A multimodal time series synchronous analysis method for body surface temperature and motor behavior of experimental animals

Abstract

The present application relates to a kind of experimental animal body surface temperature and motion behavior multimodal time sequence synchronous analysis method applied to experimental animal behavior and physiological monitoring technical field.The method includes: acquisition infrared temperature matrix and synchronous image;Obtain body surface mask and anatomical key point and carry out multi-target tracking;Based on the geometric relationship of key point, at least three anatomical functional sub-masks are automatically divided out by the intersection operation of geometric graph element and mask based on the geometric relationship of key point;Extract the part temperature vector of each sub-zone and the part local motion intensity derived from key point displacement;Time sequence correlation analysis is carried out to temperature and motion intensity sequence, characteristic time delay is solved and temperature variation gain and time-varying response characteristics are fitted, and behavior-temperature variation coupling signature is output.The present application increases dimension to anatomical sub-zone level vector by body surface temperature, quantifies the delay, gain and time constant of behavior-driven temperature variation, and provides a new means for animal physiological evaluation.

Description

Technical Field

[0001] This invention relates to the field of laboratory animal behavior and physiological monitoring technology, specifically to a multimodal temporal synchronous analysis method for laboratory animal body surface temperature and movement behavior. Background Technology

[0002] In life science research, there is a close physiological coupling between changes in body surface temperature and movement behavior in laboratory animals (such as mice and rats). For example, the rate of metabolic heat production varies in different behavioral states such as running, standing, and grooming, leading to dynamic changes in body surface temperature distribution. Simultaneously, changes in body surface temperature can also reflect physiological indicators such as stress levels, pain intensity, and drug response. Therefore, simultaneously analyzing the temporal relationship between body surface temperature and movement behavior is of significant value for research in animal behavior, pharmacology, and physiology.

[0003] In the prior art, there are methods for non-contact measurement of animal body surface temperature using infrared thermal imaging technology. For example, Chinese Patent Publication No. CN117456607B discloses a method for online recognition of multi-target mouse behavior and monitoring body temperature, which uses infrared thermal imaging to acquire mouse images, combines instance segmentation and key point detection to obtain individual contours, and extracts the average temperature within the body contour for monitoring. However, this method has the following shortcomings.

[0004] (1) Limited temperature extraction methods: Existing technologies typically extract only the average temperature within the animal's overall outline or the temperature near a few key points (such as ear temperature and tail temperature), which cannot reflect the temperature distribution differences and dynamic changes between different anatomical functional areas. Due to the different blood perfusion, metabolic heat production, and heat dissipation characteristics of different areas of the animal's body surface (such as the tail root area, trunk core area, and auricle head side area), global average temperature or isolated point temperature is difficult to accurately characterize the temperature response characteristics of different anatomical areas during behavioral processes.

[0005] (2) Insufficient temporal correlation analysis between behavior and temperature: Existing technologies mainly perform simple decision-level fusion of behavior recognition results and temperature values ​​(such as alarm when body temperature is abnormal), without quantitatively modeling the temporal characteristics such as dynamic delay and gain between behavior intensity and temperature change. It is impossible to analyze the causal delay and response characteristics of behavior-driven temperature change, and it is also difficult to output quantitative indicators with physiological explanatory power (such as temperature change gain, response time constant, etc.).

[0006] Therefore, there is an urgent need for a method that can perform multimodal temporal synchronous analysis of the surface temperature and movement behavior of experimental animals, thereby achieving a dimensionality upgrade from global temperature measurement to anatomical partition-level temperature vector measurement, and quantifying the temporal coupling characteristics between behavior and temperature. Summary of the Invention

[0007] The technical problem to be solved by this invention is to overcome the shortcomings of existing experimental animal surface temperature and movement behavior monitoring technologies, such as the single temperature extraction method and insufficient behavior-temperature correlation analysis, and to provide a synchronous analysis method and system that can automatically divide anatomical functional areas, extract site temperature vectors, and quantify the behavior-temperature temporal coupling characteristics.

