Apparatus for predicting cattle estrus behavior using biosenor data based on arrificial intelligence and mehtod therof

Abstract

The present invention provides an artificial intelligence-based estrus behavior prediction device and prediction method using livestock bio-information. An artificial intelligence-based estrus behavior prediction device using livestock bio-information according to one embodiment of the present invention comprises: a bio-data collection unit that collects bio-data of a livestock using a bio-sensor; a pre-processing unit that pre-processes the bio-data; an artificial intelligence model training unit that trains an artificial intelligence model to predict the estrus behavior of the livestock using the pre-processed bio-data; a performance indicator evaluation unit that evaluates the performance indicator of the trained artificial intelligence model; and an artificial intelligence model optimization unit that optimizes the artificial intelligence model by adjusting at least one hyperparameter based on the performance indicator. According to the artificial intelligence-based estrus behavior prediction device and method using cattle bio-information of the present invention, by accurately detecting and predicting the estrus period, the conception rate can be improved and the non-conception period reduced, thereby obtaining economic benefits.

Description

Technology Field

[0001] The present invention relates to an artificial intelligence model and algorithm for predicting whether cattle are in an estrus state, and more specifically, to an artificial intelligence-based estrus behavior prediction device and prediction method using cattle bio-information. Background Technology

[0002] The cattle industry has grown continuously from the past to the present, and cattle farming in Korea is expanding significantly. In particular, the number of Hanwoo cattle raised increased from approximately 1.5 million in 2005 to approximately 3.5 million in 2022, and the number of large-scale livestock farms is also increasing. However, as the number of livestock workers decreases and the workforce ages, managing cattle breeding is becoming more difficult. Consequently, cattle breeding management is emerging as a critical factor that directly impacts productivity and profits.

[0003] The estrous cycle of cattle is 21 days, and estrus occurs according to this cycle. Once in estrus, a cow's activity level increases by 1.3 to more than 2 times compared to normal days. Furthermore, the onset of estrus is a critical factor for breeding success. Cows typically begin estrus at night (from 6 PM to 6 AM the following day) with a probability of over 70%. This is a time when detecting and managing estrus is difficult, and it is challenging for livestock farmers to check the cow's estrous status upon waking in the morning. Consequently, it is difficult to determine the exact time of the initial onset of estrus, which makes it difficult to optimize breeding. Due to low conception rates, farmers face problems such as prolonged gestation periods, increased feed costs, and lowered productivity. Therefore, to manage the estrous cycle and breeding of cattle, it is essential to accurately predict estrus and take appropriate measures.

[0004] Therefore, one method for detecting estrus is to use pedometers. Pedometers are attached to the legs or necks of cattle to collect data on step count trends and transmit it to the cloud. Since cattle in estrus increase their step count by approximately two to four times compared to normal levels, estrus is detected through this data on step trends. However, because cattle typically move in groups, the occurrence of estrus in one cow may cause surrounding individuals to become more active as well. Additionally, while estrus can occur during the first ovulation after calving, it may not be accompanied by an increase in activity levels. For these reasons, it can be difficult to determine a cow's estrus status solely through pedometers.

[0005] Another method for detecting estrus involves attaching accelerometers to cattle and utilizing sensor values ​​generated by their movements. A 3-axis accelerometer measures acceleration along the x, y, and z axes, and detects estrus when the sum of the absolute acceleration values ​​observed during mounting behavior is higher than that of normal movement. The accelerometers are attached to the cattle's legs and necks to record activity levels and frequency over a 24-hour period. Based on this information, data analysis is conducted to display behavioral characteristics and estrus indices in graphs. This method requires advanced data processing and algorithms, and may result in inaccurate estimations or interpretations. Furthermore, it requires the attachment and maintenance of accelerometers and collars, and the equipment may strain the cattle's behavior.

[0006] In conventional technology, calculation methods utilizing acceleration sensors primarily evaluated activity levels based on the magnitude of the sum of acceleration vectors along the x, y, and z axes. However, this approach has limitations in accurately reflecting the full activity level of cattle. It is difficult to express a cow's activity level solely through changes in these axes, which can impose constraints on fully understanding and predicting the cow's actual movements and behavioral patterns. Prior art literature

[0007] Registered Patent Publication 10-2117092 (2020.05.25) The problem to be solved

[0008] The present invention aims to solve conventional problems by presenting an artificial intelligence model and algorithm that predicts estrous behavior in cattle using collected biological data.

[0009] The present invention aims to present an artificial intelligence model and algorithm capable of accurately predicting whether cattle are in estrus using biological data and predicted estrus behavior.

