Animal health monitoring system

Abstract

The invention relates to the Internet of Things technology, and discloses an animal health monitoring system which comprises wearable equipment and a cloud server. The wearable equipment comprises a processor, and an acceleration sensor, a body temperature sensor and a wireless transceiver which are connected with the processor; the processor is configured to determine the posture of the animal according to the acceleration data detected by the acceleration sensor, and obtain the body temperature of the animal from the body temperature sensor; the posture-body temperature data pair is sent to a cloud server through a wireless transceiver; the cloud server is configured to input a plurality of groups of posture-body temperature data pairs which are sequenced according to the time sequence and come from the wearable device into a pre-trained first machine learning model, and the animal health state output by the first machine learning model is obtained. According to the system, the transmission energy consumption is reduced, the endurance time is prolonged, and the accuracy of animal health state judgment is improved.

Description

Technical Field

[0001] The present application relates to Internet of Things technology, and in particular to using the Internet of Things to monitor animal health. Background Art

[0002] This section is intended to provide a background or context for understanding the embodiments of the present application and is for reference only. The applicant should not be considered to admit that this section belongs to the prior art that has been disclosed before the filing date of the present application.

[0003] In animal husbandry production, monitoring of animal health is of great significance for improving breeding efficiency. At present, IoT technology has been widely used in the field of animal health monitoring. A common practice is to wear wearable devices on animals, collect the physiological data of animals through various sensors, and transmit the data to the cloud for analysis.

[0004] For example, a temperature sensor can collect animal temperature data, and an acceleration sensor can collect animal movement data. The raw data collected by these sensors is transmitted to a cloud server via wireless communication (such as Bluetooth or long-distance radio communication) for analysis and processing to monitor the health status of the animal.

[0005] However, this monitoring method has the following problems: First, due to the need to frequently transmit a large amount of raw data to the cloud, wearable devices consume a lot of energy and have a short standby time, usually only a few days to one or two weeks. This obviously cannot meet the needs of long-term health monitoring of animals in animal husbandry production. Secondly, simply transmitting a single body temperature data or movement data to the cloud for analysis cannot fully reflect the health status of the animal, which can easily lead to misjudgment. Summary of the invention

[0006] The purpose of this application is to provide an animal health monitoring system, which not only reduces transmission energy consumption and prolongs battery life, but also improves the accuracy of judging the health status of animals.

[0007] The present application discloses an animal health monitoring system, comprising: a wearable device and a cloud server;

[0008] The wearable device includes a processor, an acceleration sensor, a body temperature sensor and a wireless transceiver connected to the processor; the processor is configured to: determine the posture of the animal according to the acceleration data detected by the acceleration sensor, and obtain the body temperature of the animal from the body temperature sensor; send the posture-body temperature data pair to the cloud server through the wireless transceiver;

[0009] The cloud server is configured to: input multiple groups of posture-body temperature data pairs from the wearable device sorted in chronological order into a pre-trained first machine learning model to obtain the animal health status output by the first machine learning model.

[0010] In a preferred embodiment, there are multiple wearable devices;

[0011] The first machine learning model is trained based on posture-body temperature data pairs from multiple wearable devices.

[0012] In a preferred example, the wearable device also includes a memory; the processor is configured to store the posture-body temperature data pairs in the memory, and send multiple groups of the posture-body temperature data pairs stored in the memory to the cloud server via the wireless transceiver at a predetermined time.

[0013] In a preferred example, the processor is further configured to: determine the sleep duration according to the current posture of the animal, control the wearable device to enter a sleep state, wake up the wearable device after the sleep duration and perform the next detection, wherein at least two different postures correspond to different sleep durations.

[0014] In a preferred embodiment, the wearable device further includes a gyroscope and a magnetometer;

[0015] The processor is further configured to determine the posture of the animal based on the data detected by the acceleration sensor, the gyroscope, and the magnetometer.

[0016] In a preferred example, the processor uses a second machine learning model in the wearable device to determine the posture of the animal based on acceleration data detected by the acceleration sensor.

[0017] In a preferred example, the processor is further configured to dynamically adjust the sampling frequencies of the acceleration sensor, the gyroscope, the magnetometer and the body temperature sensor according to the determined posture.

[0018] In a preferred example, the system further comprises an environmental parameter sensor arranged in the animal breeding environment, for measuring the environmental parameters of the environment in which the animal is located;

[0019] The environmental parameter sensor periodically uploads environmental parameters to the cloud server;

[0020] The wearable device further includes a clock, and the processor is further configured to obtain time information for determining the posture and obtaining the body temperature from the clock, form a timestamp, and send the posture-body temperature data pair, the corresponding timestamp and the identifier of the wearable device to the cloud server via the wireless transceiver;

[0021] The cloud server determines the environmental parameter sensor of the environment where the wearable device is located based on the identifier of the wearable device, and determines the environmental parameters corresponding to the posture-body temperature data pair based on the timestamp; inputs the posture-body temperature data pair and its corresponding environmental parameters into the first machine learning model to obtain the animal health status output by the first machine learning model.

[0022] In a preferred embodiment, the environmental parameters include one of the following or any combination thereof: ambient temperature, ambient humidity, ambient air pressure;

[0023] The environmental parameter sensor includes one of the following or any combination thereof:

[0024] A temperature sensor for measuring ambient temperature, a humidity sensor for measuring ambient humidity, and a pressure sensor for measuring ambient air pressure.

