A pet conflict injury prediction and pathology tracing system and method based on multi-modal sensing

Abstract

The present application relates to a kind of pet conflict injury pre-judgment and pathological tracing system and method based on multi-modal sensing.It solves the problem that pet monitoring system lacks conflict dynamics analysis and injury prediction in the prior art.It includes three-dimensional dynamics acquisition unit, injury prediction unit, pathological feature monitoring unit, tracing matching unit and diagnosis report generation unit, conflict-injury correlation knowledge base is built in injury prediction unit, tracing matching unit is connected with injury prediction unit and pathological feature monitoring unit respectively, and diagnosis report generation unit is connected with tracing matching unit.The advantages of the present application are: it realizes the leap from passive alarm to active prediction, solves the industry pain points that pet internal injury is difficult to find and cause is difficult to diagnose, and is suitable for multi-pet family, pet agency and pet medical auxiliary scene.

Description

Technical Field

[0001] This invention relates to the field of smart pet healthcare technology, specifically to a system and method for predicting pet conflict injuries and tracing pathological causes based on multimodal sensing. Background Technology

[0002] In multi-pet households, pet conflicts can easily lead to hidden injuries such as fractures, internal organ damage, and soft tissue tears. Traditional manual observation makes it difficult to predict the type of injury or trace the cause, resulting in delayed detection of internal injuries and difficulty in diagnosis. Existing pet monitoring systems can only achieve conflict alarms or basic behavioral monitoring, lacking the ability to analyze conflict dynamics, predict injuries, and match pathological characteristics. They cannot provide veterinarians with accurate etiological tracing and diagnostic basis, and cannot meet the needs of pet medical assistance. Summary of the Invention

[0003] The purpose of this invention is to address the difficulty in performing conflict dynamics analysis and injury prediction in existing pet monitoring systems. It provides a pet conflict injury prediction and pathological tracing system and method based on multimodal sensing, which represents a leap from passive alarm to active prediction. This solves the industry pain points of difficulty in detecting internal injuries and diagnosing causes in pets, and is applicable to multi-pet households, pet institutions, and pet medical assistance scenarios.

[0004] To achieve the above objectives, the present invention adopts the following technical solution: a pet conflict injury prediction and pathological source tracing system based on multimodal sensing, comprising a three-dimensional dynamics acquisition unit, an injury prediction unit, a pathological feature monitoring unit, a source tracing matching unit, and a diagnostic report generation unit. The injury prediction unit has a built-in conflict-injury association knowledge base. The source tracing matching unit is connected to the injury prediction unit and the pathological feature monitoring unit, respectively. The diagnostic report generation unit is connected to the source tracing matching unit.

[0005] The above system provides a method for predicting pet conflict injuries and tracing pathological causes based on multimodal sensing, including the following steps:

[0006] S1. Dynamic Modeling: Capture three-dimensional motion data at the moment of conflict using a three-dimensional dynamic acquisition unit, and calculate impact force, contact area, drop height, and force angle.

[0007] S2, Damage Prediction: Call the conflict-damage association knowledge base to match a list of high-probability damage types and severity levels;

[0008] S3. Feature Collection: 72-hour intensive monitoring mode is activated for the pets involved, and data on gait, breathing, eating, abdominal movement and excretion are collected at high frequency.

[0009] S4. Pathological matching: Compare real-time pathological features with typical symptoms of predicted damage and calculate feature matching score;

[0010] S5. Confidence Calculation: A weighted algorithm is used to combine the intensity of conflict and the feature matching score to output a comprehensive confidence score.

[0011] S6. Report Output: Generate a source diagnosis report based on comprehensive confidence level, marking suspected symptoms, recommended examination items, and treatment directions.

[0012] In the aforementioned method for predicting and tracing the pathological origins of pet conflicts based on multimodal sensing, the three-dimensional dynamics acquisition unit integrates a three-dimensional accelerometer, a gyroscope, a pressure sensor array, and a height sensor; step S1 specifically includes the following steps:

[0013] S11. Data preprocessing: Denoise, filter, and perform spatiotemporal alignment on the raw three-dimensional motion data collected by the sensor to remove abnormal interference data and retain valid conflict condition data.

[0014] S12. Instantaneous impact force calculation: The calculation formula is as follows:

[0015] ;

[0016] in, For instantaneous impact force;

[0017] For the pet's weight;

[0018] The instantaneous change in velocity before and after the conflict;

[0019] Duration of the conflict effect;

[0020] S13. Effective Contact Area Calculation: The contact area is calculated based on the number of effective pressure points in the pressure sensor array. The calculation formula is as follows:

[0021] ;

[0022] in, The effective contact area in the conflict;

[0023] This represents the number of effective pressure points in the pressure sensor array.

[0024] The reference sensing area for a single sensing point;

[0025] S14. Determining the Fall Height and Force Angle: Calculate the fall height difference using time-series data from the height sensor, and combine this with attitude data from the 3D gyroscope to determine the collision force angle. Complete the construction of the three-dimensional dynamic model.

