Intelligent poultry farm performance monitoring and data collection system

Abstract

An intelligent poultry farm monitoring system is disclosed, and comprises at least one house-end electronic device, at least one camera, at least one IMU device, at least one ToF device, at least one weighing device, and at least one user-end electronic device. By using the intelligent poultry farm monitoring system, a variety of monitoring data of the poultry farmed in a poultry house can be collected real time, at least including weight, activity, feed consumption, drinking water consumption, and operation condition of apparatus and / or devices. Moreover, by processing and / or analyzing the monitoring data, the system is able to conduct a poultry health prediction and / or a poultry growth prediction. Furthermore, the system is also able to generate, by analyzing and / or processing the monitoring data, an optimized feeding management policy for the poultry farmer.

Description

INTELLIGENT POULTRY FARM PERFORMANCE MONITORING AND DATA COLLECTION SYSTEMBACKGROUND OF THE INVENTION1. Field of the Invention

[0001] The present invention relates to the technology field of monitoring systems, and more particularly, to an intelligent poultry farm monitoring and data collection system.2. Description of the Prior Art

[0002] It is well known that there is a need to conduct a daily monitoring during poultry farming. Accordingly, China Patent Publication No. CN110515341A discloses an intelligent feeding management device for feeding farm, which comprises an ingredient weighing module, a feed mixing and stirring module, a feeding module, a plurality of sensors, a local computer device like smart phone, a cloud server, in which the sensors are adopted for collecting feeding data from the ingredient weighing module and the feeding module. Moreover, a poultry farmer is allowed to, by operating an intelligent feeding management platform that is shown on the smart phone thereof, remotely monitor and control the ingredient weighing module, the feed mixing and stirring module, and the feeding module.

[0003] According to the China Patent Publication No. CN110515341 A, the collected feeding data is real time uploaded to the cloud server, such that the cloud server subsequently processes the feeding data, thereby generating and providing relative information comprising monitoring images and sensing data, through the intelligent feeding management platform, to the poultry farmer. However, practical experience has revealed that, a computing interruption of the cloud server certainly takes place in case of the feeding data uploading being stopped due to the suddennetwork dropping. In such case, the intelligent feeding management platform stops to show the monitoring images and the sensing data. In addition, performance of the bird can be caused by a multitude of factors that cannot be just from feeding and water consumption along.

[0004] Therefore, above descriptions have expressed that there is still room for improvement in the intelligent feeding management device for feeding farm that is disclosed by China Patent Publication No. CN110515341 A. In view of that, inventors of the present invention have made great efforts to make inventive research and eventually provided an intelligent poultry farm monitoring system.

[0005] According to the PCT Patent Publication No. WO 2014153626A2. The patent relates to a system and method for monitoring in real time the nutriment intake (feed and / or water) of farm animals, including but not limited to poultry, turkeys, pigs and cows, based on the measurement and identification of the nutriment uptake sounds and / or the working of the nutriment supply system, while the presence and eating position of the animals is monitored by using real-time image analysis.

[0006] However, the system may only employ a limited number of sensor types, such as cameras, lacking other environmental sensors like temperature, humidity, and gas concentration. This results in the system's inability to collect comprehensive poultry house environmental data, limiting in-depth analysis and prediction of poultry health. Image analysis by the system may be limited to simple object detection and counting, lacking more detailed analysis of poultry behavior and posture. This limits the system's application in early disease warning and abnormal behavior detection. The system may not employ Al algorithms to process and analyze the collected data. This limits the system's capabilities in data mining, pattern recognition, and predictive analysis.

[0007] According to the China Patent Publication No. CN 112067040A. The patent discloses a farm inspection monitoring system, which comprises a mobile inspection subsystem and a fixed monitoring subsystem, and is characterized in that environmental data transmission and control signal transmission are earned out between the mobile inspection subsystem and the fixed monitoring subsystem through an RS485 communication interface; the fixed monitoring subsystem comprises an infrared emitter and a fixed processor, and the infrared emitter is electrically connected with the fixed processor; the mobile inspection subsystem comprises an infrared receiver and a mobile processor, the infrared receiver is electrically connected with the mobile processor, and the mobile inspection subsystem confirms the position through infrared rays received by the infrared receiver. According to the invention, local environment regulation and control can be carried out when the local environment data of the farm are abnormal.

[0008] However, the patent has low data transfer rate of infrared communication limits real-time data transmission and remote-control capabilities. The short transmission distance of infrared communication is highly susceptible to environmental factors, which can lead to unstable data transmission. The system lacks advanced Al algorithms and machine learning techniques, making it difficult to achieve intelligent analysis and decision-making. The system primarily focuses on environmental monitoring and control, lacking monitoring and analysis of individual poultry status.

[0009] According to the China Patent Publication No. CN107549049A. The invention relates to an automatic cage-rearing chicken health state monitoring device, and belongs to the technical field of poultry raising. The technical effect that the health state of chicken individuals is judged through the number and body temperature, detected by the automatic cage-rearing chicken health state monitoringdevice, of the chicken individuals, and missing inspection of manual inspection is avoided. According to the technical scheme for achieving the technical effect, the automatic cage-rearing chicken health state monitoring device comprises a vidicon, a thermal imaging camera, an inspection device, an image processing module, a display module and an alarm, wherein the vidicon and the thermal imagery camera are arranged on the inspection device, the output end of the vidicon and the output end of the thermal imagery camera are electrically connected with the image processing module, and the output end of the image processing module is electrically connected with the display module and the alarm. The automatic cage-rearing chicken health state monitoring device can be widely applied to the field of poultry raising.

[0010] However, the patent primarily focuses on the quantity and body temperature of chickens, with a relatively narrow scope of functionality. The patent primarily relies on visual and thermal imaging data for analysis, lacking consideration of other environmental factors such as humidity, temperature, and gas concentration. The patent relies heavily on manually set thresholds to judge the health status of chickens, lacking self-learning and adaptive capabilities. The visual and thermal imaging technologies used in this patent are susceptible to environmental factors such as lighting and dust, which can lead to inaccurate monitoring results.SUMMARY OF THE INVENTION

[0011] The primary objective of the present invention is to disclose an intelligent system for application in poultry farming. The intelligent poultry farm monitoring and data collection system includes practical merits of;(1) carrying out real-time monitoring in poultry environmental and performance parameters including weight, growth, activities, feed consumption, drinking water consumption, and operation condition of apparatus and / or devices;(2) conducting a poultry health prediction and / or a poultry growth prediction based on relative monitoring data; and(3) optimizing, by analyzing and / or processing relative monitoring data, poultry growth prediction model, poultry weight prediction model, feeding management policy and / or environment conditions.

[0012] In order to achieve the aforementioned objective, a first embodiment (on-device computing) of the intelligent poultry farm monitoring system is provided, which comprises:at least one first electronic device, comprising at least one camera, a first processor and a first storage module storing an application program;at least one ToF (Time of Flight) device, being integrated in the first electronic device; andat least one weighing device, being coupled to the first electronic device; wherein the first processor executes the application program so as to be configured to:acquire, by controlling the at least one camera, a plurality of poultry images from a plurality of poultry farmed in a poultry house;evaluate, by processing the plurality of poultry images, at least one poultry physical condition of each of the plurality of poultry,count, by processing the plurality of poultry images, a poultry number of the plurality of poultry;collect, by activating the at least one ToF device, a height data of the first electronic device;determine, whether at least one of the plurality of poultry is standing, sitting down or lying down according to the poultry physical condition; collect, through the at least one weighing device, a poultry feed weight data from at least one automation feeding apparatus that is disposed in the poultry house;count, by processing the poultry feed weight data, a feed consumption and / or a drinking water consumption; andcollect, through the at least one weighing device, a poultry weight data from the plurality of poultry.

[0013] In one embodiment, the intelligent poultry farm monitoring system further comprises:a PoE (Power over Ethernet) splitter, being integrated in the at least one first electronic device;at least one PoE injector, being in communication with the PoE splitter through an Ethernet cable, being linked to Internet through an Ethernet device, and being electrical connected to an AC power source via a power cable;at least one microphone, being integrated in the first electronic device; and at least one environmental sensor, being integrated in the first electronic device.

[0014] In one embodiment, the application program comprises a pre-trained poultry growth prediction model and a pre-trained poultry weight prediction model, and the first processor is further configured to:input, the plurality of poultry images, the feed consumption, the drinking water consumption, and the poultry weight data into the pre-trained poultry growth prediction model, such that the pre-trained poultry growth prediction model outputs a poultry growth prediction data; andinput, the feed consumption, the drinking water consumption, and the poultryweight data into the pre-trained poultry weight prediction model, such that the pretrained poultry weight prediction model outputs a poultry weight prediction data.

[0015] In one embodiment, the application program comprises a pre- trained weight management policy optimizing model, and the first processor is further configured to:input, the feed consumption, the drinking water consumption, and the poultry weight data into the pre-trained weight management policy optimizing model, such that the pre-trained weight management policy optimizing model outputs an optimized feeding management policy.

[0016] In one embodiment, the first electronic device is further integrated with at least one IMU (inertial measurement unit) device therein, and the first processor is further configured to:receive, an inertial data from the IMU device; andcalculate, by processing the inertial data, a camera angle of the camera.

[0017] In one embodiment, the first processor is further configured to:transmit, through the PoE splitter and the PoE injector, a monitoring data comprising the plurality of poultry images, the feed consumption, the drinking water consumption, the poultry weight data, the poultry growth prediction data, and the poultry weight prediction data to a second electronic device;wherein the second electronic device is selected from a group consisting of local server, remote server, cloud computer and cloud storage device.

[0018] In one embodiment, the second electronic device is selected from a group consisting of remote server, cloud computer and cloud storage device.

[0019] In one embodiment, the intelligent poultry farm monitoring system further comprises an electronic device capable of communicating with the first electronic device and / or the second electronic device, and the electronic device comprises aprocessor, a display module and a storage module storing a user-end application program, such that the processor executes the user-end application program so as to be configured toshow, by controlling the display module, a user operation interface;show, through the user operation interface, a dashboard comprising a plurality of statistic graphs and data charts; andshow, a real-time monitoring images of the plurality of poultry farmed in the poultry house.

[0020] In one embodiment, the dashboard is accessible via an app, a webapp, and a web browser selected from a group consisting of Chrome, Safari, Firefox, Edge, web or webapp.

[0021] In one embodiment, the first processor is further configured to:generate and transmit, in case of the camera angle is in need of adjustment, a first notification signal to the electronic device;generate and transmit, in case of the automation feeding apparatus is in need of feed adding or in need of water adding, a second notification signal to the electronic device;generate and transmit, in case of at least one of the plurality of poultry is sitting or lying down for a period of time, a third notification signal to the electronic device;generate and transmit, in case of the poultry number revealing that there is at least one of the plurality of poultry missing, a fourth notification signal to the electronic device; andgenerate and emit, in case of a sudden network dropping, an alarm signal.

[0022] In one embodiment, the camera is an electronic pan-tilt-zoom (ePTZ) camera, and the first processor is further configured to:conduct, a camera adjustment to the camera;wherein the camera adjustment is selected from a group consisting of illumination adjustment, camera angle adjustment, camera sampling rate adjustment, and basic camera parameters.

[0023] In order to achieve the aforementioned objective, a second embodiment (edge computing) of the intelligent poultry farm monitoring system is provided, which comprises:at least one first electronic device, being integrated with at least one camera, a first processor, a first storage module storing a first application program, and at least one ToF (Time of Flight) device therein;at least one weighing device, being coupled to the first electronic device; and a second electronic device in communication with the at least one first electronic device, functioning as a local computing device and comprising a second processor and a second storage module storing a second application program;wherein the first processor executes the first application program so as to be configured to:acquire, by controlling the at least one camera, a plurality of poultry images from a plurality of poultry farmed in a poultry house;collect, by activating the at least one ToF device, a height data of the first electronic device;collect, through the at least one weighing device, a poultry feed weight data from at least one automation feeding apparatus that is disposed in the poultry house;collect, through the at least one weighing device, a poultry weight data from the plurality of poultry;wherein the second processor executes the second application program so as to be configured to:evaluate, by processing the plurality of poultry images, at least one poultry physical condition of each of the plurality of poultry,count, by processing the plurality of poultry images, a poultry number of the plurality of poultry;determine, whether at least one of the plurality of poultry is standing, sitting down or lying down according to the poultry physical condition; count, by processing the poultry feed weight data, a feed consumption and / or a drinking water consumption.

[0024] In one practicable embodiment, the intelligent poultry farm monitoring system further comprises:a PoE (Power over Ethernet) splitter, being integrated in the at least one first electronic device;at least one PoE injector, being in communication with the PoE splitter through an Ethernet cable, being linked to Internet through an Ethernet device, and being electrical connected to an AC power source via a power cable;at least one microphone, being integrated in the first electronic device; and at least one environmental sensor, being integrated in the first electronic device.

[0025] In one embodiment, the second application program comprises a pretrained poultry growth prediction model and a pre-trained poultry weight prediction model, and the second processor is further configured to:input, the plurality of poultry images, the feed consumption, the drinking water consumption, and the poultry weight data into the pre-trained poultry growth prediction model, such that the pre-trained poultry growth prediction model outputs a poultry growth prediction data; andinput, the feed consumption, the drinking water consumption, and the poultry weight data into the pre-trained poultry weight prediction model, such that the pretrained poultry weight prediction model outputs a poultry weight prediction data.

[0026] In one embodiment, the second application program comprises a pretrained weight management policy optimizing model, and the second processor is further configured to:input, the feed consumption, the drinking water consumption, and the poultry weight data into the pre-trained weight management policy optimizing model, such that the pre-trained weight management policy optimizing model outputs an optimized weight management policy.

[0027] In one embodiment, the first electronic device is further integrated with at least one IMU (inertial measurement unit) device therein, and the first processor is further configured to:receive, an inertial data from the IMU device.

[0028] In one embodiment, the second processor is further configured to:calculate, by processing the inertial data, a camera angle of the camera.