[0025] In a preferred example, a wireless gateway device is further included to establish a wireless connection with the plurality of wearable devices and the plurality of environmental parameter sensors, and upload the data from the wearable devices and the environmental parameter sensors to the cloud server.

[0026] In the implementation of the present application, by integrating an acceleration sensor and a body temperature sensor in a wearable device, and having a processor determine the posture of the animal based on the acceleration data, obtain the body temperature at the same time, and upload the posture-body temperature data pair to the cloud server to input into the pre-trained first machine learning model for health status prediction, accurate monitoring and intelligent analysis of the health status of the animal can be achieved. This technical solution can greatly reduce the massive transmission of raw data and the burden of cloud processing. By generating posture-body temperature data pairs in advance at the terminal, it not only reduces the energy consumption of the device and extends the standby time, but also can continuously improve the accuracy and real-time performance of health judgment by using the continuous learning of the cloud model.

[0027] Furthermore, by using posture-temperature data pairs from multiple wearable devices for joint training, the scale of training data can be effectively expanded, and the generalization ability and prediction accuracy of machine learning models can be improved, thereby more accurately analyzing the health status of animals at the group level.

[0028] Furthermore, by setting up a memory in the wearable device and regularly uploading posture-temperature data pairs in batches, the energy consumption caused by frequent data transmission can be reduced, the battery life of the wearable device can be improved, and long-term reliable operation can be ensured under limited power conditions.

[0029] Furthermore, by dynamically determining the sleep duration of the wearable device based on the animal's current posture, the device can be put into sleep mode in a low-activity state, thereby reducing unnecessary measurements and transmissions, which can significantly reduce energy consumption and extend the working time of the device.

[0030] Furthermore, by integrating accelerometers, gyroscopes and magnetometers in wearable devices and fusing these data with a processor to determine the animal's posture, the accuracy and robustness of posture recognition can be improved, and judgment bias caused by the limitations of a single sensor can be reduced.

[0031] Furthermore, by deploying a second machine learning model on the wearable device to perform posture recognition on the acceleration data, the data can be processed quickly on the front end, reducing dependence on cloud computing resources and reducing the frequency of uploading raw data, thereby achieving more efficient edge computing and energy saving.

[0032] Furthermore, by dynamically adjusting the sampling frequencies of the accelerometer, gyroscope, magnetometer, and temperature sensor according to the determined posture, different postures can correspond to different sensor sampling frequencies, which can improve the data sampling accuracy when key behaviors occur, reduce the sampling rate when there are no important posture changes to save power consumption, and optimize the overall monitoring strategy.

[0033] Furthermore, by adding environmental parameter sensors to the system and combining the ambient temperature, ambient humidity or ambient air pressure with the posture-body temperature data pair to form time-stamped associated data and input it into the cloud model, a more comprehensive health status analysis can be obtained, improving the accuracy and adaptability of the prediction. By performing multi-dimensional correlation analysis on environmental parameters such as ambient temperature, humidity, and air pressure with animal posture-body temperature data, the external causes of abnormal animal health can be more accurately identified, thereby achieving more refined health risk prediction and intervention strategies.

[0034] Each technical feature disclosed in the above invention content, each technical feature disclosed in each implementation mode and example below, and each technical feature disclosed in the accompanying drawings can be freely combined with each other to form various new technical solutions (these technical solutions should be deemed to have been recorded in this specification), unless such combination of technical features is technically infeasible. For example, in one example, feature A+B+C is disclosed, and in another example, feature A+B+D+E is disclosed, and features C and D are equivalent technical means that play the same role. Technically, only one of them can be used, and it is impossible to use them at the same time. Feature E can be combined with feature C technically. Then, the solution of A+B+C+D should not be deemed to have been recorded because it is technically infeasible, while the solution of A+B+C+E should be deemed to have been recorded. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 It is a schematic diagram of the structure of an animal health monitoring system according to an embodiment of the present application. DETAILED DESCRIPTION

[0036] In the following description, many technical details are provided to help readers better understand the present application. However, those skilled in the art can understand that the technical solution claimed in the present application can be implemented even without these technical details and various changes and modifications based on the following embodiments.

[0037] Description of some concepts:

[0038] Wearable devices are electronic devices that can be worn on animals and integrate multiple sensors and processors to collect and process the physiological and behavioral data of animals and transmit the data to external systems through wireless communication. They are usually small in size, light in weight, and low in power consumption.

[0039] A machine learning model is a computer-implemented model that can autonomously learn patterns and associations in historical data to make predictions or decisions about new data without explicit programming of specific rules.

[0040] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below in conjunction with the accompanying drawings.

[0041] The embodiments of the present application relate to an animal health monitoring system, including: a wearable device and a cloud server. Usually there are multiple wearable devices, each of which is worn on an animal. For example, the wearable device can be an ear tag fixed on the ear of a pig, a cow, or a sheep. It can also be other types, such as a collar-type wearable device, a strap-type wearable device, an anklet-type wearable device, etc. The cloud server can be a computer or a computer cluster or a virtual host on the Internet, etc. Figure 1 is a schematic diagram of the structure of an example of a typical animal health monitoring system.

