TWI789879B - Intelligent poultry group image recognition and analysis system and method - Google Patents

Abstract

An intelligent poultry group image recognition and analysis system and method. The intelligent poultry group image recognition and analysis method includes an image identification step, an outlier analysis step, and an outlier judgment step. The image recognition step is to analyze the bird image with images of several birds, and generate a bird recognition result. The outlier analysis step is to analyze the bird identification results to mark one of the birds in the bird image as a single outlier poultry, and mark the other birds as a group bird, and analyze to obtain a single The distance between a single outlier and a group of birds. The outlier judgment step will generate outlier event data when it is judged that the outlier distance is greater than the threshold. Through the above-mentioned steps of analyzing the bird image and judging the design of outlier event data, abnormal birds can be detected early in an automated manner.

Description

Intelligent poultry group image recognition and analysis system and method

The present invention relates to an analysis method of poultry behavior, in particular to an intelligent poultry group image recognition and analysis system and method.

When birds are sick, they often have obvious clinical symptoms such as crouching, lethargy, diarrhea, and feather ruffling. These clinical symptoms will reduce the activity of the birds and easily leave the existing flock. In traditional breeding management, managers often enter the on-site poultry house to observe the behavior of the poultry to find out whether there are specific poultry that have abnormal behavior due to illness. However, the frequent entry and exit of the management staff to the poultry house will not only increase the consumption of human resources in management, but also may frighten the poultry and lead to emergency situations, which will affect the quality of life of the poultry, or affect the judgment of abnormal behavior correctness.

At present, the poultry breeding field has gradually introduced photography equipment to assist in the breeding management. The management staff can monitor the poultry house to observe the behavior of the poultry as long as they are in the office. However, it is still necessary to observe and judge the behavior of poultry by human means, which still cannot achieve the goal of automatic monitoring.

Therefore, the object of the present invention is to provide an intelligent bird flocking image recognition and analysis method that can improve at least one shortcoming of the prior art.

Therefore, the smart poultry flocking image recognition and analysis method of the present invention includes an image recognition step, an outlier analysis step, and an outlier judgment step.

In the image recognition step, an intelligent bird grouping image recognition analysis system uses a machine learning algorithm to establish a bird recognition model to analyze a bird image with images of several birds, and generate A poultry identification result, the poultry identification result including the positions of the poultry in the poultry image.

In the outlier analysis step, the intelligent poultry group image recognition analysis system analyzes the relative positional relationship of the poultry in the poultry image to the poultry image One of the birds in the image is marked as a single stray poultry, and the other poultry excluding the single stray poultry are marked as a group of poultry, and the single stray poultry and the An outlier distance between birds in the flock.

In the outlier judging step, when the intelligent poultry flocking image recognition and analysis system judges that the outlier distance is greater than a threshold value, the poultry image, the single outlier poultry and the outlier distance to generate an outlier event data.

Another object of the present invention is to provide an intelligent bird flocking image recognition and analysis system that can improve at least one shortcoming of the prior art.

Therefore, the intelligent poultry group image recognition and analysis system of the present invention is suitable for setting up in a poultry farm, and includes a camera device and an analysis server.

The camera device includes a camera unit and a control unit that is signally connected to the camera unit. The control unit controls the camera unit to capture the poultry farm image to obtain a poultry image with several poultry images.

The analysis server can analyze the bird image with a bird recognition model established by a machine learning algorithm, and generate a bird recognition result including the positions of the birds in the bird image, and The bird identification result analyzes the relative positional relationship of the birds in the bird image to mark one of the birds in the bird image as a single stray poultry, and Other birds that exclude the single outlier poultry are marked as a group of birds, and an outlier distance between the single outlier poultry and the group of birds is analyzed, and when it is judged that the outlier distance is greater than a threshold , integrating the bird image, the single outlier poultry and the outlier distance to generate an outlier event data.

Another object of the present invention is to provide an analysis server that can improve at least one shortcoming of the prior art.

Therefore, the analysis server of the present invention for intelligent bird flocking image recognition analysis includes an image recognition unit and an outlier computing unit.

