Cow core body temperature prediction method and system based on bimodal RGB-T and environmental perception

Abstract

The invention discloses a cattle core body temperature prediction method and system based on bimodal RGB-T and environmental perception. The method comprises the following steps: synchronously obtaining an infrared thermal image and a visible light color image of a target cattle; inputting the registered visible light color image and infrared thermal image into a pre-trained bimodal semantic segmentation model, and identifying and segmenting at least one key temperature measurement part of the cattle; extracting a surface temperature value of a corresponding part from the infrared thermal image based on a key temperature measurement part segmentation result obtained in the bimodal semantic segmentation step; and inputting the extracted surface temperature value into a pre-established cattle core body temperature prediction model, and calculating to obtain a core body temperature estimated value of the target cattle. According to the method, the key temperature measurement part of the cow can be automatically positioned, the high-precision body surface temperature can be obtained, and the core body temperature of the cow can be accurately predicted through the correction model, so that the health condition of the cow herd can be monitored and early warned in real time, and manual intervention and measurement errors are reduced.

Description

Technical Field

[0001] This invention relates to the fields of livestock intelligent sensing and artificial intelligence technology, and in particular to a method and system for predicting the core body temperature of cattle based on dual-modal RGB-T and environmental perception. Background Technology

[0002] Infrared thermography (IRT) plays a crucial role in monitoring core body temperature (CBT) in cattle, but it is influenced by various factors, such as ambient humidity and temperature index (THI), cattle posture and imaging distance, fur emissivity, wind speed, and direct sunlight. These factors make the relationship between surface temperature and core body temperature complex and nonlinear, increasing the difficulty of directly calculating core body temperature from surface temperature. Therefore, improving the stability and accuracy of surface temperature readings has become an important research direction.

[0003] Existing research and patents largely focus on combining single thermal imaging techniques with simple rule-based or clustering methods, such as determining rectal temperature by correlating it with specific surface temperature regions (e.g., the eye area and nasal mirror). However, joint modeling of dual-modal (combining RGB and thermal images) and environmental and geometric information is relatively rare, especially lacking assessments of device-level standardized procedures and the consistency and stability of data under different environmental and imaging angle conditions. This means that existing methods may not achieve ideal monitoring accuracy when applied in different scenarios.

[0004] Furthermore, although existing studies have proposed compensation methods for different environmental indices (such as THI) and direct sunlight conditions, most of these methods have not been systematically applied to the model, but rather treated as separate steps. This approach fails to fully utilize environmental information and limits the model's generalization ability. At the same time, current methods are relatively inadequate in handling the quantitative regression between surface temperature and core body temperature and related uncertainties, lacking specific alarm mechanisms and failing to effectively support real-time monitoring and early warning. This is particularly crucial for early detection and intervention of potential health problems in practical environments such as farms. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the existing technology. To achieve the above objective, a method and system for predicting the core body temperature of cattle based on dual-modal RGB-T and environmental perception is adopted to solve the problems mentioned in the background technology.

[0006] A method for predicting core body temperature in cattle based on bimodal RGB-T and environmental perception includes the following steps: Step S1: Simultaneously acquire infrared thermal images and visible light color images of the target cattle; Step S2: Input the registered visible light color image and infrared thermal image into the pre-trained bimodal semantic segmentation model to identify and segment at least one key temperature measurement site of the cattle. Step S3: Based on the segmentation results of the key temperature measurement parts obtained in the dual-modal semantic segmentation step, extract the surface temperature value of the corresponding parts from the infrared thermal image. The dual-modal semantic segmentation model adopts a dual encoder-single decoder architecture. The dual encoder includes a first encoder for extracting visible light image features and a second encoder for extracting infrared thermal image features. The single decoder is used to fuse and upsample the features from the dual encoder to output the semantic segmentation result. Step S4: Input the extracted surface temperature value into the pre-established cattle core body temperature prediction model to calculate the estimated core body temperature of the target cattle.