[0029] In another one practicable embodiment, the intelligent poultry farm monitoring system further comprises an electronic device capable of communicating with the second electronic device, and the electronic device comprising a processor, a display module and a storage module storing a user-end application program, such that the processor executes the user-end application program so as to be configured to:show, by controlling the display module, a user operation interface; show, through the user operation interface, a dashboard comprising a plurality of statistic graphs and data charts; andshow, a real-time monitoring images of the plurality of poultry farmed in the poultry house.

[0030] In one embodiment, the dashboard is accessible via an app, a webapp, and a web browser selected from a group consisting of Chrome, Safari, Firefox, Edge, web or webapp.

[0031] In one embodiment, the second processor is further configured to:generate and transmit, in case of the camera angle is in need of adjustment, a first notification signal to the electronic device;generate and transmit, in case of the automation feeding apparatus is in need of feed adding or in need of water adding, a second notification signal to the electronic device;generate and transmit, in case of at least one of the plurality of poultry is sitting or lying down for a period of time, a third notification signal to the electronic device;generate and transmit, in case of the poultry number revealing that there is at least one of the plurality of poultry missing, a fourth notification signal to the electronic device; andgenerate and emit, in case of a sudden network dropping, an alarm signal.

[0032] In one embodiment, the camera is an electronic pan-tilt-zoom (ePTZ) camera, and the first processor is further configured to:conduct, a camera adjustment to the camera;wherein the camera adjustment is selected from a group consisting of illumination adjustment, camera angle adjustment, camera sampling rate adjustment, and basic camera parameters.

[0033] In one embodiment, the second application program comprises a pretrained image segmentation and labelling model that is utilized for firstly segmentingeach of the plurality of poultry images, thereby generating a plurality of segmented image and then labelling each of the plurality of segmented image.

[0034] In one embodiment, the second application program comprises a pretrained background subtraction algorithm that is utilized for subtracting a background including poultry dung from each of the plurality of poultry images.

[0035] For achieving the aforementioned objective, a third embodiment (cloud computing) of the intelligent poultry farm monitoring system is provided, which comprises:at least one first electronic device, being integrated with at least one camera, a first processor, a first storage module storing a first application program, and at least one ToF (Time of Flight) device therein;at least one weighing device, being coupled to the first electronic device; and a cloud computing device in communication with the at least one first electronic device, and comprising a second processor and a second storage module storing a second application program;wherein the first processor executes the first application program so as to be configured to:acquire, by controlling the at least one camera, a plurality of poultry images from a plurality of poultry farmed in a poultry house;collect, by activating the at least one ToF device, a height data of the first electronic device;collect, through the at least one weighing device, a poultry feed weight data from at least one automation feeding apparatus that is disposed in the poultry house; andcollect, through the at least one weighing device, a poultry weight data from the plurality of poultry;wherein the second processor executes the second application program so as to be configured to:evaluate, by processing the plurality of poultry images, at least one poultry physical condition of each of the plurality of poultry;count, by processing the plurality of poultry images, a poultry number of the plurality of poultry;determine, whether at least one of the plurality of poultry is standing, sitting down or lying down according to the poultry physical condition; and count, by processing the poultry feed weight data, a feed consumption and / or a drinking water consumption.

[0036] In a practicable embodiment, the intelligent poultry farm monitoring system according to the present invention further comprises:at least one PoE (Power over Ethernet) splitter, being integrated in the at least one first electronic device;at least one PoE injector, being in communication with the PoE splitter through an Ethernet cable, being linked to Internet through an Ethernet device, and being electrical connected to an AC power source via a power cable;at least one microphone, being integrated in the first electronic device; and at least one environmental sensor, being integrated in the first electronic device.

[0037] In one embodiment, the second application program comprises a pretrained poultry growth prediction model and a pre-trained poultry weight prediction model, and the second processor is further configured to:input, the plurality of poultry images, the feed consumption, the drinking water consumption, and the poultry weight data into the pre-trained poultry growthprediction model, such that the pre-trained poultry growth prediction model outputs a poultry growth prediction data; andinput, the feed consumption, the drinking water consumption, and the poultry weight data into the pre-trained poultry weight prediction model, such that the pretrained poultry weight prediction model outputs a poultry weight prediction data.

[0038] In one embodiment, the second application program comprises a pretrained feeding management policy optimizing model, and the second processor is further configured to:input, the feed consumption, the drinking water consumption, and the poultry weight data into the pre-trained feeding management policy optimizing model, such that the pre-trained feeding management policy optimizing model outputs an optimized feeding management policy.

[0039] In one embodiment, the first electronic device is further integrated with at least one IMU (inertial measurement unit) device therein, and the first processor is further configured to:receive, an inertial data from the IMU device.

[0040] In one embodiment, the second processor is further configured to:calculate, by processing the inertial data, a camera angle of the camera.

[0041] In another one practicable embodiment, the intelligent poultry farm monitoring system further comprises an electronic device capable of communicating with the second electronic device, and the electronic device comprising a processor, a display module and a storage module storing a user-end application program, such that the processor executes the user-end application program so as to be configured to:show, by controlling the display module, a user operation interface;show, through the user operation interface, a dashboard comprising a plurality of statistic graphs and data charts; andshow, a real-time monitoring images of the plurality of poultry farmed in the poultry house.

[0042] In one embodiment, the dashboard is accessible via an app, a webapp, and a web browser selected from a group consisting of Chrome, Safari, Firefox, Edge, web or webapp.

[0043] In one embodiment, second processor is further configured to:generate and transmit, in case of the camera angle is in need of adjustment, a first notification signal to the electronic device;generate and transmit, in case of the automation feeding apparatus is in need of feed adding or in need of water adding, a second notification signal to the electronic device;generate and transmit, in case of at least one of the plurality of poultry is sitting or lying down for a period of time, a third notification signal to the electronic device;generate and transmit, in case of the poultry number revealing that there is at least one of the plurality of poultry missing, a fourth notification signal to the electronic device; andgenerate and emit, in case of a sudden network dropping, an alarm signal.

[0044] In one embodiment, the camera is an electronic pan-tilt-zoom (ePTZ) camera, and the first processor is further configured to:conduct, a camera adjustment to the camera;wherein the camera adjustment is selected from a group consisting of illumination adjustment, camera angle adjustment, camera sampling rate adjustment, and basic camera parameters.

[0045] In one embodiment, the second application program comprises a pretrained image segmentation and labelling model that is utilized for firstly segmenting each of the plurality of poultry images, thereby generating a plurality of segmented image and then labelling each of the plurality of segmented image.

[0046] In one embodiment, the second application program comprises a pretrained background subtraction algorithm that is utilized for subtracting a background including poultry dung from each of the plurality of poultry images.BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The invention as well as a preferred mode of use and advantages thereof will be best understood by referring to the following detailed description of an illustrative embodiment in conjunction with the accompanying drawings, wherein:

[0048] FIG. 1A shows a first schematic stereo diagram of an intelligent poultry farm monitoring system according to the present invention.

[0049] FIG. IB shows a second schematic stereo diagram of the intelligent poultry farm monitoring system according to the present invention.

[0050] FIG. 2 shows a third schematic stereo diagram of the intelligent poultry farm monitoring system according to the present invention.

[0051] FIG. 3 shows a fourth schematic stereo diagram of the intelligent poultry farm monitoring system according to the present invention.DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0052] To more clearly describe an intelligent poultry farm monitoring system according to the present invention, embodiments of the present invention will be described in detail with reference to the attached drawings hereinafter.

[0053] First embodiment (on-device computing)

[0054] With reference to FIG. 1A and FIG. IB, there are shown a first schematic stereo diagram and a second schematic stereo diagram of an intelligent poultry farm monitoring system according to the present invention. In first embodiment, the intelligent poultry farm monitoring system 1 comprises at least one first electronic device 11 integrated with at least one camera 12, a first processor 1 IP, a first storage module 1 IM storing a first application program, at least one IMU device 13, and at least one ToF device 14, and at least one PoE splitter 16 therein, at least one weighing device 15, at least one PoE injector 17, at least one microphone (i.e., sound sensor), and at least one environmental sensor (i.e., sensor for monitoring temperature and humidity). Briefly speaking, in the first embodiment the first electronic device 11 is a machine vision device.

[0055] For example, according to FIG. 1A and FIG. IB there are two cameras 12 integrated in the first electronic device 11. To be more specific, the PoE injector 17 is in communication with the PoE splitter 16 through an Ethernet cable, and is linked to Internet through an Ethernet device (e.g., router or Ethernet switch). It should be known that the PoE injector 17 includes an AC / DC conversion circuit and / or at least one DC / DC conversion circuit. In practical use, the PoE injector 17 is electrical connected to an AC power source via a power cable, so as to convert an AC power received from the AC power source to a DC power, and then combine an Ethernet signal and the DC power to a specific Ethernet signal. In contrast to the PoE injector 17, the PoE splitter 16 is used for separating the DC power from the specific Ethernet signal, thereby supplying the DC power to the first electronic device 11 and the camera 12. Moreover, the PoE splitter 16 also separates the Ethernet signal from the specific Ethernet signal, so as to transmit the Ethernet signal to the first electronic device 11.

[0056] As described in more detail below, the IMU device 13 comprises accelerometer and gyroscope, and is adopted for collecting an inertial data of the machine vision device. On the other hand, the ToF device 14 is adopted for collecting a height data of the machine vision device.

[0057] In addition, FIG. 1A and FIG. 1A depict that the weighing device 15 is coupled to the first electronic device 11, and consists of a load cell 151 and a digital meter 152, of which the load cell 151 is connected to a weighing platform (not shown). In practical use, the load cell 151 generates a signal in response to the body weight of a poultry (like broiler, pullets, breeders, hens, egg-laying duck, or meat duck) stays in the weighing platform. Moreover, after receiving the signal from the load cell 151, the digital meter 152 amplifies the signal, and then extracts a weight data from the amplified signal.

[0058] In practical use the first storage module 1 IM can be, but may not be limited to hard disk drive (HDD), solid state drive (SSD) or flash memory chip, and is arranged to store at least one application program including instructions for configuring the first processor IIP. In any practicable embodiments, the application program can be edited by any possible programming language such as C, python, and Nodejs. As such, after the first electronic device 11 is booted, the first processor IIP executes the application program so as to be configured to communicate and / or control the camera 12, the IMU device 13, the ToF device 14, the weighing device 15, the microphone, and the environmental sensor, thereby enabling the first electronic device 11 (i.e., machine vision device) to perform a plurality of specific functions, including: real-time image capturing, real-time data collecting and data analyzing and processing.

[0059] To be more specific, while performing the functions of real-time image capturing and data analyzing and processing, the first processor 1 IP is configured to:acquire, by controlling the at least one camera 12, a plurality of poultry images from a plurality of poultry farmed in a poultry house;evaluate, by processing the plurality of poultry images, at least one poultry physical condition of each of the plurality of poultry;count, by processing the plurality of poultry images, a poultry number of the plurality of poultry; anddetermine, whether at least one of the plurality of poultry is standing, sitting down or lying down according to the poultry physical condition.

[0060] In practical use, the at least one physical condition includes but may not limited to, body size, evaluated weight, and poultry activity condition, and the first processor 1 IP is able to determine whether each of the plurality of poultry is standing, sitting down or lying down according to the poultry physical condition. Therefore, if the poultry is found to sit or lie for a period of time, the first electronic device 11 generates and transmits a notification signal to an electronic device 3 owned by a poultry farmer.

[0061] To be more specific, after processing the poultry images, relative features are extracted from the poultry image, wherein the features include but may not be limited to, image area, width, length, segment perimeter length, back width, back height, surface area, volume, eccentricity, segment radius, and max radius (when poultry is flapping). After that, the first processor 1 IP evaluates at least one poultry physical condition of the poultry by inputting the features into at least one pre-trained Al model. For example, one specific Al model is executed by the first processor 1 IP to estimate a weight value of the poultry based on the body shape and / or the features. On the other hand, another one specific Al model is executed to estimate the poultry activity condition based on the body shape and / or the features, in which the poultry activity condition includes abnormal activity due to illness or hurt.

[0062] It is worth further explaining that, if the poultry number reveals that there is at least one of the plurality of poultry missing, the first electronic device 11 correspondingly generates and transmits an alarm signal or a notification signal to the electronic device 3 owned by the poultry farmer.

[0063] In a practicable embodiment, the application program further comprises a pre-trained background subtraction algorithm that is utilized for subtracting a background including poultry dung from each of the plurality of poultry images. Moreover, in another one practicable embodiment, the application program further comprises a pre-trained image segmentation and labelling model that is utilized for firstly segmenting each of the plurality of poultry images while the poultry number counting, thereby generating a plurality of segmented image and then labelling each of the plurality of segmented image. As such, it is more easily to obtain each poultry’ s features through feature extraction process.

[0064] It needs to further explain that, if there is the poultry weight data from the weighing device 15, the poultry weight data will also be included in the labelling. Furthermore, the application program is further set to include a pre-trained object recognition model, and the pre-trained object recognition model will automatically select object (i.e., poultry) from the inputted poultry image.

[0065] On the other hand, the first electronic device 11 is further integrated with an LED illumination module, such that an illumination light emitted by the LED illumination module can be directed to the environment of the poultry house, thereby improving the quality of the captured poultry images.

[0066] In addition, the PoE injector 17 is linked to Internet through an Ethernet like router or Ethernet switch, such that the first electronic device 11 is in communication with the electronic device 3 via Internet. Therefore, the first electronic device 11 is configured to transmit, through the PoE splitter and the PoEinjector, a monitoring data comprising the plurality of poultry images, the feed consumption, the drinking water consumption, the poultry weight data, the poultry growth prediction data, and the poultry weight prediction data to the electronic device 3.

[0067] The electronic device 3, such as a smartphone, a tablet computer, a laptop computer, a desktop computer, or an all-in-one computer, comprises a processor, a display module and a storage module storing a user-end application program, of which the processor executes the user-end application program so as to be configured to control the display module to:show, by controlling the display module, a user operation interface; show, through the user operation interface, a dashboard comprising a plurality of statistic graphs and data charts; andshow, a real-time monitoring images of the plurality of poultry farmed in the poultry house.