[0042] The wearable device includes a processor, an acceleration sensor, a body temperature sensor and a wireless transceiver connected to the processor. The processor is configured to: determine the posture of the animal according to the acceleration data detected by the acceleration sensor, and obtain the body temperature of the animal from the body temperature sensor. The posture-body temperature data pair is sent to a cloud server through the wireless transceiver. The wireless transceiver can be Bluetooth, WIFI, Zigbee or other low-power wireless communication devices.

[0043] The cloud server is configured to: input multiple groups of posture-body temperature data pairs from wearable devices sorted in chronological order into a pre-trained first machine learning model to obtain the health status of the animal output by the first machine learning model. The health status of the animal can be, for example, healthy, slightly unwell, and sick. The user can use the terminal device to access the cloud server to obtain the health status of the animal. The cloud server can also actively send a notification message to the user's terminal when it detects that the animal may be sick.

[0044] The applicant has found that an animal's posture has a strong correlation with its mental state and health status.

[0045] For example, in the case of pigs, posture is related to their mental state and health status, including:

[0046] (1) A pig sitting in a dog-like posture indicates breathing difficulties, which are commonly caused by pneumonia, heart failure, pleurisy or anemia.

[0047] (2) Standing with the head and neck stretched forward also indicates difficulty breathing. Pigs with pleurisy often stand with their backs hunched.

[0048] (3) Sick pigs with lameness are usually unwilling to stand.

[0049] (4) Tilting of the head and neck or circular motion is common in otitis media, otitis interna, meningitis, and brain abscess. The direction of the head and neck bending or circular motion is centered on the affected side.

[0050] (6) A healthy pig always has its ears erect or extended forward. If its ears droop or stick out backward, it indicates that the pig is in a bad mental state. If the pig rushes around and does not respond to external sounds, it may be deaf or blind.

[0051] Even the sleeping position of pigs can reflect their health status. Pigs usually sleep on their sides and stomachs.

[0052] 1. Side-lying is the most common sleeping position for pigs. Pigs usually lie on one side with their heads resting on their front or hind hooves.

[0053] 2. The prone position is a more relaxed position for pigs when sleeping. They will stretch their heads and front hooves and lie on the ground.

[0054] 3. Of course, in addition to these two postures, pigs also have other less common sleeping positions, such as lying on their backs and lying on their stomachs. Different pigs may have different sleeping habits, but in general, lying on their sides and lying on their stomachs are their favorite sleeping positions.

[0055] Sick pigs often stick their entire bodies to the ground and look uncomfortable. Some pigs are even too uncomfortable to sleep and sit like dogs. This is the dog sit familiar to pig farmers. Of course, just looking at the pig's sleeping posture cannot determine that the pig is sick. In addition, you also need to observe the pig's condition.

[0056] When healthy pigs sleep, their muscles are relaxed and their breathing rhythm is even; however, this is not the case for sick pigs. Sick pigs are usually very uncomfortable when they sleep. They are exhausted and fall asleep prone, often accompanied by abnormal conditions such as difficulty breathing, panting, coughing, and sitting like a dog.

[0057] This application uses pre-training and pre-learning to make a preliminary judgment on the animal's posture in the wearable device, which can not only reduce the storage of a large amount of data and the upload of raw data, but also avoid data changes caused by sensor or environmental changes. In this way, using these postures to help learn large models in the future and improve the accuracy of animal health monitoring judgments will be very meaningful.

[0058] The working principle of the accelerometer is rooted in Newton's second law, which states that the acceleration of an object is proportional to the force. In actual operation, the accelerometer can keenly sense the subtle changes in the acceleration of an object during its motion, and accurately convert this abstract physical quantity into an electrical signal that can be identified and processed for output, so as to achieve effective monitoring and in-depth analysis of the object's motion state. Not only that, the accelerometer also has a strong collaborative ability. It can work hand in hand with other types of sensors such as gyroscopes and magnetometers, and use advanced angle fusion algorithms to accurately calculate the attitude angle, which includes yaw, pitch and roll angles. These attitude angles are like the precise coordinates of the object's motion posture, which can meticulously depict the specific position and motion of the object in space.

[0059] Based on this principle, the smart ear tag of one embodiment of the present application is equipped with an acceleration sensor and is further equipped with a gyroscope. With this combination, the smart ear tag can make preliminary and accurate judgments on the details of the animal's movements and posture changes. These judgment data initially generated at the front end will be properly stored as status data by the smart ear tag system, and then uploaded to the big data model, providing rich and valuable data materials for subsequent more in-depth and complex artificial intelligence learning, thereby promoting the entire animal monitoring and analysis system to move forward in a more intelligent and efficient direction.

[0060] These motion sensors have the potential to be further integrated with temperature sensors for algorithmic fusion, which can provide a more in-depth and comprehensive analysis of the health of animals. This is because the body temperature of an animal can largely reflect its potential disease state. Take pigs, for example. Within a specific temperature range, the change in body temperature is actually an instinctive manifestation of the pigs' self-care. For example, when the pig's body temperature rises slightly, its immunity will be enhanced accordingly, thereby helping the pigs resist the invasion of external diseases. Therefore, when the pig's body temperature is below 41 degrees, it is necessary to be particularly cautious if drugs are used to reduce fever.

[0061] The normal body temperature of pigs is usually between 38-39.5 degrees Celsius, which is significantly higher than human body temperature. In the process of pig farming, whether it is giving pigs medicine or injecting vaccines, the body temperature of pigs will fluctuate. In addition, the body temperature of pigs will also rise after being stressed or engaging in strenuous activities. Even the body temperature of sows after giving birth reaches 40.3 degrees, which is considered normal.

