Task early warning method, device, system and equipment for animal physiological state

Abstract

Embodiments of the present application provide a task early warning method, device, system and equipment for animal physiological state, which relates to the technical field of data processing. The method comprises: in response to meeting a condition for triggering task analysis, determining, for each monitoring period in a plurality of continuous monitoring periods, each multi-dimensional feature of a target animal in the monitoring period; for each monitoring period, constructing a target time sequence feature sequence corresponding to the monitoring period based on each multi-dimensional feature in the monitoring period; and inputting each target time sequence feature sequence into a prediction model to obtain a prediction result of whether a task set for a specified physiological state of the target animal needs to be performed in a candidate task execution period to which a current time belongs. It can be seen that the present solution can effectively warn the task set for the specified physiological state of the animal.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a task early warning method, device, system and equipment for animal physiological states. Background Technology

[0002] Farm workers need to constantly monitor the physiological state of their animals so they can promptly implement appropriate preparatory tasks and provide artificial intervention when animals are in estrus, mating season, or ill. For example, when an animal is determined to be in mating season, preparations for mating should be initiated immediately; similarly, when an animal is determined to be ill, preparations for drug treatment should be initiated promptly. Typically, farms house a large number of animals, making it impossible for workers to respond to the physiological state of every single animal in a timely manner.

[0003] Therefore, how to effectively provide early warnings for tasks set for specific physiological states of animals, so as to assist farmers in responding to these states in a timely manner, has become a pressing technical problem that needs to be solved. Summary of the Invention

[0004] The purpose of this application is to provide a task early warning method, apparatus, system, and device for animal physiological states, so as to achieve effective early warning for tasks set for specified physiological states of animals. The specific technical solution is as follows:

[0005] In a first aspect, embodiments of this application provide a task early warning method for animal physiological states, including:

[0006] In response to the fulfillment of the conditions for triggering task analysis, for each of a series of consecutive monitoring periods, various multidimensional features of the target animal within that monitoring period are determined; wherein, each multidimensional feature is determined based on radar data collected by the target radar of the target animal within that monitoring period, and each multidimensional feature is used to describe the animal behavior of the target animal during a target statistical time period; the series of monitoring periods includes the monitoring period closest to the current time.

[0007] For each monitoring period, a target time-series feature sequence corresponding to that monitoring period is constructed based on the various multi-dimensional features within that monitoring period.

[0008] Each target temporal feature sequence is input into the prediction model to obtain a prediction result of whether a task set for a specified physiological state of the target animal needs to be executed within the candidate task execution period to which the current time belongs. The prediction model is trained based on training samples and corresponding ground truth values. The training samples include: sample temporal feature sequences of the sample animal corresponding to multiple consecutive sample periods determined based on radar data collected from the sample animal, wherein the multiple sample periods include a monitoring period with the closest sample time. The ground truth value is the annotation result of whether a task set for a specified physiological state of the sample animal needs to be executed within the candidate task execution period to which the sample time belongs.

[0009] Optionally, in one implementation, the generation of the various multidimensional features of the target animal within any monitoring period includes:

[0010] Acquire radar data collected by the target radar on the target animal during the monitoring period;

[0011] The acquired radar data is divided into multiple data segments according to the target segmentation duration; wherein the target segmentation duration is not greater than the duration of the target statistical period.

[0012] Based on the multiple data segments, a first analysis result and a second analysis result are analyzed for each data segment; wherein, the first analysis result is the behavioral classification result of the target animal within the collection time range corresponding to a data segment, and the second analysis result is the vibration intensity generated when the target animal performs an animal behavior within the collection time range corresponding to a data segment.

[0013] Based on the first and second analysis results of each data segment, multidimensional features of the target animal during the monitoring period are generated.

[0014] Optionally, in one implementation, the duration of the target statistical period is a specified multiple of the target segmentation duration;

[0015] Based on the first and second analysis results for each data segment, the multidimensional characteristics of the target animal during the monitoring period are generated, including:

[0016] For each target statistical period within the monitoring cycle, based on the first and second analysis results of each data segment corresponding to the target statistical period, the first cumulative duration of vibration intensity greater than the preset intensity and the second cumulative duration of the same animal behavior produced by the target animal are analyzed.

[0017] Based on the first cumulative duration and the second cumulative duration obtained from the analysis, as well as the first and second analysis results of each data segment corresponding to the target statistical period, the multidimensional characteristics corresponding to the target statistical period within the monitoring cycle are determined.

[0018] In this context, each data segment corresponding to the target statistical period is a data segment whose corresponding collection time range is within the target statistical period.

[0019] Optionally, in one implementation, the prediction model includes a convolutional network layer, an attention mechanism layer, and a long short-term memory layer connected in sequence;

[0020] The convolutional network layer in the prediction model is used to: extract features about local features from each target temporal feature sequence to obtain the first temporal feature sequence corresponding to the target temporal feature sequence;

[0021] The attention mechanism layer in the prediction model is used to: extract features about global time-series features for each first time-series feature sequence, and obtain the second time-series feature sequence corresponding to each first time-series feature sequence;

[0022] The long short-term memory layer in the prediction model is used to: generate a temporal change relationship of the target animal's physiological state before entering a specified physiological state based on each second temporal feature sequence; and analyze, based on the obtained temporal change relationship, whether the task for the specified physiological state of the target animal needs to be executed within the candidate task execution period to which the current time belongs.

[0023] Optionally, in one implementation, the training process of the prediction model includes:

[0024] The training samples and corresponding ground truth values ​​are input into the prediction model to be trained to obtain a prediction result of whether the task set by the specified physiological state of the sample animal needs to be executed within the candidate task execution period to which the sample time belongs.

[0025] Based on the prediction results and the true values, the model loss value of the prediction model to be trained is calculated, and in response to the model loss value being greater than a predetermined loss value, the model parameters of the prediction model to be trained are adjusted.

[0026] Optionally, in one implementation, the target animal is a breeding pig, and the specified physiological state is a mating state.

[0027] Optionally, in one implementation, the convolutional network layer is a one-dimensional convolutional network layer; the long short-term memory layer is an LSTM model.

[0028] Secondly, embodiments of this application provide a task early warning device for animal physiological states, including:

[0029] A determination module is configured to, in response to the fulfillment of conditions for triggering task analysis, determine various multidimensional features of the target animal within each of a series of consecutive monitoring periods; wherein each multidimensional feature is determined based on radar data collected from the target animal by the target radar within the monitoring period, and each multidimensional feature describes the animal behavior of the target animal during a target statistical time period; the series of monitoring periods includes the monitoring period closest to the current time.

[0030] The construction module is used to construct the target time-series feature sequence corresponding to each monitoring period based on the multi-dimensional features within that monitoring period.

[0031] The prediction module is used to input the temporal feature sequences of each target into the prediction model to obtain a prediction result of whether the task set for the specified physiological state of the target animal needs to be executed within the candidate task execution period to which the current time belongs. The prediction model is trained based on training samples and corresponding ground truth values. The training samples include: sample temporal feature sequences of the sample animal corresponding to multiple consecutive sample periods determined according to radar data collected from the sample animal, wherein the multiple sample periods include a monitoring period with the closest sample time. The ground truth values ​​are: the annotation results of whether the task set for the specified physiological state of the sample animal needs to be executed within the candidate task execution period to which the sample time belongs.