[0068] As such, the poultry farmer is able to see, by operating the user operation interface, a real-time monitoring images of the poultry from the display module of the electronic device 3, such that the poultry farmer is able to predominate a realtime activity condition of each poultry. Moreover, by operating the user operation interface shown on the screen of the display module of the electronic device 3 like smartphone, the dashboard is accessible via an app, a webapp, and a web browser selected from a group consisting of Chrome, Safari, Firefox, Edge, wherein the dashboard exhibits a plurality of statistic graphs and data charts. For example, the dashboard includes statistic graph or data chart of weight data and statistic graph or data chart of body growth data, in which the statistic graph of weight data includes a first trend line (i.e., weight growth prediction line), and the statistic graph of body growth data includes a second trend line (i.e., body growth prediction line).

[0069] As described in more detail below, in case of performing the functions of real-time data collecting and data analyzing and processing, the first processor IIP is configured to:receive, an inertial data from the IMU device 13;calculate, by processing the inertial data, a camera angle of the camera 12; collect, by activating the at least one ToF device 14, a height data of the first electronic device 11;collect, through the at least one weighing device 15, a poultry feed weight data from at least one automation feeding apparatus that is disposed in the poultry house;count, by processing the poultry feed weight data, a feed consumption and / or a drinking water consumption;collect, through the at least one weighing device 15, a poultry weight data from the plurality of poultry;collect, by activating the microphone, an environmental sound data and / or a poultry sound data; andcollect, by activating the environmental sensor, a temperature data and a humidity data of the poultry house.

[0070] Wherein weight data of at least three scenarios: when there is poultry on the weighing device 15, when the poultry is jumping onto the weighing device 15, and when there is no poultry on the weighing device 15.

[0071] Given all the above, the IMU device 13 is used for monitoring the camera angle. As such, if the camera angle is in need of adjustment, the first electronic device 11 correspondingly generates and transmits a notification signal to an electronic device 3 owned by a poultry farmer, thereby telling the poultry farmer to conduct a camera adjustment to the camera manually. In any one practicable embodiment, theelectronic device 3 can be, but may not be limited to smart phone, tablet computer, laptop computer, and desktop computer.

[0072] In a practicable embodiment, the camera 12 as shown in FIG. 1A and FIG. IB can be an electronic pan- tilt- zoom (ePTZ) camera including IR LED module (i.e., the aforesaid LED illumination module). Accordingly, the first processor IIP is further configured to enable the first electronic device 11 to perform the function of camera adjustment. For example, if a FOV (field of view) of the camera 12 fails to fully cover the whole internal of the poultry house, the first electronic device 11 generates and transmits a notification signal to the electronic device 3. In such case, the poultry farmer is allowed to, by operating the user operation interface that is shown by the display module of the electronic device 3, conduct a camera adjustment to the camera 12. In any one practicable embodiment, the camera adjustment can be, but may not be limited to LED light (illumination) adjustment, camera angle adjustment (by adjusting the electronic pan-tilt-zoom mechanism), camera sampling rate adjustment, and basic camera parameters (e.g., focus, brightness, and contrast) adjustment. For example, it is able to improve the quality of the captured poultry in a low light condition or a condition of absence of light. In addition, the camera adjustment can optional be conducting noise removing under different light source, conducting image trimming to focus on key areas and reduce file size, and conducting image compression to reduce file size.

[0073] In another one practicable embodiment, the at least one first electronic device 11 is connected to a suspended rod through a ball mount capable of rotating in any direction. To be specific, the ball mount allows multiple axis of rotation, and can be used to minimize wobbling from poultry jumping and wind turbulence.

[0074] Moreover, in first embodiment the ToF device 14 is adopted for sensing a height data of the machine vision device. In a preferable embodiment, the firstelectronic device 11 and a ground in the poultry house are spaced apart by a distance in a range between 100cm and 400cm. For example, the first electronic device 11 including two cameras 12 and the ground in the poultry house are spaced apart by 2-3.5m. Therefore, such particular arrangements bring benefits to the system 1, including:(1) having a wider field of view (FOV) for increasing the number of effective poultry samples (i.e., covering more poultry in the poultry house);(2) minimizing dust accumulation by being close to the ceiling, thereby making the poultry farmer be not needing to clean as much as well as the camera 12C acquire clearer picture overall; and(3) allowing the cameras 12 to remain in place during house cleaning because, being closer to the ceiling, they will not obstruct cleaning equipment or activities.

[0075] Besides, since the first processor 1 IP is able to count a feed consumption and / or a drinking water consumption by processing the poultry feed weight data, the first processor IIP can further determine whether the automation feeding apparatus is in need of feed adding or in need of water adding. Moreover, the first processor IIP is further configured to, in case of the automation feeding apparatus is in need of feed adding or in need of water adding, a corresponding notification signal to the electronic device 3 owned by the poultry farmer.

[0076] In first embodiment, the first electronic device 11 stores the poultry images and the monitoring data includes but may not be limited to the poultry number, the feed consumption, the drinking water consumption, the poultry weight data, the sound data, the temperature data, the humidity data, and the inertial data with time stamps in the first storage module 1 IM.

[0077] In a practicable embodiment the first processor IIP is periodically upload the monitoring data to a second electronic device like a remote server or a cloudcomputer (cloud storage device). During the data transmission, there is an option to changing the sampling period in the case that the Internet speed is too slow; for example, changing the sampling period from 10 seconds to 1 minute. Besides, if the first electronic device 11 detects that the internet is down, it will pass the data to on- device storage. Furthermore, in the case that the backlog in the on-device storage is too large and takes too long to upload, there is an option to selectively upload key portions of the data (for example, upload 1 set of data in every 6 sets of data).

[0078] Particularly, the application program comprises a pre-trained poultry growth prediction model, a pre-trained poultry weight prediction model, a pre-trained image segmentation model, a pre-trained feature extraction model and a pre-trained poultry health prediction model. By such arrangements, the first processor IIP is allowed to input the feed consumption, the drinking water consumption, and the poultry weight data into the pre-trained poultry weight prediction model, the pretrained poultry growth prediction model, the pre-trained poultry health prediction model, such that the pre-trained poultry weight prediction model outputs a poultry weight prediction data. In addition, the first processor IIP is also allowed to input the poultry images, the feed consumption, the drinking water consumption, and the poultry weight data into the pre-trained poultry growth prediction model and the pretrained poultry health prediction model, such that the pre-trained poultry growth prediction model outputs a poultry growth prediction data, and the pre-trained poultry health prediction model outputs a poultry health prediction data.

[0079] In addition, the application program further comprises a pre-trained feeding management policy optimizing model. As such, the first processor 1 IP is allowed to input the feed consumption, the drinking water consumption, and the poultry weight data into the pre-trained feeding management policy optimizing model, such that the pre-trained feeding management policy optimizing model outputs an optimizedfeeding management policy. Therefore, the optimized feeding management policy is subsequently transmitted to the electronic device 3, thereby showing to the poultry farmer.

[0080] It is needed to further explain that, when the system 1 is applied in a new poultry house, additional environment parameters including new pullet source, bird genetic variation, and new feed are collected, so as to be used in the adjustment of the algorithm (i.e., application program) of the first electronic device 11. In this case, the extracted features will also be used to add / adjust to the algorithm. Further, in a particular case, a suitable algorithm will be created based on the for the new environment parameters and / or the extracted features. Briefly speaking, the monitoring data can be used in the algorithm to predict weight data on the pure computer vision model, and this algorithm can be constantly calibrated by the data collecting device.

[0081] Second embodiment (edge computing)

[0082] With reference to FIG. 2, there is shown a third schematic stereo diagram the intelligent poultry farm monitoring system according to the present invention. In second embodiment, the intelligent poultry farm monitoring system 1 comprises at least one first electronic device 11 integrated with at least one camera 12, a first processor 1 IP, a first storage module 1 IM storing a first application program, at least one IMU device 13, and at least one ToF device 14, and at least one PoE splitter 16 therein, at least one weighing device 15, at least one PoE injector 17, at least one microphone (i.e., sound sensor), at least one environmental sensor (i.e., sensor for monitoring temperature and humidity), and a second electronic device 10 comprising a second processor 10P and a second storage module 10M storing a second application program. Briefly speaking, in second embodiment the first electronicdevice 11 is a monitoring data collecting device. On the contrary, the second electronic device 10 is in communication with the at least one first electronic device 11, and functions as a local computing device.

[0083] For example, according to FIG. 2 there are two cameras 12 integrated in the first electronic device 11. To be more specific, the PoE injector 17 is in communication with the PoE splitter 16 through an Ethernet cable, and is linked to Internet through an Ethernet device (e.g., router or Ethernet switch). It should be known that the PoE injector 17 includes an AC / DC conversion circuit and / or at least one DC / DC conversion circuit. In practical use, the PoE injector 17 is electrical connected to an AC power source via a power cable, so as to convert an AC power received from the AC power source to a DC power, and then combine an Ethernet signal and the DC power to a specific Ethernet signal. In contrast to the PoE injector 17, the PoE splitter 16 is used for separating the DC power from the specific Ethernet signal, thereby supplying the DC power to the first electronic device 11 and the camera 12. Moreover, the PoE splitter 16 also separates the Ethernet signal from the specific Ethernet signal, so as to transmit the Ethernet signal to the first electronic device 11.

[0084] In second embodiment, the IMU device 13 comprises accelerometer and gyroscope, and is adopted for collecting an inertial data of the machine vision device. On the other hand, the ToF device 14 is adopted for collecting a height data of the monitoring data collecting device (i.e., the first electronic device).

[0085] In addition, FIG. 2 depicts that the weighing device 15 is coupled to the first electronic device 11, and consists of a load cell 151 and a digital meter 152, of which the load cell 151 is connected to a weighing platform (not shown). In practical use, the load cell 151 generates a signal in response to the body weight of a poultry (like broiler, pullets, breeders, hens, egg-laying duck, or meat duck) stays in theweighing platform. Moreover, after receiving the signal from the load cell 151, the digital meter 152 amplifies the signal, and then extracts a weight data from the amplified signal.

[0086] In any possible embodiment, the first storage module 11M and the second storage module 10M all can be, but may not be limited to hard disk drive (HDD), solid state drive (SSD) or flash memory chip, wherein the first storage module 1 IM stores a first application program including instructions for configuring the first processor IIP, and the second storage module 10M stores a second application program including instructions for configuring the second processor 10P. Moreover, In any practicable embodiments, the first application program and the second application program can be edited by any possible programming language such as C, python, and Nodejs.

[0087] As such, after the first electronic device 11 is booted, the first processor 11 P executes the first application program so as to be configured to communicate and / or control the camera 12, the IMU device 13, the ToF device 14, the weighing device 15, the microphone, and the environmental sensor, Moreover, after the second electronic device 10 is booted, the second processor 10P executes the second application program so as to be configured to receive monitoring data from the first electronic device 11, and then performs functions of and data analyzing and processing.

[0088] In a word, the system 1 according to the present invention is configured to perform a plurality of specific functions, including: real-time image capturing, realtime data collecting and data analyzing and processing. To be more specific, while performing the functions of real-time image capturing and real-time data collecting, the first processor 1 IP is configured to:acquire, by controlling the at least one camera 12, a plurality of poultry images from a plurality of poultry farmed in a poultry house;collect, by activating the at least one IMU device 13, an inertial data of the first electronic device 11 ;collect, by activating the at least one ToF device 14, a height data of the first electronic device 11 ;collect, through the at least one weighing device 15, a poultry feed weight data from at least one automation feeding apparatus that is disposed in the poultry house;collect, through the at least one weighing device 15, a poultry weight data from the plurality of poultry;collect, by activating the microphone, an environmental sound data and / or a poultry sound data; andcollect, by activating the environmental sensor, a temperature data and a humidity data of the poultry house.

[0089] Correspondingly, while performing the function of data analyzing and processing, the second processor 10P is configured to:evaluate, by processing the plurality of poultry images, at least one poultry physical condition of each of the plurality of poultry;count, by processing the plurality of poultry images, a poultry number of the plurality of poultry;determine, whether at least one of the plurality of poultry is standing, sitting down or lying down according to the poultry physical condition;count, by processing the poultry feed weight data, a feed consumption and / or a drinking water consumption;receive, an inertial data from the IMU devicel3;calculate, by processing the inertial data, a camera angle of the camera 12; andcount, by processing the poultry feed weight data, a feed consumption and / or a drinking water consumption.

[0090] In practical use, the at least one physical condition includes but may not limited to, body size, evaluated weight, and poultry activity condition, and the second processor 10P is able to determine whether each of the plurality of poultry is standing, sitting down or lying down according to the poultry physical condition. Therefore, if the poultry is found to sit or lie for a period of time, the second electronic device 10 generates and transmits a notification signal to an electronic device 3 owned by a poultry farmer.

[0091] To be more specific, after processing the poultry images, relative features are extracted from the poultry image, wherein the features include but may not be limited to, image area, width, length, segment perimeter length, back width, back height, surface area, volume, eccentricity, segment radius, and max radius (when poultry is flapping). After that, the second processor 10P evaluates at least one poultry physical condition of the poultry by inputting the features into at least one pre-trained Al model. For example, one specific Al model is executed by the second processor 10P to estimate a weight value of the poultry based on the body shape and / or the features. On the other hand, another one specific Al model is executed to estimate the poultry activity condition based on the body shape and / or the features, in which the poultry activity condition includes abnormal activity due to illness or hurt.

[0092] It is worth further explaining that, if the poultry number reveals that there is at least one of the plurality of poultry missing, the second electronic device 10 correspondingly generates and transmits an alarm signal or a notification signal to the electronic device 3 owned by the poultry farmer.

[0093] In a practicable embodiment, the second application program stall in the second electronic device 10 (i.e., the local computing device) further comprises a pre-trained background subtraction algorithm that is utilized for subtracting a background including poultry dung from each of the plurality of poultry images. Moreover, in another one practicable embodiment, the application program further comprises a pre-trained image segmentation and labelling model that is utilized for firstly segmenting each of the plurality of poultry images while the poultry number counting, thereby generating a plurality of segmented image and then labelling each of the plurality of segmented image. As such, it is more easily to obtain each poultry’ s features through feature extraction process.

[0094] It needs to further explain that, if there is the poultry weight data from the weighing device 15, the poultry weight data will also be included in the labelling. Furthermore, the application program is further set to include a pre-trained object recognition model, and the pre-trained object recognition model will automatically select object (i.e., poultry) from the inputted poultry image.