[0008] To address the above problems, this invention provides a multimodal temporal synchronization analysis method for experimental animal body surface temperature and movement behavior, comprising the following steps:

[0009] S1. Data Acquisition Steps: Obtain the radiation calibration temperature matrix sequence acquired by the infrared thermal imaging camera deployed above the experimental animal housing area. It also uses a unified time reference to acquire visible light image sequences that are time-synchronized with infrared thermal imaging sequences or intrinsic grayscale image sequences output by the same thermal imaging camera.

[0010] S2, Pose and Mask Acquisition and Tracking Steps: Process each frame of visible light image sequence or intrinsic grayscale image sequence to obtain the surface projection binary mask and anatomical key point set of each detected animal individual, and maintain a consistent identity for the same animal across frames through multi-target tracking.

[0011] S3, Sub-mask generation steps for anatomical regions: For each individual with an identity, based on the topological geometric relationship of its key point set, at least three sub-masks corresponding to the anatomical functional regions are automatically divided into the body surface projection binary mask through the intersection operation of geometric primitives and the mask. The anatomical functional regions include at least the tail root region, the trunk core region, and the auricle head side region.

[0012] S4. Steps for extracting the temperature vector sequence of a anatomical region: For each anatomical region, read the set of pixel temperature samples that fall within the corresponding submask from the temperature matrix, calculate the temperature statistics of the region, and assemble them into the temperature vector of the individual in the frame.

[0013] S5. Local motion intensity sequence extraction steps: Derivation of key point velocity from the key point displacement of the same individual between adjacent frames, and grouping and converging according to the association between key points and partitions to obtain the local motion intensity of each anatomical partition, assembling it into a local motion vector, and unifying the local temperature vector sequence and the local motion vector sequence into the same time grid.

[0014] S6. Temporal Coupling Analysis Steps: For at least one anatomical region, perform temporal correlation analysis on the temperature sequence of the region and the local motion intensity sequence of the region, solve the characteristic time delay of the region, and obtain the temperature change gain and time-varying response characteristics of the region based on the characteristic time delay fitting, and output the anatomical region-level behavior-temperature change coupling signature of the individual.

[0015] S7. Output steps: Output the coupled signature and / or the temperature change analysis results derived from the coupled signature.

[0016] Furthermore, the set of key anatomical points in S2 includes at least the tip of the nose, the base of the left ear, the base of the right ear, the neck point, the center point of the trunk, and the base of the tail; the number of anatomical divisions .

[0017] Furthermore, the tail root region submask in S3 is constructed as follows: with the tail root point as the center, and... Construct a disk with radius [radius], and then perform a logical AND operation with the binary mask projected onto the body surface. The construction method for the core torso submask is as follows: the main axis and semi-major axis of the torso are defined by a vector from the neck point to the center point of the torso; the semi-minor axis is defined by multiplying the magnitude of this vector by a coefficient, forming an elliptical or rounded rectangular primitive covering the torso area. After intersection with the binary mask projected onto the body surface, the pixels occupied by the tail root area and the auricle head side area are subtracted. The construction method for the auricle head side area submask is as follows: with the left ear base point and the right ear base point as the center, respectively, the... A disk with radius is constructed and intersected with a binary mask of the body surface projection, constrained to the head-side region defined by the nose-neck line; where the scaling factor... , The ratio of the semi-axis of the ellipse is taken from a preset parameter set corresponding to the species of the experimental animal.

[0018] Furthermore, the temperature statistics calculated in S4 also include the temperature gradient within each zone. Furthermore, when the number of pixels in a certain sub-mask is lower than a preset effective threshold or the sub-mask is determined to be significantly occluded due to pose, the temperature vector element of that component in that frame is marked as invalid, and the corresponding time is removed based on the invalid mark before subsequent steps perform correlation or fitting on the sequence.

[0019] Furthermore, between S5 and S6, a posture geometry-driven temperature compensation step is included: calculating the torso principal axis angle from the keypoint set, combining the known mounting tilt angle of the infrared thermal imaging camera and the projection distance obtained from the keypoint mapping, estimating the cosine factor of the angle between the local surface normal and the line of sight; and applying compensation to the mean temperature of each zone. Among them, calibration constant , and By projecting temperature fluctuations onto the distance during the period when the individual is judged to be stationary or at low motion. Sum of cosine factors Linear regression was performed offline; time-series correlation analysis and fitting in S6 were performed after compensation. Replace the original conduct.