[0007] As a further aspect of the present invention, the specific steps in step S1 include: Environmental parameters of the cattle shed environment are collected synchronously, including at least ambient temperature and ambient humidity; in the core body temperature prediction step, the environmental parameters and the surface temperature value are input into the core body temperature prediction model.

[0008] As a further aspect of the present invention: the bimodal semantic segmentation model performs feature fusion with the corresponding layer of the single decoder at the skip connection between the first encoder and the second encoder; and / or An attention mechanism is introduced during feature fusion to adaptively weight features from different modalities and spatial locations.

[0009] As a further aspect of the present invention, the attention mechanism is a dual attention mechanism that combines channel attention and spatial attention.

[0010] As a further aspect of the present invention: the key temperature measurement areas include the periorbital region and the ear region; In the body surface temperature extraction step, the average surface temperature of the periorbital area and the average surface temperature of the ear area are extracted.

[0011] As a further aspect of the present invention: when the image contains multiple cattle, an individual differentiation step is included before or after the bimodal semantic segmentation step. Using object detection or instance segmentation methods, multiple cattle in the image are separated, and the segmentation of key temperature measurement sites and extraction of body surface temperature are performed independently for each individual.

[0012] As a further aspect of the present invention: the core body temperature prediction model is a multiple linear regression model or a nonlinear machine learning model; During the model training phase, the collected rectal temperature of cattle was used as the true label and fitted with the corresponding surface temperature value and environmental parameters. It also includes an uncertainty assessment of the output of the core body temperature estimate.

[0013] As a further aspect of the present invention: after obtaining the estimated core body temperature value, the method further includes a result output and early warning step. The estimated core body temperature is compared with a preset normal physiological body temperature range. If the value exceeds the range, an early warning signal is generated and issued.

[0014] As a further aspect of the present invention: in the data acquisition step, the infrared thermal imager and the visible light camera are coaxially mounted or calibrated to ensure consistent field of view; In the image preprocessing step, the infrared thermal image and the visible light color image are spatially aligned and registered using calibration parameters.

[0015] The second aspect of the technical solution: A prediction system employing a bovine core body temperature prediction method based on dual-modal RGB-T and environmental perception as described in any of the above claims, comprising: The data acquisition module is used to simultaneously acquire infrared thermal images and visible light color images of the target cattle; The image preprocessing module is used to perform spatial alignment and registration on the acquired images; The semantic segmentation module has a pre-trained bimodal semantic segmentation model built in, which is used to receive the registered image and output the segmentation results of key temperature measurement parts; A temperature extraction module is used to extract the surface temperature values ​​of key parts from the infrared thermal image based on the segmentation results. The core body temperature prediction module has a pre-established core body temperature prediction model for cattle, which is used to calculate the estimated core body temperature based on the surface temperature value. The output and warning module is used to output the core body temperature estimate and trigger a warning based on the comparison result.

[0016] Compared with the prior art, the present invention has the following technical advantages: By adopting the above-mentioned technical solution, non-contact and efficient temperature measurement is achieved: the core body temperature of cattle can be obtained without human contact, significantly reducing labor intensity and stress on animals, and enabling continuous automatic temperature measurement of large groups of cattle. This invention's system can be fixedly installed in the cattle shed, achieving 24 / 7 real-time monitoring and timely detection of abnormalities.

[0017] High accuracy in temperature measurement: By selecting body surface areas with high correlation to core temperature, such as the periorbital and ostomy areas, and introducing an environmental parameter correction model, the accuracy of body temperature estimation is significantly improved. Experiments show that the average error between the core body temperature predicted by the method of this invention and the actual rectal measurement value can be controlled within ±0.3℃, meeting the requirements of aquaculture production for accurate temperature measurement.

