Goat oestrus monitoring system

Abstract

The invention discloses a goat oestrus monitoring system. The goat oestrus monitoring system comprises a multi-mode infrared thermal imaging data acquisition sensing unit and a multi-dimensional environment parameter collaborative real-time sensing unit. A goat body surface temperature field and a multispectral image are acquired through an uncooled vanadium oxide focal plane array infrared detector and a multiband spectral imaging module, environmental parameters are acquired in combination with a distributed sensing network, and a motion track is recorded through depth vision and inertia measurement. The heterogeneous multi-core data processing central unit is used for preprocessing, storing and preliminarily analyzing multi-source data, the feature extraction unit is used for mining multi-dimensional features, and the intelligent decision judgment unit is used for judging an estrus state by combining a deep neural network and dynamic threshold adjustment. The system integrates multiple technologies, can accurately monitor the oestrus state of the goats in real time, provides data support for scientific breeding, and improves breeding benefits.

Description

Technical Field

[0001] The present invention relates to the field of goat behavior monitoring, and particularly to a goat estrus monitoring system. Background Art

[0002] With the large-scale and intelligent development of the livestock industry, accurately monitoring the estrus status of goats is crucial for improving breeding efficiency and increasing farming income. Traditional manual observation methods rely on the experience of breeders, which not only consume a large amount of manpower and time, but also have problems of strong subjectivity and low judgment accuracy. In large-scale farming scenarios, it is difficult to achieve real-time and accurate monitoring of a large number of goats through manual observation, easily missing the best breeding time and affecting breeding efficiency and farming benefits. At the same time, this method cannot achieve digital recording and analysis of data, making it difficult to form a scientific farming management strategy.

[0003] Although existing automated monitoring technologies have been applied, there are still many limitations. Some sensor-based monitoring systems only collect a certain physiological index or behavioral data of goats singlely, such as monitoring body temperature changes through a body temperature sensor. However, the body temperature of goats is greatly affected by environmental factors, exercise status, etc. A single index is difficult to accurately judge the estrus status, prone to false positives and false negatives. For monitoring systems based on image recognition, in complex farming environments, the image quality is interfered by factors such as light and goat occlusion, resulting in inaccurate feature extraction and unstable monitoring effects.

[0004] In addition, existing monitoring technologies lack the ability to comprehensively process and deeply analyze multi-source data. Environmental parameters such as temperature, humidity, and light in the farming environment will affect the physiological state and behavioral performance of goats, but existing systems often ignore the correlation between environmental factors and goat estrus characteristics, unable to comprehensively and accurately analyze the goat estrus status. At the same time, traditional monitoring technologies are difficult to adapt to the differences in different farming environments and goat breeds, lacking a dynamic adjustment and optimization mechanism, and difficult to meet diverse farming needs, restricting the intelligent development process of the goat farming industry. Summary of the Invention

[0005] In order to overcome the shortcomings and deficiencies of the existing technology, the present invention provides a goat estrus monitoring system.

[0006] A goat estrus monitoring system, the system includes:

[0007] A multimodal infrared thermal imaging data acquisition unit, which integrates a non-cooled focal plane infrared detector array and a multi-spectral imaging component, and is connected to the data processing unit through a fiber optic interface, and is used to collect the body surface temperature distribution and multi-spectral image data of goats;

[0008] The multi-dimensional environmental parameter collaborative real-time perception unit includes a temperature and humidity sensor array, a light intensity sensor, a barometric pressure sensor, and a wind direction and speed measurement device, which establish a communication link with the data processing unit and are used to collect environmental temperature and humidity, light, barometric pressure, and wind speed and direction parameters in real time;

[0009] The three-dimensional space motion trajectory and attitude capture unit is composed of multiple groups of depth cameras and inertial measurement units. The depth cameras perform three-dimensional modeling based on the TOF (Time of Flight) principle and are connected to the data processing unit through a gigabit Ethernet interface, and are used to record the motion trajectory and attitude information of the goats;

[0010] The heterogeneous multi-core data processing central unit is equipped with a multi-core heterogeneous processor, with a dedicated neural network acceleration chip built-in, and is configured with a large-capacity cache and a solid-state drive storage module. It is respectively connected to the multi-modal infrared thermal imaging acquisition unit, the environmental parameter collaborative perception unit, the motion trajectory capture unit, the feature extraction unit, and the decision-making judgment unit through a high-speed data bus, and is used for preprocessing and preliminary analysis of the collected data;

[0011] The multi-dimensional feature extraction and analysis unit adopts a reconfigurable computing architecture, integrates a spatio-temporal feature extraction module based on the attention mechanism, and is connected to the data processing unit through a PCI-Express interface, and is used to extract multi-dimensional feature vectors including temperature features, spectral features, environmental features, and motion features from the preprocessed data;

[0012] The intelligent decision-making judgment and result output unit includes a deep neural network model based on transfer learning and is connected to the feature extraction unit through a high-speed serial interface, and is used to judge whether the goats are in estrus according to the multi-dimensional feature vectors.

[0013] Furthermore, for the multi-modal infrared thermal imaging data acquisition unit, the pixel resolution of the uncooled vanadium oxide (VOx) focal plane array infrared detector is not less than 640×480, the temperature resolution reaches 0.05°C, the pixel size is 17μm×17μm, the multi-band spectral imaging module covers the visible and near-infrared bands of 400nm - 1100nm, includes at least 3 independent spectral channels, the autofocus actuator is based on the phase difference detection algorithm, and controls the stepping motor to drive the variable focal ratio optical lens group for focal length adjustment. The zoom ratio of the optical lens group is not less than 3:1, and the lens surface is coated with a multi-layer broadband antireflection film to ensure clear imaging of the goat body surface; at the same time, the multi-modal infrared thermal imaging data acquisition unit is also configured with a temperature calibration module based on a blackbody radiation source, and non-linearly corrects the original temperature data output by the detector according to Planck's law. The correction formula is:

[0014]

[0015] Where T calibratedis the calibrated temperature value, C1 and C2 are Planck constants, λ is the infrared radiation wavelength, and V raw is the original voltage signal output by the detector.

[0016] Furthermore, for the multi-dimensional environmental parameter collaborative real-time sensing unit, the distributed temperature and humidity sensing network includes 3 sensor nodes at different positions. Each node uses a high-precision digital temperature and humidity sensor. The temperature measurement range is -40°C to 125°C, with an accuracy of ±0.1°C, and the humidity measurement range is 0-100%RH, with an accuracy of ±1.5%RH. The Kalman filter algorithm is used to fuse the temperature and humidity data collected by each node. The state prediction equation of the Kalman filter is:

[0017]

[0018] Wherein, is the state prediction at time k based on the state at time k-1, and F k is the state transition matrix, is the optimal state estimate at time k-1, and B k is the control input matrix, and u k is the control input. The light intensity detection module uses a silicon photovoltaic cell as the sensing element. Its measurement range is 0-200000 lux, and it has an automatic gain adjustment circuit inside to automatically adjust the amplification factor according to the ambient light intensity; the air pressure monitoring device has an accuracy of ±0.1 hPa. The ultrasonic wind speed and direction measurement component uses the ultrasonic wind measurement principle. The wind direction measurement accuracy is ±3°, and the wind speed measurement range is 0-70 m / s. And this component is equipped with a wind shield to reduce the interference of environmental factors on the measurement results; the multi-dimensional environmental parameter collaborative real-time sensing unit also sets an environmental parameter correlation analysis module based on the Pearson correlation coefficient to calculate the correlation between each environmental parameter. The calculation formula is:

[0019]

[0020] Wherein, r xy is the Pearson correlation coefficient of variables x and y, and x i , y i are respectively the i-th observed values of variables x and y, are respectively the means of variables x and y, and n is the number of observed values.

