Methods and devices for real-time monitoring of the health status of free-range broiler chickens

Abstract

This disclosure relates to a method and apparatus for real-time monitoring of the health status of free-range broiler chickens. The method includes: acquiring images of a flock of free-range broiler chickens; inputting the flock images into a target detection model to obtain the poses and bounding boxes of the free-range broiler chickens in the flock images; and determining the health status of the free-range broiler chickens based on the poses and bounding boxes. This enables automatic detection of the health status of free-range broiler chickens. Furthermore, when the target detection model detects free-range broiler chickens in a flock image, it first extracts features at different scales from the flock image, then fuses these features to obtain a scale feature map with multi-scale information, and finally detects the pose and bounding box of each chicken by analyzing the scale feature map. This allows for focus on key parts of the free-range broiler chicken's body, accurately identifying similar poses and improving recognition accuracy.

Description

Technical Field

[0001] This disclosure relates to the field of image processing technology, and more specifically, to a method and apparatus for real-time monitoring of the health status of free-range broiler chickens. Background Technology

[0002] Avian influenza outbreaks are a significant factor hindering the development of the poultry industry. In related technologies, prevention of avian influenza outbreaks is generally based on physical and chemical testing and manual inspection.

[0003] Manual inspections are time-consuming and labor-intensive, and often result in misdiagnosis. While physicochemical testing has a higher accuracy rate, it is expensive and unsuitable for the aquaculture industry. Summary of the Invention

[0004] The purpose of this disclosure is to provide a method and device for real-time monitoring of the health status of free-range broiler chickens, in order to solve the above-mentioned technical problems.

[0005] To achieve the above objectives, the first aspect of this disclosure provides a method for real-time monitoring of the health status of free-range broiler chickens, the method comprising: Obtain images of free-range broiler chicken flocks; The chicken flock image is input into the target detection model to obtain the pose and detection box of the free-range broiler chickens in the chicken flock image; Based on the posture and the detection frame, the health status of the free-range broiler chickens is determined; The target detection model includes a feature extraction module, a feature fusion module, and a detection module. The target detection model is used to obtain the pose and detection boxes of free-range broilers in the chicken flock image through the following operations: The feature extraction module extracts features from the chicken flock image at different scales to obtain scale feature maps at different scales. The feature fusion module performs feature fusion processing on the scale feature maps of different scales to obtain multi-scale feature maps; The detection module determines the pose and detection box of each free-range broiler chicken in the flock image based on the multi-scale feature map.

[0006] Optionally, the feature extraction module includes a first feature extraction submodule, a second feature extraction submodule, a first feature enhancement submodule, and a third feature extraction submodule. The step of extracting features from the chicken image at different scales using the feature extraction module to obtain scale feature maps at different scales includes: The first feature extraction submodule performs first-scale feature extraction on the chicken flock image to obtain a first-scale feature map. The second feature extraction submodule performs global feature extraction on the first scale feature map to obtain the second scale feature map; The first feature enhancement submodule performs feature enhancement processing on the second scale feature map to obtain the third scale feature map; The third feature extraction submodule extracts features from the third-scale feature map to obtain the fourth-scale feature map.

[0007] Optionally, the first feature enhancement submodule includes a first attention mechanism unit, a first cross-stage partial connection unit, a first convolution unit, a second attention mechanism unit, and a second cross-stage partial connection unit. The step of performing feature enhancement processing on the second-scale feature map through the feature enhancement submodule to obtain a third-scale feature map includes: The first attention mechanism unit performs weighted processing on the second scale feature map to obtain the first scale feature sub-map; The first scale feature sub-map is enhanced by the first cross-stage partial connection unit to obtain the second scale feature sub-map. The first convolutional unit is used to extract features from the second-scale feature sub-map to obtain the third-scale feature sub-map. The third-scale feature sub-map is weighted by the second attention mechanism unit to obtain the fourth-scale feature sub-map; The third-scale feature map is obtained by performing feature enhancement processing on the fourth-scale feature map through the second cross-stage partial connection unit.

[0008] Optionally, the third feature extraction submodule includes a second convolutional unit, a third attention mechanism unit, a spatial pyramid pooling unit, and a fourth attention mechanism unit. The step of extracting features from the third-scale feature map using the third feature extraction submodule to obtain a fourth-scale feature map includes: The second convolutional unit is used to perform feature extraction processing on the third-scale feature map to obtain a fifth-scale feature sub-map; The fifth-scale feature sub-map is weighted by the third attention mechanism unit to obtain the sixth-scale feature sub-map. The sixth-scale feature sub-map is processed by the spatial pyramid pooling unit to extract features at different scales, resulting in a seventh-scale feature sub-map. The fourth attention mechanism unit performs weighted processing on the seventh-scale feature sub-map to obtain the fourth-scale feature map.

[0009] Optionally, the feature fusion module includes a feature fusion submodule and a second feature enhancement submodule. The step of performing feature fusion processing on the feature maps of different scales through the feature fusion module to obtain multi-scale feature maps includes: The feature fusion submodule performs feature fusion processing on the first scale feature map, the third scale feature map, and the fourth scale feature map to obtain a first fused feature map. The first fused feature map is enhanced by the second feature enhancement submodule to obtain a multi-scale feature map.