[0032] Optionally, in one implementation, the generation of various multidimensional features of the target animal within any monitoring period is performed through the following modules:

[0033] The acquisition module is used to acquire radar data collected by the target radar on the target animal during the monitoring period.

[0034] The segmentation module is used to segment the acquired radar data into multiple data segments according to the target segmentation duration; wherein the target segmentation duration is not greater than the duration of the target statistical period.

[0035] The analysis module is used to analyze a first analysis result and a second analysis result for each of the multiple data segments; wherein, the first analysis result is the behavioral classification result of the target animal within the collection time range corresponding to a data segment, and the second analysis result is the vibration intensity generated when the target animal performs an animal behavior within the collection time range corresponding to a data segment.

[0036] The generation module is used to generate various multidimensional features of the target animal within the monitoring period based on the first and second analysis results of each data segment.

[0037] Optionally, in one implementation, the duration of the target statistical period is a specified multiple of the target segmentation duration;

[0038] The generation module is specifically used for:

[0039] For each target statistical period within the monitoring cycle, based on the first and second analysis results of each data segment corresponding to the target statistical period, the first cumulative duration of vibration intensity greater than the preset intensity and the second cumulative duration of the same animal behavior produced by the target animal are analyzed.

[0040] Based on the first cumulative duration and the second cumulative duration obtained from the analysis, as well as the first and second analysis results of each data segment corresponding to the target statistical period, the multidimensional characteristics corresponding to the target statistical period within the monitoring cycle are determined.

[0041] In this context, each data segment corresponding to the target statistical period is a data segment whose corresponding collection time range is within the target statistical period.

[0042] Optionally, in one implementation, the prediction model includes a convolutional network layer, an attention mechanism layer, and a long short-term memory layer connected in sequence;

[0043] The convolutional network layer in the prediction model is used to: extract features about local features from each target temporal feature sequence to obtain the first temporal feature sequence corresponding to the target temporal feature sequence;

[0044] The attention mechanism layer in the prediction model is used to: extract features about global time-series features for each first time-series feature sequence, and obtain the second time-series feature sequence corresponding to each first time-series feature sequence;

[0045] The long short-term memory layer in the prediction model is used to: generate a temporal change relationship of the target animal's physiological state before entering a specified physiological state based on each second temporal feature sequence; and analyze, based on the obtained temporal change relationship, whether the task for the specified physiological state of the target animal needs to be executed within the candidate task execution period to which the current time belongs.

[0046] Optionally, in one implementation, the training process of the prediction model includes:

[0047] The training samples and corresponding ground truth values ​​are input into the prediction model to be trained to obtain a prediction result of whether the task set by the specified physiological state of the sample animal needs to be executed within the candidate task execution period to which the sample time belongs.

[0048] Based on the prediction results and the true values, the model loss value of the prediction model to be trained is calculated, and in response to the model loss value being greater than a predetermined loss value, the model parameters of the prediction model to be trained are adjusted.

[0049] Optionally, in one implementation, the target animal is a breeding pig, and the specified physiological state is a mating state.

[0050] Optionally, in one implementation, the convolutional network layer is a one-dimensional convolutional network layer; the long short-term memory layer is an LSTM model.

[0051] Thirdly, embodiments of this application provide a task early warning system for animal physiological states, including:

[0052] Target radar, used to collect radar data on target animals;

[0053] An early warning device is used to perform the task early warning method for animal physiological states provided in the first aspect above.

[0054] Fourthly, embodiments of this application provide an electronic device, including:

[0055] Memory, used to store computer programs;

[0056] The processor, when executing a program stored in memory, implements the task warning method for animal physiological states provided in the first aspect.

[0057] Fifthly, embodiments of this application provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the task warning method for animal physiological states provided in the first aspect.

[0058] This application also provides a computer program product containing instructions that, when run on a computer, enable the computer to execute the task warning method for animal physiological states provided in the first aspect.

[0059] Beneficial effects of the embodiments in this application:

[0060] As can be seen from the above, in the task early warning method for animal physiological states provided in this application embodiment, a prediction model is pre-trained based on training samples and corresponding ground truth values ​​for analysis of task early warning for animal physiological states; furthermore, this application sets a triggering condition for task analysis, so that in response to the satisfaction of the condition, for each monitoring cycle in a series of consecutive monitoring cycles, the multidimensional features of the target animal within that monitoring cycle are determined, that is, based on the radar data collected by the target radar for the target animal in each monitoring cycle, the multidimensional features within that monitoring cycle are determined, and each obtained multidimensional feature is used to describe the target animal in a target statistical period. The model predicts animal behavior within multiple monitoring periods, including the monitoring period closest to the current time. Then, for each monitoring period, a target temporal feature sequence is constructed based on the multidimensional features within that period. This sequence reflects the changing relationships of animal behavior in the target animal within the corresponding monitoring period. Finally, by analyzing the target temporal feature sequences across multiple monitoring periods using the aforementioned prediction model, the model clarifies the changing relationships of animal behavior in the target animal across consecutive monitoring periods. This determines whether a task set for a specific physiological state of the target animal needs to be executed within the candidate task execution period of the current time.

[0061] As can be seen, this scheme, by combining the multidimensional features determined for the target animal and the behavioral changes represented by the constructed time-series feature sequences of each target, provides effective data support for the analysis of subsequent prediction models, thereby improving the accuracy of the prediction results. Furthermore, the pre-trained prediction model can analyze whether a task set for a specific physiological state of the target animal needs to be executed based on the behavioral changes over multiple monitoring periods. Therefore, this scheme can effectively provide early warnings for tasks set for a specific physiological state of the target animal, assisting farmers in responding promptly to these changes. Attached Figure Description

[0062] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other embodiments can be obtained based on these drawings.

[0063] Figure 1 A flowchart illustrating a task early warning method for animal physiological states provided in an embodiment of this application;

[0064] Figure 2 A flowchart illustrating a method for generating various multidimensional features as provided in an embodiment of this application;

[0065] Figure 3 A flowchart illustrating another method for generating various multidimensional features provided in this application embodiment;

[0066] Figure 4 A flowchart illustrating a specific example provided in this application embodiment;

[0067] Figure 5 This application provides a schematic diagram of the structure of a task early warning system for animal physiological states, as shown in the embodiments of the present application.

[0068] Figure 6 This is a schematic diagram of the structure of a task early warning device for animal physiological state provided in an embodiment of this application;

[0069] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0070] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art based on this application are within the scope of protection of this application.