[0026] In the above-mentioned method for predicting and tracing the pathological origins of pet conflicts based on multimodal sensing, step S2 includes the following steps:

[0027] S21. Classification of Conflict Conditions: Based on the impact force, drop height, and force angle parameters obtained from dynamic modeling, conflict conditions are classified into four standard conditions: high-speed lateral impact, high-altitude fall, neck compression, and continuous biting.

[0028] S22. Threshold matching judgment: Retrieve the parameter thresholds from the conflict-damage association knowledge base, including the following thresholds: High-speed lateral impact judgment threshold: impact force ≥ 80N, lateral force angle ≥ 30°; High-altitude fall judgment threshold: fall height ≥ 0.5m; Neck compression judgment threshold: continuous pressure duration ≥ 3s, longitudinal pressure ≥ 50N; Continuous biting judgment threshold: contact duration ≥ 5s, local pressure ≥ 60N;

[0029] S23. Damage Severity Quantification: The severity of damage is divided into three levels: mild, moderate, and severe. The damage level coefficient is calculated based on the multiple by which the kinetic parameters exceed the standard. The calculation formula is as follows:

[0030] ;

[0031] in, This is the damage level coefficient.

[0032] These are measured dynamic parameter values.

[0033] The baseline threshold for the working condition corresponding to the conflict-damage association knowledge base;

[0034] S24. Generate a prediction list: Combine conflict conditions, damage types, and damage levels to output a structured damage prediction list.

[0035] In the above-mentioned method for predicting and tracing the pathological origins of pet conflicts based on multimodal sensing, step S3 includes the following steps:

[0036] S31. Sampling frequency tiered setting: The sampling frequency for regular monitoring is 5 times / hour. Under the 72-hour key monitoring mode, the high-frequency sampling frequency for the first 24 hours is 30 times / hour, the sampling frequency for the middle 24 hours is 20 times / hour, and the sampling frequency for the last 24 hours is 10 times / hour, so that the sampling frequency decreases gradually with the monitoring duration.

[0037] S32. Synchronous acquisition of multimodal data: Synchronously acquire pet gait trajectory data, respiratory rate, number of abdominal undulations per minute, single food intake, daily food intake frequency, excretion frequency and excrement status, and other multidimensional pathological data.

[0038] S33. Data Time Sequence Marking and Storage: Bind timestamps and vital sign numbers to each set of collected data to establish a 72-hour continuous pathological feature time sequence database, providing data support for subsequent pathological matching.

[0039] In step S3, the 72-hour intensive monitoring mode incorporates an abnormal early warning mechanism, specifically including: real-time monitoring of fluctuations in the pet's respiratory rate, gait deviation, and food and excretion volume, and setting a threshold for vital sign fluctuations; when the real-time vital sign data fluctuates by more than 25% compared to the normal baseline data, an abnormal early warning is automatically triggered, the abnormal data period and abnormal vital sign type are simultaneously marked, and the data is pushed to the source tracing and matching unit for secondary matching and verification; the formula for calculating the amplitude of vital sign fluctuations is:

[0040] ;

[0041] in, The amplitude of fluctuations in vital signs To collect vital signs data in real time, This refers to the average physical characteristics of pets under normal health conditions.

[0042] In the above-mentioned method for predicting and tracing the pathological origins of pet conflicts based on multimodal sensing, step S4 includes the following steps:

[0043] S41. Limping feature parameter extraction: Extract parameters such as pet's limb grounding time, stride, grounding pressure, and limb lift height through visual gait acquisition equipment;

[0044] S42. Quantitative Determination of Limp Type: Set limp determination criteria: Weight-bearing limp: The difference in ground pressure of the affected limb ≥30%; Weight-free limp: The percentage of time the affected limb is on the ground <20%; Toe-walking limp: The difference in limb lift height ≥40%, and the toe-touching contact area ≥80%.

[0045] S43. Injury type matching: Injury type is associated with limp type. Limping on tiptoe corresponds to foot injury, weight-bearing limp corresponds to soft tissue injury, and non-weight-bearing limp corresponds to fracture injury.

[0046] S44. Feature Matching Score Calculation: The Euclidean distance algorithm is used to calculate the matching degree between real-time pathological features and predicted lesion standard features. The calculation formula is as follows:

[0047] ;

[0048] in, The pathological feature matching score has a value range of [0,1].

[0049] The total number of characteristic indicators;

[0050] is the normalized value of the i-th real-time pathological feature;

[0051] is the normalized value of the i-th predicted damage standard feature.

[0052] In the aforementioned method for predicting pet conflict injuries and tracing pathological causes based on multimodal sensing, the confidence calculation in step S5 includes conflict intensity score calculation and comprehensive confidence weighted fusion calculation, specifically including:

[0053] S51. Conflict Intensity Score Calculation: The conflict score is calculated based on the damage level coefficient. The formula is as follows:

[0054] ;

[0055] in, The intensity of the conflict is rated, with a value ranging from [0,1].

[0056] Damage level coefficient;

[0057] S52. Weighted Fusion Calculation of Comprehensive Confidence: Combining preset weight ratios, construct the formula for calculating the confidence of causal traceability:

[0058] ;

[0059] in, The overall confidence level for tracing the source of injury is defined, with a value range of [0,1]; 0.4 is a fixed weight for the intensity of conflict, and 0.6 is a fixed weight for the consistency of pathological features.

