Intelligent sensor-based pet health ai monitoring system for grooming processes

Abstract

The application relates to the technical field of intelligent monitoring, and particularly discloses a pet health AI monitoring system for a washing and grooming process based on an intelligent sensor. The system comprises an intelligent load-bearing sensing array integrated in a washing and grooming workbench, a non-contact multi-modal physiological sensing module, an edge computing and fusion processing unit, and a health state evaluation and early warning module. Through multi-modal data fusion, dynamic motion interference filtering, and abnormality detection and trend analysis, real-time synchronous monitoring of pet heart rate, respiration, blood oxygen, and foot pressure distribution and health risk early warning are realized. The system deeply integrates the high-density intelligent load-bearing sensing array and the non-contact multi-modal physiological sensing module, and constructs an integrated sensing system capable of synchronously and continuously collecting vital signs and foot load-bearing states during the dynamic washing and grooming process of the pet.

Description

Technical Field

[0001] This invention belongs to the field of intelligent monitoring technology, specifically relating to an AI monitoring system for pet health during the grooming process based on intelligent sensors. Background Technology

[0002] In the field of animal health management and intelligent monitoring, the non-invasive assessment of pets' physiological status using sensor technology and data analysis methods has become an important research direction. This field aims to improve pet welfare and medical efficiency by continuously collecting pets' vital signs and behavioral data to achieve early warning and intervention for potential health problems.

[0003] Health monitoring in pet grooming scenarios is a crucial aspect of ensuring pet welfare and enhancing service professionalism. During grooming, pets are typically in a non-static and not fully cooperative state, which presents a unique challenge in obtaining real-time and accurate health information. An ideal technological solution should be able to continuously and imperceptibly measure the pet's core vital signs during this dynamic process, while simultaneously assessing the weight-bearing capacity of their limbs and the presence of abnormal pressure distribution in their bones and joints.

[0004] Existing technologies typically rely on single-point measurements in static or controlled environments, making them ill-suited to the dynamic and complex nature of the grooming process. Traditional monitoring methods cannot consistently and stably capture key physiological parameters such as heart rate and respiratory rate when pets are moving, struggling, or frequently changing posture, leading to data interruptions or distortions. Furthermore, existing systems lack the ability to perceive and analyze the real-time distribution of foot pressure on weight-bearing surfaces such as grooming tables, failing to effectively identify abnormal weight-bearing patterns caused by pain, joint problems, or nerve damage. This lack of monitoring capability makes it difficult for groomers to promptly detect discomfort or potential musculoskeletal issues in pets, potentially delaying the identification of health risks and even exacerbating pet suffering through improper handling.

[0005] Therefore, there is an urgent need for an intelligent system that can adapt to the dynamic characteristics of the washing and care process and achieve integrated and synchronous monitoring of vital signs and load-bearing status. Summary of the Invention

[0006] This invention provides a pet health AI monitoring system based on intelligent sensors during the pet grooming process, in order to solve the technical contradictions of existing technologies that cannot continuously and stably monitor vital signs during the dynamic pet grooming process and lack the ability to synchronously perceive and analyze the weight-bearing status of limbs.

[0007] To achieve the above objectives, this invention provides a pet health AI monitoring system based on intelligent sensors during the pet grooming process. The system includes an intelligent load-bearing sensing array integrated within the grooming workbench, a non-contact multimodal physiological sensing module, an edge computing and fusion processing unit, and a health status assessment and early warning module.

[0008] The intelligent load-bearing sensing array consists of high-density distributed pressure sensor units arranged in a matrix, completely embedded beneath the load-bearing surface of the pet's grooming workbench. Each pressure sensor unit independently collects the vertical pressure value of its coverage area and uploads the raw pressure data in real time at a sampling frequency of no less than 100 Hz. This intelligent load-bearing sensing array is configured to generate a pressure distribution heatmap of the pet's paws at any given time, and to calculate in real time the pressure center coordinates, total pressure value, and pressure distribution entropy in the paw region for each foot.

[0009] The non-contact multimodal physiological sensing module includes a millimeter-wave radar sensor and a hyperspectral imaging sensor, both mounted in a fixed spatial relationship on an adjustable bracket above the pet grooming station. The millimeter-wave radar sensor is configured to transmit frequency-modulated continuous waves to the pet area on the grooming station and receive micro-motion reflection signals from the pet's chest cavity. By performing phase demodulation and spectral analysis on these micro-motion reflection signals, micro-Doppler features related to heartbeat and respiration are extracted, and real-time heart rate and respiratory rate parameters are separated and calculated. The hyperspectral imaging sensor acquires hyperspectral image sequences of specific areas on the pet's body surface at a preset frame rate. By analyzing changes in the spectral reflectance characteristics of the skin or mucous membrane areas in the visible and near-infrared bands, physiological parameters related to blood oxygen saturation and local microcirculation status are calculated.

