Poultry leg meat weight and tibia length and width lossless prediction model construction method and system based on deep learning
By using deep learning models to non-destructively predict the weight of poultry leg meat and the length and width of tibia, the problems of high cost and human error in traditional methods are solved, enabling accurate measurement of live poultry and improving breeding efficiency.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-19
- Publication Date
- 2026-03-13
AI Technical Summary
Traditional methods for selective breeding of poultry leg meat require maintaining large-scale test populations, resulting in high costs and easy introduction of human error. They also cannot conduct continuous live monitoring, which affects breeding progress.
A lossless prediction model based on deep learning is adopted to achieve accurate and rapid measurement of poultry leg meat weight and tibia length and width through image acquisition, preprocessing, segmentation and feature extraction. The model is fitted and trained using MCNet model and Ridge machine learning.
It enables precise segmentation and rapid prediction of leg meat weight and tibia length and width in live poultry, reduces measurement costs, improves measurement efficiency, supports continuous non-destructive breeding across generations, and shortens the generation interval in breeding.
Smart Images
Figure CN121661674A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of biological breeding and artificial intelligence image recognition technology, specifically involving a method and system for constructing a non-destructive prediction model for poultry leg meat weight and tibia length and width based on deep learning. Background Technology
[0002] Leg weight and tibia length in poultry are key traits for assessing growth and development and for genetic selection, and have long been of great concern to poultry farms. Traditional methods for selecting poultry leg weight typically involve raising separate test populations of siblings, obtaining data through slaughter measurements, and then implementing pedigree selection. However, this method has three significant drawbacks: First, maintaining a large test population significantly increases feeding costs; second, the slaughtering and processing rely on manual operation, which is not only inefficient but also prone to human error, affecting data accuracy; and most importantly, continuous observation of the same individual after slaughter is impossible, making it impossible to obtain dynamic data on leg weight performance indicators at different growth and development stages, severely hindering progress in poultry breeding.
[0003] In recent years, with the rapid development of measurement technology, non-invasive measurement methods based on color imaging or radiographic techniques have provided new possibilities for animal body composition analysis. These technologies can achieve detailed observation of internal structures without damaging the organism, attracting widespread attention from breeding researchers and gradually being applied to non-destructive evaluation studies of animal carcass / slaughter performance. Against this backdrop, developing a technical model and system for the non-destructive and accurate prediction of leg meat weight and tibia length and width in live poultry based on imaging technology has significant scientific research value and application prospects for overcoming the technical bottlenecks of traditional measurement methods, such as slaughter and dissection techniques, and improving the efficiency of poultry breeding. Summary of the Invention
[0004] To address the aforementioned issues, this invention provides a method and system for constructing a non-destructive prediction model for poultry leg meat weight and tibia length and width based on deep learning. This method aims to solve the problems of low measurement efficiency, high labor costs, and inability to perform continuous live monitoring in existing methods, thereby achieving accurate, rapid, and non-destructive measurement of live poultry leg meat weight and tibia length and width.
[0005] The first technical solution of this application discloses a method for constructing a non-destructive prediction model for poultry leg meat weight and tibia length and width based on deep learning, including the following steps:
[0006] S1. Image acquisition: Scan poultry to obtain a DR image containing the complete outline of the poultry's leg area;
[0007] S2. Image preprocessing: Enhance the DR image to obtain an enhanced DR image;
[0008] S3. Image segmentation: Construct a cascaded segmentation network model with multi-scale feature fusion to segment the enhanced DR image and obtain mask images of poultry legs and tibias, namely poultry leg labels and tibia labels;
[0009] S4. Feature Extraction: Extract shape features from the poultry leg labels obtained in S3, and extract tibial length and minimum width features from the tibia labels obtained in S3 to obtain tibial length and minimum width values;
[0010] S5. Tibial length and width measurement: Tibial length and tibial width are determined based on tibial length and minimum width values;
[0011] S6. Leg weight prediction: The leg weight of the poultry to be tested is fitted and trained using the shape features of the poultry leg and the length and minimum width features of the tibia, so as to obtain a non-destructive prediction model of poultry leg meat weight based on deep learning.
