Method for pain threshold determination based on multimodal automated laser stimulation and animal behavior analysis

Abstract

This invention discloses a method and system for pain threshold determination based on multimodal automatic laser stimulation and animal behavior analysis. The method includes: using an RVC-Pose convolutional neural network model to detect key points on the animal's foot and outputting the foot's spatial coordinates in real time; positioning a laser spot on the foot and outputting continuously adjustable laser stimulation according to a preset gradient; extracting facial key points and micro-expression features using a Light-Face facial recognition algorithm, and / or reconstructing the animal's three-dimensional skeleton using a multi-view geometric reconstruction algorithm and extracting pain-related behavioral features; inputting the multimodal behavioral data into a multi-parameter fusion judgment model to automatically identify pain responses, adaptively adjusting the stimulation intensity under gradient stimulation mode, terminating stimulation and recording the pain threshold when an effective pain response is detected. This invention achieves fully automatic closed-loop control for pain threshold determination, solving the problems of inaccurate positioning, imprecise stimulation, and subjective judgment in existing technologies, significantly improving the objectivity and accuracy of pain assessment.

Description

Technical Field

[0001] This invention relates to the field of animal experimental equipment technology, specifically to a method and system for pain threshold determination based on multimodal automated laser stimulation and animal behavior analysis. Background Technology

[0002] In research fields such as biomedicine and neuroscience, accurate determination of animal pain thresholds and controllable application of pain stimulation are fundamental to studying pain mechanisms, screening analgesic drugs, and exploring neural regulation. Traditional pain stimulation methods often employ mechanical stimulation (Von Frey fiber method) and thermal stimulation, but these methods struggle to achieve precise gradient control of stimulation intensity, and the stimulation site relies on manual localization, leading to localization bias and poor repeatability of experimental data. Furthermore, pain threshold determination primarily depends on researchers visually observing animal somatic reflexes, resulting in significant subjective judgment errors and hindering standardized and automated determination, severely impacting the scientific validity and reliability of experimental results. Therefore, existing methods for animal pain stimulation and pain threshold determination suffer from inaccurate localization, imprecise quantification, subjective judgment, and inflexible control, specifically manifested as cumbersome operation, high subjectivity, low stimulation precision, and the inability to achieve automated, targeted, and quantitative stimulation. Moreover, existing systems struggle to automatically trigger stimulation based on preset conditions, failing to meet the quantitative and targeted stimulation requirements of specific experimental scenarios, thus limiting the flexibility and efficiency of experimental design.

[0003] Therefore, there is an urgent need to develop a system and method that can achieve automated and standardized precise pain threshold determination and controllable pain stimulation application, as well as automatic identification of animal body parts, gradient intensity laser stimulation, and quantitative and directional laser pain stimulation according to preset conditions, in order to break through the existing technological bottlenecks. Summary of the Invention

[0004] In view of the above-mentioned technical problems, the present invention provides a method and system for pain threshold determination based on multimodal automated laser stimulation and animal behavior analysis, aiming to build a fully automated, standardized, and intelligent pain threshold determination technology platform, realize closed-loop control of the entire process from stimulus application to behavioral perception to pain threshold determination, completely replace traditional mechanical stimulation, manual operation and subjective judgment, significantly improve the objectivity, accuracy and data richness of pain experiments, and provide a reliable, efficient and traceable experimental tool for pain mechanism research, neural regulation laws and analgesic drug screening.

[0005] To achieve the above objectives, in a first aspect, the present invention provides a method for pain threshold determination based on multimodal automated laser stimulation and animal behavior analysis, comprising the following steps:

[0006] S1 Foot Localization: This method acquires images of the animal's foot and uses a deep learning-based RVC-Pose convolutional neural network model to detect and locate key points in the foot region, outputting the animal's foot spatial coordinates in real time. The deep learning-based RVC-Pose convolutional neural network model includes: a backbone feature extraction network based on an improved CSP structure, a multi-scale feature fusion network, and a key point regression prediction network. Its overall input-output relationship is expressed as follows:

[0007]

[0008] in, For the input image features, Batch size represents the number of samples input into the network at one time; The number of feature channels in the image; For feature map height, The width of the feature map; This represents the backbone feature extraction network; This represents a multi-scale feature fusion network; This indicates a keypoint regression prediction network. The spatial coordinates of the animal's foot;

[0009] S2 laser stimulation, based on the spatial coordinates of the animal's foot, drives the XYZ three-axis displacement mechanism to position the laser spot output by the semiconductor laser emitter or programmable dot matrix laser array at the spatial coordinates of the animal's foot, and outputs laser stimulation with continuously adjustable intensity according to a preset gradient;

[0010] S3 behavior perception collects multimodal behavior data of animals, including facial images captured by a front-facing industrial camera and three-dimensional motion images captured by a multi-view synchronous camera system. The system uses a lightweight deep learning model, Light-Face, to extract facial key points and pain-related micro-expression features from the facial images, and / or uses a multi-view geometric reconstruction algorithm to reconstruct the animal's full-body three-dimensional skeleton from the three-dimensional motion images and extract pain-related behavioral features. The lightweight deep learning model, Light-Face, includes a multi-branch convolutional backbone network, a local enhancement feature module, and a bi-branch prediction network.

[0011] S4 fusion judgment and closed-loop control inputs the multimodal behavioral data into the multi-parameter fusion judgment model to automatically identify whether the animal produces a pain response; in gradient stimulation mode, the laser stimulation intensity is adaptively adjusted according to the recognition result, and stimulation is immediately terminated when an effective pain response is detected, and the current stimulation intensity is recorded as the pain threshold.

[0012] Preferably, the backbone feature extraction network incorporates a lightweight attention module that includes channel attention and coordinate attention mechanisms; the multi-scale feature fusion network adopts a bidirectional fusion structure of top-down and bottom-up. Further, the backbone feature extraction network is composed of cascaded multi-level convolutional modules, each of which can be represented as:

[0013]

[0014] in, Used to extract local spatial features; Channel compression and feature reconstruction; Batch normalization operation; This is the ReLU activation function.

[0015] Furthermore, the backbone feature extraction network is constructed based on an improved CSP structure, and its features are represented as follows:

[0016]

[0017] The input features are divided into two branches: , This indicates the convolution transformation path: one branch propagates directly, while the other performs feature transformation and then merges the results. This setup reduces computation and improves gradient propagation capabilities.

[0018] Furthermore, a lightweight attention module is introduced into the deep features, and its output is:

[0019]

[0020] Channel attention is:

[0021]

[0022] Coordinate attention is:

[0023]

[0024] in, For global average pooling; To encode position information along the horizontal and vertical directions respectively; This is element-wise multiplication.