[0018] Intelligent identification of key areas: Deep learning semantic segmentation is used to automatically locate key temperature measurement areas on cattle, which is more reliable than manual or simple threshold segmentation methods. The fusion of visible light and infrared light overcomes the problems of blurred edges and target adhesion in multi-target scenes caused by single infrared images, ensuring accurate identification of small parts such as cattle eyes and ears. The system remains robust even when cattle are partially occluded or the lighting changes.

[0019] Adaptable to complex farming environments: This invention considers the effects of temperature measurement distance, angle, hair coverage, and environmental climate, and improves the system's applicability in actual pasture environments through algorithm correction and multi-source information fusion. Whether for fixed-point monitoring of single cattle or inspection monitoring of group cattle sheds, this method can operate stably and has good scalability (e.g., it can be extended to identify individual cattle and combined with camera-based mobile inspection devices, etc.).

[0020] Early warning and management decision support: The continuous body temperature data provided by this invention can be used for early disease warning and reproductive management reference. Once an abnormal increase in a cow's body temperature is detected, it can provide an early warning of a potential febrile disease outbreak, facilitating timely intervention and treatment by veterinarians. Simultaneously, changes in the body temperature curve can also provide a basis for estrus detection and calving prediction in dairy cows, improving the accuracy and scientific nature of livestock management. This invention effectively enhances the automation and intelligence level of dairy cow health monitoring, possessing significant application value and promising prospects for widespread adoption. Attached Figure Description

[0021] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings: Figure 1 This is a schematic diagram illustrating the steps of the bovine core body temperature prediction method according to an embodiment of this application. Figure 2 This is a schematic diagram of the bimodal semantic segmentation network structure according to an embodiment of this application; Figure 3 This is a schematic diagram illustrating the identification of key temperature measurement areas in an infrared thermal imaging image of a cow, according to an embodiment of this application. Figure 4 This is a schematic diagram illustrating the relationship between bovine surface temperature and core body temperature, and a prediction model, according to an embodiment of this application. Figure 5 This is a schematic diagram of the overall structure of the system disclosed in this application. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Please refer to Figure 1 In this embodiment of the invention, a method for predicting the core body temperature of cattle based on dual-modal RGB-T and environmental perception includes the following steps: Step S1: Simultaneously acquire infrared thermal images and visible light color images of the target cattle. Specific steps include: Simultaneously collect environmental parameters of the cattle shed environment, including at least ambient temperature and ambient humidity; in the core body temperature prediction step, input the environmental parameters and the surface temperature value into the core body temperature prediction model. In this embodiment, during the data acquisition step, the infrared thermal imager and the visible light camera are coaxially mounted or calibrated to ensure consistent field of view. In the image preprocessing step, the infrared thermal image and the visible light color image are spatially aligned and registered using calibration parameters.

[0024] In a specific implementation, the data acquisition steps are as follows: an infrared thermal imager and a visible light camera are set up in the cattle shed environment (preferably the two are coaxially installed or calibrated to ensure consistent field of view), and the target cattle are imaged synchronously to obtain the corresponding infrared thermal image and visible light color image sequence.

[0025] In addition, environmental sensors can be optionally deployed to acquire parameters such as ambient temperature and humidity.

[0026] Step S2: Input the registered visible light color image and infrared thermal image into the pre-trained bimodal semantic segmentation model to identify and segment at least one key temperature measurement site of the cattle. The bimodal semantic segmentation model performs feature fusion with the corresponding layer of the single decoder at the skip connections between the first encoder and the second encoder; and / or An attention mechanism is introduced during feature fusion to adaptively weight features from different modalities and spatial locations.

[0027] In this embodiment, the attention mechanism is a dual attention mechanism that combines channel attention and spatial attention.

[0028] In a specific implementation, the image preprocessing steps are as follows: calibrating and preprocessing the acquired infrared and visible light images. This includes spatially aligning and registering the infrared and visible light images using calibration parameters to ensure that pixels at the same location correspond in both images. Simultaneously, frames at different intervals can be synchronized. If necessary, the infrared image is denoised and temperature drift corrected (e.g., using a blackbody reference source in the image or correcting based on ambient temperature) to improve the accuracy of temperature measurement.

