A method, system, device, medium and product for recognizing, positioning and detecting the posture of dead chickens

Abstract

The application discloses a dead chicken recognition, positioning and posture detection method, system, device, medium and product, and relates to the technical field of poultry health monitoring. The method comprises the following steps: acquiring a visible light image and a depth image of a to-be-tested chicken coop collected by a depth camera; inputting the visible light image of the to-be-tested chicken coop into a chicken identification model to obtain a target chicken identification result; when the target chicken identification result comprises a dead chicken identification result, determining the optimized depth values of the predicted key points of the dead chicken based on the depth image of the to-be-tested chicken coop; determining the spatial three-dimensional coordinates of the predicted key points of the dead chicken based on the optimized depth values of the predicted key points of the dead chicken respectively; determining the spatial three-dimensional coordinates of a predicted breast position point in the predicted key points of the dead chicken as the position of the dead chicken; and determining the posture information of the dead chicken based on the spatial three-dimensional coordinates of the predicted key points of the dead chicken. The application realizes the recognition, positioning and posture detection of the dead chicken.

Description

Technical Field

[0001] The present application relates to the technical field of poultry health monitoring, and in particular to a method, system, device, medium and product for identifying, locating and detecting the posture of dead chickens. Background Art

[0002] As global demand for chicken products continues to grow, chicken producers face unprecedented production pressures and management challenges. Traditional chicken farming methods rely on regular manual inspections to identify and handle dead or unhealthy chickens. This approach not only consumes significant human resources but also often fails to detect problems in a timely manner, leading to increased losses and potential health risks.

[0003] In large-scale farming environments, tens of thousands of chickens may need to be monitored daily. The causes of chicken mortality vary, including disease, environmental factors, and improper husbandry and management. Failure to promptly address the death or abnormal behavior of chickens not only affects the health of other birds but can also spread disease, compromising the safety of the entire production chain. Furthermore, chicken cages produce large amounts of harmful gases (sulfur dioxide, ammonia, etc.). High concentrations of ammonia are irritating to the human respiratory system, potentially causing inflammation of the eyes, nasal cavity, and respiratory tract, and leading to coughing, wheezing, and lung damage. Therefore, timely detection and location of dead chickens for accurate collection is crucial. Summary of the Invention

[0004] The purpose of this application is to provide a dead chicken identification, positioning and posture detection method, system, device, medium and product to achieve the identification, positioning and posture detection of dead chickens.

[0005] To achieve the above objectives, this application provides the following solutions.

[0006] In a first aspect, the present application provides a method for identifying, locating, and detecting the posture of dead chickens, comprising:

[0007] Obtaining visible light images and depth images of the chicken cage to be tested captured by the depth camera;

[0008] Inputting a visible light image of the chicken cage to be tested into a chicken recognition model to obtain a target chicken recognition result; the target chicken recognition result is a visible light image annotated with the recognition results of each chicken in the chicken cage to be tested; when the chicken is alive, the recognition result is a predicted box of the live chicken; when the chicken is dead, the recognition result includes a predicted box of the dead chicken and multiple predicted key points of the dead chicken; the chicken recognition model is obtained by training an improved YOLOv8 network;

[0009] When the target chicken identification results include identification results of dead chickens, determining the optimized depth values ​​of each predicted key point of the dead chickens based on the depth image of the chicken cage to be tested;

[0010] Determining the spatial three-dimensional coordinates of each predicted key point of the dead chicken based on the optimized depth value of each predicted key point of the dead chicken;

[0011] Determine the spatial three-dimensional coordinates of the predicted chicken breast position point among the predicted key points of the dead chicken as the position of the dead chicken;

[0012] The posture information of the dead chicken is determined based on the spatial three-dimensional coordinates of each predicted key point of the dead chicken.

[0013] In a second aspect, the present application provides a dead chicken identification, positioning and posture detection system to implement the above-mentioned dead chicken identification, positioning and posture detection method, the dead chicken identification, positioning and posture detection system includes: a depth camera and a processor; the depth camera is connected to the processor;

[0014] The depth camera is used to collect visible light images and depth images of the chicken cage to be tested;

[0015] The processor is configured to:

[0016] Inputting a visible light image of the chicken cage to be tested into a chicken recognition model to obtain a target chicken recognition result; the target chicken recognition result is a visible light image annotated with the recognition results of each chicken in the chicken cage to be tested; when the chicken is alive, the recognition result is a predicted box of the live chicken; when the chicken is dead, the recognition result includes a predicted box of the dead chicken and multiple predicted key points of the dead chicken; the chicken recognition model is obtained by training an improved YOLOv8 network;

[0017] When the target chicken identification results include identification results of dead chickens, determining the optimized depth values ​​of each predicted key point of the dead chickens based on the depth image of the chicken cage to be tested;

[0018] Determining the spatial three-dimensional coordinates of each predicted key point of the dead chicken based on the optimized depth value of each predicted key point of the dead chicken;

[0019] Determine the spatial three-dimensional coordinates of the predicted chicken breast position point among the predicted key points of the dead chicken as the position of the dead chicken;

[0020] The posture information of the dead chicken is determined based on the spatial three-dimensional coordinates of each predicted key point of the dead chicken.

[0021] In a third aspect, the present application provides a computer device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-mentioned method for identifying, locating, and detecting dead chickens.

[0022] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-mentioned method for identifying, locating and detecting dead chickens.

[0023] In a fifth aspect, the present application provides a computer program product, including a computer program, which, when executed by a processor, implements the above-mentioned dead chicken identification, positioning and posture detection method.