[0025] The multi-scale feature fusion network includes:

[0026] Top-down path:

[0027]

[0028] Bottom-up path:

[0029]

[0030] in: Indicates upsampling, For convolutional downsampling with a stride of 2, feature fusion is performed by element-wise addition.

[0031] The keypoint regression prediction network uses a convolutional regression structure:

[0032]

[0033] in, This indicates the coordinates of the key points in the output. This represents the local features in the keypoint regression prediction network.

[0034] In this application, "animal" is preferably a mouse, but is not limited thereto.

[0035] Preferably, the lightweight deep learning model Light-Face facial recognition algorithm described in this application includes a multi-branch convolutional backbone network, a local enhancement feature module, and a dual-branch prediction network; the local enhancement feature module can learn local convolutional masks and a lightweight spatial attention mechanism, and the dual-branch prediction network outputs a facial key point heatmap and micro-expression classification probability, respectively. The Light-Face facial recognition algorithm is used to recognize micro-expressions on animal faces. Its overall structure includes a backbone feature extraction network, a local enhancement feature module, a multi-scale fusion network, and a key point and expression prediction network. The overall mapping relationship is expressed as follows:

[0036]

[0037] in, To input a facial image, Batch size represents the number of samples input into the network at one time. The number of feature channels in the image; For feature map height, The width of the feature map. Output the heatmap for key points. Output the probability of each facial expression category.

[0038] Furthermore, the backbone feature extraction network adopts a multi-branch convolutional structure, and its feature extraction process is represented as follows:

[0039]

[0040] in, This is a detail branch (local texture). For context branches (global information) Shallow enhancement branch, For depthwise separable convolution, This is the GELU activation function.

[0041] The branch merging method is as follows:

[0042]

[0043] in, This is the sum of the channel dimensions.

[0044] Furthermore, the calculation method for the local enhancement feature module is as follows:

[0045]

[0046] Further integrate spatial attention:

[0047]

[0048] in, To learnable local masks, This is a spatial attention map. This is element-wise multiplication.

[0049] Feature alignment is performed using deformable convolution:

[0050]

[0051] in, This is the learned sampling offset.

[0052] Multi-scale fusion employs weighted skip connections:

[0053]

[0054] in, For learnable weights, Features at different scales.

[0055] Global context enhancement:

[0056]

[0057] Furthermore, the key point and expression prediction network employs a dual-branch output: a heatmap branch outputting a facial key point heatmap; and an expression classification branch outputting micro-expression classification probabilities. High-dimensional features are mapped through a lightweight fully connected layer. Combined with an anchor point adaptive adjustment mechanism, the accuracy of key point detection in small regions is enhanced.

[0058] Key point heatmap branches:

[0059]

[0060] Emoji Category Branches:

[0061]

[0062] The anchor point adaptive mechanism further optimizes key point prediction as follows:

[0063]

[0064] in, This is an offset predicted based on local features, used to improve the localization accuracy of small targets.

[0065] Preferably, the multi-view geometric reconstruction algorithm includes:

[0066] Select a ground reference point to fit the plane and calculate the normal vector. Rotate the coordinate system to align the ground with the Z=0 plane.

[0067] Rotate the coordinate axes in the plane according to the reference boundary direction to align them with the direction of the experimental setup.

[0068] Translate the center point to the origin and reset the ground height to zero to establish a unified world coordinate system;

[0069] The triangulation method or the least squares method is used to back-calculate the two-dimensional key points from multiple perspectives into three-dimensional space, generating the three-dimensional skeleton of the whole animal, i.e., the three-dimensional coordinates of the key points of the skeleton.

[0070] Furthermore, after obtaining the multi-camera calibration parameters and two-dimensional key points, the geometric reconstruction algorithm constructs a unified world coordinate system and achieves three-dimensional reconstruction. The overall process can be represented as follows:

[0071]

[0072] in, For multi-view two-dimensional key points; This indicates triangulation or optimized reconstruction; Indicates coordinate system rotation and alignment; This represents a translation-normalization transformation; These are the final 3D keypoint coordinates.

[0073] Selecting a set of ground reference points Fitting the plane equation:

[0074]

[0075] Solve using the least squares method:

[0076]

[0077] Obtain the plane normal vector:

[0078]

[0079] Construct rotation matrix , so that the normal vector Align to standard Z-axis direction :

[0080]

[0081] Rotate all points:

[0082]

[0083] Achieve ground and Plane alignment.

[0084] Furthermore, a reference boundary direction vector is selected. Construct a rotation matrix around the Z-axis :

[0085]

[0086] in:

[0087]

[0088] Coordinates updated to:

[0089]

[0090] Ultimately, the experimental setup was aligned with the coordinate axes.

[0091] Then select the center point and translate it:

[0092]

[0093] At the same time, the ground altitude is reset to zero:

[0094]

[0095] Furthermore, semantic parsing is performed on the three-dimensional behavioral data to extract pain-related behavioral features, including foot lifting response, paw licking behavior, escape behavior, and movement trajectory, so as to achieve spatiotemporal dynamic modeling and quantitative representation of pain response.

[0096] Furthermore, the gradient stimulation mode involves the laser output increasing progressively from low to high according to a preset gradient, with each level of stimulation lasting for a preset time. A reasonable time interval is set between two adjacent stimuli to avoid the cumulative effect of stimulation interfering with the experimental results. The system monitors the animal's (in this field, "animal" often refers to mice, but is not limited to mice; it can be extended to other suitable animals in this field) response in real time during stimulation to determine the pain threshold.

[0097] Furthermore, the fixed stimulation mode: using fixed laser power and fixed stimulation duration for single or repeated stimulation, is used to evaluate the pain response characteristics of animals at a specific intensity (such as response incidence, response latency and response intensity, etc.), and is suitable for behavioral or pharmacodynamic studies.

[0098] Preferably, during each laser stimulation process, the multi-parameter fusion judgment model employs a multi-modal fusion mechanism to simultaneously detect pain responses, including simultaneously executing the following steps:

[0099] Real-time detection of whether animals exhibit escape reflex behaviors such as lifting or retracting their feet;

[0100] Facial expression analysis: Based on facial key point detection and expression feature extraction algorithms, it identifies pain-related facial expression changes (such as squinting, facial contraction, etc.).

[0101] Pupil dynamic detection: Extracting the pupil area and analyzing its diameter changes and contraction / dilation dynamics to reflect the autonomic nerve response triggered by pain.