[0060] S53. Comprehensive Confidence Level Grading Determination: Set the source tracing determination threshold. ≥0.8 indicates a strong causal relationship, 0.6≤ <0.8 indicates moderate causal correlation, 0.4≤ <0.6 indicates a weak causal relationship. <0.4 indicates no causal relationship.

[0061] In the above-mentioned method for predicting and tracing the pathological origin of pet conflicts based on multimodal sensing, in step S6, before generating the origination diagnosis report, the origination verification is performed through the origination matching unit configuration, specifically including the following steps:

[0062] S61. Time sequence consistency verification: retrieve 72-hour time sequence pathological data to verify whether the signs of injury are consistent with the time sequence of onset after the conflict, and exclude spontaneous injuries caused by non-conflict.

[0063] S62. Multi-feature cross-validation: Cross-validation is performed using data from multiple dimensions, including gait, breathing, eating, and abdominal movement. If a single feature has a high matching degree but the multi-feature matching degree is below 0.5, the overall confidence score is adjusted using the following formula: ;

[0064] S63. Working condition tracing and verification: reverse matching of the compatibility between conflict dynamic parameters and damage types, and elimination of invalid tracing results that contradict the parameters and symptoms.

[0065] In step S6, the report output specifically includes the following steps:

[0066] S64. Symptom labeling: Based on the comprehensive confidence level classification results, label the three categories of results in sequence: highly suspected symptom, suspected symptom, and ruled out symptom, and simultaneously label the corresponding confidence level values;

[0067] S65. Examination Item Matching: Automatically matches corresponding pet medical examination items based on the predicted injury type;

[0068] S66. Treatment direction classification recommendation: According to the level of mild, moderate and severe injury, there are three corresponding treatment directions: conservative rest and care, drug intervention and surgical intervention.

[0069] S67. Report Packaging and Output: Integrates kinetic parameters, pathological feature data, matching scores, confidence results, suspected symptoms, examination and treatment recommendations to generate a standardized electronic diagnostic traceability report with a unique code, supporting data traceability and archiving.

[0070] Compared with the prior art, the advantages of the present invention are as follows:

[0071] 1. Proactively predict hidden injuries to enable early detection and intervention of internal injuries in pets, reducing the difficulty of treatment.

[0072] 2. Accurately trace the causal relationship between conflict and injury, provide data support for veterinary diagnosis, and assist in precise treatment.

[0073] 3. A tiered early warning mechanism reduces invalid alarms, improving user experience and system usability.

[0074] 4. Suitable for multi-pet households, pet institutions, and medical settings; highly versatile and valuable for application. Attached Figure Description

[0075] Figure 1 This is a system architecture block diagram of the present invention;

[0076] Figure 2 This is a flowchart of the workflow in this invention. Detailed Implementation

[0077] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0078] like Figure 1-2 As shown, a pet conflict injury prediction and pathological source tracing system based on multimodal sensing includes a three-dimensional dynamics acquisition unit, an injury prediction unit, a pathological feature monitoring unit, a source tracing matching unit, and a diagnostic report generation unit. The injury prediction unit has a built-in conflict-injury association knowledge base. The source tracing matching unit is connected to the injury prediction unit and the pathological feature monitoring unit, respectively. The diagnostic report generation unit is connected to the source tracing matching unit.

[0079] The three-dimensional dynamics acquisition unit is used to collect multi-dimensional dynamic raw sensor data in real time during pet conflict. After preprocessing, the instantaneous impact force, effective contact area, fall height and force angle are solved to construct a three-dimensional dynamics model of pet conflict and output standardized dynamic parameters of conflict conditions.

[0080] The conflict-damage association knowledge base pre-stores thresholds for various pet conflict conditions, corresponding damage types, damage level quantification benchmark parameters, and condition-damage association mapping relationships. The damage prediction unit receives parameters from the three-dimensional dynamic model, completes conflict condition classification, threshold matching judgment, and damage level quantification, and outputs a structured high-probability damage prediction list containing damage type and damage severity.

[0081] The pathological feature monitoring unit and the injury prediction unit are linked and triggered to start a 72-hour gradient key monitoring mode after a pet conflict event. The sampling frequency decreases in gradient and the time series data of multimodal physiological signs of pets such as gait, breathing, abdominal movement, eating and excretion are collected synchronously. The built-in abnormal sign early warning mechanism monitors the amplitude of the sign fluctuation in real time and marks the time period and type of abnormal signs to form a continuous time series pathological feature database.

[0082] The source tracing and matching unit is used to receive the damage prediction list and time-series pathological feature data, extract quantitative pathological feature parameters, complete the similarity matching calculation between the predicted damage and the real-time pathological features, combine the intensity of conflict and feature matching score to solve the comprehensive confidence of damage source tracing, and perform time-series consistency verification, multi-feature cross-verification and working condition source tracing review, eliminate invalid source tracing results, and output accurate damage causal relationship determination results.