[0010] Optionally, the feature fusion submodule includes a third cross-stage partial connection unit, a third convolutional unit, a fourth cross-stage partial connection unit, a fourth convolutional unit, and a fifth attention mechanism unit. The step of fusing the first-scale feature map, the third-scale feature map, and the fourth-scale feature map through the feature fusion submodule to obtain a first fused feature map includes: The fourth-scale feature map is enhanced by the third cross-stage partial connection unit to obtain the eighth-scale feature sub-map, and the eighth-scale feature sub-map is upsampled to obtain the ninth-scale feature sub-map. The third convolutional unit performs feature fusion processing on the ninth-scale feature sub-map and the third-scale feature map to obtain a first fused feature sub-map, and then performs feature extraction processing on the first fused feature sub-map to obtain a second fused feature sub-map. The second fused feature sub-image is enhanced by the fourth cross-stage partial connection unit to obtain the third fused feature sub-image, and the third fused feature sub-image is upsampled to obtain the fourth fused feature sub-image. The fourth convolutional unit performs feature fusion processing on the fourth fused feature sub-map and the first scale feature map to obtain the fifth fused feature sub-map, and then performs feature extraction processing on the fifth fused feature sub-map to obtain the sixth fused feature sub-map. The sixth fusion feature map is obtained by weighting the sixth fusion feature map through the fifth attention mechanism unit.

[0011] Optionally, the second feature enhancement submodule includes a fifth cross-stage partial connection unit, a feature extraction unit, a sixth attention mechanism unit, a fifth convolution unit, and a seventh attention mechanism unit. The step of performing feature enhancement processing on the first fused feature map through the second feature enhancement submodule to obtain a multi-scale feature map includes: The first fused feature map is enhanced by the fifth cross-stage partial connection unit to obtain the second fused feature map; The third fusion feature map is obtained by performing global feature extraction on the second fusion feature map through the feature extraction unit. The first fused feature map and the third fused feature map are fused by the sixth attention mechanism unit to obtain the fourth fused feature map, and the fourth fused feature map is weighted to obtain the fifth fused feature map. The sixth fusion feature map is obtained by performing feature enhancement processing on the fifth fusion feature map through the sixth cross-stage partial connection unit; The sixth fused feature map is processed by the fifth convolutional unit to obtain the seventh fused feature map; The seventh attention mechanism unit performs weighted processing on the seventh fusion feature map to obtain a multi-scale feature map.

[0012] Optionally, the target detection model is trained in the following manner: Acquire a sample chicken flock image, a first label, and a second label. The first label is used to indicate the true detection box of free-range broilers in the sample chicken flock image, and the second label is used to indicate the posture of free-range broilers in the sample chicken flock image. The sample chicken flock image is input into the target detection model to obtain the predicted detection box of free-range broiler chickens in the sample chicken flock image; Based on the predicted detection box and the first label, a first loss function value is determined, and based on the predicted detection box and the second label, a second loss function value is determined. The parameters of the target detection model are updated based on the first loss function value and the second loss function value.

[0013] Optionally, the value of the first loss function is determined by the following formula:

[0014]

[0015] in, This represents the value of the first loss function. A parameter representing the degree of suppression of outliers. A Represents the predicted detection box. B This represents the actual detection bounding box. b Indicates the center point of the predicted detection box. This represents the center point of the actual detection box. represents the Euclidean distance between the ground truth bounding box and the predicted bounding box, and c represents the diagonal distance between the smallest bounding box containing both the predicted and ground truth bounding boxes.

[0016] A second aspect of this disclosure provides a real-time monitoring device for the health status of free-range broiler chickens, the device comprising: Mobile devices; An image acquisition device is mounted on the mobile device; A processor is configured to control the movement of the mobile device to enable the image acquisition device located on the mobile device to acquire images at different locations, and to perform the method described in any one of the first aspects based on the acquired images.

[0017] The above technical solution allows for the direct input of images of free-range broiler chickens into a target detection model, yielding the bounding box and pose of each chicken. This allows for the determination of each chicken's health status based on its bounding box and pose, thus enabling automatic detection of the health status of free-range broiler chickens. Furthermore, when detecting free-range broiler chickens in an image, the target detection model first extracts features at different scales from the image, then fuses these features to obtain a scale feature map with multi-scale information. Finally, the pose and bounding box of each chicken are obtained by detecting these scale feature maps. This allows for focus on key body parts of the free-range broiler chickens, accurately identifying similar poses and improving recognition accuracy.

[0018] Other features and advantages of this disclosure will be described in detail in the following detailed description section. Attached Figure Description

[0019] The accompanying drawings are provided to further illustrate the present disclosure and form part of the specification. They are used together with the following detailed description to explain the present disclosure, but do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart illustrating a method for real-time monitoring of the health status of free-range broiler chickens according to an exemplary embodiment of this disclosure; Figure 2 This is a schematic diagram illustrating the effect of a CA attention mechanism according to an exemplary embodiment of the present disclosure; Figure 3 This is a structural block diagram of a target detection model according to an exemplary embodiment of the present disclosure; Figure 4 This is a schematic diagram illustrating the training process of an object detection model according to an exemplary embodiment of the present disclosure; Figure 5 This is a schematic diagram of the structure of a real-time health status monitoring device for free-range broiler chickens according to an exemplary embodiment of the present disclosure; Figure 6 This is a schematic diagram of the structure of an image acquisition device according to an exemplary embodiment of the present disclosure; Figure 7 This is a block diagram illustrating an electronic device according to an exemplary embodiment of the present disclosure; Figure 8This is a block diagram illustrating another electronic device according to an exemplary embodiment of the present disclosure.