[0102] The signal analysis unit performs real-time fusion and judgment of multi-source data:

[0103] In gradient-increment mode, if no clear pain response is detected after multiple consecutive stimulations, the system automatically moves to the next gradient stimulation. If a pain response reaching the threshold is detected, multiple stimulations are given again at the current threshold. If the response rate reaches the preset value, the laser output is immediately stopped by software control, and the current stimulation intensity is recorded. In fixed stimulation mode, the system records various response indicators during and after stimulation for subsequent statistical analysis.

[0104] Furthermore, the structured storage and analysis of the experimental data are as follows: Under the gradient-incremental stimulation mode, the laser stimulation intensity corresponding to the first elicitation of a clear pain response is defined as the pain threshold of the animal. To improve the stability and reliability of the measurement results, the same animal can be subjected to 2-3 independent tests, and the average value is taken as the final pain threshold. The system automatically records and stores all experimental data, including stimulation parameters, response type, timestamps, and image data, supporting subsequent analysis and result traceability.

[0105] Preferably, in the laser stimulation step, the dot matrix laser plate is composed of multiple independent and controllable laser emitting units. Each emitting unit of the semiconductor laser emitter outputs a wavelength of 638nm (or other near-infrared bands) and a single-point power range of 0-2W. After passing through an optical shaping system, it forms a dotted light spot with a diameter of 0.5-1mm on the animal's foot. The programmable dot matrix laser array is controlled by a single-point activation mode or a multi-point synchronous activation mode. More preferably, each laser unit is independently driven by a control module, enabling point-to-point switching control, zone control, and dynamic scanning control, and supporting continuous or discrete gradient adjustment. In this way, the system can form stimulation patterns with different intensity distributions in space, including dotted, linear, or regional stimulation.

[0106] Furthermore, the dot matrix laser array can be selected in the following modes according to experimental requirements:

[0107] Single-point activation mode: simulates traditional focused laser stimulation, used for single-point pain stimulation experiments;

[0108] Multi-point synchronous activation mode: to realize the research of complex pain spatial coding, and to form specific stimulation patterns by activating multiple points simultaneously;

[0109] Furthermore, in terms of optical structure, the dot matrix laser output can be shaped by a microlens array or a collimating lens to obtain a uniform and consistent spot size; by optimizing the laser unit spacing and working distance, multiple discrete spots with a diameter of about 0.5-1mm can be formed on the sole of an animal's foot, or a uniform stimulus can be formed in a specific area by superposition.

[0110] Preferably, the positioning accuracy of the XYZ three-axis displacement mechanism is ±0.01mm, and the response time is ≤0.2s; the drive control adopts coordinate mapping, linear interpolation trajectory planning, and PID closed-loop feedback adjustment. Further, the XYZ three-axis displacement mechanism is an orientation adjustment component, driven by a stepper motor, transmitted through a ball screw, and equipped with a linear guide to ensure motion accuracy. The positioning accuracy of each axis can reach ±0.01mm, the response time is ≤0.2s, and the maximum travel distance of the X / Y / Z axes is 150mm, 150mm, and 50mm, respectively.

[0111] In a second aspect, the inventors provide a pain threshold determination system based on multimodal automated laser stimulation and animal behavior analysis, comprising:

[0112] The precise foot positioning and recognition module includes an animal restraint box, a high-transparency glass set at the bottom of the animal restraint box, an image acquisition device, and a processing unit; the image acquisition device captures images of the animal's foot through the high-transparency glass, and the processing unit outputs the spatial coordinates of the animal's foot using an RVC-Pose convolutional neural network model.

[0113] A quantitative and directional laser stimulation module includes a semiconductor laser emitter or a programmable dot matrix laser array, an optical shaping system, an XYZ three-axis displacement mechanism, and a drive control unit; the drive control unit is connected to the output of the processing unit to receive the spatial coordinates of the animal's foot and controls the XYZ three-axis displacement mechanism to position the laser spot at the spatial coordinates of the animal's foot.

[0114] The behavior acquisition and perception module includes an acquisition unit and a 3D behavior analysis unit. The acquisition unit comprises a front-facing industrial camera and a first processing subunit, which uses the Light-Face facial recognition algorithm to extract key facial points and pain-related micro-expression features. The 3D behavior analysis unit comprises a closed experimental setup equipped with multiple high-resolution industrial cameras, a hardware synchronization triggering circuit, and a second processing subunit. The second processing subunit uses a multi-view geometric reconstruction algorithm to reconstruct the animal's full-body 3D skeleton and extract pain-related behavioral features.

[0115] A multimodal fusion judgment and closed-loop control module includes a controller and a memory. The controller is connected to a foot precise positioning and recognition module, a quantitative and directional laser stimulation module, and a behavior acquisition and sensing module, respectively. The memory stores a multi-parameter fusion judgment model. The controller is configured to execute the pain threshold measurement method described in the first aspect of the present invention. More preferably, in the quantitative and directional laser stimulation module, the optical shaping system sequentially includes a beam expander, a continuously variable neutral density attenuator, a focusing lens, and an adjustable aperture along the laser propagation path.

[0116] The multimodal fusion judgment and closed-loop control module has a built-in dedicated motion control card, and its control logic is as follows:

[0117] Coordinate mapping and command generation: Through pre-defined coordinate mapping relationships, the spatial coordinate information of the animal's foot transmitted in real time by the foot precise positioning and recognition module is converted into target position commands;

[0118] Trajectory planning and driving: The motion control card uses a linear interpolation algorithm to plan the motion of each axis and generate pulse sequences to drive the stepper motor;

[0119] Closed-loop feedback control: The encoder provides real-time feedback on the actual position of each axis, and the control card dynamically corrects the drive signal based on the deviation using a PID controller, achieving fast and stable positioning. Laser alignment: Through the above process, the laser is precisely aligned with the spatial coordinates of the animal's foot or target located by the foot-positioning and recognition module, achieving directional stimulation. Furthermore, the entire control process has a delay controlled within milliseconds and features over-limit position protection, further ensuring the reliability and accuracy of positioning.

[0120] In a third aspect of the invention, the inventors provide an electronic device comprising: a memory having program code stored thereon; and a processor connected to the memory, wherein when the program code is executed by the processor, the pain threshold determination method as described in the first aspect of the invention is implemented.

[0121] In a fourth aspect, the inventors provide a computer-readable storage medium storing program instructions that, when executed, implement the pain threshold measurement method as described in the first aspect of the invention. Through this computer-readable storage medium, a user can deploy the program instructions for the pain threshold measurement method as described in the first aspect of the invention on different computing platforms, achieving comprehensive control of various functional modules of the experimental chamber and pain threshold measurement for multimodal automated laser stimulation and animal behavior analysis.