[0083] The diagnostic report generation unit is connected to the traceability matching unit. It is used to receive the verified traceability confidence and damage determination results, complete the classification and labeling of suspected diseases, the matching of medical examination items and the classification and recommendation of personalized treatment plans, integrate dynamic parameters, pathological monitoring data, matching scores and confidence indicators to generate a standardized diagnostic traceability report with a unique traceability code and complete the archiving output.

[0084] The 3D dynamics acquisition unit integrates a 3D accelerometer, gyroscope, pressure sensor array, and height sensor. It collects multi-dimensional raw sensor data during pet conflict, performs noise reduction filtering and spatiotemporal alignment preprocessing to obtain instantaneous impact force, effective contact area, fall height, and force angle, and constructs a 3D dynamics model of the pet conflict, outputting standardized working condition parameters. The injury prediction unit communicates with the 3D dynamics acquisition unit and receives dynamics modeling parameters to classify conflict conditions, match injury types, and quantify injury severity, outputting a structured high-probability injury prediction list. The pathological feature monitoring unit connects with the injury prediction unit and synchronously collects multimodal pathological time-series data on pet gait, respiration, eating, abdominal movement, and excretion at high frequency. It identifies abnormal fluctuations in vital signs in real time and marks abnormal data, constructing a continuous dynamic pathological feature database.

[0085] The source tracing and matching unit is bidirectionally connected to the injury prediction unit and the pathological feature monitoring unit. The source tracing and matching unit performs similarity matching between real-time pathological features and predicted injury standard features, and obtains the comprehensive confidence of injury source tracing by weighting the intensity of conflict. Through a multi-dimensional source tracing verification mechanism, invalid source tracing data is eliminated to accurately determine the causal relationship between conflict and pet injury. The diagnostic report generation unit is connected to the output end of the source tracing and matching unit and receives the final source tracing judgment result. It combines the injury level and pathological feature data to complete the classification and labeling of suspected diseases, matching of medical examination items, and recommendation of graded treatment plans. It integrates the data of the whole process to generate a traceable standardized diagnostic report.

[0086] A method for predicting and tracing the pathological origin of pet conflict injuries based on multimodal sensing includes the following steps:

[0087] S1. Dynamic Modeling: Capture three-dimensional motion data at the moment of conflict using a three-dimensional dynamic acquisition unit, and calculate impact force, contact area, drop height, and force angle.

[0088] S2, Damage Prediction: Call the conflict-damage association knowledge base to match a list of high-probability damage types and severity levels;

[0089] S3. Feature Collection: 72-hour intensive monitoring mode is activated for the pets involved, and data on gait, breathing, eating, abdominal movement and excretion are collected at high frequency.

[0090] S4. Pathological matching: Compare real-time pathological features with typical symptoms of predicted damage and calculate feature matching score;

[0091] S5. Confidence Calculation: A weighted algorithm is used to combine the intensity of conflict and the feature matching score to output a comprehensive confidence score.

[0092] S6. Report Output: Generate a source diagnosis report based on comprehensive confidence level, marking suspected symptoms, recommended examination items, and treatment directions.

[0093] The three-dimensional dynamics acquisition unit integrates a three-dimensional accelerometer, a gyroscope, a pressure sensor array, and a height sensor to acquire motion and force parameters during pet conflict. Step S1 specifically includes the following steps:

[0094] S11. Data preprocessing: Denoise, filter, and perform spatiotemporal alignment on the raw three-dimensional motion data collected by the sensor to remove abnormal interference data and retain valid conflict condition data.

[0095] S12. Instantaneous impact force calculation: The calculation formula is as follows:

[0096] ;

[0097] in, For instantaneous impact force;

[0098] For the pet's weight;

[0099] The instantaneous change in velocity before and after the conflict;

[0100] Duration of the conflict effect;

[0101] S13. Effective Contact Area Calculation: The contact area is calculated based on the number of effective pressure points in the pressure sensor array. The calculation formula is as follows:

[0102] ;

[0103] in, The effective contact area in the conflict;

[0104] This represents the number of effective pressure points in the pressure sensor array.

[0105] The reference sensing area for a single sensing point;

[0106] S14. Determining the Fall Height and Force Angle: Calculate the fall height difference using time-series data from the height sensor, and combine this with attitude data from the 3D gyroscope to determine the collision force angle. Complete the construction of the three-dimensional dynamic model.

[0107] Further, step S2 includes the following steps:

[0108] S21. Classification of Conflict Conditions: Based on the impact force, drop height, and force angle parameters obtained from dynamic modeling, conflict conditions are classified into four standard conditions: high-speed lateral impact, high-altitude fall, neck compression, and continuous biting.

[0109] S22. Threshold matching judgment: Retrieve the parameter thresholds from the conflict-damage association knowledge base, including the following thresholds: High-speed lateral impact judgment threshold: impact force ≥ 80N, lateral force angle ≥ 30°; High-altitude fall judgment threshold: fall height ≥ 0.5m; Neck compression judgment threshold: continuous pressure duration ≥ 3s, longitudinal pressure ≥ 50N; Continuous biting judgment threshold: contact duration ≥ 5s, local pressure ≥ 60N;

[0110] S23. Damage Severity Quantification: The severity of damage is divided into three levels: mild, moderate, and severe. The damage level coefficient is calculated based on the multiple by which the kinetic parameters exceed the standard. The calculation formula is as follows:

[0111] ;

[0112] in, K represents the damage severity coefficient, where K∈[1,1.5) indicates mild damage, K∈[1.5,2) indicates moderate damage, and K≥2 indicates severe damage.