[0024] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0025] The present application discloses a method, system, device, medium and product for dead chicken identification, positioning and posture detection. First, a visible light image and a depth image of a chicken cage to be tested are acquired by a depth camera; then, the visible light image of the chicken cage to be tested is input into a chicken recognition model to obtain a target chicken recognition result; the target chicken recognition result is a visible light image annotated with the recognition result of each chicken in the chicken cage to be tested; when the chicken is alive, the recognition result is a prediction frame of the live chicken; when the chicken is dead, the recognition result includes a prediction frame of the dead chicken and multiple prediction key points of the dead chicken; the chicken recognition model The model is obtained by training the improved YOLOv8 network; secondly, when the target chicken recognition results include the recognition results of dead chickens, the optimized depth value of each predicted key point of the dead chicken is determined based on the depth image of the chicken cage to be tested; then, the spatial three-dimensional coordinates of each predicted key point of the dead chicken are determined based on the optimized depth value of each predicted key point of the dead chicken; thirdly, the spatial three-dimensional coordinates of the predicted chicken breast position point among the predicted key points of the dead chicken are determined as the position of the dead chicken; finally, the posture information of the dead chicken is determined based on the spatial three-dimensional coordinates of each predicted key point of the dead chicken. This application realizes the recognition, positioning and posture detection of dead chickens in sequence through a series of processing of visible light images and depth images. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0027] Figure 1 This is a flow chart of a dead chicken identification, positioning and posture detection method provided in one embodiment of the present application.

[0028] Figure 2 This is a schematic diagram of the dead chicken identification, positioning and posture detection method architecture.

[0029] Figure 3Schematic diagram of the improved YOLOv8 network structure.

[0030] Figure 4 Schematic diagram of the ConV module structure.

[0031] Figure 5 This is a schematic diagram of the C2f module structure.

[0032] Figure 6 Schematic diagram of the CA module structure.

[0033] Figure 7 This is a schematic diagram of the SPPF module structure.

[0034] Figure 8 This is a schematic diagram of the chicken identification results of the chicken identification model.

[0035] Figure 9 Schematic diagram of the harmonic mean-confidence curve of precision and recall before adding the CA module.

[0036] Figure 10 Schematic diagram of the harmonic mean-confidence curve of precision and recall after adding the CA module.

[0037] Figure 11 Schematic diagram of the accuracy-confidence curve before adding the CA module.

[0038] Figure 12 Schematic diagram of the accuracy-confidence curve after adding the CA module.

[0039] Figure 13 Schematic diagram of the principle of K-means clustering algorithm.

[0040] Figure 14 This is a diagram of the dead chicken's zero Euler angle posture.

[0041] Figure 15 This is a schematic diagram of the front structure of the dead chicken identification, positioning and posture detection system.

[0042] Figure 16 This is a schematic diagram of the back structure of the dead chicken identification, positioning and posture detection system.

[0043] Figure 17 A schematic diagram of the structure of a computer device provided in one embodiment of the present application.

[0044] Reference numerals:

[0045] Depth camera—1, IR camera—2, visible light camera—3, laser module—4, LDP sensor—5, heat dissipation vent—6, USB 3.0 extension cable—7, DC5521 charging port—8, triangular prism—9, triangular side panel—10, fixing hole—11, carrying strap—12, SD card slot—13, power switch—14, 5.5-inch display—15. DETAILED DESCRIPTION

[0046] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0047] The purpose of this application is to provide a dead chicken identification, positioning and posture detection method, system, device, medium and product, aiming to achieve the identification, positioning and posture detection of dead chickens.

[0048] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0049] In an exemplary embodiment, Figure 1 and Figure 2 As shown, a method for identifying, locating and detecting the posture of dead chickens is provided, which includes the following steps.

[0050] Step 1: Obtain the visible light image and depth image of the chicken cage to be tested captured by the depth camera.

[0051] Specifically, before step 1, the process further includes: calibrating the visible light camera in the depth camera. Specifically,

[0052] The visible light camera in the depth camera is used to simultaneously capture more than 20 sets of visible light images of the black and white chessboard calibration plate from different angles and distances. The calibration is performed using the Zhang Zhengyou calibration method to obtain the intrinsic parameter matrix and extrinsic parameter matrix of the visible light camera.

[0053] Among them, the intrinsic parameter matrix is ​​a set of parameters that describe the imaging characteristics of the visible light camera. They are the key to converting three-dimensional world coordinates into two-dimensional image coordinates, including (focal length of the visible light camera in the x direction), (focal length of the visible light camera in the y direction), (the position of the principal point on the horizontal axis (x-axis), in pixels, representing the horizontal coordinate of the center of the image), (The position of the principal point on the vertical axis (y-axis), in pixels, representing the vertical coordinate of the center of the image) and the distortion coefficient (radial distortion coefficient and tangential distortion coefficient).

[0054] Internal parameter matrix Expressed as:

[0055] .

[0056] The external parameter matrix includes: rotation matrix and translation vectors The rotation matrix describes the rotation from the world coordinate system (or camera coordinate system) to the camera coordinate system; the translation vector describes the translation from the world coordinate system to the camera coordinate system.

[0057] Step 2: Input the visible light image of the chicken cage to be tested into the chicken recognition model to obtain the target chicken recognition result.

[0058] The target chicken recognition result is a visible light image labeled with the recognition results of each chicken in the chicken cage to be tested. When the chicken is alive, the recognition result is the prediction box of the live chicken; when the chicken is dead, the recognition result includes the prediction box of the dead chicken and multiple prediction key points of the dead chicken. The chicken recognition model is obtained by training the improved YOLOv8 network.

[0059] Specifically, the number of dead chickens can be obtained based on the number of prediction boxes of dead chickens in the target chicken recognition results.

[0060] As an optional implementation, in step 2, the training process of the chicken identification model includes steps 21 to 23.

[0061] Step 21: Obtain multiple sample images and corresponding real chicken labeling results; the sample image is a visible light image of a sample chicken cage, and the real chicken labeling result is a sample image of the real labeling result of each chicken in the sample chicken cage; when the chicken is alive, the real labeling result is the real frame of the live chicken; when the chicken is dead, the real labeling result includes the real frame of the dead chicken and multiple real key points of the dead chicken.