[0020] Explanation of reference numerals in the attached figures 1. First mobile device; 2. Second mobile device; 3. Image acquisition device; 4. Base; 5. First fixing plate; 6. Temperature and humidity sensor; 7. Power supply; 8. Camera; 9. Second fixing plate; 10. Processor. Detailed Implementation

[0021] The specific embodiments of this disclosure will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit this disclosure.

[0022] It should be noted that all actions involving the acquisition of signals, information, or data in this disclosure are carried out in compliance with the relevant data protection laws and policies of the country where the location is situated, and with authorization from the owner of the relevant device.

[0023] Poultry diseases not only cause economic losses to the poultry industry, but more seriously, the influenza they develop can also threaten human health. Therefore, poultry diseases are increasingly becoming a focus of social concern. Currently, poultry monitoring methods mainly rely on physical and chemical testing and manual inspection. Manual inspection involves farmers observing the poultry's posture, feathers, comb, droppings, and sounds to make a preliminary diagnosis of their health. Manual observation is time-consuming and labor-intensive, and misdiagnosis is common. Physical and chemical testing requires experts and is expensive, making it unsuitable for the poultry industry.

[0024] In view of this, the present disclosure provides a method and apparatus for real-time monitoring of the health status of free-range broilers to overcome the above-mentioned technical problems.

[0025] The embodiments of this disclosure will be further explained below with reference to the accompanying drawings.

[0026] Figure 1 This is a flowchart illustrating a method for real-time monitoring of the health status of free-range broiler chickens according to an exemplary embodiment of this disclosure, with reference to... Figure 1 The method may include the following steps: S101: Obtain images of free-range broiler chickens.

[0027] Because overhead cameras can only capture a top-down view of the broilers, they cannot capture the specific postures of the broilers' legs and heads. This results in the broilers primarily displaying postures such as standing, drinking, and eating, making it difficult to obtain more accurate images of their movements. Therefore, this embodiment utilizes side-view cameras to capture images of free-range broilers.

[0028] S102: Input the chicken flock image into the target detection model to obtain the pose and detection box of the free-range broiler chickens in the chicken flock image.

[0029] The target detection model can be a detection model in related technologies, or it can be an improved model of a detection model in related technologies. This disclosure does not impose any restrictions on this.

[0030] For example, the object detection model in this embodiment can be an improved version of YOLOv5 (You Only Look Once v5) in related technologies. Specifically, in the YOLOv5 model in related technologies, the backbone network uses the C3 (Concentrated-Comprehensive Convolution Block) module in the traditional CNN (Convolutional Neural Networks) network structure. Because the convolutional layers within C3 are too simple, there is a problem of missed detections in global object detection, especially in this application scenario where the size of free-range broiler chickens varies, making missed detections very easy. At the same time, because the body structure of broiler chickens is simple, but there are as many as 12 poses to be detected, this leads to excessively high similarity between poses, greatly complicating the model's accurate detection. Therefore, this embodiment introduces MobileViTv3 and CA (Coordinate Attention) modules on top of the YOLOv5 model. By replacing the second C3 module in the backbone network and the medium-sized object detection head in the neck network with MobileViTv3, the characteristics of the Transformer in MobileViTv3 are utilized to extract global image features. Since CA has the advantage of simultaneously focusing on objects in both spatial and channel dimensions, adaptively learning channel weights helps the model effectively focus on useful channel information. Simultaneously, GradCAM (Gradient-weighted Class Activation Mapping) gradient calculation is used, and the results clearly show the important regions of interest, such as... Figure 2 As shown, key locations and detailed information can be focused on in various scenarios, not only locating all targets but also achieving accurate recognition of botnet-like poses. Therefore, by adding CA modules to the backbone and neck networks, leveraging the CA modules' sensitivity to targets in terms of orientation and position, the accuracy of botnet pose recognition is improved when botnets are densely packed. The improved model is denoted as MCA-YOLOv5, and its structure is as follows: Figure 3 As shown.

[0031] In order for MCA-YOLOv5 to output the pose and bounding box of each free-range broiler chicken in the input chicken flock image, it is also necessary to train MCA-YOLOv5 using training data.

[0032] For example, such as Figure 4 As shown, we can first construct the total number. Zhang's dataset contains images of free-range broiler chickens from multiple scenes, breeds, and angles, including images of indoor free-range chickens. The number of images of Zhang and outdoor free-range animals is The number of images of animals being raised outdoors in fenced enclosures is [number]. Zhang. Then, the poses and bounding boxes of free-range broiler chickens in the image dataset were manually labeled to obtain the training set. Finally, the training set was randomly divided into a training set, a test set, and a validation set in a 7:2:1 ratio, and MCA-YOLOv5 was trained based on these sets. Specifically, the training set was used to train MCA-YOLOv5, the validation set was used to determine the optimal hyperparameters of MCA-YOLOv5, and the test set was used to determine the prediction performance of MCA-YOLOv5. When the prediction performance of MCA-YOLOv5 reached the preset prediction performance, the training of MCA-YOLOv5 was considered complete.