The image recognition unit can analyze a bird image having images of several birds with a bird recognition model established by a machine learning algorithm, and generate a bird image including the birds in the bird image Bird identification results for location.

The outlier calculation unit can analyze the relative positional relationship of the birds in the bird image to the bird identification result, so as to mark one of the birds in the bird image as a single outlier poultry, and mark the other poultry excluding the single outlier poultry as a group of poultry, and analyze the outlier distance between the single outlier poultry and the group of poultry, and determine the distance between the group When the flock distance is greater than a threshold, the bird image, the single outlier poultry and the outlier distance are combined to generate an outlier event data.

The effect of the present invention is that: through the analysis server, the poultry image can be analyzed to generate the poultry identification result, and the poultry identification result can be analyzed to determine the design of the outlier event data, which can be automated, Early detection of birds with abnormal behavior.

Referring to FIG. 1 , an embodiment of the intelligent poultry group image recognition and analysis system of the present invention is suitable for setting up in a poultry farm and can communicate with a remote device 9 . The intelligent poultry flocking image recognition and analysis system includes a camera device 1 , a storage device 2 , a training server 3 , an analysis server 4 , and an early warning server 5 .

The camera device 1 is signally connected to the storage device 2 and the analysis server 4 , and includes a camera unit 11 , and a control unit 12 signally connected to the camera unit 11 . The control unit 12 will control the camera unit 11 to take pictures of the poultry farm at intervals of a predetermined time to obtain a poultry image with images of several poultry, and store the poultry image in the storage device 2.

Specifically, in the present embodiment, the camera unit 11 is a camera that takes pictures of the poultry farm from a bird's-eye view, and the control unit 12 is an embedded system (Embedded System) embedded in the camera unit 11. ).

It is worth noting that, in this embodiment, in order to avoid overlap of bird images in the bird images, the camera unit 11 is designed to take pictures from a top view angle, but the implementation is not limited in this way.

The storage device 2 is used to store the images of each poultry obtained by the camera device 1 from taking pictures of the poultry farm. The training server 3 is connected to the analysis server 4 through a signal, and a machine learning algorithm is used to build a bird identification model with several images of training birds, and the bird identification model is stored in the analysis server 4. In this embodiment, the machine learning algorithm uses a Convolutional Neural Network (CNN) algorithm, but the implementation is not limited thereto.

The analysis server 4 is signally connected to the storage device 2 and the early warning server 5 , and includes an image recognition unit 41 , and an outlier computing unit 42 signally connected to the image recognition unit 41 . The image recognition unit 41 will analyze each bird image with the bird recognition model when the camera device 1 obtains each bird image, and generate a pair of bird images corresponding to each bird image. Identification results. The outlier calculation unit 42 further analyzes and judges according to the poultry identification result to generate an outlier event data, and transmits the outlier event data to the early warning server 5 .

The warning server 5 includes a warning unit 51 and a communication unit 52 for communicating with the remote device 9 and connecting the warning unit 51 with a signal. The warning unit 51 sends a warning message to the remote device 9 via the communication unit 52 when the outlier event data is generated.

It should be noted that the signal connection mode between the camera device 1, the storage device 2, the training server 3, the analysis server 4, the early warning server 5 and the remote device 9 can be through the existing known wireless communication technology and/or wired communication technology to achieve. The wireless communication technologies are, for example but not limited to, wireless network technologies such as Bluetooth and Wi-Fi, and mobile communication network technologies such as 4G and 5G.

Referring to Fig. 2, an embodiment of the intelligent poultry group image recognition and analysis method performed by the intelligent poultry group image recognition analysis system of the present invention includes an image capture step 61, an image recognition step 62 . An outlier analysis step 63 , an outlier judgment step 64 , and an early warning step 65 .

In the step 61 of capturing images, the control unit 12 of the camera device 1 controls the camera unit 11 to take pictures of the poultry farm at intervals of the predetermined time to obtain the poultry images.

In the image recognition step 62, the image recognition unit 41 of the analysis server 4 analyzes each bird image with the bird recognition model, and generates the bird corresponding to each bird image Identification results. The bird identification result includes the positions of the birds in the bird image. In this embodiment, the image recognition unit 41 recognizes and analyzes the bird image through machine learning, but in other implementations, it can also use such as but not limited to binarization processing, edge detection The purpose of identifying poultry is achieved by means of measurement processing and feature engineering.

