Cat health intelligent detection method and system based on spatio-temporal behavior residual error analysis

Abstract

This invention discloses a method and system for intelligent feline health detection based on spatiotemporal behavioral residual analysis. The method specifically includes: collecting multimodal behavioral tuples during a cat's healthy period to construct a health behavior database; generating an ideal healthy behavior flow model by performing behavioral flow decomposition and spatiotemporal pattern mining on the health behavior database; comparing the multimodal behavioral tuples with the ideal healthy behavior flow model to construct a spatiotemporal residual tensor, and using statistical process control methods to determine the significance of the residuals, outputting residual pattern classification results; based on the residual pattern classification results, constructing specific detection models for postoperative recovery, pain stress, and urinary abnormalities respectively, and using multi-scenario fusion decision-making for conflict resolution and priority ranking, outputting a comprehensive health status and action suggestions. This invention achieves accurate monitoring and abnormal detection of feline health status through multimodal data fusion and spatiotemporal residual analysis.

Description

Technical Field

[0001] This invention relates to the interdisciplinary fields of pet intelligent monitoring, computer vision and veterinary artificial intelligence, and in particular to a method and system for intelligent detection of cat health based on spatiotemporal behavioral residual analysis. Background Technology

[0002] With the booming development of the pet economy and the widespread adoption of smart home devices, cat health monitoring has become a research hotspot for the application of artificial intelligence in vertical fields.

[0003] However, a deeper analysis of existing technologies reveals that their core judgment logic generally suffers from the following fundamental limitations: First, most existing systems suffer from single-modal perception problems, either relying solely on collar IMU data, which can only detect "the cat moved" but cannot determine "where and how the cat moved"; or relying solely on vision devices, resulting in blind spots and failure at night. A single sensor cannot fully capture the multidimensional features of a cat's behavior, and a single-modal system cannot capture all of them, leading to inaccurate monitoring results.

[0004] Secondly, most existing systems compare current data with group averages or individual historical averages. When the data is still within the "normal range" but the behavioral pattern has undergone structural changes, the system completely fails. For example, in the early stages of feline urinary tract diseases, the frequency of littering may still be within the normal range, but the "time spent loitering before littering" and "time spent licking after littering" have changed significantly. Existing threshold systems cannot capture this change in the temporal structure of behavior, thus missing the best opportunity for early diagnosis.

[0005] Furthermore, existing technology cannot answer the question of "what should ideal behavior look like in a healthy state." When a cat's postoperative activity level decreases by 50%, it is impossible to determine whether this is a normal postoperative recovery process or a sign of poor pain control, because the lack of a model of "the ideal behavioral trajectory that the cat should have at the current recovery stage" makes it impossible to accurately assess the cat's health status.

[0006] Finally, existing systems lack targeted modeling for specific medical scenarios, often applying a general behavioral analysis framework to all situations, resulting in high false alarm rates and low clinical reference value. For example, cats' behavioral characteristics and health indicators differ in different scenarios, and general models cannot accurately identify health problems in various scenarios, causing confusion for pet owners and veterinarians. Summary of the Invention

[0007] The purpose of this invention is to provide a method and system for intelligent detection of cat health based on spatiotemporal behavioral residual analysis. Through multimodal data fusion and spatiotemporal residual analysis, it achieves accurate monitoring and anomaly detection of cat health status, thereby solving at least one of the aforementioned problems in the prior art.

[0008] In a first aspect, the present invention provides a method for intelligent detection of cat health based on spatiotemporal behavioral residual analysis, the method specifically comprising: Collect cat motion data and visual spatial data, interpolate and align the cat motion data and visual spatial data in a unified time coordinate system and fuse features to generate multimodal behavior tuples; A health behavior database was constructed by collecting multimodal behavioral tuples from cats during their healthy period. By performing behavioral flow decomposition and spatiotemporal pattern mining on the health behavior database, an ideal healthy behavior flow model was generated. The multimodal behavior tuples are compared with the ideal health behavior flow model. The residuals of exercise intensity, region deviation, behavior sequence, behavior duration, posture, and toilet behavior are calculated. The spatiotemporal residual tensor is constructed, and the significance of the residuals is determined by statistical process control method. The residual pattern classification results are output. Based on the residual pattern classification results, specific detection models were constructed for postoperative recovery, pain stress, and urinary abnormalities, respectively. Conflict resolution and priority ranking were carried out through multi-scenario fusion decision-making, and comprehensive health status and action suggestions were output.

[0009] Secondly, this invention provides a cat health intelligent detection system based on spatiotemporal behavioral residual analysis, the system specifically comprising: The data fusion module is used to collect cat motion data and visual spatial data, and to interpolate, align and fuse the cat motion data and visual spatial data in a unified time coordinate system to generate multimodal behavior tuples. The feature analysis module is used to collect multimodal behavioral tuples during the cat's healthy period to build a healthy behavior library. By performing behavioral flow decomposition and spatiotemporal pattern mining on the healthy behavior library, a healthy ideal behavior flow model is generated. The pattern classification module is used to compare multimodal behavior tuples with the ideal health behavior flow model, calculate the residuals of exercise intensity, region deviation, behavior sequence, behavior duration, posture, and toilet behavior, construct the spatiotemporal residual tensor, and use statistical process control methods to determine the significance of the residuals, and output the residual pattern classification results. The scene detection module is used to construct specific detection models for postoperative recovery, pain stress and urinary abnormalities based on the residual pattern classification results. It also performs conflict resolution and priority ranking through multi-scene fusion decision-making, and outputs comprehensive health status and action suggestions.

[0010] Thirdly, the present invention provides a computer device, comprising: a memory and a processor, and a computer program stored in the memory, wherein when the computer program is executed on the processor, it implements the cat health intelligent detection method based on spatiotemporal behavioral residual analysis as described in any of the above methods.

[0011] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the intelligent cat health detection method based on spatiotemporal behavioral residual analysis as described in any of the above methods.

[0012] Compared with the prior art, the present invention has at least one of the following technical effects: 1. This invention achieves accurate monitoring and anomaly detection of cat health status through multimodal data fusion and spatiotemporal residual analysis, solving the problem of inaccurate monitoring in existing single-modal systems; 2. This invention integrates motion and visual data to generate multimodal behavioral tuples, comprehensively capturing cat behavioral characteristics and solving the problem of missing information from a single sensor; 3. This invention constructs an individualized ideal health behavior flow model, providing a dynamic benchmark for health status assessment and solving the problem of existing systems lacking ideal behavior modeling; 4. This invention constructs a spatiotemporal residual tensor through multi-dimensional residual calculation, thereby realizing the quantitative characterization of behavioral deviations and improving the sensitivity of anomaly detection; 5. This invention uses statistical process control to determine the significance of residuals and classify patterns, thereby reducing the false alarm rate and improving the accuracy of anomaly identification; 6. This invention achieves accurate identification and prioritization of postoperative recovery, pain stress, and urinary abnormalities through specific detection models and multi-scenario fusion decision-making, thereby improving its clinical reference value; 7. This invention uses a multi-stage state mechanism to construct a postoperative recovery model, realizing dynamic assessment and complication early warning during the recovery stage, and solving the problem of the lack of stage-based postoperative health assessment; 8. This invention achieves accurate determination of pain level and type through multi-dimensional pain index calculation and dynamic weight adjustment, solving the problem of lack of quantitative indicators in pain assessment; 9. This invention constructs a urinary abnormality model based on toilet behavior residuals, realizing the early identification and abnormality level determination of urinary system diseases, and solving the problem of lack of targeted modeling for urinary abnormality detection. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1This is a flowchart illustrating a cat health intelligent detection method based on spatiotemporal behavioral residual analysis according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the structure of a cat health intelligent detection system based on spatiotemporal behavioral residual analysis according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0015] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0016] In this application embodiment, the entity executing the process includes a terminal device. This terminal device includes, but is not limited to, devices capable of executing the methods disclosed in this application, such as servers, computers, smartphones, and tablets. Figure 1 A flowchart illustrating an embodiment of the intelligent cat health detection method based on spatiotemporal behavioral residual analysis disclosed in this invention is shown below: S101: Collect cat motion data and visual spatial data, interpolate and align the cat motion data and visual spatial data in a unified time coordinate system and fuse features to generate multimodal behavior tuples; S102: Collect multimodal behavioral tuples during the cat's healthy period to construct a health behavior database. By performing behavioral flow decomposition and spatiotemporal pattern mining on the health behavior database, a healthy ideal behavior flow model is generated. S103 compares the multimodal behavior tuple with the ideal health behavior flow model, calculates the residuals of exercise intensity, region deviation, behavior sequence, behavior duration, posture, and toilet behavior, constructs a spatiotemporal residual tensor, and uses statistical process control methods to determine the significance of the residuals, and outputs the residual pattern classification results. S104, based on the residual pattern classification results, constructs specific detection models for postoperative recovery, pain stress and urinary abnormalities respectively, and performs conflict resolution and priority ranking through multi-scenario fusion decision-making, outputting comprehensive health status and action suggestions.