[0062] Specifically, we manually annotated the sample images using the Labelme Image Polygonal Annotation with Python software. We first used rectangular boxes (the ground-truth boxes) to select the chickens (both dead and alive) in the sample images. If a chicken was dead, we then annotated its head, breast, left foot, and right foot using key points (i.e., multiple ground-truth key points of a dead chicken). This generated a dataset for training the chicken recognition model.

[0063] Labelme is an open-source image annotation tool widely used in computer vision and machine learning, particularly in dataset construction and object detection. It supports multiple annotation methods, such as polygons, rectangles, line segments, and points, enabling users to accurately annotate objects with complex shapes. Labelme provides a user-friendly graphical user interface that allows users to easily scale, translate, and rotate images, facilitating detailed annotation.

[0064] After the annotation is completed, the user can save the results in JSON format, so you need to write a simple Python code to convert the JSON format annotation file into a TXT file in YOLO training format.

[0065] Step 22: Build an improved YOLOv8 network.

[0066] As an optional implementation, Figure 3-Figure 7 As shown in the figure, the improved YOLOv8 network includes: backbone network, neck network and head network.

[0067] The backbone network includes: the first convolutional layer, the first C2f module, the second convolutional layer, the second C2f module, the third convolutional layer, the third C2f module, the fourth convolutional layer, the fourth C2f module, the channel attention module and the fast spatial pyramid pooling module, which are connected in sequence.

[0068] The neck network includes: a first upsampling module, a first splicing module, a fifth C2f module, a second upsampling module, a second splicing module, a sixth C2f module, a fifth convolutional layer, a third splicing module, a seventh C2f module, a sixth convolutional layer, a fourth splicing module and an eighth C2f module, which are connected in sequence; the second C2f module is connected to the second splicing module, the third C2f module is connected to the first splicing module, and the fast spatial pyramid pooling module is connected to the first upsampling module and the fourth splicing module respectively.

[0069] The head network includes: a first detection head, a second detection head and a third detection head; the sixth C2f module is connected to the first detection head, the seventh C2f module is connected to the second detection head, and the eighth C2f module is connected to the third detection head.

[0070] Specifically, in Figure 3-Figure 7 In the figure, each convolutional layer is a basic convolutional layer, denoted by ConV; CA is the channel attention module (CA); SPPF is the fast spatial pyramid pooling module (SPPF); each upsampling module is denoted by Upsample; and each concatenation module is denoted by Concat.

[0071] like Figure 4As shown in the figure, ConV includes: a two-dimensional convolutional layer (ConV2d), a batch normalization layer (BatchNorm), and a Silu activation function connected in sequence.

[0072] like Figure 5 As shown in the figure, C2f includes: basic convolution layer, segmentation module (Split), n bottleneck layers (BottleNeck), splicing module and basic convolution layer.

[0073] like Figure 6 As shown in the figure, CA includes: residual block (Residual), X-direction average pooling layer (X Avg Pool), Y-direction average pooling layer (Y Avg Pool), splicing module, multiple two-dimensional convolutional layers, non-linear layer (Non-linear), batch normalization layer, segmentation module, Sigmoid activation function and re-weighting module (Re-weight).

[0074] like Figure 7 As shown in Figure 3, SPPF includes: 2 basic convolutional layers, a two-dimensional maximum pooling layer (MaxPool2d) and a splicing module.

[0075] The backbone of the network is based on multiple basic convolutional layers and C2f modules, which gradually extract image features and introduce a CA module. This ensures that after the feature map with an input channel of 1024 passes through the CA module, the adjusted feature map maintains rich expressive power and can dynamically highlight important information related to key points, thereby improving the overall feature recognition ability.

[0076] In the backbone network of the improved YOLOv8 network, the layers are structured as follows: first, two basic convolutional layers gradually increase the feature dimension to 512, followed by successive C2f modules (including multiple residual blocks) to further enhance feature representation, and finally, a convolutional layer to increase the feature dimension to 1024. Following the CA module, a fast spatial pyramid pooling module is used to further enhance the multi-scale information processing capabilities of the feature map. As a key enhancement module, the CA module calculates the global average pooling features of each channel and generates a weight vector. This weight vector is used to dynamically adjust the activation values ​​of each channel in the feature map, thereby directing the network to focus on features that are more important for pose estimation.

[0077] In the improved YOLOv8 network's neck network, upsampling and feature concatenation further enhance keypoint detection accuracy. The neck network first upsamples the feature map to a higher resolution and then concatenates it with the feature map of the corresponding layer in the backbone network, achieving multi-level feature fusion.

[0078] Finally, the improved YOLOv8 network outputs the key point detection results (i.e., the predicted key points of the dead chicken) through the head network.

[0079] Step 23: Using each sample image as input and the corresponding real chicken labeling results as output, the improved YOLOv8 network is trained to obtain a chicken recognition model.

[0080] Specifically, to evaluate the training of the chicken recognition model, the dataset is typically divided into training, testing, and validation sets in an 8:1:1 ratio for training and improving the YOLOv8 network. The training set, which contains a large number of sample images and corresponding real-world chicken annotations, is used for model learning and parameter adjustment. It helps the model extract features and learn spatial and morphological information about objects. The validation set is used to monitor model performance during training, helping to adjust hyperparameters and prevent overfitting, thereby ensuring the model's generalization to unseen data. By using the validation set to evaluate the accuracy of the chicken recognition model during each training cycle, it is possible to determine in real time whether the model needs further adjustment. The test set is used for the final evaluation of the chicken recognition model's performance and contains data completely independent of the training process. The chicken recognition results from the test set allow an objective measurement of the model's performance in real-world applications, yielding metrics such as accuracy, recall, and the harmonic mean of precision and recall. Through multiple rounds of iterative optimization, the chicken recognition model is able to accurately identify live and dead chickens, as well as key features of dead chickens. The chicken identification results of the chicken identification model are as follows: Figure 8 shown.

[0081] The improved YOLOv8 network is obtained by improving YOLOv8. The improved YOLOv8 network can simultaneously identify multiple live chickens, dead chickens, and key points of dead chickens by quickly inferring the input visible light images.