[0021] Further, in the three-dimensional space motion trajectory and attitude capture unit, the cameras of the multi-view depth vision acquisition array based on the structured light principle are arranged in a circular array, and the included angle between adjacent cameras is 45°. Based on the multi-view stereo vision principle, the position coordinates of the goat in the three-dimensional space are calculated through the feature point matching algorithm. The depth vision acquisition array adopts the structured light technology, projects an infrared pattern of a specific pattern to assist in depth measurement, and its effective measurement distance is 0.5 m - 5 m. The depth measurement accuracy reaches ±1 mm at a distance of 1 m. The high-precision inertial measurement combination unit collects data at a sampling frequency of 100 Hz, including a three-axis accelerometer and a three-axis gyroscope. The measurement range of the accelerometer is ±16 g, and the measurement range of the gyroscope is ±2000°. The quaternion method is used to fuse the accelerometer and gyroscope data to obtain the attitude angle information of the goat. The three-dimensional space motion trajectory and attitude capture unit is also equipped with an anti-occlusion supplementary light device, which is automatically turned on when the ambient light is insufficient. The supplementary light device adopts an infrared LED array with a wavelength of 850 nm and a light-emitting angle of 120°. At the same time, the three-dimensional space motion trajectory and attitude capture unit is configured with a motion mode recognition module based on the hidden Markov model (HMM) for modeling and recognizing the motion mode of the goat. The state transition probability matrix of the hidden Markov model is:

[0022] A = [a ij N×N

[0023] where N is the number of states, and a ij represents the probability of transitioning from state i to state j.

[0024] Further, in the heterogeneous multi-core data processing central unit, the ARM processor adopts the Cortex-A72 architecture with a main frequency of 2.4 GHz. The FPGA processing unit is based on the Xilinx Virtex series chips and uses a pipeline architecture to perform parallel processing on the collected data. The preprocessing process includes median filtering to remove salt-and-pepper noise in the infrared image, image enhancement processing based on histogram equalization, and normalization processing of environmental parameters and motion data. The heterogeneous multi-core data processing central unit is also equipped with a data compression module, which uses a compression algorithm based on wavelet transform to compress and store the processed data. At the same time, a block compression strategy is introduced, and the image data is divided into sub-blocks not larger than 16×16 pixels for compression processing respectively. The data processing central unit is configured with a data fusion module based on the D-S evidence theory for fusing multi-source data. The basic probability assignment function of the D-S evidence theory is:

[0025] m: 2 Θ → [0, 1]

[0026] where Θ is the frame of discernment, 2 Θ is the power set of Θ, m is the basic probability assignment function, and it satisfies​

[0027] Furthermore, for the multi-dimensional feature extraction and analysis unit, the improved Transformer spatio-temporal feature extraction module uses the following calculation formula:

[0028]

[0029] Among them, Q is the query vector, corresponding to the time series information in the input data; K is the key vector, corresponding to the spatial position information; V is the value vector, containing the original data features; d k is the dimension of the key vector. This module captures the spatio-temporal features of the data from different angles through the multi-head attention mechanism, enhancing the feature expression ability. At the same time, this module introduces the spatio-temporal attention weight adjustment formula:

[0030]

[0031] Among them, ω t,s is the attention weight of the spatio-temporal position (t, s), σ is the activation function, n t , n s are the lengths of the time dimension and the space dimension respectively, and f t,s (x i,j ) is the feature function of the data at the spatio-temporal position (t, s).

[0032] Furthermore, the multi-dimensional feature extraction and analysis unit also includes a feature enhancement module based on the conditional generative adversarial network cGAN. Both its generator and discriminator adopt the convolutional neural network architecture. The feature extraction effect is optimized through adversarial training. The loss function formula of the generator is:

[0033]

[0034] Among them, L G is the generator loss, z is the random noise vector, p z (z) is the noise distribution, D is the discriminator, G is the generator. In addition to the noise vector, the conditional generator input also includes environmental parameters and motion features as conditional information. Its output formula is:

[0035] G(z, c) = decoder(z, c)

[0036] Among them, G(z, c) is the output of the conditional generator, z is the noise vector, c is the conditional information, and decoder is the decoding network, making the generated features more in line with the actual monitoring scenario requirements.

[0037] Further, in the intelligent decision-making judgment and result output unit, the deep neural network decision model based on transfer learning uses the pre-trained ResNet-50 as the backbone network. After the fully connected layer, two fully connected layers with 128 and 64 neurons respectively are added. The cross-entropy loss function is used for model training. The input is the multi-dimensional feature vector output by the feature extraction and analysis unit, and the output is the probability value that the goat is in estrus. This model introduces a fusion strategy based on ensemble learning, integrating multiple ResNet-50 models with different initial parameters. The output calculation formula of the ensemble model is:

[0038]

[0039] where P ensemble is the output probability of the ensemble model, M is the number of ensemble models, and P m is the output probability of the m-th independent model.

[0040] Further, the intelligent decision-making judgment and result output unit is also provided with a dynamic threshold adaptive adjustment module. This module adjusts the decision threshold by using the gradient descent algorithm with an adaptive learning rate according to the statistical analysis of historical monitoring data. The decision threshold calculation formula is:

[0041]

[0042] where θ t+1 is the decision threshold at time t + 1, θ t is the decision threshold at time t, α is the learning rate, and J(θ t ) is the loss function. By dynamically adjusting the decision threshold to meet the monitoring requirements in different breeding environments, the dynamic threshold adaptive adjustment module introduces a threshold correction mechanism based on fuzzy logic. The input of the fuzzy logic system is the difference between the current environmental parameters and historical estrus data, and the output is the threshold correction coefficient β. The decision threshold adjustment formula is updated as:

[0043] θ′ t+1 = θ t+1 ·(1 + β)

[0044] where θ′ t+1 is the finally adjusted decision threshold, and the decision threshold is adjusted more flexibly through fuzzy logic.

[0045] Beneficial effects: The present invention provides a goat estrus monitoring system, including a multi-modal infrared thermal imaging data acquisition unit, which, with the aid of advanced detectors and spectral imaging modules, accurately acquires the body surface temperature field and multi-spectral images of goats. Combining autofocus and temperature calibration technologies ensures data accuracy and provides a reliable basis for estrus monitoring; a multi-dimensional environmental parameter collaborative real-time perception unit, which collects multiple environmental parameters and conducts fusion and correlation analysis to effectively eliminate the interference of environmental factors on estrus judgment and make the monitoring results more credible. A three-dimensional space motion trajectory and posture capture unit, using depth vision and inertial measurement technologies, comprehensively records the motion trajectory and posture of goats. Combining with motion pattern recognition, it can capture the special behavior changes of goats during the estrus period. The heterogeneous multi-core data processing central unit efficiently preprocesses, compresses, and fuses multi-source data, improving data processing efficiency and reducing storage costs. The multi-dimensional feature extraction and analysis unit deeply mines data features through innovative model formulas, enhancing feature expression ability and recognition. The intelligent decision-making judgment and result output unit adopts integrated learning and dynamic threshold adjustment to improve the generalization ability of the model and adapt to different breeding environments. The overall system realizes accurate and real-time monitoring of the estrus state of goats, reduces the cost and error of manual observation, provides data support for scientific breeding, helps farmers timely grasp the estrus timing of goats, improves the breeding success rate and breeding efficiency, and promotes the intelligent and scientific development of the goat breeding industry. Brief Description of the Drawings

[0046] Figure 1 It is a diagram of the system unit composition of the present invention;

[0047] Figure 2 It is a flowchart of the system operation of the present invention. Detailed Embodiments

[0048] It should be noted that, without conflict, the embodiments in this application and the features in the embodiments can be combined with each other. The following further describes this application in detail with reference to the drawings and specific embodiments.

