Animal unbound physiological monitoring system based on pressure signals

Abstract

The invention discloses an animal non-constraint physiological monitoring system based on pressure signals, which comprises a pressure signal acquisition module, a signal processing module, a storage module, a communication module and a power supply module which are respectively connected to a processor module, the signal processing module specifically comprises a preprocessing unit, a basic feature extraction unit, a pose feature extraction unit, a physiological parameter detection unit, a sleep monitoring unit, an instantaneous health event detection unit, an emotion evaluation unit, a safety monitoring unit and a summarizing unit so as to monitor the heart rate, breath, sleep, postures, emotions and safety of a target animal. According to the invention, unbound physiological monitoring is carried out on the animal, the method is more friendly to the animal, the monitoring mode is more suitable for the animal, and the monitoring result is more accurate.

Description

Technical Field

[0001] This invention relates to animal monitoring technology, and more particularly to an animal non-restraint physiological monitoring system based on pressure signals. Background Technology

[0002] With the increasing proportion of pet ownership in cities, the health management of family pets such as dogs and cats is receiving more and more attention. Current practices largely rely on the owner's subjective observation; while common wearable monitoring products (such as collars and harnesses) can record heart rate, respiration, or body temperature, prolonged wear can cause a feeling of constriction or irritation to the fur, making them unsuitable for long-term continuous monitoring in resting environments. Therefore, there is a market demand for non-wearable, low-invasive pet vital sign monitoring devices that can operate continuously for extended periods.

[0003] Flexible piezoelectric polymer materials (such as polyvinylidene fluoride (PVDF) and its copolymers) possess excellent flexibility and biocompatibility, and are highly sensitive to time-varying mechanical stress / strain (compression), capable of converting minute vibrations caused by respiration, heartbeat, and body movement into electrical signals. Related research and product applications in the human field have demonstrated that integrating flexible piezoelectric sensors with carriers such as pillows, mattresses, or seat cushions can monitor respiration, heart rate, and even sleep states based on compression signals without the need for electrodes or patches, and can be linked to mobile terminals via wireless communication. However, animal scenarios are not simply equivalent to human scenarios. Directly applying human parameters and structures will result in systematic mismatches, specifically in physiological, postural, sleep structure, behavioral differences, and hygiene aspects. Using human-specific structures or methods can easily lead to missed detections or aliasing, resulting in inaccurate monitoring results. Furthermore, pets frequently switch between lying down, curled up, and side-lying positions, and their chest and abdomen are not always centered; their fur forms soft barriers, and their tails and limbs introduce localized impacts and movement artifacts. Animals exhibit polyphasic short sleep, short periods of arousal, and frequent patrols; human sleep scoring logic based on "long-term stability" may falsely report fragmentation or arousal.

[0004] Currently, non-wearable vital sign monitoring for pets is still in its early stages. Pet monitoring mostly relies on adhesive / wearable solutions such as accelerometers, PPG, or ECG electrodes, which have problems such as easily causing stress, hair obstruction, and poor long-term compliance, making them unfriendly to pets. Summary of the Invention

[0005] In view of the problems existing in the prior art, the purpose of this invention is to provide an animal non-restraint physiological monitoring system that is more accurate and suitable for monitoring animals.

[0006] To achieve the above-mentioned objectives, the present invention provides the following technical solution:

[0007] A pressure signal-based non-restraint physiological monitoring system for animals includes a pressure signal acquisition module, a signal processing module, a storage module, a communication module, and a power supply module, all connected to a processor module. The signal processing module specifically includes:

[0008] The preprocessing unit is used to select the corresponding filtering settings according to the target animal species and body size, and extract the respiratory cycle sequence and cardiac cycle sequence from the compression signal according to the corresponding filtering settings, and identify canine panting events or feline purring events, and perform noise reduction based on the identified events;

[0009] The basic feature extraction unit is used to calculate the time-domain features, frequency-domain features, time-frequency-domain features, and panting index / gurgling index of the target animal based on the respiratory cycle sequence and the cardiac cycle sequence.

[0010] The pose feature extraction unit is used to generate a local response map to mark the energy at each position based on the position of each pressure sensor in the pressure signal acquisition module and the energy of the acquired signal. Based on the local response map, the animal's thoracic and abdominal regions are identified as effective acquisition areas, as well as the center-periphery ratio and cross-unit coherence.

