CN116058298B

CN116058298B - Livestock behavior monitoring method and device

Info

Publication number: CN116058298B
Application number: CN202310202167.5A
Authority: CN
Inventors: 丁露雨; 李奇峰; 马为红; 蒋瑞祥; 姚春霞; 杨宝祝; 于沁杨; 余礼根; 高荣华
Original assignee: Research Center of Information Technology of Beijing Academy of Agriculture and Forestry Sciences
Current assignee: Research Center of Information Technology of Beijing Academy of Agriculture and Forestry Sciences
Priority date: 2023-03-06
Filing date: 2023-03-06
Publication date: 2023-09-12
Anticipated expiration: 2043-03-06
Also published as: CN116058298A

Abstract

The invention provides a livestock behavior monitoring method and device, belonging to the technical field of computers, wherein the method comprises the following steps: acquiring a plurality of pieces of first sensor data by continuously acquiring signals of a plurality of sensors; sliding calculation is carried out on the pieces of first sensor data based on a preset time window and a coverage rate of the preset time window, statistical data corresponding to each piece of first sensor data are obtained, parameter expansion is carried out on the pieces of first sensor data based on the statistical data corresponding to each piece of first sensor data, and a plurality of pieces of second sensor data are obtained; and acquiring a behavior classification result based on the plurality of pieces of second sensor data and the target classification model. Continuous signal acquisition, sliding calculation and parameter expansion are carried out on a plurality of sensors so as to fully obtain sensing signal characteristics of different behaviors, and expanded sensor data are input into a target classification model, so that a behavior classification result can be obtained, and multistage behaviors of livestock can be efficiently monitored.

Description

Livestock behavior monitoring method and device

Technical Field

The invention relates to the technical field of computers, in particular to a livestock behavior monitoring method and device.

Background

Livestock feeding behavior monitoring is of great importance for improving pasture management and for finding sick animals. Traditional behavior monitoring relies on manual direct observation, wastes time and labor, is inaccurate and affects normal grazing behavior of livestock. Subsequent developments have presented modern testing techniques based on wearable devices, monitoring livestock behaviour through devices such as foot rings or collars. In the related art, the method for monitoring the feeding behavior of the wearable livestock mainly comprises three categories, namely sound monitoring, pressure monitoring and acceleration monitoring.

In the related art, the method for monitoring the feeding behavior of the wearable livestock mainly comprises a single-class sensor, and only single-class behavior monitoring can be realized, but the feeding behavior of the livestock usually comprises a series of fine behaviors, for example, the feeding behavior of the cattle comprises secondary fine behaviors (multi-class behaviors) such as food coiling, chewing or swallowing, and the ruminant comprises secondary fine behaviors (multi-class behaviors) such as retching, chewing or swallowing. With the development of intelligent cultivation, the requirement of single feeding, rumination and other primary behavior monitoring cannot be met, and how to monitor the multistage behavior of livestock is a problem to be solved in the industry.

Disclosure of Invention

Aiming at the problems existing in the prior art, the embodiment of the invention provides a livestock behavior monitoring method and device.

In a first aspect, the present invention provides a method for monitoring livestock behaviour, comprising:

acquiring a plurality of pieces of first sensor data by continuously acquiring signals of a plurality of sensors;

sliding calculation is carried out on the plurality of pieces of first sensor data based on a preset time window and a preset time window coverage rate, statistical data corresponding to each piece of first sensor data are obtained, parameter expansion is carried out on the plurality of pieces of first sensor data based on the statistical data corresponding to each piece of first sensor data, and a plurality of pieces of second sensor data are obtained;

acquiring a behavior classification result based on the plurality of pieces of second sensor data and a target classification model, wherein the target classification model is used for predicting target behavior classifications corresponding to the second sensor data in a plurality of behavior classifications, and the plurality of behavior classifications comprise at least one multi-level behavior classification;

the plurality of sensors are sensors of different types, the sensors are used for monitoring behavior characteristics of livestock, the first sensor data are sensor signals obtained by collecting signals of the plurality of sensors at a first moment, and the first moment is a collection moment corresponding to the first sensor data.

Optionally, according to the method for monitoring livestock behavior provided by the present invention, the sliding calculation is performed on the plurality of pieces of first sensor data based on a preset time window and a coverage rate of the preset time window, and statistical data corresponding to each piece of first sensor data is obtained, including:

noise reduction and filtering processing is carried out on the plurality of pieces of first sensor data;

sliding calculation is carried out on the basis of the preset time window, the coverage rate of the preset time window and the plurality of pieces of first sensor data after noise reduction filtering processing, and statistical data corresponding to the pieces of first sensor data are obtained;

the statistical data includes one or more of the following parameters: mean, standard deviation, median, value range, complex vector, skewness, or warp.

Optionally, according to the method for monitoring livestock behavior provided by the present invention, the parameter expansion is performed on the plurality of pieces of first sensor data based on the statistical data corresponding to the pieces of first sensor data, to obtain a plurality of pieces of second sensor data, including:

acquiring a plurality of pieces of third sensor data by adding the statistical data to the first sensor data based on the statistical data corresponding to each piece of first sensor data;

Based on each parameter class in the plurality of pieces of third sensor data, carrying out relative importance degree analysis to obtain a relative importance degree mean value and a relative importance degree value corresponding to each parameter class;

filtering parameters belonging to target parameter classes in the plurality of pieces of third sensor data based on the relative importance average value and the relative importance value corresponding to each parameter class, and obtaining the plurality of pieces of second sensor data;

the target parameter class is a parameter class in which the relative importance value in each parameter class is smaller than or equal to the relative importance mean value.