[0113] These are measured dynamic parameter values.

[0114] The baseline threshold for the working condition corresponding to the conflict-damage association knowledge base;

[0115] S24. Generate a prediction list: Combine conflict conditions, damage types, and damage levels to output a structured damage prediction list.

[0116] Specifically, step S3 includes the following steps:

[0117] S31. Sampling frequency tiered setting: The sampling frequency for regular monitoring is 5 times / hour. Under the 72-hour key monitoring mode, the high-frequency sampling frequency for the first 24 hours is 30 times / hour, the sampling frequency for the middle 24 hours is 20 times / hour, and the sampling frequency for the last 24 hours is 10 times / hour, so that the sampling frequency decreases gradually with the monitoring duration.

[0118] S32. Synchronous acquisition of multimodal data: Synchronously acquire pet gait trajectory data, respiratory rate, number of abdominal undulations per minute, single food intake, daily food intake frequency, excretion frequency and excrement status, and other multidimensional pathological data.

[0119] S33. Data Time Sequence Marking and Storage: Bind timestamps and vital sign numbers to each set of collected data to establish a 72-hour continuous pathological feature time sequence database, providing data support for subsequent pathological matching.

[0120] In step S3, the 72-hour intensive monitoring mode incorporates an abnormal early warning mechanism, specifically including: real-time monitoring of fluctuations in the pet's respiratory rate, gait deviation, and food and excretion volume, and setting a threshold for vital sign fluctuations; when the real-time vital sign data fluctuates by more than 25% compared to the normal baseline data, an abnormal early warning is automatically triggered, the abnormal data period and abnormal vital sign type are simultaneously marked, and the data is pushed to the source tracing and matching unit for secondary matching and verification; the formula for calculating the amplitude of vital sign fluctuations is:

[0121] ;

[0122] in, The amplitude of fluctuations in vital signs To collect vital signs data in real time, This refers to the average physical characteristics of pets under normal health conditions.

[0123] In detail, step S4 includes the following steps:

[0124] S41. Limping feature parameter extraction: Extract parameters such as pet's limb grounding time, stride, grounding pressure, and limb lift height through visual gait acquisition equipment;

[0125] S42. Quantitative Determination of Limp Type: Set limp determination criteria: Weight-bearing limp: The difference in ground pressure of the affected limb ≥30%; Weight-free limp: The percentage of time the affected limb is on the ground <20%; Toe-walking limp: The difference in limb lift height ≥40%, and the toe-touching contact area ≥80%.

[0126] S43. Injury type matching: Injury type is associated with limp type. Limping on tiptoe corresponds to foot injury, weight-bearing limp corresponds to soft tissue injury, and non-weight-bearing limp corresponds to fracture injury.

[0127] S44. Feature Matching Score Calculation: The Euclidean distance algorithm is used to calculate the matching degree between real-time pathological features and predicted lesion standard features. The calculation formula is as follows:

[0128] ;

[0129] in, S represents the pathological feature matching score, with a value range of [0,1]. The closer S is to 1, the higher the feature matching degree.

[0130] The total number of characteristic indicators;

[0131] is the normalized value of the i-th real-time pathological feature;

[0132] is the normalized value of the i-th predicted damage standard feature.

[0133] In addition, the confidence calculation in step S5 includes the calculation of the conflict intensity score and the weighted fusion calculation of the overall confidence, specifically including:

[0134] S51. Conflict Intensity Score Calculation: The conflict score is calculated based on the damage level coefficient. The formula is as follows:

[0135] ;

[0136] in, The intensity of the conflict is rated, with a value ranging from [0,1].

[0137] For damage level coefficient, when When ≥3, Take the maximum value of 1;

[0138] S52. Weighted Fusion Calculation of Comprehensive Confidence: Combining preset weight ratios, construct the formula for calculating the confidence of causal traceability:

[0139] ;

[0140] in, The overall confidence level for tracing the source of injury is defined, with a value range of [0,1]; 0.4 is a fixed weight for the intensity of conflict, and 0.6 is a fixed weight for the consistency of pathological features.

[0141] S53. Comprehensive Confidence Level Grading Determination: Set the source tracing determination threshold. ≥0.8 indicates a strong causal relationship, 0.6≤ <0.8 indicates moderate causal correlation, 0.4≤ <0.6 indicates a weak causal relationship. <0.4 indicates no causal relationship.

[0142] In step S6, before generating the source tracing diagnostic report, source tracing verification is performed through the source tracing matching unit configuration, specifically including the following steps:

[0143] S61. Time sequence consistency verification: retrieve 72-hour time sequence pathological data to verify whether the signs of injury are consistent with the time sequence of onset after the conflict, and exclude spontaneous injuries caused by non-conflict.