[0029] Step S3: Based on the segmentation results of the key temperature measurement parts obtained in the dual-modal semantic segmentation step, extract the surface temperature value of the corresponding parts from the infrared thermal image. The dual-modal semantic segmentation model adopts a dual encoder-single decoder architecture. The dual encoder includes a first encoder for extracting visible light image features and a second encoder for extracting infrared thermal image features. The single decoder is used to fuse and upsample the features from the dual encoder to output the semantic segmentation result. In a specific implementation, the bimodal image semantic segmentation step is as follows: the registered visible light image and infrared thermal image are used as inputs and fed into a pre-trained bimodal semantic segmentation model to automatically identify the key temperature measurement sites of the cattle.

[0030] Specifically, the segmentation model employs an improved U-Net convolutional neural network architecture, comprising two independent encoder branches for visible light (RGB) and infrared thermal (thermal imaging), extracting feature maps from the two modalities respectively, and then fusing them during the decoding stage, such as... Figure 2 As shown.

[0031] By fusing modal features at the skip connections of the encoder and introducing an attention mechanism (such as the Channel-Spatial Attention Module CBAM), the model can fully utilize the complementary information of visible light and infrared data to enhance the ability to identify target parts. The decoder upsamples the fused features layer by layer to restore them to the original resolution, and outputs a semantic segmentation result of the same size as the input image, where different pixel categories correspond to different parts of the cow.

[0032] This invention designates the periorbital and ear regions of cattle as key temperature measurement sites, which are then labeled in the segmentation results (e.g., defining category labels such as "periorbital," "ear," and "background"). For images containing multiple cattle, instance segmentation or object detection methods can be combined to differentiate the segmentation results based on individual affiliation. For example, the bounding box of each cattle can be detected first, and then semantic segmentation can be applied within the bounding box to achieve independent identification of the key parts of each individual.

[0033] Step S4: Input the extracted surface temperature value into the pre-established cattle core body temperature prediction model to calculate the estimated core body temperature of the target cattle.

[0034] In this embodiment, the key temperature measurement sites include the periorbital area and the ear area; In the body surface temperature extraction step, the average surface temperature of the periorbital area and the average surface temperature of the ear area are extracted.

[0035] In this embodiment, when the image contains multiple cattle, an individual differentiation step is included before or after the bimodal semantic segmentation step: Using object detection or instance segmentation methods, multiple cattle in the image are separated, and the segmentation of key temperature measurement sites and extraction of body surface temperature are performed independently for each individual.

[0036] In this embodiment, the core body temperature prediction model is a multiple linear regression model or a nonlinear machine learning model; During the model training phase, the collected rectal temperature of cattle was used as the true label and fitted with the corresponding surface temperature value and environmental parameters. It also includes an uncertainty assessment of the output of the core body temperature estimate.

[0037] In this embodiment, after obtaining the core body temperature estimate, the method further includes result output and early warning steps: The estimated core body temperature is compared with a preset normal physiological body temperature range. If the value exceeds the range, an early warning signal is generated and issued.

[0038] In a specific implementation, the surface temperature extraction step is as follows: Based on the semantic segmentation results, the surface temperature values ​​of key parts of the cattle are extracted from the infrared thermal image. For the identified peri-eye and ear regions, the set of pixel temperature values ​​for these regions is extracted from the corresponding infrared thermal image. Preferably, the average temperature of all pixels in each region can be calculated to obtain the average surface temperature of that part. For example, the average temperature of the peri-eye region is denoted as Teye, and the average temperature of the ear region is denoted as Tear. In some cases, the maximum temperature within the region or a specific statistical indicator can also be used as a representative value. In this embodiment, using the average temperature helps to reduce the influence of infrared image noise and local anomalies on the temperature measurement results. If there are multiple key parts, their temperature features can be extracted separately. At the same time, external parameters such as ambient temperature and relative humidity are recorded.