[0010] The problems of the present invention are not limited to those mentioned above, and other unmentioned problems will be clearly understood by those skilled in the art from the description below. means of solving the problem

[0011] An artificial intelligence-based estrus behavior prediction device using cattle bio-information according to one embodiment of the present invention is,

[0012] A bio-data collection unit that collects bio-data of cattle using a bio-sensor;

[0013] A preprocessing unit for preprocessing the above biological data;

[0014] An artificial intelligence model learning unit that trains an artificial intelligence model to predict the estrus behavior of the cattle using the preprocessed biological data;

[0015] A performance indicator evaluation unit for evaluating performance indicators of the above-mentioned artificial intelligence model; and

[0016] It includes an artificial intelligence model optimization unit that optimizes the artificial intelligence model by adjusting at least one hyperparameter based on the above performance indicators.

[0018] Preferably,

[0019] The above biosensor is installed within the rumen of the cattle, and the biological data is characterized as being the movement acceleration of the cattle and the biological temperature of the cattle.

[0021] Preferably,

[0022] The above pretreatment unit is configured to transform the above biological data into a predetermined data form and rearrange it.

[0024] Preferably,

[0025] It further includes a behavior labeling unit that labels the movement behavior of the cattle when collecting the above biometric data, and

[0026] The trained artificial intelligence model predicts whether the labeled movement behavior of the cattle is estrus behavior.

[0028] Preferably,

[0029] It further includes a cattle estrus determination unit that determines whether the cattle are in an estrus state based on the collected real-time biological data and predicted estrus behavior of the cattle using the optimized artificial intelligence model.

[0031] Preferably,

[0032] The above performance indicator is characterized by being at least one of accuracy, recall, and F1 score.

[0034] An artificial intelligence-based estrus behavior prediction method using cattle bio-information according to one embodiment of the present invention is,

[0035] A bio-data collection step for collecting bio-data of cattle using a biosensor;

[0036] A preprocessing step for preprocessing the above biological data;

[0037] An artificial intelligence model training step for training an artificial intelligence model to predict the estrus behavior of the cattle using the preprocessed biological data;

[0038] A performance indicator evaluation step for evaluating performance indicators of the above-mentioned artificial intelligence model; and

[0039] It includes an artificial intelligence model optimization step of optimizing the artificial intelligence model by adjusting at least one hyperparameter based on the above performance indicators.

[0041] Preferably,

[0042] The above biosensor is installed within the rumen of the cattle, and the biological data is characterized as being the movement acceleration of the cattle and the biological temperature of the cattle.

[0044] Preferably,

[0045] The above preprocessing step includes a step of transforming the bio-data into a predetermined data form and rearranging it.

[0047] Preferably,

[0048] The method further includes a behavior labeling step for labeling the movement behavior of the cattle when collecting the above biometric data, and

[0049] The trained artificial intelligence model predicts whether the labeled movement behavior of the cattle is estrus behavior.

[0051] Preferably,

[0052] It further includes a cattle estrus determination step that determines whether the cattle are in an estrus state based on the collected real-time biological data and predicted estrus behavior of the cattle using the optimized artificial intelligence model.

[0054] Preferably,

[0055] The above performance indicator is characterized by being at least one of accuracy, recall, and F1 score.

[0057] Specific details of other embodiments are included in the detailed description and drawings. Effects of the invention

[0058] According to the artificial intelligence-based estrus behavior prediction device and method using livestock bio-information of the present invention, the welfare of livestock can be improved by optimizing the livestock barn environment and operation and enhancing conditions within the barn. Furthermore, it can resolve the labor shortage problem caused by the aging of livestock farmers and simultaneously have a positive impact on the local economy.

[0059] According to the artificial intelligence-based estrus behavior prediction device and method using cattle bio-information of the present invention, by accurately detecting and predicting the estrus period, the conception rate can be improved and the non-conception period reduced, thereby obtaining economic benefits.

[0060] According to the artificial intelligence-based estrus behavior prediction device and method using cattle bio-information of the present invention, there is a significant advantage in being able to predict and prepare for the estrus of cattle in advance.

[0061] However, the effects of the present invention are not limited to those mentioned above, and other unmentioned effects will be clearly understood by those skilled in the art from the description below. Brief explanation of the drawing