[0049] As Figure 1 shown, a goat estrus monitoring system, the system includes:

[0050] A multi-modal infrared thermal imaging data acquisition unit, which integrates an uncooled vanadium oxide (VOx) focal plane array infrared detector and a multi-band spectral imaging module, is equipped with a variable focal ratio optical lens group and an autofocus actuator, and is connected to the data processing central unit through a high-speed fiber optic data transmission link, and is used for collecting high-precision temperature field distribution and multi-spectral image data of the goat body surface;

[0051] Specifically, the multi-modal infrared thermal imaging data acquisition unit is the basic module for the entire monitoring system to obtain core data. Its core component is the uncooled vanadium oxide (VOx) focal plane array infrared detector. With the high sensitivity of vanadium oxide material to infrared radiation, it can achieve a pixel resolution of not less than 640×480 and a temperature resolution of 0.05°C. The pixel size reaches 17μm×17μm, which enables the detector to accurately capture the subtle temperature changes on the goat's body surface. The multi-band spectral imaging module covers the visible and near-infrared bands from 400nm to 1100nm and contains at least 3 independent spectral channels, capable of synchronously collecting different spectral information and providing multi-dimensional data support for analyzing the physiological state of goats. The variable focal ratio optical lens group has a variable ratio of not less than 3:1. In cooperation with the autofocus actuator based on the phase difference detection algorithm, the focal length is adjusted by controlling the stepper motor. Whether the goat is moving nearby or resting in the distance, clear imaging can be ensured, and the multi-layer broadband anti-reflection film on the lens surface further improves the imaging quality.

[0052] The significance of this unit in the system is that it provides key evidence for judging the estrus state of goats by collecting high-precision temperature field distribution and multi-spectral image data. During the estrus period of goats, specific changes will occur in their body surface temperature, and infrared thermal imaging can visually present these temperature differences; while multi-spectral images can reflect the spectral characteristic changes of tissues such as the skin and hair of goats. These combined pieces of information contribute to a more accurate judgment of the estrus state. For example, in the actual application of a certain goat farm, this unit successfully captured that the ear temperature of a female goat in the pre-estrus period increased by 0.3 - 0.5°C compared to normal. At the same time, in the multi-spectral image, the reflection spectrum of its facial skin showed characteristics different from those in the non-estrus period, providing strong data support for subsequent accurate judgment.

[0053] During the actual working process, this unit continuously scans the goat's body surface at preset time intervals. Once a goat is detected entering the shooting field of view, the autofocus actuator is quickly activated. After completing the focusing operation, the uncooled vanadium oxide detector and the multi-band spectral imaging module work together to transmit the collected temperature data and image data to the data processing central unit in real time through a high-speed fiber optic data transmission link with extremely low latency and extremely high stability, saving time for subsequent data processing and analysis.

[0054] The multi-dimensional environmental parameter collaborative real-time sensing unit is composed of a distributed temperature and humidity sensing network, a high-precision light intensity detection module, an atmospheric pressure monitoring device, and an ultrasonic wind speed and direction measurement component. It uses the industrial-grade RS485 communication protocol to establish a stable communication connection with the data processing central unit to achieve real-time synchronous acquisition of parameters such as temperature, humidity, light, air pressure, wind speed, and wind direction in the breeding environment;

[0055] Specifically, the multi-dimensional environmental parameter collaborative real-time perception unit is mainly responsible for collecting various key parameters in the breeding environment to eliminate environmental interference factors for the system to accurately judge the estrus state of goats. This unit consists of a distributed temperature and humidity sensing network, a high-precision light intensity detection module, an atmospheric pressure monitoring device, and an ultrasonic wind speed and direction measurement component. The distributed temperature and humidity sensing network contains at least 3 sensor nodes at different positions. Each node uses a high-precision digital temperature and humidity sensor. The temperature measurement range is -40°C - 125°C, and the accuracy can reach ±0.1°C. The humidity measurement range is 0 - 100%RH, and the accuracy is ±1.5%RH. By using the Kalman filtering algorithm to fuse and process the data of each node, it can effectively eliminate the data fluctuations caused by sensor errors and local environmental differences, and obtain accurate temperature and humidity data. The light intensity detection module uses a silicon photocell sensing element, with a measurement range of 0 - 200000 lux. It is equipped with an automatic gain adjustment circuit, which can automatically adjust the amplification factor according to the ambient light intensity to ensure accurate measurement under different lighting conditions. The atmospheric pressure monitoring device has an accuracy of ±0.1 hPa. The ultrasonic wind speed and direction measurement component uses the ultrasonic wind measurement principle, with a wind direction measurement accuracy of ±3°, a wind speed measurement range of 0 - 70 m / s, and is equipped with a wind shield to reduce the interference of environmental factors on the measurement results.

[0056] This unit plays an indispensable role in the system. Environmental factors have a significant impact on the physiological state and behavioral performance of goats. If environmental parameters are not considered and the estrus state is judged only based on the physiological indicators and behavioral data of goats themselves, misjudgment is very likely to occur. For example, in a high-temperature and high-humidity environment, even if a goat is not in estrus, its body temperature may rise and its activity level will decrease accordingly; the change in light intensity will affect the biological clock of goats and thus affect their behavior patterns. This unit collects environmental parameters in real time and sets up an environmental parameter correlation analysis module based on the Pearson correlation coefficient to calculate the correlation between various environmental parameters, helping the system to reasonably correct the influence of environmental factors when analyzing the estrus state of goats and ensuring the accuracy of the judgment results. In a mountain goat farm, when sudden weather changes occur and the environmental temperature, humidity, and air pressure change drastically, this unit timely collects and transmits data, and the system analyzes the physiological data of goats in combination with these environmental parameters, effectively avoiding the misjudgment of estrus caused by environmental changes.

[0057] During operation, each component continuously collects data at a stable frequency. The nodes of the temperature and humidity sensing network constantly collect the temperature and humidity information of the surrounding environment. The light intensity detection module senses the light changes in real time. The atmospheric pressure monitoring device and the ultrasonic wind speed and direction measurement component also synchronously obtain the air pressure, wind speed, and wind direction data. All data is transmitted to the data processing central unit in a stable and reliable manner through the industrial-grade RS485 communication protocol, providing an accurate environmental data basis for the system to comprehensively analyze the estrus state of goats.