[0020] Furthermore, in S6: the time series correlation analysis includes calculating the normalized cross-correlation coefficient within a preset delay window and taking the optimal delay that maximizes the correlation coefficient. Characteristic time delay The fitting process to obtain temperature-varying gain and time-varying response characteristics includes: aligning the local motion intensity of the affected area according to the optimal delay, and fitting the finite impulse response model using regularized least squares method. Output impulse response coefficient group And calculate the integral gain. and by Equivalent time constant derived from shape The coupling signature must contain at least a partition identifier. Characteristic time delay Integral gain and equivalent time constant .

[0021] Furthermore, in S2: the binary mask of body surface projection is obtained through an instance segmentation network, and the set of anatomical key points is obtained through a key point detection network; the instance segmentation network is an instance segmentation network trained with experimental animal mask annotation data, and the key point detection network is a deep learning pose estimation network trained with experimental animal key point annotation data; the multi-object tracking algorithm is StrongSort, DeepSORT, or ByteTrack.

[0022] This invention also provides a multimodal temporal synchronization analysis system for experimental animal body surface temperature and movement behavior, comprising:

[0023] The imaging acquisition module is configured to acquire radiation image sequences and calibrate them into a temperature matrix using an infrared thermal imaging camera deployed above the laboratory animal housing space. And to acquire a sequence of visible light images or an intrinsic grayscale image sequence output by the same thermal imaging camera that is synchronized with the acquisition time.

[0024] The segmentation and detection module is configured to run the instance segmentation network for each frame to output the surface projection binary mask of each individual animal, and run the key point detection network to output the set of anatomical key points, which include at least the nose tip point, left ear base point, right ear base point, neck point, trunk center point and tail root point.

[0025] The tracking module is configured to run a multi-target tracking algorithm to maintain the identity and centroid trajectory of the same animal across frames.

[0026] The anatomical partition generation module is configured to automatically divide at least three anatomical functional partition sub-masks within the body surface projection binary mask by using the topological geometric relationship of the key point set and the intersection operation of geometric primitives and the mask. The anatomical functional partitions include the tail root region, the trunk core region, and the auricle head side region.

[0027] The location temperature extraction module is configured to read temperature sample sets from the temperature matrix partition by partition, calculate the temperature statistics of each partition, and assemble them into a location temperature vector sequence.

[0028] The part motion extraction module is configured to derive the local motion intensity sequence of the part from the inter-frame displacement of key points and unify the part temperature vector sequence and the part motion vector sequence into the same time grid.

[0029] The temporal coupling parsing module is configured to perform a time-series correlation analysis with delay estimation on at least one anatomical partition using the corresponding components of the partition's temperature sequence and the local motion intensity sequence, solve for the optimal delay, fit the temperature-varying gain and time-varying response characteristics, and output the anatomical partition-level behavior-temperature-varying coupling signature.

[0030] The output module is configured to output coupled signatures and / or temperature change analysis results.

[0031] Compared with the prior art, the present invention has the following beneficial effects:

[0032] 1. Based on the topological geometric relationship of key points, this invention automatically divides anatomical regions with physiological semantics, such as the tail root region, trunk core region, and auricle head side region, through the intersection operation of geometric primitives and body surface masks. It upgrades the body surface temperature information from a global single scalar to a multi-dimensional part temperature vector, which can more accurately reflect the temperature distribution differences and dynamic changes of different anatomical regions during the behavior process, and provides a spatial analytical basis for behavior-temperature coupling analysis.

