X-ray machine-based poultry meat proportion rapid sorting method, device and medium

Abstract

This invention relates to a method, apparatus, and medium for rapid sorting of poultry meat percentage based on X-ray imaging. The method includes: Step S1: acquiring X-ray images of the poultry to be tested; Step S2: identifying the sternal region and skin boundary based on the X-ray image; Step S3: drawing a normal at the sternal apex, obtaining the intersection of the normal at the sternal apex and the skin boundary as the first surface point, and obtaining the initial bone-skin distance based on the distance from the sternal apex to the first surface point; Step S4: obtaining the angle between the X-ray machine's visual axis direction and the normal direction of the first surface point as the consistency difference angle, obtaining the sternal integrity ratio and sternal principal axis attitude angle based on the sternal region, and correcting the initial bone-skin distance based on the consistency difference angle, sternal integrity ratio, and sternal principal axis attitude angle to obtain the corrected bone-skin distance; Step S5: obtaining the meat percentage grading result based on the corrected bone-skin distance. Compared with the prior art, this invention can achieve poultry meat percentage sorting based on X-ray imaging.

Description

Technical Field

[0001] This invention relates to the field of poultry sorting, and in particular to a rapid sorting method, apparatus and medium for poultry meat percentage based on X-ray machine. Background Technology

[0002] In the modern poultry farming and processing industry, the meat content of poultry is a crucial sorting indicator. Current technologies typically involve slaughtering the animals first, followed by sorting based on fat content or meat development. For example, existing technologies like Chinese patent CN108548831A use X-ray imaging to determine fat content. However, in recent years, with increasingly fierce competition in the food industry, traditional frozen meat has lower profit margins, while fresh meat, especially chilled meat, requires a high degree of coordination between supply and demand. If fresh meat cannot be delivered to customers within the stipulated time after slaughter, it will be sold as lower-priced chilled or even frozen meat, significantly impacting the company's production costs.

[0003] Traditional methods for measuring the fat content and development of slaughtered meat require slaughtering before measurement, which can easily lead to such problems. Therefore, those skilled in the art need a method to measure the fat content of live poultry, even if the measurement is not very accurate, it can still meet the needs of this field.

[0004] The technical solutions of existing technologies such as the aforementioned Chinese patent CN108548831A cannot be applied to the detection of live poultry because the internal characteristics of live animals are more complex, including blood vessels and air, which greatly affect the error of X-ray images. In response, some existing technologies have attempted to use CT images for detection. For example, Chinese patent CN113545796A discloses a method for detecting the quality traits of pork in live animals using computed tomography, which mentions that it can measure the fat content in the muscle. However, such existing technologies also have certain drawbacks. First, CT machines are much more expensive than X-ray machines, and their reliability is lower and their lifespan is shorter. For farms, upgrading to CT machines for detection would have a significant impact on their already low profit margins. Second, detecting fat in the muscle cannot fully meet the requirements because it does not completely measure fat content, but rather measures the overall development. In poultry, muscle and fat are often intertwined, and simply measuring fat content may not meet the sorting requirements.

[0005] Based on this, those skilled in the art are dedicated to developing a method for sorting poultry meat percentages using an X-ray machine. Summary of the Invention

[0006] The purpose of this invention is to provide a method, apparatus, and medium for rapid sorting of poultry meat percentage based on X-ray machine in order to solve the problems existing in the prior art.

[0007] The objective of this invention can be achieved through the following technical solutions:

[0008] A rapid sorting method for poultry meat percentage based on X-ray machine includes:

[0009] Step S1: Acquire X-ray images of the poultry to be tested;

[0010] Step S2: Identify the sternal region and skin boundary based on X-ray image recognition;

[0011] Step S3: Draw a normal at the apex of the sternum, and obtain the intersection of the normal at the apex of the sternum and the skin boundary as the first surface point. Obtain the initial bone-skin distance based on the distance from the apex of the sternum to the first surface point.

