Intelligent sow health early warning system and method based on auditory meatus core temperature monitoring

Abstract

The invention provides an intelligent sow health early warning system and method based on auditory meatus core temperature monitoring. The intelligent sow health early warning system comprises a wearable ear tag, a data analysis module and a health early warning module. The wearable ear tag is used for collecting core temperature data of the ear canal of the sow; the data analysis module is used for receiving the core temperature data, performing time sequence analysis on the core temperature data and constructing a sow health assessment model; and the health early warning module is used for locking the sow with abnormal health and performing abnormal early warning by adopting a preset early warning rule based on an evaluation result of the sow health evaluation model. According to the technical scheme, real-time and accurate monitoring of the health condition of the sow is realized, and the breeding management efficiency and the disease prevention capability are effectively improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent breeding and health monitoring, and particularly relates to an intelligent sow health early warning system and method based on ear canal core temperature monitoring. Background Art

[0002] Low efficiency of traditional body temperature monitoring: Existing ear tags mostly adopt infrared temperature measurement technology. However, the ear canal environment is complex and is easily interfered by external temperatures, resulting in insufficient accuracy (±0.5°C error) and the inability to obtain core temperature data. Manual inspection relies on handheld temperature measurement equipment, which is time-consuming and has a high missed inspection rate, and continuous 24-hour monitoring cannot be achieved. Poor generalization of health early warning models: Existing systems mostly give early warnings based on a single temperature threshold and do not combine multi-dimensional data such as exercise volume. The false alarm rate is as high as over 30%. Disease prediction relies on manual experience and lacks the ability of dynamic time series analysis, making it difficult to detect early sub-healthy states in a timely manner. Insufficient equipment adaptability and reliability: The ear tags are heavy (some products exceed 15 g), and the wearing comfort is poor, resulting in frequent shaking and falling off of sows.

[0003] As an important physiological parameter, the core temperature of the ear canal can sensitively reflect the health status of sows (such as estrus, infection, stress response, etc.). However, there is currently a lack of an intelligent sow health early warning system and method based on ear canal core temperature monitoring. Summary of the Invention

[0004] The present invention provides an intelligent sow health early warning system and method based on ear canal core temperature monitoring, which deeply integrates precise sensing technology, biometric modeling, and artificial intelligence algorithms, and solves key problems such as low monitoring accuracy, late early warning, and poor equipment reliability in traditional breeding.

[0005] An intelligent sow health early warning system based on ear canal core temperature monitoring includes: a wearable ear tag, a data analysis module, and a health early warning module;

[0006] The wearable ear tag is used to collect core temperature data of the sow's ear canal;

[0007] The data analysis module is used to receive the core temperature data and perform time series analysis on the core temperature data to construct a sow health assessment model;

[0008] The health early warning module is used to lock sows with abnormal health and give abnormal early warnings based on the evaluation results of the sow health assessment model by using a preset early warning rule.

[0009] Preferably, the wearable ear tag includes:

[0010] A temperature data acquisition module, which is used to collect core temperature data of the sow's ear canal based on a temperature sensor;

[0011] A control module for performing primary filtering and dynamic temperature compensation on the core temperature data to obtain compensated core temperature data;

[0012] A Bluetooth communication module for transmitting the compensated core temperature data;

[0013] A power supply module for powering the temperature sensor based on a button battery.

[0014] Preferably, the process of constructing the temperature sensor includes:

[0015] Based on three-dimensional scanning technology, obtaining anatomical structure data of the sow ear canal at different growth stages and physiological states, and constructing a three-dimensional ear canal model;

[0016] Integrating a micro-thermistor on a flexible substrate and wrapping it from the inside out with a plasticizable material and skin-friendly silica gel to obtain an initial temperature sensor;

[0017] Based on the three-dimensional ear canal model and the ear canal variation range, adjusting the shape of the initial temperature sensor to obtain the final temperature sensor.

[0018] Preferably, the data analysis module includes:

[0019] A data receiving unit for receiving the core temperature data;

[0020] A multi-scale window analysis unit for capturing short-term fluctuation characteristics of the core temperature data based on a preset time scale;

[0021] A long short-term memory network unit for capturing long-term dependence characteristics of the core temperature data based on an LSTM extended architecture;

[0022] A feature classification unit for extracting key features of the short-term fluctuation characteristics and the long-term dependence characteristics, and classifying the key features using an improved random forest algorithm to obtain a classification result;

[0023] An evaluation model construction unit for performing sow health level evaluation based on the classification result and completing the construction of the sow health evaluation model.