[0033] In training MCA-YOLOv5 using the training set, images of free-range broiler chickens, labeled broiler poses, and broiler detection boxes from the training set can be input into MCA-YOLOv5 to obtain predicted broiler detection boxes. Then, based on the predicted and labeled broiler detection boxes, a first loss function value is determined, and based on the predicted and labeled broiler poses, a second loss function value is determined. Finally, the parameters of MCA-YOLOv5 are updated based on the first and second loss function values. In other words, the object detection model can be trained in the following way: Acquire a sample chicken flock image, a first label, and a second label. The first label indicates the true detection bounding box of the free-range broiler chickens in the sample chicken flock image, and the second label indicates the true pose of the free-range broiler chickens in the sample chicken flock image. Input the sample chicken flock image into the target detection model to obtain the predicted detection bounding box of the free-range broiler chickens in the sample chicken flock image. Determine a first loss function value based on the predicted detection bounding box and the first label, and determine a second loss function value based on the predicted detection bounding box and the second label. Update the parameters of the target detection model based on the first loss function value and the second loss function value.

[0034] It should be understood that, in images of free-range broiler chickens, the number of healthy chickens significantly exceeds the number of unhealthy chickens; that is, there are significantly more chicken poses representing healthy chickens than representing unhealthy chickens, resulting in a class imbalance problem. Simultaneously, due to the simple structure and diverse behaviors of broiler chickens, the problem of difficult and easy samples also arises. To overcome these two problems, this embodiment introduces the Focal-loss function, the formula of which is as follows:

[0035] Given an offset variable , express Location, express Location, Let the displacement deviation loss value be represented. Then, the gradient of the Focal loss function can be shown as follows:

[0036] Combining the DIOU loss function with the Focal-loss function allows us to... Replace with , The formula for calculating the loss value of the DIOU loss function is as follows:

[0037] Replaced It is no longer a constant, but infinitely close to 0, at the same time Since the gradient is also close to 0, the overall gradient will be smaller after multiplication, reducing the impact of reweighting. The impact of smaller bounding boxes. To address this issue, the IOU values ​​are used to reweight the DIOU loss, resulting in the final desired Focal-DIOU function relationship, as follows:

[0038] In the formula , It is a parameter that controls the degree of suppression of outliers.

[0039] In other words, among the possible methods, the value of the first loss function can be determined by the following formula:

[0040]

[0041] in, This represents the value of the first loss function. A parameter representing the degree of suppression of outliers.A Represents the predicted detection box. B This represents the actual detection bounding box. b Indicates the center point of the predicted detection box. This represents the center point of the actual detection box. This represents the Euclidean distance between the actual detection box and the predicted detection box. c This represents the diagonal distance between the smallest bounding box containing the predicted detection box and the ground truth detection box.

[0042] After training the target detection model using the above training method, the chicken flock image can be input into the target detection model to obtain the pose and detection box of the free-range broiler chicken in the chicken flock image.

[0043] S103: Based on the posture and the detection frame, determine the health status of the free-range broiler chicken.

[0044] For example, free-range broilers are considered healthy when they are standing, eating, drinking, running, preening, fighting, resting, flapping their wings, or walking. They are considered unhealthy when they are standing with their necks tucked in, limping, or burying their heads under their wings.

[0045] The target detection model includes a feature extraction module, a feature fusion module, and a detection module. The target detection model is used to obtain the pose and detection boxes of free-range broilers in the chicken flock image through the following operations: The feature extraction module extracts features from the chicken flock image at different scales to obtain scale feature maps at different scales; the feature fusion module performs feature fusion processing on the scale feature maps at different scales to obtain multi-scale feature maps; and the detection module determines the pose and detection box of each free-range broiler chicken in the chicken flock image based on the multi-scale feature maps.

[0046] To better understand the real-time health monitoring method for free-range broiler chickens in this disclosure, the following provides a detailed explanation of the processing of chicken flock images by the target detection model.

[0047] In a possible implementation, the feature extraction module includes a first feature extraction submodule, a second feature extraction submodule, a first feature enhancement submodule, and a third feature extraction submodule. The step of extracting features from the chicken image at different scales using the feature extraction module to obtain scale feature maps at different scales includes: The first feature extraction submodule extracts features at a first scale from the chicken image to obtain a first-scale feature map; the second feature extraction submodule extracts global features from the first-scale feature map to obtain a second-scale feature map; the first feature enhancement submodule enhances the features of the second-scale feature map to obtain a third-scale feature map; and the third feature extraction submodule extracts features from the third-scale feature map to obtain a fourth-scale feature map.

[0048] For example, such as Figure 3 As shown, the first feature extraction submodule can be composed of three CBS (Cross-Stage-Partial-connections) and one C3 (Concentrated-Comprehensive ConvolutionBlock). Thus, after the chicken flock image is transmitted to the first feature extraction submodule, it can be processed by CBS1, CBS2, C3 and CBS3 in sequence to obtain the first scale feature map.

[0049] The second feature extraction submodule can be composed of MobileViTv3. Thus, after the first-scale feature map is transmitted to the second feature extraction submodule, the characteristics of Transformer in MobileViTv3 can be used to extract global features from the first-scale feature map and obtain the second-scale feature map.