[0010] The edge computing and fusion processing unit is deployed locally on the pet grooming workbench. It first receives real-time pressure data streams from the intelligent load-bearing sensing array and real-time physiological parameter data streams from the non-contact multimodal physiological sensing module. Within this edge computing and fusion processing unit, a spatiotemporal alignment and data fusion engine is run. Based on a unified timestamp, this engine synchronizes asynchronous data from different sensors and, according to a preset pet body model, spatially correlates and maps pressure distribution data with physiological sensing data.

[0011] Furthermore, the edge computing and fusion processing unit includes a dynamic motion interference filtering submodule. This submodule analyzes the macroscopic limb movement components and abrupt change patterns in pressure distribution data in millimeter-wave radar signals to identify and mark the periods when the pet struggles violently or undergoes significant changes in body position. For the physiological parameter data during these periods, it initiates a noise suppression algorithm based on adaptive Kalman filtering to filter out motion artifacts and ensure the calculation stability of parameters such as heart rate and respiratory rate during dynamic processes.

[0012] The health status assessment and early warning module receives multimodal time-series data after fusion and filtering. This module includes a weight-bearing pattern anomaly detection submodule and a physiological state trend analysis submodule. The weight-bearing pattern anomaly detection submodule calculates the following core indicators based on real-time acquired four-legged pressure distribution data: the asymmetry index of pressure distribution between each leg, the trajectory drift of the pressure center of a single leg per unit time, and the time difference between the peak pressure occurrences of the two forelegs or two hind legs. This submodule incorporates a weight-bearing pattern benchmark model trained on a large amount of healthy pet grooming data. By comparing the real-time calculated core indicators with the normal range thresholds provided by the benchmark model, if any indicator continuously exceeds the threshold range for more than 3 seconds, it is determined to be a weight-bearing pattern anomaly, and a first-level early warning signal is generated. This first-level early warning signal indicates that the pet may have limb pain, joint discomfort, or neurological damage.

[0013] The physiological state trend analysis submodule performs time-series analysis on the filtered heart rate, respiratory rate, and blood oxygen saturation parameters. This submodule calculates the time-domain indices of heart rate variability, the short-term volatility of respiratory rate, and the moving average of blood oxygen saturation. Simultaneously, it uses a physiological state prediction model based on a long short-term memory network to make rolling predictions of physiological parameter trends over the next 30 seconds. This submodule compares real-time parameters, their short-term trends, and predicted trends with preset species-specific physiological safety intervals. When a sharp increase in heart rate or respiratory rate is detected, a decreasing trend in blood oxygen saturation is observed, or the prediction model outputs a future risk probability exceeding 70%, a secondary warning signal is generated. This secondary warning signal indicates that the pet may be in a state of stress, pain, or potential cardiopulmonary dysfunction.

[0014] Furthermore, the system also includes an early warning synthesis and interaction module. This module receives primary and secondary early warning signals from the health status assessment and early warning module. The module contains a rule engine that synthesizes early warning signals according to a preset priority logic based on their type, level, and concurrency. When both primary and secondary early warning signals are received simultaneously, the system determines it to be a high-risk state and generates the highest-level audible and visual alarm command. Simultaneously, it sends detailed alarm information containing specific abnormal indicators and recommended countermeasures to the associated mobile terminal via a wireless communication interface. If only a single type of early warning signal is received, a corresponding level of alert information is generated. All early warning events, raw data snapshots, and processing logs are stored in a local secure storage unit for subsequent traceability and analysis.

[0015] In one embodiment of the present invention, the millimeter-wave radar sensor operates in the 60 GHz to 64 GHz frequency band, with a bandwidth of not less than 2 GHz, to meet the high-precision detection requirements for micrometer-level displacement of the thoracic cavity. The hyperspectral imaging sensor operates in the spectral range of 500 nm to 900 nm, with a spectral resolution better than 5 nm and a spatial resolution of not less than 1 million pixels, to ensure that characteristic spectral absorption differences related to blood oxygenation parameters can be captured.

[0016] In one embodiment of the present invention, the adaptive Kalman filter algorithm in the dynamic motion interference filtering submodule does not have fixed values ​​for its process noise covariance matrix and measurement noise covariance matrix. Instead, these values ​​are dynamically adjusted based on the real-time identified pet motion intensity level. The motion intensity level is determined jointly by analyzing the total variance of the pressure sensor array data and the Doppler spectral width of the millimeter-wave radar signal. When the motion intensity level is high, the algorithm automatically increases the process noise covariance, making the filter more confident in the predicted values; when the motion intensity level is low, it decreases the process noise covariance, making the filter more confident in the measured values, thereby achieving optimal filtering under motion interference.