[0039] The core body temperature prediction step involves inputting the extracted surface temperatures and environmental parameters from one or more key body parts into a pre-established bovine core body temperature prediction model to calculate the estimated core body temperature (Tcore) of the bovine. The prediction model can employ either a mathematical regression model or a machine learning model.

[0040] Linear / nonlinear regression models are trained using a large amount of sample data. For example, multiple linear regression or support vector machines are used to fit a mapping formula based on the relationship between bovine rectal temperature (true core temperature) measured in historical experiments and the corresponding teeth, tears, and environmental parameters. In one example, the following correction relationship can be established: Where Tenv represents ambient temperature, RHenv represents ambient relative humidity, and α, β, γ, δ, and ε are regression coefficients. Training these coefficients using methods such as least squares allows the model to achieve good fit to data under different environments. Besides linear models, nonlinear methods such as artificial neural networks can also be used to improve prediction accuracy under complex conditions. During the model building phase, collected rectal temperatures of cattle are used as tags and trained with infrared thermometry data to ensure that the Tcore estimate closely matches the actual core body temperature.

[0041] The output and application steps are as follows: The predicted core body temperature (Tcore) of the cattle is output and compared with the normal physiological body temperature range. If a cow's Tcore exceeds the normal range (e.g., above 39.5℃ or below the lower limit), the system can determine that the cow may have a fever, be sick, or have an abnormal body temperature, and immediately issue an early warning signal (such as an audible alarm or SMS notification to the manager). This invention can also store multiple consecutively monitored body temperature data to form a daily body temperature variation curve for each cow, which can be viewed and analyzed by livestock farmers to promptly grasp the health dynamics of the herd. The entire process can be automatically completed by a computer system, achieving 24 / 7 non-contact monitoring of cattle body temperature.

[0042] Example 2: Image Processing and Segmentation Recognition; The central processing unit (CPU) preprocesses the acquired image data: using system calibration parameters, it performs geometric correction on the infrared thermal image and the visible light image, aligning the two images to the same coordinate system. Subsequently, it performs filtering, noise reduction, and temperature calibration on the infrared image. Since the infrared sensor may be affected by ambient temperature drift, the processing unit reads data from the ambient temperature sensor and compensates for and corrects the temperature values ​​in the infrared image (e.g., adjusting based on a calibration curve provided by the manufacturer or a blackbody reference patch placed in the image). The corrected infrared thermal image ensures accurate and reliable temperature readings. Then, the system uses the registered visible light color image and the infrared pseudo-color thermal image as two-channel inputs, feeding them into the dual-modal deep learning segmentation model of this invention. Figure 2As shown, the diagram illustrates the specific structure of the model, which includes a dual-branch encoder and a cascaded decoder. Encoder A on the left targets RGB visible light images and consists of several convolutional and downsampling layers. Each layer is followed by a residual unit and a channel attention module to extract morphological and texture features of cattle. Encoder B on the right targets infrared thermal images. Its structure is similar, but it has slightly fewer convolutional layers (because thermal images have relatively simpler details). Each layer also includes an attention module to highlight features in temperature-sensitive areas. Both encoders generate feature maps at different scales, which are connected to the corresponding scale decoder layers via skip connections. In the decoding stage, each level first fuses the RGB and infrared feature maps through a modality fusion unit (e.g., convolution after concatenation), and uses a block convolutional attention mechanism (CBAM) to adjust the weights of the fused features to highlight modal information beneficial to the segmentation task. Then, upsampling and convolution are performed to reconstruct spatial details. This layer-by-layer fusion and upsampling process finally yields a segmentation result map of the same size as the input at the output layer. The pixel values ​​of the segmentation result image are divided into several categories, with different colors used to mark the object regions in the image. In this embodiment, the categories include: background, cow outline, cow eye region, cow ear region, etc. We focus particularly on the two key temperature measurement areas: the "cow eye region" and the "cow ear region." Figure 3 As shown in the diagram, the segmentation recognition effect is illustrated: in the visible light image, the cow's head organs are clearly visible; in the segmentation results output by the model, the area around the cow's eyes is accurately delineated (e.g., Figure 3 (As shown in the red highlighted area), the cow's ears are also marked in a different color. Even in real-world scenes where the cows are tilted at certain angles or under varying lighting conditions, the bimodal model can still distinguish the eye and ear regions from the background and other areas using visible light texture and infrared temperature differences, demonstrating good robustness. For multiple cows appearing in the same scene, the segmentation network can identify the eye and ear regions of all cows. To distinguish individuals, object detection algorithms can be further combined to obtain the positional boundaries of each cow, and then the segmentation results can be mapped to each individual according to their spatial location, achieving key feature recognition for multiple targets.