[0082] To improve the accuracy and adaptability of the model, the improved YOLOv8 network adds a call-joining (CA) module. This module dynamically adjusts the weights of feature channels, enhancing the expressiveness of key features and effectively suppressing the interference of redundant information.

[0083] In the improved YOLOv8 network, the CA module adaptively adjusts the weights of feature channels to improve the model's responsiveness to important features. The implementation of the CA module includes the following steps: first, adaptive height and width average pooling is performed on the input feature map to capture global information; then, channel attention weights are generated through a series of convolution operations and activation functions. During this process, the input feature map is split into two directions (horizontal (X) and vertical (Y)), and channel weights are generated through convolution for each direction. Finally, the CA module multiplies the input feature map with the generated channel attention weights to enhance the representation of important features while suppressing redundant information. This effectively improves the detection accuracy and robustness of the improved YOLOv8 network in complex scenarios.

[0084] In the process of training the same dataset, by comparing the performance indicators of the model without and with the CA module, a significant improvement was observed. Specifically, Figure 9 and Figure 10 As shown in the evaluation of the harmonic mean of precision and recall versus confidence curves, the chicken recognition model without the CA module achieved a harmonic mean of precision and recall across all categories of 0.99 at a confidence threshold of 0.620. However, after adding the CA module, the harmonic mean of precision and recall for the chicken recognition model increased to 1.00 at a confidence threshold of 0.741. This demonstrates that the introduction of the CA module significantly improved the overall performance of the chicken recognition model in the key point detection task, particularly enhancing its discriminative ability when accurately judging complex postures.

[0085] In the evaluation of the precision-confidence curve, Figure 11 and Figure 12 As shown in the figure, the accuracy of the model without the CA module remained high at 1.0 for all categories at a confidence threshold of 0.705. However, after adding the CA module, the accuracy of the chicken recognition model further improved to 1.0, and the confidence threshold was increased to 0.790. This change shows that the CA module not only strengthens the chicken recognition model's confidence in correct predictions but also improves its stability and accuracy at higher confidence levels. This means that when processing real-world data, the chicken recognition model can more effectively identify and locate key points, reducing the probability of false detections and missed detections.

[0086] In summary, the introduction of the CA module significantly improved the performance of the chicken identification model across various metrics, particularly the harmonic mean of precision and recall, and accuracy, demonstrating its effectiveness in feature extraction and importance weighting. This not only improved the performance of the chicken identification model on the training set but also provided greater reliability and accuracy for subsequent applications.

[0087] The principle of the CA module includes:

[0088] 1. Coordinate information embedding: Global pooling is performed on the input feature map along the horizontal and vertical directions respectively to generate feature vectors in the two directions. After concatenating the two feature vectors, channel attention weights are generated through convolution and nonlinear activation functions. Specifically as follows:

[0089] Let the input feature map be , where is the number of channels, and are the height and width of the input feature map respectively.

[0090] The feature vector generated by horizontal global pooling is: , where and are the pixel values of the pixel points at height and width

[0091] in the input feature map. The feature vector generated by vertical global pooling is: ,

[0092] . Concatenation and convolution: After concatenating and , channel attention weights are generated through convolution and Sigmoid activation function:

[0093] . where is the Sigmoid activation function,

[0094] is the convolution operation.

[0095] 2. Attention weight application: The generated channel attention weights are applied to the channel dimension of the input feature map respectively to enhance the model's perception ability of spatial position and inter-channel relationship. The channel attention weights are applied to the channel dimension of the input feature map respectively to obtain the enhanced feature map :

[0096] .To implement the model in production, the trained chicken recognition model was converted from a .pt (PyTorch Model File) to a .onnx (Open Neural Network Exchange) model. Finally, the .onnx model was converted to a .rknn model using a Ubuntu virtual machine configured with the RKNN (Rockchip Neural Network) virtual environment, enabling efficient execution on an Orange Pi 5 motherboard with a built-in AI accelerator, the Neural Processing Unit (NPU).

[0097] The converted RKNN model is effectively deployed on the device to implement intelligent recognition. First, the RKNN API library is called and the corresponding algorithm is written to ensure smooth model operation. The specific process is as follows: first, the model parameters are set to optimize the inference process; then, the model is read and initialized to prepare for subsequent data processing. Next, the system can acquire visible light images in real time, perform inference using the trained RKNN model, and obtain timely inference results.

[0098] Automated technology detects dead chickens and identifies key points in real-time visible light images. This process not only improves monitoring efficiency but also reduces the need for manual intervention, enabling timely identification and resolution of poultry health issues, effectively ensuring farm safety and economic profitability. Furthermore, the introduction of an automated identification system provides data support for livestock management, enabling more informed decisions and management strategies.

[0099] Step 3: When the target chicken identification results include the identification results of dead chickens, the optimized depth value of each predicted key point of the dead chicken is determined based on the depth image of the chicken cage to be tested.

[0100] As an optional implementation, step 3 includes steps 31 to 35.

[0101] Step 31: Determine any predicted key point of the dead chicken as the current key point.

[0102] Step 32: In the depth image of the chicken coop to be tested, with the current key point as the center, extract the initial depth value of each pixel in the preset neighborhood to obtain the depth data set to be clustered of the current key point.

[0103] Specifically, since the depth camera obtains the registered visible light image and depth image, the initial depth value of each key point can be directly obtained through the depth image according to the predicted key points of the dead chicken obtained in step 2.

[0104] Step 32 is specifically as follows: Extract the depth information near the key point: With the current key point as the center, extract the initial depth value of each pixel in a local area (i.e., a preset neighborhood, such as a 5×5 neighborhood). The depth image is usually a two-dimensional matrix, and each pixel stores an initial depth value. When extracting the initial depth value of each pixel in the preset neighborhood, it is necessary to ensure that the current key point is within the valid range of the depth image (i.e., does not exceed the boundary of the depth image). The initial depth values ​​of each pixel in the preset neighborhood are extracted to form the depth data set to be clustered for the current key point. ,in, ( ) is the initial depth value of the i-th pixel in the preset neighborhood of the current key point. If there are invalid depth values ​​in the preset neighborhood, they need to be removed when constructing the dataset to ensure the purity of the dataset. It will be used as input to the K-means clustering algorithm.