[0095] Moreover, the first electronic device 11 is further integrated with an LED illumination module, such that an illumination light emitted by the LED illumination module can be directed to the environment of the poultry house, thereby improving the quality of the captured poultry images.

[0096] In addition, the PoE injector 17 is linked to Internet through an Ethernet like router or Ethernet switch, such that the first electronic device 11 is in communication with the second electronic device 10 via Internet or local LAN. Therefore, the first electronic device 11 is configured to transmit, through the PoE splitter and the PoE injector, a monitoring data comprising the plurality of poultry images, the poultry weight data, and the environmental data to the second electronic device 10.

[0097] The electronic device 3, such as a smartphone, a tablet computer, a laptop computer, a desktop computer, or an all-in-one computer, comprises a processor, a display module and a storage module storing a user-end application program, of which the processor executes the user-end application program so as to be configured to control the display module to:show, by controlling the display module, a user operation interface; show, through the user operation interface, a dashboard comprising a plurality of statistic graphs and data charts; andshow, a real-time monitoring images of the plurality of poultry farmed in the poultry house.

[0098] As such, the poultry farmer is able to see, by operating the user operation interface, a real-time monitoring images of the poultry from the display module of the electronic device 3, such that the poultry farmer is able to predominate a realtime activity condition of each poultry. Moreover, by operating the user operation interface shown on the screen of the display module of the electronic device 3 like smartphone, the dashboard is accessible via an app, a webapp and a web browser selected from a group consisting of Chrome, Safari, Firefox, Edge, wherein the dashboard exhibits a plurality of statistic graphs and data charts. For example, the dashboard includes statistic graph or data chart of weight data and statistic graph or data chart of body growth data, in which the statistic graph of weight data includes a first trend line (i.e., weight growth prediction line), and the statistic graph of body growth data includes a second trend line (i.e., body growth prediction line).

[0099] Given all the above, the IMU device 13 is used for monitoring the camera angle. As such, if the camera angle is in need of adjustment, the second electronic device 10 correspondingly generates and transmits a notification signal to an electronic device 3 owned by a poultry farmer, thereby telling the poultry farmer toconduct a camera adjustment to the camera manually. In any one practicable embodiment, the electronic device 3 can be, but may not be limited to smart phone, tablet computer, laptop computer, and desktop computer.

[0100] In a practicable embodiment, the camera 12 as shown in FIG. 2 can be an electronic pan-tilt-zoom (ePTZ) camera including IR LED module (i.e., the aforesaid LED illumination module). Accordingly, the first processor 1 IP is further configured to enable the first electronic device 11 to perform the function of camera adjustment. For example, if a FOV (field of view) of the camera 12 fails to fully cover the whole internal of the poultry house, the second electronic device 10 generates and transmits a notification signal to the electronic device 3. In such case, the poultry farmer is allowed to, by operating the user operation interface that is shown by the display module of the electronic device 3, conduct a camera adjustment to the camera 12 by remotely controlling the first electronic device 11. In any one practicable embodiment, the camera adjustment can be, but may not be limited to LED light (illumination) adjustment, camera angle adjustment (by adjusting the electronic pan-tilt-zoom mechanism), camera sampling rate adjustment, and basic camera parameters (e.g., focus, brightness, and contrast) adjustment. For example, it is able to improve the quality of the captured poultry in a low light condition or a condition of absence of light. In addition, the camera adjustment can optional be conducting noise removing under different light source, conducting image trimming to focus on key areas and reduce file size, and conducting image compression to reduce file size.

[0101] In another one practicable embodiment, the at least one first electronic device 11 is connected to a suspended rod through a ball mount capable of rotating in any direction. To be specific, the ball mount allows multiple axis of rotation, and can be used to minimize wobbling from poultry jumping and wind turbulence.

[0102] Moreover, in second embodiment the ToF device 14 is adopted for sensing a height data of the first electronic device 11. In a preferable embodiment, the first electronic device 11 and a ground in the poultry house are spaced apart by a distance in a range between 100cm and 400cm. For example, the first electronic device 11 including two cameras 12 and the ground in the poultry house are spaced apart by 2-3.5m.

[0103] Besides, since the second processor 10P is able to count a feed consumption and / or a drinking water consumption by processing the poultry feed weight data, the second processor 10P can further determine whether the automation feeding apparatus is in need of feed adding or in need of water adding. Moreover, the second processor 10P is further configured to, in case of the automation feeding apparatus is in need of feed adding or in need of water adding, a corresponding notification signal to the electronic device 3 owned by the poultry farmer.

[0104] In second embodiment, the first electronic device 11 temporarily stores the poultry images and the monitoring data with time stamps in the first storage module 11M, and periodically transmits the poultry images and the monitoring data to the second electronic device 10. For this reason, the second processor 10P is further configured to, in case of a sudden network dropping, generate and emit an alarm signal to the electronic device 3. During the data transmission, there is an option to changing the sampling period in the case that the Internet speed is too slow; for example, changing the sampling period from 10 seconds to 1 minute. Besides, if the first electronic device 11 detects that the internet is down, it will pass the data to on-device storage. Furthermore, in the case that the backlog in the on-device storage is too large and takes too long to upload, there is an option to selectively upload key portions of the data (for example, upload 1 set of data in every 6 sets of data).

[0105] Moreover, the second processor 10P can be further configured to periodically upload the poultry images and the monitoring data to a could electronic device like a remote server or a cloud computer (cloud storage device).

[0106] Particularly, the second application program comprises a pre-trained poultry growth prediction model, a pre-trained poultry weight prediction model and a pre-trained poultry health prediction model. By such arrangements, the second processor 10P is allowed to input the feed consumption, the drinking water consumption, and the poultry weight data into the pre-trained poultry weight prediction model, such that the pre-trained poultry weight prediction model outputs a poultry weight prediction data. In addition, the second processor 10P is also allowed to input the poultry images, the feed consumption, the drinking water consumption, and the poultry weight data into the pre-trained poultry growth prediction model and the pre-trained poultry health prediction model, such that the pre-trained poultry growth prediction model outputs a poultry growth prediction data, and the pre- trained poultry health prediction model outputs a poultry health prediction data.

[0107] In addition, the second application program further comprises a pre-trained feeding management policy optimizing model. As such, the second processor 10P is allowed to input the feed consumption, the drinking water consumption, and the poultry weight data into the pre-trained feeding management policy optimizing model, such that the pre-trained feeding management policy optimizing model outputs an optimized feeding management policy. Therefore, the optimized feeding management policy is subsequently transmitted to the electronic device 3, thereby showing to the poultry farmer.

[0108] It is needed to further explain that, when the system 1 is applied in a new poultry house, additional environment parameters including new pullet source, birdgenetic variation, and new feed are collected, so as to be used in the adjustment of the algorithm (i.e., application program) of the second electronic device 10. In this case, the extracted features will also be used to add / adjust to the algorithm. Further, in a particular case, a suitable algorithm will be created based on the for the new environment parameters and / or the extracted features. Briefly speaking, the monitoring data can be used in the algorithm to predict weight data on the pure computer vision model, and this algorithm can be constantly calibrated by the data collecting device.

[0109] Third embodiment (Cloud computing)

[0110] With reference to FIG. 3, there is shown a fourth schematic stereo diagram of the intelligent poultry farm monitoring system according to the present invention. In third embodiment, the intelligent poultry farm monitoring system 1 comprises at least one first electronic device 11 integrated with at least one camera 12, a first processor 1 IP, a first storage module 1 IM storing a first application program, at least one IMU device 13, and at least one ToF device 14, and at least one PoE splitter 16 therein, at least one weighing device 15, at least one PoE injector 17, at least one microphone (i.e., sound sensor), at least one environmental sensor (i.e., sensor for monitoring temperature and humidity), and a cloud computing device 1 A comprising a third processor 1AP and a third storage module 1AM storing a third application program. Briefly speaking, in third embodiment the first electronic device 11 is a monitoring data collecting device. On the contrary, the cloud computing device 1A is in communication with the at least one first electronic device 11 , and is configured for conducting cloud data processing and / or analyzing.

[0111] For example, according to FIG. 3 there are two cameras 12 integrated in the first electronic device 11. To be more specific, the PoE injector 17 is incommunication with the PoE splitter 16 through an Ethernet cable, and is linked to Internet through an Ethernet device (e.g., router or Ethernet switch). It should be known that the PoE injector 17 includes an AC / DC conversion circuit and / or at least one DC / DC conversion circuit. In practical use, the PoE injector 17 is electrical connected to an AC power source via a power cable, so as to convert an AC power received from the AC power source to a DC power, and then combine an Ethernet signal and the DC power to a specific Ethernet signal. In contrast to the PoE injector 17, the PoE splitter 16 is used for separating the DC power from the specific Ethernet signal, thereby supplying the DC power to the first electronic device 11 and the camera 12. Moreover, the PoE splitter 16 also separates the Ethernet signal from the specific Ethernet signal, so as to transmit the Ethernet signal to the first electronic device 11.

[0112] In third embodiment, the IMU device 13 comprises accelerometer and gyroscope, and is adopted for collecting an inertial data of the machine vision device. On the other hand, the ToF device 14 is adopted for collecting a height data of the monitoring data collecting device (i.e., the first electronic device).

[0113] In addition, FIG. 3 depicts that the weighing device 15 is coupled to the first electronic device 11, and consists of a load cell 151 and a digital meter 152, of which the load cell 151 is connected to a weighing platform (not shown). In practical use, the load cell 151 generates a signal in response to the body weight of a poultry (like broiler, pullets, breeders, hens, egg-laying duck, or meat duck) stays in the weighing platform. Moreover, after receiving the signal from the load cell 151, the digital meter 152 amplifies the signal, and then extracts a weight data from the amplified signal.

[0114] In any possible embodiment, the first storage module 11M and the third storage module 1AM all can be, but may not be limited to hard disk drive (HDD),solid state drive (SSD) or flash memory chip, wherein the first storage module 1 IM stores a first application program including instructions for configuring the first processor HP, and the third storage module 1AM stores a second application program including instructions for configuring the third processor 1AP. Moreover, In any practicable embodiments, the first application program and the second application program can be edited by any possible programming language such as C, python, and Nodejs.

[0115] As such, after the first electronic device 11 is booted, the first processor 1 IP executes the first application program so as to be configured to communicate and / or control the camera 12, the IMU device 13, the ToF device 14, the weighing device 15, the microphone, and the environmental sensor, Moreover, after the cloud computing device 1A is booted, the third processor 1AP executes the third application program so as to be configured to receive monitoring data from the first electronic device 11, and then performs functions of and data analyzing and processing.

[0116] In a word, the system 1 according to the present invention is configured to perform a plurality of specific functions, including; real-time image capturing, realtime data collecting and data analyzing and processing. To be more specific, while performing the functions of real-time image capturing and real-time data collecting, the first processor 1 IP is configured to:acquire, by controlling the at least one camera 12, a plurality of poultry images from a plurality of poultry farmed in a poultry house;collect, by activating the at least one IMU device 13, an inertial data of the first electronic device 11 ;collect, by activating the at least one ToF device 14, a height data of the first electronic device 11 ;collect, through the at least one weighing device 15, a poultry feed weight data from at least one automation feeding apparatus that is disposed in the poultry house;collect, through the at least one weighing device 15, a poultry weight data from the plurality of poultry;collect, by activating the microphone, an environmental sound data and / or a poultry sound data; andcollect, by activating the environmental sensor, a temperature data and a humidity data of the poultry house.

[0117] Correspondingly, while performing the function of data analyzing and processing, the third processor 1AP is configured to:evaluate, by processing the plurality of poultry images, at least one poultry physical condition of each of the plurality of poultry;count, by processing the plurality of poultry images, a poultry number of the plurality of poultry;determine, whether at least one of the plurality of poultry is standing, sitting down or lying down according to the poultry physical condition;count, by processing the poultry feed weight data, a feed consumption and / or a drinking water consumption;receive, an inertial data from the IMU devicel3;calculate, by processing the inertial data, a camera angle of the camera 12; and count, by processing the poultry feed weight data, a feed consumption and / or a drinking water consumption.

[0118] Wherein weight data of at least three scenarios: when there is poultry on the weighing device 15, when the poultry is jumping onto the weighing device 15, and when there is no poultry on the weighing device 15.

[0119] In practical use, the at least one physical condition includes but may not limited to, body size, evaluated weight, and poultry activity condition, and the second processor 10P is able to determine whether each of the plurality of poultry is standing, sitting down or lying down according to the poultry physical condition. Therefore, if the poultry is found to sit or lie for a period of time, the cloud computing device 1A generates and transmits a notification signal to an electronic device 3 owned by a poultry farmer.

[0120] To be more specific, after processing the poultry images, relative features are extracted from the poultry image, wherein the features include but may not be limited to, image area, width, length, segment perimeter length, back width, back height, surface area, volume, eccentricity, segment radius, and max radius (when poultry is flapping). After that, the third processor 1AP evaluates at least one poultry physical condition of the poultry by inputting the features into at least one pre-trained Al model. For example, one specific Al model is executed by the third processor 1AP to estimate a weight value of the poultry based on the body shape and / or the features. On the other hand, another one specific Al model is executed to estimate the poultry activity condition based on the body shape and / or the features, in which the poultry activity condition includes abnormal activity due to illness, discomfort, or injuries.

[0121] It is worth further explaining that, if the poultry number reveals that there is at least one of the plurality of poultry missing, the cloud computing device 1A correspondingly generates and transmits an alarm signal or a notification signal to the electronic device 3 owned by the poultry farmer.

[0122] In a practicable embodiment, the third application program stall in the cloud computing device 1A further comprises a pre-trained background subtraction algorithm that is utilized for subtracting a background including poultry dung fromeach of the plurality of poultry images. Moreover, in another one practicable embodiment, the application program further comprises a pre-trained image segmentation and labelling model that is utilized for firstly segmenting each of the plurality of poultry images while the poultry number counting, thereby generating a plurality of segmented image and then labelling each of the plurality of segmented image. As such, it is more easily to obtain each poultry’s features through feature extraction process.