[0062] FIG. 1 is a schematic diagram illustrating the configuration of an artificial intelligence-based estrus behavior prediction device using livestock bio-information according to one embodiment of the present invention. FIG. 2 is a schematic diagram illustrating a process for predicting estrus behavior in cattle according to one embodiment of the present invention. FIG. 3 is a schematic diagram illustrating the structure of an artificial intelligence model (NLinear) used to predict the estrus behavior of cattle according to one embodiment of the present invention. FIGS. 4a and 4b are drawings showing the results of estrus behavior prediction using an artificial intelligence model (NLinear) used to predict estrus behavior of cattle according to one embodiment of the present invention. Figures 5a and 5b are diagrams showing the results of predicting estrus behavior using an MLP model compared to the artificial intelligence model (NLinear) of the present invention. Figures 6a and 6b are diagrams showing the results of predicting estrus behavior using an LSTM model compared to the artificial intelligence model (NLinear) of the present invention. FIG. 7 is a flowchart illustrating an artificial intelligence-based estrus behavior prediction method using livestock bio-information according to one embodiment of the present invention. FIG. 8 is a drawing illustrating an exemplary computing device capable of implementing a device and / or system according to various embodiments of the present invention. Specific details for implementing the invention

[0063] The advantages and features of the present invention and the methods for achieving them will become clear by referring to the embodiments described below in detail together with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below but may be implemented in various different forms. These embodiments are provided merely to ensure that the disclosure of the present invention is complete and to fully inform those skilled in the art of the scope of the invention, and the present invention is defined only by the scope of the claims. Throughout the specification, the same reference numerals refer to the same components.

[0064] The embodiments described herein will be described with reference to cross-sectional and / or plan views, which are exemplary illustrations of the invention. In the drawings, the thickness of the components is exaggerated for effective description of the technical content. Accordingly, the components illustrated in the drawings are schematic in nature, and the shapes of the components illustrated in the drawings are intended to illustrate specific forms of the components and are not intended to limit the scope of the invention. Although terms such as first, second, third, etc., have been used to describe various components in the various embodiments of this specification, these components should not be limited by such terms. These terms are used merely to distinguish one component from another. The embodiments described and illustrated herein also include their complementary embodiments.

[0065] The terms used herein are for describing the embodiments and are not intended to limit the invention. In this specification, the singular form includes the plural form unless specifically stated otherwise in the text. As used herein, "comprises" and / or "comprising" do not exclude the presence or addition of one or more other components, steps, actions, and / or elements to the mentioned components, steps, actions, and / or elements.

[0066] Unless otherwise defined, all terms used in this specification (including technical and scientific terms) may be used in a meaning commonly understood by those skilled in the art to which the present invention pertains. Additionally, terms defined in commonly used dictionaries are not to be interpreted ideally or excessively unless explicitly and specifically defined otherwise.

[0067] Hereinafter, the concept of the present invention and embodiments according thereto will be described in detail with reference to the drawings.

[0069] FIG. 1 is a schematic diagram illustrating the configuration of an artificial intelligence-based estrus behavior prediction device using livestock bio-information according to one embodiment of the present invention.

[0070] An artificial intelligence-based estrus behavior prediction device (100) using livestock bio-information according to one embodiment of the present invention includes a bio-data collection unit (110), a preprocessing unit (120), an artificial intelligence model learning unit (130), a performance indicator evaluation unit (140), and an artificial intelligence model optimization unit (150).

[0071] The bio-data collection unit (110) collects bio-data of cattle using a bio-sensor.

[0072] In one embodiment, the biosensor is installed in the rumen of a cattle.

[0073] In one embodiment, the biological data are the movement acceleration of the cattle and the biological temperature of the cattle.

[0074] The preprocessing unit (120) preprocesses the biological data.

[0075] In one embodiment, the pretreatment unit (120) is configured to rearrange the bio-data by transforming it into a predetermined data form.

[0076] The artificial intelligence model learning unit (130) trains an artificial intelligence model to predict the estrus behavior of cattle using preprocessed biological data.

[0077] The performance indicator evaluation unit (140) evaluates the performance indicators of the learned artificial intelligence model.

[0078] In the present invention, the performance indicators include at least one of accuracy, recall, and F1 score.

[0079] The artificial intelligence model optimization unit (150) optimizes the artificial intelligence model by adjusting at least one hyperparameter based on performance indicators.

[0080] In the present invention, the optimization of an artificial intelligence model refers to the improvement of the performance of an artificial intelligence model, and the artificial intelligence optimization unit may be referred to as the artificial intelligence performance improvement unit.

[0081] Alternatively, in the present invention, optimization refers to a process of adjusting at least one hyperparameter, and the optimization unit may refer to a parameter adjustment unit.

[0082] In one embodiment, an artificial intelligence-based estrus behavior prediction device (100) using livestock bio-information further includes a behavior labeling unit (not shown) that labels the movement behavior of the livestock when collecting bio-data, and an artificial intelligence model learning unit (130) is configured to predict whether the movement behavior of the livestock labeled by the artificial intelligence model is estrus behavior.

[0083] In one embodiment, the artificial intelligence-based estrus behavior prediction device (100) using cattle bio-information further includes a cattle estrus determination unit (not shown) that determines whether the cattle are in an estrus state based on collected real-time bio-data and predicted estrus behavior of the cattle using an optimized artificial intelligence model.