[0043] Example 3: Temperature calculation and core body temperature conversion; Once the areas around the eyes and ears of the cattle are identified, the processing unit extracts the corresponding temperature data for these areas from the infrared thermal image. For example... Figure 3 As shown, the temperature at the bull's eye location appears slightly higher than the surrounding area in the infrared image (due to the rich blood supply near the eye socket and significant heat dissipation in the hairless area), while the temperature of the ear is affected by blood flow and the environment, and is generally slightly lower than the temperature around the eye. The system will Figure 3The average temperature of the peri-eye pixels marked in red is calculated to obtain the average surface temperature (Teye) of the cow's peri-eye area. The average temperature of the ear pixels marked in blue is calculated to obtain the Tear. Simultaneously, the ambient temperature (Tenv) is read as approximately 20.5℃ and the humidity (RHenv) as approximately 60%. These data are used as input to the core body temperature prediction model pre-established in this invention. The model is obtained through learning from a large amount of sample data. For example, for Holstein cows in this embodiment, rectal temperature and corresponding Teye, Tear, and environmental parameters were collected from 100 cows at different seasons and lactation stages. 80% of the dataset was used to train the multivariate regression model, and the remaining 20% ​​was used to validate the model's accuracy. Training results show that peri-eye temperature and rectal temperature are positively correlated (correlation coefficient above 0.8), followed by ear temperature. Adding ambient temperature and humidity terms effectively improves the model's fitting accuracy under extreme high and low temperature conditions. The final model form is as follows: Figure 4 As shown in the diagram, this illustrates the relationship between surface temperature and core body temperature in cattle, along with a predictive model. During the inference phase, the real-time collected data for Teye, Tear, Tenv, and RHenv are substituted into the model for calculation, yielding an estimated value for the cow's core body temperature, Tcore. Figure 3 The individual shown had a calculated Tcore of 38.8℃. Comparing this result with the normal body temperature range for dairy cows (38.0~39.3℃), it can be determined that the cow's body temperature is slightly higher than the average but within the normal range, and there is no abnormality. If the Tcore exceeds the threshold (e.g., ≥39.5℃), the system marks the individual as suspected of having a fever and highlights it on the user interface, while simultaneously triggering the alarm module to remind a veterinarian to check. To verify the effectiveness of the method of the present invention, the above system was deployed in a farm to continuously monitor dozens of dairy cows for a week, and the rectal temperature of some cows was manually measured at regular intervals every day for comparison. The results show that the mean square error between the core body temperature predicted by the present invention and the actual rectal temperature is between 0.2~0.3℃, and the maximum error in a single measurement does not exceed 0.5℃, which can well reflect the actual body temperature changes of the cows. In particular, under an environmental range with a daily maximum temperature of 35℃ and a minimum temperature of 15℃, the system still maintains stable temperature measurement accuracy through environmental parameter correction. This proves the reliability and practical value of the method of the present invention in real complex environments.