[0017] In this embodiment, a six-axis inertial measurement unit (accelerometer + gyroscope) deployed within the cat's collar collects raw data of triaxial acceleration and triaxial angular velocity at a sampling frequency of 50Hz. A Kalman filter algorithm is used for attitude calculation, extracting the following core motion features: Instantaneous velocity: The cat's velocity at each moment is calculated by integrating the acceleration and removing the drift. Step count: A peak detection algorithm is used to identify gait cycles and accumulate steps; Attitude angles: Extract pitch angle, roll angle, and yaw angle to determine the cat's posture, such as lying down, standing, or crouching. Exercise intensity: Calculate the root mean square of the resultant acceleration to quantify the intensity of the exercise; Postural stability: Analyzes the frequency and amplitude of fluctuations in posture angles to determine the presence of pain-related features such as trembling and stiffness.

[0018] Full-body video streams of cats were captured at 30fps using 2-3 RGB-D depth cameras deployed in a home environment. A lightweight object detection algorithm (YOLOv5s) was employed to detect and track the cats, extracting the following core spatial features: 3D motion trajectory: Real-time positioning of the cat in the home spatial coordinate system to generate a continuous motion trajectory; Activity area heat map: Divide the home space into grids (e.g., 0.5m × 0.5m), count the time the cat stays in each grid, and generate an activity area distribution heat map; Behavioral posture recognition: Spatiotemporal graph convolutional network (ST-GCN) is used to recognize fine-grained behaviors, including walking, running, jumping, grooming, eating, drinking, toileting (entering the litter box), lingering before excretion, burying after excretion, lying still, curling up, etc. Path curvature: Calculates the degree of curvature of the motion trajectory to determine whether there are gait abnormalities (such as limping causing path distortion). Duration of behavior: Record the start and end times and duration of each behavior.

[0019] All sensor data are unified to the same time coordinate system, and an interpolation alignment method is used to achieve precise alignment between IMU data (50Hz) and visual data (30Hz). Based on this, the motion features extracted by IMU and the spatial features extracted by vision are fused to form a multimodal behavior tuple.

[0020] Construct a cat health record database to store static data such as breed, age, weight, sex, spaying / neutering status, medical history, and surgical records (e.g., dates of spaying / neutering surgery and orthopedic surgery). This data will be used for individualized parameter adjustments during subsequent modeling of ideal health behaviors.

[0021] The core idea of ​​the healthy ideal behavior flow modeling algorithm is to construct "the ideal behavior trajectory that the cat should have at the current time, in the current recovery stage, and in the current situation" based on the cat's historical behavior data in a healthy state and combined with individual profile parameters.

[0022] Specifically, a health baseline is first established. During the first N days (e.g., 14 days) of each cat's initial use of the system, and within the period confirmed by a veterinarian to be in good health, all multimodal behavioral data are collected to construct a "health behavior database" specific to that cat. Then, feature decomposition is performed on each behavioral tuple in the health behavior database to obtain motion flow (e.g., velocity sequence, step count sequence), posture flow (e.g., posture angle sequence), trajectory flow (e.g., position sequence), and behavioral temporal flow (e.g., behavior category sequence, behavior duration sequence).

[0023] Temporal pattern mining algorithms (such as PrefixSpan) are used to extract periodic behavioral patterns from health behavior streams, including: Daily activity rhythms: typical activity levels at different times (early morning, morning, afternoon, evening, and late night); Toilet behavior patterns: average lingering time before using the toilet, average time during using the toilet, burying time after using the toilet, and daily toilet frequency distribution; Eating and drinking patterns: eating time, duration of each eating session, and estimated water intake (based on the duration and frequency of drinking behavior); Activity area preferences: preferred rest areas, activity areas, and secluded areas at different times; Behavioral transition patterns: the probability matrix of transitions between behaviors, such as typical sequences like "rest → grooming → rest" and "eating → grooming → rest".

[0024] Based on the above mining pattern, combined with contextual factors such as current time, number of days post-surgery (if applicable), and ambient temperature, an ideal behavioral flow for the current moment is generated.

[0025] The system compares real-time collected multimodal behavioral tuples with a healthy ideal behavior flow model, calculating residuals for motion intensity, region deviation, behavior sequence, behavior duration, posture, and toileting behavior. These residual indicators are constructed as a three-dimensional spatiotemporal residual tensor (time × behavior type × residual dimension), and statistical process control (SPC) methods (such as control chart analysis) are used to determine the significance of the residuals. For example, if the duration of a cat's licking after using the toilet consistently exceeds the upper control limit of the healthy model, the residual is considered significantly abnormal. The system outputs residual pattern classification results, labeling the abnormal behavior type and the time window in which it occurred.

[0026] Based on the residual pattern classification results, specific detection models were constructed for three clinical scenarios: Postoperative recovery detection model: If the residual of movement intensity and the residual of regional deviation decrease simultaneously, and the residual of posture shows an increase in the frequency of curling up, it is judged as normal postoperative recovery; if the residual of movement intensity is consistently lower than 50% of the healthy model and is accompanied by abnormal residual of licking duration, it indicates poor pain control.

[0027] Pain stress detection model: If the postural residuals show a significant increase in the frequency of hunching and the behavioral sequence residuals show a shortened activity path, then it is determined to be acute pain stress.

[0028] Urinary tract abnormality detection model: If the toilet behavior residual shows an increase in the duration of lingering before urination and a decrease in the duration of licking after urination, and the behavior duration residual shows a prolongation of the time spent urinating in a single session, it indicates a risk of urinary system disease.

[0029] Through a multi-scenario fusion decision-making module, conflict detection results are prioritized (e.g., urinary abnormalities have higher priority than postoperative recovery), and weighted factors such as residual significance and historical health records are considered to output a final health status assessment and action recommendations. For example, when the system detects both postoperative recovery residuals and urinary abnormality residuals, if the urinary-related residuals are more significant, urinary problems are prioritized and the postoperative care plan is adjusted accordingly.

[0030] In some embodiments, step S101 above, which involves collecting cat motion data and visual spatial data, interpolating and aligning the cat motion data and visual spatial data in a unified time coordinate system, and fusing features to generate multimodal behavior tuples, specifically includes: Raw data of triaxial acceleration and triaxial angular velocity were collected by a six-axis inertial measurement unit deployed in the cat's collar. After attitude calculation by Kalman filtering algorithm, instantaneous velocity, step count, attitude angle, motion intensity and attitude stability were extracted as core motion features. Visual spatial data is collected by depth cameras deployed in the home environment. After target detection and tracking, the three-dimensional motion trajectory, activity area heat map, behavior posture recognition, path curvature and behavior duration are extracted as core spatial features. The core motion features and core spatial features are unified to the same time coordinate system and interpolated, aligned, and fused at the feature level to generate multimodal behavior tuples containing velocity, steps, attitude angle, 3D coordinates, behavior category, and behavior duration.

[0031] In this embodiment, a six-axis inertial measurement unit (IMU) is integrated into the cat's collar. This unit consists of a three-axis accelerometer and a three-axis gyroscope, used to acquire raw data during the cat's movement in real time. During acquisition, the accelerometer records the cat's acceleration changes in three-dimensional space, and the gyroscope records the cat's angular velocity changes. The raw data is used to calculate the cat's attitude using a Kalman filter algorithm. This algorithm eliminates sensor noise and compensates for dynamic errors by fusing acceleration and angular velocity information, thereby accurately calculating the cat's instantaneous attitude. Based on the calculated attitude data, the following core motion features are further extracted: Instantaneous velocity: The cat's real-time movement velocity is calculated by integrating acceleration and combining it with attitude compensation; Step count: Identify the cat's gait cycle based on the periodic fluctuation characteristics of acceleration and count the number of steps per unit time; Posture angles: including pitch angle, roll angle and yaw angle, reflecting the spatial posture of a cat's body; Exercise intensity: The intensity of a cat's activity is calculated by combining acceleration amplitude and frequency; Postural stability: The stability of a cat's body posture is assessed based on the rate of change of angular velocity.

[0032] A depth camera with 3D spatial perception capability is deployed in a home environment to collect the cat's visual spatial data in real time. During the data collection process, a target detection algorithm is first used to identify the cat's position in the image, and a target tracking algorithm is used to maintain the continuity of the cat's identity. Based on the detection and tracking results, the following core spatial features are extracted: 3D motion trajectory: Calculate the cat's movement path in 3D space by changing its position between consecutive frames; Activity area heat map: Statistical analysis of the time cats spend in different spatial areas per unit of time generates a heat map reflecting their activity preferences; Behavior and posture recognition: Using deep learning models to classify cat postures (such as walking, sitting, lying down, etc.); Path curvature: Calculates the degree of curvature of the movement trajectory, reflecting whether the cat's movement is linear or circuitous; Behavior duration: Records the time it takes for a cat to complete a certain behavior (such as littering or eating).

[0033] The above features supplement information about cats’ activity from the perspectives of spatial distribution and behavioral patterns.