[0012] Furthermore, the cascaded segmentation network model with multi-scale feature fusion described in S3 is an MCNet model based on the OpenCV method, which includes a ResNet residual network, a multi-scale feature extraction module, and an upsampling module.
[0013] Furthermore, the multi-scale feature extraction module described in S3 includes four dilated convolutions with dilation rates of 3, 5, 6, and 9, as well as two 1*1 convolutional layers. Each dilated convolution corresponds to a BN and a ReLU layer.
[0014] Furthermore, the shape features of the poultry leg described in S4 include the leg's area, perimeter, length, width, circumcircle radius, circumcircle ellipticity, and the side length of the equivalent square of the circumcircle rectangle. Feature extraction is performed using the OpenCV algorithm. The OpenCV findContours function is used to apply the outermost contour pattern retrieval and contour compression algorithms to detect the outer contour in the target image and generate its corresponding filling mask for subsequent pixel-level calculations. The area of the region is obtained by counting the number of non-zero pixels in the mask. The perimeter of the target contour is calculated using the arcLength function. The area, length, and width of the minimum circumcircle rectangle are calculated using the minAreaRect function. The circumcircle radius is calculated using the minEnclosingCircle function. Ellipse fitting is performed using the fitEllipse function to calculate the major and minor axes of the circumcircle ellipse and obtain the circumcircle ellipticity. The side length of the equivalent square of the circumcircle rectangle is calculated using the sqrt(height * width) function based on the width and height of the circumcircle rectangle.
[0015] Furthermore, the determination of tibial length and tibial width based on tibial length and minimum width values as described in S5 specifically involves:
[0016] (1) Length measurement:
[0017] a. Equal-width slice division: Divide the tibial length value and minimum width value data into two equal parts along the width direction, and generate two images perpendicular to the long axis of the tibia for detection and calculation respectively;
[0018] b. Dynamic detection and positioning: The constant width straight line sliding algorithm is used to move along the long axis of the tibia, monitor the intersection area of the tibial mask in real time, and record the coordinates of the first non-zero intersection point in the proximal end and the last non-zero intersection point in the distal end;
[0019] c. Length calculation: The spatial pixel distance between the two endpoints is calculated by connecting the two coordinate lines and used as the tibial length;
[0020] (2) Minimum width measurement:
[0021] During the sliding window process described in (1)-b, the intersection area between each translation and the tibia region is calculated in real time, and the minimum pixel width during all cross translations is recorded as the minimum width of the tibia.
[0022] The second technical solution of this application discloses a non-destructive prediction model for poultry leg meat weight and tibia length and width based on deep learning, obtained according to the above construction method, including:
[0023] Image acquisition module: used to acquire DR images of the poultry under test that contain the outline of the complete leg region;
[0024] Image preprocessing module: used to enhance images to obtain enhanced DR images;
[0025] Image segmentation module: used to segment enhanced DR images to obtain poultry leg and tibia labels;
[0026] Feature extraction module: used to extract shape features of the leg region and length and width features of the tibia;
[0027] Tibial length and width measurement module: used to calculate the tibial length and width values using the tibial length and width characteristics;
[0028] Leg weight prediction module: Used to non-destructively predict the weight of poultry legs using extracted shape features and tibia length and width features.
[0029] Furthermore, a deep learning-based method for predicting the leg weight of live poultry includes using the deep learning-based non-destructive prediction model for the leg weight and tibia length and width of the poultry to be tested to predict the tibia length, width, and leg weight of the poultry to be tested.
[0030] And a computer device including a memory, a processor, and a computer program stored in the memory and executable on the processor, the processor being used to execute the model.
[0031] The third technical solution of this application discloses the application of the above model in the field of poultry breeding.