[0062] Judging from the temperature rise of sick pigs, it can be roughly divided into the following typical stages: the first is the mild fever stage, with the body temperature between 40 and 41 degrees. This stage is common in local inflammations such as mastitis and gastroenteritis after pig delivery, as well as certain chronic infectious diseases, such as chronic swine fever, paratyphoid, etc.; the second is the moderate fever stage, with the body temperature in the range of 41-42 degrees, which often occurs in acute viral infectious diseases, such as acute swine fever, swine influenza, etc., or local infectious diseases with a greater impact such as pneumonia; the third is the high fever stage, with the body temperature between 42 and 43 degrees, which is usually closely related to acute infectious diseases and inflammation, such as swine fever, pneumonia, lobar pneumonia, pleuropneumonia and other diseases; the last is the extremely high fever stage, with the body temperature reaching above 43 degrees, which often occurs in serious acute infectious diseases, such as swine erysipelas, anthrax, acute septicemic streptococcus and other critical diseases. Through accurate monitoring and stage judgment of pig body temperature changes, combined with data obtained by motion sensors, it can provide extremely critical and accurate basis for pig health management and disease prevention, help the pig farming industry develop in a more scientific and efficient direction, and ensure the health of the pig herd and maximize breeding benefits.

[0063] For example, you can choose some typical pig postures, such as sitting with an arched back, standing, walking, trotting, lying on the side, lying prone, looking up, looking down, etc., and summarize the original data into posture data before uploading it for processing.

[0064] In order to effectively reduce energy consumption and save electricity, the data can be stored first and then uploaded for processing. Specifically, in the data collection stage, when the amount of data accumulates to a certain scale, it is then uploaded and processed in a centralized manner, which can avoid the power loss caused by frequent data transmission. In addition, the sampling frequency can be dynamically adjusted according to the posture of the animal to achieve energy-saving optimization. For example, when a pig is in a resting state, its posture is relatively stable and the movement amplitude is very small. At this time, we can put the entire sensor module into sleep mode and suspend data sampling, thereby greatly reducing power consumption. When the pig starts to walk fast, its exercise intensity and speed increase significantly, and its body posture changes frequently and violently. In this case, the sampling frequency can be increased accordingly to accurately capture its movement data and posture change details. When the pig is lying on its side or performing non-violent exercise, its posture changes are relatively mild, so the sampling frequency can be appropriately reduced. On the premise of ensuring that its basic state can be effectively monitored, the amount of data generated can be reduced as much as possible, thereby reducing power consumption, extending the battery life of the sensor, and improving the energy efficiency performance of the overall system.

[0065] Optionally, in one embodiment, the first machine learning model is trained based on posture-body temperature data pairs from multiple wearable devices. The posture-body temperature data pairs of multiple animals can be continuously monitored, and the health status of the animals can be manually determined (i.e., the posture-body temperature data pairs are marked) to form training data, and supervised training is performed on the first machine learning model.

[0066] As long as the first machine learning model can identify the health status of an animal based on posture-body temperature data pairs of multiple animals sorted in time, a variety of models can be used to implement the first machine learning model. The following are some non-limiting examples.

[0067] The LSTM (Long Short-Term Memory) model is an option. It can effectively process time series data and learn the long-term dependencies between animal behavior and body temperature changes. LSTM can receive posture-body temperature sequence data in a continuous time window and predict the health status of the animal.

[0068] GRU (Gated Recurrent Unit) network is another option. Compared with LSTM, GRU has a simpler structure, fewer parameters, faster training speed, and can also effectively capture time series features. For resource-constrained scenarios, GRU is a good choice.

[0069] Temporal convolutional networks (TCNs) can also be considered. TCNs process time series data through causal convolutions, which has better parallelism and faster training than recurrent neural networks. It can effectively extract local pattern features of animal behavior and physiological indicators.

[0070] The encoder part of the Transformer model can also be used. Through the self-attention mechanism, it can capture long-distance dependencies in the sequence, which is particularly suitable for processing animal behavior patterns over a longer time span.

[0071] Optionally, in one embodiment, the wearable device further includes a memory. The processor is configured to store the posture-body temperature data pairs in the memory, and send the plurality of posture-body temperature data pairs stored in the memory to the cloud server via the wireless transceiver at a predetermined time. For example, the data may be reported once every hour or once a day.

[0072] Optionally, in one embodiment, the processor is also configured to dynamically determine an appropriate sleep duration based on the current posture of the animal, and perform low-power management of the wearable device accordingly. When the system detects through an accelerometer (and optional gyroscope and magnetometer) that the animal is in a specific posture (e.g., lying still, standing still, moving slowly, or running / walking with high activity), the processor will allocate different sleep durations according to a pre-set strategy. For example, when it is detected that the animal is in a relatively stable, low-activity posture (such as lying down to rest) for a long time, the processor can put the wearable device into a longer sleep state to achieve energy saving. When the animal is in a highly active state (e.g., eating or moving quickly), the processor can shorten the sleep duration to ensure that the device wakes up more frequently and collects data, thereby timely capturing key dynamic changes that may affect health status.