[0012] Step S4: Obtain the angle between the X-ray machine's visual axis direction and the normal direction of the first surface point as the consistency difference angle. Based on the sternal region, obtain the sternal integrity ratio and the sternal principal axis attitude angle. Based on the consistency difference angle, the sternal integrity ratio, and the sternal principal axis attitude angle, correct the initial bone-skin distance to obtain the corrected bone-skin distance.

[0013] Step S5: Obtain the meat percentage grading results based on the corrected bone-skin distance.

[0014] The sternal region in step S2 is obtained by segmentation using a segmentation model, wherein the segmentation model includes:

[0015] The backbone feature extraction module is configured to extract the backbone features F of the image based on the X-ray image;

[0016] The boundary awareness module is configured to extract a gradient map G from the X-ray image using the Laplace operator, and then fuse the gradient map G with the image backbone features F to obtain a first fused map F. Bam ;

[0017] The multi-scale detail recovery module is configured to process the image backbone features F through dilated convolution to obtain the receptive field detail map F. ms And will experience the details of the wild map F ms The second fused image F is obtained by fusing it with the image backbone features F. mdr , the first fusion graph F Bam Second fusion diagram F mdr The fusion yields the sternal region and its mask image.

[0018] The loss function of the segmentation model is:

[0019] L=λ1×L BCE+λ2×L edge +λ3×L structure ;

[0020] ;

[0021] L edge =‖▽Mask-∇GT‖1;

[0022] L structure =1-SSIM(Mask,GT);

[0023] Where: L is the loss function of the segmentation model, λ1 is the coefficient of the binary cross-entropy loss, L BCE The loss is a binary cross-entropy loss, λ² is the coefficient of the edge-preserving loss, and L... edge The edge preservation loss is λ3, which is the coefficient of the structural similarity loss. structure For structural similarity loss, N is the total number of pixels in a training batch, y i p is the true label of the i-th pixel. i SSIM(·) represents the probability that the i-th pixel belongs to the target region, GT represents the true label, and ▽ represents the gradient sign.

[0024] The sternal region is:

[0025] F out =Conv 1*1 (Concat(F Bam ,F mdr ));

[0026] The mask image is:

[0027] Mask = JH(F out );

[0028] Among them, F out For the sternal region, JH (·) is the activation function.

[0029] The apex of the sternum is the point in the sternal region that is closest to the skin boundary.

[0030] The corrected bone-skin distance is:

[0031] D'=D*C;

[0032] C=C θ ×C S ×C n ;

[0033] ;

[0034] C S=min(1,A S / A ref );

[0035] C n =cos(α);

[0036] Where: D' is the corrected bone-skin distance, D is the initial bone-skin distance, and C is the comprehensive evaluation coefficient. θ C is the attitude angle evaluation coefficient. S C is the evaluation coefficient for sternal integrity. n The coefficient for consistency difference angle is θ, where θ is the angle between the principal axis of the sternum and the preset direction, θ0 is the ideal attitude angle, σ is the standard deviation, and A is the coefficient for consistency difference angle. S A represents the area of ​​the sternal region. ref α represents the standard sternal area, and α represents the concordance difference angle.

[0037] The standard sternal area is obtained based on the average value of the sternal area under the same breed and the same number of days in the standard posture.

[0038] In step S5, the corrected bone-skin distance is compared with the threshold interval to obtain the grading result.

[0039] A rapid sorting device for poultry meat percentage based on X-ray machine includes a memory, a processor, and a program stored in the memory. When the processor executes the program, it implements the method described above.

[0040] A storage medium having a program stored thereon, which, when executed, implements the method described above.

[0041] Compared with the prior art, the present invention has the following beneficial effects:

[0042] 1. On the one hand, by designing the bone-skin distance as the sorting basis for meat ratio, the limitation of existing technology that X-ray images cannot detect the lean meat and fat content of live poultry is solved. On the other hand, by designing a correction based on the consistency difference angle, the sternal integrity ratio and the sternal principal axis posture angle for the initial bone-skin distance, the problem of low accuracy caused by the easy movement of live poultry when taking X-ray images can be solved.