[0058] The three-dimensional space motion trajectory and attitude capture unit includes a multi-view depth vision acquisition array based on the principle of structured light and a high-precision inertial measurement combination unit. The cameras of the depth vision acquisition array are arranged in a circular layout, and data is transmitted to the data processing central unit through Gigabit Ethernet interfaces. The inertial measurement combination unit collects triaxial acceleration and triaxial angular velocity data at a high-frequency sampling rate to jointly complete the recording of the goat's motion trajectory in the three-dimensional space and the acquisition of attitude information.

[0059] Specifically, the three-dimensional space motion trajectory and attitude capture unit focuses on recording the motion and attitude information of the goat in space, providing a basis for judging the estrus state of the goat from the aspect of behavior science. This unit includes a multi-view depth vision acquisition array based on the principle of structured light and a high-precision inertial measurement combination unit. The cameras of the multi-view depth vision acquisition array are distributed in a circular array, and the included angle between adjacent cameras is 45°. Based on the principle of multi-view stereo vision, the position coordinates of the goat in the three-dimensional space are calculated through a feature point matching algorithm. The structured light technology is adopted to project an infrared pattern of a specific pattern to assist in depth measurement. The effective measurement distance is 0.5m - 5m, and the depth measurement accuracy reaches ±1mm at a distance of 1m, which can accurately restore the spatial position and body contour of the goat. The high-precision inertial measurement combination unit collects data at a sampling frequency of 100Hz and includes a triaxial accelerometer and a triaxial gyroscope. The measurement range of the accelerometer is ±16g, and the measurement range of the gyroscope is ±2000° / s. The quaternion method is used to fuse the accelerometer and gyroscope data to obtain the attitude angle information of the goat in real time, such as roll angle, pitch angle, and yaw angle. In addition, this unit is also equipped with an anti-occlusion supplementary lighting device, which adopts an infrared LED array with a wavelength of 850nm and a light-emitting angle of 120°, and is automatically turned on when the ambient light is insufficient or the goats block each other to ensure the continuity of data collection; at the same time, a motion mode recognition module based on the Hidden Markov Model (HMM) is configured to model and recognize the motion mode of the goat.

[0060] This unit is of great significance in goat estrus monitoring. During the estrus period of goats, their behavior patterns will change significantly, such as increased activity, frequent walking, mounting other goats, etc. By accurately capturing the motion trajectory and attitude of goats through this unit, these abnormal behaviors can be detected in a timely manner. For example, in a large-scale goat farm, after a female goat entered the estrus period, the length of its daily motion trajectory increased by nearly twice compared with usual, and it showed multiple behaviors of trying to mount other female goats. These behavior information was accurately recorded by this unit and transmitted to the system. Combining the data collected by other units, the system quickly and accurately judged that the female goat was in the estrus state.

[0061] During actual operation, the multi-view depth vision acquisition array and the high-precision inertial measurement combined unit work synchronously. The depth vision acquisition array continuously scans the activity area of the goat to capture the spatial position changes of the goat; the inertial measurement combined unit real-time senses the dynamic posture of the goat. The data collected by both are quickly transmitted to the data processing central unit through the Gigabit Ethernet interface. When the environmental light conditions are poor, the anti-occlusion supplementary lighting device automatically starts to ensure that the vision acquisition is not affected; the motion pattern recognition module based on the hidden Markov model then conducts real-time analysis on the collected motion data to identify various motion patterns of the goat, providing rich and accurate behavioral data for the system to judge the estrus state.

[0062] The heterogeneous multi-core data processing central unit is equipped with an ARM and FPGA heterogeneous multi-core processor architecture, integrated with a dedicated neural network acceleration chip, a large-capacity high-speed cache module, and a solid-state storage array. It conducts data interaction with the high-resolution multi-modal infrared thermal imaging data acquisition and sensing unit, the multi-dimensional environmental parameter collaborative real-time sensing unit, the three-dimensional space motion trajectory and attitude capture unit, the feature extraction and analysis unit, and the decision-making judgment and output unit through a high-speed parallel data bus, realizing preprocessing, storage, and preliminary analysis of the collected data;

[0063] Specifically, the heterogeneous multi-core data processing central unit is the data processing core of the entire monitoring system, undertaking the important tasks of preprocessing, storing, and preliminary analyzing the multi-source data collected. It is equipped with an ARM and FPGA heterogeneous multi-core processor architecture. The ARM processor adopts the Cortex-A72 architecture with a main frequency of 2.4 GHz, being good at handling complex logic control and operating system tasks; the FPGA processing unit is based on the Xilinx Virtex series chips, having powerful parallel processing capabilities and being suitable for high-speed data operations. This unit is integrated with a dedicated neural network acceleration chip, which can greatly improve the computing efficiency of deep learning models; it is equipped with a large-capacity high-speed cache module and a solid-state storage array, capable of meeting the fast storage and reading requirements of a large amount of data. Through the high-speed parallel data bus, it establishes an efficient data interaction channel with the high-resolution multi-modal infrared thermal imaging data acquisition and sensing unit, the multi-dimensional environmental parameter collaborative real-time sensing unit, the three-dimensional space motion trajectory and attitude capture unit, the feature extraction and analysis unit, and the decision-making judgment and output unit.

[0064] In the system, the role of this unit is crucial. The data transmitted from each acquisition unit often has problems such as noise and inconsistent formats and cannot be directly used for analysis. For example, the infrared thermal imaging data may have salt-and-pepper noise, and the units and magnitudes of environmental parameters and motion data are also different. This unit uses a pipeline architecture to process the acquired data in parallel. First, it removes the salt-and-pepper noise in the infrared image through median filtering and enhances the image based on histogram equalization to make the image details clearer; then it normalizes the environmental parameters and motion data, converting different types of data into a unified format and magnitude for subsequent analysis. At the same time, the data compression module uses a compression algorithm based on wavelet transform and introduces a block compression strategy, dividing the image data into sub-blocks no larger than 16×16 pixels for compression processing respectively, which significantly reduces the data storage space while ensuring data quality. In addition, the data fusion module based on D-S evidence theory fuses multi-source data from different acquisition units, mines the potential associations between the data, and provides a more comprehensive and accurate data basis for subsequent feature extraction and decision-making. In practical applications, this unit can process a large amount of monitoring data in a short time, ensuring the real-time performance and accuracy of the system.

[0065] During operation, the heterogeneous multi-core data processing central unit constantly monitors the high-speed parallel data bus. Once it receives the data transmitted from the acquisition unit, it immediately starts the corresponding data processing process. The ARM processor is responsible for coordinating the work of each processing module and scheduling the data flow; the FPGA processing unit uses its parallel computing advantage to perform operations such as fast filtering, enhancement, and normalization on the data. A part of the processed data is stored in the solid-state storage array for subsequent reference and analysis; another part is transmitted to the feature extraction and analysis unit to prepare for further mining of data features. The whole process is efficient and orderly, ensuring the fast processing and effective utilization of data by the system.