[0073] Furthermore, this posture-based dynamic sleep duration control can also be optimized in combination with historical data and environmental information. For example, if the system finds through long-term monitoring that an animal is generally in a low-activity, resting state during a certain period of time (such as the early morning), the device can automatically increase the sleep duration during this period to maximize battery life; conversely, if the animal is more active or is in the stage of a potential disease attack during a certain period of time, the system can shorten the sleep cycle to increase the data sampling frequency, so as to detect abnormal conditions in a timely manner.

[0074] Through this strategy, the power supply and data acquisition frequency of the wearable device in different states are dynamically optimized, achieving the goal of minimizing power consumption while ensuring monitoring accuracy, thereby extending battery life and reducing maintenance and replacement costs. This intelligent low-power management solution lays the foundation for the long-term and stable operation of the system in actual large-scale livestock scenarios. Optionally, in one embodiment, the wearable device also includes other sensors, such as a gyroscope, a magnetometer, a heart rate sensor, a sound sensor, and the like. The processor is also configured to determine the posture of the animal based on the data detected by the accelerometer, the gyroscope, and the magnetometer, so that the determined animal posture is more accurate than using the accelerometer alone.

[0075] Optionally, in one embodiment, the processor may use a second machine learning model in the wearable device to determine the posture of the animal based on acceleration data detected by the acceleration sensor. The second machine learning model may be implemented in a variety of ways. The following are some non-limiting examples:

[0076] Traditional machine learning models:

[0077] Random Forest: After extracting features from acceleration data (such as mean, variance, frequency domain features, peak features, etc.), random forest is used to classify feature vectors. This method is easy to implement and has good performance and robustness for small-scale data.

[0078] Support Vector Machine (SVM): The acceleration sequence is also converted into a feature vector through feature engineering, and SVM is used for multi-class posture recognition. SVM has a good classification effect on high-dimensional data and has high training efficiency when the data scale is moderate.

[0079] Deep Learning Models:

[0080] Convolutional Neural Network (CNN): The acceleration data can be regarded as a one-dimensional time series and input into a one-dimensional convolutional neural network to automatically learn features and output posture classification results. CNN can reduce the workload of manually designing features in feature extraction.

[0081] Long short-term memory network (LSTM) or gated recurrent unit (GRU): In view of the temporal correlation of time series data, LSTM or GRU is used to model the acceleration signal, which can better capture the temporal persistence and dynamic changes of animal posture.

[0082] Temporal Convolutional Network (TCN): It efficiently models time series through one-dimensional convolution and dilated convolution, balancing performance and computational efficiency in capturing long-term dependencies and short-term features.

[0083] Combining CNN with LSTM (i.e. first extracting local features through CNN and then modeling global timing information through LSTM) is also a common strategy, which can extract effective spatial features and temporal features in the model at the same time.

[0084] Lightweight Models and Embedded Optimizations:

[0085] Considering the computing power and energy consumption limitations of wearable devices, some lightweight models can be selected or the above models can be pruned, quantized, and distilled. For example:

[0086] MicroNets: Such as MicroCNN or quantized LSTM, these deep learning architectures that have undergone model compression and optimization still maintain relatively good recognition accuracy when running on embedded devices.

[0087] TinyML Tools: Compress and quantize neural networks through TinyML toolchains (such as TensorFlow Lite Micro) to make them suitable for resource-constrained hardware platforms, so that they can run efficiently in wearable devices.

[0088] In summary, we can choose from traditional machine learning models (such as random forest, SVM) to deep learning models (CNN, LSTM, GRU, etc.) and their lightweight versions to implement the second machine learning model to accurately and efficiently identify animal postures based on the data from the acceleration sensor. When selecting a specific model, we should comprehensively consider the model accuracy, computing and storage resources, and the actual operating conditions and application scenarios of the wearable device.

[0089] Optionally, for the training of the second machine model, the wearable device can upload the raw data of the accelerometer to the cloud at the beginning, and then train the second machine model based on the big data in the cloud to judge the posture of the animal according to the acceleration data. Although this process consumes a lot of power, it is a pre-training process. The trained second machine learning model is implanted into the wearable device, and the wearable device will not upload the raw data of the accelerometer in the future.

[0090] Optionally, in one embodiment, the processor may also use a non-machine learning algorithm to determine the posture of the animal based on the outputs of the accelerometer, gyroscope, and magnetometer.

[0091] Optionally, in one embodiment, the processor is further configured to dynamically adjust the sampling frequency of the acceleration sensor, the gyroscope, the magnetometer, and the body temperature sensor according to the determined posture.

[0092] Optionally, in one embodiment, the system also includes an environmental parameter sensor arranged in the animal breeding environment, which is used to measure the environmental parameters of the animal's environment. The environmental parameter sensor periodically uploads the environmental parameters to the cloud server. The wearable device also includes a clock, and the processor is also configured to obtain time information for determining the posture and obtaining the body temperature from the clock, form a timestamp, and send the posture-body temperature data pair, the corresponding timestamp and the identification of the wearable device to the cloud server through a wireless transceiver. The cloud server determines the environmental parameter sensor of the environment where the wearable device is located based on the identification of the wearable device, and determines the environmental parameters corresponding to the posture-body temperature data pair based on the timestamp. The posture-body temperature data pair and its corresponding environmental parameters are input into the first machine learning model to obtain the animal health status output by the first machine learning model.