[0050] The first feature enhancement submodule can consist of two Coordinate Attention (CA) units, two Coordinate Shift Block (CBS) units, and one Coordinate Shift Block (C3) unit. Thus, after the second-scale feature map is transmitted to the first feature enhancement submodule, it can be processed sequentially by CA1, CBS1, C3, CA2, and CBS2 to obtain the third-scale feature map. That is, in this embodiment, the first feature enhancement submodule can include a first attention mechanism unit, a first cross-stage partial connection unit, a first convolution unit, a second attention mechanism unit, and a second cross-stage partial connection unit. The step of performing feature enhancement processing on the second-scale feature map through the feature enhancement submodule to obtain the third-scale feature map can include: The first attention mechanism unit performs weighted processing on the second-scale feature map to obtain a first-scale feature sub-map; the first cross-stage partial connection unit performs feature enhancement processing on the first-scale feature sub-map to obtain a second-scale feature sub-map; the first convolution unit performs feature extraction processing on the second-scale feature sub-map to obtain a third-scale feature sub-map; the second attention mechanism unit performs weighted processing on the third-scale feature sub-map to obtain a fourth-scale feature sub-map; and the second cross-stage partial connection unit performs feature enhancement processing on the fourth-scale feature sub-map to obtain a third-scale feature map.

[0051] The third feature extraction submodule can consist of a C3, a SPPF (Spatial Pyramid Pooling), and two CAs. Thus, after the third-scale feature map is transmitted to the third feature extraction submodule, it can be processed sequentially by C3, CA1, SPPF, and CA2 to obtain the fourth-scale feature map. That is, the third feature extraction submodule in this embodiment can include a second convolutional unit, a third attention mechanism unit, a spatial pyramid pooling unit, and a fourth attention mechanism unit. The step of extracting features from the third-scale feature map using the third feature extraction submodule to obtain the fourth-scale feature map includes: The second convolutional unit extracts features from the third-scale feature map to obtain a fifth-scale feature sub-map; the third attention mechanism unit weights the fifth-scale feature sub-map to obtain a sixth-scale feature sub-map; the spatial pyramid pooling unit extracts features from the sixth-scale feature sub-map at different scales to obtain a seventh-scale feature sub-map; and the fourth attention mechanism unit weights the seventh-scale feature sub-map to obtain a fourth-scale feature map.

[0052] In a possible implementation, the feature fusion module includes a feature fusion submodule and a second feature enhancement submodule. The step of performing feature fusion processing on the feature maps of different scales through the feature fusion module to obtain multi-scale feature maps may include: The feature fusion submodule performs feature fusion processing on the first scale feature map, the third scale feature map, and the fourth scale feature map to obtain a first fused feature map; the second feature enhancement submodule performs feature enhancement processing on the first fused feature map to obtain a multi-scale feature map.

[0053] For example, such as Figure 3As shown, the feature fusion submodule can consist of two CBSs, two C3s, and one CA. Therefore, after the first-scale feature map, the third-scale feature map, and the fourth-scale feature map are transmitted to the feature fusion submodule, they can be processed by CBS1, C31, CBS2, C32, and CA to obtain the first fused feature map. That is, the feature fusion submodule in this embodiment includes a third cross-stage partial connection unit, a third convolution unit, a fourth cross-stage partial connection unit, a fourth convolution unit, and a fifth attention mechanism unit. The feature fusion process performed by the feature fusion submodule on the first-scale feature map, the third-scale feature map, and the fourth-scale feature map to obtain the first fused feature map may include: The fourth-scale feature map is enhanced by the third cross-stage partial connection unit to obtain an eighth-scale feature sub-map, which is then upsampled to obtain a ninth-scale feature sub-map. The ninth-scale feature sub-map and the third-scale feature map are fused by the third convolutional unit to obtain a first fused feature sub-map, which is then upsampled to obtain a second fused feature sub-map. The second fused feature sub-map is enhanced by the fourth cross-stage partial connection unit to obtain a third fused feature sub-map, which is then upsampled to obtain a fourth fused feature sub-map. The fourth fused feature sub-map and the first-scale feature map are fused by the fourth convolutional unit to obtain a fifth fused feature sub-map, which is then upsampled to obtain a sixth fused feature sub-map. The sixth fused feature sub-map is weighted by the fifth attention mechanism unit to obtain a first fused feature map.

[0054] The second feature enhancement submodule can consist of two CBSs, one MobileViTv3, two CAs, and one C3. Thus, after the first fused feature map is transmitted to the second feature enhancement submodule, it can be processed sequentially by CBS1, MobileViTv3, CA1, CBS2, C3, and CA2 to obtain a multi-scale feature map. That is, in this embodiment, the second feature enhancement submodule includes a fifth cross-stage partial connection unit, a feature extraction unit, a sixth attention mechanism unit, a fifth convolution unit, and a seventh attention mechanism unit. The step of performing feature enhancement processing on the first fused feature map through the second feature enhancement submodule to obtain a multi-scale feature map can include: The first fused feature map is enhanced by the fifth cross-stage partial connection unit to obtain a second fused feature map; the second fused feature map is then subjected to global feature extraction by the feature extraction unit to obtain a third fused feature map; the first fused feature map and the third fused feature map are then fused by the sixth attention mechanism unit to obtain a fourth fused feature map, and the fourth fused feature map is then weighted to obtain a fifth fused feature map; the fifth fused feature map is then enhanced by the sixth cross-stage partial connection unit to obtain a sixth fused feature map; the sixth fused feature map is then extracted by the fifth convolution unit to obtain a seventh fused feature map; and the seventh fused feature map is then weighted by the seventh attention mechanism unit to obtain a multi-scale feature map.