[0071] To address the aforementioned technical problems, embodiments of this application provide a method, apparatus, system, and device for task early warning based on animal physiological states. This method is applicable to various application scenarios where early warnings are issued for tasks set based on animal physiological states. For example, a veterinary hospital may issue early warnings for tasks set based on the physiological states of pet cats; another example is a livestock farm issuing early warnings for tasks set based on the physiological states of breeding pigs. Furthermore, this method can be applied to various electronic devices such as laptops and desktop computers (hereinafter referred to as electronic devices). Therefore, embodiments of this application do not limit the application scenarios or the executing entity of this method.

[0072] This application provides a task early warning method for animal physiological states, which may include the following steps:

[0073] In response to the fulfillment of the conditions for triggering task analysis, for each of a series of consecutive monitoring periods, various multidimensional features of the target animal within that monitoring period are determined; wherein, each multidimensional feature is determined based on radar data collected by the target radar of the target animal within that monitoring period, and each multidimensional feature is used to describe the animal behavior of the target animal during a target statistical time period; the series of monitoring periods includes the monitoring period closest to the current time.

[0074] For each monitoring period, a target time-series feature sequence corresponding to that monitoring period is constructed based on the various multi-dimensional features within that monitoring period.

[0075] Each target temporal feature sequence is input into the prediction model to obtain a prediction result of whether a task set for a specified physiological state of the target animal needs to be executed within the candidate task execution period to which the current time belongs. The prediction model is trained based on training samples and corresponding ground truth values. The training samples include: sample temporal feature sequences of the sample animal corresponding to multiple consecutive sample periods determined based on radar data collected from the sample animal, wherein the multiple sample periods include a monitoring period with the closest sample time. The ground truth value is the annotation result of whether a task set for a specified physiological state of the sample animal needs to be executed within the candidate task execution period to which the sample time belongs.

[0076] As can be seen from the above, in the task early warning method for animal physiological states provided in this application embodiment, a prediction model is pre-trained based on training samples and corresponding ground truth values ​​for analysis of task early warning for animal physiological states; furthermore, this application sets a triggering condition for task analysis, so that in response to the satisfaction of the condition, for each monitoring cycle in a series of consecutive monitoring cycles, the multidimensional features of the target animal within that monitoring cycle are determined, that is, based on the radar data collected by the target radar for the target animal in each monitoring cycle, the multidimensional features within that monitoring cycle are determined, and each obtained multidimensional feature is used to describe the target animal in a target statistical period. The model predicts animal behavior within multiple monitoring periods, including the monitoring period closest to the current time. Then, for each monitoring period, a target temporal feature sequence is constructed based on the multidimensional features within that period. This sequence reflects the changing relationships of animal behavior in the target animal within the corresponding monitoring period. Finally, by analyzing the target temporal feature sequences across multiple monitoring periods using the aforementioned prediction model, the model clarifies the changing relationships of animal behavior in the target animal across consecutive monitoring periods. This determines whether a task set for a specific physiological state of the target animal needs to be executed within the candidate task execution period of the current time.

[0077] As can be seen, this scheme, by combining the multidimensional features determined for the target animal and the behavioral changes represented by the constructed time-series feature sequences of each target, provides effective data support for the analysis of subsequent prediction models, thereby improving the accuracy of the prediction results. Furthermore, the pre-trained prediction model can analyze whether a task set for a specific physiological state of the target animal needs to be executed based on the behavioral changes over multiple monitoring periods. Therefore, this scheme can effectively provide early warnings for tasks set for a specific physiological state of the target animal, assisting farmers in responding promptly to these changes.

[0078] The following description, in conjunction with the accompanying drawings, details a task early warning method for animal physiological states provided by an embodiment of this application.

[0079] Figure 1 This application provides a flowchart illustrating a task warning method for animal physiological states, comprising the following steps S101-S103:

[0080] S101: In response to the conditions for triggering task analysis being met, for each of multiple consecutive monitoring periods, determine the various multidimensional features of the target animal within that monitoring period.

[0081] The various multidimensional features are determined based on radar data collected from the target animal by the target radar within the monitoring period, and each multidimensional feature is used to describe the animal behavior of the target animal during a target statistical period; the monitoring period includes the monitoring period closest to the current time.

[0082] In this application, considering the advantages of radar in data acquisition, such as being unaffected by light, weather, or obstructions, radar data collected by the target radar from the target animal is used as the basis for analysis when monitoring animal behavior. For example, the aforementioned radar data can be multi-channel time-series data from a millimeter-wave radar.

[0083] It should be noted that the collected radar data is obtained continuously from the target radar. To facilitate the analysis of the temporal correlation between the animal behavior and the corresponding specified physiological state, radar data within a monitoring cycle is used as a unit of analysis. For example, at the end of each monitoring cycle, based on the radar data collected by the target radar for the target animal within that monitoring cycle, multidimensional features describing the animal behavior under a specific statistical period are determined, resulting in various multidimensional features of the target animal within that monitoring cycle. Thus, upon meeting the conditions for triggering task analysis, the multidimensional features from the monitoring cycles closest to the current time can be directly retrieved, eliminating the need to re-analyze radar data collected from multiple monitoring cycles each time the above conditions are met, thereby improving the analytical efficiency of the task warning method.

[0084] The monitoring period can be 24 hours (i.e., one day), or 8 hours, etc., and each monitoring period can include multiple target statistical time periods. For example, if the monitoring period is 24 hours, the target statistical time periods included in the monitoring period can be 1 hour, 2 hours, 5 hours, etc. In other words, the duration of a monitoring period can be understood as a predetermined multiple of the duration of a target statistical time period within that monitoring period.

[0085] The multidimensional features of animal behavior under each target statistical period can be understood as: features that describe the same animal behavior under the same target statistical period from different dimensions. In other words, the multidimensional features are used to characterize the same animal behavior from different dimensions, and different multidimensional features are used to describe animal behavior under different target statistical periods.

[0086] Furthermore, the specific implementation process for generating various multidimensional features of the target animal within any monitoring period is detailed in steps S201-S204 below, and will not be elaborated here.

[0087] Optionally, the conditions for triggering task analysis are as follows: The task analysis process is triggered when the current time is the end time of the most recent monitoring period (e.g., a monitoring period lasts 24 hours); or, the task analysis process is triggered when the current time is a predetermined time (e.g., 7 AM daily). In this case, the radar data to be analyzed can be radar data from the previous day (i.e., radar data collected between 7 AM today and 7 AM the previous day), or, at 7 AM, radar data collected before 6 AM within the past 24 hours. However, it should be noted that the multidimensional features of each of the multiple consecutive monitoring periods to be analyzed must include the monitoring period most recent to the current time to improve the real-time performance of the obtained prediction results.

[0088] To improve the targeting of animal behavior monitoring, for example, the target animals are raised in a one-animal-one-pen manner. Accordingly, the aforementioned target radar is installed by means of suspension at a designated height directly above the target pen where the target animal belongs.

[0089] S102: For each monitoring period, construct the target time-series feature sequence corresponding to that monitoring period based on the various multi-dimensional features within that monitoring period.

[0090] In this application, for each monitoring period, the various multidimensional features within the monitoring period are spliced ​​together in the order of the target statistical time period to which each multidimensional feature belongs, so as to construct the target time series feature sequence corresponding to the monitoring period.