Optionally, according to the livestock behavior monitoring method provided by the invention, the plurality of sensors include an inertial sensor and a pressure sensor;

the obtaining a plurality of pieces of first sensor data by continuously collecting signals of a plurality of sensors includes:

acquiring a first sensing signal of the inertial sensor and a second sensing signal of the pressure sensor, judging whether the first sensing signal is larger than or equal to an inertial sensing trigger threshold value, and judging whether the second sensing signal is larger than or equal to a pressure sensing trigger threshold value;

and if the first sensing signal is determined to be greater than or equal to the inertial sensing trigger threshold or the second sensing signal is determined to be greater than or equal to the pressure sensing trigger threshold, continuously acquiring the plurality of sensors based on a preset frequency to acquire the plurality of pieces of first sensor data.

Optionally, according to the method for monitoring livestock behavior provided by the present invention, before the behavior classification result is obtained based on the plurality of second sensor data and the target classification model, the method further includes:

judging whether a sensing signal missing condition exists in the continuous signal acquisition process based on the plurality of pieces of first sensor data, and acquiring a target sensor signal category, wherein the target sensor signal category is used for representing the sensing signal missing condition in the continuous signal acquisition process;

and determining a target classification model from a plurality of preset classification models based on the target sensor signal category, wherein the plurality of preset classification models correspond to different sensor signal categories.

Optionally, according to the livestock behavior monitoring method provided by the invention, in a case that the inertial sensing signal and the pressure sensing signal are not missing in the process of the continuous signal acquisition represented by the target sensor signal category, the target classification model is a first classification model in the plurality of preset classification models;

the obtaining a behavior classification result based on the plurality of second sensor data and the target classification model includes:

acquiring target behavior classifications corresponding to the second sensor data based on the second sensor data and the first classification model;

Determining a behavior classification result based on target behavior classifications corresponding to the pieces of second sensor data;

the first classification model is used for predicting target behavior classification corresponding to the second sensor data in a plurality of first behavior classifications;

the plurality of first behavioral classifications includes: standing feeding coil behavior classification, standing feeding chewing behavior classification, walking feeding coil chewing behavior classification, standing ruminant chewing behavior classification, lying ruminant chewing behavior classification, walking ruminant chewing behavior classification, drinking behavior classification, walking behavior classification, standing behavior classification and lying behavior classification.

Optionally, according to the livestock behavior monitoring method provided by the invention, in a case that the pressure sensing signal is absent and the inertial sensing signal is not absent in the process of the continuous signal acquisition represented by the target sensor signal category, the target classification model is a second classification model in the plurality of preset classification models;

acquiring target behavior classifications corresponding to the pieces of second sensor data based on the pieces of second sensor data and the second classification model;

the second classification model is used for predicting target behavior classification corresponding to the second sensor data in a plurality of second behavior classifications;

the plurality of second behavioral classifications includes: feeding behavior classification, head raising behavior classification, head lowering behavior classification, walking behavior classification, lying behavior classification and standing behavior classification.

Optionally, according to the livestock behavior monitoring method provided by the present invention, in a case that the inertial sensing signal is absent and the pressure sensing signal is not absent in the process of the continuous signal acquisition represented by the target sensor signal category, the target classification model is a third classification model among the plurality of preset classification models, and the third classification model includes a first differential unit, an attapulgite detection unit, a second differential unit and a classification unit;

performing differential processing on pressure sensing signals corresponding to each piece of second sensor data according to time sequence by the first differential unit to obtain a plurality of first differential values;

Performing concave-convex point detection on the first differential values through the concave-convex point detection unit to determine a plurality of zero points;

calculating distances between non-adjacent zero points for the plurality of zero points according to time sequence through the second differential unit, and obtaining a distance sequence, wherein the distance sequence comprises a plurality of distance values;

analyzing a threshold interval in which each distance value in the distance sequence is positioned according to a preset value interval configuration by the classification unit, and acquiring target behavior classifications corresponding to each piece of second sensor data;

the preset value interval configuration comprises a first value interval, a second value interval and a third value interval, wherein the upper limit value of the first value interval is equal to the lower limit value of the second value interval, the upper limit value of the second value interval is smaller than the lower limit value of the third value interval, the first value interval is used for representing the value range of the distance value corresponding to the feeding chewing behavior, the second value interval is used for representing the value range of the distance value corresponding to the ruminant chewing behavior, and the third value interval is used for representing the value range of the distance value corresponding to the drinking behavior.

Optionally, according to the livestock behavior monitoring method provided by the invention, the behavior classification result includes target behavior classifications corresponding to the second sensor data;

after the behavior classification result is obtained based on the plurality of pieces of second sensor data and the target classification model, the method further comprises:

correcting the target behavior classification corresponding to each piece of second sensor data based on the state transition probability matrix and the preset data sequence model, and obtaining corrected behavior classification corresponding to each piece of second sensor data;

the state transition probability matrix is used for representing state transition probabilities among the behavior classifications.