[0085] FIG. 2 is a schematic diagram illustrating a process for predicting estrus behavior in cattle according to one embodiment of the present invention.

[0086] Figure 2 illustrates a process for predicting estrus behavior in cattle, in which information on the activity level and body temperature of cattle is collected as an initial step (210).

[0087] In one embodiment, an acceleration sensor may be used to measure the activity level of a cattle. An acceleration sensor is attached to the cattle to measure the sensor value generated by the movement of the cattle.

[0088] Body temperature information is collected using a wireless capsule device within the rumen. The wireless capsule device within the rumen collects body temperature information from sensors (e.g., temperature sensors) within the rumen using wireless communication.

[0089] As a specific example, the collected biological data may include x-axis minimum acceleration, y-axis minimum acceleration, z-axis minimum acceleration, x-axis maximum acceleration, y-axis maximum acceleration, z-axis maximum acceleration, x-axis average acceleration, y-axis average acceleration, z-axis average acceleration, overall minimum acceleration, overall maximum acceleration, overall average acceleration, and gastric temperature.

[0090] In addition, when movement behavior of cattle occurs, the event is labeled (220). The movement behavior is provided to an artificial intelligence model and used as training data to determine estrus behavior.

[0091] The collected biometric data and labeled movement behaviors of cattle are transmitted to a server via relays and routers and stored in a database (DB).

[0092] Then, the collected biometric data is appropriately preprocessed for training an artificial intelligence model (230).

[0093] As an example of preprocessing, the preprocessing unit (120) may be configured to convert randomly generated measurement date and time data collected from the biosensor into a DateTime data type and then rearrange them.

[0094] The time unit of the data is set to a 10-minute interval, identical to the unit of data collection, and is expressed in year-month-day-hour-minute-second format. As a result of experiments with various sets of collected data, it was found that the presence or absence of estrus can be effectively predicted within the 10-minute collection interval even when using all collected biological data; therefore, all gastric temperature and acceleration data are utilized.

[0095] As another example of preprocessing, the preprocessing unit (120) may be configured to classify the collected cattle movements into fine movements, medium movements, and large movements. By utilizing data regarding the degree of classified movements, the artificial intelligence model can improve the performance of predicting the estrus state.

[0096] The artificial intelligence model training unit (241) trains an artificial intelligence model to predict the estrus behavior of cattle using preprocessed biological data. In the present invention, a linear model called an NLinear model is selected as the artificial intelligence model.

[0097] The performance indicator evaluation unit (242) evaluates the performance indicators of the learned artificial intelligence model.

[0098] In the present invention, the performance indicators include at least one of accuracy, recall, and F1 score.

[0099] When in estrus, the correct answer data has a value of 1, and when not, it has a value of 0. The prediction value is the final prediction of estrus state (1) if the estrus state model prediction value between 0 and 1 exceeds a threshold of 0.5, and if it does not exceed, it is the final prediction of not being in estrus state (0).

[0100] The confusion matrix for the classification results is Table 1 It is the same as.

[0101] [Table 1] Confusion Matrix

[0102]

[0103] In Table 1, a True Positive is when the model correctly predicts the cow's actual estrus state. A True Negative is when the model correctly predicts that the cow is not in estrus when the cow is not actually in estrus. A False Negative is when the model incorrectly predicts that the cow is not in estrus when the cow is not actually in estrus. A False Positive is when the model incorrectly predicts that the cow is in estrus when the cow is not actually in estrus.

[0104] Performance metrics are:

[0105] 1. Accuracy

[0106]

[0107] Accuracy refers to the proportion of True Positives and True Negatives in the confusion matrix; in other words, it is an indicator that evaluates how accurately a model predicts the estrus status of cows.

[0108] 2. Recall

[0109]

[0110] Recall refers to the ratio of times an AI model correctly predicts a cow as being in estrus when the cow is actually in estrus, and this can be calculated by dividing the True Positives in the confusion matrix by the sum of True Positives and False Negatives.

[0111] The objective of the present invention is to detect estrus in cattle. Since the loss is greater when an error occurs in judging a cow in estrus as not being in estrus than when an error occurs in judging a cow in estrus as being in estrus when it is not, the recall rate can be considered an important evaluation metric.

[0112] 3. F1 Score

[0113]

[0114] The F1 score is the harmonic mean of precision and recall. Since cattle have an average of 21 days of normal time and 18 hours of estrus duration, it is highly unbalanced, consisting mostly of non-estrus states.

[0115] The data used in this invention also shows a severe imbalance between classes, with only 1.6% of the total 218,300 data points being in estrus. When using imbalanced data, the F1 score provides a more reasonable evaluation result than accuracy by considering precision and recall simultaneously.