[0091] It should be noted that each element in the constructed target time-series feature sequence represents a feature vector of multidimensional features under a target statistical period.

[0092] S103: Input the temporal feature sequences of each target into the prediction model to obtain the prediction result of whether the task set for the specified physiological state of the target animal needs to be executed within the candidate task execution period of the current time.

[0093] The prediction model is trained based on training samples and corresponding ground truth values. The training samples include: sample time sequence features of sample animals corresponding to multiple consecutive sample periods determined by radar data collected from the sample animals. Among the multiple sample periods, there is a monitoring period with the closest sample time. The ground truth values ​​are: the annotation results of whether the task set for the specified physiological state of the sample animal should be executed for the candidate task execution period to which the sample time belongs.

[0094] In this application, a prediction model is pre-trained based on training samples and corresponding ground truth values. The specific training process is as follows: steps A1-A2, which will not be elaborated here.

[0095] Each constructed target time-series feature sequence can reflect the changes in animal behavior of the target animal within the corresponding monitoring period. However, a single target time-series feature sequence can provide limited effective data and may even contain useless interfering data. Therefore, the constructed target time-series feature sequences can be simultaneously input into the prediction model to provide a data foundation for subsequent analysis. By analyzing the target time-series feature sequences of multiple monitoring periods through the prediction model, the changes in animal behavior of the target animal within consecutive monitoring periods can be clarified. This allows for the determination of whether a task set for a specific physiological state of the target animal should be executed within the candidate task execution period of the current time, thereby improving the accuracy of the prediction model's analysis of the changes in animal behavior of the target animal and further enhancing the accuracy of the prediction results output by the prediction model.

[0096] The aforementioned current time period for candidate task execution can be understood as: within a specified time period after the task analysis was triggered, such as the day the task analysis was triggered, or within two days of the task analysis being triggered. This is all reasonable.

[0097] Optionally, in one implementation, the target animal is a breeding pig, and the specified physiological state is the mating state. Correspondingly, the aforementioned animal behaviors may include lying down, feeding, mounting, etc.

[0098] As can be seen from the above, in the task early warning method for animal physiological states provided in this application embodiment, a prediction model is pre-trained based on training samples and corresponding ground truth values ​​for analysis of task early warning for animal physiological states; furthermore, this application sets a triggering condition for task analysis, so that in response to the satisfaction of the condition, for each monitoring cycle in a series of consecutive monitoring cycles, the multidimensional features of the target animal within that monitoring cycle are determined, that is, based on the radar data collected by the target radar for the target animal in each monitoring cycle, the multidimensional features within that monitoring cycle are determined, and each obtained multidimensional feature is used to describe the target animal in a target statistical period. The model predicts animal behavior within multiple monitoring periods, including the monitoring period closest to the current time. Then, for each monitoring period, a target temporal feature sequence is constructed based on the multidimensional features within that period. This sequence reflects the changing relationships of animal behavior in the target animal within the corresponding monitoring period. Finally, by analyzing the target temporal feature sequences across multiple monitoring periods using the aforementioned prediction model, the model clarifies the changing relationships of animal behavior in the target animal across consecutive monitoring periods. This determines whether a task set for a specific physiological state of the target animal needs to be executed within the candidate task execution period of the current time.

[0099] As can be seen, this scheme, by combining the multidimensional features determined for the target animal and the behavioral changes represented by the constructed time-series feature sequences of each target, provides effective data support for the analysis of subsequent prediction models, thereby improving the accuracy of the prediction results. Furthermore, the pre-trained prediction model can analyze whether a task set for a specific physiological state of the target animal needs to be executed based on the behavioral changes over multiple monitoring periods. Therefore, this scheme can effectively provide early warnings for tasks set for a specific physiological state of the target animal, assisting farmers in responding promptly to these changes.

[0100] Alternatively, in one implementation, such as Figure 2 As shown, the generation method of each multidimensional feature of the target animal in any monitoring period includes the following steps S201-S204:

[0101] S201: Acquire radar data collected by the target radar on the target animal during the monitoring period;

[0102] S202: According to the target segmentation time, the acquired radar data is divided into multiple data segments;

[0103] The target segmentation duration shall not exceed the duration of the target statistical period;

[0104] S203: Based on multiple data segments, analyze the first and second analysis results for each data segment;

[0105] The first analysis result is the behavioral classification result of the target animal within the collection time range corresponding to a data segment, and the second analysis result is the vibration intensity generated when the target animal performs an animal behavior within the collection time range corresponding to a data segment.

[0106] S204: Based on the first and second analysis results of each data segment, generate the multidimensional characteristics of the target animal within the monitoring period.

[0107] In this implementation, for any given monitoring period, radar data collected by the target radar from the target animal within that monitoring period is acquired. The acquired radar data is then divided into multiple data segments, with the segmentation duration not exceeding the target statistical time period. The target segmentation duration can be 1 second, 10 seconds, 1 minute, 3 minutes, etc. The target statistical time period can be 1 hour, 2 hours, etc.

[0108] For example, a target sliding window with a window length equal to the target segmentation time and a sliding step size of a predetermined step size is set. The acquired radar data is arranged in the order of acquisition time to obtain a corresponding one-dimensional radar data sequence. Then, the target segmentation time is used as the interception unit and the predetermined step size is used as the displacement increment, so that the target sliding window slides sequentially along the time axis of the obtained radar data sequence. During the sliding process, the radar data within the coverage area of ​​each interception unit (window) is intercepted in real time, and the resulting subsets of radar data windows are used as multiple data segments.

[0109] Then, for each of the multiple acquired data segments, based on the radar data contained in the data segment, the animal behavior of the target animal within the acquisition time range corresponding to the data segment is identified, the behavior classification result corresponding to the data segment is obtained, and the obtained behavior classification result is used as the first analysis result of the data segment.

[0110] Furthermore, in the analysis of the physiological state of a target animal, vibration intensity is a core quantitative indicator reflecting the animal's activity characteristics and physiological function. Under normal physiological conditions, the vibration intensity generated by a target animal performing a behavior remains within a relatively stable range. However, when specific physiological states such as disease, stress, estrus, or mating occur, the vibration intensity exhibits significant abnormal fluctuations. In other words, the vibration intensity generated by a target animal performing a behavior can help predictive models effectively analyze the specific physiological state that the target animal is currently in or will soon be in.

[0111] Therefore, for each of the multiple acquired data segments, based on the radar data contained in the data segment, the radar response intensity received within the acquisition time range corresponding to the data segment is analyzed as the vibration intensity generated when the target animal performs an animal behavior within the acquisition time range corresponding to the data segment. The greater the radar response intensity, the greater the vibration intensity generated when the target animal performs an animal behavior. Then, the analyzed vibration intensity is used as the second analysis result of the data segment.

[0112] In this way, based on the first and second analysis results of each data segment, the generated multidimensional features of the target animal during the monitoring period can effectively characterize the physiological state of the target animal during the monitoring period. That is, the first and second analysis results of each data segment obtained during a monitoring period are used as a multidimensional feature obtained during the monitoring period.