[0122] The above aspects of the present invention can be applied to basic research on pain mechanisms, screening and efficacy evaluation of analgesic drugs, evaluation of pain-related disease models, and evaluation of the effects of non-pharmacological treatments for pain, filling the technical gap in the prior art regarding traditional methods for capturing somatic, facial, and autonomic nerve responses, as well as quantifying spontaneous and induced pain.

[0123] Unlike existing technologies, the above technical solution has at least the following beneficial effects:

[0124] 1. Precise positioning, no human error: The global shutter camera and deep learning algorithm of the image acquisition device realize the precise recognition of the animal's foot coordinates. Combined with the XYZ three-axis electric displacement stage (positioning accuracy ±0.01mm) to realize the directional adjustment of the laser, the control module has a built-in motion control card to complete the direct conversion of coordinates and drive signals. The positioning process is fully automated from hardware to software, completely avoiding human positioning error and improving the accuracy of stimulation.

[0125] 2. Standardization and automation of pain threshold determination: Gradient intensity laser stimulation is used, and a dedicated algorithm is used to automatically identify the foot lifting and retraction reflex, eliminating the need for manual observation and judgment. The multimodal fusion judgment and closed-loop control module automatically records the stimulation intensity when the animal produces a pain reflex, achieving standardization and automation of pain threshold determination, and greatly improving the repeatability and scientific validity of experimental data.

[0126] 3. Quantitative and directional controllable, flexible experimental design: The quantitative and directional laser stimulation module is equipped with complete optical components to achieve quantitative control of the laser spot. The XYZ three-axis electric displacement stage enables directional adjustment. The preset drive control unit supports time triggering and external event triggering (compatible with BNC level interface + RS485 serial port), which can be linked with various external physiological image acquisition devices. At the same time, it can precisely control the intensity, duration and number of laser stimulations to meet the pain stimulation needs in different experimental scenarios, thereby improving the flexibility and efficiency of experimental design.

[0127] 4. Robust hardware safety and high system integration: Equipped with multiple safety features such as physical relays, emergency stop buttons, and light shields, the system automatically cuts off power when the laser loses its target, ensuring experimental safety. Each module is connected via a standardized wired interface, resulting in high system integration and easy operation. No manual intervention from professionals is required, making it widely applicable to animal pain-related experiments in fields such as biomedicine and neuroscience, reducing the workload of experimental personnel and improving experimental efficiency.

[0128] 5. Achieving fully automated closed-loop research: Through the fully automated linkage of image recognition, quantitative and directional adjustment, laser stimulation, reflection detection, and trigger control, the four major pain points of existing equipment—inaccurate positioning, imprecise quantification, subjective judgment, and inflexible control—are solved. This upgrades the research platform from manual / semi-automatic measurement to a fully automated closed-loop research platform, providing more reliable experimental equipment for pain mechanism research, analgesic drug screening, and exploration of neural regulation patterns.

[0129] The above description of the invention is merely an overview of the technical solution of this application. In order to enable those skilled in the art to better understand the technical solution of this application and to implement it based on the description and drawings, and to make the above-mentioned objectives and other objectives, features and advantages of this application easier to understand, the following description is provided in conjunction with the specific embodiments and drawings of this application. Attached Figure Description

[0130] The accompanying drawings are only used to illustrate the principles, implementation methods, applications, features, and effects of specific embodiments of this application and other related content, and should not be considered as limitations on this application.

[0131] In the accompanying drawings of the instruction manual:

[0132] Figure 1 This is an overall flowchart of Embodiment 1 of the present invention;

[0133] Figure 2 This is a hardware schematic diagram of Example 1;

[0134] Figure 3 This is a diagram of the main control board in Example 1;

[0135] Figure 4 This is a schematic diagram of the RVC-Pose convolutional neural network structure in Example 1;

[0136] Figure 5 This is a schematic diagram of the network structure of the Light-Face facial recognition algorithm in Example 1.

[0137] The reference numerals used in the above figures are explained as follows:

[0138] 1. Animal restraint box; 2. First infrared camera; 3. Head holder; 4. High-transmittance glass; 5. Second infrared camera; 6. Laser emitter; 7. XYZ three-axis displacement gimbal power supply; 8. Equipment power supply; 9. XYZ three-axis displacement platform; 10. Main control circuit board; 11. Frame structure; 12. Dot matrix laser plate; 13. Three-dimensional movable replacement bucket. Detailed Implementation

[0139] To illustrate the possible application scenarios, technical principles, implementable specific solutions, and achievable objectives and effects of this application in detail, the following description, in conjunction with the listed specific embodiments and accompanying drawings, provides a detailed explanation. The embodiments described herein are merely illustrative of the technical solutions of this application and are therefore intended to limit the scope of protection of this application.

[0140] In this document, the term "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The term "embodiment" appearing in various places throughout the specification does not necessarily refer to the same embodiment, nor does it specifically limit its independence or connection with other embodiments. In principle, in this application, as long as there are no technical contradictions or conflicts, the technical features mentioned in each embodiment can be combined in any way to form corresponding implementable technical solutions.

[0141] Unless otherwise defined, the technical terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the use of related terms herein is merely for the purpose of describing particular embodiments and is not intended to limit this application.

[0142] In the description of this application, the term "and / or" is used to describe the logical relationship between objects, indicating that three relationships can exist. For example, A and / or B means: A exists, B exists, and A and B exist simultaneously. Additionally, the character " / " in this document generally indicates that the preceding and following objects have an "or" logical relationship.

[0143] In this application, terms such as “first” and “second” are used only to distinguish one entity or operation from another, and do not necessarily require or imply any actual quantity, hierarchy or order relationship between these entities or operations.

[0144] Unless otherwise specified, the use of terms such as “comprising,” “including,” “having,” or other similar expressions in this application is intended to cover non-exclusive inclusion, which does not exclude the presence of additional elements in a process, method, or product that includes the stated elements, such that a process, method, or product that includes a list of elements may include not only those defined elements but also other elements not expressly listed, or elements inherent to such a process, method, or product.

[0145] As understood in the Examination Guidelines, in this application, expressions such as "greater than," "less than," and "exceeding" are understood to exclude the stated number; expressions such as "above," "below," and "within" are understood to include the stated number. Furthermore, in the description of the embodiments in this application, "multiple" means two or more (including two), and similar expressions related to "multiple" are also understood in this way, such as "multiple groups" and "multiple times," unless otherwise explicitly specified.