[0024] Preferably, the process of improving the random forest algorithm includes:

[0025] Training a random forest network using an existing feature set to obtain the average value of the reduction in node purity when each feature is used to split nodes in all decision trees during the training process;

[0026] Calculating a feature importance score based on the average value of the reduction in node purity;

[0027] Normalize the importance scores of the features to obtain feature sampling weights;

[0028] During each node split, sample the features to be sampled according to the feature sampling weights to obtain a feature subset with weighted random sampling;

[0029] In the feature subset, select the feature with the maximum information gain as the split point to complete the improvement of the random forest algorithm.

[0030] Preferably, the health warning module includes:

[0031] An evaluation result tracking unit for locating healthy abnormal sows based on the evaluation results and obtaining the abnormal duration and the degree of abnormality;

[0032] A warning unit for performing hierarchical warnings based on the abnormal duration and the degree of abnormality.

[0033] The present invention also provides an intelligent sow health warning method based on ear canal core temperature monitoring for implementing the system, including:

[0034] Collect the core temperature data of the sow's ear canal;

[0035] Receive the core temperature data and perform time series analysis on the core temperature data to construct a sow health assessment model;

[0036] Based on the evaluation results of the sow health assessment model, use a preset warning rule to lock healthy abnormal sows and perform abnormal warnings.

[0037] Preferably, the method for constructing the sow health assessment model includes:

[0038] Receive the core temperature data;

[0039] Capture the short-term fluctuation characteristics of the core temperature data based on a preset time scale;

[0040] Capture the long-term dependence characteristics of the core temperature data based on the LSTM extended architecture;

[0041] Extract the key features of the short-term fluctuation characteristics and the long-term dependence characteristics, and classify the key features using an improved random forest algorithm to obtain a classification result;

[0042] Based on the classification result, perform a sow health level assessment to complete the construction of the sow health assessment model.

[0043] Compared with the prior art, the beneficial effects of the present invention are as follows: By directly collecting the core temperature data of the sow's ear canal through the wearable ear tag, compared with traditional body surface temperature measurement, it can better reflect the true body temperature status of the sow and improve the accuracy of temperature measurement. The system can continuously monitor the sow's body temperature for 24 hours, promptly detect abnormal body temperature, and avoid delaying the treatment time. The data analysis module performs time series analysis on the core temperature data, can capture the subtle changes in the sow's body temperature, and construct a more accurate health assessment model. The health assessment model can more comprehensively evaluate the health status of the sow, reduce false alarms and missed reports, and improve the reliability of early warning. This system is an important part of the intelligent management of the breeding industry and can be linked with other intelligent devices (such as intelligent feeding systems, environmental control systems) to achieve comprehensive intelligent management of the farm. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions of the present invention, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0045] Figure 1 It is a schematic structural diagram of an intelligent sow health early warning system based on core ear canal temperature monitoring according to an embodiment of the present invention;

[0046] Figure 2 It is a curve graph showing the change of the core temperature of the sow's ear canal over time according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0047] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art without creative efforts based on the embodiments of the present invention belong to the scope of protection of the present invention.

[0048] To make the above objects, features, and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below with reference to the drawings and specific embodiments.

[0049] Embodiment 1

[0050] As Figure 1 、 Figure 2 shown, an intelligent sow health early warning system based on core ear canal temperature monitoring includes: a wearable ear tag, a data analysis module, and a health early warning module;

[0051] The wearable ear tag is used to collect the core temperature data of the sow's ear canal.

[0052] A further embodiment is that the wearable ear tag includes: a temperature data acquisition module, a control module, a Bluetooth communication module, and a power supply module.

[0053] The temperature data acquisition module is used to collect the core temperature data of the sow's ear canal based on a temperature sensor; in this embodiment, a further embodiment is that the process of constructing the temperature sensor includes:

[0054] Based on three-dimensional scanning technology, obtain the anatomical structure data of the sow's ear canal in different growth stages and physiological states, and construct a three-dimensional model of the ear canal; specifically, use a structured light scanner (accuracy ±0.05mm) and a thermal imager to collect synchronously to solve the problem of in vivo motion artifacts. Among them, structured light scanning generates point cloud data of the sow's ear canal through grating projection and phase decoding. That is, the projector projects a coded stripe pattern onto the ear canal, and two high-resolution cameras (binocular system) capture the stripes deformed by the ear canal surface, and calculate the absolute phase value through the phase shift method to establish a pixel-three-dimensional coordinate mapping:

[0055]

[0056] (I0~I3) are intensity images collected by the four-step phase shift method).