[0055] It is worth noting that, except for the number 3 in C3, the numbers in the other modules, submodules, or units indicate which position. For example, CA1 in the second feature addition submodule indicates the first CA in the second feature addition submodule, and CBS2 in the second feature addition submodule indicates the second CBS in the second feature addition submodule.

[0056] The above technical solution allows for the direct input of images of free-range broiler chickens into a target detection model, yielding the bounding box and pose of each chicken. This allows for the determination of each chicken's health status based on its bounding box and pose, thus enabling automatic detection of the health status of free-range broiler chickens. Furthermore, when detecting free-range broiler chickens in an image, the target detection model first extracts features at different scales from the image, then fuses these features to obtain a scale feature map with multi-scale information. Finally, the pose and bounding box of each chicken are obtained by detecting these scale feature maps. This allows for focus on key body parts of the free-range broiler chickens, accurately identifying similar poses and improving recognition accuracy.

[0057] Based on the same concept, this disclosure also provides a device for real-time monitoring of the health status of free-range broiler chickens, the device including: Mobile devices; The mobile device can be any device that enables the image acquisition device to acquire images from different locations, and this disclosure does not impose any limitations on this.

[0058] For example, a mobile device can be as follows Figure 5As shown, the device includes a base 4, a first moving device 1, and a second moving device 2. The first moving device 1 is mounted on the base 5 and is used to move up and down along the base 5. The second moving device 2 is mounted on the first moving device 1 and is used to move left and right along the first moving device 1. Therefore, when the image acquisition device 3 is mounted on the second moving device 2, the image acquisition device 3 can acquire images from different positions by moving the first moving device 1 up and down and / or moving the second moving device 2 left and right.

[0059] Image acquisition device 3 is mounted on the mobile device; For example, the image acquisition device 3 can be as follows: Figure 6 As shown, the device includes a first fixing plate 5, a second fixing plate 9, a power supply 7, and a camera 8. The first fixing plate 5 is mounted on the second moving device 2, the camera 8 is mounted on the first fixing plate 5 via the second fixing plate 9, and the power supply 7 is mounted on the first fixing plate 5 and electrically connected to the camera 8. Therefore, when the first moving device 1 moves the second moving device 2 up and down, the image acquisition device 3 can move up and down with the second moving device 2, and / or, when the second moving device 2 moves left and right, the image acquisition device 3 can move left and right with the second moving device 2, thereby achieving image acquisition at different positions.

[0060] The processor 10 is used to control the movement of the mobile device so that the image acquisition device 3 located on the mobile device can acquire images at different locations, and perform the steps of any of the above-described methods for real-time monitoring of the health status of free-range broilers based on the acquired images.

[0061] Through the above technical solution, the image acquisition device 3 can move freely in four directions. This allows breeders to track one or more free-range broilers using the real-time health monitoring device, and also enables large-scale monitoring based on a single camera, reducing monitoring costs. Furthermore, in the real-time health monitoring method for free-range broilers, images of the flock can be directly input into the target detection model to obtain the bounding box and posture of each chicken in the flock image. The health status of each chicken can then be determined based on its bounding box and posture. This enables automatic detection of the health status of free-range broilers. On the other hand, when the target detection model detects free-range broilers in the flock image, it first extracts features at different scales from the flock image, then fuses these features to obtain a scale feature map with multi-scale information. Finally, it detects the posture and bounding box of each chicken by analyzing the scale feature map. This allows for focus on key parts of the free-range broiler's body, accurately identifying similar postures and improving recognition accuracy.

[0062] In a possible implementation, in order to understand the environmental conditions of free-range broilers in a timely manner, the mobile device is also equipped with a temperature and humidity sensor 6.

[0063] Based on the same concept, embodiments of this disclosure also provide a computer-readable medium having a computer program stored thereon, which, when executed by a processing device, implements the steps of any of the above-described methods for real-time monitoring of the health status of free-range broilers.

[0064] Based on the same concept, this disclosure also provides an electronic device that may include: A storage device on which computer programs are stored; A processing device is used to execute the computer program in the storage device to implement the steps of any of the above-described methods for real-time monitoring of the health status of free-range broilers.

[0065] Based on the same concept, this disclosure also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of any of the above-described methods for real-time monitoring of the health status of free-range broilers.

[0066] Figure 7 This is a block diagram illustrating an electronic device 700 according to an exemplary embodiment. Figure 7 As shown, the electronic device 700 may include a processor 701 and a memory 702. The electronic device 700 may also include one or more of a multimedia component 703, an input / output (I / O) interface 704, and a communication component 705.

[0067] The processor 701 controls the overall operation of the electronic device 700 to complete all or part of the steps in the aforementioned method for real-time monitoring of the health status of free-range broiler chickens. The memory 702 stores various types of data to support the operation of the electronic device 700. This data may include, for example, instructions for any application or method operating on the electronic device 700, and application-related data such as contact data, sent and received messages, images, audio, video, etc. The memory 702 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. Multimedia component 703 may include a screen and an audio component. The screen may be, for example, a touchscreen, and the audio component is used to output and / or input audio signals. For example, the audio component may include a microphone for receiving external audio signals. The received audio signals may be further stored in memory 702 or transmitted via communication component 705. The audio component also includes at least one speaker for outputting audio signals. I / O interface 704 provides an interface between processor 701 and other interface modules, such as a keyboard, mouse, buttons, etc. These buttons may be virtual or physical buttons. Communication component 705 is used for wired or wireless communication between the electronic device 700 and other devices. Wireless communication, such as Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G, 4G, NB-IoT, eMTC, or other 5G technologies, or combinations thereof, is not limited here. Therefore, the corresponding communication component 705 may include: a Wi-Fi module, a Bluetooth module, an NFC module, etc.