[0034] Since the acquisition frequency and timestamps of motion data and visual spatial data may differ, it is necessary to unify the two types of data into the same time coordinate system. Using the time axis of the visual spatial data as a reference, interpolation is performed on the motion data. For each visual frame corresponding to a given time point, if valid motion data samples exist, they are used directly; if there are time gaps, the motion feature values ​​at that moment are estimated through linear interpolation. The aligned motion features and spatial features are then combined in a time series to generate a multimodal behavior tuple.

[0035] In some embodiments, step S102 above, which involves collecting multimodal behavioral tuples during the cat's healthy period to construct a health behavior database, and generating an ideal healthy behavior flow model by performing behavioral flow decomposition and spatiotemporal pattern mining on the health behavior database, specifically includes: Within the first preset number of days after the target cat is confirmed to be in a healthy state, multimodal behavioral tuples are continuously collected to build a health behavior database exclusive to the target cat; Each multimodal behavior tuple in the health behavior database is decomposed into motion flow, posture flow, trajectory flow, and behavior temporal flow, and velocity sequence, step count sequence, posture angle sequence, position sequence, behavior category sequence, and behavior duration sequence are extracted respectively. Based on velocity sequences, step sequences, attitude angle sequences, position sequences, behavior category sequences, and behavior duration sequences, a temporal pattern mining algorithm is used to extract daily activity rhythms, toileting behavior patterns, eating and drinking patterns, activity area preferences, and behavior transition patterns, forming an individualized spatiotemporal behavior pattern library. Based on the spatiotemporal behavior pattern library, and combined with at least one situational factor among the current time, postoperative days, and ambient temperature, a healthy ideal behavior flow model is generated that includes ideal exercise intensity, ideal activity area distribution, ideal behavior sequence, and ideal behavior duration at the current moment.

[0036] In this embodiment, within the first preset number of days (e.g., 14 consecutive days) after the target cat is diagnosed as healthy by a veterinarian, the daily behavior data of the cat is continuously recorded through a deployed multimodal data acquisition system. During the collection process, the system simultaneously acquires the cat's motion data (such as instantaneous speed, steps, and posture angles) and visual spatial data (such as 3D trajectory, activity area, and behavior category). Following the aforementioned multimodal behavior tuple generation method, the two types of data are aligned on the time axis and fused into a high-dimensional tuple containing features such as speed, steps, posture angles, location, behavior category, and behavior duration. After each day's collection is completed, all multimodal behavior tuples for that day are stored in chronological order to form a raw health behavior database specific to the target cat. This database needs to cover the cat's entire 24-hour activity cycle, including daytime active periods and nighttime resting periods, to ensure the completeness and representativeness of the behavior data.

[0037] Each multimodal behavior tuple in the health behavior database is structurally decomposed into the following four categories of behavior flows: Motion flow: Extracting velocity sequence (recording the cat's movement speed at each moment) and step sequence (counting the number of steps per unit time). Attitude Flow: Extracts the attitude angle sequence (records the dynamic changes of the cat's pitch, roll, and yaw angles); Trajectory Flow: Extracting location sequences (recording the coordinate changes of cats in three-dimensional space) and activity area heatmaps (statistically analyzing the distribution of cats' dwell time in each area); Behavior time sequence: Extract the behavior category sequence (label the cat's behavior type in chronological order, such as walking, sitting, and toileting) and the behavior duration sequence (record the start and end times of each behavior).

[0038] Based on the decomposed time series, time series pattern mining algorithms (such as frequent pattern mining based on sliding windows or deep learning time series models) are used to extract individualized behavioral patterns from the following dimensions: Daily activity rhythm: Analyze the peak and trough periods of a cat's activity over a 24-hour period to identify its inherent biological clock patterns (such as being active in the early morning and resting in the afternoon). Toilet behavior patterns: statistically analyze the distribution range of parameters such as toilet frequency, time spent loitering before using the toilet, and time spent licking after using the toilet, and establish baseline characteristics of toilet behavior; Feeding and drinking patterns: Record data such as the number of times a cat eats each day, the duration of each meal, and the frequency of drinking to identify the cat's eating preferences and patterns; Activity area preference: Based on the activity area heat map, divide the cat's high-frequency activity areas (such as windowsills and cat beds) and low-frequency activity areas to quantify its space utilization habits; Behavioral transition patterns: Analyze the triggering conditions and transition probabilities of a cat switching from one behavior (such as walking) to another (such as lying down) to establish a behavioral state transition map.

[0039] When generating a healthy ideal behavior flow model, the system needs to dynamically adjust the spatiotemporal behavior pattern library in conjunction with current contextual factors. Specific contextual factors include: Current time: Match the behavior pattern to the corresponding time period according to the cat's biological clock (e.g., reduce the intensity of activity at night). Postoperative days: If the cat is in the postoperative recovery period, historical data of similar surgeries will be retrieved from the pattern library to adjust the ideal range of activity area and exercise intensity (e.g., restrict jumping behavior within 3 days after surgery). Ambient temperature: When the ambient temperature is too high or too low, refer to historical data on the cat's behavior changes at similar temperatures (such as reduced activity in high temperatures and increased curling time in low temperatures) to adjust the ideal behavior parameters.

[0040] By weighted fusion of the aforementioned contextual factors, the most matching sub-pattern set is selected from the spatiotemporal behavior pattern library to generate a health ideal behavior flow model for the current moment. This model is output in the form of a time series, including ideal exercise intensity (such as the upper limit of speed), ideal activity area distribution (such as the range of high-frequency activity areas), ideal behavior sequence (such as the typical sequence of "eating → toileting → resting"), and ideal behavior duration (such as the median duration of a single squatting session), providing a dynamic benchmark for subsequent residual analysis.

[0041] In some embodiments, step S103 above, which involves comparing the multimodal behavior tuple with the ideal health behavior flow model, calculating the motion intensity residual, region deviation residual, behavior sequence residual, behavior duration residual, posture residual, and toilet behavior residual, and constructing a spatiotemporal residual tensor, specifically includes: The real-time collected multimodal behavior tuples are compared point by point with the ideal health behavior flow model on the same time axis. The residuals of the movement intensity between the actual movement intensity and the ideal movement intensity, the residuals of the region deviation between the actual activity area and the ideal activity area, the residuals of the behavior sequence between the actual behavior sequence and the ideal behavior sequence, the residuals of the behavior duration between the actual behavior duration and the ideal behavior duration, the residuals of the posture between the actual posture angle and the ideal posture angle, and the residuals of the toilet behavior between the actual toilet behavior structure and the ideal toilet behavior structure are calculated respectively. The residuals of all dimensions are organized according to the time and space dimensions to form a spatiotemporal residual tensor.

[0042] In this embodiment, the real-time collected multimodal behavior tuples and the ideal health behavior flow model are placed on the same time axis. For each time point, the cat's actual movement intensity data at that moment is extracted from the multimodal behavior tuples, while the corresponding ideal movement intensity data at the same moment is obtained from the ideal health behavior flow model. By comparing the actual movement intensity and the ideal movement intensity, the difference between them is calculated; this difference is the movement intensity residual.

[0043] Similarly, on the same timeline, the cat's actual activity area at that moment is determined from the multimodal behavior tuple. This can be achieved by combining the cat's positional information from visual spatial data with pre-defined region divisions. Simultaneously, the cat's ideal activity area at that moment is obtained from the healthy ideal behavior flow model. By comparing the actual activity area with the ideal activity area, the region deviation residual is calculated based on the degree of positional difference between the two areas.

[0044] On the timeline, observe the actual behavioral sequences of the cat in the multimodal behavior tuples, such as walking, jumping, and resting. Simultaneously, examine the ideal behavioral sequences of the cat within the same time period in the healthy ideal behavior flow model. Compare the actual and ideal behavioral sequences to identify the differences, and calculate the behavioral sequence residuals based on these differences.

[0045] For each specific behavior, the actual duration of the behavior is obtained from the multimodal behavior tuple, and the ideal duration of the same behavior is obtained from the healthy ideal behavior flow model. The difference between the actual behavior duration and the ideal behavior duration is the behavior duration residual.

[0046] At each point in time along the timeline, the cat's actual posture angle data is extracted from the multimodal behavior tuple. This can be obtained through sensor information from motion data or visual analysis to determine the angles of different parts of the cat's body. Simultaneously, the cat's ideal posture angle data at the same moment is obtained from the healthy ideal behavior flow model. By comparing the actual posture angles with the ideal posture angles, the angle difference between them is calculated; this difference is the posture residual.

[0047] For cat toileting behavior, actual toileting behavior structure information is extracted from multimodal behavior tuples, including the duration of specific behavioral stages such as pre-toilet probing time, toileting time, and post-toilet licking time, as well as the sequence of behaviors. Simultaneously, ideal toileting behavior structure information is obtained from a healthy ideal behavior flow model. By comparing the actual and ideal toileting behavior structures, and considering differences in the duration of each stage and the sequence of behaviors, toileting behavior residuals are calculated comprehensively.