[0032] Furthermore, the application includes using the model to continuously collect and measure the leg meat weight of poultry during the growth period, selecting poultry breeding populations based on the predicted leg meat weight values, thereby achieving continuous non-destructive breeding of live generations based on the leg meat weight phenotype.
[0033] Beneficial effects:
[0034] 1. Achieve precise leg segmentation for live poultry.
[0035] 2. It can quickly and accurately predict the weight of poultry leg meat.
[0036] 3. It can quickly measure the length and width of poultry tibias.
[0037] 4. It achieves a good balance in terms of scanning efficiency, accuracy and cost, and has the potential for large-scale application in actual breeding farm environments. Attached Figure Description
[0038] Figure 1 This is a diagram of the MCNet segmentation model architecture;
[0039] Figure 2 This is a schematic diagram of the multi-scale feature extraction module structure;
[0040] Figure 3 This is an overall structural diagram;
[0041] Figure 4 It is a masked image of a chicken leg and shinbone; the left is the chicken leg and the right is the shinbone. Detailed Implementation
[0042] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.
[0043] Unless otherwise specifically stated, the relative arrangement, numerical expressions, and values of the components and steps described in these embodiments do not limit the scope of this application. It should also be understood that, for ease of description, the dimensions of the various parts shown in the drawings are not drawn to actual scale. Techniques, methods, and devices known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and devices should be considered part of the specification. In all examples shown and discussed herein, any specific values should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values. It should be noted that similar reference numerals and letters in the following drawings denote similar items; therefore, once an item is defined in one drawing, it need not be further discussed in subsequent drawings.
[0044] Unless otherwise specified, the technical terms in this specification have the same meaning as those generally understood by those skilled in the art; however, in case of any conflict, the definitions in this specification shall prevail.
[0045] The first embodiment of this application, using chickens as an example, discloses a non-destructive prediction method based on deep learning for the weight of a live chicken's leg and the length and width of its tibia. Figure 3 As shown, it includes the following steps:
[0046] S1. Image Acquisition: Scan the chicken to obtain a DR image containing the complete outline of the chicken leg region;
[0047] S2. Image preprocessing: Enhance the DR image to obtain an enhanced DR image;
[0048] S3. Image segmentation: Construct a cascaded segmentation network model with multi-scale feature fusion to segment the enhanced DR image and obtain mask images of chicken leg and tibia, namely chicken leg label and tibia label.
[0049] S4. Feature Extraction: Using the OpenCV algorithm, extract the area, perimeter, length, width, circumcircle radius, outer ellipticity, and side length of the equivalent square of the circumcircle rectangle from the chicken leg label obtained in S3; extract the tibia length and width features from the tibia label obtained in S3.
[0050] S5. Weight Prediction: The slaughter weight of the chicken leg is fitted and trained using the shape features of the chicken leg and the length and width features of the tibia to obtain a chicken leg weight prediction model based on deep learning.
[0051] S1 uses an imaging device to scan the chicken under test to obtain a DR image containing the outline of the complete chicken leg area; S2 uses thresholding in digital image processing to enhance the brightness of the acquired DR image, thereby obtaining clearer DR data.
[0052] In S3, such as Figure 1 As shown, by constructing a cascaded segmentation network model with multi-scale feature fusion, image segmentation is achieved on the enhanced DR image, obtaining mask images of the chicken leg and tibia respectively, as shown. Figure 4 As shown, the chicken leg label and shin label are used. The length and width features of the shin label are extracted using the multi-scale feature extraction module of the model.