In a second aspect, the present invention also provides a livestock performance monitoring apparatus, comprising:

the first acquisition module is used for acquiring a plurality of pieces of first sensor data by carrying out continuous signal acquisition on a plurality of sensors;

the second acquisition module is used for carrying out sliding calculation on the plurality of pieces of first sensor data based on a preset time window and a preset time window coverage rate, acquiring statistical data corresponding to each piece of first sensor data, carrying out parameter expansion on the plurality of pieces of first sensor data based on the statistical data corresponding to each piece of first sensor data, and acquiring a plurality of pieces of second sensor data;

A third obtaining module, configured to obtain a behavior classification result based on the plurality of second sensor data and a target classification model, where the target classification model is configured to predict a target behavior classification corresponding to the second sensor data from a plurality of behavior classifications, and the plurality of behavior classifications includes at least one multi-level behavior classification;

According to the livestock behavior monitoring method and device, various behavior characteristics of livestock can be monitored through the plurality of sensors of different types, continuous signal acquisition can be conducted on the plurality of sensors, a plurality of pieces of first sensor data can be obtained, sliding calculation can be conducted on the plurality of pieces of first sensor data, statistical data corresponding to the plurality of pieces of first sensor data are obtained, so that sensing signal characteristics of different behaviors are fully obtained, parameter expansion can be conducted on the plurality of pieces of first sensor data based on the statistical data corresponding to the plurality of pieces of first sensor data, a plurality of pieces of second sensor data can be obtained, and then target behavior classification corresponding to the second sensor data can be predicted in the plurality of behavior classifications through the target classification model, so that behavior classification results can be obtained.

Drawings

In order to more clearly illustrate the invention or the technical solutions of the prior art, the following description will briefly explain the drawings used in the embodiments or the description of the prior art, and it is obvious that the drawings in the following description are some embodiments of the invention, and other drawings can be obtained according to the drawings without inventive effort for a person skilled in the art.

FIG. 1 is a schematic flow chart of a livestock behavior monitoring method according to the present invention;

FIG. 2 is a schematic view of a sensor installation provided by the present invention;

FIG. 3 is a second flow chart of the livestock behavior monitoring method according to the present invention;

FIG. 4 is a schematic diagram of a livestock performance monitoring apparatus according to the present invention;

fig. 5 is a schematic structural diagram of an electronic device provided by the present invention.

Reference numerals:

201: an inertial sensor; 202: a pressure sensor; 203: a wearing part; 204: an air bag.

Detailed Description

In order to facilitate a clearer understanding of various embodiments of the present invention, some relevant background knowledge is first presented as follows.

In the related art, the method for monitoring the feeding behavior of the wearable livestock mainly comprises three categories, namely sound monitoring, pressure monitoring and acceleration monitoring. For example, the livestock feeding sound signal collection device can be fixed on the forehead of the livestock, and the amplitude, time domain and frequency domain characteristics of the sound signal are combined with a preset judging function and a specific threshold value to identify the feeding behavior of the livestock. For example, the livestock chewing behavior monitoring device can be designed based on the pressure monitoring principle, and the threshold change of the first posture and the second posture is set by utilizing the pressure change generated by the reciprocating motion of the lower jaw so as to further identify the behavior characteristic change of the mouth of the livestock. For example, acceleration sensing methods are often used to detect the movement of livestock, and are more widely used in production (e.g., cows often monitor the amount of movement by acceleration sensing methods, and develop oestrus with changes in the amount of movement).

The inertial sensing method such as acceleration can better realize primary behavior recognition such as walking, lying and the like; pressure sensing has advantages in chew behavior recognition. The behavior recognition targets applicable to different sensing methods are different, and a single sensing method and a simple threshold model cannot meet the efficient and synchronous recognition of multi-level behaviors. For example, the pressure sensor can recognize the chewing behavior, but cannot determine whether the chewing behavior under the feeding condition or the chewing behavior under the ruminant condition occurs at this time, and it is necessary to make a determination in combination with other methods. Feeding chewing is of great significance to the estimation of feed intake, while the number of ruminant chewing is helpful to the evaluation of livestock health, and the application direction of chewing behaviors under different conditions is different.

In order to overcome the defects, the invention provides a livestock behavior monitoring method and device, which can input expanded sensor data into a target classification model by carrying out continuous signal acquisition, sliding calculation and parameter expansion on a plurality of sensors so as to acquire behavior classification results, and can realize efficient monitoring of multistage behaviors of livestock.

For the purpose of making the objects, technical solutions and advantages of the present invention more apparent, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings, and it is apparent that the described embodiments are some embodiments of the present invention, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.

Fig. 1 is a schematic flow chart of a livestock behavior monitoring method according to the present invention, and as shown in fig. 1, an execution subject of the livestock behavior monitoring method may be an electronic device. The method comprises the following steps:

step 101, acquiring a plurality of pieces of first sensor data by continuously acquiring signals of a plurality of sensors;

Specifically, in order to realize monitoring multistage behaviors of livestock, a plurality of different types of sensors can be arranged on the body of the livestock, various behavior characteristics of the livestock can be monitored through the plurality of different types of sensors, continuous signal acquisition is performed on the plurality of sensors, a plurality of pieces of first sensor data can be acquired, and the first sensor data are sensor signals acquired by performing signal acquisition on the plurality of sensors at corresponding moments.

Optionally, fig. 2 is a schematic view of the sensor installation provided by the present invention, as shown in fig. 2, where the plurality of sensors includes: an inertial sensor 201 and a pressure sensor 202, the inertial sensor 201 being for monitoring behavioral characteristics of the livestock neck, the pressure sensor 202 being for monitoring behavioral characteristics of the livestock mouth;

The inertial sensor 201 is fixed to the neck of the livestock by a wearing piece 203, the pressure sensor 202 is fixed to the levator nasolabial muscle of the livestock by the wearing piece 203, and an air bag 204 is provided between the pressure sensor 202 and the levator nasolabial muscle of the livestock.