[0033] 2. This invention performs time-series correlation analysis on the temperature sequence and the local motion intensity sequence of a body part. It solves for the optimal delay time of behavior-driven temperature changes by normalizing the cross-correlation coefficient, and uses a finite impulse response model to fit the temperature change gain and equivalent time constant, outputting an anatomical partition-level coupling signature containing delay, gain, and time constant. These quantitative indicators have clear physiological significance (delay reflects the motion-heat generation-radiation transfer time, gain reflects temperature response sensitivity, and time constant reflects the rate of thermal decay), providing new objective parameters for physiological assessment of animal stress states and drug responses.

[0034] 3. This invention forms a complete automated processing chain from data acquisition, individual detection and tracking, anatomical partition generation, temperature and motion extraction to temporal coupling analysis, without the need for manual intervention. It is suitable for long-term, large-scale experimental animal monitoring scenarios and significantly reduces the cost of manual annotation.

[0035] 4. The present invention may also include posture geometry-driven temperature compensation, which automatically estimates the measurement distance and viewing angle by utilizing the geometric relationship of key points, and calibrates the compensation coefficient through offline regression of static segment data, effectively eliminating the influence of animal posture changes on the accuracy of infrared temperature measurement, and further improving the accuracy of temperature measurement. Attached Figure Description

[0036] Figure 1 This is a flowchart illustrating the multimodal temporal synchronization analysis method for experimental animal body surface temperature and movement behavior in an embodiment of the present invention.

[0037] Figure 2 This is a structural block diagram of the multimodal temporal synchronization analysis system for experimental animal body surface temperature and movement behavior in an embodiment of the present invention. Detailed Implementation

[0038] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. This embodiment uses SPF-grade C57BL / 6J mice as experimental animals to describe in detail the implementation process of the method and system of the present invention. However, the embodiments of the present invention are not limited thereto and can also be applied to other experimental animals such as rats and guinea pigs.

[0039] Example 1

[0040] I. Experimental Setup and Data Acquisition

[0041] The experimental apparatus used in this embodiment includes:

[0042] Infrared thermal imaging camera: 320×256 pixels resolution, 30fps frame rate, operating wavelength 7.5–13μm, factory radiation calibration coefficients are built into the camera firmware.

[0043] Visible light camera: 1280×1024 pixels resolution, 30fps frame rate.

[0044] Synchronous Trigger Controller: A multi-channel hardware trigger signal generator that simultaneously outputs TTL pulses to the external trigger interfaces of both the infrared and visible light cameras, ensuring that the frame acquisition time error between the two cameras is less than 1ms.

[0045] The cage is made of transparent polymethyl methacrylate (PMMA) with an open top and dimensions of 400mm×250mm×200mm. The camera is mounted 800mm directly above the cage, with the infrared camera's optical axis pointing vertically downwards.

[0046] II. Method and Flow

[0047] Please see Figure 1 The multimodal temporal synchronization analysis method for experimental animal body surface temperature and movement behavior provided in this embodiment includes the following steps.

[0048] S1. Data Acquisition Steps

[0049] The synchronization trigger controller is activated, and the infrared thermal imaging camera and visible light camera simultaneously acquire image sequences at 30fps. The infrared camera outputs a 14-bit digital signal, which is converted into a temperature matrix after radiometric calibration. Each pixel value represents the radiation temperature (in °C) at that point. The data acquisition lasted 10 minutes, resulting in 18,000 frames. A timestamp for each frame was also recorded. ( (Frame number).

[0050] S2. Attitude and Mask Acquisition and Tracking Steps

[0051] For each visible light image frame, inference is first performed using a pre-trained YOLOv8-seg instance segmentation network. The training dataset for this network contains 5000 visible light images labeled with binary masks of mouse body surface projections (i.e., each pixel in each image is labeled to indicate whether it represents a mouse body surface), enabling the network to output a binary mask of the body surface projection for each mouse. Simultaneously, a pre-trained YOLOv8-pose keypoint detection network (training dataset contains 5000 visible light images labeled with 6 keypoints) is used to output the pixel coordinates of the 6 anatomical keypoints:

[0052] tip of nose

[0053] left ear base point

[0054] Right ear base point

[0055] Neck point

[0056] center point of the torso

[0057] tail root point

[0058] Subsequently, the StrongSort multi-target tracking algorithm was run, using the centroid of the keypoint set and the mask as the observation, to assign a unique identifier to each mouse. ( , (Number of mice in the current frame), and generate centroid trajectory. .