[0011] The physiological parameter detection unit is used to calculate the respiratory rate and heart rate of the target animal based on the time-domain and frequency-domain characteristics of the respiratory cycle sequence and cardiac cycle sequence of the effective acquisition area.

[0012] The sleep monitoring unit is used to divide the respiratory cycle sequence and heart cycle sequence into several short-term analysis windows, and extract the sleep state segment and depth based on the time domain and frequency domain characteristics of the short-term analysis windows, and determine the sleep posture based on the center-periphery ratio and cross-unit coherence.

[0013] The instantaneous health event detection unit is used to identify instantaneous health events based on time-frequency domain features;

[0014] An emotion assessment unit is used to identify the current emotion of a target animal based on time-domain features, panting index / gurgling index, and heart rate.

[0015] The safety monitoring unit is used to monitor dynamic evidence in real time to indicate whether the target animal is present or not, and to calculate the duration of presence based on the dynamic evidence.

[0016] The aggregation unit is used to send the results calculated by each unit to the processor module, so that the processor module can send the data to an external terminal and / or cloud server through the communication module for data interaction.

[0017] Furthermore, the pressure signal acquisition module specifically includes:

[0018] A carrier for the target animal to contact or lie on;

[0019] Several pressure sensors are installed at different locations within the carrier to collect pressure signals generated by the interaction between the target animal and the carrier when in-situ evidence is detected.

[0020] Furthermore, the preprocessing unit specifically includes:

[0021] The canine signal preprocessing subunit, when the user-preset target animal species is canine, searches for the corresponding cardiac frequency range and median window duration from the body size-filter parameter mapping table based on the user-preset target animal body size. Then, it extracts the respiratory cycle sequence from the compression signal using bandpass filtering within the canine respiratory frequency range, and extracts the cardiac cycle sequence from the compression signal using bandpass filtering within the searched cardiac frequency range. Next, it performs baseline suppression on the respiratory and cardiac cycle sequences, achieved through sliding median subtraction using the searched median window duration and first-order high-pass filtering. Finally, it calculates the short-time energy of each segment in the respiratory and cardiac cycle sequences, and uses adaptive thresholding to identify and remove noise. Segments with short-time energy greater than a preset threshold in the surrounding segments are marked as tail / limb impact events rather than noise. Segments with short-time energy rising to a preset energy threshold, with the main energy concentrated in 2–8 Hz and dominated by the central region, are marked as canine panting events rather than noise.

[0022] The feline signal preprocessing subunit, when the user-preset target animal species is feline, finds the corresponding median window duration from the body size-filter parameter mapping table based on the user-preset target animal body size. Then, it extracts the respiratory cycle sequence from the pressure signal using bandpass filtering within the feline respiratory frequency range, and extracts the cardiac cycle sequence from the pressure signal using bandpass filtering within the feline cardiac frequency range. Next, it performs baseline suppression on the respiratory and cardiac cycle sequences by sliding median subtraction based on the found median window duration and first-order high-pass filtering. Then, it identifies segments in the cardiac cycle sequence where the signal energy in the 25–50 Hz frequency band continuously increases and exceeds a preset threshold as feline purring events, and filters out these events from the cardiac cycle sequence. Finally, it calculates the short-time energy of each segment of the respiratory and cardiac cycle sequences and uses adaptive thresholding to identify and remove noise from each component.

[0023] Furthermore, the feature extraction unit specifically includes:

[0024] The temporal feature extraction subunit is used to extract the corresponding main peak position, coefficient of variation, stability, zero crossover rate, mean and variance of the envelope, and activity index based on short-time energy from the respiratory cycle sequence and the cardiac cycle sequence, respectively.

[0025] The frequency domain feature extraction subunit is used to extract the corresponding spectral peak frequency, bandwidth, and harmonic energy ratio from the respiratory cycle sequence and the cardiac cycle sequence, respectively.

[0026] The time-frequency domain feature extraction subunit is used to calculate the energy distribution, instantaneous frequency, and Teager energy operator for respiratory cycle sequences and cardiac cycle sequences, respectively.

[0027] The panting index extraction subunit is used to calculate the ratio of the energy of the panting event frequency band to the total energy of the respiratory cycle sequence when the target animal is a canine, and use it as the panting index.

