieces are converted into one or more expression forms to obtain one or more image conversion data of the meat pieces corresponding to the one or more expression forms.

3. The method according to claim 2, wherein: The expression forms include dual-energy form, grayscale form, RGB form, atomic number form and multiple base material forms.

4. The method according to claim 1, wherein: The method further comprises: The scanned image data of the meat piece or the image conversion data of the meat piece is calibrated to be unified to an imaging standard corresponding to the trained meat quality composition calculation model.

5. The method according to claim 1, wherein: The method further comprises: performing image processing on the scanned image data of the meat piece to divide at least one area to be inspected from the scanned image data of the meat piece; and The image conversion data of the meat piece corresponding to the at least one area to be detected is input into the trained meat quality composition calculation model to calculate the meat quality composition information of the at least one area to be detected.

6. The method according to claim 1, wherein: The training includes: Performing the dual-energy X-ray scanning on the meat sample to obtain scanned image data of the meat sample, and converting the expression form of the scanned image data of the meat sample to obtain image conversion data of the meat sample; Performing high-precision measurement on the meat sample to obtain the true value of the meat quality component of the meat sample; and Based on the image conversion data of the meat sample and based on the true value of the meat quality component, the meat quality component calculation model is trained to obtain the trained meat quality component calculation model.

7. The method according to claim 6, wherein: The training also includes: The scanned image data of the plurality of meat samples or the image conversion data of the meat samples are calibrated to be unified under the same reference imaging standard.

8. The method according to claim 6, wherein: The training also includes: Converting the scanned image data of the meat sample into one or more expression forms to obtain one or more image conversion data of the meat sample corresponding to the one or more expression forms; and Based on the image conversion data of the plurality of meat samples, a plurality of the trained meat quality composition calculation models are obtained.

9. The method according to claim 6, wherein: The training also includes: For each of the plurality of meat samples, establishing a mapping relationship between the scanned image data of the meat sample and a true value of the meat quality component; and Based on the mapping relationship among the multiple meat samples, the meat quality composition calculation model is trained by calculating the difference between the calculated value of the meat quality composition of the same meat sample and the true value of the meat quality composition, and updating the model parameters of the meat quality composition calculation model based on the difference.

10. The method according to any one of claims 6 to 9, wherein: The method for performing the high-precision measurement comprises: Using a high-precision imaging device to obtain the true value of the meat composition of the meat sample; and / or The meat sample is cut manually and measured to obtain the true value of the meat quality component of the meat sample.

11. The method according to any one of claims 6 to 9, wherein: The true value of the meat quality component includes at least one of the muscle weight, fat weight, bone weight, total weight, muscle weight percentage, fat weight percentage, and bone weight percentage of the meat sample.

12. The method according to any one of claims 1 to 9, wherein: The calculated meat composition information includes at least one of muscle weight, fat weight, bone weight, total weight, muscle weight percentage, fat weight percentage, and bone weight percentage; as well as The method further includes: performing output processing on the calculated meat composition information to externally output at least one of muscle weight, fat weight, bone weight, total weight, muscle weight percentage, fat weight percentage and bone weight percentage.

13. A meat quality component calculation system, the system executing the method according to any one of claims 1 to 12, the system comprising: an image scanning module, the image scanning module being configured to perform a dual-energy X-ray scan on the meat piece or the meat sample to obtain scanned image data of the meat piece or the meat sample; a conversion module, the conversion module being configured to convert the expression form of the scanned image data to obtain image conversion data of the meat piece or the meat sample; A true value acquisition module, wherein the true value acquisition module is configured to perform high-precision measurement on the meat sample to obtain a true value of a meat quality component of the meat sample; and A meat quality composition calculation model is provided, wherein the meat quality composition calculation model is trained based on the image conversion data of a plurality of the meat samples and the true value of the meat quality composition to obtain a trained meat quality composition calculation model, and the meat quality composition information of the meat piece is calculated based on the image conversion data of the meat piece using the trained meat quality composition calculation model.

14. The system according to claim 13, further comprising: A calibration module is configured to calibrate the scanned image data or the image conversion data to unify them to the same reference imaging standard.

15. The system according to claim 13, further comprising: An image division module is configured to perform image processing on the scanned image data of the meat piece to divide at least one area to be detected from the scanned image data of the meat piece, so that the system obtains image conversion data of the meat piece corresponding to the at least one area to be detected.

16. The system according to claim 13, further comprising: A mapping establishment module is configured to establish, for each of the meat samples, a mapping relationship between the scanned image data of the meat sample and the true value of the meat quality component.

17. The system according to any one of claims 13 to 16, further comprising: An output processing module is configured to perform output processing on the calculated meat quality composition information to output at least one of muscle weight, fat weight, bone weight, total weight, muscle weight percentage, fat weight percentage and bone weight percentage to the outside.

Images