[0043] 2. Using a segmentation model to segment the sternal region can improve segmentation accuracy. The boundary perception module uses the Laplace operator to extract gradient maps, which can effectively enhance the recognition of skin boundaries and reduce noise interference. The multi-scale detail restoration module expands the receptive field through dilated convolution, which can capture multi-scale features, thereby more accurately segmenting the sternal region and providing a reliable basis for subsequent bone-skin distance calculation. On the other hand, it can enhance the robustness of the model. The multi-module fusion design improves the model's adaptability to complex backgrounds in X-ray images, making it suitable for live poultry detection and overcoming the error problems of traditional methods in live applications.

[0044] 3. The loss function combines binary cross-entropy loss, edge preservation loss, and structural similarity loss, which can simultaneously focus on pixel-level classification accuracy, edge sharpness, and structural similarity, making model training more stable, avoiding overfitting, and improving segmentation quality. Edge preservation loss ensures that the segmentation boundary is consistent with the true label, and structural similarity loss ensures the integrity of the overall structure, thereby generating a more accurate sternal mask map and reducing subsequent calculation errors.

[0045] 4. Specific calculation formulas for the sternal region and mask image are provided. The results are directly output through convolution and activation functions, which simplifies the segmentation process, realizes fast real-time processing, meets the speed requirements of sorting methods, and makes the algorithm easier to implement and optimize, thereby improving the repeatability and engineering application value of the method.

[0046] 5. The sternal vertex is defined as the point in the sternal region that is closest to the skin boundary. Based on geometric relationships, the sternal vertex can be found automatically and accurately, avoiding subjective errors caused by human intervention. Regardless of how the bird's posture changes, the reference point can be stably determined, ensuring the reliability of the initial bone-skin distance calculation.

[0047] 6. The corrected bone-skin distance incorporates the consistency difference angle, sternal integrity ratio, and sternal principal axis posture angle as correction factors. On the one hand, by comprehensively correcting the initial bone-skin distance through multiple coefficients, the measurement deviation caused by the movement of live poultry, changes in posture, or X-ray shooting angle can be effectively offset, improving the accuracy of meat ratio estimation. On the other hand, the exponential and cosine function design in the formula makes the correction process smooth, adaptable to different scenarios, and ensures the consistency and reliability of sorting results under different conditions. Attached Figure Description

[0048] Figure 1 This is a schematic diagram of the main steps of the method of the present invention. Detailed Implementation

[0049] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.

[0050] A rapid sorting method for poultry meat percentage based on X-ray machine, such as... Figure 1 As shown, it includes:

[0051] Step S1: Acquire X-ray images of the poultry to be tested;

[0052] Specifically, by guiding birds into the collection area and controlling their posture as much as possible, X-ray images of their chest areas are collected and transmitted to the image processing system.

[0053] Step S2: Based on X-ray image recognition, the sternal region and skin boundary are obtained. The sternal region is segmented using a segmentation model, which includes:

[0054] 1. The backbone feature extraction module is configured to extract the backbone features F of the image based on the X-ray image;

[0055] Specifically, the backbone feature extraction network uses a simplified UNet structure, the encoder adopts a ResBlock structure, and the image backbone feature F is:

[0056] ;

[0057] Where: Backbone(·) is the operator symbol of the backbone feature extraction network, I is the X-ray image, H is the height of the X-ray image, and W is the width of the X-ray image;

[0058] 2. The boundary awareness module is configured to use the Laplace operator to extract a gradient map G=∇I from the X-ray image, and then fuse the gradient map G with the image backbone features F to obtain the first fused map F. Bam :

[0059] F Bam =F*σ(Conv(G));

[0060] Where: Conv(·) is the convolution operator.

[0061] 3. The multi-scale detail restoration module is configured to process the image backbone features F through dilated convolution to obtain the receptive field detail map F. ms And will experience the details of the wild map F ms The second fused image F is obtained by fusing it with the image backbone features F. mdr , the first fusion graph F Bam Second fusion diagram F mdr The fusion yields the sternal region and its mask image.

[0062] Among them, the sensory field detail image F ms for:

[0063] ;

[0064] Where: d is the expansion rate, and r is the scale value;

[0065] The sternal region is:

[0066] F out =Conv 1*1 (Concat(F Bam ,F mdr ));

[0067] The mask image is:

[0068] Mask = JH(F out );

[0069] Among them, F out For the sternal region, JH (·) is the activation function.