[0048] After calculating the various residuals mentioned above, the residuals of all dimensions are organized according to the time and spatial dimensions. Using the time dimension as the horizontal axis, the various residuals calculated at each time point (motor intensity residual, region deviation residual, behavior sequence residual, behavior duration residual, posture residual, and toilet behavior residual) are arranged in a specific order. Simultaneously, considering the spatial dimension, if some residuals are related to the cat's specific spatial location (such as region deviation residual), they are integrated with spatial location information during organization. In this way, all residual data are integrated into a data structure with spatiotemporal characteristics, namely the spatiotemporal residual tensor. This spatiotemporal residual tensor can comprehensively reflect the differences between the cat's actual behavior and ideal healthy behavior in time and space. This embodiment can effectively compare multimodal behavior tuples with the ideal healthy behavior flow model, calculate various residuals, and construct the spatiotemporal residual tensor, providing a crucial data processing step for intelligent cat health detection.

[0049] In some embodiments, step S103 above, which involves using statistical process control methods to determine the significance of the residuals and outputting the residual pattern classification results, specifically includes: Based on historical residual data collected during the healthy period, the mean and standard deviation of residuals in each dimension are calculated, and individualized residual control limits are set. Each residual value in the spatiotemporal residual tensor is compared with the residual control limit to determine whether the current residual has reached a significant residual level. If the current residual reaches a significant residual level, extract the spatial distribution features, temporal distribution features, and behavioral correlation features of the residual from the spatiotemporal residual tensor. The spatial distribution features include the spatial variance of the residual and the location and area of ​​the residual cluster. The temporal distribution features include the ratio of the nighttime residual to the daytime residual. The behavioral correlation features include the degree of increase in the residual when a specific behavior occurs. Spatial distribution features, temporal distribution features, and behavioral correlation features are input into the classifier for classification and recognition, and the classification result is output as at least one residual pattern among diffuse, focal, temporal, behavioral correlation, and mixed types.

[0050] In this embodiment, when the cat is in a healthy period, we have collected multimodal behavioral tuples and constructed a healthy behavior database. Based on this, we calculated the spatiotemporal residual tensor and obtained historical residual data. For each dimension of the historical residual data, such as exercise intensity residual and region deviation residual, we calculated its mean and standard deviation. The mean is calculated by adding all historical residual values ​​for that dimension and dividing by the number of residual values. The standard deviation measures the dispersion of these residual values ​​around the mean. Based on the calculated mean and standard deviation, we set individualized residual control limits for the residuals of each dimension. Generally, the range of control limits can be set by adding or subtracting a certain multiple of the standard deviation from the mean. For example, using the mean plus or minus three times the standard deviation can cover most of the residual fluctuation range under normal circumstances. Exceeding this range may indicate an abnormal situation.

[0051] Each residual value in the spatiotemporal residual tensor is compared one by one with the previously set residual control limits for the corresponding dimension. For each residual value, it is checked whether it exceeds the range of the control limits. If a residual value is greater than the upper control limit or less than the lower control limit, then the current residual is determined to have reached a significant residual level. This indicates that the cat's behavior in that dimension differs significantly from the ideal healthy behavior, possibly suggesting a problem with the cat's health. For example, if the control limit range for the regional deviation residual is -2 to 2, and the currently calculated regional deviation residual value is 5, then the regional deviation residual can be determined to have reached a significant residual level.

[0052] Once the current residual is determined to have reached a significant residual level, it is necessary to extract the relevant features of the residual from the spatiotemporal residual tensor in order to further analyze the residual pattern, as shown in Table 1 below; Table 1. Residual pattern classification results:

[0053] The extracted spatial distribution features, temporal distribution features, and behavioral association features are input into a pre-trained classifier. The classifier can employ machine learning algorithms such as decision trees and support vector machines. During training, a large amount of sample data with known residual pattern types is used to train the classifier, enabling it to learn the correspondence between different feature combinations and residual patterns. When new feature data is input, the classifier classifies and identifies the residual pattern based on its learned knowledge, outputting a classification result for at least one residual pattern among diffuse, focal, temporal, behavioral association, and mixed patterns. The scene detection logic will differ for different residual patterns, as shown in Table 2 below. Table 2. Correlation Logic between Residual Patterns and Scene Detection:

[0054] This embodiment can effectively use statistical process control methods to determine the significance of residuals and output accurate residual pattern classification results, providing an important basis for subsequent specific detection and comprehensive health status assessment of cats for different health problems.

[0055] In some embodiments, in step S104 above, the specific detection models constructed based on the residual pattern classification results for postoperative recovery, pain stress, and urinary abnormalities, and the conflict resolution and priority ranking performed through multi-scenario fusion decision-making to output comprehensive health status and action suggestions, specifically includes: Based on the residual pattern classification results, a postoperative recovery specific detection model is constructed using a multi-stage state mechanism; Based on the residual pattern classification results, a pain stress-specific detection model was constructed by calculating the pain index. Based on the residual pattern classification results, a urinary abnormality-specific detection model is constructed by calculating the urinary abnormality index. Based on the output results of the postoperative recovery specific detection model, the pain stress specific detection model and the urinary abnormality specific detection model, the conflict resolution matrix is ​​used to identify and process the information overlap, causal relationship, independent events and primary and secondary relationships between multiple scenarios, and obtain the conflict identification results of multiple scenarios. The results of conflict identification in multiple scenarios are classified according to the priority ranking rules based on urgency, and a decision tree is used to hierarchically determine the overall health status and action recommendations.

[0056] Furthermore, the postoperative recovery-specific detection model constructed using a multi-stage state mechanism based on the residual pattern classification results specifically includes: Based on the type of surgery and the number of days after surgery, the postoperative recovery process is divided into the acute phase, the early recovery phase, the functional recovery phase, and the rehabilitation consolidation phase. Differentiated ideal activity baselines and pain control thresholds are set for each recovery phase, forming a multi-stage state machine. Based on a multi-stage state machine, the residual pattern classification results are used as input signals and integrated into the pain control assessment process. Different weight coefficients are assigned to different residual patterns in the residual pattern classification results. The residual patterns are combined with the posture residual, behavior sequence residual and regional deviation residual for weighted fusion, and the pain control status of each recovery stage is output. Based on the pain control status at each recovery stage, the real-time detected residual change trajectory is matched with a preset complication feature pattern library. When the matching degree reaches a preset matching degree threshold, an early warning output for the corresponding complication type is triggered. The complication feature pattern library includes residual combination features corresponding to wound infection mode, intestinal obstruction mode and pain loss of control mode.

[0057] In this embodiment, the entire postoperative recovery process is meticulously divided into four stages based on the type of surgery performed on the cat and the number of days post-surgery: the acute phase, the early recovery phase, the functional recovery phase, and the rehabilitation consolidation phase, as shown in Table 3 below. For each recovery stage, differentiated ideal activity baselines and pain control thresholds are set. The ideal activity baseline is derived from statistical data on the activity levels of a large number of healthy cats during the corresponding postoperative recovery stage; it reflects the activity level a cat should achieve during normal recovery at that stage. The pain control threshold is used to determine whether the cat is in a state of good pain control at that stage. If the cat's activity level, posture, and other behaviors exceed this threshold range, it may indicate poor pain control. By setting these parameters for each stage, a complete multi-stage state machine is formed, providing a framework for subsequent evaluation and detection.

[0058] Table 3. Division of Recovery Phases

[0059] Based on the constructed multi-stage state machine, the residual pattern classification results are incorporated as input signals into the pain control assessment process. Since different residual patterns reflect pain control status to varying degrees, corresponding weight coefficients are assigned to different residual patterns according to their classification results, as shown in Table 4 below. For example, focal residual patterns may reflect localized pain more significantly and are given a higher weight; while diffuse residual patterns may reflect abnormalities in the overall physical condition and are given a relatively lower weight. Table 4 Pain Control Assessment:

[0060] The residuals of posture, behavior sequence, and regional deviation are weighted and fused together. Posture residuals reflect whether the cat's posture is normal, such as whether there are abnormal postures like curling up or stiffness; behavior sequence residuals reflect the coherence and rationality of the cat's behavior, such as whether there are frequent interruptions or abnormal behavioral sequences; regional deviation residuals indicate the degree of deviation between the cat's activity area and the ideal area. These residuals are weighted according to their respective weight coefficients to obtain a comprehensive assessment value. Based on this assessment value, the pain control status at each recovery stage is output. For example, if the comprehensive assessment value is high, it indicates that the cat may be in a state of poor pain control and needs further attention and examination; if the assessment value is low, it indicates that the cat's pain control is good and the recovery process is normal.

[0061] Based on the pain control status at each recovery stage, the real-time detected residual change trajectory is continuously monitored. The residual change trajectory is then matched with a pre-defined complication feature pattern library. The complication feature pattern library is derived from the collection and analysis of residual data from a large number of cats with different complications. It includes residual combination features corresponding to wound infection patterns, intestinal obstruction patterns, and pain loss of control patterns, as shown in Table 5 below. Table 5. Complication Warning:

[0062] When the real-time detected residual change trajectory matches a certain pattern in the complication feature pattern library to a preset matching degree threshold, an early warning output for the corresponding complication type is triggered. The preset matching degree threshold is set based on the accuracy and reliability of the actual detection. When the matching degree exceeds this threshold, it indicates that the cat is likely to have the corresponding complication, and the pet owner and veterinarian should be notified in time for further examination and treatment.