[0146] In the description of the embodiments of this application, the space-related expressions used, such as "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "vertical," "top," "bottom," "inner," "outer," "clockwise," "counterclockwise," "axial," "radial," and "circumferential," indicate the orientation or positional relationship based on the orientation or positional relationship shown in the specific embodiments or drawings. They are only for the purpose of describing the specific embodiments of this application or for the reader's understanding, and do not indicate or imply that the device or component referred to must have a specific position, a specific orientation, or be constructed or operated in a specific orientation. Therefore, they should not be construed as limitations on the embodiments of this application.

[0147] Unless otherwise expressly specified or limited, the terms "installation," "connection," "linking," "fixing," and "setting," as used in the description of the embodiments of this application, should be interpreted broadly. For example, "connection" can be a fixed connection, a detachable connection, or an integral setting; it can be a mechanical connection, an electrical connection, or a communication connection; it can be a direct connection or an indirect connection through an intermediate medium; it can be the internal connection of two components or the interaction between two components. For those skilled in the art to which this application pertains, the specific meaning of the above terms in the embodiments of this application can be understood according to the specific circumstances.

[0148] Example 1

[0149] A multimodal perception-based animal pain threshold determination and quantitative directional laser pain stimulation system, such as Figure 1 As shown, the system includes the following steps:

[0150] S1, Foot Precision Positioning and Recognition Module:

[0151] A high-definition infrared industrial camera is installed at the bottom of the experimental chamber to capture images of the animal's foot through a high-transmittance glass. Combined with the RVC-Pose convolutional neural network model based on deep learning, the animal's foot area is detected and key points are located with high precision, realizing real-time acquisition of the foot's spatial position and providing accurate spatial coordinate information for subsequent laser stimulation.

[0152] S2, Quantitative and Directional Laser Stimulation Module:

[0153] A semiconductor laser emitter or a programmable dot matrix laser array is used as the pain stimulus source, and a continuously adjustable gradient intensity laser signal is output under the drive of the control module. The semiconductor laser emitter achieves precise control of the laser spot size and spatial position through an optical shaping system and a high-precision XYZ three-axis displacement mechanism, so that the laser can act precisely on the sole of the animal's foot with preset intensity, duration and spatial coordinates, realizing quantitative and directional pain stimulus output.

[0154] S3, Multi-scenario Animal Behavior Acquisition and Multimodal Perception Module:

[0155] The system is configured with two switchable experimental scenario structures to adapt to different research paradigms:

[0156] The head-fixation experimental mode employs a head-fixation structure (such as...) Figure 2 The animal restraint box of the head holder 3 shown in the figure, combined with the front-mounted industrial camera to acquire animal facial images, realizes the detection of facial key points and the recognition of pain-related micro-expressions through the lightweight deep learning model Light-Face facial recognition algorithm; at the same time, combined with the pupil detection module, the eye area is segmented and tracked in real time, and autonomic nerve response indicators such as pupil diameter change and contraction / dilation rate are extracted to realize the refined analysis of pain response.

[0157] The free-movement three-dimensional behavior analysis mode replaces the animal restraint box with a head clamp 3 with a cylindrical or enclosed animal behavior experimental device that uses multi-view synchronous acquisition (such as...). Figure 2The three-dimensional active replacement tank 13 shown is equipped with multiple high-resolution industrial cameras (preferably four or more) around the tank to simultaneously capture images of the experimental animals from different fixed angles. Each camera achieves time alignment through a hardware synchronization triggering mechanism and transmits multiple image data to the back-end processing unit. Through multi-view RVC-Pose deep learning algorithm and geometric reconstruction algorithm, the real-time reconstruction of the animal's whole-body three-dimensional skeleton is achieved. Based on the three-dimensional skeleton information, the three-dimensional behavioral data is further semantically parsed based on a lightweight visual language model to extract pain-related behavioral features, including foot lifting response, paw licking behavior, escape behavior, and movement trajectory, thereby realizing spatiotemporal dynamic modeling and quantitative representation of pain response.

[0158] S4, Multimodal Fusion Decision and Closed-Loop Control Module:

[0159] The system uses a software platform to preset experimental paradigms (including gradient stimulation mode or fixed stimulation mode) and integrates the following multimodal data: plantar reflex behavior (2D / 3D), facial expression features, pupil dynamic changes, and three-dimensional movement behavior sequences to construct a multi-parameter fusion judgment model, enabling automatic identification and classification of animal pain responses. In gradient stimulation mode, the system automatically adjusts the stimulation intensity based on real-time detection results and terminates stimulation immediately upon detecting a valid pain response, recording the corresponding stimulation intensity as the pain threshold, thus achieving fully automated closed-loop pain threshold measurement. Simultaneously, the system performs structured storage and analysis of data throughout the experiment, enabling pain threshold calculation, behavioral statistical analysis, and traceable management of experimental results.

[0160] like Figure 2As shown, the experimental enclosure includes two animal restraint boxes 1, a first infrared camera 2, a high-transmittance glass 4, a second infrared camera 5 and third to sixth infrared cameras (not shown), a laser emitter 6, a dot matrix laser plate 12, a power supply 8, an XYZ three-axis displacement gimbal power supply 7, an XYZ three-axis displacement gimbal 9, a main control circuit board 10, and a frame structure 11. The frame structure 11 is constructed using a combination of aluminum profiles and acrylic sheets. The interior of the enclosure is equipped with copper mesh and sound insulation cotton to achieve electromagnetic shielding and noise reduction, thereby improving the stability of system operation and imaging quality. The XYZ three-axis displacement gimbal 9 is connected to the analysis equipment via a wired connection, and a communication architecture is constructed with wired connection as the primary method and wireless connection as a secondary method. One of the two animal restraint boxes is an animal restraint box with a head clamp 3 used for head-fixed experimental mode. It uses the lightweight deep learning model Light-Face facial recognition algorithm to detect facial key points and recognize pain-related micro-expressions. At the same time, it combines a pupil detection module to perform real-time segmentation and tracking of the eye area, and extract autonomic nerve response indicators such as pupil diameter changes and contraction / dilation rates to achieve refined analysis of pain response. Another type is a cylindrical or enclosed animal behavior experimental device for multi-view synchronous acquisition, such as, but not limited to, an animal restraint box using a three-dimensional activity replacement bucket 13. Multiple high-resolution industrial cameras (preferably four or more) are arranged around the box to synchronously acquire images of the experimental animals from different fixed angles. Each camera achieves time alignment through a hardware synchronous triggering mechanism and transmits multiple image data to the back-end processing unit. Through multi-view RVC-Pose deep learning algorithm and geometric reconstruction algorithm, the real-time reconstruction of the animal's whole-body three-dimensional skeleton is realized. Based on the three-dimensional skeleton information, the three-dimensional behavioral data is further semantically parsed based on a lightweight visual language model to extract pain-related behavioral features, including foot lifting response, paw licking behavior, escape behavior, and movement trajectory, so as to realize spatiotemporal dynamic modeling and quantitative characterization of pain response.