[0059] S3, Steps for Generating Anatomical Partition Submasks

[0060] For each identified mouse, three anatomical functional area sub-masks are constructed in each frame based on keypoint coordinates:

[0061] (1) Tail root sub-mask :Tail root point Centered on, radius (Right now Construct a disk, then combine it with a binary mask projected onto the body surface. Perform a pixel-by-pixel logical AND operation to obtain the tail root region mask.

[0062] (2) Submask of the core area of ​​the torso : with neck point and the center point of the torso Define the direction of the torso's principal axis, with a principal axis length of... With the midpoint of the principal axis as the center of the ellipse, and the semi-major axis as... The semi-minor axis is Construct an elliptical primitive. Then connect this elliptical primitive with... After the intersection operation, the pixels already occupied by the tail root region and the auricle head side region are subtracted pixel by pixel to obtain the torso core region mask.

[0063] (3) Submask of the auricle head region : Taking the left ear base point respectively and right ear base point Centered on, radius Pixels are used as disks, and two disks are merged and combined with... Intersection operation is performed, and the area is further constrained to the head side region defined by the line connecting the tip of the nose and the neck (pixel position is determined by the equation of a straight line) to obtain the auricular head side region mask.

[0064] The above parameters , Pixels, ellipse half-axis ratio All measurements were pre-set for the C57BL / 6J mouse size. For other species, a small number of calibration frames can be pre-calculated to obtain the corresponding scaling coefficients.

[0065] S4, Steps for extracting temperature vector sequences at specific locations

[0066] For each anatomical region ( ), in the temperature matrix Read the corresponding submask from the middle Temperature values ​​of all pixels within Calculate the temperature statistics for this zone:

[0067] Average temperature of each zone

[0068] highest temperature in the zone

[0069] Temperature gradient within the zone

[0070] The average temperature of the three partitions of the frame is assembled into a site temperature vector. If the number of pixels in a sub-mask of a certain region is less than 10 (the preset effective threshold), or if the region is detected to be severely occluded (e.g., the mask area of ​​the torso core region is less than 50% of the historical average area), then the component is marked as invalid (NaN) and the corresponding time point will be automatically removed during subsequent analysis.

[0071] S5. Steps for extracting local motion intensity sequences of body parts

[0072] For each mouse with a unique identifier, calculate the displacement velocity of key points between adjacent frames. Using the torso center point... For example, its speed ,in s.

[0073] Then, the key points are grouped according to their anatomical regions:

[0074] Key point for attribution of the tail root region: Tail root point

[0075] Key point for assigning the core area of ​​the torso: Neck point Center point of the torso

[0076] Key point for classifying the lateral region of the auricle: left ear base point Right ear base point , tip of nose (Because the tip of the nose and the ear area move in unison)

[0077] For each partition Calculate the average velocity of all assigned key points within the partition (if there are no key points within the partition, take the centroid velocity) to obtain the local motion intensity of the location. Assembled into part motion vectors .

[0078] The temperature vector sequence and motion vector sequence of the affected area are aligned by timestamps to form synchronized multivariate time-series data.

[0079] S6. Steps for resolving temporal coupling

[0080] With the core area of ​​the torso ( Taking this as an example, extract the temperature sequence of this partition. and motion intensity sequence ( First, the sequence is normalized to zero mean.

[0081] (1) Solving for the characteristic time delay : In the preset delay window Within the frame (corresponding to -1s to +1s), calculate the normalized cross-correlation coefficient:

[0082]

[0083] Take The largest Then the characteristic time delay (seconds). In this embodiment, the core area of ​​the torso was measured. s (meaning that the temperature begins to change significantly about 0.5 seconds after the movement).

[0084] (2) Fitting temperature-varying gain and equivalent time constant: Align the motion intensity sequence according to the optimal delay, let The finite impulse response (FIR) model is fitted using the regularized least squares method:

[0085]

[0086] in (Corresponding to a 0.5s historical window). Ridge regression (L2 regularization coefficient) is used. Solve for the impulse response coefficient set Calculate the integral gain. This reflects the temperature change caused by a unit motion intensity under steady-state conditions (unit: ℃·s / pixel). The equivalent time constant is calculated based on the centroid of the impulse response. .