[0070] Using a segmentation model to segment the sternal region can improve segmentation accuracy. The boundary perception module uses the Laplace operator to extract gradient maps, which can effectively enhance the recognition of skin boundaries and reduce noise interference. The multi-scale detail restoration module expands the receptive field through dilated convolution, which can capture multi-scale features and thus segment the sternal region more accurately, providing a reliable basis for subsequent bone-skin distance calculation. On the other hand, it can enhance the robustness of the model. The multi-module fusion design improves the model's adaptability to complex backgrounds in X-ray images, making it suitable for live poultry detection and overcoming the error problems of traditional methods in live applications.

[0071] Furthermore, in this embodiment, the loss function of the segmentation model is:

[0072] L=λ1×L BCE +λ2×L edge +λ3×L structure ;

[0073] ;

[0074] L edge =‖∇Mask-∇GT‖1;

[0075] L structure =1-SSIM(Mask,GT);

[0076] Where: L is the loss function of the segmentation model, λ1 is the coefficient of the binary cross-entropy loss, L BCEThe loss is a binary cross-entropy loss, λ² is the coefficient of the edge-preserving loss, and L... edge The edge preservation loss is λ3, which is the coefficient of the structural similarity loss. structure For structural similarity loss, N is the total number of pixels in a training batch, y i p is the true label of the i-th pixel. i SSIM(·) represents the probability that the i-th pixel belongs to the target region, GT represents the true label, and ∇ represents the gradient sign.

[0077] The loss function combines binary cross-entropy loss, edge preservation loss, and structural similarity loss, which can simultaneously focus on pixel-level classification accuracy, edge sharpness, and structural similarity, making model training more stable, avoiding overfitting, and improving segmentation quality. Edge preservation loss ensures that the segmentation boundary is consistent with the true label, and structural similarity loss ensures the integrity of the overall structure, thereby generating a more accurate sternal mask map and reducing subsequent calculation errors.

[0078] Step S3: Draw a normal at the apex of the sternum, and obtain the intersection of the normal at the apex of the sternum and the skin boundary as the first surface point. Obtain the initial bone-skin distance based on the distance from the apex of the sternum to the first surface point.

[0079] In this embodiment, the sternal vertex is the point in the sternal region that is closest to the skin boundary. Of course, in some other embodiments, it can also be obtained in other ways, such as template matching. The initial bone-skin distance can be obtained by multiplying the number of pixels by the unit length of the pixels. For example, 90 pixels equals 1 cm.

[0080] Step S4: Obtain the angle between the X-ray machine's visual axis direction and the normal direction of the first surface point as the consistency difference angle. Based on the sternal region, obtain the sternal integrity ratio and the sternal principal axis attitude angle. Then, based on the consistency difference angle, the sternal integrity ratio, and the sternal principal axis attitude angle, correct the initial bone-skin distance to obtain the corrected bone-skin distance. The corrected bone-skin distance is:

[0081] D'=D*C;

[0082] C=C θ ×C S ×C n ;

[0083] ;

[0084] C S =min(1,A S / A ref );

[0085] C n =cos(α);

[0086] Where: D' is the corrected bone-skin distance, D is the initial bone-skin distance, and C is the comprehensive evaluation coefficient. θ C is the attitude angle evaluation coefficient. S C is the evaluation coefficient for sternal integrity. n The coefficient for consistency difference angle is θ, where θ is the angle between the principal axis of the sternum and the preset direction, θ0 is the ideal attitude angle, σ is the standard deviation, and A is the coefficient for consistency difference angle. S A represents the area of ​​the sternal region. ref α represents the standard sternal area, and α represents the concordance difference angle.

[0087] In this embodiment, the standard sternal area is obtained based on the average value of the sternal area under the same breed and the same number of days in the standard posture.