Step 102, performing sliding calculation on the plurality of pieces of first sensor data based on a preset time window and a coverage rate of the preset time window, obtaining statistical data corresponding to each piece of first sensor data, and performing parameter expansion on the plurality of pieces of first sensor data based on the statistical data corresponding to each piece of first sensor data, so as to obtain a plurality of pieces of second sensor data.

Specifically, after the plurality of pieces of first sensor data are obtained, sliding calculation can be performed on the plurality of pieces of first sensor data, statistical data corresponding to each piece of first sensor data are obtained, so that sensing signal characteristics of different behaviors are obtained sufficiently (so that in the process of classifying and predicting a subsequent classification model, recognition accuracy of different behaviors is improved, misjudgment occurring during recognition of similar behaviors or mixed behaviors is reduced), and parameter expansion can be performed on the plurality of pieces of first sensor data based on the statistical data corresponding to each piece of first sensor data, so that the plurality of pieces of second sensor data are obtained.

Step 103, obtaining a behavior classification result based on the plurality of pieces of second sensor data and a target classification model, wherein the target classification model is used for predicting a target behavior classification corresponding to the second sensor data in a plurality of behavior classifications, and the plurality of behavior classifications comprise at least one multi-level behavior classification.

Specifically, after the plurality of pieces of second sensor data are acquired, a target behavior classification corresponding to the second sensor data can be predicted in a plurality of behavior classifications by a target classification model to acquire a behavior classification result, wherein the plurality of behavior classifications include at least one multi-level behavior classification, and the corresponding multi-level behavior classification can be identified under the condition that the domestic animal generates multi-level behaviors. Synchronous recognition of multi-level behaviors is realized, meanwhile, recognition accuracy of different behaviors is improved, and misjudgment occurring during recognition of similar behaviors or mixed behaviors is reduced.

According to the livestock behavior monitoring method, various behavior characteristics of livestock can be monitored through the plurality of sensors of different types, continuous signal acquisition can be carried out on the plurality of sensors, a plurality of pieces of first sensor data can be obtained, sliding calculation can be carried out on the plurality of pieces of first sensor data, statistical data corresponding to the plurality of pieces of first sensor data are obtained so as to fully obtain sensing signal characteristics of different behaviors, parameter expansion can be carried out on the plurality of pieces of first sensor data based on the statistical data corresponding to the plurality of pieces of first sensor data, a plurality of pieces of second sensor data can be obtained, and target behavior classification corresponding to the second sensor data can be predicted in the plurality of behavior classifications through a target classification model so as to obtain a behavior classification result.

Specifically, noise in the plurality of pieces of first sensor data can be eliminated through noise reduction and filtering processing, the plurality of pieces of first sensor data are smoothed, sliding calculation can be performed on the plurality of pieces of first sensor data after the noise reduction and filtering processing based on a preset time window and a preset time window coverage rate, and statistical data corresponding to each piece of first sensor data can be obtained.

It will be appreciated that by sliding calculations, sensor signal characteristics of different behaviors can be fully obtained, which can assist in improving behavior recognition accuracy.

Specifically, by adding the statistical data to the first sensor data, a plurality of pieces of third sensor data can be obtained, the third sensor data not only includes the original first sensor data but also includes corresponding statistical data, further, relative importance degree analysis can be performed on each parameter class in the plurality of pieces of third sensor data, a relative importance degree mean value and a relative importance degree value corresponding to each parameter class are obtained, the relative importance degree mean value is obtained by averaging the relative importance degree values corresponding to each parameter class, and further, parameters belonging to a target parameter class in the plurality of pieces of third sensor data can be filtered based on the relative importance degree mean value and the relative importance degree value corresponding to each parameter class, so as to obtain a plurality of pieces of second sensor data.

It can be understood that the data processing amount of the subsequent behavior classification can be reduced and the operation efficiency can be improved by filtering out the parameters belonging to the target parameter class in the plurality of pieces of third sensor data.

Specifically, by judging whether the first sensing signal is greater than or equal to the inertial sensing trigger threshold and judging whether the second sensing signal is greater than or equal to the pressure sensing trigger threshold, signal acquisition can be performed under the condition that the first sensing signal is greater than or equal to the inertial sensing trigger threshold or the second sensing signal is greater than or equal to the pressure sensing trigger threshold, and the data quantity can be reduced, the power consumption can be reduced and the endurance time can be prolonged due to the proper trigger threshold and the triggered data acquisition period.

Optionally, in the case that the first sensing signal is determined to be greater than or equal to the inertial sensing trigger threshold or the second sensing signal is determined to be less than the pressure sensing trigger threshold, signal acquisition may be performed on the inertial sensor and the pressure sensor during continuous signal acquisition.

Alternatively, in the case where the first sensing signal is determined to be less than the inertial sensing trigger threshold or the second sensing signal is determined to be greater than or equal to the pressure sensing trigger threshold, signal acquisition may be performed on the pressure sensor during continuous signal acquisition without performing signal acquisition on the inertial sensor.

Specifically, in the process of continuous signal acquisition of the plurality of sensors, a sensing signal missing condition may occur, for different sensing signal missing conditions, in order to enable the selected target classification model to adapt to the sensing signal to be processed, a target sensor signal class may be determined, where different sensor signal classes correspond to different sensing signal missing conditions, that is, different sensor signal classes correspond to different sensing signals to be processed, for example, in the case that the inertial sensing signal and the pressure sensing signal are not missing in the process of continuous signal acquisition represented by the target sensor signal class, the sensing signal to be processed includes the inertial sensing signal and the pressure sensing signal; for example, in the event that the pressure sensing signal is absent and the inertial sensing signal is not absent during the target sensor signal class characterizing continuous signal acquisition, the sensing signal to be processed includes the inertial sensing signal; for example, in the event that the inertial sensing signal is absent and the pressure sensing signal is not absent during the target sensor signal class characterizing continuous signal acquisition, the sensing signal to be processed includes the pressure sensing signal.