The outlier analysis step 63 includes a substep of obtaining outlier poultry and a substep of obtaining outlier distance.

In the sub-step of acquiring outlier poultry, the outlier calculation unit 42 of the analysis server 4 calculates an average center position according to the positions of the poultry in the poultry image, and analyzes and calculates each The distance between the position of the bird and the average center position is obtained to obtain a bird farthest from the average center position, and the bird is marked as a single stray poultry, and the single stray poultry will be excluded Other birds were tagged as a flock of birds.

In the sub-step of obtaining the outlier distance, the outlier calculation unit 42 calculates the position of a group center according to the positions of the birds in the group, and calculates the distance between the position of the single outlier poultry and the group center position to get the outlier distance.

In this embodiment, the calculation of the distance is obtained by using the Euclidean distance (Euclidean distance), specifically, for example, point A (x1, y1) and point B (x2, y2) are respectively on the xy plane If there are two points, add the square of the difference between the x coordinates (x2-x1) ² and the square of the difference between the y coordinates (y2-y1) ² and then open the square root to get the Euclidean distance. However, in other variations of the present invention, for example, but not limited to, Mahalanobis distance (Mahalanobis distance) can also be used for calculation.

In the outlier judgment step 64, the outlier calculation unit 42 will integrate the bird image, the single outlier poultry and the outlier distance to generate the outlier distance when it is judged that the outlier distance is greater than a threshold value. Outlier event data. What needs to be added is that the setting of the threshold needs to consider factors such as stocking density, stocking environment, and poultry activity characteristics in order to achieve the accuracy of outlier judgment.

In the early warning step 65, the early warning unit 51 of the early warning server 5 generates the early warning message when the outlier event data is generated.

To sum up, the analysis server 4 can analyze the poultry image to generate the poultry identification result, and analyze the poultry identification result to obtain the distance between the single outlier poultry and the outlier, and then judge the outlier. The design of the outlier event data can detect birds with abnormal behaviors in an automated way, which helps to reduce the dependence on manual labor in feeding and management. In addition, when the outlier event data is generated by the early warning server 5, the design of generating the early warning message can achieve active alarm, which is a very innovative and practical creation. Therefore, the purpose of the present invention can indeed be achieved.

But the above-mentioned ones are only embodiments of the present invention, and should not limit the scope of the present invention. All simple equivalent changes and modifications made according to the patent scope of the present invention and the content of the patent specification are still within the scope of the present invention. Within the scope covered by the patent of the present invention.

1: camera device

11: Camera unit

12: Control unit

2: storage device

3: Training server

4: Analysis server

41: Image recognition unit

42: Outlier computing unit

5: Early warning server

51: Early warning unit

52: Communication unit

61: Capture image step

62: Image recognition steps

63: Outlier analysis step

64: Outlier Judgment Step

65: Early warning step

9: Remote device

Other features and effects of the present invention will be clearly presented in the implementation manner with reference to the drawings, wherein: Fig. 1 is a functional block diagram illustrating an embodiment of the intelligent poultry flocking image recognition and analysis system of the present invention; and Fig. 2 is a flow chart illustrating an embodiment of the intelligent poultry flocking image recognition and analysis method of the present invention.

61: Capture image step

62: Image recognition steps

63: Outlier analysis step

64: Outlier Judgment Step

65: Early warning step

Claims (8)