[0066] The multi-dimensional feature extraction and analysis unit adopts a reconfigurable computing hardware architecture, with an improved Transformer spatio-temporal feature extraction module and a generative adversarial network feature enhancement module built in. It is connected to the heterogeneous multi-core data processing central unit through a high-speed PCI-Express interface, extracts high-dimensional feature vectors that fuse information such as temperature, spectrum, environment, and motion from the preprocessed data, and enhances and optimizes the features;

[0067] Specifically, the multi-dimensional feature extraction and analysis unit is a key link for the system to achieve accurate judgment. Its main function is to extract high-dimensional feature vectors that integrate information such as temperature, spectrum, environment, and motion from the preprocessed data, and enhance and optimize the features. This unit adopts a reconfigurable computing hardware architecture, has flexible computing resource allocation capabilities, and can adapt to the processing requirements of different types of data. It is built with an improved Transformer spatio-temporal feature extraction module and a generative adversarial network feature enhancement module, and is connected to the heterogeneous multi-core data processing central unit through a high-speed PCI-Express interface to ensure the high speed and stability of data transmission.

[0068] The improved Transformer spatio-temporal feature extraction module is based on the Transformer architecture and can deeply analyze data from both time and space dimensions. When processing goat estrus monitoring data, it uses the time series information as the query vector, the spatial position information as the key vector, and the original data features as the value vector. Through the multi-head attention mechanism, it captures the spatio-temporal features of the data from different perspectives. For example, when analyzing the infrared thermal imaging data of goats, this module can focus on the change trend of the body surface temperature of goats over time, as well as the spatial distribution differences of temperatures in different parts, so as to extract more representative spatio-temporal features. At the same time, the introduced spatio-temporal attention weight adjustment mechanism can dynamically adjust the attention weights of different spatio-temporal positions according to the importance of the data, highlighting the key features related to the estrus state of goats.

[0069] The generative adversarial network feature enhancement module consists of a generator and a discriminator, both of which adopt a convolutional neural network architecture. The generator generates richer and more representative features by learning the feature distribution of the original data; the discriminator is responsible for distinguishing the generated features from the original real features. Through the adversarial training of the two, the feature extraction effect is continuously optimized. In goat estrus monitoring, this module can generate features that are more in line with the actual monitoring scenario requirements according to conditional information such as environmental parameters and motion features. For example, when the environmental temperature is low, the generative adversarial network can enhance the features related to goat estrus in a low-temperature environment, enabling the system to more accurately identify the estrus state under different environmental conditions. The high-dimensional feature vectors extracted and enhanced by this unit provide high-quality input data for the subsequent intelligent decision-making judgment and result output unit, significantly improving the accuracy and reliability of system judgment.

[0070] The intelligent decision-making judgment and result output unit includes a deep neural network decision-making model based on transfer learning and a dynamic threshold adaptive adjustment module. It receives the feature vectors output by the feature extraction and analysis unit through a high-speed serial communication interface. After being analyzed and calculated by the deep neural network model, it combines the dynamically adjusted decision threshold to judge whether the goat is in estrus and outputs the final judgment result.

[0071] Specifically, the intelligent decision-making judgment and result output unit is the final execution module of the entire goat estrus monitoring system, undertaking the important task of judging whether a goat is in estrus based on the extracted feature vectors and outputting the final result. This unit includes a deep neural network decision model based on transfer learning and a dynamic threshold adaptive adjustment module. It receives the multi-dimensional feature vectors output by the feature extraction and analysis unit through a high-speed serial communication interface, uses the powerful data analysis ability of the deep neural network model for calculation, and combines the dynamically adjusted decision threshold to obtain the judgment result.

[0072] The deep neural network decision model based on transfer learning uses the pre-trained ResNet-50 as the backbone network, and adds two fully connected layers with 128 and 64 neurons respectively after the fully connected layer. The pre-trained ResNet-50 network has learned rich feature expression capabilities from a large amount of image data. Through transfer learning, it can quickly adapt to the characteristics of goat estrus monitoring data, reducing the time and data volume required for model training. This model takes the multi-dimensional feature vector as the input, and through the calculation and feature transformation of multiple layers of neural networks, finally outputs the probability value of the goat being in estrus. To improve the generalization ability of the model, a fusion strategy based on ensemble learning is introduced, integrating multiple ResNet-50 models with different initial parameters, and synthesizing the output results of multiple models, so that the system can maintain stable judgment performance when facing different breeding environments and goat individual differences.

[0073] The dynamic threshold adaptive adjustment module adjusts the decision threshold using the gradient descent algorithm with an adaptive learning rate based on the statistical analysis of historical monitoring data. Factors such as different breeding environments, goat breeds, and seasonal changes will all affect the judgment criteria for goat estrus status. For example, in different seasons, the normal physiological indicators and behavioral performances of goats will vary, and the corresponding estrus judgment thresholds also need to be adjusted. This module dynamically adjusts the decision threshold by continuously learning historical data and analyzing the feature differences between estrus and non-estrus states of goats in different situations. At the same time, a threshold correction mechanism based on fuzzy logic is introduced, taking the difference degree between the current environmental parameters and historical estrus data as the input, and outputting the threshold correction coefficient to further flexibly adjust the decision threshold. Finally, the intelligent decision-making judgment and result output unit compares the probability value calculated by the deep neural network model with the adjusted decision threshold. If the probability value is greater than the decision threshold, it is determined that the goat is in estrus; otherwise, it is determined that the goat is not in estrus, and the judgment result is presented to the breeding personnel in an intuitive manner through the output device, providing timely and accurate basis for breeding decisions.

[0074] Preferably, for the multi-modal infrared thermal imaging data acquisition unit, the pixel resolution of the uncooled vanadium oxide (VOx) focal plane array infrared detector is not less than 640×480, the temperature resolution reaches 0.05°C, the pixel size is 17μm×17μm, the multi-band spectral imaging module covers the visible and near-infrared bands of 400nm - 1100nm, includes at least 3 independent spectral channels, the autofocus actuator is based on the phase difference detection algorithm, and adjusts the focal length by controlling the stepper motor to drive the variable focal ratio optical lens group. The zoom ratio of the optical lens group is not less than 3:1, and the lens surface is coated with a multi-layer broadband antireflection film to ensure clear imaging of the goat body surface; at the same time, this unit is also equipped with a temperature calibration module based on a blackbody radiation source, which performs non-linear correction on the original temperature data output by the detector according to Planck's law. The correction formula is:

[0075]

[0076] where, T calibrated is the calibrated temperature value, C1 and C2 are Planck constants, λ is the infrared radiation wavelength, and V raw is the original voltage signal output by the detector.

[0077] Preferably, for the multi-dimensional environmental parameter collaborative real-time perception unit, the distributed temperature and humidity sensing network includes 3 sensor nodes at different positions. Each node uses a high-precision digital temperature and humidity sensor. The temperature measurement range is -40°C - 125°C, the accuracy is ±0.1°C, the humidity measurement range is 0 - 100%RH, and the accuracy is ±1.5%RH. The temperature and humidity data collected by each node are fused through the Kalman filtering algorithm. The state prediction equation of the Kalman filter is:

[0078]

[0079] where, is the state prediction at time k based on the state at time k-1, F k is the state transition matrix, is the optimal state estimate at time k-1, B k is the control input matrix, u k is the control input. The light intensity detection module uses a silicon photocell as the sensing element. Its measurement range is 0 - 200000lux, and it is built with an automatic gain adjustment circuit that can automatically adjust the amplification factor according to the ambient light intensity; the accuracy of the barometric pressure monitoring device reaches ±0.1hPa. The ultrasonic wind speed and direction measurement component uses the ultrasonic wind measurement principle. The wind direction measurement accuracy is ±3°, and the wind speed measurement range is 0 - 70m / s. And this component is equipped with a wind shield to reduce the interference of environmental factors on the measurement results; this unit also sets an environmental parameter correlation analysis module based on the Pearson correlation coefficient to calculate the correlation between environmental parameters. The calculation formula is:

[0080]

[0081] where r xy is the Pearson correlation coefficient of variables x and y, and x i , y i are the i-th observed values of variables x and y, respectively, are the means of variables x and y, respectively, and n is the number of observed values.