It will be appreciated that for different sensor signal absence conditions, in order to enable the selected target classification model to adapt to the sensor signal to be processed, the target classification model may be determined among a plurality of preset classification models based on the target sensor signal class, and implementing the selected target classification model may be adapted to process sensor data.

Specifically, in the case where the inertial sensor signal and the pressure sensor signal are not missing during the continuous signal acquisition represented by the target sensor signal class (i.e., in the case where the target sensor signal class is the first sensor signal class), the multi-level hybrid behavior may be identified, the inertial sensor signal and the pressure sensor signal are included in the processed sensor signal, and in order to enable the selected target classification model to adapt to the sensor signal to be processed, the first classification model may be determined as the target classification model among a plurality of preset classification models, the first classification model is used to predict the target behavior classification corresponding to the second sensor data among a plurality of first behavior classifications, the plurality of first behavior classifications include the multi-level hybrid behavior classification, so that the selected first behavior classification is applicable to processing the sensor data in this case (i.e., in the case where the target sensor signal class is the first sensor signal class).

It is understood that the feeding behavior of livestock includes feeding, rumination, wandering, lying, drinking, and excretion, among others. Under grazing conditions, livestock's grazing behavior is mainly composed of feeding and ruminating behaviors, and other behaviors depend on feeding behavior, and various mixed behaviors, such as feeding while walking, often occur. The multi-level mixing behavior can be identified by a first classification model.

It is understood that standing feeding coil behavior classification is a multi-stage mixed behavior classification that characterizes livestock to stand while feeding coils; standing feeding chewing behavior classification is a multi-level mixed behavior classification that characterizes livestock to chew while standing; the walking feeding coil chewing behavior classification is a multi-stage mixed behavior classification for representing that livestock walk and feed feeding coil chewing is performed at the same time; standing ruminant chewing behavior classification is a multi-level mixed behavior classification that characterizes livestock to stand while ruminant chewing; the lying ruminant chewing behavior classification is a multi-stage mixed behavior classification that characterizes a livestock to lie while being ruminant chewed; the walking ruminant chewing behavior classification is a multi-level mixed behavior classification that characterizes a livestock to walk while being ruminantd.

Therefore, by synchronously monitoring key fine actions such as food coiling, food intake chewing, ruminant chewing and the like, on one hand, the accuracy of behavior recognition can be improved, and on the other hand, production management such as food intake estimation, health evaluation and the like can be realized by integrating the occurrence condition of the fine actions.

Specifically, in the case where the pressure sensing signal is absent and the inertial sensing signal is not absent in the process of the continuous signal acquisition is represented by the target sensor signal class (i.e., in the case where the target sensor signal class is the second sensor signal class), the motion behavior may be identified, the inertial sensing signal included in the sensing signal to be processed, in order to enable the selected target classification model to adapt to the sensing signal to be processed, the second classification model may be determined as the target classification model in a plurality of preset classification models, the second classification model is used for predicting the target behavior classification corresponding to the second sensor data in a plurality of second behavior classifications, the plurality of second behavior classifications include the motion behavior classification, and implementing the selected second behavior classification may be applicable to processing the sensor data in this case (i.e., in the case where the target sensor signal class is the second sensor signal class).

Specifically, in the case where the inertial sensing signal is absent and the pressure sensing signal is not absent during the continuous signal acquisition represented by the target sensor signal class (i.e., in the case where the target sensor signal class is the third sensor signal class), the eating behavior may be identified, the pressure sensing signal included in the sensing signal to be processed, in order to enable the selected target classification model to adapt to the sensing signal to be processed, a third classification model may be determined as a target classification model among a plurality of preset classification models based on the target sensor signal class, the third classification model being used to predict a target behavior class corresponding to the second sensor data among a plurality of third behavior classifications, including a feeding chewing behavior class, a ruminant chewing behavior class, and a drinking behavior class, to implement the selected third behavior classification being applicable to processing the sensor data in this case (i.e., in the case where the target sensor signal class is the third sensor signal class).

The first differential unit is used for carrying out differential processing and concave-convex point detection on the pressure sensing signals corresponding to the second sensor data according to the time sequence, so that the zero point of the waveform corresponding to the time sequence characteristic data can be found, further, the distance values among the non-adjacent zero points can be calculated for the plurality of zero points according to the time sequence through the second differential unit, a distance sequence can be obtained, the distance sequence comprises a plurality of distance values, the distance values can represent the period duration of the action, and the greater the distance value is, the longer the period duration of the action is.

Generally, the period of feeding chew is shorter than the period of ruminant chew, which is shorter than the period of drinking. The method can set proper preset value interval configuration, the preset value interval configuration comprises a first value interval, a second value interval and a third value interval, the upper limit value of the first value interval is equal to the lower limit value of the second value interval, the upper limit value of the second value interval is smaller than the lower limit value of the third value interval, the first value interval is used for representing the value range of the distance value corresponding to the feeding chewing behavior, the second value interval is used for representing the value range of the distance value corresponding to the ruminant chewing behavior, and the third value interval is used for representing the value range of the distance value corresponding to the drinking behavior.