[0044] The second aspect of the technical solution: A prediction system employing a bovine core body temperature prediction method based on dual-modal RGB-T and environmental perception as described in any of the above claims, comprising: The data acquisition module is used to simultaneously acquire infrared thermal images and visible light color images of the target cattle; The image preprocessing module is used to perform spatial alignment and registration on the acquired images; The semantic segmentation module has a pre-trained bimodal semantic segmentation model built in, which is used to receive the registered image and output the segmentation results of key temperature measurement parts; A temperature extraction module is used to extract the surface temperature values ​​of key parts from the infrared thermal image based on the segmentation results. The core body temperature prediction module has a pre-established core body temperature prediction model for cattle, which is used to calculate the estimated core body temperature based on the surface temperature value. The output and warning module is used to output the core body temperature estimate and trigger a warning based on the comparison result.

[0045] Example 4: System Hardware Architecture; like Figure 5 As shown in the figure, this is a schematic diagram of the overall system structure, which shows the hardware layout and signal flow of the infrared thermal imager, visible light camera, environmental sensor and data processing unit. The cattle core body temperature monitoring system in this embodiment includes an infrared thermal imager, a visible light camera, an environmental parameter acquisition module, and a central processing unit.

[0046] The infrared thermal imager is a non-contact temperature measurement camera using the 8~14μm long-wave infrared band, mounted on a bracket above the cattle shed, covering multiple areas where cattle are active. The visible light camera is installed side by side with the infrared camera, and after calibration, it is ensured that the image captured by the visible light camera is highly overlapping with that of the infrared camera.

[0047] The environmental parameter acquisition module includes temperature and humidity sensors, anemometers, etc., which are used to record information such as ambient temperature, relative humidity and air velocity in the cattle shed in real time.

[0048] The central processing unit can be an industrial computer or server, connected to infrared cameras, visible light cameras and sensor modules, and is responsible for the synchronous acquisition, processing and analysis of data.

[0049] In practical applications, infrared and visible light cameras can be set to timed shooting or continuous video modes. In this embodiment, synchronous infrared and visible light images are captured every 5 minutes, and the data is sent to the central processing unit for storage and processing.

[0050] Example 5: System Application and Expansion The temperature measurement system of this invention can be integrated into livestock production in various ways. For example, infrared and visible light cameras can be installed at the entrance of the cattle shed, automatically capturing images and measuring temperature as the cattle pass by, thus achieving group screening; or they can be installed above the feeding passage to monitor the body temperature of each cow as it lowers its head to eat. Once an abnormal temperature is detected in a cow, the system can record its identity through an electronic ear tag worn by the cow, facilitating timely isolation and examination by staff. Furthermore, this system can interface with a ranch's Internet of Things (IoT) platform, allowing temperature data to be wirelessly transmitted to the cloud and correlated with data such as milk production and activity levels of the cows to generate a comprehensive health assessment report. In addition, the method of this invention can also be extended to the health monitoring of other large livestock (such as beef cattle and alpacas), requiring only adjustment of model parameters and retraining of the segmentation network according to different animals. Therefore, the technical solution provided by this invention has broad applicability and provides strong support for the development of modern smart animal husbandry. Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention. The scope of the invention is defined by the appended claims and their equivalents, all of which should be included within the scope of protection of the invention.

Claims

1. A method for predicting core body temperature in cattle based on dual-modal RGB-T and environmental perception, characterized in that, Includes the following steps: Step S1: Simultaneously acquire infrared thermal images and visible light color images of the target cattle; Step S2: Input the registered visible light color image and infrared thermal image into the pre-trained bimodal semantic segmentation model to identify and segment at least one key temperature measurement site of the cattle. Step S3: Based on the segmentation results of the key temperature measurement parts obtained in the dual-modal semantic segmentation step, extract the surface temperature value of the corresponding parts from the infrared thermal image. The dual-modal semantic segmentation model adopts a dual encoder-single decoder architecture. The dual encoder includes a first encoder for extracting visible light image features and a second encoder for extracting infrared thermal image features. The single decoder is used to fuse and upsample the features from the dual encoder to output the semantic segmentation result. Step S4: Input the extracted surface temperature value into the pre-established cattle core body temperature prediction model to calculate the estimated core body temperature of the target cattle.