[0063] This embodiment can effectively construct a postoperative recovery-specific detection model based on residual pattern classification results using a multi-stage state machine, enabling the assessment of pain control status and early warning of complications during the postoperative recovery process in cats, providing a scientific and effective method for postoperative health monitoring in cats.

[0064] Furthermore, the construction of a pain stress-specific detection model based on the residual pattern classification results and by calculating the pain index specifically includes: Pain index analysis was performed based on postural residuals, behavioral sequence residuals, movement intensity residuals, and movement duration residuals to obtain multiple pain features, including stiffness index, tremor index, behavioral fragmentation index, behavioral persistence index, behavioral avoidance index, recovery delay index, rest dependence index, and self-care ability index. A comprehensive pain feature vector is constructed based on multiple pain features. The key features are used for initial screening to determine severe pain and neuropathic pain, the combined features are used for moderate pain, and the edge features are used for mild discomfort. The weights of the comprehensive pain feature vector are dynamically adjusted based on the residual pattern classification results, and the pain level and pain type are output.

[0065] In this embodiment, the relationship between each residual and pain characteristics is shown in Table 6 below; Table 6. Pain Feature Extraction:

[0066] After obtaining multiple pain features, a comprehensive pain feature vector is constructed based on these features. The comprehensive pain feature vector is a dataset containing multiple dimensions, with each dimension corresponding to a pain feature, which can comprehensively reflect the cat's pain state characteristics.

[0067] The weights of the comprehensive pain feature vector are dynamically adjusted based on the residual pattern classification results. The residual pattern classification results reflect the difference pattern between the cat's current behavioral state and its normal state. Different residual patterns reflect different degrees of pain, as shown in Table 7 below. Table 7 Pain Index Calculation:

[0068] In the example in Table 7, the pain index = 0.60 × pain feature vector + 0.40 × residual pattern.

[0069] If the residual pattern classification results show a large change in the cat's posture residuals, it indicates that posture is a more significant indicator of pain. In this case, the weights of the stiffness index and trembling index in the pain feature vector can be appropriately increased to make the determination of pain level and type more accurate. In this way, the accurate pain level and pain type are finally output (as shown in Tables 8, 9 and 10 below), providing an important basis for the cat's health monitoring and subsequent treatment. Table 8 Pain Scale:

[0070] Table 9 Pain Type Identification:

[0071] Based on the comprehensive pain feature vector, the pain level is determined as shown in Table 10 below, which is divided into 4 levels of pain level. Table 10 Pain Rating Matrix:

[0072] Layer 1: Key Feature Determination: F1>0.8 and F7>0.8 → Severe pain; F2>0.7→Neurotic pain, proceed to level 2; Other → Enter Level 2: Layer 2: Determination of combined features; (F5+F6)>1.4 or (F3+F7)>1.2 → Moderate pain; Other → Enter Level 3: Layer 3: Edge feature determination; (F1+F8)>1.0→Mild discomfort; Other → Painless; Layer 4: Residual Mode Correction: Temporal or behaviorally related pain levels → Pain severity increased by one level; No clear pattern and output "mild" → downgraded to painless; This embodiment can effectively construct a pain stress-specific detection model based on residual pattern classification results and by calculating the pain index, thereby accurately determining the pain level and type of cats and providing strong support for cat health management.

[0073] Furthermore, the pain index analysis based on posture residuals, behavioral sequence residuals, movement intensity residuals, and movement duration residuals yields multiple pain characteristics, specifically including: Time-series statistics were performed on the attitude residuals to extract the stiffness index, which characterizes attitude stiffness and attitude fluctuation anomaly, and the jitter index, which characterizes jitter frequency. Behavioral transfer entropy analysis and behavioral segment length analysis were performed on the behavioral sequence residuals to extract the behavioral fragmentation index, which characterizes the degree of frequent interruption of behavior, and the behavioral persistence index, which characterizes the ability to maintain behavior. We conducted high-intensity behavior proportion change analysis and activity recovery slope analysis on the residuals of exercise intensity to extract the behavior avoidance index to characterize pain-related behavior avoidance and the recovery delay index to characterize the degree of activity recovery sluggishness. We conducted a change analysis on the proportion of resting time and the change analysis on the duration of self-care behavior in the behavioral duration residuals, and extracted the resting dependence index to characterize the degree of pain-induced rest and the self-care ability index to characterize the impact of pain on daily living ability.

[0074] In this embodiment, the stiffness index is extracted by analyzing the length of time the cat's body posture remains relatively fixed and the amplitude of posture changes within the time series. When a cat is in pain, its body often exhibits a relatively stiff posture, such as a fixed degree of curling up and minimal changes in limb extension angles. By statistically analyzing the percentage of time the cat's body posture remains relatively unchanged within a certain period and the range of posture changes, combined with a pre-set standard threshold, when the percentage of time remains unchanged exceeds a certain proportion and the amplitude of posture changes is less than a specific value, it can be determined that the cat's body stiffness has increased, and thus the stiffness index, used to characterize posture stiffness, can be extracted.

[0075] The extraction of the tremor index is also based on the time-series statistics of posture residuals. During the recording of the cat's posture data, close attention is paid to minute tremors, either locally or overall. When a cat trembles involuntarily due to pain, these characteristics will be reflected in the posture residuals. By setting an appropriate frequency monitoring range, the number of times the cat's body trembles per unit time is counted, and the amplitude of the tremors is analyzed. When the number of tremors exceeds the normal range and the amplitude reaches a certain level, a tremor index, representing the frequency of tremors, is extracted from this data.

[0076] Behavioral transition entropy analysis and behavioral fragment length analysis were performed on the behavioral sequence residuals. Behavioral transition entropy analysis was used to study the transition patterns and uncertainties between cat behaviors. Under normal circumstances, cat behaviors exhibit a certain degree of coherence and regularity; for example, the transition from play to rest behavior follows a certain transition and logic. However, when a cat is in pain, its behavioral sequence becomes chaotic, and the transitions between behaviors become frequent and irregular. By calculating the behavioral transition entropy, an increase in entropy value indicates an increase in the uncertainty of cat behavior transitions, i.e., a greater degree of frequent behavioral interruptions. This allows for the extraction of a behavioral fragmentation index to characterize the degree of frequent behavioral interruptions.

[0077] Behavioral segment length analysis focuses on the duration of individual cat behaviors. Under normal circumstances, certain cat behaviors, such as eating and playing, have relatively stable durations. When a cat is in pain, the duration of these behaviors shortens, making it difficult to maintain them for extended periods. By statistically analyzing the length of each cat's behavioral segment and comparing it to the length of behavioral segments under normal conditions, a behavioral persistence index, which characterizes the ability to maintain behavior, is extracted when the length of a behavioral segment is significantly shortened.

[0078] We conducted high-intensity behavior percentage change analysis and activity recovery slope analysis on the residuals of exercise intensity. The high-intensity behavior percentage change analysis calculated the proportion of time a cat spent engaging in high-intensity exercise (such as running and jumping) within a certain period of time relative to its total activity time. When a cat is healthy, the high-intensity behavior percentage remains relatively stable. When a cat is in pain, it actively avoids high-intensity behaviors, leading to a decrease in the high-intensity behavior percentage. By monitoring changes in the high-intensity behavior percentage, when the percentage significantly decreases, we extracted a behavior avoidance index to characterize pain-related behavior avoidance.

[0079] Activity recovery slope analysis studies the trend of a cat's activity level returning to normal after a period of activity. Under normal circumstances, cats can recover to their normal activity level relatively quickly after activity, and their activity recovery curve has a certain slope. When a cat is in pain, its recovery ability decreases, the recovery speed slows down, and the slope of the recovery curve becomes smaller. By calculating the activity recovery slope and comparing it with normal conditions, a recovery delay index is extracted to characterize the degree of slowness in activity recovery when the slope becomes significantly smaller.

[0080] We conducted behavioral duration residual analysis on changes in the proportion of resting time and changes in self-care behavior time. The analysis of changes in the proportion of resting time involved calculating the percentage of time a cat spent resting within a given period. When a cat is healthy, the proportion of resting time remains relatively stable. When a cat is in pain, due to discomfort, it is more inclined to rest, leading to an increase in the proportion of resting time. By monitoring changes in the proportion of resting time, when the proportion significantly increases, a resting dependence index, used to characterize the degree of pain-induced rest, was extracted.

[0081] Self-care behavior duration variation analysis focuses on the time cats spend performing self-care behaviors (such as eating, drinking, and grooming). Under normal circumstances, cats can successfully complete these self-care behaviors, and the time required is relatively stable. When a cat is in pain, physical pain may affect the performance of self-care behaviors, leading to an increase or decrease in the duration of these behaviors (e.g., difficulty eating due to pain leading to prolonged eating time, or reduced grooming time due to discomfort). By statistically analyzing the changes in the duration of cats' self-care behaviors and comparing them with normal conditions, when the changes exceed a certain range, a self-care ability index is extracted to characterize the impact of pain on daily living abilities.