[0123] It needs to further explain that, if there is the poultry weight data from the weighing device 15, the poultry weight data will also be included in the labelling. Furthermore, the application program is further set to include a pre-trained object recognition model, and the pre-trained object recognition model will automatically select object (i.e., poultry) from the inputted poultry image.

[0124] Moreover, the first electronic device 11 is further integrated with an LED illumination module, such that an illumination light emitted by the LED illumination module can be directed to the environment of the poultry house, thereby improving the quality of the captured poultry images.

[0125] In addition, the PoE injector 17 is linked to Internet through an Ethernet like router or Ethernet switch, such that the first electronic device 11 is in communication with the cloud computing device 1 A via Internet. Therefore, the first electronic device 11 is configured to transmit, through the PoE splitter and the PoE injector, a monitoring data comprising the plurality of poultry images, the poultry weight data, and the environmental data to the cloud computing device 1A.

[0126] The electronic device 3, such as a smartphone, a tablet computer, a laptop computer, a desktop computer, or an all-in-one computer, comprises a processor, a display module and a storage module storing a user-end application program, ofwhich the processor executes the user-end application program so as to be configured to control the display module to:show, by controlling the display module, a user operation interface; show, through the user operation interface, a dashboard comprising a plurality of statistic graphs and data charts; andshow, a real-time monitoring images of the plurality of poultry farmed in the poultry house.

[0127] As such, the poultry farmer is able to see, by operating the user operation interface, a real-time monitoring images of the poultry from the display module of the electronic device 3, such that the poultry farmer is able to predominate a realtime activity condition of each poultry. Moreover, by operating the user operation interface shown on the screen of the display module of the electronic device 3 like smartphone, the dashboard is accessible via an app, a webapp, and a web browser selected from a group consisting of Chrome, Safari, Firefox, Edge, wherein the dashboard exhibits a plurality of statistic graphs and data charts. For example, the dashboard includes statistic graph or data chart of weight data and statistic graph or data chart of body growth data, in which the statistic graph of weight data includes a first trend line (i.e., weight growth prediction line), and the statistic graph of body growth data includes a second trend line (i.e., body growth prediction line).

[0128] Given all the above, the IMU device 13 is used for monitoring the camera angle. As such, if the camera angle is in need of adjustment, the cloud computing device 1A correspondingly generates and transmits a notification signal to an electronic device 3 owned by a poultry farmer, thereby telling the poultry farmer to conduct a camera adjustment to the camera manually. In any one practicable embodiment, the electronic device 3 can be, but may not be limited to smart phone, tablet computer, laptop computer, and desktop computer.

[0129] In a practicable embodiment, the camera 12 as shown in FIG. 3 can be an electronic pan-tilt-zoom (ePTZ) camera including IR LED module (i.e., the aforesaid LED illumination module). Accordingly, the first processor 1 IP is further configured to enable the first electronic device 11 to perform the function of camera adjustment. For example, if a FOV (field of view) of the camera 12 fails to fully cover the whole internal of the poultry house, the cloud computing device 1 A generates and transmits a notification signal to the electronic device 3. In such case, the poultry farmer is allowed to, by operating the user operation interface that is shown by the display module of the electronic device 3, conduct a camera adjustment to the camera 12 by remotely controlling the first electronic device 11. In any one practicable embodiment, the camera adjustment can be, but may not be limited to LED light (illumination) adjustment, camera angle adjustment (by adjusting the electronic pan-tilt-zoom mechanism), camera sampling rate adjustment, and basic camera parameters (e.g., focus, brightness, and contrast) adjustment. For example, it is able to improve the quality of the captured poultry in a low light condition or a condition of absence of light. In addition, the camera adjustment can optional be conducting noise removing under different light source, conducting image trimming to focus on key areas and reduce file size, and conducting image compression to reduce file size.

[0130] In another one practicable embodiment, the at least one first electronic device 11 is connected to a suspended rod through a ball mount capable of rotating in any direction. To be specific, the ball mount allows multiple axis of rotation, and can be used to minimize wobbling from poultry jumping and wind turbulence.

[0131] Moreover, in third embodiment the ToF device 14 is adopted for sensing a height data of the first electronic device 11. In a preferable embodiment, the first electronic device 11 and a ground in the poultry house are spaced apart by a distance in a range between 100cm and 400cm. For example, the first electronic device 11including two cameras 12 and the ground in the poultry house are spaced apart by 2-3.5m.

[0132] Besides, since the third processor 1AP is able to count a feed consumption and / or a drinking water consumption by processing the poultry feed weight data, the third processor 1 AP can further determine whether the automation feeding apparatus is in need of feed adding or in need of water adding. Moreover, the third processor 1AP is further configured to, in case of the automation feeding apparatus is in need of feed adding or in need of water adding, a corresponding notification signal to the electronic device 3 owned by the poultry farmer.

[0133] In second embodiment, the first electronic device 11 temporarily stores the poultry images and the monitoring data with time stamps in the first storage module 11M, and periodically transmits the poultry images and the monitoring data to the cloud computing device 1A. For this reason, the third processor 1AP is further configured to, in case of a sudden network dropping, generate and emit an alarm signal to the electronic device 3. During the data transmission, there is an option to changing the sampling period in the case that the Internet speed is too slow; for example, changing the sampling period from 10 seconds to 1 minute. Besides, if the first electronic device 11 detects that the internet is down, it will pass the data to on-device storage. Furthermore, in the case that the backlog in the on-device storage is too large and takes too long to upload, there is an option to selectively upload key portions of the data (for example, upload 1 set of data in every 6 sets of data).

[0134] Particularly, the third application program comprises a pre-trained poultry growth prediction model, a pre-trained poultry weight prediction model and a pretrained poultry health prediction model. By such arrangements, the third processor 1AP is allowed to input the feed consumption, the drinking water consumption, and the poultry weight data into the pre-trained poultry weight prediction model, suchthat the pre-trained poultry weight prediction model outputs a poultry weight prediction data. In addition, the third processor 1AP is also allowed to input the poultry images, the feed consumption, the drinking water consumption, and the poultry weight data into the pre-trained poultry growth prediction model and the pretrained poultry health prediction model, such that the pre-trained poultry growth prediction model outputs a poultry growth prediction data, and the pre-trained poultry health prediction model outputs a poultry health prediction data.

[0135] In addition, the third application program further comprises a pre-trained feeding management policy optimizing model. As such, the third processor 1AP is allowed to input the feed consumption, the drinking water consumption, and the poultry weight data into the pre-trained feeding management policy optimizing model, such that the pre-trained feeding management policy optimizing model outputs an optimized feeding management policy. Therefore, the optimized feeding management policy is subsequently transmitted to the electronic device 3, thereby showing to the poultry farmer.

[0136] It is needed to further explain that, when the system 1 is applied in a new poultry house, additional environment parameters including age, region breed, light cycle, vaccination information, location, new pullet source, pullet mix, hen age, probiotics, litter composition, local temperature, local humidity, ventilation cycle, ventilation speed, poultry integrator, bird genetic variation, and new feed are collected, so as to be used in the adjustment of the algorithm (i.e., application program) of the third electronic device 1A. In this case, the extracted features will also be used to add / adjust to the algorithm. Further, in a particular case, a suitable algorithm will be created based on the for the new environment parameters and / or the extracted features. Briefly speaking, the monitoring data can be used in thealgorithm to predict weight data on the pure computer vision model, and this algorithm can be constantly calibrated by the data collecting device.

[0137] In one embodiment, the system facilitates real-time algorithmic adjustment for multiple computer vision-only devices deployed within a single poultry house or across multiple poultry houses on a farm. Specifically, a data collection system is operable to interface with a plurality of computer vision-based camera systems, each configured to implement a common algorithm. The data collection system is further configured to continuously acquire real-world weighing data reflective of ground truth values, through the use of the weighing device 15.

[0138] This acquired data is processed to perform corrections to the algorithm, thereby accounting for localized variations within the environment. The correction can be small if the variation is small, and can be adjusted within a fixed time, such as one hour. This small model operations, where the number of parameters are not too high, can be suitable for slight adjustments in the poultry flock cycle, such as a change in feed. In the case of multiple adjustments, then the number of corrections, as well as the sample size from the data collector, will be larger.

[0139] The updated algorithm can then be propagated to the computer vision-based camera systems via edge computing or cloud-based infrastructure, ensuring that the updated weight prediction data is consistently applied across all camera systems.

[0140] By utilizing a centralized data collection system in conjunction with multiple computer vision-based devices, this implementation reduces the need for individual recalibration of each camera system while maintaining accuracy and scalability in predicting poultry weight across large-scale poultry operations.

[0141] Fourth embodiment

[0142] The environmental sensor comprises a temperature sensor, monitors temperature changes within the poultry house to assess thermal comfort of chickens and prevent heat stress; a humidity sensor, monitors humidity changes within thepoultry house to prevent respiratory diseases caused by excessive humidity; a light intensity sensor, monitors light intensity to simulate natural light, influencing the physiological cycle and egg production of chickens; a gas concentration sensor, monitors the concentration of harmful gases such as ammonia and carbon dioxide to ensure air quality.

[0143] The intelligent poultry farm monitoring system 1 also comprises a poultry activity sensor, and the poultry activity sensor further comprises an accelerometer, monitors the frequency and intensity of poultry movement to assess activity levels and reflect overall health; a gyroscope, monitors changes in poultry posture to determine if the bird is sick or injured; an image sensor, through visual analysis, monitors poultry behavior such as eating, drinking, and roosting. The image sensor could replace by the camera 12.

[0144] The camera 12 captures images of poultry for behavioral analysis, individual identification, and population counting. The time-of-Flight (ToF) device 14 measures the distance between the poultry and the sensor for spatial positioning and height measurement. The weight sensor measures feed and water consumption and the weighing device 15 measures the poultry weight. The processor processes data from various sensors, performing analysis, calculations, and control functions.

[0145] The system's functionalities cover multiple aspects to enhance the efficiency and accuracy of poultry management. Among them, poultry behavior analysis is a key module. Through image analysis technology, the unique characteristics of each chicken, such as feather color and patterns, can be identified to achieve individual recognition. Additionally, the system can classify chicken actions, including eating, drinking, resting, and activity, to assess their health status and productivity. Meanwhile, through abnormal behavior detection, the system canquickly identify abnormal situations such as frequent feather pecking or fighting, facilitating early intervention.

[0146] In terms of poultry health monitoring, the system can provide real-time assessments of health status through visual inspection functions, analyzing feather conditions, eye conditions, and weight changes. At the same time, abnormal behavior detection can quickly identify symptoms such as decreased appetite or reduced activity, helping to identify potential disease risks early.

[0147] Environmental monitoring focuses on providing suitable growth conditions. By continuously monitoring temperature and humidity, the system ensures a stable environment in the poultry house. Simultaneously, the system can detect the concentration of harmful gases in the air, such as ammonia and carbon dioxide, to maintain good air quality and protect the health of poultry.

[0148] In terms of feeding management, the system integrates the monitoring and control functions of feed and water, not only tracking consumption but also automatically adjusting supply to ensure the nutritional needs of poultry. This function, combined with automation technology, reduces manpower input and improves management efficiency.

[0149] In terms of technical implementation, the system utilizes image analysis technology, employing deep learning models such as convolutional neural networks (CNNs) and recurrent neural networks (RNNs) to achieve individual recognition and behavior classification. Sensor fusion technology integrates different data sources to improve monitoring accuracy. Machine learning algorithms (such as support vector machines and random forests) further enhance data analysis and prediction capabilities. Data processing techniques including data cleaning, feature engineering, and data mining effectively extract valuable information, and cloud platformssupport remote monitoring, data storage, and analysis, providing comprehensive system support.

[0150] Individual identification technology leverages image analysis and deep learning to achieve accurate recognition and tracking of poultry. Firstly, feature extraction is a core step. Through deep learning models such as convolutional neural networks (CNNs), unique features of poultry can be extracted from images, including feather patterns, beak shapes, and leg features. Furthermore, by combining keypoint detection techniques like OpenPose, the system can locate important body parts of the poultry, such as eyes, beaks, and feet, and establish a unique identification feature based on the relative positions and distances between these keypoints.

[0151] For individual tracking, Kalman filters are used to predict the possible location of the poultry in the next frame, and combined with object detection results for updates to maintain tracking continuity. Additionally, deep learning tracking techniques such as DeepSORT, by combining appearance features and motion information, can achieve accurate multi-object tracking, ensuring that the dynamic behaviors of each chicken can be accurately identified and tracked in complex scenes. These technologies collectively enhance the accuracy and stability of individual identification.

[0152] Deep learning serves as the core technology for feature extraction in image analysis, enabling high-precision poultry identification through multi-layer image processing and feature learning.

[0153] Firstly, data collection and annotation form the foundation. High-definition cameras are used to collect multi-angle images of poultry, ensuring coverage of various lighting conditions, backgrounds, and action scenarios. Subsequently, the data is manually annotated to mark the feature locations of the poultry, such asfeather patterns, beak shapes, and leg features, preparing a high-quality labeled dataset for training deep learning models.

[0154] Next, deep learning model design and training are conducted. Suitable deep learning architectures for image processing, such as ResNet or EfficientNet, are selected as the base models and pre-trained to accelerate convergence. In the feature extraction stage, convolutional layers are responsible for learning local features of poultry in images, such as textures and edges, while fully connected layers or embedding layers further abstract these features, transforming them into highdimensional feature vectors for distinguishing different individuals.

[0155] To enhance the generalization ability of the model, data augmentation techniques, including rotation, cropping, flipping, and lighting adjustment, are implemented to simulate diversity in real-world environments. Additionally, regularization techniques such as Dropout and L2 regularization are used to prevent overfitting.

[0156] Keypoint detection assists the feature extraction process. By integrating models like OpenPose, the locations of key body points such as the eyes, beak, and feet of the poultry are extracted. These keypoints can be used as inputs for geometric features, further constructing identification vectors with position and distance features to enhance individual discrimination.

[0157] Finally, model optimization and deployment are conducted. After model training, optimization algorithms such as Adam or SGD are used to adjust parameters to improve accuracy. Simultaneously, combining quantization techniques or edge computing, the model is deployed to edge devices or cloud platforms to enable realtime image processing and feature extraction, ensuring the system can operate stably in real-time scenarios.