[0116] The artificial intelligence model optimization unit (150) optimizes the artificial intelligence model by adjusting at least one hyperparameter (243) based on performance indicators (244).

[0117] The task of finding the optimal model is performed by adjusting hyperparameters based on performance metrics.

[0118] Finally, the optimized model (260) determines whether the data is in the estrus state of the cattle when real-time biological data (250) is input (270).

[0120] FIG. 3 is a schematic diagram illustrating the structure of an artificial intelligence model (NLinear) used to predict the estrus behavior of cattle according to one embodiment of the present invention.

[0121] FIG. 3 illustrates the structure of an artificial intelligence model used to predict the estrus behavior of cattle using biological data, wherein the collected biological data (310) is preprocessed before being input into the artificial intelligence model (320).

[0122] In one embodiment of the present invention, the artificial intelligence model is configured to predict the estrus state (332) at a future point in time using data (331) at a past point in time L. Here, L is the sequence length of the input data and refers to the length of the past point in time referenced to calculate the prediction value.

[0123] Input data of length L consists of gastric temperature and all acceleration data. T is the sequence length of the output data, representing the length of future time points predicted by the model. For the given input data, the model outputs the degree of estrus for each time point T as a value between 0 and 1.

[0124] In one embodiment of the present invention, NLinear, a single linear model, was used as an artificial intelligence model for predicting the estrus behavior of cattle. When the value at the time to be predicted shows an upward or downward trend, the difference in distribution between datasets is large, and when normalizing with the mean and variance of the training data, the prediction accuracy may be low as it deviates from the distribution.

[0125] In the present invention as well, the difference in distribution between normal and estrus periods is large on average, which may result in a decrease in performance.

[0126] Therefore, we use an NLinear artificial intelligence model, which is a method of training by subtracting the last value of the input data and then adding it back.

[0127] The mean squared error between the model prediction value at time T and the actual estrus state of cattle at time T was used as the loss function for model training.

[0128]

[0129] In the loss function above, y i is the actual estrus state at time i. is the model-predicted estrus state at time i.

[0130] In the present invention, approximately 218,300 total data points were divided in chronological order and used as 70% for training, 10% for verification, and 20% for evaluation.

[0131] We designed an efficient training method that uses an EarlyStopping feature to terminate training early when the loss function value no longer decreases.

[0132] In addition, to optimize learning, the batch size, learning rate, and number of epochs were evaluated by changing them within the ranges of {512, 1024, 2048}, {0.0001, 0.0005, 0.001}, and {20, 50, 100}, respectively, and no significant difference in performance was found. In the present invention, the batch size was set to 1024, the learning rate to 0.0005, and the number of epochs to 100.

[0133] To perform model optimization, the model performance was measured by changing (1) the sequence length L of the input data, (2) the sequence length T of the output data, and (3) the composition of the input data.

[0134] Experiments were conducted by setting the sequence length L of the input data to {6, 36, 96, 168, 336, 3024}, respectively. These represent 1 hour, 6 hours, the default input sequence length of NLinear, and the typical estrous cycle of a cow, respectively.

[0135] The evaluation results show that performance is lowest when L is 6, and subsequent values ​​show similar performance, suggesting that consideration of factors other than L is more important. In the present invention, L was set to 168.

[0136] Experiments were conducted by setting the sequence length T of the output data to {6, 36, and 144}, respectively. These represent 1 hour, 6 hours, and 24 hours. The experimental results showed that performance was low when T became very large, such as 144, indicating that high performance could not be expected for long sequences. However, since the present invention prioritizes detecting the onset of estrus over long-term prediction, demonstrating high performance in short-term predictions is sufficiently meaningful. In this invention, T was set to 6.

[0137] Three cases were compared regarding the composition of input data: (i) using gastric temperature and all acceleration data, (ii) not using gastric temperature, and (iii) not using average acceleration. As a result, all three cases recorded similar performance. This suggests that excluding a small number of data terms does not affect learning because all data terms show a large change during the estrus period and prediction is performed on a univariate.

[0138] In the present invention, input data was constructed using all data terms.

[0139] To summarize, the estrus prediction model set the length of the input data sequence to 168 and the length of the output data sequence to 6, and constructed the input data using gastric temperature and all acceleration information.

[0141] This paper presents an experimental example of the process of training an artificial intelligence model for predicting estrus behavior using cattle biological data.

[0142] The length of the input sequence L was set to 168, which is slightly longer than one day, and the length of the output sequence T was set to 6, which is one hour. The Mean Squared Error was used as the loss function for training, and the input data consisted of gastric temperature and all acceleration information. Training was performed with the batch size, learning rate, and number of training iterations set to 1024, 0.0005, and 100, respectively. The estrus behavior prediction performance on the evaluation data after training is shown in Table 2 below.