[0082] Furthermore, the construction of a urinary abnormality-specific detection model based on the residual pattern classification results and by calculating the urinary abnormality index specifically includes: Based on toilet behavior residuals, behavior sequence residuals, and regional deviation residuals, a temporal structure analysis is performed on each toilet event to extract the pre-toilet wandering time, the time during toileting, and the post-toilet burying time. These are then compared with the corresponding health baseline mean of the ideal health behavior flow model to output a urinary abnormality feature vector. The urinary abnormality feature vector includes the degree of wandering abnormality, the degree of urination abnormality, the degree of burying abnormality, the degree of toilet frequency abnormality, the degree of licking the urinary area after toileting abnormality, and the degree of urination location abnormality. When the residual pattern classification result contains focal patterns and the spatial clustering area of ​​the focal patterns is located in the litter box area, the weight of the structural abnormality dimension in the urinary abnormality feature vector is adjusted, and a weighted fusion calculation is performed in combination with the urinary abnormality feature vector to output the urinary abnormality index. Based on the threshold range of the urinary tract abnormality index, output at least one urinary tract abnormality level among normal, mild abnormality, moderate abnormality, and severe abnormality. Combine the temporal type in the residual pattern classification result to correct the urinary tract abnormality level and output the urinary tract abnormality judgment result.

[0083] In this embodiment, the temporal structure of each toilet-use event of the cat is analyzed based on the toilet behavior residual, behavior sequence residual and region deviation residual, as shown in Table 11 below; Table 11. Extraction of features of urinary tract abnormalities:

[0084] During a cat's normal littering process, its littering behavior follows a certain temporal pattern, including stages such as pacing before littering, urination during littering, and burying after littering. By analyzing the residuals of littering behavior, the start and end points of the littering behavior on the timeline can be accurately located; combined with the residuals of the behavior sequence, the position of the littering behavior in the entire behavior sequence and its order with other behaviors can be clarified; and by using the regional deviation residuals, it can be determined whether the specific area where the littering behavior occurs corresponds to the normal area.

[0085] Based on the above analysis, the pre-toilet wandering time, toilet time, and post-toilet burying time were accurately extracted for each toilet-use event. Simultaneously, the mean health baseline for these durations was obtained from the ideal health behavior flow model. The ideal health behavior flow model is constructed by collecting multimodal behavioral tuples during the cat's healthy period, containing normal patterns and parameters of various behaviors in a healthy cat. Therefore, the mean health baseline it provides has high accuracy and representativeness.

[0086] The extracted actual durations are compared with the corresponding health baseline mean. For example, if the actual duration of lingering before using the litter box is significantly longer than the health baseline mean, it indicates that the cat may be exhibiting abnormal lingering behavior before using the litter box; if the duration of lingering during urination differs significantly from the health baseline mean, it may suggest a problem with the cat's urination process; and if the duration of burying the feline after using the litter box is abnormal, it may also reflect an abnormality in the cat's urinary system. Through this comparison method, a urinary abnormality feature vector is output, which includes multiple dimensions such as the degree of abnormal lingering, the degree of abnormal urination, the degree of abnormal burying, the degree of abnormal litter box frequency, the degree of abnormal behavior of licking the urinary area after using the litter box, and the degree of abnormality in urination location. Among them, the degree of abnormal wandering reflects the degree of deviation of the cat's wandering behavior before using the litter box from the normal state; the degree of abnormal urination reflects the abnormal situation of the cat's urination process; the degree of abnormal burying indicates the degree of abnormality of the cat's burying behavior after using the litter box; the degree of abnormal littering frequency is determined by comparing the number of times the cat uses the litter box over a period of time with the normal frequency; the degree of abnormal licking of the urinary tract after using the litter box is determined by observing the difference between the frequency and duration of the cat's licking of the urinary tract after using the litter box and the normal situation; and the degree of abnormal urination location is judged by the deviation of the cat's urination location from the normal urination area such as the litter box.

[0087] When the residual pattern classification results contain focal patterns, and the spatial clustering area of ​​these focal patterns is located in the litter box area, it means that the cat's abnormal behavior is mainly concentrated in this specific area, and is highly correlated with urinary system problems. In this case, the weight of the structural anomaly dimension in the urinary abnormality feature vector is adjusted. The structural anomaly dimension mainly reflects the anomalies in the temporal structure and spatial location of the cat's toileting behavior. Since the anomalies are concentrated in the litter box area, it indicates that the importance of structural anomalies in urinary abnormalities has increased. Therefore, the weight of this dimension is appropriately increased to more accurately reflect the degree of abnormality in the cat's urinary system.

[0088] After adjusting the weights, a weighted fusion calculation is performed using the urinary tract abnormality feature vector. The degree of abnormality in each dimension is summed according to the adjusted weights to obtain a comprehensive value, namely the urinary tract abnormality index. This index comprehensively considers multiple abnormalities in a cat's toileting behavior, and can more comprehensively and accurately reflect the health status of the cat's urinary system.

[0089] As shown in Table 12 below, based on the threshold range of the calculated urinary tract abnormality index, at least one urinary tract abnormality level is output: normal, mild abnormality, moderate abnormality, and severe abnormality. Different threshold ranges are preset, with each range corresponding to a urinary tract abnormality level. For example, when the urinary tract abnormality index is in a low range, it is judged as normal; when the index exceeds a certain value but is within the moderate range, it is judged as mild abnormality; as the index further increases, it is judged as moderate abnormality and then severe abnormality, respectively. Table 12 Clinical significance of various urinary abnormalities:

[0090] As shown in Table 13 below, the urinary tract abnormality level is corrected by combining the temporal pattern in the residual pattern classification results. The temporal pattern reflects the trend of the cat's abnormal behavior over time, such as whether the abnormal behavior appears suddenly or gradually worsens. If the temporal pattern shows that the cat's urinary tract abnormal behavior appears suddenly and is relatively severe, then even if the urinary tract abnormality index is in the mild abnormality range, the judgment result may be corrected to moderate abnormality; conversely, if the abnormal behavior appears gradually and develops slowly, even if the index is in the moderate abnormality range, it may be corrected to mild abnormality. Table 13 Residual Model Correction:

[0091] This correction method can more accurately determine abnormalities in a cat's urinary system and output the final urinary abnormality diagnosis, providing a more reliable reference for pet owners and veterinarians.

[0092] Furthermore, based on the output results of the postoperative recovery-specific detection model, the pain stress-specific detection model, and the urinary abnormality-specific detection model, a conflict resolution matrix is ​​used to identify and process information overlap, causal relationships, independent events, and primary / secondary relationships among multiple scenarios to obtain multi-scenario conflict identification results, specifically including: Based on the output results of the postoperative recovery specific detection model, the pain stress specific detection model, and the urinary abnormality specific detection model, the current scenario combination type is identified. The scenario combination type includes a dual scenario combination of postoperative recovery and pain stress, a dual scenario combination of postoperative recovery and urinary abnormality, a dual scenario combination of pain stress and urinary abnormality, and a triple scenario combination of postoperative recovery, pain stress, and urinary abnormality. The scenario combination type is input into a preset conflict resolution matrix for matching. The conflict resolution matrix is ​​used to define the conflict type labels corresponding to different scenario combinations. The conflict type labels include at least one of the following: information overlap type, causal relationship type, independent event type, and primary and secondary relationship type. By associating conflict type labels with corresponding scene combination information, a multi-scene conflict recognition result is generated, which includes conflict type identifiers, a list of scenes involved in the conflict, and the original output parameters of each scene.

[0093] In one possible implementation, the method for determining the scenario combination type includes: when poor postoperative pain control and moderate to severe pain stress coexist, the scenario combination type is information overlap type; when severe urinary tract abnormalities and moderate pain stress coexist, the scenario combination type is causal relationship type; when postoperative recovery complications and moderate urinary tract abnormalities coexist, the scenario combination type is independent event type; when severe pain stress and mild urinary tract abnormalities coexist, the scenario combination type is primary and secondary relationship type.

[0094] In this embodiment, the parameters output by each scene detection algorithm are shown in Table 14 below; Table 14 Scene Detection Output Definitions:

[0095] After identifying the scenario combination type, it is input into a pre-defined conflict resolution matrix for matching. When multiple scenarios are triggered simultaneously, conflicts may exist (such as postoperative pain and chronic pain coexisting), requiring identification and handling, as shown in Table 15 below. This conflict resolution matrix is ​​pre-constructed based on extensive data analysis and real-world case studies. Its function is to define the conflict type labels corresponding to different scenario combinations. Conflict type labels mainly include at least one of the following: information overlap type, causal relationship type, independent event type, and primary-secondary relationship type. Information overlap type refers to different scenarios using some of the same information when assessing the cat's health status. For example, postoperative recovery and pain stress may both involve the cat's activity level data. Causal relationship type indicates that the occurrence of one scenario is a result of another scenario. For example, the cat experiences pain due to urinary abnormalities, which in turn affects postoperative recovery. Independent event type means that the scenarios are independent of each other, with no obvious correlation or influence. Primary-secondary relationship type refers to the existence of primary and secondary scenarios among multiple scenarios, with the primary scenario having a more critical impact on the cat's health status. By matching the scenario combination type with the conflict resolution matrix, the conflict type label corresponding to the current scenario combination can be quickly and accurately determined, thereby clarifying the relationship between different scenarios.