An intelligent poultry group image recognition and analysis method, comprising: an image recognition step, which enables an intelligent poultry group image recognition and analysis system to analyze a poultry identification model established by a machine learning algorithm. bird images of images of birds and generate a bird identification result including the positions of the birds in the bird image; an outlier analysis step that enables the intelligent bird The bird grouping image recognition analysis system analyzes the relative positional relationship of the birds in the bird image on the bird recognition result, so as to mark one of the birds in the bird image as A single stray poultry, and other poultry excluding the single stray poultry are marked as a group of poultry, and an outlier distance between the single stray poultry and the group of poultry is obtained by analysis, the distance The group analysis step includes a sub-step of obtaining outlier poultry, so that the intelligent poultry group image recognition and analysis system can calculate an average center position according to the positions of the poultry in the poultry image, and calculate each the distance of the position of a bird from the mean center position, and the bird furthest from the mean center position is marked as the single stray bird and the other birds are marked as the group birds; and a The flock judgment step is to make the intelligent poultry flocking image recognition and analysis system integrate the poultry image, the single outlier poultry and the outlier distance to generate a Outlier event data. The intelligent poultry grouping image recognition and analysis method as described in claim 1, wherein the outlier analysis step also includes the following sub-steps: a sub-step of obtaining outlier distance, making the intelligent poultry grouping image The identification and analysis system calculates a group center position according to the positions of the birds in the group, and calculates the distance between the position of the single outlier poultry and the center position of the group to obtain the outlier distance. The intelligent poultry flocking image recognition and analysis method as described in claim 1 further includes an image capture step and an early warning step, the image capture step is to make the intelligent poultry flocking image The identification and analysis system takes a video of a poultry farm to obtain the image of the poultry, and the early warning step will cause the intelligent poultry group image identification and analysis system to generate an early warning message when the outlier event data is generated. An intelligent poultry group image recognition and analysis system is suitable for setting up in a poultry farm, and includes: a camera device, including a camera unit, and a control unit connected to the camera unit by a signal, and the control unit will controlling the camera unit to capture images of the poultry farm to obtain a bird image with images of several birds; and an analysis server capable of analyzing the bird recognition model established by a machine learning algorithm an image of birds, and generating a bird recognition result including the positions of the birds in the bird image, and analyzing the bird recognition result for the relative positions of the birds in the bird image relation to mark one of the birds in the bird image as a single stray poultry, and to mark the other birds excluding the single stray poultry as a group of poultry, and analyze to obtain an outlier distance between the single outlier poultry and the group of poultry, and when it is judged that the outlier distance is greater than a threshold value, the image of the poultry, the single outlier poultry and the outlier distance to generate an outlier event data; the analysis server will calculate an average center position based on the positions of the birds in the bird image, and calculate the relationship between the position of each bird and the The distance from the average center position, and the bird farthest from the average center position is marked as the single outlier poultry, and the other birds are marked as the group birds. In the intelligent poultry flocking image recognition and analysis system described in claim 4, the analysis server will also calculate a group center position according to the positions of the poultry in the group of poultry, and calculate the single The distance between the position of the outlier poultry and the center of the flock is calculated to obtain the outlier distance. The intelligent poultry flocking image recognition and analysis system as described in claim 4 can communicate with a remote device, and the intelligent poultry flocking image recognition and analysis system also includes an early warning server, the early warning server The device includes a communication unit for communicating with the remote device, and an early warning unit connected to the communication unit by a signal, and the early warning unit will send an early warning message to the remote device through the communication unit when the outlier event data is generated. end device. An analysis server for image recognition and analysis of intelligent poultry clusters, including: an image recognition unit capable of analyzing an image of several birds with a bird recognition model established by a machine learning algorithm bird images, and generate a bird identification including the location of the birds in the bird image recognition results; and an outlier computing unit, capable of analyzing the relative positional relationship of the birds in the bird image with respect to the bird recognition results, so that one of the birds in the bird image One is marked as a single stray poultry, and other poultry excluding the single stray poultry are marked as a group of poultry, and an outlier distance between the single stray poultry and the group of poultry is obtained by analysis , and when it is judged that the outlier distance is greater than a threshold, the bird image, the single outlier poultry and the outlier distance are combined to generate an outlier event data; the outlier computing unit will Calculate the position in the bird image to obtain an average center position, and calculate the distance between the position of each bird and the average center position, and mark the bird farthest from the average center position as the single Remove only stray birds and mark other birds as members of the flock. The analysis server as described in claim item 7, wherein, the outlier calculation unit calculates a group center position according to the positions of the poultry in the group, and calculates the position and the position of the single outlier poultry The distance between the center positions of the group to obtain the outlier distance.

Images