[0028] The Gurgle Index Extraction Subunit is used to calculate the ratio of the energy of the feline gurgle event frequency band to the total energy of the heartbeat cycle sequence when the target animal is a feline, and this ratio is used as the Gurgle Index.

[0029] Furthermore, the pose feature extraction unit specifically includes:

[0030] The response map generation subunit is used to generate a local response map that marks the energy at each location by mapping the pressure signal energy of each pressure sensor in the pressure signal acquisition module to its physical location.

[0031] The response map parameter calculation subunit is used to calculate the sum of energy of the pressure sensors in the predefined central region divided by the sum of energy in the surrounding region, as the center-periphery ratio, based on the local response map; calculate the coherence function of different pressure sensors in the respiratory or cardiac frequency band, as the cross-unit coherence; calculate the average point of the position-weighted calculation with the energy of each pressure sensor as the weight, as the response centroid; and calculate the principal component direction of the energy distribution, as the principal axis.

[0032] The effective sampling region extraction sub-unit is used to obtain the position of the pressure sensor with cross-unit coherence higher than the threshold along the main axis with the response centroid as the center of the animal's chest and abdomen as the center, and this position is taken as the effective sampling region. When a pressure sensor with energy significantly higher than the effective sampling region but located in the edge region appears in the local response map, and the coherence of the pressure sensor with the pressure signal of the pressure sensor in the effective sampling region is lower than the threshold, it is identified as an artifact and edge interference from the limbs or tail and is not taken as the effective sampling region.

[0033] Furthermore, the physiological parameter detection unit specifically includes:

[0034] The respiratory rate estimation subunit is used to calculate the respiratory cycle and then the respiratory rate based on the envelope signal and main peak position of the respiratory cycle sequence in the effective sampling area.

[0035] The heart rate estimation subunit is used to calculate the heart rate based on the envelope signal and main peak position of the heart rate cycle sequence in the effective sampling area.

[0036] The signal verification subunit is used to determine whether the respiratory rate or heart rate simultaneously meets the following criteria: the activity index is lower than the exercise threshold, the coefficient of variation and stability are better than the set standard, the bandwidth of the main peak of the signal spectrum is lower than the threshold and the harmonic energy ratio fluctuation value is less than the threshold. If so, the estimation is determined to be accurate; otherwise, the estimation is determined to be incorrect and the estimation is repeated.

[0037] Furthermore, the sleep detection unit specifically includes:

[0038] The signal division subunit is used to divide the respiratory cycle sequence and the cardiac cycle sequence into continuous short-time analysis windows, respectively.

[0039] The sleep judgment subunit checks whether each short-term analysis window simultaneously meets the following criteria: the activity index of the respiratory cycle sequence and the cardiac cycle sequence of the corresponding window are both below the motion threshold, the zero crossover rate is below the threshold, and the stability is above the threshold. If so, the short-term analysis window is determined to be a potential sleep window. When the number of consecutive potential sleep windows exceeds a preset minimum duration threshold, the total events of the potential sleep window are defined as a continuous sleep state segment. The remaining potential sleep windows are classified as wakefulness state segments.

[0040] The sleep level assessment subunit is used to evaluate the depth and stability of breathing based on the mean and variance of the respiratory cycle sequence envelope. A continuous sleep state segment with a mean higher than the threshold and a variance lower than the threshold is a deep sleep stage, while the opposite is a light sleep stage or a restless stage.

[0041] The sleep posture determination subunit is used to determine that the animal is curled up when the center-periphery ratio is greater than the threshold, and otherwise is stretched out. When the cross-unit coherence is greater than the threshold, it is determined that the animal is rolling over as a whole, and otherwise is determined to be local limb twitching.

[0042] Furthermore, the instantaneous health event detection unit specifically includes:

[0043] The instantaneous health event identification subunit is used to mark the current time as the occurrence time of an instantaneous health event when the output value of the Teager energy operator in the time-frequency domain features exceeds a threshold.

[0044] The instantaneous health event analysis subunit is used to acquire signals within a preset time window before and after the occurrence time, generate a wavelet energy distribution time spectrum using continuous wavelet transform, classify events by matching the wavelet energy distribution time spectrum with a preset event energy fingerprint, and calculate the instantaneous frequency of the signal within the preset time window to determine the accuracy of the instantaneous health event classification.