[0068] In an exemplary embodiment, the electronic device 700 may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above-described method for real-time monitoring of the health status of free-range broilers.

[0069] In another exemplary embodiment, a computer-readable storage medium including program instructions is also provided, which, when executed by a processor, implement the steps of the above-described method for real-time monitoring of the health status of free-range broilers. For example, the computer-readable storage medium may be the memory 702 including the program instructions, which may be executed by the processor 701 of the electronic device 700 to complete the above-described method for real-time monitoring of the health status of free-range broilers.

[0070] Figure 8 This is a block diagram illustrating an electronic device 800 according to an exemplary embodiment. For example, the electronic device 800 may be provided as a server. (Refer to...) Figure 8 The electronic device 800 includes a processor 822, which may be one or more, and a memory 832 for storing computer programs executable by the processor 822. The computer program stored in the memory 832 may include one or more modules, each corresponding to a set of instructions. Furthermore, the processor 822 may be configured to execute the computer program to perform the aforementioned method for real-time monitoring of the health status of free-range broiler chickens.

[0071] Additionally, the electronic device 800 may also include a power supply component 826 and a communication component 850. The power supply component 826 can be configured to perform power management of the electronic device 800, and the communication component 850 can be configured to enable communication of the electronic device 800, such as wired or wireless communication. Furthermore, the electronic device 800 may also include an input / output (I / O) interface 858. The electronic device 800 can operate on an operating system stored in the memory 832.

[0072] In another exemplary embodiment, a computer-readable storage medium including program instructions is also provided, which, when executed by a processor, implement the steps of the above-described method for real-time monitoring of the health status of free-range broilers. For example, the non-transitory computer-readable storage medium may be the memory 832 including the program instructions, which may be executed by the processor 822 of the electronic device 800 to complete the above-described method for real-time monitoring of the health status of free-range broilers.

[0073] In another exemplary embodiment, a computer program product is also provided, the computer program product comprising a computer program executable by a programmable device, the computer program having a code portion for performing the above-described method for real-time monitoring of the health status of free-range broilers when executed by the programmable device.

[0074] The preferred embodiments of the present disclosure have been described in detail above with reference to the accompanying drawings. However, the present disclosure is not limited to the specific details of the above embodiments. Within the scope of the technical concept of the present disclosure, various simple modifications can be made to the technical solutions of the present disclosure, and these simple modifications all fall within the protection scope of the present disclosure.

[0075] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, this disclosure will not describe the various possible combinations separately.

[0076] Furthermore, various different embodiments of this disclosure can be combined in any way, as long as they do not violate the spirit of this disclosure, they should also be regarded as the content disclosed in this disclosure.

Claims

1. A method for real-time monitoring of the health status of free-range broiler chickens, characterized in that, The method includes: Acquire images of free-range broiler chickens using a side-view camera; The chicken flock image is input into the target detection model to obtain the pose and detection box of the free-range broiler chickens in the chicken flock image; Based on the posture and the detection frame, the health status of the free-range broiler chickens is determined; The target detection model includes a feature extraction module, a feature fusion module, and a detection module. The target detection model is used to obtain the pose and detection boxes of free-range broilers in the chicken flock image through the following operations: The feature extraction module extracts features from the chicken flock image at different scales to obtain scale feature maps at different scales. The feature fusion module performs feature fusion processing on the scale feature maps of different scales to obtain multi-scale feature maps; The detection module determines the pose and detection box of each free-range broiler chicken in the flock image based on the multi-scale feature map. The target detection model is trained in the following manner: Acquire a sample chicken flock image, a first label, and a second label. The first label is used to indicate the true detection box of free-range broilers in the sample chicken flock image, and the second label is used to indicate the true posture of free-range broilers in the sample chicken flock image. The sample chicken flock image is input into the target detection model to obtain the predicted detection box of free-range broiler chickens in the sample chicken flock image; Based on the predicted detection box and the first label, a first loss function value is determined, and based on the predicted detection box and the second label, a second loss function value is determined. The parameters of the target detection model are updated based on the first loss function value and the second loss function value.

2. The method for real-time monitoring of the health status of free-range broiler chickens according to claim 1, characterized in that, The feature extraction module includes a first feature extraction submodule, a second feature extraction submodule, a first feature enhancement submodule, and a third feature extraction submodule. The feature extraction module performs feature extraction at different scales on the chicken image to obtain scale feature maps at different scales, including: The first feature extraction submodule performs first-scale feature extraction on the chicken flock image to obtain a first-scale feature map. The second feature extraction submodule performs global feature extraction on the first scale feature map to obtain the second scale feature map; The first feature enhancement submodule performs feature enhancement processing on the second scale feature map to obtain the third scale feature map; The third feature extraction submodule extracts features from the third-scale feature map to obtain the fourth-scale feature map.