[0057] Paste reflective marker points (diameter 2mm) on the ear, and construct a motion trajectory equation through 6 high-speed infrared cameras (1000fps).

[0058] Then use the improved time series ICP algorithm to align the point cloud data of different growth stages:

[0059]

[0060] Among them, R t represents the rotation matrix (3×3 orthogonal matrix) at time t, t t represents the translation vector (3×1 vector) at time t, represents the coordinate (3D vector) of the i-th point at time t, represents the coordinate of the corresponding point at time t+Δt, λ represents the regularization coefficient, which controls the smoothness of the rigid body transformation and avoids sudden changes in the model between adjacent stages. ||·|| F is the Frobenius norm (the square root of the sum of the squares of the matrix elements).

[0061] Calculate the differential geometric properties of the point cloud through the moving least squares method (MLS):

[0062] Retain κ combFeature points greater than 0.2 are retained, and key anatomical structures such as the cochlear helix and vascular depressions are preserved. Among them, κ1 represents the first principal curvature, κ2 represents the second principal curvature, and κ comb is the combined curvature, which is used for threshold judgment of feature retention.

[0063] Motion compensation is achieved using the motion estimation equation and the IPC algorithm, and feature extraction is performed using the least squares method.

[0064] After obtaining the point cloud data of the sow's ear canal at different growth stages through three-dimensional scanning, the offset optimization algorithm is used to perform solid modeling on the point cloud. This strategy eliminates the modeling error caused by individual morphological differences by dynamically adjusting the offset between the scanned data and the standard ear canal model, ensuring that the model can reflect both the group commonality and retain individual characteristics. Specifically, an implicit surface function F(x) is defined, combined with the radial basis function (RBF) and growth constraints:

[0065]

[0066] φ(r) = r 3 represents the cubic radial basis function, P(x) represents the quadratic polynomial to compensate for the global shape, G(t, x) = α(t)·||x|| represents the time growth term, α(t)·||x||, and α(t) is the time-dependent growth rate. w i represents the weight coefficient of the i-th control point.

[0067] An energy minimization objective function is established:

[0068]

[0069] d k represents the signed distance from the point cloud k to the surface (d > 0 outside, d < 0 inside), represents the gradient vector of the implicit function, n represents the normal field of the point cloud, β represents the weight of the smoothing term, and γ represents the weight of the distance constraint. The normal field-guided optimization algorithm improves the surface smoothness by 40%.

[0070] The finite element method is used for discrete solution, and the mesh size is adaptive (0.1 mm in the narrow area of the ear canal and 0.5 mm in the periphery), completing the solid modeling and obtaining the three-dimensional model of the sow's ear canal.

[0071] Using the parametric growth tensor, a growth dynamic constraint is established for the three-dimensional model of the sow's ear canal. Specifically, the ear canal growth velocity field is defined:

[0072]

[0073] g radial (t) represents the radial growth rate, g axial (t) represents the axial growth rate (along the depth direction of the ear canal), gspiral (t) represents the spiral growth rate (controlling the cochlear morphological evolution), e r ,e θ ,e z is the unit basis vector of the cylindrical coordinate system.

[0074] And dynamically adjust the network topology based on the sow growth rate to complete the growth dynamic constraint of the three-dimensional model of the sow ear canal.

[0075] Finally, output the three-dimensional solid model of the sow ear canal:

[0076] V k represents the vertex set, F k represents the face set, A k represents the attribute set (curvature, thickness, and growth stage label).

[0077] Cluster analyze the three-dimensional solid models of the ear canals of pigs in different physiological states such as pregnancy and lactation, extract key parameters such as ear canal volume and curvature radius, and establish a database of ear canal morphological variations.

[0078] Optimize the "safe threshold range" of the ear canal model based on the genetic algorithm to determine the boundary conditions for the shape adjustment of the sensor, and avoid fitting failure caused by sudden changes in ear canal size.

[0079] Integrate the micro-thermistor on a flexible substrate, and wrap it from the inside to the outside with a plastic material and skin-friendly silicone to obtain an initial temperature sensor. In this embodiment, a composite substrate of polyurethane foam and a conductive polymer (such as PEDOT:PSS) is used to construct a porous microstructure through 3D printing technology, which not only realizes the high-precision embedding of the thermistor, but also improves the ductility (stretch ratio up to 200%) and heat conduction efficiency of the substrate. Add a nano-silver antibacterial coating to the outer layer of the silicone, and reduce the risk of ear canal friction and allergy through surface micro-texture design (biomimetic shark skin structure).