[0143] [Table 2] Training performance of the estrous behavior prediction model

[0144]

[0145] The reason the performance metrics in Table 2 appear very high is, firstly, that the short length of the sequences to be predicted means that a single linear layer is sufficient to produce accurate prediction results. Secondly, due to severe class imbalance, only 1.6% of the data is in estrus; this implies that even if all data are assumed not to be in estrus, the accuracy can still reach nearly 99%. Therefore, the meaningful metric in this experiment is the F1 score. Finally, results were derived by replacing the predicted values ​​with 1.0 or 0.0, which indicates that significant performance differences can occur depending on the threshold set.

[0146] However, the recall, a metric for accurately predicting estrus during the actual estrus period, reached 0.96, and the MSE loss from the training was merely 0.0011; this indicates that the actual performance is excellent, not just the metrics. Experiments were conducted by substituting the predicted values ​​with a threshold of 1.0 and 0.0, respectively. However, if the threshold is set too low, the output sequence length T is short, resulting in a very high prediction success rate (due to the extremely small error) and potentially leading to confused results. Therefore, the threshold was set to 0.8 for the experiments.

[0147] A notable point is that among the experimental result metrics shown below, since EarlyStopping was used, there are data points where models that have completed a certain level of training exhibit the same metrics even when each condition is changed.

[0148] The following Table 3 is a table of results verified by varying the output sequence length T.

[0149] [Table 3] Learning results according to changes in output sequence length

[0150]

[0151] Except for the output sequence, the remaining settings were set identically with the input sequence length L = 168 and the threshold = 0.8. As can be seen in the results in Table 3, the metrics deteriorate as the length of the output sequence increases. If one wishes to increase the length of T, it would be necessary to lower the threshold criterion or adjust other learning methods.

[0152] [Table 4] Learning results according to changes in input sequence length

[0153]

[0154] The results obtained by varying the input sequence length L are shown in Table 4. Although the F1-score appears to increase slightly as the input sequence length increases, it does not show a significant difference in actual metrics. Therefore, while the input sequence length alone does not show a significant performance difference, it must be longer than the output sequence length T, and there is a possibility of achieving higher performance when the input sequence length and other values ​​are appropriately adjusted.

[0155] [Table 5] Learning results according to changes in threshold point

[0156]

[0157] Table 5 shows the experimental results when the threshold for determining the degree of estrus was varied. All other configurations were set identically, and it was observed that recall decreased to some extent as the threshold value was increased. However, since there was no significant difference when only the threshold value was changed while other settings were fixed, this value does not hold much significance as a standalone value; rather, there is potential for improvement when adjusted organically in conjunction with other values.

[0158] The differences in learning performance according to the configuration of input data are shown in Table 6. These are the experimental results when using gastric temperature and all acceleration information, when excluding gastric temperature, and when excluding the average acceleration. Since maximum and minimum acceleration data are used even when the average acceleration is excluded, it can be assumed that there is no performance degradation even if the average acceleration is excluded; however, it can be confirmed that performance drops when gastric temperature is excluded when the overall sequence length is shortened.

[0159] Therefore, unless acceleration is completely excluded, internal temperature is a feature that must be included; furthermore, since the performance reduction resulting from removing the feature is judged to be a greater loss compared to the increased efficiency, it would be desirable to use all features.

[0160] [Table 6] Training results according to input data configuration (L = 96, T = 6)

[0161]

[0163] FIGS. 4a and 4b are drawings showing the results of estrus behavior prediction using an artificial intelligence model (NLinear) used to predict estrus behavior of cattle according to one embodiment of the present invention.

[0164] The learning model proposed in this invention demonstrated prediction performance as shown in Figures 4a and 4b, and the confusion matrix was observed as shown in Table 7. Due to the characteristics of the NLinear model, errors in the prediction values ​​mainly occurred at the last data point, and it was confirmed that accurate estrous behavior was predicted at almost all time points with small errors. In addition, it was confirmed that the prediction values ​​were very stable when not in estrus, while the error values ​​were relatively large in prediction results that included the actual estrus period. Given that the recall and F1 scores were significantly high as well as the accuracy, this implies that the accuracy of behavior prediction was improved by effectively identifying attributes that aid in predicting estrous behavior, even though the number of estrous state data points was extremely small.

[0165] [Table 7] Confusion matrix for predicting estrus behavior using the NLinear model of the present invention (for cattle for evaluation)

[0166]

[0168] Figures 5a and 5b are diagrams showing the results of predicting estrus behavior using an MLP model compared to the artificial intelligence model (NLinear) of the present invention.