[0105] Step 33: Use the K-means clustering algorithm to cluster the depth data set to be clustered of the current key point to obtain multiple clusters corresponding to the current key point; each cluster includes the clustered depth values ​​of one or more pixel points.

[0106] However, due to the presence of noise in the depth image and the interference of obstacles such as wire in the caged chicken environment, the initial depth value obtained may be invalid or not the true depth information of the dead chicken. In order to solve this problem, the K-means clustering algorithm is used to perform cluster analysis on the initial depth values ​​near the key points, so as to accurately estimate the depth value of the key points. The principle of the K-means clustering algorithm is as follows Figure 13 shown.

[0107] Step 33 specifically includes:

[0108] Step 1: Initialize K-means clustering: Determine the number of clusters K1 (usually set to 2 or 3, but the specific value can be adjusted based on the complexity of the scene). Randomly select K1 initial cluster centers and randomly select K1 initial depth values ​​from the depth dataset of the key point to be clustered as the initial centers.

[0109] Step 2: Assign labels: the initial depth value of each pixel in the depth dataset to be clustered for the current key point , calculate its distance from all initial cluster centers (using Euclidean distance) and assign it to the cluster to which the nearest initial cluster center belongs. The distance calculation formula is:

[0110] .

[0111] in, For the After assigning labels, the initial depth value of each pixel will belong to a cluster.

[0112] Step 3: Update the cluster center: For each cluster, calculate the mean of all its initial depth values ​​and use the mean as the new cluster center. The update formula is:

[0113] .

[0114] in, For the cluster, Cluster The number of depth values ​​in the update. The updated cluster centers will be used for the next iteration.

[0115] Step 4: Check convergence: Compare the changes of the current cluster center with the previous cluster center. If the change of the cluster center is less than a preset threshold , or the maximum number of iterations, 100, is reached, the algorithm stops; otherwise, it returns to step 2 to continue iterating. After convergence, K-means clustering is completed, and the dataset is divided into K1 clusters.

[0116] Step 34: Determine the cluster with the largest number of pixels among the clusters corresponding to the current key point as the main cluster corresponding to the current key point.

[0117] Specifically, the main cluster represents the main depth distribution near the key point, which can effectively exclude the interference of noise and outliers.

[0118] Step 35: sorting the clustered depth values ​​of the pixels in the main cluster corresponding to the current key point, and determining the clustered depth value at the median of the sorting as the optimized depth value of the current key point.

[0119] Specifically, the median is selected as the depth estimate of the key point. The median is robust to outliers and can further improve the accuracy of depth information.

[0120] Step 4: Determine the spatial three-dimensional coordinates of each predicted key point of the dead chicken based on the optimized depth value of each predicted key point of the dead chicken.

[0121] As an optional implementation, step 4 includes steps 41 and 42.

[0122] Step 41: Obtain an intrinsic parameter matrix of a visible light camera in a depth camera that captures a visible light image of the chicken coop to be tested.

[0123] Step 42: Determine the spatial three-dimensional coordinates of each predicted key point of the dead chicken based on the optimized depth value and the internal parameter matrix of each predicted key point of the dead chicken.

[0124] Specifically, the calculation formula for the spatial three-dimensional coordinates of any key point includes:

[0125] .

[0126] .

[0127] .

[0128] in, is the x-axis component of the three-dimensional coordinate of the key point, in millimeters (mm); is the x-coordinate value of the key point in the two-dimensional coordinate system with the center of the visible light image as the origin, in pixels; is the z-axis component of the three-dimensional coordinates of the key point, in millimeters (mm); The y-axis component of the three-dimensional spatial coordinate of the key point, in millimeters (mm); is the y-coordinate value of the key point in the two-dimensional coordinate system with the center of the visible light image as the origin, in pixels; The optimized depth value of the key point, in millimeters (mm).

[0129] The three-dimensional coordinates of the predicted chicken head position are obtained through the calculation formula of the three-dimensional coordinates , predict the spatial three-dimensional coordinates of the chicken breast position point , predict the spatial three-dimensional coordinates of the chicken's left foot position And predict the spatial three-dimensional coordinates of the chicken's right foot position point .

[0130] in, To predict the x-axis component of the three-dimensional coordinates of the chicken head position point; The y-axis component of the three-dimensional coordinates of the predicted chicken head position point; To predict the z-axis component of the three-dimensional coordinates of the chicken head position point; The x-axis component of the spatial three-dimensional coordinates of the predicted chicken breast position point; The y-axis component of the spatial three-dimensional coordinates of the predicted chicken breast position point; To predict the z-axis component of the spatial three-dimensional coordinates of the chicken breast position point; Predict the x-axis component of the three-dimensional coordinates of the chicken's left foot position; Predict the y-axis component of the three-dimensional coordinates of the chicken's left foot position; Predict the z-axis component of the three-dimensional coordinates of the chicken's left foot position; Predict the x-axis component of the three-dimensional coordinates of the chicken's right foot position point; Predict the y-axis component of the three-dimensional coordinates of the chicken's right foot position point; Predict the z-axis component of the three-dimensional coordinates of the chicken's right foot position.

[0131] Step 5: Determine the spatial three-dimensional coordinates of the predicted chicken breast position point among the predicted key points of the dead chicken as the position of the dead chicken.

[0132] Step 6: Determine the posture information of the dead chicken based on the spatial three-dimensional coordinates of each predicted key point of the dead chicken.

[0133] Specifically, the dead chicken zero Euler angle posture is as follows Figure 14 As shown, The angle between the vector and its projection on the xz plane is the pitch angle. The angle between the vector and its projection on the xz plane is the roll angle. The angle between the vector and its projection on the yz plane is the yaw angle.