[0088] A comprehensive evaluation coefficient C is introduced, which is composed of the product of three factors: posture angle evaluation coefficient, sternal integrity evaluation coefficient, and consistency difference angle coefficient. The comprehensive evaluation coefficient C is used as a correction factor and is directly multiplied by the initial bone-skin distance D to obtain the corrected bone-skin distance D.

[0089] Based on this, a mathematical model is used to quantify and compensate for measurement errors caused by incorrect bird posture, incomplete body parts, or poor shooting angles, as follows:

[0090] 1. The posture angle evaluation coefficient is mainly used to compensate for the error caused by body tilt.

[0091] If the bird's body tilts or rotates during the photoshoot, meaning there is an angle between the sternal axis and the preset ideal direction, then the initial bone-skin distance measured from the X-ray image will not be the true vertical distance, but rather a tilted, projected distance that is longer than the true value.

[0092] Therefore, the attitude angle evaluation coefficient, which takes the form of a Gaussian function, is used. When the attitude is perfect, C is not corrected. When the attitude deviates, its value is less than 1; the greater the deviation, the smaller the attitude angle evaluation coefficient.

[0093] The initial distance D is corrected by a coefficient less than 1, thus compensating for the "stretched" projected distance caused by tilting, making the result D′ closer to the true vertical bone-skin distance. The standard deviation σ controls the sensitivity to angular deviation.

[0094] 2. The sternal integrity evaluation coefficient is mainly used to compensate for errors caused by incompleteness in the sternal region.

[0095] In X-ray images, the area of ​​the identified sternum region may be smaller than its true, complete area due to factors such as bird posture, occlusion, or imperfect image segmentation. An incomplete region can lead to inaccurate vertex localization, thus affecting the reliability of the reference point for distance measurement.

[0096] Therefore, the sternal integrity evaluation coefficient is used, which is expressed as an area ratio. When the sternal region is complete or nearly complete, no correction is made. When the sternal region is incomplete, the more missing parts there are, the smaller the coefficient becomes.

[0097] The sternal integrity evaluation coefficient serves as a reliability weight. If the detected sternal region is incomplete, it indicates that the basis of this measurement is unreliable. In this case, the initial measurement result D is discounted, and its weight is reduced, so that the final result D′ will not deviate too much due to the error of the reference point.

[0098] 3. The consistency difference angle coefficient is mainly used to compensate for errors caused by incorrect shooting angles.

[0099] If the X-ray machine's visual axis is not perpendicular to the bird's surface, i.e., the shooting angle is incorrect, the positioning of the "first surface point" on the skin boundary will be deviated. This will lead to inaccurate measurement of the "initial bone-skin distance".

[0100] Therefore, no correction is made when the shot is perfectly vertical; when the shot is at an angle, the larger the angle, the smaller the coefficient.

[0101] The consistency difference angle coefficient is based on the principles of geometric optics. It compensates for surface point positioning errors caused by non-perpendicular shooting by correcting the initial distance with a coefficient less than 1, making it closer to the ideal measurement value under perpendicular shooting.

[0102] This approach eliminates the strong reliance on the poultry's posture during photography. Regardless of the poultry's posture during photography, as long as the system can identify key features such as the sternum region and skin boundaries, it can use this correction model to correct an erroneous "raw observation" D into a standardized measurement that more closely approximates the true physiological characteristics. Through multi-parameter fusion analysis, the accuracy and stability of meat percentage prediction during online live poultry sorting are significantly improved, addressing the limitation of traditional methods that cannot be applied due to the variability of live animal postures.

[0103] Step S5: Obtain the meat percentage grading result based on the corrected bone-skin distance, wherein the corrected bone-skin distance is compared with the threshold range to obtain the grading result.

[0104] Specifically, based on the sample calibration results, the measured distance is mapped to the fat development level and classified into grades. For example, in this embodiment, the corrected bone-skin distance is less than 0.3cm as grade 1, greater than 0.5cm as grade 3, and the rest as grade 2.

[0105] By constructing a sample library, X-ray image ranging is paired with the actual fat layer to establish linear or nonlinear regression models. The models can be continuously updated by collecting samples to adapt to poultry groups of different breeds, ages, and sexes. The system measurement error is controlled within ±0.05cm.