And analyzing the threshold value interval of each distance value in the distance sequence to judge the target behavior classification corresponding to each piece of second sensor data, thereby realizing the acquisition of multi-stage behavior classification under the condition.

Specifically, livestock behavior typically has a time series continuity, which can be corrected by the time series probability of behavior occurrence according to this feature. For example, if the recognized behavioral result is almost food intake mastication for a period of time, and the probability of occurrence of ruminant mastication is extremely low at some time in the middle of the period of time, if the recognized behavioral result is ruminant mastication, the large probability is misjudgment, and it can be corrected to food intake mastication.

Therefore, the target behavior classification corresponding to each piece of second sensor data can be corrected based on the state transition probability matrix and the preset data sequence model, and the accuracy of the behavior classification is improved.

Fig. 3 is a second flowchart of the livestock behavior monitoring method according to the present invention, as shown in fig. 3, the livestock behavior monitoring method includes steps 301 to 306:

in step 301, a plurality of pieces of first sensor data are acquired in case it is detected that the sensor signal reaches a threshold value for triggering signal acquisition.

Specifically, let the trigger threshold of the inertial sensor be the acceleration signal a1, the trigger threshold of the pressure sensor be the pressure signal a2, and when the signal of the inertial sensor is greater than or equal to a1 or the signal of the pressure sensor is greater than or equal to a2, n pieces of data with time sequence characteristics are continuously acquired at the frequency of β (Hz). And after the acquisition of the n pieces of data is finished, detecting whether the sensor signal reaches a threshold value for triggering signal acquisition or not again, entering a dormant state if the sensor signal does not reach the trigger threshold value, continuously acquiring the n pieces of data with time sequence characteristics again if the sensor signal reaches the trigger threshold value, and the like.

Step 302, determining and marking the sensor signal class (i.e., determining the target sensor signal class) based on the dimensions and the order of the acquired signals.

Specifically, let the dimension of the obtained detection signal be null be Na, give the complete occupation dimension of the multi-element sensing signal be Ta, and give a fixed occupation order to the Ta dimension parameter. When na=0, it is labeled as category one (i.e., the first sensor signal category), i.e., the detected signal includes inertial sensing, pressure sensing, and other equivalent complete sensor category signals. When Na is not equal to 0, the missing sensor signal category is judged by combining the bit sequence, the pressure sensing signal is marked as a category two (namely, the second sensor signal category) when the pressure sensing signal is missing, and the inertia sensing signal is marked as a category three (namely, the third sensor signal category) when the inertia sensing signal is missing. The marked different categories, during the behavior recognition process, the corresponding output objects are different.

Step 303, performing noise reduction filtering, feature extraction and parameter expansion processing on the plurality of pieces of first sensor data.

Specifically, wavelet filtering and high-low pass filtering are sequentially used to perform noise reduction and smoothing processing on the original signal. Setting n obtained sensor signals, each of which contains m parameters, and recording each data after noise reduction and smoothing processing, which comprises m-dimensional parameters as a data matrix. To better obtain the sensor signal characteristics, to +. >The second time length is a time windowFor the coverage of the time window, the Mean (Mean), standard Deviation (SD), median (Median), range (range), resultant vector and the like are respectively calculated for m inputs in a sliding way, and parameter expansion is carried out to obtain new calculated +.>Parameters are obtained by common methods such as filtration, wrapping or embedding>The relative importance of each of the individual calculation parameters. Calculate->Taking the mean value Im of the relative importance degree of the parameters, and taking the +.>The parameters are used as characteristic parameters to form n strips, each comprising +>A new data set of dimensional parameters (i.e. a plurality of second sensor data) is noted ∈>. The new data will be used for the next step of behavior recognition.

Step 304, determining a target classification model from a plurality of preset classification models based on the target sensor signal category, and performing behavior recognition based on the target classification model.

Optionally, in the case that the target sensor signal class is the first sensor signal class, the multi-stage mixed behavior may be identified, the processing of the sensing signals including the inertial sensing signal and the pressure sensing signal may be performed, and in order to enable the selected target classification model to adapt to the sensing signal to be processed, a first classification model may be determined as the target classification model among a plurality of preset classification models based on the target sensor signal class, the first classification model being used for predicting the target behavior classification corresponding to the second sensor data among a plurality of first behavior classifications, the plurality of first behavior classifications including the multi-stage mixed behavior classification.

Alternatively, the first classification model may be a pre-trained ensemble learning classification model, and the plurality of first behavioral classifications may include 10 behavioral classifications including standing feeding coil behavior classification, standing feeding chewing behavior classification, walking feeding coil chewing behavior classification, standing ruminant chewing behavior classification, lying ruminant chewing behavior classification, walking ruminant chewing behavior classification, drinking behavior classification, walking behavior classification, standing behavior classification, and lying behavior classification.

Optionally, in the case that the target sensor signal class is the second sensor signal class, the motion behavior may be identified, the to-be-processed sensor signal includes an inertial sensor signal, and in order to enable the selected target classification model to adapt to the to-be-processed sensor signal, a second classification model may be determined as a target classification model among a plurality of preset classification models based on the target sensor signal class, the second classification model being used for predicting a target behavior classification corresponding to the second sensor data among a plurality of second behavior classifications, the plurality of second behavior classifications including the motion behavior classification.

Optionally, the second classification model may be a pre-trained ensemble learning classification model, and the plurality of second behavior classifications may include 6 behavior classifications, such as a feeding behavior classification, a head raising behavior classification, a head lowering behavior classification, a walking behavior classification, a lying behavior classification, and a standing behavior classification.