[0134] As an optional implementation, the predicted key points include: predicted chicken head position point, predicted chicken breast position point, predicted chicken left foot position point and predicted chicken right foot position point; posture information includes: pitch angle, roll angle and yaw angle.

[0135] Step 6 includes steps 61 to 63.

[0136] Step 61: Determine the pitch angle of the dead chicken based on the three-dimensional spatial coordinates of the predicted chicken head position point and the three-dimensional spatial coordinates of the predicted chicken breast position point.

[0137] Specifically, the calculation formula for the pitch angle is:

[0138] .

[0139] in, is the pitch angle.

[0140] Step 62: Determine the rolling angle of the dead chicken based on the three-dimensional spatial coordinates of the predicted left foot position point and the three-dimensional spatial coordinates of the predicted right foot position point of the dead chicken.

[0141] Specifically, the calculation formula of the roll angle is:

[0142] .

[0143] in, is the roll angle.

[0144] Step 63: Determine the yaw angle of the dead chicken based on the three-dimensional spatial coordinates of the predicted chicken head position point and the three-dimensional spatial coordinates of the predicted chicken breast position point.

[0145] Specifically, the calculation formula for the yaw angle is:

[0146] .

[0147] in, is the yaw angle.

[0148] In fact, after obtaining the Euler angle (i.e., posture information) of the dead chicken, the Euler angle of the dead chicken is also normalized to ensure that the value of the Euler angle is between -180° and 180°.

[0149] The method of this application is based on deep learning and depth cameras. The improved YOLOv8 network is trained to obtain a chicken recognition model. By embedding the chicken recognition model on the main control panel, the recognition and key point detection of dead chickens are realized. At the same time, the depth information of key points is accurately obtained using depth images and K-means clustering algorithms. The posture zero Euler angle is customized based on the actual situation of the chicken cage, and a dead chicken physical model is constructed. The dead chicken rotation posture solution formula is derived to achieve accurate solution from two-dimensional plane to three-dimensional posture. This method has the advantages of low computing power requirements, accurate dead chicken recognition, and good posture detection effect. It can be deployed on a low-cost edge computing platform. The calculated Euler angle is transmitted to the dead chicken picking robot through wireless or wired communication, which can achieve accurate grasping of dead chickens, effectively reducing the inspection pressure of large-scale farms, and providing important technical support for promoting the intelligent development and unmanned management of caged chicken cages.

[0150] In an exemplary embodiment, Figure 15 and Figure 16 As shown, a dead chicken identification, positioning and posture detection system is provided to implement the above-mentioned dead chicken identification, positioning and posture detection method. The dead chicken identification, positioning and posture detection system includes: a depth camera 1 and a processor (arranged inside the system, not shown in the figure); the depth camera 1 is connected to the processor.

[0151] The depth camera 1 is used to collect visible light images and depth images of the chicken cage to be tested.

[0152] The chicken identification model is deployed in the processor, which is used to:

[0153] The visible light image of the chicken cage to be tested is input into the chicken recognition model to obtain the target chicken recognition result; the target chicken recognition result is a visible light image annotated with the recognition result of each chicken in the chicken cage to be tested. When the chicken is alive, the recognition result is the prediction box of the live chicken; when the chicken is dead, the recognition result includes the prediction box of the dead chicken and multiple prediction key points of the dead chicken. The chicken recognition model is obtained by training the improved YOLOv8 network.

[0154] When the target chicken identification results include the identification results of dead chickens, the optimized depth values ​​of the predicted key points of the dead chickens are determined based on the depth image of the chicken cage to be tested.

[0155] The spatial three-dimensional coordinates of each predicted key point of the dead chicken are determined based on the optimized depth value of each predicted key point of the dead chicken.

[0156] The spatial three-dimensional coordinates of the predicted chicken breast position point among the predicted key points of the dead chicken are determined as the position of the dead chicken.

[0157] The posture information of the dead chicken is determined based on the spatial three-dimensional coordinates of each predicted key point of the dead chicken.

[0158] Specifically, the dead chicken identification, location, and posture detection system includes a depth camera 1, heat dissipation vents 6, a USB 3.0 extension cable 7, a DC5521 charging port 8, a triangular prism 9, triangular side panels 10, mounting holes 11, a hand strap 12, an SD card slot 13, a power switch 14, a 5.5-inch display 15, and a processor. The depth camera 1 includes an IR camera 2, a visible light camera 3, a laser module 4, and an LDP sensor 5.

[0159] In depth camera 1, two IR cameras 2 are used for stereo vision, capturing depth information by capturing infrared images. Visible light camera 3 captures color images or videos of the chicken coop, capturing the scene's texture for target recognition. Laser module 4 is used for laser ranging, calculating distance by emitting a laser beam and measuring the return time. The Laser Detection and Ranging (LDP) sensor, also known as a lidar, detects distance through laser reflection, offering high spatial resolution and accuracy. The IR cameras 2, laser module 4, and LDP sensor 5 work together to capture highly accurate depth images of the chicken coop.

[0160] The heat dissipation holes 6 are used to dissipate the heat generated by the main control board, depth camera and other hardware during operation in a timely manner to prevent heat accumulation during the operation of the device and cause safety hazards.

[0161] The USB 3.0 extension cable 7 is used to extend the USB 3.0 interface of the main control board to the outside of the device, so as to facilitate wired communication between the device and other peripherals.

[0162] The DC5521 charging port 8 is used to charge or power the device so that the device can operate normally in the short or long term.

[0163] The triangular prism 9 is designed to connect the main body of the dead chicken identification, positioning and posture detection system with the fixed panel below. It is used to prevent the upper and lower structures of the device from being in the same straight line, and is also beneficial to maintaining the stability of the device when it is fixed.

[0164] The design of the triangular side panel 10 is similar to that of a triangular prism, which enhances the stability of the dead chicken identification, positioning and posture detection system. It is also used to tilt the dead chicken identification, positioning and posture detection system when it is idle, preventing lens wear and accidental touch caused by direct contact between the camera, display screen, left switch slot, right charging cable and USB extension cable and the storage desktop.