[0096] Table 15 Conflict Scene Recognition Matrix:

[0097] After matching the conflict resolution matrix and determining the conflict type labels, the conflict type labels are associated with the corresponding scenario combination information. The scenario combination information includes a list of specific scenarios involved in the conflict, clearly identifying which scenarios were combined to cause the conflict. It also records the original output parameters of each scenario; these parameters are the specific data obtained by the model when evaluating each scenario. For example, for the dual scenario combination of postoperative recovery and pain stress, the parameters related to the cat's postoperative recovery level output by the postoperative recovery model and the parameters related to the cat's pain level output by the pain stress model are recorded. By closely linking the conflict type labels with this scenario combination information, a multi-scenario conflict identification result is generated, containing conflict type identifiers, a list of scenarios involved in the conflict, and the original output parameters of each scenario. This result comprehensively and clearly presents the conflict situation between different scenarios, providing an accurate basis for subsequent grading based on urgency priority rules and using decision trees to hierarchically determine comprehensive health status and action recommendations.

[0098] The results of multi-scenario conflict identification are graded according to an urgency priority ranking rule. This urgency priority ranking rule is based on the degree of harm and urgency of different health problems to the cat. For example, urinary tract diseases may rapidly worsen if not treated promptly, thus their priority may be higher; while slow postoperative recovery may have a relatively smaller impact on the cat in the short term, thus its priority is relatively lower. Four levels of urgency are defined for ranking and user prompts, as shown in Table 16 below.

[0099] Table 16 Urgency Priority Ranking:

[0100] After classifying the multi-scenario conflict identification results according to the urgency priority ranking rule, a decision tree-based hierarchical determination is used to comprehensively assess the health status and action recommendations: Level 1: Emergency Event Judgment If any scenario P0 is triggered in an emergency → the overall status is "emergency medical treatment"; If a complication warning is triggered in any scenario → the overall status = "Complication Risk"; Other → Enter Level 2: Layer 2: Multi-scene overlay judgment; If ≥2 scenarios output "moderate" or above → overall status = "multi-system anomaly"; If ≥2 scenarios output "mild" or above → overall status = "general discomfort"; Other → Enter the 3rd floor; Layer 3: Single-scene determination: If postoperative pain is poorly controlled → Overall condition = "Poor postoperative pain control"; If the pain stress pain level is ≥ moderate → the overall state = "pain state" + pain type; If the urinary tract abnormality level is ≥ moderate → the overall status = "urinary tract abnormality" + structural abnormality index; Other → Overall Status = "Slight Abnormality" or "Healthy"; The decision tree is pre-constructed based on different classification levels and various possible scenario combinations. It can progressively determine the cat's overall health status based on the input classification results and scenario information, and provide corresponding action suggestions. The parameters of the final fused decision output are shown in Table 17 below.

[0101] Table 17 Output Parameter Definitions:

[0102] Through the above methods, accurate comprehensive health status and action recommendations are ultimately generated, providing valuable reference for pet owners and veterinarians.

[0103] Reference Figure 2 An embodiment of the present invention provides a cat health intelligent detection system 2 based on spatiotemporal behavioral residual analysis, wherein the system 2 specifically includes: The data fusion module 201 is used to collect cat motion data and visual spatial data, and to perform interpolation alignment and feature fusion on the cat motion data and visual spatial data in a unified time coordinate system to generate multimodal behavior tuples. Feature analysis module 202 is used to collect multimodal behavioral tuples during the cat's healthy period to build a healthy behavior library. By performing behavioral flow decomposition and spatiotemporal pattern mining on the healthy behavior library, a healthy ideal behavior flow model is generated. The pattern classification module 203 is used to compare the multimodal behavior tuples with the ideal healthy behavior flow model, calculate the residuals of exercise intensity, region deviation, behavior sequence, behavior duration, posture, and toilet behavior, construct the spatiotemporal residual tensor, and use statistical process control methods to determine the significance of the residuals and output the residual pattern classification results. The scene detection module 204 is used to construct specific detection models for postoperative recovery, pain stress and urinary abnormalities based on the residual pattern classification results, and to resolve conflicts and prioritize them through multi-scene fusion decision-making, and output comprehensive health status and action suggestions.

[0104] It is understandable that, such as Figure 1 The content of the illustrated embodiments of the cat health intelligent detection method based on spatiotemporal behavioral residual analysis is applicable to the embodiments of this cat health intelligent detection system based on spatiotemporal behavioral residual analysis. The specific functions implemented by the embodiments of this cat health intelligent detection system based on spatiotemporal behavioral residual analysis are as follows: Figure 1 The illustrated embodiment of the intelligent cat health detection method based on spatiotemporal behavioral residual analysis is the same, and the beneficial effects achieved are the same as those shown. Figure 1 The beneficial effects achieved by the illustrated embodiment of the intelligent cat health detection method based on spatiotemporal behavioral residual analysis are also the same.

[0105] It should be noted that the information interaction and execution process between the above systems are based on the same concept as the method embodiments of the present invention. For details on their specific functions and technical effects, please refer to the method embodiments section, which will not be repeated here.

[0106] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0107] Reference Figure 3 The present invention also provides a computer device 3, including: a memory 302 and a processor 301, and a computer program 303 stored in the memory 302. When the computer program 303 is executed on the processor 301, it implements the cat health intelligent detection method based on spatiotemporal behavioral residual analysis as described in any of the above methods.

[0108] The computer device 3 may be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer device 3 may include, but is not limited to, a processor 301 and a memory 302. Those skilled in the art will understand that... Figure 3 The computer device 3 is merely an example and does not constitute a limitation on the computer device 3. It may include more or fewer components than shown in the figure, or combine certain components, or different components, such as input / output devices, network access devices, etc.

[0109] The processor 301 can be a Central Processing Unit (CPU), but it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor.

[0110] In some embodiments, the memory 302 may be an internal storage unit of the computer device 3, such as a hard disk or memory of the computer device 3. In other embodiments, the memory 302 may be an external storage device of the computer device 3, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the computer device 3. Furthermore, the memory 302 may include both internal and external storage units of the computer device 3. The memory 302 is used to store the operating system, applications, boot loader, data, and other programs, such as the program code of the computer program. The memory 302 can also be used to temporarily store data that has been output or will be output.

[0111] This invention also provides a computer-readable storage medium storing a computer program thereon. When the computer program is run by a processor, it implements the intelligent cat health detection method based on spatiotemporal behavioral residual analysis as described in any of the above methods.

[0112] In this embodiment, if the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include at least: any entity or device capable of carrying computer program code to a photographing device / terminal device, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium. Examples include USB flash drives, portable hard drives, magnetic disks, or optical disks. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electrical carrier signals or telecommunication signals.

[0113] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0114] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0115] In the embodiments disclosed in this application, it should be understood that the disclosed devices / terminal equipment and methods can be implemented in other ways. For example, the device / terminal equipment embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling or direct coupling or communication connection may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

Claims

1. A method for intelligent detection of feline health based on spatiotemporal behavioral residual analysis, characterized in that, The method specifically includes: Collect cat motion data and visual spatial data, interpolate and align the cat motion data and visual spatial data in a unified time coordinate system and fuse features to generate multimodal behavior tuples; A health behavior database was constructed by collecting multimodal behavioral tuples from cats during their healthy period. By performing behavioral flow decomposition and spatiotemporal pattern mining on the health behavior database, an ideal healthy behavior flow model was generated. The multimodal behavior tuples are compared with the ideal health behavior flow model. The residuals of exercise intensity, region deviation, behavior sequence, behavior duration, posture, and toilet behavior are calculated. The spatiotemporal residual tensor is constructed, and the significance of the residuals is determined by statistical process control method. The residual pattern classification results are output. Based on the residual pattern classification results, specific detection models were constructed for postoperative recovery, pain stress, and urinary abnormalities, respectively. Conflict resolution and priority ranking were carried out through multi-scenario fusion decision-making, and comprehensive health status and action suggestions were output.