[0113] Considering that when target radar collects vibration signals from target animals as radar data, it will inevitably capture various environmental clutter interferences. These interferences are unrelated to the analysis of the target animal's physiological state and will drown out or distort the effective signal. For example, the slight vibrations of fixed structures such as fences and walls will form stable background clutter; airflow vibrations generated by the operation of ventilation equipment, vibration transmission from personnel walking, and the activities of other animals will form random interference information; electromagnetic noise generated by electrical equipment and wireless communication equipment in the farm will be coupled into the radar data.

[0114] Based on this, optionally, before step S202 above, the above task early warning method may further include: filtering and denoising the acquired radar data.

[0115] Accordingly, step S202 above may include the following steps:

[0116] According to the target segmentation time, the filtered and denoised radar data is divided into multiple data segments.

[0117] Thus, through the above-mentioned filtering and denoising processing, the identification accuracy of effective data related to the target animal is improved, and the signal-to-noise ratio of radar data is reduced, providing a reliable data foundation for subsequent radar data analysis. The above-mentioned filtering and denoising processing can be frequency-domain based (e.g., low-pass filtering, high-pass filtering, or band-pass filtering) or time-domain based (e.g., moving average filtering, median filtering), etc., and this application does not impose specific limitations.

[0118] In this implementation, radar data belonging to the same data segment are analyzed from multiple dimensions by dividing the data into segments. This yields multi-dimensional features of the physiological state of the target animal within the acquisition time range corresponding to the same data segment, providing multi-dimensional data information for subsequent prediction models and improving the accuracy of prediction results.

[0119] Optionally, in one implementation, the duration of the target statistical period is a specified multiple of the target segmented duration, such as... Figure 3 As shown, step S204 above may include the following steps S2041-S2042:

[0120] S2041: For each target statistical period within the monitoring cycle, based on the first and second analysis results of each data segment corresponding to the target statistical period, analyze the first cumulative duration of vibration intensity greater than the preset intensity, and the second cumulative duration of the same animal behavior produced by the target animal.

[0121] S2042: Based on the first cumulative duration and the second cumulative duration obtained from the analysis, as well as the first and second analysis results of each data segment corresponding to the target statistical period, determine the multidimensional characteristics corresponding to the target statistical period within the monitoring cycle.

[0122] In this context, each data segment corresponding to the target statistical period is a data segment whose corresponding collection time range is within the target statistical period.

[0123] In this implementation, it is considered that the radar data collected by the target radar may be subject to instantaneous environmental interference (such as: people passing by briefly, equipment shaking momentarily, etc.). Such interference may cause the vibration intensity to deviate from the normal threshold for a short time, but the generation of such radar data is unrelated to the target animal itself; and the single animal behavior performed by the target animal may be triggered by random factors (such as: accidental environmental stimuli, etc.) and does not have the representational meaning of physiological state.

[0124] Therefore, in order to improve the effectiveness of the collected multidimensional data, each monitoring cycle is divided into various target statistical periods. The duration of each target statistical period is a specified multiple of the target segmentation duration. Then, for each target statistical period within the monitoring cycle, based on the first and second analysis results of each data segment corresponding to the target statistical period, the first cumulative duration of vibration intensity greater than the preset intensity and the second cumulative duration of the same animal behavior produced by the target animal are analyzed.

[0125] In other words, by separately statistically analyzing the first and second cumulative durations, the data interference caused by occasional behaviors or events on the analysis of the physiological state of the target animal is reduced, thereby improving the validity and accuracy of the multidimensional features obtained within a monitoring period.

[0126] Subsequently, based on the first and second cumulative durations obtained from the analysis, and the first and second analysis results of each data segment corresponding to the target statistical period (i.e., each data segment whose corresponding collection time range is within the target statistical period), the multidimensional features corresponding to the target statistical period within the monitoring cycle are determined. In other words, for each target statistical period within the monitoring cycle, the multidimensional features corresponding to that target statistical period can be: the first cumulative duration counted under that target statistical period, the second cumulative duration counted under that target statistical period, the first analysis result corresponding to each data segment under that target statistical period, and the second analysis result corresponding to each data segment under that target statistical period. For example, if the monitoring cycle includes 3 target statistical periods, and each target statistical period includes 5 data segments, then the multidimensional features corresponding to each target statistical period in the monitoring cycle are: the first and second analysis results of each data segment included in the target statistical period, and, based on the first and second analysis results of each data segment: the first cumulative duration of vibration intensity greater than a preset intensity, and the second cumulative duration of the same animal behavior produced by the target animal.

[0127] In this implementation, a target statistical period is used as a statistical unit to count the first cumulative duration and the second cumulative duration. This reduces the data interference caused by occasional behaviors or events on the analysis of the physiological state of the target animal and improves the accuracy of the multidimensional features obtained within a monitoring period.

[0128] The preset intensity can be determined based on the vibration intensity generated by the target animal when it performs an animal behavior under normal physiological conditions, or it can be determined based on the average vibration intensity generated by the target animal when it performs an animal behavior under normal physiological conditions. Both of these are reasonable.

[0129] In another embodiment of this application, the prediction model includes a convolutional network layer, an attention mechanism layer, and a long short-term memory layer connected in sequence;

[0130] The convolutional network layer in the prediction model is used to: extract features about local features from each target temporal feature sequence to obtain the first temporal feature sequence corresponding to the target temporal feature sequence;

[0131] The attention mechanism layer in the prediction model is used to: extract features about global temporal features for each first temporal feature sequence, and obtain the second temporal feature sequence corresponding to each first temporal feature sequence;

[0132] The long short-term memory layer in the prediction model is used to: generate a temporal change relationship of the target animal’s own physiological state before entering a specified physiological state based on each second temporal feature sequence; and analyze, based on the obtained temporal change relationship, whether the task for the specified physiological state of the target animal needs to be executed within the candidate task execution period to which the current time belongs.

[0133] In this embodiment, the convolutional network layer can extract features about local features for each target temporal feature sequence, and then perform nonlinear combination and enhancement of the multidimensional features corresponding to the target temporal feature sequence based on the extracted local features to obtain the first temporal feature sequence corresponding to the target temporal feature sequence.

[0134] The attention mechanism layer can be used to extract global temporal features for each first temporal feature sequence to obtain global temporal features for each local feature, and generate attention weights corresponding to each local temporal feature. Then, based on the obtained attention weights, a second temporal feature sequence corresponding to each first temporal feature sequence is generated. In other words, through the above attention mechanism layer, the prediction model can dynamically assign different attention weights to local features in the first temporal feature sequence that belong to different time periods (such as day and night). That is, by allocating attention weights, the key time periods that are strongly correlated with the label during the day (i.e., the time period corresponding to the first cumulative duration) are extracted, while the time periods with greater interference are weakened. This is to adapt to the characteristic that the effective information of the target animal is different at different time periods (for example, at night, there is less interference from humans, so a larger attention weight can be assigned to local features belonging to night), so as to help the prediction model ignore some interference information and improve the anti-interference ability of the prediction model.