[0158] The implementation of deep learning models focuses on efficiently extracting features from images. It begins with data preparation and annotation. High-definition cameras collect multi-angle, multi-scene images of poultry to ensure coverage of various lighting conditions, backgrounds, and poses. These image data undergo meticulous manual annotation, including feather patterns, beak shapes, and leg features, with each chicken assigned a unique label (ID). Such data annotation provides high-quality input for model training and lays the foundation for feature learning.

[0159] In the model design phase, deep learning architectures such as ResNet or EfficientNet are selected. These architectures, centered around multi-layer convolutional neural networks (CNNs), effectively extract low-level features (such as edges and textures) and high-level features (such as shapes and semantic information) from images. Convolutional layers are responsible for detecting local features, while pooling layers further compress the feature space to accelerate computation while preserving important information. The final fully connected layer transforms features into high-dimensional embedding vectors, which are used to distinguish individual chickens.

[0160] To further enhance the model's ability to distinguish between similar’ individuals, feature embedding techniques are implemented. Using contrastive learning methods (such as Siamese Networks or Triplet Loss), the model learns the differences between different individuals. Especially in cases where feather patterns are highly similar, these techniques can significantly improve model accuracy. Additionally, data augmentation techniques (such as rotation, flipping, and lighting adjustment) are used to simulate diversity in real- world scenarios, making the model more generalizable.

[0161] Model optimization and deployment are also key stages in the implementation process. During training, advanced optimization algorithms such as Adam or SGD are used to adjust parameters. After training, the model can be deployed to edge devices or cloud platforms to enable real-time image processing, providing stable and efficient support for chicken identification.

[0162] The implementation of keypoint detection focuses on accurately locating poultry body parts, which is crucial for identification and tracking. Initially, a high-quality dataset encompassing diverse scenarios is constructed, and manual annotations are made to mark the positions of key body parts such as eyes, beaks, and feet. This annotated data not only aids model learning but also ensures the accuracy of detection results.

[0163] In the implementation process, efficient keypoint detection models like OpenPose or MediaPipe are selected. These models generate "heatmaps" through multiple convolutional operations, where the highlighted areas in the heatmap correspond to the locations of keypoints. Before inputting the image into the model, it undergoes scaling and normalization to ensure consistent size and reduce the impact of lighting or contrast. Simultaneously, background removal techniques such as Mask R-CNN effectively eliminate interference from non-target objects in the image, focusing solely on the poultry itself.

[0164] After detection, geometric features are constructed using the positions and distances of keypoints. By calculating the relative distances between keypoints, such as the distance from the eye to the tip of the beak and the distance from the tip of the beak to the foot, geometric-based identification features are further established. In addition to distances, the angles between keypoints can also be calculated to capture the uniqueness of the poultry's body structure. These geometric features serve as identification vectors to assist in individual differentiation.

[0165] If the image contains multiple poultry, object detection techniques such as YOLO or Faster R-CNN must be combined for multi-object segmentation. The detection model first separates the regions of each poultry and then performs keypoint detection on the segmented image to ensure that the parts of each poultry can be independently identified. Ultimately, the combination of these techniques enables efficient and accurate poultry keypoint detection and identification, providing comprehensive individual tracking and management functions for farms.

[0166] Individual tracking technology aims to accurately and continuously identify and track specific target chickens in images, even in dynamic and complex environments. Core technologies include traditional Kalman filtering and deep learning-based tracking methods. The following details the implementation process of these methods.

[0167] Kalman Filtering Implementation

[0168] Kalman filtering is a linear prediction technique widely used in target tracking. Its core lies in modeling the historical movement trajectory of a target to predict its potential position in the next frame, and then updating the prediction based on actual detection results, achieving smooth and accurate tracking.

[0169] Firstly, a motion model of the chicken is established to describe its position (e.g., coordinates) and velocity in the image. The motion model typically adopts a simple linear motion assumption, i.e., the chicken's velocity does not change drastically in a short period. The state vector of the model includes position information (such as [x,y] coordinates) and velocity (such as [Xx, Yy]).

[0170] In each frame of the image, the Kalman filter is divided into two steps: prediction and update. In the prediction step, the expected position of the target in the current frame is calculated based on the state vector of the previous frame and the motion model. In the update step, the actual position observation provided by thetarget detector is combined through weighted calculation to update the predicted value and the observed value, reducing noise interference and improving accuracy.

[0171] Furthermore, when there are multiple chickens in the image, the Hungarian algorithm is combined for multi-target assignment. This algorithm calculates the distance between the predicted position and the detection result, matching the detection result with the corresponding tracking object to solve the multi-target tracking problem.

[0172] Deep Learning Tracking Implementation

[0173] Deep learning-based tracking methods, such as DeepSORT, achieve accurate multi-object tracking by combining appearance features and motion information, especially suitable for scenarios with similar-looking chickens and complex backgrounds.

[0174] The implementation process first requires an efficient object detection model (such as YOLO or Faster R-CNN) to detect chicken regions in each frame of the image, outputting bounding boxes. Next, appearance features are extracted for each target within the bounding box. This part usually uses a pre-trained deep learning model (such as ResNet) to convert the image into a high-dimensional feature vector, describing the unique appearance information of the chicken, such as feather texture and color patterns.

[0175] In the tracking stage, DeepSORT combines the motion prediction results of the Kalman filter with the appearance features extracted by deep learning, and uses the Hungarian algorithm to achieve target matching. This method not only considers the spatial position of the target (motion information) but also introduces appearance similarity as an auxiliary, enabling stable tracking even when the target is occluded or overlapped.

[0176] To improve the robustness of the tracking system, re-identification (RelD) technology can be introduced. When a chicken temporarily disappears (e.g., due to occlusion or moving out of the frame), the system can re-identify its identity based on appearance features, achieving long-term stable tracking.

[0177] Implementation of Activity Analysis

[0178] Activity analysis aims to comprehensively understand the daily activity status of poultry and accurately detect anomalies through posture estimation and behavior classification. This process combines advanced image processing and machine learning techniques to achieve dynamic monitoring and classification of poultry behavior.

[0179] Posture Estimation Implementation

[0180] Posture estimation is the foundation of activity analysis, primarily through models like OpenPose for detecting and locating the poultry's skeleton. These models, based on convolutional neural networks (CNNs), segment the poultry in the image and extract key body points such as the head, wingtips, and feet. By connecting these keypoints, a skeleton diagram can be generated to clearly display the poultry's posture.

[0181] In implementation, the posture estimation model is used to judge whether the poultry is standing, walking, squatting, or lying down. These posture types are trained using a pre-labeled dataset, where the model learns the feature patterns of each posture. When image data is input into the model, the system can immediately output the posture judgment result of the poultry.

[0182] Furthermore, the activity level of the poultry is calculated using the results of posture estimation. This process combines the frequency of posture changes and the distance of position movement. For example, frequent standing and walking postures indicate higher activity levels in poultry, while long periods of lying downmay indicate health problems. By accumulating the total daily activity, farm managers can quickly understand the overall vitality of the flock.

[0183] Behavior Classification Implementation

[0184] Behavior classification is built on the basis of posture estimation, identifying behavior patterns by analyzing the sequence of poultry movements. This process uses sequential models (such as LSTM or GRU) to handle the time dependence of a series of poultry movements, thereby judging whether they are foraging, drinking, or resting. Sequential models can capture the continuity between actions, providing more accurate behavior classification results than single-frame analysis.

[0185] In implementation, the system first extracts action sequences through object detection and posture estimation, and then inputs these sequences into a sequential model for classification. For example, foraging behavior is usually accompanied by frequent head-lowering and short-distance movement, while drinking involves a specific angle change of the head towards the water trough. The model automatically classifies the behavior into the corresponding category based on these features.

[0186] Additionally, anomaly behavior detection is a key part of behavior classification. The system learns the data pattern of normal behavior to establish a behavioral baseline for healthy poultry. When an abnormal pattern is detected, such as prolonged stillness, frequent feather pecking, or repetitive repetitive behavior, the system triggers an alert. This function is particularly important for early detection of diseases or stress responses, helping farms take timely intervention measures.

[0187] Implementation of Pose Estimation

[0188] Pose estimation for poultry involves analyzing images using deep learning techniques to extract key body points and construct a skeletal model, thereby determining the posture. The entire process includes data preparation, modelselection, training, inference, and result application. Here is a detailed description of the implementation.

[0189] Data Preparation and Annotation

[0190] The first step in implementing pose estimation is to prepare a large dataset of annotated images. High-definition cameras are used to capture multi-angle images of poultry under various lighting conditions and environments, ensuring coverage of multiple possible postures such as standing, squatting, walking, or lying down. Manual annotations are performed on the images to record the coordinates of each chicken's key points, including the head, wingtips, leg joints, and tail. These annotated data form the basis for training the model and are related to the accuracy and generalization ability of the final model.

[0191] To enhance the diversity of the dataset, data augmentation techniques can be employed, such as image flipping, rotation, contrast adjustment, and occlusion simulation, to simulate various situations that may occur in real-world scenarios.

[0192] Model Selection and Design

[0193] In terms of model selection for pose estimation, commonly used frameworks include OpenPose, DeepLabCut, or MediaPipe. These models are based on deep learning, using convolutional neural networks (CNNs) to extract features from images and generate "keypoint heatmaps" to locate the body parts of the poultry.

[0194] OpenPose is a popular multi-person pose estimation framework that can simultaneously detect multiple chickens and construct their skeletons. The model input is a standardized image, which, after multiple convolutional operations, outputs a heatmap for each keypoint. The highlighted areas in the heatmap correspond to the possible locations of the keypoints. In addition, OpenPose also uses part affinity fields (PAFs) to determine the connections between keypoints, thereby generating a complete skeletal structure.

[0195] Model Training and Optimization

[0196] In the model training phase, the annotated dataset is used for supervised learning. The model gradually adjusts its parameters to improve prediction accuracy by calculating the error between the predicted keypoint position and the annotated position (e.g., mean squared error, MSE).

[0197] During the training process, a multi-scale feature extraction strategy is used, which extracts image features at different resolutions to help the model better identify small targets (such as chicken leg joints). At the same time, regularization techniques such as Dropout are used to prevent overfitting and ensure that the model maintains stable performance in different environments.

[0198] Behavior Classification Implementation

[0199] Behavior classification for poultry focuses on analyzing sequences of poultry actions, combining image processing, sequential modeling, and anomaly detection techniques to comprehensively understand poultry's daily behavior patterns. The following is a detailed description of the implementation.

[0200] Data Preparation and Behavior Annotation

[0201] The first step in behavior classification is to establish an annotated dataset covering a variety of behaviors. By capturing poultry action images with a camera and labeling their behaviors based on expert knowledge as foraging, drinking, resting, etc., each image sequence should correspond to a behavior label, and the start and end times of the behavior should be recorded.

[0202] To improve the model's understanding of behavioral details, image data can be converted into action sequence data, and key features can be extracted, such as changes in poultry posture, position movement, and head direction. Combining image data from different scenarios (such as day and night changes, lighting differences) can improve the model's generalization ability.

[0203] Feature Extraction and Action Modeling

[0204] The key to behavior classification lies in feature extraction, extracting spatial and temporal features by processing image sequences. Spatial features describe the current state of the poultry, such as posture, position, and skeletal information; temporal features capture the continuous changes in actions, such as direction and speed of movement.

[0205] Deep learning models (such as 3D convolutional neural networks, 3D-CNN) can directly process image sequences to extract spatiotemporal features. Specifically, 3D-CNN uses three-dimensional convolution kernels to simultaneously capture the spatial patterns and temporal dependencies of images, modeling behavior. In addition, skeletal data generated based on pose estimation can be used as input for sequential models (such as LSTM or GRU), and the model can identify behavior types based on the dynamic changes of skeletal points.

[0206] Training and Optimization of Behavior Classification Models

[0207] The training of behavior classification models uses supervised learning methods. The dataset is divided into training sets, validation sets, and test sets for model training, hyperparameter tuning, and performance evaluation. The loss function (such as cross-entropy loss) is used to measure the accuracy of the model's predicted behavior category, and the model gradually improves its classification ability as training progresses.

[0208] To enhance the model's stability and generalization performance, data augmentation techniques (such as random cropping, rotation, or brightness adjustment) can be used to generate more diverse training samples. In addition, using the attention mechanism can improve the model's performance in long-sequence analysis and help the model focus on key action features.

[0209] Anomaly Behavior Detection

[0210] An extension of behavior classification is anomaly behavior detection, which establishes a baseline model of normal behavior by learning the daily behavior patterns of poultry. The anomaly detection system monitors real-time behavior data and compares it to the baseline model to identify behaviors that deviate from the normal range.

[0211] For example, abnormal behaviors such as prolonged stillness or frequent feather pecking can be detected by analyzing the dynamic changes in behavioral features. These features include low activity levels, missing key actions, or irregular behavior frequencies. To improve accuracy, anomaly detection models often combine unsupervised learning methods (such as autoencoders or isolation forest) to learn the implicit distribution of data and discover abnormal patterns.

[0212] Real-time Processing and Application

[0213] The inference of behavior classification models needs to run in real-time scenarios on farms to ensure that poultry behavior can be quickly identified. To achieve efficient processing, the model can be deployed to edge computing devices, combined with optimization algorithms (such as quantization techniques) to reduce the computational load.

[0214] Classification results can be used for farm management decisions, such as triggering alerts when abnormal behaviors are detected, or adjusting feed supply strategies based on foraging behavior data. At the same time, long-term behavior monitoring data can also be used for health analysis and poultry welfare improvement, providing data support for farming efficiency.

[0215] Through behavior classification technology, farms can accurately grasp the daily activities and health status of the flock, improve management efficiency and reduce risks.

[0216] Specifically, in order to more accurately conduct visual analysis, monitor chicken behavior, individual identification, individual tracking, behavior classification and appearance feature analysis, the present invention further includes the following camera structure.

[0217] As shown in FIG. 4, the camera 12 according to one embodiment of the present invention largely includes an upper base (30) and a shell (20) assembled to the upper base (30).

[0218] In the case of the upper base (30), it is a space where the camera 12 is attached and is usually formed in a disc shape with a certain thickness, but there is no particular limitation on the shape. However, since the shell (20) has a dome type shape, it is most preferable to form a disc shape that generally matches the above shape.