[0087] In this embodiment, the core area of ​​the torso ℃·s / pixel s.

[0088] Output Coupling Signature Similarly, calculations were performed on the tail root region and the head lateral region of the auricle to obtain a complete set of coupled signatures.

[0089] S7, Output Steps

[0090] The coupled signature set is output in JSON format, and a temperature change analysis report is generated, including: delay-gain scatter plots for each partition, dynamic curves of time constant changes with behavioral state, etc. The experimental monitoring interface can also display the temperature vector and motion intensity waveforms of each partition of the mouse in real time, as well as the predicted temperature change trend.

[0091] III. Attitude geometry driven temperature compensation (optional enhancement)

[0092] In another embodiment, to further improve temperature measurement accuracy, an attitude geometry-driven temperature compensation step is added between S5 and S6:

[0093] (1) Calculate the trunk principal axis angle from key points Assuming the local normal to the body surface is approximately perpendicular to the principal axis of the torso (within the horizontal plane), then the angle between the line of sight and the normal... The camera can be mounted with a tilt angle. (This embodiment) (Vertical downward) and projection distance Estimate: .in The depth of the center point of the torso can be calculated by assuming the animal's body surface is located on the bottom plane of the cage, based on camera calibration parameters and pixel coordinates.

[0094] (2) Apply compensation to the average temperature of the zones: Temperature fluctuation data during the stationary phase (centroid velocity < 2 pixels / frame for 30 consecutive frames) was used for linear regression calibration. and , Take the average distance of the stationary segment. After calibration, the temperature series in S6 is changed to... Analysis can effectively reduce temperature fluctuations caused by changes in posture.

[0095] IV. Temperature Change Decoupling and Early Warning (Optional Enhancement)

[0096] In yet another optional implementation (this implementation is not a necessary technical feature of the present invention, but is only for further explanation), the behavior-driving components are calculated based on the FIR model. Then, calculate the residuals. Set partition-specific thresholds. For the core area of ​​the torso When the mouse's behavioral state is determined to be stationary (center of mass velocity < 1 pixel / frame) and If the temperature change persists for more than 5 seconds, a non-motion-related temperature change warning will be issued, indicating that there may be stress or pathological conditions.

[0097] V. System Implementation Examples

[0098] Please see Figure 2 This embodiment also provides a multimodal temporal synchronization analysis system for experimental animal body surface temperature and movement behavior, including:

[0099] Imaging acquisition module: Infrared thermal imaging camera and visible light camera, driven by synchronous trigger controller, acquire data and transmit it to host computer.

[0100] Segmentation and detection module: YOLOv8-seg and YOLOv8-pose models deployed on GPU servers, outputting masks and key points.

[0101] Tracking module: Online multi-target tracking implemented using the StrongSort algorithm, maintaining IDs and trajectories.

[0102] Anatomical partition generation module: Automatically generates sub-masks for the tail root region, trunk core region, and auricle head side region according to the geometric construction method described in S3.

[0103] Location temperature extraction module: Reads the pixel temperature of each partition from the temperature matrix, calculates statistics and assembles the temperature vector.

[0104] Partial motion extraction module: The motion intensity of each region is derived from the velocity of key points.

[0105] Temporal coupling parsing module: Implements cross-correlation delay estimation and FIR fitting, and outputs coupling signature.

[0106] Output module: Displays and stores the results.

[0107] The modules communicate with each other via the ROS (Robot Operating System) message bus, and the real-time processing frame rate can reach 30fps.

[0108] VI. Experimental Results

[0109] The above method was used to record the behavior and analyze the temperature of 10 C57BL / 6J mice for 10 minutes. The results showed that:

[0110] The temperature vectors of anatomical regions showed significant differences under different behavioral states such as running, standing still, and grooming, with the core trunk region exhibiting the largest temperature fluctuation. ), followed by the lateral area of ​​the auricle ( ), the tail root area is the smallest ( ).