[0135] The Long Short-Term Memory (LSTM) layer is used to concatenate each obtained second temporal feature sequence into a sequence with a longer time span, so that the concatenated sequence can characterize the temporal changes in the target animal's physiological state before entering the specified physiological state, providing more basic analytical data for subsequent model analysis. Then, based on the obtained temporal changes, the prediction results of whether the task for the specified physiological state of the target animal should be executed within the candidate task execution period of the current time are analyzed.

[0136] Alternatively, in one implementation, the convolutional network layer is a one-dimensional convolutional network layer, and the long short-term memory layer is an LSTM (Long Short-Term Memory) model.

[0137] In this embodiment, the prediction model may include a convolutional network layer, an attention mechanism layer, and a long short-term memory layer connected in sequence. By combining the above-mentioned convolutional network layer, attention mechanism layer, and long short-term memory layer, it is possible to capture local correlations, global temporal dependencies, and dynamic trends of physiological states, and fully utilize and understand the temporal dependencies and dynamic trends between the multidimensional features of the target animal to improve the accuracy of the obtained prediction results.

[0138] In another embodiment of this application, the training process of the prediction model includes the following steps A1-A2:

[0139] Step A1: Input the training samples and corresponding ground truth values ​​into the prediction model to be trained, so as to obtain the prediction result of whether the task set by the specified physiological state of the sample animal needs to be executed within the candidate task execution period of the sample time.

[0140] Step A2: Based on the prediction results and the true values, calculate the model loss value of the prediction model to be trained, and adjust the model parameters of the prediction model to be trained in response to the model loss value being greater than the predetermined loss value.

[0141] The prediction model in this embodiment is trained based on training samples and corresponding ground truth values. The training samples include: sample time sequence features of sample animals corresponding to multiple consecutive sample periods determined by radar data collected from the sample animals. Among the multiple sample periods, there is a monitoring period with the closest sample time. The ground truth values ​​are: the annotation results of whether the task set for the specified physiological state of the sample animal should be executed for the candidate task execution period to which the sample time belongs.

[0142] Thus, after inputting the acquired training samples and corresponding ground truth values ​​into the prediction model to be trained, the prediction model analyzes the changes in the physiological state of the sample animal as represented by the temporal feature sequence of the samples, and predicts whether the task set by the specified physiological state of the sample animal needs to be executed within the candidate task execution period to which the sample time belongs. Then, based on the obtained prediction results and the ground truth values ​​of the training samples corresponding to the prediction results, the model loss value of the prediction model to be trained is calculated; thereby, by judging the degree of difference between the model loss value and the predetermined loss value, it is determined whether the model parameters of the prediction model to be trained need to be adjusted, that is, whether the prediction model to be trained can end training.

[0143] When the model loss value is greater than the predetermined loss value, the model parameters of the prediction model to be trained need to be adjusted. At this time, the model parameters of the prediction model to be trained can be adjusted using the backpropagation algorithm based on the degree of difference between the model loss value and the predetermined loss value, until the model loss value is no greater than the predetermined loss value, and then the training ends.

[0144] To facilitate understanding, a specific example is provided below to illustrate the task warning method for animal physiological states provided in this application. For example... Figure 4 As shown, the specific process of this example includes the following steps S401-S408:

[0145] S401: Acquire radar data about the target breeding pig;

[0146] S402: Filter and denoise radar data;

[0147] S403: Extract vibration intensity from the processed radar data according to a window length of 1; where the window length of 1 is the target segmentation duration in this application;

[0148] S404: Extract pig behavior from the processed radar data according to a window length of 1; wherein, pig behavior is the animal behavior in this application;

[0149] S405: Accumulate the cumulative duration of strong activity according to window length 2 and the cumulative duration of each type of pig behavior; wherein, window length 2 is the target statistical period in this application, the cumulative duration of strong activity is the first cumulative duration in this application, and the cumulative duration of each type of pig behavior is the second cumulative duration in this application;

[0150] S406: The feature learning layer in the hierarchical temporal deep learning model uses a convolutional neural network to obtain daily features for data with a window length of 2; wherein, the hierarchical temporal deep learning model is the prediction model in this application, the feature learning layer in the hierarchical temporal deep learning model is the convolutional network layer of the prediction model in this application, and the daily features are the first temporal feature sequence in this application;

[0151] S407: The attention mechanism layer in the hierarchical temporal deep learning model extracts effective information from daily features and ignores interference information through dynamic weight allocation: the extracted daily features are the second temporal feature sequence in this application.

[0152] S408: The temporal pattern learning layer in the hierarchical temporal deep learning model uses a long short-term memory network to extract the temporal change relationship of various features in the pre-mating period from the received and refined daily features, so as to output the result of whether the target breeding pig is about to enter the mating time; wherein, the temporal pattern learning layer is the long short-term memory layer in this application.

[0153] In this specific example, multi-channel time-series data of millimeter-wave radar of the target breeding pig is acquired as radar data collected for the target breeding pig. Then, the acquired radar data is filtered and denoised to remove clutter interference. After that, for the processed radar data, the vibration intensity within a window length of 1 is continuously calculated using a sliding window. For each data segment within a window length of 1, a deep learning model is used to identify pig behavior. Then, the cumulative duration of strong activity within a window length of 2 and the cumulative duration of each type of pig behavior are statistically analyzed.

[0154] Using the duration corresponding to window length 2 as the statistical unit, the vibration intensity, pig behavior, cumulative duration of strong activity, and cumulative duration of each type of pig behavior obtained under each window length 2 are used as multidimensional features under that window length 2.

[0155] Subsequently, multi-dimensional features accumulated over multiple days are input into a trained hierarchical temporal deep learning model, on a daily basis, to predict the mating time of breeding pigs. Specifically:

[0156] The feature learning layer uses a one-dimensional convolutional neural network to process the feature vectors of multi-dimensional features for each day, enabling it to learn the local correlations between daily features and perform non-linear combination and enhancement of multi-dimensional features to obtain higher-level daily features.

[0157] The attention mechanism layer enables the model to dynamically assign different weights to data from different time periods in the daily features, adapting to the different effective information of the breeding pigs at different times. For example, there is less human interference at night, so the weight of that time period can be adjusted more finely, thereby ignoring some interference information and improving the model's anti-interference ability.

[0158] The temporal pattern learning layer extracts daily high-level feature vectors through a convolutional neural network and inputs them into a long short-term memory network to extract longer-term temporal changes in various features before mating. Based on the extracted longer-term temporal changes, it analyzes whether the target breeding pig is about to enter the mating state and whether the tasks set for this mating state need to be executed.

[0159] As can be seen above, this specific example improves the data dimensionality and increases data diversity by extracting multi-dimensional features from radar data to analyze the physiological state of target breeding pigs. Furthermore, considering that the prediction of breeding time for breeding pigs may last for several days and there may be useless interference within a day, a hierarchical temporal deep learning model is adopted to take into account both the learning of feature trends before and after and the masking of interference periods of breeding pigs throughout the day, so as to improve the accuracy of breeding time prediction.