2. The method according to claim 1, characterized in that, The process involves collecting cat motion data and visual spatial data, interpolating and aligning the data in a unified time coordinate system, and fusing features to generate multimodal behavior tuples. Specifically, this includes: Raw data of triaxial acceleration and triaxial angular velocity were collected by a six-axis inertial measurement unit deployed in the cat's collar. After attitude calculation by Kalman filtering algorithm, instantaneous velocity, step count, attitude angle, motion intensity and attitude stability were extracted as core motion features. Visual spatial data is collected by depth cameras deployed in the home environment. After target detection and tracking, the three-dimensional motion trajectory, activity area heat map, behavior posture recognition, path curvature and behavior duration are extracted as core spatial features. The core motion features and core spatial features are unified to the same time coordinate system and interpolated, aligned, and fused at the feature level to generate multimodal behavior tuples containing velocity, steps, attitude angle, 3D coordinates, behavior category, and behavior duration.

3. The method according to claim 1, characterized in that, The process involves collecting multimodal behavioral tuples from cats during their healthy period to construct a health behavior database. This database is then used for behavior flow decomposition and spatiotemporal pattern mining to generate an ideal healthy behavior flow model. Specifically, this includes: Within the first preset number of days after the target cat is confirmed to be in a healthy state, multimodal behavioral tuples are continuously collected to build a health behavior database exclusive to the target cat; Each multimodal behavior tuple in the health behavior database is decomposed into motion flow, posture flow, trajectory flow, and behavior temporal flow, and velocity sequence, step count sequence, posture angle sequence, position sequence, behavior category sequence, and behavior duration sequence are extracted respectively. Based on velocity sequences, step sequences, attitude angle sequences, position sequences, behavior category sequences, and behavior duration sequences, a temporal pattern mining algorithm is used to extract daily activity rhythms, toileting behavior patterns, eating and drinking patterns, activity area preferences, and behavior transition patterns, forming an individualized spatiotemporal behavior pattern library. Based on the spatiotemporal behavior pattern library, and combined with at least one situational factor among the current time, postoperative days, and ambient temperature, a healthy ideal behavior flow model is generated that includes ideal exercise intensity, ideal activity area distribution, ideal behavior sequence, and ideal behavior duration at the current moment.

4. The method according to claim 3, characterized in that, The process of comparing multimodal behavior tuples with a healthy ideal behavior flow model, calculating residuals for motion intensity, region deviation, behavior sequence, behavior duration, posture, and toileting behavior, and constructing a spatiotemporal residual tensor specifically includes: The real-time collected multimodal behavior tuples are compared point by point with the ideal health behavior flow model on the same time axis. The residuals of the movement intensity between the actual movement intensity and the ideal movement intensity, the residuals of the region deviation between the actual activity area and the ideal activity area, the residuals of the behavior sequence between the actual behavior sequence and the ideal behavior sequence, the residuals of the behavior duration between the actual behavior duration and the ideal behavior duration, the residuals of the posture between the actual posture angle and the ideal posture angle, and the residuals of the toilet behavior between the actual toilet behavior structure and the ideal toilet behavior structure are calculated respectively. The residuals of all dimensions are organized according to the time and space dimensions to form a spatiotemporal residual tensor.

5. The method according to claim 4, characterized in that, The method of using statistical process control to determine the significance of residuals and output residual pattern classification results specifically includes: Based on historical residual data collected during the healthy period, the mean and standard deviation of residuals in each dimension are calculated, and individualized residual control limits are set. Each residual value in the spatiotemporal residual tensor is compared with the residual control limit to determine whether the current residual has reached a significant residual level. If the current residual reaches a significant residual level, extract the spatial distribution features, temporal distribution features, and behavioral correlation features of the residual from the spatiotemporal residual tensor. The spatial distribution features include the spatial variance of the residual and the location and area of ​​the residual cluster. The temporal distribution features include the ratio of the nighttime residual to the daytime residual. The behavioral correlation features include the degree of increase in the residual when a specific behavior occurs. Spatial distribution features, temporal distribution features, and behavioral correlation features are input into the classifier for classification and recognition, and the classification result is output as at least one residual pattern among diffuse, focal, temporal, behavioral correlation, and mixed types.

6. The method according to claim 1, characterized in that, Based on the residual pattern classification results, specific detection models are constructed for postoperative recovery, pain stress, and urinary abnormalities, respectively. Conflict resolution and prioritization are performed through multi-scenario fusion decision-making, outputting comprehensive health status and action suggestions, specifically including: Based on the residual pattern classification results, a postoperative recovery specific detection model is constructed using a multi-stage state mechanism; Based on the residual pattern classification results, a pain stress-specific detection model was constructed by calculating the pain index. Based on the residual pattern classification results, a urinary abnormality-specific detection model is constructed by calculating the urinary abnormality index. Based on the output results of the postoperative recovery specific detection model, the pain stress specific detection model and the urinary abnormality specific detection model, the conflict resolution matrix is ​​used to identify and process the information overlap, causal relationship, independent events and primary and secondary relationships between multiple scenarios, and obtain the conflict identification results of multiple scenarios. The results of conflict identification in multiple scenarios are classified according to the priority ranking rules based on urgency, and a decision tree is used to hierarchically determine the overall health status and action recommendations.

7. The method according to claim 6, characterized in that, The postoperative recovery specificity detection model, based on the residual pattern classification results and constructed using a multi-stage state mechanism, specifically includes: Based on the type of surgery and the number of days after surgery, the postoperative recovery process is divided into the acute phase, the early recovery phase, the functional recovery phase, and the rehabilitation consolidation phase. Differentiated ideal activity baselines and pain control thresholds are set for each recovery phase, forming a multi-stage state machine. Based on a multi-stage state machine, the residual pattern classification results are used as input signals and integrated into the pain control assessment process. Different weight coefficients are assigned to different residual patterns in the residual pattern classification results. The residual patterns are combined with the posture residual, behavior sequence residual and regional deviation residual for weighted fusion, and the pain control status of each recovery stage is output. Based on the pain control status at each recovery stage, the real-time detected residual change trajectory is matched with a preset complication feature pattern library. When the matching degree reaches a preset matching degree threshold, an early warning output for the corresponding complication type is triggered. The complication feature pattern library includes residual combination features corresponding to wound infection mode, intestinal obstruction mode and pain loss of control mode.

8. The method according to claim 6, characterized in that, The pain stress-specific detection model, constructed based on residual pattern classification results and by calculating a pain index, specifically includes: Pain index analysis was performed based on postural residuals, behavioral sequence residuals, movement intensity residuals, and movement duration residuals to obtain multiple pain features, including stiffness index, tremor index, behavioral fragmentation index, behavioral persistence index, behavioral avoidance index, recovery delay index, rest dependence index, and self-care ability index. A comprehensive pain feature vector is constructed based on multiple pain features. The key features are used for initial screening to determine severe pain and neuropathic pain, the combined features are used for moderate pain, and the edge features are used for mild discomfort. The weights of the comprehensive pain feature vector are dynamically adjusted based on the residual pattern classification results, and the pain level and pain type are output.

9. The method according to claim 6, characterized in that, The method for constructing a urinary abnormality-specific detection model based on residual pattern classification results by calculating a urinary abnormality index specifically includes: Based on toilet behavior residuals, behavior sequence residuals, and regional deviation residuals, a temporal structure analysis is performed on each toilet event to extract the pre-toilet wandering time, the time during toileting, and the post-toilet burying time. These are then compared with the corresponding health baseline mean of the ideal health behavior flow model to output a urinary abnormality feature vector. The urinary abnormality feature vector includes the degree of wandering abnormality, the degree of urination abnormality, the degree of burying abnormality, the degree of toilet frequency abnormality, the degree of licking the urinary area after toileting abnormality, and the degree of urination location abnormality. When the residual pattern classification result contains focal patterns and the spatial clustering area of ​​the focal patterns is located in the litter box area, the weight of the structural abnormality dimension in the urinary abnormality feature vector is adjusted, and a weighted fusion calculation is performed in combination with the urinary abnormality feature vector to output the urinary abnormality index. Based on the threshold range of the urinary tract abnormality index, output at least one urinary tract abnormality level among normal, mild abnormality, moderate abnormality, and severe abnormality. Combine the temporal type in the residual pattern classification result to correct the urinary tract abnormality level and output the urinary tract abnormality judgment result.

10. A cat health intelligent detection system based on spatiotemporal behavioral residual analysis, characterized in that, The system specifically includes: The data fusion module is used to collect cat motion data and visual spatial data, and to interpolate, align and fuse the cat motion data and visual spatial data in a unified time coordinate system to generate multimodal behavior tuples. The feature analysis module is used to collect multimodal behavioral tuples during the cat's healthy period to build a healthy behavior library. By performing behavioral flow decomposition and spatiotemporal pattern mining on the healthy behavior library, a healthy ideal behavior flow model is generated. The pattern classification module is used to compare multimodal behavior tuples with the ideal health behavior flow model, calculate the residuals of exercise intensity, region deviation, behavior sequence, behavior duration, posture, and toilet behavior, construct the spatiotemporal residual tensor, and use statistical process control methods to determine the significance of the residuals, and output the residual pattern classification results. The scene detection module is used to construct specific detection models for postoperative recovery, pain stress and urinary abnormalities based on the residual pattern classification results. It also performs conflict resolution and priority ranking through multi-scene fusion decision-making, and outputs comprehensive health status and action suggestions.

Images