Optionally, in the case that the target sensor signal class is the third sensor signal class, the dietary behavior may be identified, the to-be-processed sensor signal includes a pressure sensor signal, and in order to enable the selected target classification model to adapt to the to-be-processed sensor signal, a third classification model may be determined as a target classification model among a plurality of preset classification models based on the target sensor signal class, the third classification model being used for predicting a target behavior classification corresponding to the second sensor data among a plurality of third behavior classifications, the plurality of third behavior classifications including the dietary behavior classification.

Alternatively, the third classification model may be a pre-trained tree model or a pre-trained support vector machine model, and the plurality of third behavior classifications may include a rolling behavior classification, a chewing behavior classification, a drinking behavior classification, and the like.

Optionally, after the plurality of pieces of second sensor data output by the third classification model are acquired, performing differential processing according to time sequence based on pressure sensing signals corresponding to the pieces of fourth sensor data when the plurality of pieces of fourth sensor data exist in the plurality of pieces of second sensor data, and acquiring a plurality of first differential values, wherein the target behaviors corresponding to the fourth sensor data are classified into masticatory behavior classifications;

Performing concave-convex point detection on the first differential values to determine a plurality of non-adjacent zero points;

performing differential processing according to time sequence based on a plurality of non-adjacent zero points to obtain a plurality of second differential values;

updating the target behavior classification corresponding to each piece of fourth sensor data into the ruminant chewing behavior classification or the feeding chewing behavior classification based on a preset differential threshold value and a plurality of second differential values;

and determining a behavior classification result based on the target behavior classification corresponding to each piece of second sensor data.

For example, h pieces of data with time sequence characteristics (i.e. pressure sensing signals corresponding to the fourth pieces of sensor data) can be extracted, and a sequence can be formed according to the continuous sequence of sampling pointsThe first differentiating unit may perform the differentiating process using the following formula (1), by the following formula (1), for the +.>Pressure sensor signal->And->Pressure sensor signal->Performing differential processing calculation to obtain a first differential value +.>The concave-convex point detection unit may perform concave-convex point detection by the following formula (2), and determine a plurality of zero points by performing concave-convex point detection (L is a measurement error within a detection limit of the sensor) by the following formula (2): />

Further, two non-adjacent zero points can be detected with respect to the plurality of zero points in time series by the second differential unit (by To determine the zero point) to determine the distance sequence +.>The preset value interval configuration comprises a first value interval, a second value interval and a third value interval, wherein the lower limit value of the first value interval is a first threshold value, the upper limit value of the first value interval and the lower limit value of the second value interval are a second threshold value, the upper limit value of the second value interval is a third threshold value, the lower limit value of the third value interval is a fourth threshold value, and the fourth threshold value is larger than the third threshold value. By comparing the distance values with the first threshold value, the second threshold value, the third threshold value and the fourth threshold value, the threshold intervals where the distance values are located in the distance sequence can be analyzed, and the target behavior classification corresponding to each piece of second sensor data can be determined.

Step 305 corrects the identified behavior using the timing probabilities.

Specifically, for the above-mentioned plurality of behavior classifications, the total classification number of the plurality of behavior classifications may be Num, the jth behavior classification among the plurality of behavior classificationsGiven->The state transition probability sequence of (2) is +.>Wherein->Representation->Probability of the corresponding behavior being converted into the 1 st behavior class corresponding behavior, +.>Representation->Probability of the corresponding behavior being converted into the corresponding behavior of the 2 nd behavior class, and so on, ++ >Representation->Probability of the corresponding behavior being converted into a Num-th behavior classification corresponding behavior, wherein +.>Representation->The probability of the corresponding behavior being converted to other behavior (non-target behavior). And->The sum of all state transition probabilities in (a) is 1. Based on the state transition probability sequence of each behavior classification, a state transition probability matrix can be constructed, and the target behavior classification corresponding to each piece of second sensor data can be corrected by using the state transition probability matrix and a preset data sequence model (such as a Viterbi algorithm in a hidden Markov chain) to obtain corrected behavior classification corresponding to each piece of second sensor data.

And 306, conditioning output, short-term storage and transmission are performed based on the corrected behavior classification corresponding to each piece of second sensor data.

Specifically, the end result of conditioning the output includes two parts: part of the behavior recognition results are behavior recognition results (comprising corrected behavior classifications corresponding to the second sensor data) processed by the model, the behavior recognition results form a union, and the union is conditioned and output according to a time sequence; another part of the original data set X. The two parts of output data are stored for a short period in a front coverage mode by different files, and data calling and transmission under weak signals are convenient. When wired transmission can be adopted, each behavior identification result and an original data set X can be selected and transmitted according to the need; when wireless transmission is employed, only the respective behavior recognition results may be transmitted.

The livestock behavior monitoring device provided by the invention is described below, and the livestock behavior monitoring device described below and the livestock behavior monitoring method described above can be referred to correspondingly.