[0082] Preferably, for the three-dimensional space motion trajectory and attitude capture unit, the cameras of the multi-view depth vision acquisition array based on the structured light principle are arranged in a circular array, and the included angle between adjacent cameras is 45°. Based on the multi-view stereo vision principle, the position coordinates of the goat in the three-dimensional space are calculated through a feature point matching algorithm. The depth vision acquisition array uses structured light technology to project an infrared pattern of a specific pattern to assist depth measurement. Its effective measurement distance is 0.5 m - 5 m, and the depth measurement accuracy reaches ±1 mm at a distance of 1 m. The high-precision inertial measurement combination unit collects data at a sampling frequency of 100 Hz and includes a three-axis accelerometer and a three-axis gyroscope. The measurement range of the accelerometer is ±16 g, and the measurement range of the gyroscope is ±2000°. The quaternion method is used to fuse the accelerometer and gyroscope data to obtain the attitude angle information of the goat; the three-dimensional space motion trajectory and attitude capture unit is also provided with an anti-occlusion supplementary lighting device, which is automatically turned on when the ambient light is insufficient. The supplementary lighting device uses an infrared LED array with a wavelength of 850 nm and a light-emitting angle of 120°; at the same time, this unit is configured with a motion mode recognition module based on the hidden Markov model (HMM) for modeling and recognizing the motion mode of the goat. The state transition probability matrix of the hidden Markov model is:

[0083] A = [a ij N×N

[0084] where N is the number of states, and a ij represents the probability of transitioning from state i to state j.

[0085] ​Preferably, for the heterogeneous multi-core data processing central unit, the ARM processor adopts the Cortex-A72 architecture with a main frequency of 2.4 GHz. The FPGA processing unit is based on Xilinx Virtex series chips and uses a pipeline architecture to perform parallel processing on the acquired data. The preprocessing process includes median filtering to remove salt-and-pepper noise in the infrared image, image enhancement processing based on histogram equalization, and normalization processing of environmental parameters and motion data. This unit also has a data compression module that compresses and stores the processed data using a compression algorithm based on wavelet transform. At the same time, a block compression strategy is introduced, and the image data is divided into sub-blocks no larger than 16×16 pixels for compression processing respectively. The data processing central unit is configured with a data fusion module based on D-S evidence theory for fusing multi-source data. The basic probability assignment function of D-S evidence theory is:

[0086] m:2 Θ →[0,1]

[0087] where Θ is the frame of discernment, 2 Θ is the power set of Θ, m is the basic probability assignment function, and satisfies

[0088] Preferably, for the multi-dimensional feature extraction and analysis unit, the improved Transformer spatio-temporal feature extraction module uses the following calculation formula:

[0089]

[0090] where Q is the query vector corresponding to the time series information in the input data; K is the key vector corresponding to the spatial position information; V is the value vector containing the original data features; d k is the dimension of the key vector. This module captures the spatio-temporal features of the data from different angles through the multi-head attention mechanism to enhance the feature expression ability. At the same time, this module introduces a spatio-temporal attention weight adjustment formula:

[0091]

[0092] where ω t,s is the attention weight of the spatio-temporal position (t, s), σ is the activation function, n t , n s are the lengths of the time dimension and the space dimension respectively, f t,s (x i,j ) is the feature function of the data at the spatio-temporal position (t, s). The attention weights of different spatio-temporal positions are dynamically adjusted through this formula to highlight the key features.

[0093] Preferably, the multi-dimensional feature extraction and analysis unit further includes a feature enhancement module based on a conditional generative adversarial network (cGAN). Both its generator and discriminator adopt a convolutional neural network architecture. The feature extraction effect is optimized through adversarial training. The loss function formula of the generator is:

[0094]

[0095] where L G is the generator loss, z is a random noise vector, p z (z) is the noise distribution, D is the discriminator, G is the generator. In addition to the noise vector, the conditional generator input also includes environmental parameters and motion features as conditional information. Its output formula is:

[0096] G(z,c) = decoder(z,c)

[0097] where G(z,c) is the output of the conditional generator, z is the noise vector, c is the conditional information, and decoder is the decoding network, making the generated features more in line with the requirements of the actual monitoring scenario.

[0098] Preferably, in the intelligent decision-making judgment and result output unit, the deep neural network decision model based on transfer learning uses a pre-trained ResNet-50 as the backbone network. After the fully connected layer, two fully connected layers with 128 and 64 neurons respectively are added. The cross-entropy loss function is used for model training. The input is the multi-dimensional feature vector output by the feature extraction and analysis unit, and the output is the probability value that the goat is in estrus. This model introduces a fusion strategy based on ensemble learning, integrating multiple ResNet-50 models with different initialization parameters. The output calculation formula of the ensemble model is:

[0099]

[0100] where P ensemble is the output probability of the ensemble model, M is the number of ensemble models, P m is the output probability of the m-th independent model. The generalization ability of the model is improved through ensemble learning.

[0101] Preferably, the intelligent decision-making judgment and result output unit also has a dynamic threshold adaptive adjustment module. This module adjusts the decision threshold using the gradient descent algorithm with an adaptive learning rate based on the statistical analysis of historical monitoring data. The decision threshold calculation formula is:

[0102]

[0103] where θ t+1 is the decision threshold at time t + 1, θ tis the decision threshold at time t, α is the learning rate, and J(θ t ) is the loss function. The decision threshold is dynamically adjusted to meet the monitoring requirements in different breeding environments. Further, the dynamic threshold adaptive adjustment module introduces a threshold correction mechanism based on fuzzy logic. The input of the fuzzy logic system is the difference between the current environmental parameters and the historical estrus data, and the output is the threshold correction coefficient β. The decision threshold adjustment formula is updated as follows:

[0104] θ′ t+1 = θ t+1 ·(1 + β)

[0105] where θ′ t+1 is the finally adjusted decision threshold, and the decision threshold is adjusted more flexibly through fuzzy logic.

[0106] As Figure 2 shown, a goat estrus monitoring system operates in the following steps:

[0107] Step S1: The multi-modal infrared thermal imaging data acquisition and perception unit collects multi-modal images of the goat body surface at preset time intervals, the multi-dimensional environmental parameter collaborative real-time perception unit synchronously collects environmental parameters, and the three-dimensional space motion trajectory and posture capture unit records the goat's motion trajectory and posture information in real time;

[0108] Step S2: The collected image data, environmental parameters, and motion data are transmitted to the heterogeneous multi-core data processing central unit. The data processing central unit performs preprocessing operations such as noise reduction, enhancement, and normalization on the data, and stores the processed data;

[0109] Step S3: The multi-dimensional feature extraction and analysis unit obtains the preprocessed data from the data processing central unit, and extracts multi-dimensional feature vectors containing temperature, spectrum, environment, and motion information through an improved Transformer spatio-temporal feature extraction module and a feature enhancement module based on a conditional generative adversarial network;

[0110] Step S4: Then, the extracted multi-dimensional feature vectors are input into the intelligent decision-making judgment and result output unit. The deep neural network decision model in the decision-making judgment unit analyzes the feature vectors and calculates the probability value that the goat is in the estrus state;

[0111] Step S5: The dynamic threshold adaptive adjustment module in the intelligent decision-making judgment and result output unit adjusts the decision threshold according to historical monitoring data, using the gradient descent algorithm with an adaptive learning rate combined with the fuzzy logic threshold correction mechanism, and compares the calculated probability value with the adjusted decision threshold;

[0112] Step S6: Determine whether the goat is in estrus according to the comparison result. If the probability value is greater than the decision threshold, it is determined that the goat is in estrus; otherwise, it is determined that the goat is not in estrus, and the judgment result is output.