[0161] like Figure 3 As shown, the main control circuit board 10 serves as the core control unit of the system and is equipped with multiple types of physical interfaces for data interaction and control signal transmission with various functional modules; the preset trigger module is adapted to a standardized external linkage signal interface for receiving external trigger signals and controlling the system to perform corresponding operations; the high-transmittance glass 4 has a transmittance of not less than 92% in the infrared laser band used, and its size is 15cm×15cm with a thickness of 5mm.

[0162] The above modules work together to form a fully automatic closed-loop control system, which is used to determine the pain threshold of animals and to provide quantitative and targeted pain stimulation.

[0163] like Figure 4As shown, the deep learning-based RVC-Pose convolutional neural network model includes a backbone feature extraction network, a feature fusion network, and a keypoint prediction network. Its overall input-output relationship can be expressed as:

[0164]

[0165] in, Input image features; Batch size represents the number of samples input into the network at one time; The number of feature channels in the image; For feature map height, The width of the feature map; This represents a backbone feature fusion network; Represents a feature fusion network; This indicates a keypoint prediction network. To output the coordinates of key points.

[0166] The backbone feature extraction network consists of cascaded convolutional modules, each of which can be represented as:

[0167]

[0168] in, Used to extract local spatial features; Channel compression and feature reconstruction; Batch normalization operation; This is the ReLU activation function.

[0169] The backbone network is built on an improved CSP structure, and its characteristics are as follows:

[0170]

[0171] The input features are divided into two branches: , This represents the convolutional transformation path, where one branch propagates directly, while the other undergoes feature transformation and is then fused. This reduces computational cost and improves gradient propagation capabilities.

[0172] A lightweight attention mechanism is introduced into deep features, and its output is:

[0173]

[0174] Channel attention is:

[0175]

[0176] Coordinate attention is:

[0177]

[0178] in, For global average pooling; To encode position information along the horizontal and vertical directions respectively; This is element-wise multiplication.

[0179] The feature fusion network employs a multi-scale fusion approach:

[0180] Top-down path:

[0181]

[0182] Bottom-up path:

[0183]

[0184] in: Indicates upsampling, For convolutional downsampling with a stride of 2, feature fusion is performed by element-wise addition.

[0185] Further channel alignment:

[0186]

[0187] The keypoint prediction network employs a convolutional regression structure:

[0188]

[0189] in, This indicates the coordinates of the key points in the output. This represents local features in a keypoint regression prediction network.

[0190] The laser stimulation module includes a semiconductor laser emitter and an optical modulation component; wherein, the output wavelength of the semiconductor laser emitter is 638nm, the power is adjustable in the range of 0–2W, and the laser intensity is continuously gradient-adjusted under the drive of the control module, and its adjustment range covers the animal pain response threshold range (0.1–2W) to achieve quantitative pain stimulation output of different intensities;

[0191] The optical modulation assembly, along the laser propagation path, includes, in sequence, a beam expander, a neutral density attenuator, a focusing lens, and an adjustable aperture, and its specific structure is as follows:

[0192] The beam expander is preferably a Galilean structure, used to expand the laser beam diameter to approximately 5 mm, thereby reducing the beam divergence angle and improving subsequent focusing stability. The neutral density attenuator is a continuously variable structure with an optical density adjustment range of 0.1–3.0, and is rotated and adjusted by a stepper motor to achieve fine control of the laser output power. The focusing lens is preferably a plano-convex lens with a focal length of 75 mm, used to focus the expanded laser beam. The aperture of the adjustable stop is adjustable from 0.1 to 2 mm, used to limit the final output spot size. After the above optical modulation, the laser forms a dotted spot with a diameter of 0.5–1 mm on the sole of the animal's foot. The spot size can be finely adjusted by adjusting the relative distance between the focusing lens and the target surface and the aperture of the stop to ensure the spatial accuracy and consistency of the stimulation area.

[0193] Programmable dot matrix laser arrays are used to achieve multi-point synchronous or programmable spatially distributed optical stimulation output;

[0194] The fractional laser array consists of multiple independently controllable laser emitting units, each with an output wavelength of 638nm (or other near-infrared bands) and a single-point power range of 0–2W. Each laser unit is independently driven by a control module, enabling point-to-point switching control, zone control, and dynamic scanning control, and supporting continuous or discrete gradient adjustment. Thus, the system can create stimulation patterns with different intensity distributions in space, including point, line, or regional stimulation. The fractional laser array can be selected in the following modes according to experimental requirements:

[0195] Single-point activation mode: simulates traditional focused laser stimulation for single-point pain stimulation experiments; multi-point synchronous activation mode: enables research on complex pain spatial coding, forming specific stimulation patterns through simultaneous activation of multiple points.

[0196] In terms of optical structure, the dot matrix laser output can be shaped by a microlens array or a collimating lens to obtain a uniform and consistent spot size; by optimizing the laser unit spacing and working distance, multiple discrete spots with a diameter of about 0.5–1 mm can be formed on the sole of an animal's foot, or a uniform stimulus can be formed in a specific area by superposition.

[0197] The XYZ three-axis displacement mechanism is a directional adjustment component driven by a stepper motor, transmitted through a ball screw, and equipped with linear guides to ensure motion accuracy. The positioning accuracy of each axis can reach ±0.01mm, the response time is ≤0.2s, and the maximum travel of the X / Y / Z axes is 150mm, 150mm, and 50mm, respectively.

[0198] The control module incorporates a dedicated motion control card, whose control logic is as follows: Coordinate mapping and command generation: Through pre-calibrated coordinate mapping relationships, the foot coordinate information transmitted in real time by the image recognition module is converted into target position commands; Trajectory planning and driving: The motion control card uses a linear interpolation algorithm to plan the motion of each axis, generating pulse sequences to drive the stepper motors; Closed-loop feedback control: The encoder provides real-time feedback on the actual position of each axis, and the control card dynamically corrects the drive signal based on the deviation using a PID controller, achieving fast and stable positioning; Laser alignment: Through the above process, the laser is precisely aligned with the animal's foot or target area located by the image recognition module, achieving directional stimulation. Furthermore, the entire control process has a delay controlled within milliseconds and features position over-limit protection, further ensuring the reliability and accuracy of positioning.