[0017] As one embodiment of the present invention, the process of establishing the load-bearing mode benchmark model is as follows: In the initial stage of system deployment or calibration, a large amount of plantar pressure data of healthy pets of different breeds and sizes under calm bathing conditions are collected to form a training dataset; for each measurement in the dataset, the core indicators are extracted as feature vectors; the feature vector set is clustered using a Gaussian mixture model to determine the feature distribution of various healthy load-bearing modes; finally, the feature value range covering 95% of healthy samples is set as the dynamic threshold range of each core indicator, and a load-bearing mode anomaly detection submodule is embedded.

[0018] In one embodiment of the present invention, the system adopts a modular power supply and communication architecture. The intelligent load-bearing sensing array and the non-contact multimodal physiological sensing module are uniformly powered by the DC power supply inside the washing and care workbench, and connected to the edge computing and fusion processing unit through a wired communication bus to ensure low latency and high reliability of the data link. The edge computing and fusion processing unit, the health status assessment and early warning module, and the early warning synthesis and interaction module can be integrated into an industrial-grade edge computing device, which also provides wireless LAN access for alarm push and data synchronization.

[0019] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0020] 1. This invention deeply integrates a high-density intelligent load-bearing sensing array with a non-contact multimodal physiological sensing module to construct an integrated sensing system capable of synchronously and continuously collecting vital signs and foot load-bearing status during the dynamic grooming process of pets. This system fundamentally overcomes the limitations of traditional static monitoring methods, achieving comprehensive real-time sensing of heart rate, respiration, blood oxygenation, and limb pressure distribution, providing an unprecedented multi-dimensional data foundation for comprehensively assessing the health status of pets during grooming.

[0021] 2. This invention innovatively designs a dynamic motion interference filtering submodule, which employs a Kalman filter algorithm based on motion intensity adaptation. This effectively identifies and suppresses motion artifacts caused by the pet's struggles and movement that interfere with the calculation of core physiological parameters. This technology ensures the continuity and accuracy of key vital sign data such as heart rate and respiratory rate in the uncontrolled dynamic scenario of pet grooming, solving the core technical problem of easily distorted and interrupted physiological signals in dynamic environments.

[0022] 3. This invention achieves intelligent analysis from raw data to health risk insights through the collaborative work of the load-bearing mode anomaly detection submodule and the physiological state trend analysis submodule. The system not only compares static thresholds of load-bearing parameters in real time, but also uses time-series analysis and predictive models to anticipate deteriorating trends in physiological states. This dual mechanism combining immediate anomaly detection and trend warning significantly enhances the system's ability to detect potential health risks early and its forward-looking warning capabilities, enabling pet groomers to take timely intervention measures and effectively protect pet welfare and safety.

[0023] 4. The system architecture of this invention emphasizes edge computing and localized processing. All core data fusion, filtering, analysis, and early warning generation are completed locally at the washing station, greatly reducing reliance on cloud networks and ensuring the real-time performance and privacy of data processing. Simultaneously, the modular design makes the system easy to deploy, maintain, and upgrade, exhibiting excellent practicality and scalability. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the overall technical solution architecture of the pet health AI monitoring system based on intelligent sensors during the washing and care process proposed in this invention;

[0025] Figure 2 This is a schematic diagram of the core principle framework of the dynamic motion interference filtering submodule and multimodal data fusion in this invention;

[0026] Figure 3 This is a schematic diagram of the multi-level interaction relationship and data flow between the non-contact multimodal physiological sensing module and the intelligent load-bearing sensing array in the present invention.

[0027] Figure 4This is a logical flowchart of the health status assessment and early warning module in this invention. Detailed Implementation

[0028] Example 1: Please refer to the appendix Figure 1 To be continued Figure 4 The pet health AI monitoring system based on intelligent sensors described in this invention consists of four core functional modules: an intelligent load-bearing sensing array, a non-contact multimodal physiological sensing module, an edge computing and fusion processing unit, and a health status assessment and early warning module. These modules achieve high-precision data synchronization via a wired communication bus and a unified time base, and complete all sensing, processing, and decision-making processes locally, ensuring the system's real-time performance, robustness, and privacy security in the dynamic and uncontrolled environment of pet grooming. Please refer to the appendix. Figure 1 This diagram is a schematic diagram of the overall technical solution architecture of this system, which clearly shows the physical deployment location, data flow and logical interaction relationship of each functional module.