[0165] The fixing hole 11 is used to fix the dead chicken identification, positioning and posture detection system to the dead chicken picking robot to realize the automated management of the chicken cage.

[0166] The hand strap 12 is used for moving the dead chicken identification, positioning and posture detection system to facilitate taking.

[0167] The SD card slot 13 is used to insert and remove the SD card, which is convenient for upgrading and reinstalling the software system of the dead chicken identification, positioning and posture detection system.

[0168] The power switch 14 is used to start or shut down the entire dead chicken identification, positioning and posture detection system.

[0169] The 5.5-inch display screen 15 is used to display the collection interface when collecting data, which is convenient for on-site inspection of the quality of the collected data; after being embedded with the chicken recognition model, it is used to display the chicken recognition results and depth images in the form of visible light images, which is convenient for management personnel to check.

[0170] The housing of the dead chicken identification, location, and posture detection system in this application adopts an integrated design with an overall trapezoidal structure, aiming to improve the ease of installation and operational stability of the dead chicken identification, location, and posture detection system. The main body of the dead chicken identification, location, and posture detection system is a rectangular parallelepiped, with a fixed panel located at the rear end of the dead chicken identification, location, and posture detection system. Combined with the following two innovative designs, the safety and effectiveness of the dead chicken identification, location, and posture detection system are ensured both in operation and in idle states.

[0171] First, a triangular prism connects the main body of the dead chicken identification, location, and posture detection system to the lower fixed panel, preventing a 90° angle between the upper and lower structures. The main body of the dead chicken identification, location, and posture detection system accounts for over 90% of the system's weight. This helps prevent breakage between the main body and the lower fixed panel, thereby improving the stability of the dead chicken identification, location, and posture detection system in its fixed state. This design effectively reduces the risk of tipping over due to a high center of gravity, ensuring the safety of the equipment during use.

[0172] Secondly, the triangular side panel design, similar in function to a triangular prism, further enhances the overall stability of the dead chicken identification, positioning, and posture detection system. It ensures the parallelism between the front camera and rear faces of the system, ensuring that the front face of the system is in a vertical plane when fixed. This structurally avoids excessive errors in subsequent spatial coordinate calculations. Furthermore, the triangular side panels support the dead chicken identification, positioning, and posture detection system when idle, preventing direct contact between the camera, display, left switch slot, right charging cable, and USB extension cable, and the desktop. This prevents lens wear and accidental touches, improving the durability of the device.

[0173] Furthermore, the provision of fixing holes 11 allows the dead chicken identification, location, and posture detection system to be fixed to the dead chicken collection robot, facilitating automated chicken house management and further enhancing the practicality and functionality of the device. Overall, the trapezoidal housing design balances aesthetics and practicality, providing sturdy protection and support for the device in practical applications.

[0174] An embedded system control board (Orange Pi 5), based on the powerful RK3588S chip, serves as the core control and data processing unit (i.e., processor) of the dead chicken identification, location, and posture detection system. This processor, anchored within the system through a structural design, boasts high performance and multitasking capabilities, making it suitable for high-load data processing scenarios. Based on a 64-bit ARM multi-core architecture, its core processing unit supports a variety of high-speed communication interfaces (such as USB, Ethernet, and HDMI), enabling stable inter-module data transmission and collaborative operation. The processor receives, processes, and integrates signals from various functional modules. Its design ensures high reliability and stability while offering flexible scalability. Choosing the Orange Pi 5 as the hardware platform maximizes the performance of the RK3588S chip while optimizing overall costs and delivering an excellent price-performance ratio. This design ensures efficient and stable operation of the dead chicken identification, location, and posture detection system in a variety of application scenarios, making it a key component in implementing the core functions of the dead chicken identification, location, and posture detection system.

[0175] In an exemplary embodiment, a computer device is provided, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement a method for identifying, locating, and detecting the posture of dead chickens.

[0176] In an exemplary embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, a method for identifying, locating, and detecting the posture of a dead chicken is implemented.

[0177] In an exemplary embodiment, a computer program product is provided, including a computer program, which, when executed by a processor, implements a method for identifying, locating, and detecting the posture of a dead chicken.

[0178] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 17 As shown. The computer device includes a processor, a memory, an input / output interface (I / O) and a communication interface. The processor, memory and input / output interface are connected via a system bus, and the communication interface is connected to the system bus via the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a method for identifying, locating and detecting the posture of a dead chicken is implemented.

[0179] Those skilled in the art will understand that Figure 17 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0180] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the above-mentioned embodiments. In particular, any reference to memory, database, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0181] The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processors involved in the various embodiments provided herein may include, but are not limited to, general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic units, data processing logic units based on quantum computing, and the like.

[0182] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0183] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0184] This document uses specific examples to illustrate the principles and implementation methods of this application. The description of the above examples is only intended to help understand the method and core concept of this application. At the same time, for those skilled in the art, based on the concept of this application, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A method for identifying, locating and detecting the posture of dead chickens, characterized in that: The dead chicken identification, positioning and posture detection method comprises: Obtaining visible light images and depth images of the chicken cage to be tested captured by the depth camera; Inputting a visible light image of the chicken cage to be tested into a chicken recognition model to obtain a target chicken recognition result; the target chicken recognition result is a visible light image annotated with the recognition results of each chicken in the chicken cage to be tested; when the chicken is alive, the recognition result is a predicted box of the live chicken; when the chicken is dead, the recognition result includes a predicted box of the dead chicken and multiple predicted key points of the dead chicken; the chicken recognition model is obtained by training an improved YOLOv8 network; When the target chicken identification results include identification results of dead chickens, determining the optimized depth values ​​of each predicted key point of the dead chickens based on the depth image of the chicken cage to be tested; Determining the spatial three-dimensional coordinates of each predicted key point of the dead chicken based on the optimized depth value of each predicted key point of the dead chicken; Determine the spatial three-dimensional coordinates of the predicted chicken breast position point among the predicted key points of the dead chicken as the position of the dead chicken; Determine the posture information of the dead chicken based on the spatial three-dimensional coordinates of each predicted key point of the dead chicken; The improved YOLOv8 network includes: a backbone network, a neck network and a head network; The backbone network includes: a first convolutional layer, a first C2f module, a second convolutional layer, a second C2f module, a third convolutional layer, a third C2f module, a fourth convolutional layer, a fourth C2f module, a channel attention module and a fast spatial pyramid pooling module connected in sequence.