3. The method for real-time monitoring of the health status of free-range broiler chickens according to claim 2, characterized in that, The first feature enhancement submodule includes a first attention mechanism unit, a first cross-stage partial connection unit, a first convolution unit, a second attention mechanism unit, and a second cross-stage partial connection unit. The step of performing feature enhancement processing on the second-scale feature map through the feature enhancement submodule to obtain a third-scale feature map includes: The first attention mechanism unit performs weighted processing on the second scale feature map to obtain the first scale feature sub-map; The first scale feature sub-map is enhanced by the first cross-stage partial connection unit to obtain the second scale feature sub-map. The first convolutional unit is used to extract features from the second-scale feature sub-map to obtain the third-scale feature sub-map. The third-scale feature sub-map is weighted by the second attention mechanism unit to obtain the fourth-scale feature sub-map; The third-scale feature map is obtained by performing feature enhancement processing on the fourth-scale feature map through the second cross-stage partial connection unit.

4. The method for real-time monitoring of the health status of free-range broiler chickens according to claim 2, characterized in that, The third feature extraction submodule includes a second convolutional unit, a third attention mechanism unit, a spatial pyramid pooling unit, and a fourth attention mechanism unit. The step of extracting features from the third-scale feature map using the third feature extraction submodule to obtain a fourth-scale feature map includes: The second convolutional unit is used to perform feature extraction processing on the third-scale feature map to obtain a fifth-scale feature sub-map; The fifth-scale feature sub-map is weighted by the third attention mechanism unit to obtain the sixth-scale feature sub-map. The sixth-scale feature sub-map is processed by the spatial pyramid pooling unit to extract features at different scales, resulting in a seventh-scale feature sub-map. The fourth attention mechanism unit performs weighted processing on the seventh-scale feature sub-map to obtain the fourth-scale feature map.

5. The method for real-time monitoring of the health status of free-range broiler chickens according to claim 4, characterized in that, The feature fusion module includes a feature fusion submodule and a second feature enhancement submodule. The feature fusion module performs feature fusion processing on the feature maps of different scales to obtain multi-scale feature maps, including: The feature fusion submodule performs feature fusion processing on the first scale feature map, the third scale feature map, and the fourth scale feature map to obtain the first fused feature map; The first fused feature map is enhanced by the second feature enhancement submodule to obtain a multi-scale feature map.

6. The method for real-time monitoring of the health status of free-range broiler chickens according to claim 5, characterized in that, The feature fusion submodule includes a third cross-stage partial connection unit, a third convolutional unit, a fourth cross-stage partial connection unit, a fourth convolutional unit, and a fifth attention mechanism unit. The feature fusion submodule performs feature fusion processing on the first-scale feature map, the third-scale feature map, and the fourth-scale feature map to obtain a first fused feature map, including: The fourth-scale feature map is enhanced by the third cross-stage partial connection unit to obtain the eighth-scale feature sub-map, and the eighth-scale feature sub-map is upsampled to obtain the ninth-scale feature sub-map. The third convolutional unit performs feature fusion processing on the ninth-scale feature sub-map and the third-scale feature map to obtain a first fused feature sub-map, and then performs feature extraction processing on the first fused feature sub-map to obtain a second fused feature sub-map. The second fused feature sub-image is enhanced by the fourth cross-stage partial connection unit to obtain the third fused feature sub-image, and the third fused feature sub-image is upsampled to obtain the fourth fused feature sub-image. The fourth convolutional unit performs feature fusion processing on the fourth fused feature sub-map and the first scale feature map to obtain the fifth fused feature sub-map, and then performs feature extraction processing on the fifth fused feature sub-map to obtain the sixth fused feature sub-map. The sixth fusion feature map is obtained by weighting the sixth fusion feature map through the fifth attention mechanism unit.

7. The method for real-time monitoring of the health status of free-range broiler chickens according to claim 5, characterized in that, The second feature enhancement submodule includes a fifth cross-stage partial connection unit, a feature extraction unit, a sixth attention mechanism unit, a fifth convolution unit, and a seventh attention mechanism unit. The second feature enhancement submodule performs feature enhancement processing on the first fused feature map to obtain a multi-scale feature map, including: The first fused feature map is enhanced by the fifth cross-stage partial connection unit to obtain the second fused feature map; The third fusion feature map is obtained by performing global feature extraction on the second fusion feature map through the feature extraction unit. The sixth attention mechanism unit performs feature fusion processing on the first fusion feature map and the third fusion feature map to obtain a fourth fusion feature map, and then performs weighted processing on the fourth fusion feature map to obtain a fifth fusion feature map; The sixth fusion feature map is obtained by performing feature enhancement processing on the fifth fusion feature map through the sixth cross-stage partial connection unit; The sixth fused feature map is processed by the fifth convolutional unit to obtain the seventh fused feature map; The seventh attention mechanism unit performs weighted processing on the seventh fusion feature map to obtain a multi-scale feature map.

8. The method for real-time monitoring of the health status of free-range broiler chickens according to claim 1, characterized in that, The first loss function value is determined by the following formula: in, This represents the value of the first loss function. A parameter representing the degree of suppression of outliers. A Represents the predicted detection box. B This represents the actual detection bounding box. b Indicates the center point of the predicted detection box. This represents the center point of the actual detection box. represents the Euclidean distance between the ground truth bounding box and the predicted bounding box, and c represents the diagonal distance between the smallest bounding box containing both the predicted and ground truth bounding boxes.

9. A real-time health status monitoring device for free-range broiler chickens, characterized in that, The device includes: Mobile devices; An image acquisition device is mounted on the mobile device; A processor is configured to control the movement of the mobile device so that the image acquisition device located on the mobile device acquires images at different locations, and to perform the method according to any one of claims 1 to 8 based on the acquired images.

Images