[0053] The cascaded segmentation network model for multi-scale feature fusion is the MCNet model, which includes a ResNet residual network, a multi-scale feature extraction module, and an upsampling module (e.g., Figure 1 (As shown). The decoding layer of this model includes ordinary convolutional and deconvolutional neural network layers. Skip connections are used to achieve feature fusion between the encoding and decoding layers, and feature transfer is performed by incorporating contextual information. The MCNet model includes two 3x3 convolutional layers, one 1x1 convolutional layer, and corresponding BN and activation layers for each convolutional layer, used to extract deep features. The input features are passed through two 3x3 convolutional layers and one 1x1 convolutional layer respectively, and the two parts are added together to output the extracted deep features. This is used for accurate segmentation of chicken legs and tibias in input DR images.
[0054] In a further implementation, such as Figure 2 As shown, the multi-scale feature extraction module of the model includes four dilated convolutions with dilation rates of 3, 5, 6, and 9, as well as two 1*1 convolutional layers. Each dilated convolution corresponds to a BN and a ReLU layer, which are used to extract features of the length and width of the tibia.
[0055] In a further embodiment, the shape features of the chicken leg in S4 include the area, perimeter, length, width, circumcircle radius, external ellipticity, and side length of the equivalent square of the circumcircle rectangle.
[0056] Furthermore, the fitting training is performed using the Ridge machine learning model, that is, the extracted chicken leg and tibia features and the slaughter weight of the chicken leg are used to fit and train the Ridge model, and the machine learning is used to predict the weight of the chicken leg, thereby obtaining a deep learning-based prediction model for the weight of the chicken leg and the length and width of the tibia.
[0057] The prediction method of this application will be further explained below through specific embodiments.
[0058] Example 1: Prediction of Chicken Leg Weight
[0059] Using 36-day-old broiler chickens as the data base, this model was used to perform mask segmentation of the leg and tibia regions, predict leg weight, and measure tibia length and width. The Pearson correlation between the predicted leg weight and the leg weight measured at slaughter was 82.58%, RMSE was 9.08, and MAE was 7.39. The comparison between the prediction results and the slaughter value is shown in Table 1.
[0060] Table 1 Comparison of Prediction Results and Slaughter Value
[0061] Sample number Leg muscle weight prediction results (g) Chicken leg slaughter weight (g) 52881 116.88 114 55723 113.18 102 55726 125.82 122 56033 119.66 113 73200 133.53 126 75682 101.72 98 75711 126.30 121 5612 144.43 150 5678 139.32 151 72257 125.04 124
[0062] The experimental results show that the system has achieved high prediction accuracy and can be practically applied in actual breeding processes. On-site, only images of live chickens need to be captured; the system will then calculate and output the predicted leg weight. It can predict the leg weight of a large number of chickens, and after all measurements are taken, the chickens are ranked from highest to lowest leg weight based on their "image prediction value," and those selected for breeding are kept according to the breeding plan. Through continuous selection across multiple generations, the average leg weight will be effectively improved, significantly reducing feed conversion rate and eliminating the need for slaughter throughout the entire process. This achieves continuous, non-destructive selection across live generations, significantly accelerating genetic progress and saving testing costs. This innovative method avoids the harm to chickens caused by traditional slaughter testing, achieves continuous, non-destructive data collection during critical growth periods, and provides high-density phenotypic data support for genomic selection. In the whole-genome selection breeding stage, using this system to predict leg weight, combined with genotype data for genome-wide association analysis (GWAS), allows for the screening of loci significantly associated with leg weight, further uncovering key genetic information regulating leg development, significantly shortening the generation interval, and thus achieving efficient and precise breeding goals.
[0063] Therefore, those skilled in the art will recognize that although embodiments of the present invention have been shown and described in detail herein, many other variations or modifications conforming to the principles of the present invention can be directly determined or derived from the disclosure of the present invention without departing from the spirit and scope of the invention. Therefore, the scope of the present invention should be understood and recognized as covering all such other variations or modifications.