[0160] Corresponding to the above method embodiments, this application also provides a task early warning system for animal physiological states, such as... Figure 5 As shown, the system includes:

[0161] Target Radar 510 is used to collect radar data on target animals;

[0162] The early warning device 520 is used to perform a task early warning method for animal physiological state provided in the embodiments of this application.

[0163] As can be seen, in this scheme, the early warning device, by combining the multidimensional features determined for the target animal and the behavioral changes represented by the constructed temporal feature sequences of each target, provides effective data support for the analysis of subsequent prediction models, thereby improving the accuracy of the prediction results. Furthermore, the pre-trained prediction model can analyze whether a task set for a specific physiological state of the target animal needs to be executed based on the behavioral changes over multiple monitoring periods. Therefore, this scheme can effectively provide early warnings for tasks set for a specific physiological state of the target animal, assisting farmers in responding promptly to these changes.

[0164] Corresponding to the above method embodiments, this application also provides a task early warning device for animal physiological states, such as... Figure 6 As shown, the device includes:

[0165] The determination module 610 is configured to, in response to the fulfillment of conditions for triggering task analysis, determine various multidimensional features of the target animal within each of a series of consecutive monitoring periods; wherein each multidimensional feature is determined based on radar data collected from the target animal by the target radar within the monitoring period, and each multidimensional feature describes the animal behavior of the target animal during a target statistical time period; the series of monitoring periods includes the monitoring period closest to the current time.

[0166] The construction module 620 is used to construct the target time series feature sequence corresponding to each monitoring period based on the multi-dimensional features within that monitoring period.

[0167] The prediction module 630 is used to input the temporal feature sequences of each target into the prediction model to obtain a prediction result of whether the task set for the specified physiological state of the target animal needs to be executed within the candidate task execution period to which the current time belongs; wherein, the prediction model is trained based on training samples and corresponding ground truth values; the training samples include: sample temporal feature sequences of the sample animal corresponding to multiple consecutive sample periods determined according to radar data collected from the sample animal, wherein the multiple sample periods include a monitoring period with the closest sample time; the ground truth value is: the annotation result of whether the task set for the specified physiological state of the sample animal needs to be executed within the candidate task execution period to which the sample time belongs.

[0168] Optionally, in one implementation, the generation of various multidimensional features of the target animal within any monitoring period is performed through the following modules:

[0169] The acquisition module is used to acquire radar data collected by the target radar on the target animal during the monitoring period.

[0170] The segmentation module is used to segment the acquired radar data into multiple data segments according to the target segmentation duration; wherein the target segmentation duration is not greater than the duration of the target statistical period.

[0171] The analysis module is used to analyze a first analysis result and a second analysis result for each of the multiple data segments; wherein, the first analysis result is the behavioral classification result of the target animal within the collection time range corresponding to a data segment, and the second analysis result is the vibration intensity generated when the target animal performs an animal behavior within the collection time range corresponding to a data segment.

[0172] The generation module is used to generate various multidimensional features of the target animal within the monitoring period based on the first and second analysis results of each data segment.

[0173] Optionally, in one implementation, the duration of the target statistical period is a specified multiple of the target segmentation duration;

[0174] The generation module is specifically used for:

[0175] For each target statistical period within the monitoring cycle, based on the first and second analysis results of each data segment corresponding to the target statistical period, the first cumulative duration of vibration intensity greater than the preset intensity and the second cumulative duration of the same animal behavior produced by the target animal are analyzed.

[0176] Based on the first cumulative duration and the second cumulative duration obtained from the analysis, as well as the first and second analysis results of each data segment corresponding to the target statistical period, the multidimensional characteristics corresponding to the target statistical period within the monitoring cycle are determined.

[0177] In this context, each data segment corresponding to the target statistical period is a data segment whose corresponding collection time range is within the target statistical period.

[0178] Optionally, in one implementation, the prediction model includes a convolutional network layer, an attention mechanism layer, and a long short-term memory layer connected in sequence;

[0179] The convolutional network layer in the prediction model is used to: extract features about local features from each target temporal feature sequence to obtain the first temporal feature sequence corresponding to the target temporal feature sequence;

[0180] The attention mechanism layer in the prediction model is used to: extract features about global time-series features for each first time-series feature sequence, and obtain the second time-series feature sequence corresponding to each first time-series feature sequence;

[0181] The long short-term memory layer in the prediction model is used to: generate a temporal change relationship of the target animal's physiological state before entering a specified physiological state based on each second temporal feature sequence; and analyze, based on the obtained temporal change relationship, whether the task for the specified physiological state of the target animal needs to be executed within the candidate task execution period to which the current time belongs.

[0182] Optionally, in one implementation, the training process of the prediction model includes:

[0183] The training samples and corresponding ground truth values ​​are input into the prediction model to be trained to obtain a prediction result of whether the task set by the specified physiological state of the sample animal needs to be executed within the candidate task execution period to which the sample time belongs.

[0184] Based on the prediction results and the true values, the model loss value of the prediction model to be trained is calculated, and in response to the model loss value being greater than a predetermined loss value, the model parameters of the prediction model to be trained are adjusted.

[0185] Optionally, in one implementation, the target animal is a breeding pig, and the specified physiological state is a mating state.

[0186] Optionally, in one implementation, the convolutional network layer is a one-dimensional convolutional network layer; the long short-term memory layer is an LSTM model.

[0187] This application also provides an electronic device, such as... Figure 7 As shown, it includes:

[0188] Memory 701 is used to store computer programs;

[0189] When the processor 702 executes the program stored in the memory 701, it implements the task warning method for the physiological state of animals as described in any of the above embodiments.

[0190] Furthermore, the aforementioned electronic device may also include a communication bus and / or a communication interface, with the processor 702, the communication interface, and the memory 701 communicating with each other via the communication bus.

[0191] The communication bus mentioned in the above electronic devices can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of illustration, only one thick line is used to represent it in the diagram, but this does not mean that there is only one bus or one type of bus.

[0192] The communication interface is used for communication between the aforementioned electronic devices and other devices.

[0193] The memory may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0194] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0195] In another embodiment provided in this application, a computer-readable storage medium is also provided, which stores a computer program that, when executed by a processor, implements the steps of any of the above-described task warning methods for the physiological state of an animal.

[0196] In another embodiment provided in this application, a computer program product containing instructions is also provided, which, when run on a computer, causes the computer to execute any of the task warning methods for the physiological state of an animal in the above embodiments.

[0197] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a solid-state drive (SSD), etc.

[0198] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0199] The various embodiments in this specification are described in a related manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the device embodiments, electronic device embodiments, computer-readable storage medium embodiments, and computer program product embodiments are basically similar to the method embodiments, so the descriptions are relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0200] The above description is merely a preferred embodiment of this application and is not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application are included within the scope of protection of this application.