Fig. 4 is a schematic structural diagram of a livestock behavior monitoring device provided by the present invention, as shown in fig. 4, the livestock behavior monitoring device includes: a first acquisition module 401, a second acquisition module 402, and a third acquisition module 403, wherein:

a first acquisition module 401, configured to acquire a plurality of pieces of first sensor data by performing continuous signal acquisition on a plurality of sensors;

the second obtaining module 402 is configured to perform sliding calculation on the plurality of pieces of first sensor data based on a preset time window and a coverage rate of the preset time window, obtain statistical data corresponding to each piece of first sensor data, and perform parameter expansion on the plurality of pieces of first sensor data based on the statistical data corresponding to each piece of first sensor data, so as to obtain a plurality of pieces of second sensor data;

a third obtaining module 403, configured to obtain a behavior classification result based on the plurality of second sensor data and a target classification model, where the target classification model is configured to predict a target behavior classification corresponding to the second sensor data from a plurality of behavior classifications, and the plurality of behavior classifications includes at least one multi-level behavior classification;

Fig. 5 is a schematic structural diagram of an electronic device according to the present invention, and as shown in fig. 5, the electronic device may include: processor 510, communication interface (Communications Interface) 520, memory 530, and communication bus 540, wherein processor 510, communication interface 520, memory 530 complete communication with each other through communication bus 540. Processor 510 may invoke logic instructions in memory 530 to perform a livestock performance monitoring method comprising:

Further, the logic instructions in the memory 530 described above may be implemented in the form of software functional units and may be stored in a computer-readable storage medium when sold or used as a stand-alone product. Based on this understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art or in a part of the technical solution, in the form of a software product stored in a storage medium, comprising several instructions for causing a computer device (which may be a personal computer, a server, a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a magnetic disk, or an optical disk, or other various media capable of storing program codes.

In another aspect, the present invention also provides a computer program product comprising a computer program, the computer program being storable on a non-transitory computer readable storage medium, the computer program, when executed by a processor, being capable of performing the livestock performance monitoring method provided by the above methods, the method comprising:

In yet another aspect, the present invention also provides a non-transitory computer readable storage medium having stored thereon a computer program which when executed by a processor is implemented to perform the livestock performance monitoring method provided by the above methods, the method comprising:

The apparatus embodiments described above are merely illustrative, wherein the elements illustrated as separate elements may or may not be physically separate, and the elements shown as elements may or may not be physical elements, may be located in one place, or may be distributed over a plurality of network elements. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art will understand and implement the present invention without undue burden.

From the above description of the embodiments, it will be apparent to those skilled in the art that the embodiments may be implemented by means of software plus necessary general hardware platforms, or of course may be implemented by means of hardware. Based on this understanding, the foregoing technical solution may be embodied essentially or in a part contributing to the prior art in the form of a software product, which may be stored in a computer readable storage medium, such as ROM/RAM, a magnetic disk, an optical disk, etc., including several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the method described in the respective embodiments or some parts of the embodiments.

Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present invention, and are not limiting; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method of monitoring livestock performance, comprising:

The plurality of sensors are sensors of different types, the sensors are used for monitoring behavior characteristics of livestock, the first sensor data are sensor signals obtained by collecting signals of the plurality of sensors at a first moment, the first moment is a collecting moment corresponding to the first sensor data, the plurality of sensors comprise inertial sensors and pressure sensors, the inertial sensors are used for monitoring the behavior characteristics of the necks of the livestock, and the pressure sensors are used for monitoring the behavior characteristics of the necks of the livestock;

before the behavior classification result is obtained based on the plurality of pieces of second sensor data and the target classification model, the method further comprises:

determining a target classification model in a plurality of preset classification models based on the target sensor signal category, wherein the plurality of preset classification models correspond to different sensor signal categories;

The condition of absence of a sensor signal includes any one of the following conditions:

the inertial sensing signal and the pressure sensing signal are not missing;

or, the pressure sensing signal is absent and the inertial sensing signal is not absent;

or, the inertial sensing signal is absent and the pressure sensing signal is not absent;

under the condition that the inertial sensing signals and the pressure sensing signals are not missing in the process of representing the continuous signal acquisition by the target sensor signal category, the target classification model is a first classification model in the plurality of preset classification models;

the plurality of first behavioral classifications includes: standing feeding coil behavior classification, standing feeding chewing behavior classification, walking feeding coil chewing behavior classification, standing ruminant chewing behavior classification, lying ruminant chewing behavior classification, walking ruminant chewing behavior classification, drinking behavior classification, walking behavior classification, standing behavior classification and lying behavior classification;

Under the condition that the pressure sensing signal is absent and the inertial sensing signal is not absent in the process of representing the continuous signal acquisition by the target sensor signal category, the target classification model is a second classification model in the plurality of preset classification models;

the plurality of second behavioral classifications includes: feeding behavior classification, head raising behavior classification, head lowering behavior classification, walking behavior classification, lying behavior classification and standing behavior classification;

under the condition that inertial sensing signals are absent and pressure sensing signals are not absent in the continuous signal acquisition process of the target sensor signal category representation, the target classification model is a third classification model in the plurality of preset classification models, and the third classification model comprises a first differential unit, an attapulgite detection unit, a second differential unit and a classification unit;

2. The method for monitoring livestock behavior according to claim 1, wherein the sliding calculation is performed on the plurality of pieces of first sensor data based on a preset time window and a preset time window coverage rate, and statistical data corresponding to each piece of first sensor data is obtained, including:

3. The method for monitoring livestock behaviors according to claim 2, wherein the parameter expanding the plurality of pieces of first sensor data based on the statistical data corresponding to the pieces of first sensor data to obtain the plurality of pieces of second sensor data includes:

4. The method of claim 1, wherein the acquiring a plurality of first sensor data by performing continuous signal acquisition on a plurality of sensors comprises:

5. The livestock performance monitoring method of any of claims 1-4, wherein the performance classification result comprises a target performance classification corresponding to each piece of second sensor data;

6. A livestock performance monitoring apparatus, comprising:

the inertial sensing signal and the pressure sensing signal are not missing;