[0199] like Figure 5 As shown, the Light-Face facial recognition algorithm is used to identify micro-expressions on animal faces. Its overall structure includes a backbone feature extraction network, a local enhancement feature module, a multi-scale fusion network, and a key point and expression prediction network. The overall mapping relationship is expressed as follows:

[0200]

[0201] in, To input a facial image, Batch size represents the number of samples input into the network at one time. The number of feature channels in the image; For feature map height, The width of the feature map. Output the heatmap for key points. Output the probability of each facial expression category.

[0202] The backbone network employs a multi-branch convolutional structure, and its feature extraction process is represented as follows:

[0203]

[0204] in, This is a detail branch (local texture). For context branches (global information) Shallow enhancement branch, For depthwise separable convolution, This is the GELU activation function.

[0205] The branch merging method is as follows:

[0206]

[0207] in, This is the sum of the channel dimensions.

[0208] The local enhancement feature module is calculated as follows:

[0209]

[0210] Further integrate spatial attention:

[0211]

[0212] in, To learnable local masks, This is a spatial attention map. This is element-wise multiplication.

[0213] Feature alignment is performed using deformable convolution:

[0214]

[0215] in, This is the learned sampling offset.

[0216] Multi-scale fusion employs weighted skip connections:

[0217]

[0218] in, For learnable weights, Features at different scales.

[0219] Global context enhancement:

[0220]

[0221] The keypoint and expression prediction network employs a dual-branch output: a heatmap branch outputting a heatmap of facial keypoints; and an expression classification branch outputting micro-expression classification probabilities. High-dimensional features are mapped through lightweight fully connected layers. An adaptive anchor point adjustment mechanism enhances the accuracy of keypoint detection in small regions.

[0222] Key point heatmap branches:

[0223]

[0224] Emoji Category Branches:

[0225]

[0226] The anchor point adaptive mechanism further optimizes key point prediction as follows:

[0227]

[0228] in, This is an offset predicted based on local features, used to improve the localization accuracy of small targets.

[0229] After obtaining multi-camera calibration parameters and 2D key points, the geometric reconstruction algorithm constructs a unified world coordinate system and achieves 3D reconstruction. The overall process can be represented as follows:

[0230]

[0231] in, For multi-view two-dimensional key points; This indicates triangulation or optimized reconstruction; Indicates coordinate system rotation and alignment; This represents a translation-normalization transformation; These are the final 3D keypoint coordinates.

[0232] Selecting a set of ground reference points Fitting the plane equation:

[0233]

[0234] Solve using the least squares method:

[0235]

[0236] Obtain the plane normal vector:

[0237]

[0238] Construct rotation matrix , so that the normal vector Align to standard Z-axis direction :

[0239]

[0240] Rotate all points:

[0241]

[0242] Achieve ground and Plane alignment.

[0243] Reference boundary direction vector Construct a rotation matrix around the Z-axis :

[0244]

[0245] in:

[0246]

[0247] Coordinates updated to:

[0248]

[0249] Ultimately, the experimental setup was aligned with the coordinate axes.

[0250] Then select the center point. To translate:

[0251]

[0252] At the same time, the ground altitude is reset to zero:

[0253]

[0254] Furthermore, based on a lightweight visual language model, semantic analysis is performed on the three-dimensional behavioral data to extract pain-related behavioral features, including foot lifting response, paw licking behavior, escape behavior, and movement trajectory, thereby achieving spatiotemporal dynamic modeling and quantitative representation of pain response.

[0255] The gradient stimulation mode involves progressively increasing laser output from low to high according to a preset gradient, with each stimulation level lasting for a preset duration. A reasonable time interval is set between adjacent stimulations to avoid the cumulative effect of stimulation interfering with the experimental results. The system monitors the animal's response in real time during stimulation to determine the pain threshold.

[0256] Fixed stimulation mode: Single or repeated stimulation is performed using fixed laser power and fixed stimulation duration to assess the pain response characteristics of animals at a specific intensity (such as response incidence, response latency and response intensity), and is suitable for behavioral or pharmacodynamic studies.

[0257] The multi-parameter fusion judgment model involves the system simultaneously detecting pain responses using a multi-modal fusion mechanism during each laser stimulation process, including:

[0258] Behavioral response detection: Real-time identification of whether animals exhibit escape reflex behaviors such as lifting or withdrawing their feet;

[0259] Facial expression analysis: Based on facial key point detection and expression feature extraction algorithms, it identifies pain-related facial expression changes (such as squinting, facial contraction, etc.).

[0260] Pupil dynamic detection: Extracting the pupil area and analyzing its diameter changes and contraction / dilation dynamics to reflect the autonomic nerve response triggered by pain.

[0261] The signal analysis unit performs real-time fusion and judgment of multi-source data:

[0262] In gradient-increment mode, if no clear pain response is detected after multiple consecutive stimulations, the system automatically moves to the next gradient stimulation. If a pain response reaching the threshold is detected, multiple stimulations are given again at the current threshold. If the response rate reaches the preset value, the laser output is immediately stopped by software control, and the current stimulation intensity is recorded. In fixed stimulation mode, the system records various response indicators during and after stimulation for subsequent statistical analysis.

[0263] The experimental data was structured and stored for analysis as follows: Under the gradient-incremental stimulation mode, the laser stimulation intensity corresponding to the first elicitation of a clear pain response was defined as the animal's pain threshold. To improve the stability and reliability of the measurement results, the same animal could be tested independently 2-3 times, and the average value was taken as the final pain threshold. The system automatically recorded and stored all experimental data, including stimulation parameters, response type, timestamps, and image data, supporting subsequent analysis and result traceability.

[0264] Example 2

[0265] An electronic device includes a processor and a memory. The memory stores program instructions for implementing the animal pain threshold determination and quantitative directional laser pain stimulation system of Embodiment 1. When the processor executes the program instructions, it can implement the following functional modules: a plantar precise positioning and recognition module, a quantitative directional laser stimulation module, a behavior acquisition and perception module, and a multimodal fusion determination and closed-loop control module.

[0266] Example 3

[0267] A computer-readable storage medium is provided, pre-stored with computer-executable program instructions for implementing all functions of the multimodal perception-based animal pain threshold determination and quantitative directional laser pain stimulation system of Embodiment 1. Through this computer-readable storage medium, users can deploy the system described in this invention on different computing platforms to achieve comprehensive control of each functional module of the experimental chamber and automatic determination of animal pain thresholds and multimodal behavioral analysis.