[0113] There are many deficiencies in the existing goat estrus monitoring technologies, while this system achieves a breakthrough with innovative design and advanced technologies.

[0114] I. Precise and comprehensive data collection, breaking through the traditional single limitation;

[0115] Traditional manual observation and some single-sensor monitoring methods have problems such as low efficiency, strong subjectivity, and single indicators, and are prone to misjudgment due to environmental interference. The multi-modal infrared thermal imaging data collection unit of this goat estrus monitoring system integrates a non-cooled vanadium oxide focal plane array infrared detector and a multi-band spectral imaging module, which can accurately collect the temperature field distribution and multi-spectral images of the goat's body surface. The pixel size reaches 17μm×17μm, and the temperature resolution is 0.05°C, which can capture subtle physiological changes during the estrus period. The multi-dimensional environmental parameter collaborative real-time perception unit, through a distributed sensing network and a variety of monitoring components, synchronously collects multi-dimensional environmental parameters such as temperature, humidity, and light, fuses the data using the Kalman filter, and can also analyze the parameter correlation, effectively eliminating the interference of environmental factors, and changing the situation where existing technologies ignore the environmental impact or only simply collect data. The three-dimensional space motion trajectory and attitude capture unit, combined with a multi-view depth vision acquisition array and an inertial measurement combination unit, based on structured light technology and hidden Markov model, accurately records the three-dimensional motion trajectory and attitude of the goat, and identifies special behavior patterns during the estrus period, making up for the deficiencies of traditional behavior monitoring that are not comprehensive and inaccurate.

[0116] II. Efficient and intelligent data processing, solving the shortcoming of traditional analysis;

[0117] Existing monitoring systems have problems such as data isolation and poor algorithm adaptability in data processing. The heterogeneous multi-core data processing central unit of this system adopts an ARM and FPGA heterogeneous multi-core architecture, and through pipeline parallel processing, performs preprocessing such as noise reduction, enhancement, and normalization on multi-source data, and uses wavelet transform compression algorithm and D-S evidence theory to fuse data, mining the potential associations of data, improving data quality and processing efficiency. The multi-dimensional feature extraction and analysis unit based on the attention mechanism uses an improved Transformer spatio-temporal feature extraction module and a conditional generative adversarial network feature enhancement module to deeply extract and enhance features from the spatio-temporal dimension, highlighting key information, and can extract estrus-related features more accurately than traditional methods, improving the data utilization value.

[0118] III. Scientific and reliable decision-making judgment, optimizing the traditional judgment mode;

[0119] Traditional monitoring technologies have strong subjectivity in decision-making judgments and fixed thresholds, making it difficult to adapt to complex aquaculture scenarios. The intelligent decision-making judgment and result output unit of this system is based on a deep neural network decision model for transfer learning. With the pre-trained ResNet-50 as the backbone and combined with ensemble learning, it improves the generalization ability of the model. The dynamic threshold adaptive adjustment module uses the adaptive learning rate gradient descent algorithm and fuzzy logic to dynamically adjust the decision threshold according to historical data and environmental changes, making the judgment criteria more in line with the actual situation, avoiding misjudgments and missed judgments caused by traditional fixed thresholds, providing a timely and accurate basis for aquaculture decisions, and significantly improving the scientific and intelligent level of goat aquaculture.

[0120] Although the embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various equivalent changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A goat estrus monitoring system, characterized in that The system includes: A multi-modal infrared thermal imaging data acquisition unit, which integrates an uncooled focal plane infrared detector array and a multi-spectral imaging component, is connected to the data processing unit through a fiber optic interface, and is used to collect the body surface temperature distribution and multi-spectral image data of goats; A multi-dimensional environmental parameter collaborative real-time perception unit, which includes a temperature and humidity sensor array, a light intensity sensor, a barometric pressure sensor, and a wind direction and speed measurement device, establishes a communication link with the data processing unit, and is used to collect environmental temperature and humidity, light, barometric pressure, and wind speed and direction parameters in real time; A three-dimensional space motion trajectory and attitude capture unit, which consists of multiple groups of depth cameras and inertial measurement units. The depth cameras perform three-dimensional modeling based on the TOF (Time of Flight) principle and are connected to the data processing unit through a gigabit Ethernet interface, and are used to record the motion trajectory and attitude information of goats; A heterogeneous multi-core data processing central unit, which is equipped with a multi-core heterogeneous processor, an embedded dedicated neural network acceleration chip, and a large-capacity cache and solid-state drive storage module. It is respectively connected to the multi-modal infrared thermal imaging acquisition unit, the environmental parameter collaborative perception unit, the motion trajectory capture unit, the feature extraction unit, and the decision-making judgment unit through a high-speed data bus, and is used to preprocess and preliminarily analyze the collected data; A multi-dimensional feature extraction and analysis unit, which adopts a reconfigurable computing architecture and integrates a spatio-temporal feature extraction module based on an attention mechanism, is connected to the data processing unit through a PCI-Express interface, and is used to extract multi-dimensional feature vectors including temperature features, spectral features, environmental features, and motion features from the preprocessed data; An intelligent decision-making judgment and result output unit, which includes a deep neural network model based on transfer learning, is connected to the feature extraction unit through a high-speed serial interface, and is used to judge whether a goat is in estrus according to the multi-dimensional feature vector.

2. The goat estrus monitoring system according to claim 1, characterized in that, For the multi-modal infrared thermal imaging data acquisition unit, the multi-band spectral imaging module covers the visible and near-infrared bands of 400nm - 1100nm, includes at least 3 independent spectral channels, the autofocus actuator is based on the phase difference detection algorithm, and controls the stepper motor to drive the variable focal ratio optical lens group for focal length adjustment. The zoom ratio of the optical lens group is not less than 3:1, and the lens surface is coated with multiple layers of broadband antireflection films to ensure clear imaging of the goat body surface; at the same time, the multi-modal infrared thermal imaging data acquisition unit is also equipped with a temperature calibration module based on a blackbody radiation source, and non-linearly corrects the original temperature data output by the detector according to Planck's law. The correction formula is: Among them, T calibrated is the calibrated temperature value, C1 and C2 are Planck constants, λ is the infrared radiation wavelength, and V raw is the original voltage signal output by the detector.