[0029] The intelligent load-bearing sensing array is fully embedded beneath the load-bearing surface of the grooming workbench. Its structure consists of a matrix arrangement of high-density distributed pressure sensor units. The physical coverage area of ​​this intelligent load-bearing sensing array is sufficient to accommodate the four-legged projection range of common small to medium-sized dogs and cats when standing, sitting, lying down, or moving slightly, with typical dimensions of 800 mm in length and 600 mm in width. Each pressure sensor unit has a sensing area of ​​10 mm × 10 mm, with a center-to-center spacing of 15 mm between adjacent units, forming a high-density grid with a spatial resolution of approximately 4.4 sampling points per square centimeter. All pressure sensor units utilize piezoresistive thin-film sensing materials, exhibiting excellent linearity, fast response characteristics, and water resistance. Each unit independently collects the vertical pressure value of its coverage area at a sampling frequency of no less than 100 Hz. The raw pressure data is encapsulated in 16-bit integer format and appended with a timestamp accurate to the microsecond level before being uploaded in real time to the edge computing and fusion processing unit via an internal high-speed serial bus.

[0030] During operation, the intelligent load-bearing sensing array continuously generates pressure distribution heatmaps of the pet's paws at any given time. These heatmaps are represented as a two-dimensional matrix, with each element corresponding to the measured pressure value of a pressure sensor unit, expressed in kilopascals (kPa). The system identifies four main pressure clusters using a connected component analysis algorithm, corresponding to the left forefoot, right forefoot, left hindfoot, and right hindfoot. For each foot pressure cluster, the system calculates three key indicators in real time: first, the total pressure borne by the foot, obtained by summing the pressure values ​​of all effective sensor units within the cluster; second, the coordinates of the foot pressure center, calculated using a weighted average method, where the spatial coordinates are weighted by the pressure value of each unit, as shown in the following formula:

[0031] ;

[0032] in, For the first Pressure values ​​of each sensor unit Let it be its physical position in the array coordinate system. The third is the number of effective sensor units contained within the foot pressure cluster. The fourth is the pressure distribution entropy in the plantar region, used to quantify the degree of concentration or dispersion of pressure distribution; the calculation formula is:

[0033] ;

[0034] in, This is the normalized local pressure value. This represents the number of sub-regions into which the sole of the foot is divided. A higher distribution entropy value indicates more dispersed pressure, which may reflect foot discomfort or unstable support; conversely, a lower value indicates concentrated pressure, which may be normal weight-bearing or a behavior to avoid localized pain.

[0035] The non-contact multimodal physiological sensing module is mounted on an adjustable bracket above the pet grooming station. The height of this bracket is electrically adjustable from 500 mm to 1200 mm to accommodate the chest height of pets of different sizes. The module comprises a millimeter-wave radar sensor and a hyperspectral imaging sensor, both spatially fixed in relative positions with their field of view centered on the central area of ​​the grooming station. The millimeter-wave radar sensor operates in the 60 GHz to 64 GHz frequency band, emitting frequency-modulated continuous wave signals with a bandwidth of at least 2 GHz, a range resolution better than 7.5 cm, and a velocity resolution up to 0.1 m / s.

[0036] The non-contact multimodal physiological sensor emits electromagnetic waves into the pet's chest cavity and receives reflected signals caused by slight chest wall movements. The system performs deskewing, fast Fourier transform, and phase demodulation on the echo signals to extract the micro-Doppler frequency shift components related to heartbeat and respiration. Through spectral peak detection and autocorrelation analysis, the heart rate and respiratory rate are separated, and a real-time parameter sequence with timestamps is output.

[0037] The hyperspectral imaging sensor operates within a spectral range of 500 nm to 900 nm, divided into 80 continuous spectral channels, with a spectral resolution better than 5 nm and a spatial resolution of 1280 × 1024 pixels. This sensor acquires a sequence of hyperspectral images of specific regions on a pet's body surface at a rate of 5 frames per second. The system first performs dark current correction, flat-field correction, and radiometric calibration on the raw images to obtain reflectance cube data. Subsequently, spectral reflectance values ​​are extracted from preset characteristic bands, and blood oxygen saturation parameters are calculated using the ratio method or a multiple linear regression model. For example, by calculating the reflectance ratio of the 760 nm to 805 nm bands, the influence of skin pigmentation and thickness can be effectively eliminated, revealing tissue blood oxygen levels. The system analyzes the dynamic changes in the visible light spectrum to assess the filling speed and stability of local microcirculation, serving as an auxiliary physiological indicator.

[0038] The edge computing and fusion processing unit is deployed in an industrial-grade chassis inside the washing and care workbench. It employs an ARM architecture multi-core processor and a dedicated AI acceleration chip, running a real-time operating system. This edge computing and fusion processing unit synchronously receives pressure data streams from the intelligent load-bearing sensing array and physiological parameter data streams from the non-contact multimodal physiological sensing module via a wired communication bus. The spatiotemporal alignment and data fusion engine running within the unit first timestamps all input data, using linear interpolation or spline interpolation methods to align low-frequency physiological parameters to the time axis of high-frequency pressure data, ensuring that each frame of pressure data is associated with corresponding physiological state information.