[0144] S62. Multi-feature cross-validation: Cross-validation is performed using data from multiple dimensions, including gait, breathing, eating, and abdominal movement. If a single feature has a high matching degree but the multi-feature matching degree is below 0.5, the overall confidence score is adjusted using the following formula: ;

[0145] S63. Working condition traceability verification: reverse matching of the compatibility between conflict dynamic parameters and damage types, eliminating invalid traceability results that contradict the parameters and symptoms, and ensuring the accuracy of traceability.

[0146] In step S6, the report output specifically includes the following steps:

[0147] S64. Symptom labeling: Based on the comprehensive confidence level classification results, label the three categories of results in sequence: highly suspected symptom, suspected symptom, and ruled out symptom, and simultaneously label the corresponding confidence level values;

[0148] S65. Examination Item Matching: Based on the predicted injury type, the system automatically matches the corresponding pet medical examination items. Rib / internal organ injuries are matched with B-ultrasound and X-ray examinations; fractures / joint dislocations are matched with 3D imaging scans; soft tissue injuries are matched with surface imaging examinations; and respiratory disorders are matched with cardiopulmonary function tests.

[0149] S66. Treatment direction classification recommendation: According to the level of mild, moderate and severe injury, there are three corresponding treatment directions: conservative rest and care, drug intervention and surgical intervention.

[0150] S67. Report Packaging and Output: Integrates kinetic parameters, pathological feature data, matching scores, confidence results, suspected symptoms, examination and treatment recommendations to generate a standardized electronic diagnostic traceability report with a unique code, supporting data traceability and archiving.

[0151] 1. Dynamic Quantitative Analysis: Breaking through the limitations of traditional conflict identification, it quantifies the intensity, angle, and height of conflict through three-dimensional data to achieve accurate prediction of damage type.

[0152] 2. Refined differentiation of pathological features: Refine the types of gait, breathing and other features to accurately match the typical manifestations of different injuries and reduce the misdiagnosis rate.

[0153] 3. Weighted confidence mechanism: Combining the intensity of conflict with the degree of pathological matching, quantifying causal correlation, and outputting graded early warning and diagnostic suggestions.

[0154] 4. Medical-grade traceability report: Integrates prediction, matching, and confidence data to provide veterinarians with evidence of etiology and direction for examination, thereby improving diagnostic efficiency.

[0155] In summary, the principle of this invention lies in the following: It achieves its functions through a three-dimensional dynamics acquisition unit, an injury prediction unit, a pathological feature monitoring unit, a source tracing and matching unit, and a diagnostic report generation unit. The three-dimensional dynamics acquisition unit collects spatial coordinates, motion velocity vectors, limb postures, and contact collision data of multiple pets to construct a three-dimensional dynamics model of the conflict. The injury prediction unit, based on a built-in conflict-injury association knowledge base, calculates force parameters according to the dynamic data to predict the type and severity of injury. The pathological feature monitoring unit collects pet gait symmetry, respiratory waveforms, eating and swallowing movements, and abdominal undulation characteristics. The source tracing and matching unit performs similarity matching between the predicted injury type and the monitored pathological features, verifies the causal relationship, and outputs a comprehensive confidence level. The diagnostic report generation unit outputs a source tracing report containing the conflict process, predicted injury, feature matching degree, and treatment suggestions. This invention achieves a leap from passive alarm to active prediction, solving the industry pain points of difficulty in detecting internal injuries in pets and difficulty in diagnosing the cause of the disease. It is suitable for multi-pet households, pet institutions, and pet medical assistance scenarios.

[0156] The specific embodiments described herein are merely illustrative of the spirit of the invention. Those skilled in the art to which this invention pertains may make various modifications or additions to the described specific embodiments or use similar methods to substitute them, without departing from the spirit of the invention or exceeding the scope defined by the appended claims.

Claims

1. A pet conflict injury prediction and pathological source tracing system based on multimodal sensing, comprising a three-dimensional dynamics acquisition unit, an injury prediction unit, a pathological feature monitoring unit, a source tracing and matching unit, and a diagnostic report generation unit, characterized in that, The damage prediction unit has a built-in conflict-damage association knowledge base, the source matching unit is connected to the damage prediction unit and the pathological feature monitoring unit respectively, and the diagnostic report generation unit is connected to the source matching unit.

2. The pet conflict injury prediction and pathological tracing system based on multimodal sensing according to claim 1, characterized in that, The three-dimensional dynamics acquisition unit integrates a three-dimensional accelerometer, a gyroscope, a pressure sensor array, and a height sensor. It acquires multi-dimensional raw sensor data during pet conflict, performs noise reduction filtering and spatiotemporal alignment preprocessing to obtain instantaneous impact force, effective contact area, fall height, and force angle, and constructs a three-dimensional dynamics model of the pet conflict, outputting standardized working condition parameters. The injury prediction unit communicates with the three-dimensional dynamics acquisition unit and receives dynamics modeling parameters to classify conflict conditions, match injury types, and quantify injury severity, outputting a structured high-probability injury prediction list. The pathological feature monitoring unit connects with the injury prediction unit and synchronously acquires multi-modal pathological time-series data on pet gait, respiration, eating, abdominal movement, and excretion at high frequency, identifying abnormal fluctuations in vital signs and marking abnormal data in real time, constructing a continuous dynamic pathological feature database.