[0169] The comparison results of other models with respect to the artificial intelligence prediction model NLinear model are shown in Figures 5a, 5b, 6a, and 6b.

[0170] The results of predicting estrous behavior using a universal model, the MLP (Multi Layer Perceptron) (Figs. 5a and 5b), are illustrated.

[0171] For the MLP model (Figs. 5a and 5b), the input layer receives 42 input sequences, passes through 5 hidden layers, and is configured to calculate the estrus probability at the next time step in an output layer consisting of 2 units, and training was performed with a batch size of 1000, a learning rate of 0.0001, and 200 training iterations.

[0172] Figures 5a and 5b illustrate the performance on biological data from cattle for evaluation, representing the actual estrus records and the model's estrus prediction values ​​(estrus prediction probabilities) for a single input sequence (10-minute intervals). The horizontal axis represents the time flow of the biological data, and the vertical axis represents the estrus prediction probability. Experimental results showed that the MLP model exhibited a low recall rate, frequently predicting estrus even when the individual was not in an estrus state.

[0174] Figures 6a and 6b are diagrams showing the results of predicting estrus behavior using an LSTM model compared to the artificial intelligence model (NLinear) of the present invention.

[0175] An LSTM (Long Short-Term Memory) model, which is mainly used for time series data processing (Figs. 6a and 6b), was used.

[0176] In the case of the LSTM model in Figures 6a and 6b, it was configured to receive 42 input sequences and calculate the estrus probability at the next time step, and training was performed with a batch size of 432, a learning rate of 0.0001, and 200 training iterations.

[0177] This plot illustrates the performance on biological data of evaluation cattle, representing the actual estrus records and the model's estrus prediction values ​​(estrus prediction probabilities) for a single input sequence (10-minute intervals). The horizontal axis represents the time flow of the biological data, and the vertical axis represents the estrus prediction probability. Experimental results showed that the LSTM model tended to predict estrus around the time of the estrus state, but its accuracy was slightly lower.

[0179] FIG. 7 is a flowchart illustrating an artificial intelligence-based estrus behavior prediction method using livestock bio-information according to one embodiment of the present invention.

[0180] An artificial intelligence-based estrus behavior prediction method using livestock bio-information according to one embodiment of the present invention includes a bio-data collection step (S710), a preprocessing step (S720), an artificial intelligence model training step (S730), a performance indicator evaluation step (S740), and an artificial intelligence model optimization step (S750).

[0181] The biological data collection step (S710) collects biological data of cattle using a biosensor.

[0182] In one embodiment, the biosensor is installed in the rumen of the cattle, and the biological data may include the movement acceleration of the cattle and the biological temperature of the cattle.

[0183] The preprocessing step (S720) preprocesses the biological data.

[0184] In one embodiment, the preprocessing step includes a step of transforming the biological data into a predetermined data form and rearranging it.

[0185] The artificial intelligence model training step (S730) trains an artificial intelligence model to predict the estrus behavior of cattle using the preprocessed biological data.

[0186] The performance indicator evaluation step (S740) evaluates the performance indicators of the learned artificial intelligence model.

[0187] The artificial intelligence model optimization step (S750) optimizes the artificial intelligence model by adjusting at least one hyperparameter based on performance indicators.

[0188] In one embodiment, the method further includes a behavior labeling step for labeling the movement behavior of cattle when collecting biological data; and an artificial intelligence model is trained to predict whether the labeled movement behavior of the cattle is an estrus behavior.

[0189] In one embodiment, the method further includes a cattle estrus determination step that determines whether the cattle are in estrus based on collected real-time biological data and predicted estrus behavior using an optimized artificial intelligence model.

[0190] In one embodiment, the performance indicator is at least one of accuracy, recall, and F1 score.

[0192] FIG. 8 is a drawing illustrating an exemplary computing device capable of implementing a device and / or system according to various embodiments of the present invention.

[0193] Referring to FIG. 8, an exemplary computing device (800) capable of implementing devices according to some embodiments of the present disclosure will be described in more detail.

[0194] A computing device (800) may include one or more processors (810), a bus (850), a communication interface (870), a memory (830) for loading a computer program (891) executed by the processor (810), and a storage (890) for storing the computer program (891). However, only components related to embodiments of the present disclosure are illustrated in FIG. 8.

[0195] Therefore, a person skilled in the art to which this disclosure belongs will understand that other general-purpose components may be included in addition to the components shown in FIG. 8.

[0196] The processor (810) controls the overall operation of each component of the computing device (800). The processor (810) may be configured to include a CPU (Central Processing Unit), an MPU (Micro Processor Unit), an MCU (Micro Controller Unit), a GPU (Graphic Processing Unit), or any form of processor (810) well known in the art of the present disclosure. Additionally, the processor (810) may perform operations for at least one application or program for executing the method according to the embodiments of the present disclosure. The computing device (800) may have one or more processors (810). The computing device (800) may refer to artificial intelligence (AI).