[0111] Delay obtained from timing coupling analysis It exhibits stability across different individuals, with an average core area of ​​the trunk. s; Integral gain Positively correlated with exercise intensity; equivalent time constant Significantly shortened under stress ( ( ), which can be used as a non-invasive evaluation indicator of physiological state.

[0112] The attitude compensation step reduced the average temperature variance in the stationary section by 42%, effectively improving the stability of the temperature measurement data.

[0113] In summary, this invention enables multimodal temporal synchronous analysis of experimental animal body surface temperature and movement behavior, and outputs physiologically meaningful quantitative coupled signatures, providing a new technical means for monitoring animal behavior and physiology.

[0114] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A multimodal temporal synchronous analysis method for analyzing the body surface temperature and movement behavior of experimental animals, characterized in that, Includes the following steps: S1. Data Acquisition Steps: Obtain the radiation calibration temperature matrix sequence acquired by the infrared thermal imaging camera deployed above the experimental animal housing area. And its unified time reference, and acquire a visible light image sequence that is time-synchronized with the infrared thermal imaging sequence or an intrinsic grayscale image sequence output by the same thermal imaging camera; S2, Pose and Mask Acquisition and Tracking Steps: Process the visible light image sequence or intrinsic grayscale image sequence of each frame to obtain the surface projection binary mask and anatomical key point set of each detected animal individual, and maintain a consistent identity for the same animal across frames through multi-target tracking. S3. Anatomical sub-mask generation step: For each individual with the identity identifier, based on the topological geometric relationship of its key point set, at least three sub-masks corresponding to the anatomical functional regions are automatically divided inside the binary mask projected on the body surface through the intersection operation of geometric primitives and the mask. The anatomical functional regions include at least the tail root region, the trunk core region, and the auricle head side region. S4. Location temperature vector sequence extraction step: For each anatomical region, read the pixel temperature sample set that falls within the corresponding submask from the temperature matrix, calculate the temperature statistics of the region and assemble them into the location temperature vector of the individual in the frame. S5. Local motion intensity sequence extraction steps: Derivation of key point velocity from the key point displacement of the same individual between adjacent frames, and grouping and converging according to the association relationship between key points and partitions to obtain the local motion intensity of each anatomical partition, assembling it into a local motion vector, and unifying the local temperature vector sequence and the local motion vector sequence into the same time grid. S6. Temporal Coupling Analysis Steps: For at least one anatomical region, perform temporal correlation analysis on the temperature sequence of the region and the local motion intensity sequence of the region to solve the characteristic time delay of the region, and obtain the temperature change gain and time change response characteristics of the region based on the characteristic time delay, and output the anatomical region-level behavior-temperature change coupling signature of the individual. S7. Output step: Output the coupled signature and / or the temperature change analysis result derived from the coupled signature.

2. The method according to claim 1, characterized in that, The set of key anatomical points in S2 includes at least the tip of the nose, the base of the left ear, the base of the right ear, the neck point, the center point of the trunk, and the base of the tail; the number of anatomical divisions .

3. The method according to claim 1, characterized in that, The tail root sub-mask in S3 is constructed as follows: with the tail root point as the center, and... Construct a disk with radius [radius], and then perform a logical AND operation with the binary mask projected onto the body surface. ; The construction method of the core area submask of the torso is as follows: the main axis direction and semi-major axis of the torso are defined by the vector from the neck point to the center point of the torso, and the semi-minor axis is defined by multiplying the magnitude of the vector by a coefficient to form an elliptical or rounded rectangular primitive covering the torso area. After the intersection operation with the binary mask of the body surface projection, the pixels occupied by the tail root area and the auricle head side area are deducted. The auricular head side region sub-mask is constructed as follows: with the left ear base point and the right ear base point as the center, respectively, and... After constructing a disk with radius, perform an intersection operation with the binary mask of the body surface projection, and constrain it to be within the head side region divided by the nose-neck line; The proportionality coefficient mentioned above , The ratio of the semi-axis of the ellipse is taken from a preset parameter set corresponding to the species of the experimental animal.