Claims
1. A method for constructing a non-destructive prediction model for poultry leg meat weight and tibia length and width based on deep learning, characterized in that, Includes the following steps: S1. Image Acquisition: Scan poultry to obtain a DR image containing the complete outline of the poultry's leg area; S2. Image preprocessing: Enhance the DR image to obtain an enhanced DR image; S3. Image segmentation: Construct a cascaded segmentation network model with multi-scale feature fusion to segment the enhanced DR image and obtain mask images of poultry legs and tibias, namely poultry leg labels and tibia labels; S4. Feature Extraction: Extract shape features from the poultry leg labels obtained in S3, and extract tibial length and minimum width features from the tibia labels obtained in S3 to obtain tibial length and minimum width values; S5. Tibial length and width measurement: Tibial length and tibial width are determined based on tibial length and minimum width values; S6. Leg weight prediction: The leg weight of the poultry to be tested is fitted and trained using the shape features of the poultry leg and the length and minimum width features of the tibia, so as to obtain a non-destructive prediction model of poultry leg meat weight based on deep learning.
2. The method for constructing the prediction model according to claim 1, characterized in that, The cascaded segmentation network model for multi-scale feature fusion described in S3 is the MCNet model based on the OpenCV method, which includes a ResNet residual network, a multi-scale feature extraction module, and an upsampling module.
3. The method for constructing the prediction model according to claim 1 or 2, characterized in that, The multi-scale feature extraction module described in S3 includes four dilated convolutions with dilation rates of 3, 5, 6, and 9, as well as two 1*1 convolutional layers. Each dilated convolution corresponds to a BN and a ReLU layer.
4. The method for constructing the prediction model according to claim 1, characterized in that, The poultry leg shape features described in S4 include the leg's area, perimeter, length, width, circumcircle radius, external ellipticity, and the side length of the equivalent square of the circumcircle rectangle.
5. The method for constructing the prediction model according to claim 1, characterized in that, The determination of tibial length and tibial width based on tibial length and minimum width values, as described in S5, is as follows: (1) Length measurement: a. Equal-width slice division: Divide the tibial length value and minimum width value data into two equal parts along the width direction, and generate two images perpendicular to the long axis of the tibia for detection and calculation respectively; b. Dynamic detection and positioning: The constant width straight line sliding algorithm is used to move along the long axis of the tibia, monitor the intersection area of the tibial mask in real time, and record the coordinates of the first non-zero intersection point in the proximal end and the last non-zero intersection point in the distal end; c. Length calculation: The spatial pixel distance between the two endpoints is calculated by connecting the two coordinate lines and used as the tibial length; (2) Minimum width measurement: During the sliding window process described in (1)-b, the intersection area between each translation and the tibia region is calculated in real time, and the minimum pixel width during all cross translations is recorded as the minimum width of the tibia.
6. A non-destructive prediction model for poultry leg meat weight and tibia length and width based on deep learning, obtained by the construction method according to any one of claims 1-5, characterized in that, include: Image acquisition module: used to acquire DR images of the poultry under test that contain the outline of the complete leg region; Image preprocessing module: used to enhance images to obtain enhanced DR images; Image segmentation module: used to segment enhanced DR images to obtain poultry leg and tibia labels; Feature extraction module: used to extract shape features of the leg region and length and width features of the tibia; Tibial length and width measurement module: used to calculate the tibial length and width values using the tibial length and width characteristics; Leg weight prediction module: Used to non-destructively predict the weight of poultry legs using extracted shape features and tibia length and width features.
7. A non-destructive prediction method for poultry leg meat weight and tibia length and width based on deep learning, characterized in that, This includes using the deep learning-based non-destructive prediction model for poultry leg meat weight and tibia length and width described in claim 6 to predict the tibia length, width, and leg meat weight of the poultry to be tested.
8. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, the processor being configured to execute the model of claim 6.
9. The application of the model according to claim 6 in the field of poultry breeding.
10. The application according to claim 9, characterized in that, This includes using the model to continuously collect leg meat weight data of poultry during their growth period, selecting breeding poultry based on predicted leg meat weight values, and thus achieving continuous non-destructive breeding across generations based on the leg meat weight phenotype.