[0039] Furthermore, this spatiotemporal alignment and data fusion engine spatially maps pressure distribution data with physiological sensor data based on a pre-defined pet body model. For example, when the hyperspectral imaging sensor focuses on the left ear region, the system establishes a spatial correlation between the blood oxygenation parameter of that left ear region and the pressure distribution data of the left forefoot; when the millimeter-wave radar locks onto the midline of the chest cavity, its heart rate parameter is correlated with the total pressure value of both forefoot foes. This spatial mapping mechanism provides a structured data foundation for subsequent multimodal cross-validation and anomaly detection.

[0040] The dynamic motion interference filtering submodule is the core of this system for achieving stable calculation of physiological parameters under dynamic environments. This submodule first analyzes the macroscopic limb movement components in the millimeter-wave radar signal—by detecting low-frequency, high-energy components in the Doppler spectrum—to determine whether there are significant changes in body position or struggling behavior. This submodule monitors abrupt change patterns in the intelligent load-bearing sensing array data, including instantaneous increases in the total pressure variance and jump-like displacements of the single-foot pressure center trajectory. When any of these conditions is triggered, the system marks the current time period as a "high motion interference period."

[0041] During periods of high motion interference, the system initiates a noise suppression algorithm based on adaptive Kalman filtering. The state vector of this algorithm includes heart rate, respiratory rate, and their first derivatives, and the prediction model employs a constant velocity model. The key innovation lies in its process noise covariance matrix. With measurement noise covariance matrix It's not a fixed value, but rather dynamically adjusted based on the pet's real-time exercise intensity level. (Exercise intensity level) Determined jointly by the following formulas:

[0042] ;

[0043] in, This represents the total variance of the current pressure data. This represents the baseline variance at rest. The 3dB bandwidth of the Doppler spectrum for millimeter-wave radar. This is the reference bandwidth in the resting state; This is the weighting coefficient, with a value of 0.6. When When the activity is deemed high-intensity, the system automatically increases the intensity. The value makes the filter rely more on state predictions than on contaminated measurements; when When this is determined to be low-intensity exercise, reduce... This enhances the confidence in the measured values. The adaptive mechanism significantly improves the robustness of the filter in dynamic scenes.

[0044] The health status assessment and early warning module receives multimodal time-series data after fusion and filtering. It contains two parallel sub-modules: a weight-bearing pattern anomaly detection sub-module and a physiological state trend analysis sub-module. The weight-bearing pattern anomaly detection sub-module continuously calculates three core indicators based on real-time quadruped pressure distribution data. The first indicator is the asymmetry index of pressure distribution between the legs, defined as:

[0045] ;

[0046] in, , , , These are the total pressure values ​​for the left forefoot, right forefoot, left hindfoot, and right hindfoot, respectively. The first term represents the total pressure of all four legs. The second term represents the trajectory drift of the pressure center of a single leg per unit time, taking the maximum value among the four legs and calculating the root mean square of its Euclidean distance within a 1-second sliding window. The third term represents the time difference between the peak pressure values ​​of the two forelegs or two hind legs, calculated by detecting the local maxima of the pressure waveforms of each leg and calculating the peak time offset of the forelegs on the same side or the forelegs on the opposite side at the same position.

[0047] The load-bearing pattern anomaly detection submodule's built-in load-bearing pattern benchmark model was established through a calibration process during the initial system deployment. During calibration, operators guided at least 20 healthy pets of different breeds, weighing between 3000 grams and 25 kilograms, to remain calmly standing on the grooming table for at least 30 seconds. The system collected all pressure data and extracted the three core indicators mentioned above to form a feature vector. Subsequently, a Gaussian mixture model was used to cluster the feature vector set, identifying the healthy load-bearing pattern distribution of different groups, such as small dogs, medium dogs, large dogs, and cats. Finally, the feature value range covering 95% of the healthy samples was set as a dynamic threshold; for example, the asymmetry index threshold was 0.25, the trajectory drift threshold was 8 mm / s, and the pressure peak time difference threshold was 120 milliseconds. When any indicator continuously exceeds the corresponding threshold for more than 3 seconds, the system determines it as an abnormal load-bearing pattern, generating a level one warning signal, indicating possible limb pain, arthritis, or nerve damage.