[0093] Environmental parameters may include environmental temperature, environmental humidity, environmental air pressure, etc. Environmental parameter sensors may include temperature sensors for measuring environmental temperature, humidity sensors for measuring environmental humidity, pressure sensors for measuring environmental air pressure, etc.

[0094] A timestamp is data representing time, not necessarily a time value. For example, if the environmental parameter sensor is used periodically, the timestamp can be an integer representing the number of cycles that have passed since the agreed start time.

[0095] Optionally, in one embodiment, in order to improve the accuracy and applicability of the system in assessing the health status of animals, the system can further expand the scope of collection and utilization of environmental parameters. In this embodiment, not only basic environmental parameter sensors such as temperature, humidity, and air pressure are included, but more types of sensors can also be deployed to obtain more comprehensive environmental information, such as:

[0096] Harmful gas sensor: used to detect the concentration of harmful or irritating gases (such as ammonia and hydrogen sulfide) in the breeding environment to assess the possible negative impact of the breeding environment on the animal's respiratory system and overall health.

[0097] Dust / particle sensor: used to detect the concentration of particulate matter in the environment (such as PM2.5, PM10), so as to assess the air quality level of the breeding environment and its potential impact on animal health.

[0098] Light sensor: used to measure the light intensity and light cycle in the enclosure so as to correlate the ambient light conditions with the animal's behavior (eating, exercise, resting cycles).

[0099] Acoustic sensors: used to monitor noise levels in pens or pastures and assess the impact of the sound environment on animal stress levels and health.

[0100] In this embodiment, the environmental parameter sensor will upload multi-dimensional environmental data including ambient temperature, humidity, air pressure, gas concentration, number of dust particles, light intensity and noise level to the cloud server at set time intervals (for example, every few minutes to several hours, and the specific interval can be flexibly adjusted according to actual needs).

[0101] Wearable devices still use built-in clocks to timestamp posture and temperature data, and transmit posture-temperature data pairs with timestamps and device identification to cloud servers. Based on the identification information of the wearable device, the cloud server matches and associates the posture-temperature data pairs corresponding to the device with the environmental data at the same timestamp (or close to the timestamp) measured by the environmental parameter sensor corresponding to its location.

[0102] After obtaining complete data association, the cloud server inputs multimodal data including posture, body temperature and various environmental parameters into the pre-trained first machine learning model. The model has used these environmental parameters as part of the input during the training process to improve the accuracy of health status prediction. For example, if the model recognizes that an animal has abnormally high body temperature and atypical posture (such as laziness, shortness of breath, etc.) in an environment with high temperature and humidity or high concentration of harmful gases, the probability of the model issuing an early warning of its health status will increase accordingly.

[0103] This embodiment has the following technical effects:

[0104] 1. Improve prediction accuracy and scenario adaptability: With more comprehensive environmental parameter data, the model can better distinguish short-term abnormal behaviors caused by environmental changes from real health problems caused by diseases and other reasons, thereby reducing false positives or negatives.

[0105] 2. Support accurate decision-making and intervention: Based on the comprehensive analysis of environmental data and animal status data, keepers can promptly identify unhealthy feeding conditions (such as overcrowding, poor ventilation, high concentrations of harmful gases, etc.) and make targeted adjustments (increase ventilation, control light, improve pen cleanliness, etc.).

[0106] 3. Promote large-scale intelligent feeding management: In large-scale breeding scenarios, the integration of environmental parameter sensors, wearable devices and cloud-based intelligent analysis can help provide dynamic and detailed health monitoring and advice for each animal. By continuously accumulating data and training models, the system can continuously improve its understanding of the complex relationship between environmental changes and animal health, thereby achieving a higher level of automated and intelligent livestock management.

[0107] In summary, through richer environmental parameter collection and deep data fusion, a more comprehensive, accurate and scalable foundation is provided for animal health monitoring and early warning, enabling the first machine learning model to make more reliable predictions about animal health status in complex and changing breeding environments.

[0108] Optionally, in one embodiment, the system further includes a wireless gateway device, which establishes wireless connections with multiple wearable devices and multiple environmental parameter sensors, and uploads data from the wearable devices and environmental parameter sensors to a cloud server. The wireless gateway device is not required. For example, if the wearable device can directly access the Internet (e.g., through an ultra-low power cellular network), the wireless gateway device can be omitted.

[0109] Optionally, in one embodiment, the system adopts a multi-layer wireless network architecture, including multiple wireless gateway devices. Each wireless gateway device is responsible for covering wearable devices and environmental parameter sensors in a specific area and establishing a stable wireless connection with these devices. The wireless gateway device adopts an industrial-grade design, has waterproof and dustproof functions, and can be installed in an appropriate location in the animal breeding environment.

[0110] The wireless gateway device supports a variety of wireless communication protocols, including low-power Bluetooth (BLE), WIFI, ZigBee, and LoRa. For short-range communication, the gateway uses BLE, WIFI, or ZigBee protocols to exchange data with wearable devices and environmental parameter sensors. These protocols have low power consumption characteristics and are suitable for long-term operation of battery-powered devices. For scenarios that require long-distance coverage, the gateway can use the LoRa protocol to achieve wireless communication within a range of several kilometers.