3. The pet conflict injury prediction and pathological tracing system based on multimodal sensing according to claim 2, characterized in that, The source tracing and matching unit is bidirectionally connected to the injury prediction unit and the pathological feature monitoring unit. The source tracing and matching unit performs similarity matching between real-time pathological features and predicted injury standard features, and obtains the comprehensive confidence of injury source tracing by weighting the intensity of conflict. It also eliminates invalid source tracing data through a multi-dimensional source tracing verification mechanism to accurately determine the causal relationship between conflict and pet injury. The diagnostic report generation unit is connected to the output of the source tracing and matching unit and receives the final source tracing judgment result. It combines the injury level and pathological feature data to complete the classification and labeling of suspected diseases, matching of medical examination items, and recommendation of graded treatment plans. It integrates the data of the whole process to generate a traceable standardized diagnostic report.

4. A method applicable to the system according to any one of claims 1-3, characterized in that, Includes the following steps: S1. Dynamic Modeling: Capture three-dimensional motion data at the moment of conflict using a three-dimensional dynamic acquisition unit, and calculate impact force, contact area, drop height, and force angle. S2, Damage Prediction: Call the conflict-damage association knowledge base to match a list of high-probability damage types and severity levels; S3. Feature Collection: 72-hour intensive monitoring mode is activated for the pets involved, and data on gait, breathing, eating, abdominal movement and excretion are collected at high frequency. S4. Pathological matching: Compare real-time pathological features with typical symptoms of predicted damage and calculate feature matching score; S5. Confidence Calculation: A weighted algorithm is used to combine the intensity of conflict and the feature matching score to output a comprehensive confidence score. S6. Report Output: Generate a source diagnosis report based on comprehensive confidence level, marking suspected symptoms, recommended examination items, and treatment directions.

5. The method for predicting pet conflict injuries and tracing pathological causes based on multimodal sensing according to claim 4, characterized in that, The three-dimensional dynamics acquisition unit integrates a three-dimensional accelerometer, a gyroscope, a pressure sensor array, and a height sensor; step S1 specifically includes the following steps: S11. Data preprocessing: Denoise, filter, and perform spatiotemporal alignment on the raw three-dimensional motion data collected by the sensor to remove abnormal interference data and retain valid conflict condition data. S12. Instantaneous impact force calculation: The calculation formula is as follows: ； in, For instantaneous impact force; For the pet's weight; The instantaneous change in velocity before and after the conflict; Duration of the conflict effect; S13. Effective Contact Area Calculation: The contact area is calculated based on the number of effective pressure points in the pressure sensor array. The calculation formula is as follows: ； in, The effective contact area in the conflict; This represents the number of effective pressure points in the pressure sensor array. The reference sensing area for a single sensing point; S14. Determining the Fall Height and Force Angle: Calculate the fall height difference using time-series data from the height sensor, and combine this with attitude data from the 3D gyroscope to determine the collision force angle. Complete the construction of the three-dimensional dynamic model.

6. The method for predicting and tracing the pathological origin of pet conflict injuries based on multimodal sensing according to claim 5, characterized in that, Step S2 includes the following steps: S21. Classification of Conflict Conditions: Based on the impact force, drop height, and force angle parameters obtained from dynamic modeling, conflict conditions are classified into four standard conditions: high-speed lateral impact, high-altitude fall, neck compression, and continuous biting. S22. Threshold matching judgment: Retrieve the parameter thresholds from the conflict-damage association knowledge base, including the following thresholds: High-speed lateral impact judgment threshold: impact force ≥ 80N, lateral force angle ≥ 30°; High-altitude fall judgment threshold: fall height ≥ 0.5m; Neck compression judgment threshold: continuous pressure duration ≥ 3s, longitudinal pressure ≥ 50N; Continuous biting judgment threshold: contact duration ≥ 5s, local pressure ≥ 60N; S23. Damage Severity Quantification: The severity of damage is divided into three levels: mild, moderate, and severe. The damage level coefficient is calculated based on the multiple by which the kinetic parameters exceed the standard. The calculation formula is as follows: ； in, This is the damage level coefficient. These are measured dynamic parameter values. The baseline threshold for the working condition corresponding to the conflict-damage association knowledge base; S24. Generate a prediction list: Combine conflict conditions, damage types, and damage levels to output a structured damage prediction list.