[0080] Based on the three-dimensional ear canal model and the ear canal variation range, adjust the shape of the initial temperature sensor to obtain the final temperature sensor. In this embodiment, based on the sensor stiffness matrix and the soft tissue stiffness matrix of the sow ear canal, a contact mechanics model is constructed to solve the contact pressure distribution and perform virtual wearing simulation.

[0081] The control module is used to perform primary filtering and dynamic temperature compensation on the core temperature data to obtain the compensated core temperature data; the built-in low-power processing unit is used to process the collected temperature data in real time and perform preliminary filtering. In this embodiment, based on the heat generation rate of blood flow metabolism and the thermal expansion coupling coefficient of the sow ear, a heat conduction-deformation coupling equation is constructed to complete the dynamic temperature compensation of the core temperature data.

[0082] A Bluetooth communication module is used to transmit the compensated core temperature data. Specifically, a preset wireless communication protocol is adopted to segment the compressed temperature data to generate multiple data segments. The check value of each data segment is calculated through a redundancy check mechanism, and the check value is appended to the end of the data segment. During the transmission process, the signal interference intensity is monitored in real time. If the interference intensity is higher than the preset threshold, the data retransmission mechanism is triggered. The lost data segment is located according to the packet loss information feedback by the receiving end, and the lost data segment is retransmitted from the source end. An error correction algorithm is used to verify the check value of the received data segment. If the check values do not match, the error correction process is triggered. The error part in the data segment is repaired through the error correction process to generate a complete and accurate data segment. The repaired data segments are recombined to restore the complete compressed temperature data.

[0083] A power supply module is used to supply power to the temperature sensor based on a button battery.

[0084] A data analysis module is used to receive the core temperature data and perform a time series analysis on the core temperature data to construct a sow health assessment model. A further implementation method is that the data analysis module includes:

[0085] A data receiving unit is used to receive the core temperature data. Specifically, a preset distributed storage architecture is adopted to receive the uploaded temperature data to obtain the original temperature data. The original temperature data is classified according to the sow individual identification to generate a classified temperature data set. For the classified temperature data set, an independent data storage unit is constructed to generate a storage unit identification. The classified temperature data is stored in the corresponding independent storage unit through the storage unit identification. If data loss occurs during the storage process, the lost data is retrieved from the source end according to the storage unit identification. A preset data verification algorithm is used to perform integrity verification on the temperature data in the storage unit. According to the verification result, if the data is complete, the storage is completed; if the data is incomplete, the data repair process is triggered.

[0086] A multi-scale window analysis unit is used to capture the short-term fluctuation characteristics of the core temperature data based on a preset time scale. In this embodiment, multiple time scale windows are preset to ensure 50% overlap between the windows. The window size is adaptively adjusted according to the data sampling frequency and signal characteristics. The short-term fluctuation characteristics include time domain characteristics, frequency domain characteristics, and non-linear characteristics. Among them, the time domain characteristics include calculating the mean, standard deviation, skewness, and kurtosis within each window. The frequency domain characteristics include performing a fast Fourier transform on the core temperature data within the window to obtain the main frequency and spectral entropy. The non-linear characteristics include sample entropy for measuring the complexity of the time series and Lyapunov exponent for evaluating the chaotic characteristics. The characteristics of different windows are spliced into a high-dimensional vector, and t-SNE is used to map the high-dimensional features to a low-dimensional space, retaining the local structure information.

[0087] Long short-term memory network units are used to capture the long-term dependence features of core temperature data based on the LSTM extended architecture; in this embodiment, an exponential gate (sLSTM) and a new storage mixing technology are introduced into the LSTM, allowing the model to revise storage decisions and enhancing the control of the information flow. The memory unit is extended from a scalar to a matrix (mLSTM), increasing the storage capacity, and a covariance update rule is introduced to support parallel processing. The sLSTM and mLSTM are integrated into a residual block, and the xLSTM architecture is constructed by stacking to enhance the nonlinear expression ability of the model.

[0088] Specifically, integrating the sLSTM and mLSTM into the residual block alleviates the problem of gradient vanishing:

[0089] H t = LayerNorm(H t-1 + sLSTM(H t-1 ) + mLSTM(H t-1 ))

[0090] LayerNorm: Layer normalization, H t represents the hidden state matrix.