[0111] To improve communication reliability, each wireless gateway device (referred to as "gateway") can be equipped with mesh network function. Gateways can communicate with each other to form a self-organizing network. When a gateway fails or the signal is weak, data can be transmitted through other gateways to ensure the stable operation of the system. At the same time, each wearable device and environmental parameter sensor can also establish connections with multiple gateways at the same time to achieve communication redundancy.

[0112] The wireless gateway device has local storage and computing capabilities. When the network connection is unstable, the gateway can temporarily store the data uploaded by the device and upload it to the cloud server in batches after the network is restored. The gateway can also pre-process and compress the data to reduce the network transmission load. In addition, the gateway also has a simple anomaly detection function that can quickly detect and report device failures or abnormal conditions.

[0113] Each wireless gateway device can be equipped with Ethernet and 4G / 5G communication modules, which can access the Internet via wired or wireless means to transmit data to the cloud server. The gateway will automatically select the best network connection method and switch to the wireless network when the wired network is unavailable to ensure the continuity of data transmission.

[0114] The system adopts a layered data transmission strategy. Wearable devices and environmental parameter sensors will choose different transmission methods according to the importance and urgency of the data. For ordinary monitoring data, batch transmission can be used to reduce the frequency of communication. For abnormal data or alarm information detected, it will be immediately reported to the cloud server through the gateway to ensure timely detection and handling of problems.

[0115] Optionally, in one embodiment, the system may also utilize image recognition technology to realize automatic annotation of posture data, so as to more efficiently train the first machine learning model.

[0116] Specifically, in this embodiment, the system also includes a set of video monitoring devices, such as a camera installed in the animal breeding environment. The implementation of this embodiment is as follows:

[0117] 1. Training of the third machine learning model:

[0118] First, prepare a large number of animal image datasets (for example, these images can be from different individuals, different postures and different breeding environments of pigs). Manually annotate the animal postures in these images (for example, standing, lying, lame, eating, sitting, etc.) to form a set of high-quality training datasets.

[0119] The labeled image data set is used to supervise the training of the third machine learning model, so that it can automatically recognize the posture of the animal based on a single frame image. Since the training is based on a large amount of manually labeled image data, the third machine learning model has a strong posture recognition ability and good robustness to complex environments. The third machine learning model can adopt a convolutional neural network structure, such as ResNet, MobileNet and other image recognition networks.

[0120] 2. Data collection and automatic annotation process:

[0121] In actual use, cameras are deployed in animal breeding environments to continuously capture videos of multiple animals wearing wearable devices.

[0122] At the same time, the wearable device worn on the animal (including accelerometers and optional gyroscopes, magnetometers and other sensors) will record the animal's status at a certain sampling frequency and upload the data to the cloud server wirelessly.

[0123] In the cloud server or edge computing unit, the video stream captured by the camera is decomposed into continuous image frames, and these image frames are input into the trained third machine learning model to automatically identify the animal posture corresponding to each time point. Since the identity of individual animals can be identified in the scene captured by the camera (for example, through specific identification, fence positioning, image recognition algorithm or other positioning means), the system can match the posture recognized by the image with the specific timestamp and sensor data of the specific wearable device (corresponding to the specific animal).

[0124] In other words, for each timestamp, the animal posture label obtained from the camera image can be matched one-to-one with the acceleration data uploaded by the wearable device at the same time (e.g., a sequence of acceleration vectors within a predetermined time period before the time). In this way, the acceleration sensor data (and optional gyroscope and magnetometer data) are automatically labeled using the posture recognition results from the third machine learning model without manual intervention.

[0125] 3. Train the second machine learning model:

[0126] Through the above-mentioned automatic labeling method, a large number of "posture-acceleration data" samples are generated quickly and efficiently. These automatically labeled data are input into the second machine learning model for training, so that the second machine learning model can predict the posture of the animal based on the data using acceleration (and optional gyroscope and magnetometer). Due to the large amount of training data and accurate labeling, the second machine learning model has stronger generalization ability and higher posture recognition accuracy for different environments and different animals.

[0127] 4. Push the trained second machine learning model to the animal’s wearable device.

[0128] The second machine learning model can be continuously optimized by collecting data using a camera and an accelerometer, and the optimized second machine learning model can be regularly sent to each wearable device, thereby continuously improving the accuracy of the second machine learning model in detecting animal postures.

[0129] This embodiment has the following effects:

[0130] 1. Reduce the cost and difficulty of manual labeling: The traditional method requires manual labeling of the posture corresponding to the acceleration sensor data, which is time-consuming and error-prone. By first training a third machine learning model to automatically label the image, and then realizing automatic labeling through the time synchronization of the video and the acceleration sensor data, the workload and cost of manual labeling are greatly reduced.

[0131] 2. Improve the consistency and quality of data annotation: The third machine learning model has been trained based on a large amount of annotated image data, and its posture recognition accuracy and consistency are higher than manual visual observation. Using this model to automatically label the acceleration sensor data can improve the quality and consistency of the training data, thereby improving the prediction performance of the final first machine learning model.

[0132] 3. Accelerate model iteration and optimization: After the automatic labeling process is implemented, when the system adds or updates acceleration sensor data, new high-quality training data can be quickly obtained without additional manual labeling. This enables the first machine learning model to iterate and optimize faster, and adapt more quickly to different scenarios or different groups of animals, thereby improving the flexibility and adaptability of the system.

[0133] Through this embodiment, an efficient and automated closed-loop process of training data generation and model optimization can be built, thereby improving the overall accuracy and reliability of the animal health monitoring system.