[0048] The physiological state trend analysis submodule performs in-depth time-series analysis on heart rate, respiratory rate, and blood oxygen saturation. This submodule first calculates time-domain indices of heart rate variability, including the standard deviation (SDNN) and root mean square deviation (RMSSD) between adjacent heartbeats; it then calculates the standard deviation of respiratory rate within a 10-second window as short-term volatility; and performs a 30-second moving average on blood oxygen saturation to smooth transient noise. More importantly, this submodule embeds a physiological state prediction model based on a long short-term memory network. This model takes multi-parameter time-series data from the past 60 seconds as input and predicts the changing trends of each parameter and the probability of potential risks within the next 30 seconds. During the training phase, the model uses a large labeled dataset, including scenarios such as normal bathing / care, mild stress, moderate pain, and severe cardiopulmonary abnormalities, and outputs future risk probability values ​​(between 0 and 1).

[0049] This physiological state trend analysis submodule compares real-time parameters, short-term trends, and prediction results with preset species-specific physiological safety ranges. For example, for adult dogs, the safe upper limit for heart rate is 180 beats / minute. If the real-time heart rate reaches 190 beats / minute and shows an upward trend, while the prediction model outputs a 75% probability that the heart rate will exceed 200 beats / minute in the next 30 seconds, a level two warning is triggered. Similarly, if blood oxygen saturation continuously drops from 98% to below 94%, and the predicted trend shows further deterioration, a level two warning is also triggered. The level two warning signal indicates that the pet may be in a state of acute stress, pain attack, or cardiopulmonary compensatory imbalance.

[0050] The early warning synthesis and interaction module receives level one and level two early warning signals and processes them according to a preset rule engine. The rule logic is as follows: if only a level one early warning is present, a yellow icon and the text "Caution: Abnormal Limb Weight-Bearing" are displayed on the LED screen in front of the washing station; if only a level two early warning is present, an orange icon and the text "Caution: Abnormal Physiological State" are displayed; if both level one and level two early warnings exist simultaneously, or if any early warning lasts for more than 10 seconds, a high-risk state is determined, triggering an audible and visual alarm (a buzzer sounds at a frequency of 1000 Hz, and a red warning light flashes), and pushing detailed alarm information to associated mobile terminals via wireless LAN, including the abnormality type, specific indicator values, occurrence time, and recommended measures. All early warning events, raw data snapshots taken 5 seconds before triggering, and complete processing logs are encrypted and stored in a secure storage unit on a local solid-state drive for at least 30 days, supporting subsequent retrospective analysis and veterinary diagnostic reference.

[0051] The system adopts a modular power supply and communication architecture. The intelligent load-bearing sensing array and the non-contact multimodal physiological sensing module are uniformly powered by a 24V DC power supply inside the pet care workbench, which has overcurrent, overvoltage, and short-circuit protection functions. All sensor modules are connected to the edge computing device via an industrial-grade CAN bus or gigabit Ethernet. The communication protocol uses a custom binary format, including data type identifiers, timestamps, checksums, and data payloads, to ensure transmission efficiency and reliability. The edge computing device integrates all core processing modules and provides Wi-Fi 6 wireless access, supporting the synchronization of non-sensitive data (such as device status and usage statistics) with the cloud management platform. However, all core data related to pet health is strictly retained locally, complying with animal medical data privacy regulations.

[0052] In summary, this embodiment, through the deep integration of high-density pressure sensing, non-contact physiological monitoring, adaptive motion filtering, and multimodal health assessment, constructs an intelligent system that can stably, accurately, and proactively monitor pet health status even in real-world dynamic grooming scenarios. This system not only solves the problems of data distortion and interruption in dynamic environments using traditional methods, but also achieves early identification of pain, stress, and cardiopulmonary risks through both load-bearing and physiological dimensions, providing solid technical support for improving the safety and welfare of pet grooming.

[0053] Example 2: Based on the aforementioned examples, this example further optimizes the deployment strategy and data fusion logic of the non-contact multimodal physiological sensing module to adapt to the special needs of simultaneous grooming of multiple pets or large dog breeds. Specifically, this example adds a second set of non-contact multimodal physiological sensing modules above the grooming workbench. The two modules are respectively mounted on independently rotatable and tiltable gimbal supports, forming a dual-view collaborative monitoring architecture. Please refer to the appendix. Figure 3The diagram illustrates the multi-level interaction relationships and data flow. In this embodiment, the data flow is expanded to dual inputs, and the system needs to have target recognition and sensor scheduling capabilities.

[0054] When the system detects two or more pets on the grooming table (identified by an intelligent load-bearing sensing array that identifies more than four major pressure clusters, or by multi-target segmentation in a hyperspectral image), the edge computing and fusion processing unit initiates a multi-target tracking mode. The system first uses depth information from a hyperspectral imaging sensor or distance-angle maps from millimeter-wave radar to spatially locate and distinguish multiple individual pets. Subsequently, two sets of non-contact multimodal physiological sensing modules automatically allocate monitoring targets based on their respective field-of-view coverage: prioritizing the high-resolution sensor for larger or more active individuals, while the other covers the remaining pets. The pan-tilt mount is fine-tuned according to the pet's real-time position to ensure that the chest cavity and body surface feature areas are always within the optimal detection range of the sensors.