2. The dead chicken identification, positioning and posture detection method according to claim 1, characterized in that: The training process of the chicken identification model includes: Obtain multiple sample images and corresponding real chicken labeling results; the sample image is a visible light image of a sample chicken cage, and the real chicken labeling result is a sample image of the real labeling result of each chicken in the sample chicken cage; when the chicken is alive, the real labeling result is the real frame of the live chicken; when the chicken is dead, the real labeling result includes the real frame of the dead chicken and multiple real key points of the dead chicken; Construct the improved YOLOv8 network; The improved YOLOv8 network is trained with each sample image as input and the corresponding real chicken labeling result as output to obtain the chicken recognition model.

3. The dead chicken identification, positioning and posture detection method according to claim 2, characterized in that: The neck network includes: a first upsampling module, a first splicing module, a fifth C2f module, a second upsampling module, a second splicing module, a sixth C2f module, a fifth convolutional layer, a third splicing module, a seventh C2f module, a sixth convolutional layer, a fourth splicing module, and an eighth C2f module connected in sequence; the second C2f module is connected to the second splicing module, the third C2f module is connected to the first splicing module, and the fast spatial pyramid pooling module is connected to the first upsampling module and the fourth splicing module respectively; The head network includes: a first detection head, a second detection head and a third detection head; the sixth C2f module is connected to the first detection head, the seventh C2f module is connected to the second detection head, and the eighth C2f module is connected to the third detection head.

4. The dead chicken identification, positioning and posture detection method according to claim 1, characterized in that: Based on the depth image of the chicken cage to be tested, the optimized depth values ​​of each predicted key point of the dead chicken are determined, including: Determine any predicted key point of the dead chicken as the current key point; In the depth image of the chicken coop to be tested, with the current key point as the center, the initial depth value of each pixel in the preset neighborhood is extracted to obtain the depth data set to be clustered of the current key point; The K-means clustering algorithm is used to cluster the depth data set of the current key point to obtain multiple clusters corresponding to the current key point; each cluster includes the clustered depth values ​​of one or more pixels; The cluster with the largest number of pixels among the clusters corresponding to the current key point is determined as the main cluster corresponding to the current key point; The clustered depth values ​​of the pixels in the main cluster corresponding to the current key point are sorted, and the clustered depth value at the median of the sorting is determined as the optimized depth value of the current key point.

5. The dead chicken identification, positioning and posture detection method according to claim 1, characterized in that: Determine the spatial three-dimensional coordinates of each predicted key point of the dead chicken based on the optimized depth value of each predicted key point of the dead chicken, including: Obtaining an intrinsic parameter matrix of a visible light camera in a depth camera that collects visible light images of the chicken coop to be tested; The spatial three-dimensional coordinates of each predicted key point of the dead chicken are determined according to the optimized depth value of each predicted key point of the dead chicken and the internal parameter matrix.

6. The dead chicken identification, positioning and posture detection method according to claim 1, characterized in that: The predicted key points include: a predicted chicken head position point, a predicted chicken breast position point, a predicted chicken left foot position point, and a predicted chicken right foot position point; the posture information includes: a pitch angle, a roll angle, and a yaw angle; Based on the three-dimensional spatial coordinates of each predicted key point of the dead chicken, the posture information of the dead chicken is determined, including: Determine the pitch angle of the dead chicken based on the spatial three-dimensional coordinates of the predicted chicken head position point and the predicted chicken breast position point; Determine the rolling angle of the dead chicken based on the spatial three-dimensional coordinates of the predicted left foot position point of the dead chicken and the spatial three-dimensional coordinates of the predicted right foot position point of the dead chicken; The yaw angle of the dead chicken is determined based on the spatial three-dimensional coordinates of the predicted chicken head position point and the predicted chicken breast position point of the dead chicken.

7. A dead chicken identification, positioning and posture detection system for implementing the dead chicken identification, positioning and posture detection method according to any one of claims 1 to 6, characterized in that: The dead chicken identification, positioning and posture detection system includes: a depth camera and a processor; the depth camera is connected to the processor; The depth camera is used to collect visible light images and depth images of the chicken cage to be tested; The processor is configured to: Inputting a visible light image of the chicken cage to be tested into a chicken recognition model to obtain a target chicken recognition result; the target chicken recognition result is a visible light image annotated with the recognition results of each chicken in the chicken cage to be tested; when the chicken is alive, the recognition result is a predicted box of the live chicken; when the chicken is dead, the recognition result includes a predicted box of the dead chicken and multiple predicted key points of the dead chicken; the chicken recognition model is obtained by training an improved YOLOv8 network; When the target chicken identification results include identification results of dead chickens, determining the optimized depth values ​​of each predicted key point of the dead chickens based on the depth image of the chicken cage to be tested; Determining the spatial three-dimensional coordinates of each predicted key point of the dead chicken based on the optimized depth value of each predicted key point of the dead chicken; Determine the spatial three-dimensional coordinates of the predicted chicken breast position point among the predicted key points of the dead chicken as the position of the dead chicken; The posture information of the dead chicken is determined based on the spatial three-dimensional coordinates of each predicted key point of the dead chicken.

8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the dead chicken identification, positioning, and posture detection method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the dead chicken identification, positioning and posture detection method according to any one of claims 1 to 6 is implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the dead chicken identification, positioning and posture detection method according to any one of claims 1 to 6 is implemented.

Images