[0091] The xLSTM network is constructed by stacking multiple residual blocks:

[0092] where L is the number of residual block layers (default L = 6).

[0093] Using the constructed xLSTM network for long-term dependence feature extraction:

[0094] A temporal convolutional module TCN is added before the xLSTM to extract short-term local features, and dilated convolution is used to expand the receptive field. Multi-head self-attention is introduced into each layer of the xLSTM, and the attention outputs of different layers are weighted and fused to obtain the weighted fusion features. The long-term dependence of the weighted fusion features is captured through the memory matrix of the mLSTM, and the global dependence relationship of the time series is modeled using the covariance matrix to complete the extraction of long-term dependence features.

[0095] Feature classification units are used to extract the key features of short-term fluctuation features and long-term dependence features, and the improved random forest algorithm is used to classify the key features to obtain the classification results; in this embodiment, a further implementation method is that the process of improving the random forest algorithm includes:

[0096] Training a random forest network using the existing feature set to obtain the average value of the reduction in node purity when each feature is used to split nodes in all decision trees during the training process;

[0097] Calculating the feature importance score based on the average value of the reduction in node purity;

[0098] Normalize the feature importance scores to obtain feature sampling weights;

[0099] At each node split, sample the features to be sampled according to the feature sampling weights to obtain a feature subset for weighted random sampling;

[0100] In the feature subset, select the feature that maximizes the information gain as the split point to complete the improvement of the random forest algorithm.

[0101] An evaluation model construction unit is used to evaluate the health level of sows based on the classification results to complete the construction of the sow health evaluation model.

[0102] A health warning module is used to lock the sows with abnormal health and issue abnormal warnings based on the evaluation results of the sow health evaluation model using preset warning rules.

[0103] A further implementation manner is that the health warning module includes:

[0104] An evaluation result tracking unit is used to locate the sows with abnormal health based on the evaluation results and obtain the duration and degree of abnormality; specifically, count the length of consecutive abnormal time, record the start and end times [t start , t end of each abnormal segment, and calculate the duration. Calculate the abnormality intensity based on the deviation between the evaluation result and the preset health threshold, and comprehensively score each abnormal segment:

[0105]

[0106] where τ is the attenuation coefficient (default τ = 10) for weighting recent abnormalities.

[0107] A warning unit is used to issue hierarchical warnings based on the duration and degree of abnormality.

[0108] Specifically, according to the abnormal duration Δt and the abnormal degree I total , define three-level warnings:

[0109] Low-risk warning: Δt ∈ [T1, T2) and I total < I1,

[0110] Medium-risk warning: Δt ∈ [T2, T3) or I total ∈ [I1, I2),

[0111] High-risk warning: Δt ≥ T3 or I toal ≥ I2.

[0112] The duration ranges T1, T2, and T3 can be user-defined.

[0113] The early warning content includes: 1. Sow ID; 2. Abnormal start and end time; 3. Abnormal duration; 4. Abnormal degree score; 5. Early warning level.

[0114] The early warning information is pushed in the following ways:

[0115] Mobile push: Send it to the breeding management APP in real time;

[0116] SMS notification: Send high-risk early warnings to the administrator's mobile phone;

[0117] Acoustic and optical alarm: Trigger the acoustic and optical alarm device in the farm.

[0118] The model of the present invention continuously learns the individual temperature data and dynamically updates the reference value and the health threshold.

[0119] The system provides a manual marking function (confirming actual events such as infection and parturition), and feeds back to optimize the model.

[0120] Regularly optimize the algorithm using all individual data to ensure both group applicability and individual accuracy.

[0121] The present invention can accurately identify the estrus period, pre-parturition state and disease risk of sows, timely provide early warning signals to breeding management personnel, and optimize production management. It supports docking with the existing breeding management system to form a closed loop of health management. The data is stored in the cloud for a long time, supports the generation of multi-dimensional health reports, and provides a scientific basis for breeding optimization.

[0122] Embodiment 2

[0123] The present invention also provides an intelligent sow health early warning method based on ear canal core temperature monitoring for implementing the system, including:

[0124] Collect the core temperature data of the sow's ear canal;

[0125] Receive the core temperature data, perform time series analysis on the core temperature data, and construct a sow health assessment model;

[0126] Based on the evaluation results of the sow health assessment model, use the preset early warning rules to lock the sows with health abnormalities and give early warnings.