7. The method for predicting and tracing the pathological origin of pet conflict injuries based on multimodal sensing according to claim 6, characterized in that, Step S3 includes the following steps: S31. Sampling frequency tiered setting: The sampling frequency for regular monitoring is 5 times / hour. Under the 72-hour key monitoring mode, the high-frequency sampling frequency for the first 24 hours is 30 times / hour, the sampling frequency for the middle 24 hours is 20 times / hour, and the sampling frequency for the last 24 hours is 10 times / hour, so that the sampling frequency decreases gradually with the monitoring duration. S32. Synchronous acquisition of multimodal data: Synchronously acquire pet gait trajectory data, respiratory rate, number of abdominal undulations per minute, single food intake, daily food intake frequency, excretion frequency and excrement status, and other multidimensional pathological data. S33. Data time series marking and storage: Bind timestamps and vital sign numbers to each set of collected data to establish a 72-hour continuous pathological feature time series database to provide data support for subsequent pathological matching. In step S3, the 72-hour intensive monitoring mode incorporates an abnormal early warning mechanism, specifically including: real-time monitoring of the pet's respiratory rate, gait deviation, and fluctuations in food intake and excretion, and setting a threshold for vital sign fluctuations; when the real-time vital sign data fluctuates by more than 25% compared to the normal baseline data, an abnormal early warning is automatically triggered, the abnormal data period and abnormal vital sign type are simultaneously marked, and the data is pushed to the source tracing and matching unit for secondary matching and verification; wherein, the formula for calculating the amplitude of vital sign fluctuations is: ； in, The amplitude of fluctuations in vital signs To collect vital signs data in real time, This refers to the average physical characteristics of pets under normal health conditions.

8. The method for predicting and tracing the pathological origin of pet conflict injuries based on multimodal sensing according to claim 7, characterized in that, Step S4 includes the following steps: S41. Limping feature parameter extraction: Extract parameters such as pet's limb grounding time, stride, grounding pressure, and limb lift height through visual gait acquisition equipment; S42. Quantitative Determination of Limp Type: Set limp determination criteria: Weight-bearing limp: The difference in ground pressure of the affected limb ≥30%; Weight-free limp: The percentage of time the affected limb is on the ground <20%; Toe-walking limp: The difference in limb lift height ≥40%, and the toe-touching contact area ≥80%. S43. Injury type matching: Injury type is associated with limp type. Limping on tiptoe corresponds to foot injury, weight-bearing limp corresponds to soft tissue injury, and non-weight-bearing limp corresponds to fracture injury. S44. Feature Matching Score Calculation: The Euclidean distance algorithm is used to calculate the matching degree between real-time pathological features and predicted lesion standard features. The calculation formula is as follows: ； in, The pathological feature matching score has a value range of [0,1]. The total number of characteristic indicators; is the normalized value of the i-th real-time pathological feature; The normalized value of the i-th predicted damage standard feature; The confidence calculation in step S5 includes the calculation of the conflict intensity score and the weighted fusion calculation of the overall confidence, specifically including: S51. Conflict Intensity Score Calculation: The conflict score is calculated based on the damage level coefficient. The formula is as follows: ； in, The intensity of the conflict is rated, with a value ranging from [0,1]. Damage level coefficient; S52. Weighted Fusion Calculation of Comprehensive Confidence: Combining preset weight ratios, construct the formula for calculating the confidence of causal traceability: ； in, The overall confidence level for tracing the source of injury is defined, with a value range of [0,1]; 0.4 is a fixed weight for the intensity of conflict, and 0.6 is a fixed weight for the consistency of pathological features. S53. Comprehensive Confidence Level Grading Determination: Set the source tracing determination threshold. ≥0.8 indicates a strong causal relationship, 0.6≤ <0.8 indicates moderate causal correlation, 0.4≤ <0.6 indicates a weak causal relationship. <0.4 indicates no causal relationship.

9. The method for predicting and tracing the pathological origin of pet conflict injuries based on multimodal sensing according to claim 8, characterized in that, In step S6, before generating the source tracing diagnostic report, source tracing verification is performed through the source tracing matching unit configuration, specifically including the following steps: S61. Time sequence consistency verification: retrieve 72-hour time sequence pathological data to verify whether the signs of injury are consistent with the time sequence of onset after the conflict, and exclude spontaneous injuries caused by non-conflict. S62. Multi-feature cross-validation: Cross-validation is performed using data from multiple dimensions, including gait, breathing, eating, and abdominal movement. If a single feature has a high matching degree but the multi-feature matching degree is below 0.5, the overall confidence score is adjusted using the following formula: ; S63. Working condition tracing and verification: reverse matching of the compatibility between conflict dynamic parameters and damage types, and elimination of invalid tracing results that contradict the parameters and symptoms.

10. The method for predicting and tracing the pathological origin of pet conflict injuries based on multimodal sensing according to claim 8, characterized in that, In step S6, the report output specifically includes the following steps: S64. Symptom labeling: Based on the comprehensive confidence level classification results, label the three categories of results in sequence: highly suspected symptom, suspected symptom, and ruled out symptom, and simultaneously label the corresponding confidence level values; S65. Examination Item Matching: Automatically matches corresponding pet medical examination items based on the predicted injury type; S66. Treatment direction classification recommendation: According to the level of mild, moderate and severe injury, there are three corresponding treatment directions: conservative rest and care, drug intervention and surgical intervention. S67. Report Packaging and Output: Integrates kinetic parameters, pathological feature data, matching scores, confidence results, suspected symptoms, examination and treatment recommendations to generate a standardized electronic diagnostic traceability report with a unique code, supporting data traceability and archiving.

Images