[0055] In this mode, the spatiotemporal alignment and data fusion engine needs to handle multi-target, multi-source heterogeneous data streams. The system establishes an independent data channel for each individual pet, including a dedicated stress distribution subarray, physiological parameter sequences, and fused state vectors. The dynamic motion interference filtering submodule also operates independently for each individual, calculating their respective motion intensity levels and applying adaptive filtering. The health status assessment and early warning module generates independent level one and level two early warning signals for each individual. The early warning synthesis and interaction module presents the health status of each pet on the display interface in a split-screen or multi-tab format, and highlights the corresponding area when any pet triggers a high-risk early warning.

[0056] For large dog breeds (weighing over 25 kg), this embodiment strengthens the range and structure of the intelligent load-bearing sensing array. The maximum range of the pressure sensor unit has been increased from 50 kPa to 100 kPa, and the sensing film substrate uses reinforced polyimide composite material to withstand higher pressure without plastic deformation. The physical dimensions of the array have also been expanded to 1200 mm in length and 800 mm in width to ensure that all four paws of a large dog remain fully within the sensing area when extended. The load-bearing mode benchmark model has also been supplemented with calibration data for large dogs, and the threshold ranges of various core indicators have been recalculated. For example, the asymmetry index threshold has been relaxed to 0.30 to accommodate their natural load-bearing asymmetry.

[0057] This embodiment significantly improves the system's applicability and robustness by introducing multi-target collaborative monitoring and a large dog adaptation mechanism, enabling it to serve complex grooming environments with multiple scenarios and breeds, such as pet grooming chain stores and animal hospitals, while maintaining accurate monitoring of the health status of each individual pet.

Claims

1. A pet health AI monitoring system based on intelligent sensors during the grooming process, characterized in that, include: The intelligent load-bearing sensing array, integrated inside the washing and care workbench, consists of high-density distributed pressure sensor units arranged in a matrix. It is used to collect vertical pressure data of the pet's paws at a sampling frequency of no less than 100 Hz, generate a pressure distribution heat map, and calculate the pressure center coordinates, total pressure value, and pressure distribution entropy of each paw in real time. A non-contact multimodal physiological sensing module is installed above the washing and care workbench, including a millimeter-wave radar sensor and a hyperspectral imaging sensor; The edge computing and fusion processing unit is deployed locally on the washing and care workbench to receive real-time pressure data streams from the intelligent load-bearing sensing array and real-time physiological parameter data streams from the non-contact multimodal physiological sensing module. The health status assessment and early warning module is used to receive multimodal time-series data after fusion and filtering. The health status assessment and early warning module includes a load-bearing mode abnormality detection submodule and a physiological state trend analysis submodule. The early warning synthesis and interaction module is used to receive the first-level and second-level early warning signals from the health status assessment and early warning module; The early warning synthesis and interaction module is equipped with a rule engine, which performs synthesis processing according to the type, level and concurrency status of the early warning signal and according to a preset priority logic. Upon receiving a Level 1 or Level 2 warning signal, a highest-level audible and visual alarm command is generated and sent via a wireless communication interface to the associated mobile terminal, containing detailed alarm information including specific abnormal indicators and recommended countermeasures.

2. The pet health AI monitoring system based on intelligent sensors during the grooming process according to claim 1, characterized in that, The hyperspectral imaging sensor is used to acquire hyperspectral image sequences of specific areas on the pet's body surface. By analyzing the changes in spectral reflectance characteristics of the skin or mucous membrane area in the visible and near-infrared bands, physiological parameters related to blood oxygen saturation and local microcirculation status are calculated. The edge computing and fusion processing unit includes a spatiotemporal alignment and data fusion engine and a dynamic motion interference filtering submodule; The spatiotemporal alignment and data fusion engine synchronizes asynchronous data from different sensors based on a unified timestamp, and spatially maps pressure distribution data and physiological sensor data according to a preset pet body model. The dynamic motion interference filtering submodule analyzes the macroscopic limb motion components and abrupt change patterns in pressure distribution data in millimeter-wave radar signals to identify and mark the periods when the pet struggles violently or changes its body position significantly. Based on the physiological parameter data during these periods, it activates a noise suppression algorithm based on adaptive Kalman filtering to filter out motion artifacts.