4. The method according to claim 1, characterized in that, The temperature statistics calculated in S4 also include the temperature gradient within each zone. ; Furthermore, when the number of pixels in a certain sub-mask is lower than a preset effective threshold or the sub-mask is determined to be significantly occluded due to pose, the temperature vector element of that component in that frame is marked as invalid. Subsequent steps will remove the corresponding time based on the invalid mark before performing correlation or fitting on the sequence.

5. The method according to claim 1, characterized in that, Between S5 and S6, there is also an attitude geometry-driven temperature compensation step: The trunk principal axis angle is calculated from the set of key points. Combined with the known mounting angle of the infrared thermal imaging camera and the projection distance obtained by mapping from the key points, the cosine factor of the angle between the local normal of the body surface and the line of sight is estimated. Apply compensation to the average temperature of each zone: Among them, calibration constant , and By projecting temperature fluctuations onto the distance during the period when the individual is judged to be stationary or at low motion. Sum of cosine factors Obtained offline by performing linear regression; The temporal correlation analysis and fitting in S6 are compensated for... Replace the original conduct.

6. The method according to claim 1, characterized in that, In S6: The time-series correlation analysis includes calculating the normalized cross-correlation coefficient within a preset delay window and selecting the optimal delay that maximizes the correlation coefficient. The characteristic time delay ; The fitting process to obtain temperature-varying gain and time-varying response characteristics includes: aligning the local motion intensity of the site according to the optimal delay, and fitting the finite impulse response model using the regularized least squares method. Output impulse response coefficient group And calculate the integral gain. and by Equivalent time constant derived from shape ; The coupling signature includes at least a partition identifier. Characteristic time delay Integral gain and equivalent time constant .

7. The method according to claim 1, characterized in that, In S2: The body surface projection binary mask is obtained through an instance segmentation network, and the set of anatomical key points is obtained through a key point detection network. The instance segmentation network is an instance segmentation network trained with experimental animal mask annotation data, and the key point detection network is a deep learning pose estimation network trained with experimental animal key point annotation data. The multi-target tracking algorithm is StrongSort, DeepSORT, or ByteTrack.

8. A multimodal temporal synchronous analysis system for analyzing the surface temperature and movement behavior of experimental animals, characterized in that, include: The imaging acquisition module is configured to acquire radiation image sequences and calibrate them into a temperature matrix using an infrared thermal imaging camera deployed above the laboratory animal housing space. And to acquire a sequence of visible light images or an intrinsic grayscale image sequence output by the same thermal imaging camera that is synchronized with the acquisition time. The segmentation and detection module is configured to run the instance segmentation network for each frame to output the surface projection binary mask of each individual animal, and run the key point detection network to output the set of anatomical key points, wherein the key points include at least the nose tip point, the base point of the left ear, the base point of the right ear, the neck point, the center point of the trunk, and the tail root point. The tracking module is configured to run a multi-target tracking algorithm to maintain the identity and centroid trajectory of the same animal across frames. The anatomical partition generation module is configured to automatically divide at least three anatomical functional partition sub-masks within the body surface projection binary mask by using the topological geometric relationship of the key point set and the intersection operation of the geometric primitives and the mask. The anatomical functional partitions include the tail root region, the trunk core region, and the auricle head side region. The location temperature extraction module is configured to read temperature sample sets from the temperature matrix partition by partition, calculate the temperature statistics of each partition, and assemble them into a location temperature vector sequence. The part motion extraction module is configured to derive a local motion intensity sequence of parts from the inter-frame displacement of key points and unify the part temperature vector sequence and the part motion vector sequence into the same time grid. The temporal coupling parsing module is configured to perform a time-series correlation analysis with delay estimation on at least one anatomical partition using the corresponding components of the partition's temperature sequence and the local motion intensity sequence, solve for the optimal delay, fit the temperature-varying gain and time-varying response characteristics, and output the anatomical partition-level behavior-temperature-varying coupling signature. The output module is configured to output the coupled signature and / or temperature change analysis results.

Images