[0219] In the case of the shell (20), it is usually formed in a dome type that forms a certain space so that the camera 12 can be stored. The shell (20) should be formed in a transparent window form so that the camera 12 introduced inside can adjust the position to be monitored and accurately photographed. Therefore, in the case of the shell (20), a space (21) corresponding to a transparent window that can secure the field of view of the camera unit is provided.

[0220] It is desirable that the inner diameter of the shell (20) matches the outer diameter of the upper base (30) so that the fastening is completely accomplished. In addition, a rubber ring (5) may be additionally provided between the shell (20) and the upper base (30) to enable mutual compression fixation. The gap between the joining surfaces of the shell (20) and the upper base (30) can be more perfectly sealed through the rubber ring (5).

[0221] In general, the structural features of the camera 12 according to one embodiment of the present invention are that a plurality of bulge portion (210) areformed on the inner side surface of the shell (20), a plurality of track are formed on the outer side surface of the upper base (30), and when the shell (20) and the upper base (30) are horizontally rotated, the bulge portion (210) and the track are mutually connected and vertically fixed, and the complete connection is established by the up-and-down movement of the external locking device provided on the outer side of the shell (20) and the upper base (30).

[0222] As described above, in the case of the camera 12, a more perfect assembly structure is formed through double connection of the inner connection and the outer connection (the inner connection and the outer connection are performed simultaneously). Therefore, let's examine the technical features of the present invention in the order of inner and outer coupling through the additional drawings below.

[0223] 1. Hereinafter, the inner coupling will be described first.

[0224] FIGs 5 and 6 are internal cross-sectional views showing the appearance after the internal coupling of the shell (20) and the upper base (30) constituting the camera 12 according to one embodiment of the present invention is fastened.

[0225] As can be seen from FIGs. 5 and 6, in order to perfectly fasten the shell (20) and the upper base (30), bulge portion (210) and track corresponding to the technical features of the present invention are provided. The bulge portion (210) are provided in multiple numbers at a certain interval on the inner surface of the shell (20), and the track are provided in multiple numbers at a certain interval on the outer surface of the upper base (30). Since the camera 12 is fastened in a form in which the bulge portion (210) are fitted into the track , the interval between the bulge portion (210) and the interval between the track must be the same, and the bulge portion (210) and the track must be positioned in the shell (20) and the upper base (30) at mutually corresponding positions, respectively.

[0226] In the case of the camera 12, the shell (20) and the upper base (30) are combined, and the diameter of the inner surface of the shell (20) and the diameter of the outer surface of the upper base (30) are the same, so that the shell (20) is covered on the upper base (30). In the drawing, it appears that the assembly is performed by applying pressure from above to below, but the drawing is drawn in the correct direction to help the understanding of a general technician. When the dome type camera is actually installed, the upper base (30) is fixed to the ceiling or the side wall, and the shell (20) is installed so that it faces the ground or faces in a direction perpendicular to the ground.

[0227] The shell (20) of the camera 12 is equipped with a bulge portion (210). The bulge portion (210) is provided in the direction of the side wall of the shell (20) and can be formed in multiple pieces along the inner surface of the shell (20). The degree of freedom of the position of the bulge portion (210) provided in the shell (20) is guaranteed, but it is ideal to form an angle of 120° between them and to form three places (three pieces).

[0228] In response to this, the track provided in the upper base (30) of the camera 12 corresponds to a concavely dug structure. This is because the bulge portion (210) and the track are combined so that the shell (20) and the upper base (30) are mutually coupled and fixed.

[0229] Similarly, it can be formed in multiple pieces along the outer surface of the upper base (30). The degree of freedom of the track provided in the upper base (30) is guaranteed, but for perfect waterproofing and balanced assembly, it is ideal to form a 120° angle between the track so as to correspond to the shell (20) and to form them in three places (three pieces).

[0230] The entire joining process of the shell (20) and the upper base (30) is as follows. The first joining process is performed in the form that the shell (20) is puton the upper base (30). In this case, it is desirable to set the preposition so that the bulge portion (210) and the track face each other and then join them for the subsequent steps. After that, the assembler rotates the shell (20) left and right by a predetermined amount so that the bulge portion (210) is positioned in a position where it perfectly faces the track and is temporarily joined.

[0231] After that, when the assembler rotates the shell (20), the third joining process, i.e., the inner joining process, in which the bulge portion (210) moves along the path of the track that is inclined diagonally downward based on the horizontal plane, is carried out in earnest. In the case of the rubber ring (5), since it is implemented with a material that has elasticity or flexibility for waterproofing, when the shell (20) is rotated to a certain extent, the bulge portion (210) flexibly moves along the path of the track and is compressed.

[0232] As the bulge portion (210) is drawn into the diagonal track according to the horizontal rotational movement of the shell (20), it is tightened according to the horizontal and vertical movement of the bulge portion (210). Since the length of the track is not long in the present invention, the rotation distance of the shell (20) is also relatively short, so that joining is very easy. Finally, through rotation, the bulge portion (210) is settled in the cavity (311) formed at the end of the track path. In the case of the cavity (311), since it has a relatively concave shape compared to the path of the track , when the bulge portion (210) is settled in the cavity (311), it is fixed without movement and takes its place in the cavity (311). Accordingly, since the joint structure of the shell (20) and the upper base (30) does not loosen without artificial manipulation, that is, without external force, a stable assembled state is maintained.

[0233] 2. Hereinafter, the external coupling will be described.

[0234] FIGs. 7 and 8 are perspective views illustrating the process of fastening the external locking device of the camera 12 according to one embodiment of the present invention.

[0235] The shell (20) and the upper base (30) are each provided with a fastening confirmation track (50) that can check whether they are accurately fastened to each other. The shape of the fastening confirmation track (50) is not limited, but is generally provided in a semicircular shape on the lower side of the shell (20) and the upper outer side of the upper base (30), forming a single circle. Therefore, the user can check whether the assembly of the shell (20) and the upper base (30) is consistent through the shape confirmation as described above.

[0236] In addition, a indicator notch (60) that indicates the rotational coupling direction of the shell (20) may be additionally provided on the lower side of the shell (20). In general, the indicator notch (60) has a half-arrow shape. By rotating the shell (20) in the direction of the arrow and fastening it with the upper base (30), an effect of preventing incorrect assembly in advance occurs. Through the above process, a stable assembly structure is finally formed, and complete fastening and perfect waterproofing of the camera 12 are possible.

[0237] FIG. 9 is a perspective view showing the appearance after the shell (20) and upper base (30) constituting the camera 12 according to one embodiment of the present invention are fastened.

[0238] In summary, the camera 12 according to one embodiment of the present invention is simple and easy to assemble through a bulge portion-track fastening method and external locking without a separate tool (toolless) and without using a multiple bolt fastening method (boltless, bolt holeless).

[0239] On the contrary, when separating the shell (20) and upper base (30), the removal process is very easy in the reverse order of the above process. Since theremoval process includes the same technical features, a detailed description thereof will be omitted.

[0240] Accordingly, when checking the unique effect of the camera 12 including the technical features of the present invention, there is an advantage in that it can form a smooth and neat design and appearance due to the technical features of not using a multiple bolt fastening method (Boltless, Bolt holeless). And could be easily embedded in the first electronic device 11.

[0241] A vertical compression fixing structure is formed through a relatively easy horizontal combination called a rotary combination, and since the rotation distance is very short, it does not cause any burden to the user performing the assembly, the bulge portion (210) is stably positioned in the cavity (311) located at the end of the track , and due to the characteristics of the double combination structure in which the inner and outer combinations are performed simultaneously, a perfect waterproof state can be implemented in which foreign substances or rainwater do not enter between the shell (20) and the upper base (30).

[0242] Also, since it is easy to install in places where the working environment is relatively difficult, such as ceilings or walls, and the user repeats the process of attaching and detaching the shell and upper base several times and precisely adjusts the angle or direction of the internal camera unit to reduce shooting errors in the location to be monitored, the work time is significantly shortened compared to the conventional technology of attaching a number of bolts and nuts, so the stress and risk factors caused by the work of the worker can be significantly reduced. Since separate tools and additional assembly parts are not required, economic benefits can be realized in that the cost is reduced.

[0243] The camera according to the present invention has been described with reference to the embodiments shown in the drawings, but this is only an example,and anyone skilled in the art will understand that various modifications and equivalent other embodiments are possible from this. Therefore, the true scope of technical protection should be determined by the technical idea of the attached patent claims.

[0244] In some embodiments, the present invention further comprises the method or steps described below.

[0245] Step 1; Sensor Deployment and Installation

[0246] 1-1: Evenly distribute various environmental sensors (such as temperature, humidity, light intensity, gas concentration) and poultry activity sensors (such as accelerometers, gyroscopes, image sensors) throughout the chicken coop.

[0247] 1-2: Ensure that key areas (such as drinkers, feeders, perches) have sufficient sensors to accurately monitor poultry behavior.

[0248] 1-3: Select and install sensor locations based on the chicken coop's structural design (ventilation system, lighting system, etc.) to guarantee comprehensive environmental data.

[0249] Step 2: Data Collection and Preprocessing

[0250] 2-1: Stall operating the sensors and collect data regularly, including environmental data (temperature, humidity, light intensity, gas concentration) and poultry behavior data (movement frequency, posture, activity level).

[0251] 2-2: Collect image data of poultry using image sensors, preprocess the images to remove noise, and standardize them to adapt to deep learning models.

[0252] 2-3: Use data cleaning techniques to handle missing values or outliers in the data, ensuring the quality of subsequent processing.

[0253] Step 3: Data Fusion and Analysis

[0254] 3-1: Apply the Kalman filter algorithm to fuse data from different sensors (environmental sensors, activity sensors, image sensors) to improve data accuracy and reliability.

[0255] 3-2: Use the Hidden Markov Model (HMM) to model the poultry's behavioral states (such as eating, drinking, resting, activity), and calculate the transition probability of behaviors based on sensor data.

[0256] 3-3: Utilize deep learning algorithms (such as LSTM, RNN) to analyze time- series data, learn poultry behavior patterns, and identify abnormal behavior patterns.

[0257] Step 4: Behavior Recognition and Anomaly Detection

[0258] 4-1: Apply convolutional neural networks (CNNs) to extract features from image data, identify unique features of poultry (such as feather patterns, beak shape, leg features), and achieve individual identification through deep learning models.

[0259] 4-2: Utilize pose estimation models (such as OpenPose) to analyze changes in poultry posture and calculate activity levels based on behavioral changes (such as eating, walking, resting).

[0260] 4-3: Detect abnormal behaviors such as prolonged stillness or frequent feather pecking, and set thresholds. When these behaviors exceed the set range, send an alert to notify the manager.

[0261] Step 5: Real-time Monitoring and Alert System

[0262] 5-1: Set up a real-time monitoring system to present sensor data and analysis results to managers in real-time. Use dashboards to display the poultry's health status, activity level, and environmental conditions.

[0263] 5-2: When the system detects abnormal poultry behavior or unsuitable environmental conditions, automatically send alerts to the manager (such as via SMS,app notification, or email) to remind them to check the poultry or environmental conditions.

[0264] 5-3: After the alert is issued, the system will provide specific recommendations to help managers take corrective actions (such as adjusting temperature, humidity, or adjusting the poultry's diet and care).

[0265] Step 6: Environmental Optimization and Management Recommendations

[0266] 6-1: The system continuously monitors environmental parameters in the chicken coop and provides environmental optimization recommendations based on sensor data. If environmental conditions (such as temperature or humidity) reach dangerous levels, the system will prompt the manager to adjust the chicken coop ventilation system or strengthen cooling / heating measures.

[0267] 6-2: The system analyzes the correlation between poultry behavior data and environmental data, and gives early warnings for abnormal behaviors (such as prolonged stillness or fatigue) caused by environmental factors (such as excessive temperature), avoiding the impact of adverse environmental conditions on poultry health.

[0268] Step 7: Data Accumulation and Long-term Planning

[0269] 7-1: The system will continuously accumulate poultry behavior data and environmental data and conduct long-term analysis. This data will provide strong support for farm management, helping managers make production decisions, health management, and resource allocation.

[0270] 7-2: Based on historical data and behavior patterns, the system can predict the health risks and disease probability of the flock, and generate long-term health and production reports to provide data support for managers to assist in future planning.

[0271] Step 8: System Optimization and Feedback Mechanism

[0272] 8-1: Regularly evaluate the system's performance and optimize the model based on manager feedback and actual application results. This includes adjusting deep learning model parameters, improving anomaly detection algorithms, or retraining the prediction model based on new data.

[0273] 8-2: The system should have the ability to learn on its own, constantly adjusting and optimizing based on newly collected data and manager operations, maintaining accurate monitoring of poultry health and behavior.

[0274] Step 9: Report Generation and Data Visualization

[0275] 9-1: Based on the data analysis results, the system will automatically generate daily, weekly, or monthly reports detailing the poultry's health status, environmental conditions, behavior patterns, etc. The report should be concise and easy to understand, facilitating decision-making by managers.

[0276] 9-2: Data visualization tools will help managers visually view environmental data, health status, behavior patterns, etc., and conduct comparisons and analyses to assist in decision-making.