3. The estrus monitoring system for goats according to claim 1, characterized in that, For the multi-dimensional environmental parameter collaborative real-time perception unit, the distributed temperature and humidity sensing network includes 3 sensor nodes at different positions. Each node uses a high-precision digital temperature and humidity sensor. The temperature measurement range is -40°C - 125°C, the accuracy is ±0.1°C, the humidity measurement range is 0 - 100%RH, and the accuracy is ±1.5%RH. The Kalman filter algorithm is used to fuse the temperature and humidity data collected by each node. The state prediction equation of the Kalman filter is: Among them, is the state prediction at time k based on the state at time k - 1, and F k is the state transition matrix, is the optimal state estimate at time k - 1, and B k is the control input matrix, and u k is the control input. The light intensity detection module uses a silicon photocell as the sensing element, with a measurement range of 0 - 200000 lux, and an automatic gain adjustment circuit is built-in to automatically adjust the amplification factor according to the ambient light intensity; the air pressure monitoring device has an accuracy of ±0.1 hPa. The ultrasonic wind speed and direction measurement component uses the ultrasonic wind measurement principle, with a wind direction measurement accuracy of ±3°, a wind speed measurement range of 0 - 70 m / s, and the component is equipped with a wind shield to reduce the interference of environmental factors on the measurement results; the multi-dimensional environmental parameter collaborative real-time perception unit is also provided with an environmental parameter correlation analysis module based on the Pearson correlation coefficient, which is used to calculate the correlation between environmental parameters, and the calculation formula is: where r xy is the Pearson correlation coefficient of variables x and y, x i , y i are the i-th observations of variables x and y respectively, are the means of variables x and y respectively, and n is the number of observations.

4. A goat estrus monitoring system according to claim 1, characterized in that, The three-dimensional space motion trajectory and attitude capture unit: The cameras of the multi-view depth vision acquisition array based on the structured light principle are distributed in a circular array, and the included angle between adjacent cameras is 45°. Based on the multi-view stereo vision principle, the position coordinates of the goat in the three-dimensional space are calculated through a feature point matching algorithm. The depth vision acquisition array uses structured light technology to project an infrared pattern of a specific pattern to assist depth measurement. Its effective measurement distance is 0.5m - 5m, and the depth measurement accuracy reaches ±1mm at a distance of 1m. The high-precision inertial measurement combination unit collects data at a sampling frequency of 100Hz, including a three-axis accelerometer and a three-axis gyroscope. The measurement range of the accelerometer is ±16g, and the measurement range of the gyroscope is ±2000°. The quaternion method is used to fuse the accelerometer and gyroscope data to obtain the attitude angle information of the goat; The three-dimensional space motion trajectory and attitude capture unit is also equipped with an anti-occlusion supplementary light device, which is automatically turned on when the ambient light is insufficient. The supplementary light device uses an infrared LED array with a wavelength of 850nm and a light-emitting angle of 120°; At the same time, the three-dimensional space motion trajectory and attitude capture unit is configured with a motion mode recognition module based on the Hidden Markov Model (HMM) for modeling and recognizing the motion mode of the goat. The state transition probability matrix of the Hidden Markov Model is: A = [a ij N×N ​ where N is the number of states, a ij represents the probability of transitioning from state i to state j.

5. The estrus monitoring system for goats according to claim 1, wherein The heterogeneous multi-core data processing central unit: The ARM processor adopts the Cortex-A72 architecture with a main frequency of 2.4GHz. The FPGA processing unit is based on the Xilinx Virtex series chips and uses a pipeline architecture to perform parallel processing on the collected data. The preprocessing process includes median filtering to remove salt-and-pepper noise in the infrared image, image enhancement processing based on histogram equalization, and normalization processing of environmental parameters and motion data. The heterogeneous multi-core data processing central unit is also equipped with a data compression module, which uses a compression algorithm based on wavelet transform to compress and store the processed data. At the same time, a block compression strategy is introduced, and the image data is divided into sub-blocks no larger than 16×16 pixels for compression processing respectively. The data processing central unit is configured with a data fusion module based on the D-S evidence theory for fusing multi-source data. The basic probability assignment function of the D-S evidence theory is: m:2 Θ →[0,1] Among them, Θ is the identification framework, 2 Θ is the power set of Θ, and m is the basic probability assignment function, satisfying 6. The estrus monitoring system for goats according to claim 1, wherein The multi-dimensional feature extraction and analysis unit: The improved Transformer spatio-temporal feature extraction module uses the following calculation formula: Among them, Q is the query vector corresponding to the time series information in the input data; K is the key vector corresponding to the spatial position information; V is the value vector containing the original data features; d k is the dimension of the key vector. This module captures the spatio-temporal features of the data from different perspectives through the multi-head attention mechanism to enhance the feature expression ability. At the same time, this module introduces a spatio-temporal attention weight adjustment formula: Among them, ω t,s is the attention weight of the spatio-temporal position (t, s), σ is the activation function, n t , n s are the lengths of the time dimension and the space dimension respectively, and f t,s (x i,j ) is the feature function of the data at the spatio-temporal position (t, s).

7. A goat estrus monitoring system according to claim 1, characterized in that, The multi-dimensional feature extraction and analysis unit also includes a feature enhancement module based on the conditional generative adversarial network (cGAN). Both its generator and discriminator adopt the convolutional neural network architecture, and the feature extraction effect is optimized through adversarial training. The loss function formula of the generator is: Among them, L G is the generator loss, z is the random noise vector, p z (z) is the noise distribution, D is the discriminator, G is the generator. In addition to the noise vector, the conditional generator input also includes environmental parameters and motion features as conditional information, and its output formula is: G(z,c) = decoder(z,c) Among them, G(z,c) is the output of the conditional generator, z is the noise vector, c is the conditional information, and decoder is the decoding network, making the generated features more in line with the actual monitoring scenario requirements.

8. The estrus monitoring system for goats according to claim 1, characterized in that In the intelligent decision-making judgment and result output unit, the deep neural network decision-making model based on transfer learning uses the pre-trained ResNet-50 as the backbone network. After the fully connected layer, two fully connected layers with 128 and 64 neurons respectively are added. The cross-entropy loss function is used for model training. The input is the multi-dimensional feature vector output by the feature extraction and analysis unit, and the output is the probability value that the goat is in the estrus state. This model introduces a fusion strategy based on ensemble learning to integrate multiple ResNet-50 models with different initial parameters. The output calculation formula of the integrated model is as follows: where, P ensemble is the output probability of the integrated model, M is the number of integrated models, and P m is the output probability of the m-th independent model.

9. The estrus monitoring system for goats according to claim 1, characterized in that, The intelligent decision-making judgment and result output unit is also equipped with a dynamic threshold adaptive adjustment module. This module adjusts the decision threshold by using the gradient descent algorithm with an adaptive learning rate according to the statistical analysis of historical monitoring data. The decision threshold calculation formula is as follows: where θ t+1 is the decision threshold at time t+1, θ t is the decision threshold at time t, α is the learning rate, J(θ t ) is the loss function. To adapt to the monitoring requirements in different breeding environments by dynamically adjusting the decision threshold, the dynamic threshold adaptive adjustment module introduces a threshold correction mechanism based on fuzzy logic. The input of the fuzzy logic system is the difference between the current environmental parameters and the historical estrus data, and the output is the threshold correction coefficient β. The decision threshold adjustment formula is updated as follows: θ′ t+1 = θ t+1 ·(1 + β) Among them, θ′ t+1 is the finally adjusted decision threshold, and the decision threshold is adjusted more flexibly through fuzzy logic.

Images