Claims

1. A method for task alerting to an animal physiological state, characterized by, include: In response to the fulfillment of conditions for triggering task analysis, for each of multiple consecutive monitoring periods, various multidimensional features of the target animal within that monitoring period are determined. Each multidimensional feature is determined based on radar data collected from the target animal by the target radar within that monitoring period, and each multidimensional feature describes the animal behavior of the target animal during a target statistical time period. The multiple monitoring periods include the monitoring period closest to the current time. Each multidimensional feature includes: a first analysis result and a second analysis result for each data segment corresponding to a target statistical time period, and a first cumulative duration of vibration intensity greater than a preset intensity and a second cumulative duration of the same animal behavior produced by the target animal, determined based on the first and second analysis results of the data segments corresponding to the target statistical time period. The first analysis result of each data segment is the behavioral classification result of the target animal within the collection time range corresponding to the data segment. The second analysis result of each data segment is the vibration intensity generated when the target animal performs an animal behavior within the collection time range corresponding to the data segment. Each data segment corresponding to the target statistical period is: a data segment in the collected radar data whose corresponding collection time range is within the target statistical period. The vibration intensity is determined by: based on the radar data contained in a data segment, analyzing the radar response intensity received within the collection time range corresponding to the data segment, which is taken as the vibration intensity generated when the target animal performs an animal behavior within the collection time range corresponding to the data segment. The greater the radar response intensity, the greater the vibration intensity generated when the target animal performs an animal behavior. For each monitoring period, a target time-series feature sequence corresponding to that monitoring period is constructed based on the various multi-dimensional features within that monitoring period. Each target temporal feature sequence is input into the prediction model to obtain a prediction result of whether a task set for a specified physiological state of the target animal needs to be executed within the candidate task execution period to which the current time belongs. The prediction model is trained based on training samples and corresponding ground truth values. The training samples include: sample temporal feature sequences of the sample animal corresponding to multiple consecutive sample periods determined based on radar data collected from the sample animal, wherein the multiple sample periods include a monitoring period with the closest sample time. The ground truth value is the annotation result of whether a task set for a specified physiological state of the sample animal needs to be executed within the candidate task execution period to which the sample time belongs.

2. The method according to claim 1, characterized in that, The generation methods for the various multidimensional features of the target animal during any monitoring period include: Acquire radar data collected by the target radar on the target animal during the monitoring period; The acquired radar data is divided into multiple data segments according to the target segmentation duration; wherein the target segmentation duration is not greater than the duration of the target statistical period. Based on the multiple data segments, analyze the first and second analysis results for each data segment; Based on the first and second analysis results of each data segment, multidimensional features of the target animal during the monitoring period are generated.

3. The method according to claim 2, characterized in that, The duration of the target statistical period is a specified multiple of the target segmented duration; Based on the first and second analysis results for each data segment, the multidimensional characteristics of the target animal during the monitoring period are generated, including: For each target statistical period within the monitoring cycle, based on the first and second analysis results of each data segment corresponding to the target statistical period, the first cumulative duration of vibration intensity greater than the preset intensity and the second cumulative duration of the same animal behavior produced by the target animal are analyzed. Based on the first cumulative duration and the second cumulative duration obtained from the analysis, as well as the first and second analysis results of each data segment corresponding to the target statistical period, the multidimensional characteristics corresponding to the target statistical period within the monitoring cycle are determined.

4. The method according to claim 1, characterized in that, The prediction model comprises a convolutional network layer, an attention mechanism layer, and a long short-term memory layer connected in sequence. The convolutional network layer in the prediction model is used to: extract features about local features from each target temporal feature sequence to obtain the first temporal feature sequence corresponding to the target temporal feature sequence; The attention mechanism layer in the prediction model is used to: extract features about global time-series features for each first time-series feature sequence, and obtain the second time-series feature sequence corresponding to each first time-series feature sequence; The long short-term memory layer in the prediction model is used to: generate a temporal change relationship of the target animal's physiological state before entering a specified physiological state based on each second temporal feature sequence; and analyze, based on the obtained temporal change relationship, whether the task for the specified physiological state of the target animal needs to be executed within the candidate task execution period to which the current time belongs.

5. The method according to claim 1, characterized in that, The training process of the prediction model includes: The training samples and corresponding ground truth values ​​are input into the prediction model to be trained to obtain a prediction result of whether the task set by the specified physiological state of the sample animal needs to be executed within the candidate task execution period to which the sample time belongs. Based on the prediction results and the true values, the model loss value of the prediction model to be trained is calculated, and in response to the model loss value being greater than a predetermined loss value, the model parameters of the prediction model to be trained are adjusted.

6. The method according to claim 1, characterized in that, The target animal is a breeding pig, and the specified physiological state is the mating state.

7. The method according to claim 4, characterized in that, The convolutional network layer is a one-dimensional convolutional network layer; the long short-term memory layer is an LSTM model.

8. A task early warning device for animal physiological states, characterized in that, include: A determination module, in response to conditions for triggering task analysis, determines various multidimensional features of the target animal within each of multiple consecutive monitoring cycles. Each multidimensional feature is determined based on radar data collected from the target animal by the target radar within that monitoring cycle, and each multidimensional feature describes the animal behavior of the target animal during a target statistical time period. The multiple monitoring cycles include the monitoring cycle closest to the current time. Each multidimensional feature includes: a first analysis result and a second analysis result for each data segment corresponding to a target statistical time period; and a first cumulative duration of vibration intensity greater than a preset intensity and a second cumulative duration of the same animal behavior produced by the target animal, determined based on the first and second analysis results of the data segments corresponding to the target statistical time period. The first analysis result of each data segment is the behavioral classification result of the target animal within the collection time range corresponding to the data segment, and the second analysis result of each data segment is the vibration intensity generated by the target animal when it performs an animal behavior within the collection time range corresponding to the data segment. Each data segment corresponding to the target statistical period is a data segment in the collected radar data whose corresponding collection time range is within the target statistical period. The method for determining the vibration intensity includes: based on the radar data contained in a data segment, analyzing the radar response intensity received within the collection time range corresponding to the data segment, which is used as the vibration intensity generated by the target animal when it performs an animal behavior within the collection time range corresponding to the data segment. The greater the radar response intensity, the greater the vibration intensity generated by the target animal when it performs an animal behavior. The construction module is used to construct the target time-series feature sequence corresponding to each monitoring period based on the multi-dimensional features within that monitoring period. The prediction module is used to input the temporal feature sequences of each target into the prediction model to obtain a prediction result of whether the task set for the specified physiological state of the target animal needs to be executed within the candidate task execution period to which the current time belongs. The prediction model is trained based on training samples and corresponding ground truth values. The training samples include: sample temporal feature sequences of the sample animal corresponding to multiple consecutive sample periods determined according to radar data collected from the sample animal, wherein the multiple sample periods include a monitoring period with the closest sample time. The ground truth values ​​are: the annotation results of whether the task set for the specified physiological state of the sample animal needs to be executed within the candidate task execution period to which the sample time belongs.

9. A task early warning system for animal physiological states, characterized in that, include: Target radar, used to collect radar data on target animals; A warning device for performing the method according to any one of claims 1-7.

10. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor, when executing a program stored in memory, implements the method described in any one of claims 1-7.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method described in any one of claims 1-7.

Images