[0277] A method for intelligent poultry farm performance monitoring and data collection, comprising the steps of:a) sensor deployment and data collection:(i) deploying a plurality of sensors within the poultry house, wherein the sensors include:at least one environmental sensor configured to collect data related to environmental parameters within the poultry house, including temperature, humidity, and gas concentration;at least one activity sensor configured to detect and collect poultry movement data, wherein the activity sensor includes at least one of an accelerometer, a gyroscope, or an image sensor;at least one weighing device configured to collect data related to the weight of the poultry or the weight of feed being consumed by the poultry; (ii) collecting the following data using the deployed sensors:environmental data from the environmental sensor, including temperature, humidity, and gas concentration;activity data from the activity sensor, including movement frequency, posture, and activity levels of the poultry;poultry weight data from the weighing device, including the weight of the poultry or the amount of feed consumed by the poultry;b) data preprocessing:(iii) preprocessing the collected data by:filtering noise in the environmental data to eliminate irrelevant fluctuations or measurement errors, thereby ensuring the reliability of temperature, humidity, and gas concentration readings;standardizing the activity data to correct for sensor biases and convert raw movement data into useful behavioral information, such as identifying specific periods of inactivity or excessive movement;cleaning the poultry weight data to remove any anomalous readings caused by sensor malfunctions or measurement errors;c) data fusion and behavior analysis:(iv) fusing the collected data using a data fusion algorithm to combine the environmental data, activity data, and poultry weight data, wherein the data fusion algorithm is configured to:apply a kalman filter to integrate noisy sensor data from different sources (environmental sensors, activity sensors, weighing devices) in order to provide a more accurate and reliable set of data for further analysis;use a hidden markov model (hmm) to model the behavioral states of the poultry based on the activity data, including states such as resting, eating, drinking, or being active, the hmm is configured to calculate state transition probabilities based on sensor input, allowing for the prediction of behavioral patterns over time;d) anomaly detection:(v) analyzing the processed data to detect abnormal behaviors or conditions, comprising the steps of:identifying stationary behavior by detecting periods of inactivity exceeding a predefined threshold, indicating potential health issues or distress in the poultry;detecting abnormal movement patterns such as excessive or frequent feather pecking behavior, indicative of stress or aggression;setting behavior thresholds for abnormal behaviors, wherein the system compares real-time data with predefined thresholds, and if any of the thresholds are exceeded, an alert is triggered to notify the manager of the poultry house;e) real-time data analysis and alerts:(vi) real-time processing and monitoring of the poultry's health and behavior by:monitoring the poultry's health status continuously through the collection of activity and weight data, ensuring that any sudden changes in behavior are quickly detected;sending alerts to a farm manager when abnormal poultry behavior or unsuitable environmental conditions are detected, wherein the alert systemnotifies the manager via a communication medium such as sms, app notification, or email;providing corrective actions or recommendations based on the detected anomalies, such as adjusting environmental parameters (temperature, humidity) or modifying poultry care practices (diet, movement opportunities); f) long-term data analysis and decision support:(vii) accumulating the collected data over time and performing long-term analysis to:identify long-term behavioral trends and health patterns, enabling the farm manager to make informed decisions about poultry management, such as adjusting feeding schedules or altering living conditions based on historical data;predict potential health risks by analyzing the relationship between the poultry's behavior and environmental factors, enabling proactive management of poultry health and welfare.

[0278] In an another embodiment, the method for intelligent poultry farm performance monitoring and data collection, comprising the steps of:collecting environmental data from environmental sensors within a poultry house, wherein the environmental data includes at least temperature, humidity, and gas concentration values;collecting activity data from activity sensors, wherein the activity data includes movement frequency, posture, and activity levels of individual poultry;collecting weight data from weighing devices, wherein the weight data includes a body weight of individual poultry and feed consumption weight from feeding stations;capturing image data using image sensors to provide visual records of poultry and their behavior;preprocessing the collected data by filtering environmental data to remove noise and correct measurement errors, standardizing activity data to adjust for sensor- specific biases, cleaning weight data to eliminate outliers and ensure consistency in readings, and processing image data to normalize image formats and enhance clarity for feature extraction;fusing the preprocessed environmental, activity, weight, and image data using:a Kalman filter to integrate multi- sensor data and improve accuracy, and a Hidden Markov Model (HMM) to model poultry behavioral states;applying machine learning techniques to analyze the fused data, comprises:using Convolutional Neural Networks (CNNs) to extract unique features of poultry from image data, such as feather patterns, beak shapes, and leg characteristics;employing pose estimation algorithms to detect posture changes and calculate activity levels;utilizing Long Short-Term Memory (LSTM) networks or Recurrent Neural Networks (RNNs) to analyze time-series data, identify normal behavior patterns, and detect deviations indicating abnormal behavior;continuously monitoring the analyzed data to detect anomalies, comprises prolonged stillness, frequent feather pecking, and environmental parameter deviations beyond acceptable thresholds;triggering an alert system when anomalies are detected, wherein the alerts comprise descriptions of the abnormality and recommended corrective actions, such as adjusting environmental conditions or feed;transmitting the processed data and alerts to a centralized management interface, displaying current health status and activity levels of poultry, real-time environmental conditions in the poultry house, and historical trends for behavioral and environmental data; andgenerating actionable insights and periodic reports based on historical data to assist in long-term planning for health management and production optimization, and predict potential risks such as disease or stress based on behavior and environmental correlations.

[0279] In another embodiment, the present invention further comprises a thermal imaging camera module to enhance the system's capabilities. A thermal imaging or thermographic camera is a device that detects infrared radiation emitted by objects and converts it into a thermal image, displaying variations in surface temperature. Unlike conventional cameras, which rely on visible light, infrared radiation is emitted by all objects as long as the temperature is above absolute zero, allowing for the identification of temperature differences across an object or within an environment, with or without visible light.

[0280] In the context of poultry monitoring, a thermal imaging camera can capture heat images of birds by detecting the radiated heat from their bodies. These images provide a temperature map of the bird, revealing areas of heat concentration corresponding to its core body regions. Since feathers often act as insulation, they typically appear cooler than the bird’s skin or other uncovered parts, enabling differentiation between the body and feather layers.

[0281] By overlaying the thermal images with high-resolution visual camera images, a composite view of the chicken can be created. This technique allows for even more accurate measurement of the bird's true body size by isolating the thermal "core" regions that represent the actual body structure, separate from the cooler feathered areas.

[0282] In operation, the thermal imaging camera module is operably connected to the camera, which coordinates the collection and transmission of data. Thermal images, along with data from other modules such as the camera modules, weighing devices, and environmental sensors, are transmitted for processing. The processing of this combined data can occur:On-Device Processing: Directly within the core camera system, utilizing the integrated processor and storage modules.Edge Computing: Through a local computing device that receives and processes the combined data from at least one poultry house.Cloud Computing: Via a remote server or cloud platform for centralized and scalable data analysis.

[0283] The thermal imaging camera module allows for the extraction of additional features, significantly improving the system's analytical capabilities. Specifically:The thermal images provide precise delineation of the poultry’s body core by distinguishing it from the cooler feather layers that insulate the bird. This differentiation enables more accurate measurements of the actual body size.The additional images can enable better feature extraction through the use of thermal boundaries to define critical body parts on the bird, such as head, torso, and limbs.Thermal images can reveal temperature anomalies, such as hot spots from inflammation or cooler regions, that may indicate health issues or stress. Whencombined with visual and environmental data, this enhances the system’s ability to monitor poultry welfare.The combination of heat and visual data can improve the analysis of behaviors like resting or activity levels.

[0284] Therefore, above description has introduced the intelligent poultry farm monitoring system according to the present invention completely. Moreover, It is worth particularly explained that the above description is made on embodiments of the present invention. However, the embodiments are not intended to limit scope of the present invention, and all equivalent implementations or alterations within the spirit of the present invention still fall within the scope of the present invention.

Claims

What is claimed is:Third embodiment (Cloud computing)1. An intelligent poultry farm performance monitoring and data collection system, comprising:at least one first electronic device, being integrated with at least one camera, a first processor, a first storage module storing a first application program, and at least one ToF (Time of Flight) device therein;at least one weighing device, being coupled to the first electronic device; and a cloud computing device 1A in communication with the at least one first electronic device, and comprising a second processor and a second storage module storing a second application program;wherein the first processor executes the first application program so as to be configured to:acquire, by controlling the at least one camera, a plurality of poultry images from a plurality of poultry farmed in a poultry house;collect, by activating the at least one ToF device, a height data of the first electronic device;collect, through the at least one weighing device, a poultry feed weight data from at least one automation feeding apparatus that is disposed in the poultry house; andcollect, through the at least one weighing device, a poultry weight data from the plurality of poultry;wherein the second processor executes the second application program so as to be configured to:evaluate, by processing the plurality of poultry images, at least one poultry physical condition of each of the plurality of poultry;count, by processing the plurality of poultry images, a poultry number of the plurality of poultry;determine, whether at least one of the plurality of poultry is standing, sitting down or lying down according to the poultry physical condition; and count, by processing the poultry feed weight data, a feed consumption and / or a drinking water consumption.

2. The intelligent poultry farm performance monitoring and data collection system of claim 1, further comprising:at least one PoE (Power over Ethernet) splitter, being integrated in the at least one first electronic device;at least one PoE injector, being in communication with the PoE splitter through an Ethernet cable, being linked to Internet through an Ethernet device, and being electrical connected to an AC power source via a power cable;at least one microphone, being integrated in the first electronic device; and at least one environmental sensor, being integrated in the first electronic device.

3. The intelligent poultry farm performance monitoring and data collection system of claim 2, wherein the second application program comprises a pre-trained poultry growth prediction model and a pre-trained poultry weight prediction model, and the second processor is further configured to:input, the plurality of poultry images, the feed consumption, the drinking water consumption, and the poultry weight data into the pre-trained poultry growth prediction model, such that the pre-trained poultry growth prediction model outputs a poultry growth prediction data; andinput, the feed consumption, the drinking water consumption, and the poultryweight data into the pre-trained poultry weight prediction model, such that the pretrained poultry weight prediction model outputs a poultry weight prediction data.

4. The intelligent poultry farm performance monitoring and data collection system of claim 2, wherein the second application program comprises a pre-trained feeding management policy optimizing model, and the second processor is further configured to:input, the feed consumption, the drinking water consumption, and the poultry weight data into the pre-trained feeding management policy optimizing model, such that the pre-trained feeding management policy optimizing model outputs an optimized feeding management policy.

5. The intelligent poultry farm performance monitoring and data collection system of claim 1, wherein the first electronic device is further integrated with at least one IMU (inertial measurement unit) device therein, and the first processor is further configured to:receive, an inertial data from the IMU device.

6. The intelligent poultry farm performance monitoring and data collection system of claim 5, wherein the second processor is further configured to:calculate, by processing the inertial data, a camera angle of the camera.

7. The intelligent poultry farm performance monitoring and data collection system of claim 5, further comprising an electronic device capable of communicating with the second electronic device, and the electronic device comprising a processor, a display module and a storage module storing a user-end application program, suchthat the processor executes the user-end application program so as to be configured to:show, by controlling the display module, a user operation interface; show, through the user operation interface, a dashboard comprising a plurality of statistic graphs and data charts; andshow, a real-time monitoring images of the plurality of poultry farmed in the poultry house.

8. The intelligent poultry farm performance monitoring and data collection system of claim 7, wherein the dashboard is accessible via an app, a webapp, and a web browser selected from a group consisting of Chrome, Safari, Firefox, Edge, web or webapp.

9. The intelligent poultry farm performance monitoring and data collection system of claim 7, wherein the second processor is further configured to:generate and transmit, in case of the camera angle is in need of adjustment, a first notification signal to the electronic device;generate and transmit, in case of the automation feeding apparatus is in need of feed adding or in need of water adding, a second notification signal to the electronic device;generate and transmit, in case of at least one of the plurality of poultry is sitting or lying down for a period of time, a third notification signal to the electronic device;generate and transmit, in case of the poultry number revealing that there is at least one of the plurality of poultry missing, a fourth notification signal to the electronic device; andgenerate and emit, in case of a sudden network dropping, an alarm signal.

10. The intelligent poultry farm performance monitoring and data collection system of claim 7, wherein the camera is an electronic pan-tilt- zoom (ePTZ) camera, and the first processor is further configured to:conduct, a camera adjustment to the camera;wherein the camera adjustment is selected from a group consisting of illumination adjustment, camera angle adjustment, camera sampling rate adjustment, and basic camera parameters.

11. The intelligent poultry farm performance monitoring and data collection system of claim 2, wherein the second application program comprises a pre-trained image segmentation and labelling model that is utilized for firstly segmenting each of the plurality of poultry images, thereby generating a plurality of segmented image and then labelling each of the plurality of segmented image.

12. The intelligent poultry farm performance monitoring and data collection system of claim 2, wherein the second application program comprises a pre-trained background subtraction algorithm that is utilized for subtracting a background including poultry dung from each of the plurality of poultry images.

13. The intelligent poultry farm performance monitoring and data collection system of claim 1, wherein a camera comprises an upper base and a shell assembled to the upper base, and the upper base is a space where the camera is attached, and the shell has a dome type shape.

14. A method for intelligent poultry farm performance monitoring and data collection,comprising the steps of:collecting environmental data from environmental sensors within a poultry house, wherein the environmental data includes at least temperature, humidity, and gas concentration values;collecting activity data from activity sensors, wherein the activity data includes movement frequency, posture, and activity levels of individual poultry;collecting weight data from weighing devices, wherein the weight data includes a body weight of individual poultry and feed consumption weight from feeding stations;capturing image data using image sensors to provide visual records of poultry and their behavior;preprocessing the collected data by filtering environmental data to remove noise and correct measurement errors, standardizing activity data to adjust for sensor- specific biases, cleaning weight data to eliminate outliers and ensure consistency in readings, and processing image data to normalize image formats and enhance clarity for feature extraction;fusing the preprocessed environmental, activity, weight, and image data using:a Kalman filter to integrate multi- sensor data and improve accuracy, and a Hidden Markov Model (HMM) to model poultry behavioral states;applying machine learning techniques to analyze the fused data, comprises:using Convolutional Neural Networks (CNNs) to extract unique features of poultry from image data, such as feather patterns, beak shapes, and leg characteristics;employing pose estimation algorithms to detect posture changes and calculate activity levels;utilizing Long Short-Term Memory (LSTM) networks or Recurrent NeuralNetworks (RNNs) to analyze time-series data, identify normal behavior patterns, and detect deviations indicating abnormal behavior;continuously monitoring the analyzed data to detect anomalies, comprises prolonged stillness, frequent feather pecking, and environmental parameter deviations beyond acceptable thresholds;triggering an alert system when anomalies are detected, wherein the alerts comprise descriptions of the abnormality and recommended corrective actions, such as adjusting environmental conditions or feed;transmitting the processed data and alerts to a centralized management interface, displaying current health status and activity levels of poultry, real-time environmental conditions in the poultry house, and historical trends for behavioral and environmental data; andgenerating actionable insights and periodic reports based on historical data to assist in long-term planning for health management and production optimization, and predict potential risks such as disease or stress based on behavior and environmental correlations.

Images