[0106] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A method for quickly sorting poultry meat based on the proportion of X-ray machine, characterized in that, The method comprises the following steps: Step S1: collecting an X-ray image of the farmed bird to be tested; Step S2: identifying a sternum region and a skin boundary based on the X-ray image; Step S3: drawing a normal line at the apex of the sternum, obtaining the intersection point of the normal line of the apex of the sternum and the skin boundary as a first surface point, and obtaining an initial bone-skin distance based on the distance from the apex of the sternum to the first surface point; Step S4: obtaining an angle between the X-ray machine viewing axis direction and the first surface point normal direction as a consistency difference angle, obtaining a sternum integrity ratio and a sternum main axis attitude angle based on the sternum region, and correcting the initial bone-skin distance based on the consistency difference angle, the sternum integrity ratio, and the sternum main axis attitude angle to obtain a corrected bone-skin distance; Step S5: obtaining a meat proportion grading result based on the corrected bone-skin distance.

2. The method according to claim 1, wherein, The sternum region in step S2 is obtained by a segmentation model, wherein the segmentation model comprises: a backbone feature extraction module configured to extract image backbone features F based on the X-ray image; The boundary perception module is configured to extract a gradient map G based on the X-ray image using a Laplace operator, and fuse the gradient map G with the image backbone feature F to obtain a first fusion map F Bam ; The multi-scale detail recovery module is configured to process the image stem feature F through a hole convolution to obtain a receptive field detail map F ms , and fuse the receptive field detail map F ms with the image stem feature F to obtain a second fusion map F mdr , and fuse the first fusion map F Bam and the second fusion map F mdr to obtain a sternum region and a mask map Mask thereof.

3. The method according to claim 2, wherein, The loss function of the segmentation model is: L = λ1 x L BCE + λ2 x L edge + λ3 x L structure ; ； L edge =‖∇Mask-∇GT‖1; L structure =1-SSIM(Mask,GT); wherein: L is the loss function of the segmentation model, λ1 is the coefficient of the binary cross-entropy loss, L BCE is the binary cross-entropy loss, λ2 is the coefficient of the edge-preserving loss, L edge is the edge-preserving loss, λ3 is the coefficient of the structural similarity loss, L structure is the structural similarity loss, N is the total number of pixels in a training batch, y i is the true label of the i-th pixel, p i is the probability that the model predicts the i-th pixel belongs to the target region, SSIM(·) is the structural similarity index, GT is the true label, and ∇ is the gradient symbol.

4. The method according to claim 2, wherein, The sternum region is: F out =Conv 1*1 (Concat(F Bam ,F mdr )); The mask graph Mask is: Mask = JH(F out ) ; where F out is the thoracic region, and JH (·) is an activation function.

5. The method according to claim 2, wherein the method is characterized by, The apex of the sternum is the point in the sternum region closest to the skin boundary.

6. The method according to claim 1, wherein the method is characterized by, The corrected bone-skin distance is: D'=D*C; C=C θ x C S x C n ; ； C S =min(1,A S / A ref ) C n = cos(a); Wherein: D' is the corrected bone-skin distance, D is the initial bone-skin distance, C is the comprehensive evaluation coefficient, C θ is the posture angle evaluation coefficient, C S is the sternum integrity evaluation coefficient, C n is the consistency difference angle coefficient, θ is the included angle between the main axis of the sternum and the preset direction, θ0 is the ideal posture angle, σ is the standard deviation, A S is the sternum area, A ref is the standard sternum area, and α is the consistency difference angle.

7. The method according to claim 6, wherein the method is characterized by, The standard sternum area is obtained based on the mean value of the sternum area under the same standard posture of the same breed and the same number of days.

8. The method according to claim 1, wherein the method is characterized by, In step S5, the corrected bone-skin distance is compared with a threshold interval to obtain the grading result. 9.A poultry meat percentage based X-ray machine rapid sorting device, comprising a memory, a processor, and a program stored in the memory, characterized in that, The processor executes the program to implement the method of any one of claims 1-8.

10. A storage medium having stored thereon a program, characterized by The program is executed to implement the method of any one of claims 1-8.

Images