3. The pet health AI monitoring system based on intelligent sensors during the grooming process according to claim 1, characterized in that, The load-bearing mode abnormality detection submodule calculates the asymmetry index of pressure distribution between each foot, the trajectory drift of the single foot pressure center per unit time, and the time difference of the pressure peak of the two forefeet or two hind feet based on real-time four-leg pressure distribution data. It also compares the real-time calculated core indicators with the normal range threshold through the built-in load-bearing mode benchmark model. When any indicator exceeds the threshold range for more than 3 seconds, a first-level warning signal indicating that the limb may be abnormal is generated. The physiological state trend analysis submodule performs time series analysis on the filtered heart rate, respiratory rate, and blood oxygen saturation parameters, calculates the time-domain index of heart rate variability, the short-term volatility of respiratory rate, and the moving average of blood oxygen saturation, and uses a physiological state prediction model based on a long short-term memory network to predict the trend of physiological parameters over the next 30 seconds. The real-time parameters, their short-term trends, and the predicted trends are compared with a preset species-specific physiological safety range. When a sharp increase in heart rate or respiratory rate, a decreasing trend in blood oxygen saturation, or a prediction model outputting a future risk probability exceeding 70% is detected, a secondary warning signal indicating an abnormal physiological state is generated.

4. The pet health AI monitoring system based on intelligent sensors during the grooming process according to claim 2, characterized in that, The adaptive Kalman filtering algorithm in the dynamic motion interference filtering submodule dynamically adjusts its process noise covariance matrix and measurement noise covariance matrix according to the real-time identified pet motion intensity level. The exercise intensity level is determined by jointly analyzing the total variance of the pressure sensor array data and the Doppler spectral width of the millimeter-wave radar signal. When the intensity level of the motion is high, the algorithm automatically increases the process noise covariance, making the filter more confident in the predicted values. When the motion intensity level is low, the process noise covariance is reduced, making the filter more confident in the measurement values.

5. The pet health AI monitoring system based on intelligent sensors during the grooming process according to claim 3, characterized in that, The process of establishing the load-bearing mode benchmark model is as follows: In the initial stage of system deployment or calibration, a large amount of plantar pressure data of healthy pets of different breeds and sizes under calm washing conditions is collected to form a training dataset. For each measurement in the dataset, extract the core metrics as feature vectors; Cluster analysis of the feature vector set was performed using a Gaussian mixture model to determine the feature distribution of various healthy load-bearing modes. The feature value range covering 95% of healthy samples is set as the dynamic threshold range of various core indicators, and an abnormal load-bearing mode detection submodule is embedded.

6. The pet health AI monitoring system based on intelligent sensors during the grooming process according to claim 1, characterized in that, The millimeter-wave radar sensor operates in the 60 GHz to 64 GHz frequency band, with a bandwidth of not less than 2 GHz. The hyperspectral imaging sensor operates in the spectral range of 500 nanometers to 900 nanometers, with a spectral resolution better than 5 nanometers and a spatial resolution of no less than 1 million pixels. The calculation process for the pressure center coordinates is as follows: the pressure values ​​of each pressure sensor unit are used as weights to calculate a weighted average of their physical positions in the array coordinate system. The calculation process of the pressure distribution entropy in the sole region is as follows: the sole region is divided into multiple sub-regions, and the information entropy is calculated based on the normalized local pressure values.

7. The pet health AI monitoring system based on intelligent sensors during the grooming process according to claim 1, characterized in that, The system adopts a modular power supply and communication architecture; The intelligent load-bearing sensing array and the non-contact multimodal physiological sensing module are powered by a unified DC power supply inside the washing and care workbench and are connected to the edge computing and fusion processing unit through a wired communication bus. An edge computing and fusion processing unit, a health status assessment and early warning module, and an early warning synthesis and interaction module are integrated into an industrial-grade edge computing device, which provides wireless local area network access functionality.

8. The pet health AI monitoring system based on intelligent sensors during the grooming process according to claim 1, characterized in that, The early warning synthesis and interaction module is also used for: When only a Level 1 warning signal is received, a prompt message indicating abnormal limb weight-bearing is generated; When only a Level 2 warning signal is received, a prompt message indicating an abnormal physiological state is generated; All warning events, raw data snapshots, and processing logs are stored in a local secure storage unit.

9. The pet health AI monitoring system based on intelligent sensors during the grooming process according to claim 1, characterized in that, The physiological state trend analysis submodule uses a physiological state prediction model based on long short-term memory networks, taking multi-parameter time-series data from the past 60 seconds as input, to predict the changing trends of each parameter and the probability of potential risks within the next 30 seconds. The non-contact multimodal physiological sensing module is mounted on an adjustable bracket, the height of which can be electrically adjusted within the range of 500 mm to 1200 mm.

10. The pet health AI monitoring system based on intelligent sensors during the grooming process according to claim 1, characterized in that, In the intelligent load-bearing sensing array, the sensing area of ​​each pressure sensor unit is 10 mm × 10 mm, and the center-to-center distance between adjacent units is 15 mm.

Images