[0197] The memory (830) stores various data, commands and / or information. The memory (830) may load one or more programs (891) from storage (890) to execute a method according to embodiments of the present disclosure. The memory (830) may be implemented as a volatile memory such as RAM, but the technical scope of the present disclosure is not limited thereto.

[0198] The bus (850) provides communication functions between components of the computing device (800). The bus (850) can be implemented as various types of buses, such as an address bus, a data bus, and a control bus.

[0199] The communication interface (870) supports wired and wireless internet communication of the computing device (800). Additionally, the communication interface (870) may support various communication methods other than internet communication. To this end, the communication interface (870) may be configured to include a communication module well known in the art of the present disclosure.

[0200] According to some embodiments, the communication interface (870) may be omitted.

[0201] Storage (890) can store one or more programs (891) and various data non-temporarily.

[0202] Storage (890) may be configured to include non-volatile memory such as ROM (Read Only Memory), EPROM (Erasable Programmable ROM), EEPROM (Electrically Erasable Programmable ROM), flash memory, a hard disk, a removable disk, or any form of computer-readable recording medium well known in the art to which this disclosure belongs.

[0203] A computer program (891) may include one or more instructions that cause a processor (810) to perform a method / operation according to various embodiments of the present disclosure when loaded into memory (830). That is, the processor (810) may perform a method / operation according to various embodiments of the present disclosure by executing the one or more instructions.

[0205] Although preferred embodiments of the present invention have been illustrated and described above, the present invention is not limited to the specific embodiments described above. Various modifications are possible by those skilled in the art without departing from the essence of the invention as claimed in the patent claims, and such modifications should not be understood individually from the technical spirit or perspective of the present invention.

Claims

Claim 1 An AI-based estrus behavior prediction device using cattle bio-information, comprising: a bio-data collection unit that collects bio-data including movement acceleration data and body temperature data of cattle using a bio-sensor installed in the rumen of cattle; a pre-processing unit configured to transform and rearrange the bio-data into a predetermined data form, including data from a past time point L with a sequence length L of input data; an AI model learning unit that takes the pre-processed bio-data as input and trains an AI model to predict the degree of estrus for a future preset time point T, which is the sequence length of output data, as a value between 0 and 1; and an AI model optimization unit configured to optimize the model by adjusting the hyperparameters of the AI ​​model, by determining the estrus state of the cattle when the predicted value for the future time point T output from the AI ​​model exceeds a preset threshold, and calculating a prediction error by comparing the determination result with the correct answer value of the learning data. Claim 2 delete Claim 3 delete Claim 4 The AI-based estrus behavior prediction device using cattle bio-information according to claim 1, further comprising a behavior labeling unit that labels the movement behavior of the cattle when collecting the bio-data, and wherein the learned AI model predicts whether the labeled movement behavior of the cattle is estrus behavior. Claim 5 delete Claim 6 An AI-based estrus behavior prediction device using cattle bio-information according to claim 1, wherein the AI ​​model optimization unit calculates a prediction error by comparing the judgment result with the correct answer value of the training data, and iteratively optimizes the AI ​​model by adjusting hyperparameters based on the prediction error. Claim 7 A method for predicting estrus behavior based on artificial intelligence using cattle bio-information, comprising: a step of collecting biological data including movement acceleration and biological temperature of cattle using a biosensor installed in the rumen of cattle; a step of a preprocessing unit preprocessing the biological data to include time series data from past time point L, the sequence length of input data L; a step of an artificial intelligence model learning unit learning to predict the degree of estrus for a future preset time point T, the sequence length of output data, as a value between 0 and 1, using the preprocessed biological data as input; and a step of an artificial intelligence model optimization unit determining the state of estrus when the predicted value for the future time point T output from the artificial intelligence model exceeds a preset threshold, calculating a prediction error by comparing the determination result with the correct answer value of the learning data, and optimizing the artificial intelligence model by adjusting hyperparameters based on the prediction error. Claim 8 delete Claim 9 delete Claim 10 The method of claim 7 further comprises a behavior labeling step of labeling the movement behavior of the cattle when collecting the biometric data; and the learned artificial intelligence model predicts whether the labeled movement behavior of the cattle is an estrus behavior. Claim 11 delete Claim 12 A method for predicting estrus behavior based on artificial intelligence using cattle bio-information according to claim 7, wherein the step of optimizing the artificial intelligence model comprises the step of calculating a prediction error by comparing the judgment result with the correct answer value of the training data, and iteratively optimizing the artificial intelligence model by adjusting hyperparameters based on the prediction error.

Images