2. The method for predicting core body temperature in cattle based on dual-modal RGB-T and environmental perception as described in claim 1, characterized in that, The specific steps in step S1 include: Environmental parameters of the cattle shed environment are collected synchronously, including at least ambient temperature and ambient humidity; in the core body temperature prediction step, the environmental parameters and the surface temperature value are input into the core body temperature prediction model.

3. The method for predicting core body temperature in cattle based on dual-modal RGB-T and environmental perception as described in claim 1, characterized in that, The bimodal semantic segmentation model performs feature fusion with the corresponding layer of the single decoder at the skip connections between the first encoder and the second encoder; and / or An attention mechanism is introduced during feature fusion to adaptively weight features from different modalities and spatial locations.

4. The method for predicting core body temperature in cattle based on dual-modal RGB-T and environmental perception as described in claim 3, characterized in that, The attention mechanism described is a dual attention mechanism that combines channel attention and spatial attention.

5. The method for predicting core body temperature in cattle based on dual-modal RGB-T and environmental perception as described in claim 1, characterized in that, The key temperature measurement areas include the area around the eyes and the ear area; In the body surface temperature extraction step, the average surface temperature of the periorbital area and the average surface temperature of the ear area are extracted.

6. The method for predicting core body temperature in cattle based on dual-modal RGB-T and environmental perception as described in claim 1, characterized in that, When an image contains multiple cattle, an individual differentiation step is also included before or after the bimodal semantic segmentation step: Using object detection or instance segmentation methods, multiple cattle in the image are separated, and the segmentation of key temperature measurement sites and extraction of body surface temperature are performed independently for each individual.

7. The method for predicting core body temperature in cattle based on dual-modal RGB-T and environmental perception as described in claim 1, characterized in that, The core body temperature prediction model is a multiple linear regression model or a nonlinear machine learning model. During the model training phase, the collected rectal temperature of cattle was used as the true label and fitted with the corresponding surface temperature value and environmental parameters. It also includes an uncertainty assessment of the output of the core body temperature estimate.

8. The method for predicting core body temperature in cattle based on dual-modal RGB-T and environmental perception as described in claim 1, characterized in that, After obtaining the core body temperature estimate, the process also includes result output and early warning steps: The estimated core body temperature is compared with a preset normal physiological body temperature range. If the value exceeds the range, an early warning signal is generated and issued.

9. The method for predicting core body temperature in cattle based on dual-modal RGB-T and environmental perception as described in claim 1, characterized in that, In the data acquisition step, the infrared thermal imager and the visible light camera are coaxially mounted or calibrated to ensure consistent field of view; In the image preprocessing step, the infrared thermal image and the visible light color image are spatially aligned and registered using calibration parameters.

10. A prediction system employing the bovine core body temperature prediction method based on dual-modal RGB-T and environmental perception as described in any one of claims 1 to 9, characterized in that, include: The data acquisition module is used to simultaneously acquire infrared thermal images and visible light color images of the target cattle; The image preprocessing module is used to perform spatial alignment and registration on the acquired images; The semantic segmentation module has a pre-trained bimodal semantic segmentation model built in, which is used to receive the registered image and output the segmentation results of key temperature measurement parts; A temperature extraction module is used to extract the surface temperature values ​​of key parts from the infrared thermal image based on the segmentation results. The core body temperature prediction module has a pre-established core body temperature prediction model for cattle, which is used to calculate the estimated core body temperature based on the surface temperature value. The output and warning module is used to output the core body temperature estimate and trigger a warning based on the comparison result.

Images