[0134] In each embodiment of the present application, the processor may be a central processing unit (CPU), a graphic processing unit (GPU), a digital signal processor (DSP), a microcontroller unit (MCU), a neural network processor (NPU), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), or other programmable logic devices. The aforementioned memory may be a random access memory (RAM), a flash memory, a hard disk, or a solid state drive.

[0135] It should be noted that, in the present application, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is any such actual relationship or order between these entities or operations. Moreover, the term "include", "comprise" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, article or equipment including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or equipment. In the absence of more restrictions, the elements defined by the sentence "include one" do not exclude the existence of other identical elements in the process, method, article or equipment including the elements. In the present application, if it is mentioned that a certain action is performed according to a certain element, it means at least the meaning of performing the action according to the element, which includes two situations: performing the action only according to the element and performing the action according to the element and other elements. Multiple, multiple, multiple, etc. expressions include 2, 2 times, 2 kinds and more than 2, more than 2 times, more than 2 kinds.

[0136] This specification includes combinations of the various embodiments described herein. Individual references to embodiments (e.g., "one embodiment" or "some embodiments" or "preferred embodiments"); however, these embodiments are not mutually exclusive unless indicated as mutually exclusive or it is clear to a person skilled in the art that they are mutually exclusive. It should be noted that the word "or" is used in this specification in a non-exclusive sense unless the context clearly indicates or requires otherwise.

[0137] All documents mentioned in this specification are considered to be included in the disclosure of this application as a whole, so that they can be used as a basis for modification when necessary. In addition, it should be understood that the above is only a preferred embodiment of this specification and is not intended to limit the scope of protection of this specification. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of one or more embodiments of this specification should be included in the scope of protection of one or more embodiments of this specification.

[0138] In some cases, the actions or steps described in the claims may be performed in a different order than in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or sequential order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

Claims

1. An animal health monitoring system, characterized in that: include: Wearable devices and cloud servers; The wearable device includes a processor, an acceleration sensor, a body temperature sensor and a wireless transceiver connected to the processor; the processor is configured to: determine the posture of the animal according to the acceleration data detected by the acceleration sensor, and obtain the body temperature of the animal from the body temperature sensor; Sending the posture-body temperature data pair to the cloud server via the wireless transceiver; The cloud server is configured to: input multiple groups of posture-body temperature data pairs from the wearable device sorted in chronological order into a pre-trained first machine learning model to obtain the animal health status output by the first machine learning model.

2. The animal health monitoring system according to claim 1, characterized in that: There are multiple wearable devices; The first machine learning model is trained based on posture-body temperature data pairs from multiple wearable devices.

3. The animal health monitoring system according to claim 1, characterized in that: The wearable device also includes a memory; the processor is configured to store the posture-body temperature data pairs in the memory, and send multiple groups of the posture-body temperature data pairs stored in the memory to the cloud server via the wireless transceiver at a predetermined time.

4. The animal health monitoring system according to claim 1, characterized in that: The processor is also configured to: determine the sleep duration according to the current posture of the animal, control the wearable device to enter a sleep state, wake up the wearable device after the sleep duration and perform the next detection, wherein at least two different postures correspond to different sleep durations.

5. The animal health monitoring system according to claim 1, characterized in that: The wearable device also includes a gyroscope and a magnetometer; The processor is further configured to determine the posture of the animal based on the data detected by the acceleration sensor, the gyroscope, and the magnetometer.

6. The animal health monitoring system according to claim 1, characterized in that: The processor uses a second machine learning model in the wearable device to determine the posture of the animal based on the acceleration data detected by the acceleration sensor.

7. The animal health monitoring system according to claim 5, characterized in that: The processor is also configured to dynamically adjust the sampling frequencies of the acceleration sensor, the gyroscope, the magnetometer, and the body temperature sensor according to the determined posture.

8. The animal health monitoring system according to claim 1, characterized in that: The system further comprises an environmental parameter sensor arranged in the animal breeding environment, for measuring the environmental parameter of the environment in which the animal is located; The environmental parameter sensor periodically uploads environmental parameters to the cloud server; The wearable device further includes a clock, and the processor is further configured to obtain time information for determining the posture and obtaining the body temperature from the clock, form a timestamp, and send the posture-body temperature data pair, the corresponding timestamp and the identifier of the wearable device to the cloud server via the wireless transceiver; The cloud server determines the environmental parameter sensor of the environment where the wearable device is located based on the identifier of the wearable device, and determines the environmental parameters corresponding to the posture-body temperature data pair based on the timestamp; inputs the posture-body temperature data pair and its corresponding environmental parameters into the first machine learning model to obtain the animal health status output by the first machine learning model.

9. The animal health monitoring system according to claim 8, characterized in that: The environmental parameters include one of the following or any combination thereof: ambient temperature, ambient humidity, ambient air pressure; The environmental parameter sensor includes one of the following or any combination thereof: A temperature sensor for measuring ambient temperature, a humidity sensor for measuring ambient humidity, and a pressure sensor for measuring ambient air pressure.

10. The animal health monitoring system according to claim 8, characterized in that: It also includes a wireless gateway device, which establishes wireless connections with multiple wearable devices and multiple environmental parameter sensors, and uploads data from the wearable devices and the environmental parameter sensors to the cloud server.

Images