[0127] A further implementation method is that the method for constructing the sow health assessment model includes:

[0128] Receive the core temperature data;

[0129] Capture the short-term fluctuation characteristics of the core temperature data based on a preset time scale;

[0130] Capture the long-term dependence characteristics of the core temperature data based on the LSTM extended architecture;

[0131] Extract the key features of short-term fluctuation features and long-term dependence features, and use the improved random forest algorithm to classify the key features to obtain the classification results;

[0132] Based on the classification results, conduct the health level assessment of sows and complete the construction of the sow health assessment model.

[0133] The embodiments described above are only descriptions of the preferred embodiments of the present invention, and do not limit the scope of the present invention. Without departing from the design spirit of the present invention, various deformations and improvements made by those of ordinary skill in the art to the technical solutions of the present invention shall fall within the protection scope determined by the claims of the present invention.

Claims

1. An intelligent sow health early warning system based on ear canal core temperature monitoring, characterized in that: include: Wearable ear tags, data analysis modules, and health warning modules; The wearable ear tag is used to collect core temperature data of the sow's ear canal; The data analysis module is used to receive the core temperature data, perform time series analysis on the core temperature data, and construct a sow health assessment model; The health warning module is used to lock in abnormally healthy sows and issue abnormal warnings based on the evaluation results of the sow health evaluation model and using preset warning rules.

2. The system according to claim 1, characterized in that The wearable ear tag comprises: A temperature data acquisition module is used to collect the core temperature data of the sow's ear canal based on a temperature sensor; A control module, used for performing primary filtering and dynamic temperature compensation on the core temperature data to obtain compensated core temperature data; Bluetooth communication module for transmitting compensated core temperature data; A power supply module is used to supply power to the temperature sensor based on a button battery.

3. The system according to claim 2, characterized in that The process of building the temperature sensor includes: Based on 3D scanning technology, the anatomical structure data of the sow ear canal at different growth stages and physiological states are obtained, and a 3D model of the ear canal is constructed; The micro-thermistor is integrated on a flexible substrate and wrapped from the inside to the outside with plastic material and skin-friendly silicone to obtain an initial temperature sensor; Based on the ear canal three-dimensional model and the ear canal variation range, the initial temperature sensor is adjusted in shape to obtain the final temperature sensor.

4. The system according to claim 1, characterized in that The data analysis module includes: A data receiving unit, configured to receive the core temperature data; A multi-scale window analysis unit, used to capture short-term fluctuation characteristics of the core temperature data based on a preset time scale; A long short-term memory network unit, for capturing long-term dependency characteristics of the core temperature data based on an LSTM extended architecture; A feature classification unit, used to extract key features of the short-term fluctuation feature and the long-term dependency feature, and classify the key features using an improved random forest algorithm to obtain a classification result; The evaluation model construction unit is used to evaluate the sow health level based on the classification result and complete the construction of the sow health evaluation model.

5. The system according to claim 4, characterized in that The process of improving the random forest algorithm includes: Use the existing feature set to train a random forest network and obtain the average value of the reduction in node purity when each feature is used to split nodes in all decision trees during training; Calculate feature importance scores based on the average of node purity reduction; Normalizing the feature importance scores to obtain feature sampling weights; When each node is split, the features to be sampled are sampled according to the feature sampling weights to obtain a weighted randomly sampled feature subset; In the feature subset, the feature with the maximum information gain is selected as the splitting point to improve the random forest algorithm.

6. The system according to claim 1, characterized in that The health warning module includes: An evaluation result tracking unit, used for locating healthy abnormal sows based on the evaluation results, and obtaining the duration and degree of abnormality; The early warning unit is used to provide graded early warnings based on the duration and severity of the abnormality.

7. An intelligent sow health early warning method based on ear canal core temperature monitoring, used to implement the system according to any one of claims 1 to 6, characterized in that: include: Collect core temperature data from the sow’s ear canal; receiving the core temperature data, performing time series analysis on the core temperature data, and constructing a sow health assessment model; Based on the evaluation results of the sow health evaluation model, preset warning rules are adopted to lock in abnormally healthy sows and issue abnormal warnings.

8. The method according to claim 7, characterized in that The method for constructing the sow health assessment model comprises: receiving the core temperature data; capturing short-term fluctuation characteristics of the core temperature data based on a preset time scale; Capturing the long-term dependency characteristics of the core temperature data based on the LSTM extended architecture; Extracting key features of the short-term fluctuation features and the long-term dependence features, and classifying the key features using an improved random forest algorithm to obtain classification results; Based on the classification results, the sow health level is assessed to complete the construction of the sow health assessment model.

Images