[0268] Finally, it should be noted that although the above embodiments have been described in the text and drawings of this application, this should not limit the scope of patent protection of this application. Any technical solutions that are based on the essential concept of this application and utilize the content described in the text and drawings of this application, resulting in equivalent structural or procedural substitutions or modifications, as well as the direct or indirect application of the technical solutions of the above embodiments to other related technical fields, are all included within the scope of patent protection of this application.

Claims

1. A method for pain threshold determination based on multimodal automated laser stimulation and animal behavior analysis, characterized in that, Includes the following steps: S1 Foot Localization: This method acquires images of the animal's foot and uses a deep learning-based RVC-Pose convolutional neural network model to detect and locate key points in the foot region, outputting the animal's foot spatial coordinates in real time. The deep learning-based RVC-Pose convolutional neural network model includes: a backbone feature extraction network based on an improved CSP structure, a multi-scale feature fusion network, and a key point regression prediction network. Its overall input-output relationship is as follows: ； in, Input image features; Batch size represents the number of samples input into the network at one time. The number of feature channels in the image; The feature map height; The width of the feature map; This represents the backbone feature extraction network; This represents a multi-scale feature fusion network; This indicates a keypoint regression prediction network. The spatial coordinates of the animal's foot; S2 laser stimulation, based on the spatial coordinates of the animal's foot, drives the XYZ three-axis displacement mechanism to position the laser spot output by the semiconductor laser emitter or programmable dot matrix laser array at the spatial coordinates of the animal's foot, and outputs laser stimulation with continuously adjustable intensity according to a preset gradient; S3 behavior perception collects multimodal behavior data of animals, including facial images and three-dimensional motion images; uses the Light-Face facial recognition algorithm to extract facial key points and pain-related micro-expression features from the facial images, and / or uses a multi-view geometric reconstruction algorithm to reconstruct the animal's full-body three-dimensional skeleton from the three-dimensional motion images and extract pain-related behavior features; S4 fusion judgment and closed-loop control inputs the multimodal behavioral data into the multi-parameter fusion judgment model to automatically identify whether the animal produces a pain response; in gradient stimulation mode, the laser stimulation intensity is adaptively adjusted according to the recognition result, and stimulation is terminated when an effective pain response is detected, and the current stimulation intensity is recorded as the pain threshold.

2. The pain threshold determination method according to claim 1, characterized in that, The backbone feature extraction network includes a lightweight attention module with channel attention and coordinate attention mechanisms; the multi-scale feature fusion network adopts a bidirectional fusion structure of top-down and bottom-up.

3. The pain threshold determination method according to claim 1, characterized in that, The Light-Face facial recognition algorithm includes a multi-branch convolutional backbone network, a local enhancement feature module, and a dual-branch prediction network. The local enhancement feature module can learn local convolutional masks and lightweight spatial attention mechanisms, and the dual-branch prediction network outputs facial key point heatmaps and micro-expression classification probabilities, respectively.

4. The pain threshold determination method according to claim 1, characterized in that, The multi-view geometric reconstruction algorithm includes: Select a ground reference point to fit the plane and calculate the normal vector. Rotate the coordinate system to align the ground with the Z=0 plane. Rotate the coordinate axes in the plane according to the reference boundary direction to align them with the direction of the experimental setup. Translate the center point to the origin and reset the ground height to zero to establish a unified world coordinate system; The triangulation method or the least squares method is used to back-calculate the two-dimensional key points from multiple perspectives into three-dimensional space to generate the three-dimensional skeleton of the animal.

5. The pain threshold determination method according to claim 1, characterized in that, The multi-parameter fusion decision model is executed synchronously in each laser stimulation step: Detection of foot lifting or retraction reflex behavior, recognition of pain expressions such as squinting or facial contraction, and analysis of pupil diameter changes; When a pain response reaching the preset response rate is detected in the gradient increment mode, the laser output is terminated and the current stimulation intensity is recorded.

6. The pain threshold determination method according to claim 1, characterized in that, In the laser stimulation step, the output wavelength of the semiconductor laser emitter is 638nm, the power adjustment range is 0-2W, and after passing through the optical shaping system, it forms a dotted light spot with a diameter of 0.5-1mm on the sole of the animal's foot; the programmable dot matrix laser array is controlled in a single-point activation mode or a multi-point synchronous activation mode.

7. The pain threshold determination method according to claim 1, characterized in that, The positioning accuracy of the XYZ three-axis displacement mechanism is ±0.01mm, and the response time is ≤0.2s; the drive control adopts coordinate mapping, linear interpolation trajectory planning and PID closed-loop feedback adjustment.

8. A pain threshold determination system based on multimodal automated laser stimulation and animal behavior analysis, characterized in that, include: The precise foot positioning and recognition module includes an animal restraint box, a high-transparency glass set at the bottom of the animal restraint box, an image acquisition device, and a processing unit; the image acquisition device captures images of the animal's foot through the high-transparency glass, and the processing unit outputs the spatial coordinates of the animal's foot using an RVC-Pose convolutional neural network model. A quantitative and directional laser stimulation module includes a semiconductor laser emitter or a programmable dot matrix laser array, an optical shaping system, an XYZ three-axis displacement mechanism, and a drive control unit; the drive control unit is connected to the output of the processing unit to receive the spatial coordinates of the animal's foot and control the XYZ three-axis displacement mechanism to position the laser spot at the spatial coordinates of the animal's foot. The behavior acquisition and perception module includes an acquisition unit and a three-dimensional behavior analysis unit. The acquisition unit consists of a front-facing industrial camera and a first processing subunit. The first processing subunit uses the Light-Face facial recognition algorithm to extract key facial points and pain-related micro-expression features. The three-dimensional behavior analysis unit consists of a closed experimental device equipped with multiple high-resolution industrial cameras, a hardware synchronous triggering circuit, and a second processing subunit. The second processing subunit uses a multi-view geometric reconstruction algorithm to reconstruct the animal's full-body three-dimensional skeleton and extract pain-related behavioral features. as well as A multimodal fusion judgment and closed-loop control module includes a controller and a memory; the controller is connected to a foot precise positioning and recognition module, a quantitative directional laser stimulation module and a behavior acquisition and sensing module respectively; the memory stores a multi-parameter fusion judgment model; the controller is configured to execute the pain threshold measurement method according to any one of claims 1 to 7, wherein, in the quantitative directional laser stimulation module, the optical shaping system sequentially includes a beam expander, a continuously variable neutral density attenuator, a focusing lens and an adjustable aperture along the laser propagation path.

9. An electronic device, characterized in that, include: A memory that stores program code; A processor, connected to the memory, and when the program code is executed by the processor, implements the pain threshold determination method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It stores program instructions that, when executed, implement the pain threshold determination method as described in any one of claims 1 to 7.

Images