[0045] Furthermore, the emotion assessment unit specifically includes:

[0046] The dynamic baseline calculation subunit is used to calculate the dynamic baseline based on heart rate variability entropy.

[0047] The emotion recognition subunit determines the current emotion as excitement / exercise if both the panting index and activity index are above a threshold when the target animal species is canine, and as heat stress when both are below a threshold.

[0048] Furthermore, the safety monitoring unit specifically includes:

[0049] The dynamic evidence monitoring subunit is used to monitor dynamic evidence in real time. The dynamic evidence is of three types: first, upper / lower boundary events, i.e., energy transitions that occur within a very short period of time; second, periodic physiological wave evidence, i.e., the stable existence of relevant narrow band spectral lines or envelopes of respiratory cycle sequences or cardiac cycle sequences over a continuous period of time; and third, low-frequency energy duty cycle, i.e., the continuous overthreshold of low-frequency band energy relative to the environmental baseline.

[0050] The in-service duration calculation subunit is used to update the in-service probability with a fixed step size. When any dynamic evidence is established, the in-service probability is increased, and when dynamic evidence is missing, it is decreased at a set rate. When the in-service probability reaches the upper threshold, it is determined to be in-service, and when it is below the lower threshold, it is determined to be out of service. The in-service duration is accumulated to obtain the total in-service duration.

[0051] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention specifically incorporates canine panting and feline purring vibration recognition based on animal characteristics, resulting in more accurate monitoring results. Furthermore, considering the active nature of animals, this invention employs adaptive recognition of chest and abdominal positions, improving the SNR of chest and abdominal micro-vibrations and suppressing tail / limb artifacts under common animal lying / curling / side-lying and soft fur isolation conditions. Overall, the monitoring is more targeted and animal-friendly, achieving stable, long-term, and scalable continuous monitoring of vital signs without increasing animal burden, demonstrating significant practical value and industrialization prospects. Attached Figure Description

[0052] Figure 1 This is a structural diagram of the animal non-restraint physiological monitoring system based on pressure signals provided in an embodiment of the present invention;

[0053] Figure 2 This is a structural diagram of the signal processing module provided in an embodiment of the present invention. Detailed Implementation

[0054] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0055] This invention provides an animal non-restraint physiological monitoring system based on pressure signals, such as... Figure 1 As shown, it includes a pressure signal acquisition module, a signal processing module, a storage module, a communication module, and a power supply module, which are respectively connected to the processor module.

[0056] The processor module is connected to each other and serves as the processing center for each module, controlling each module and converting data.

[0057] The storage module is controlled by the processor module and is used to store pressure signals, processing data and / or monitoring results, supporting cyclic overwriting and breakpoint caching.

[0058] The communication module is controlled by the processor module and is used to exchange data between the monitoring results and an external terminal and / or a cloud server; the communication method can be Wi-Fi, Bluetooth, cellular or a combination thereof.

[0059] The power module is controlled by the processor module and is used to supply power to each module and to achieve low power management.

[0060] The pressure signal acquisition module specifically includes a carrier and several pressure sensors. The carrier is used for the target animal to contact or lie down. The carrier can be any one or a combination of a mat, nest, floor mat, mattress, lying platform, transport box liner, cage bedding, or wearable / spreadable flexible carrier. The pressure sensors are located at different parts of the carrier and are used to collect pressure signals (i.e., time-varying mechanical stress / strain signals) generated by the interaction between the target animal and the carrier when in-situ evidence is detected. The pressure sensor module and the carrier are preferably indirectly in contact through a mechanical coupling structure, but direct contact is also possible. The mechanical coupling structure is a mechanical coupling layer, which can be a layer of highly elastic foam material, an elastomer material, or a gel material. This mechanical coupling layer forms at least two different mechanical impedances in the central and peripheral regions of the carrier. The compressive modulus and / or thickness and / or porosity of the central region are different from those of the peripheral region, so as to improve the mechanical transmission gain of the animal's chest and abdomen micro-vibrations and attenuate the local impact and swinging interference generated by the animal's tail in the micro-vibration frequency band of 0.05–20 Hz. The pressure sensor can be composed of one or more pressure sensor units forming a module, and its geometric arrangement is not limited, and can be a regular matrix, strip, ring, partition, or non-uniform distribution. The pressure sensor is preferably a piezoelectric thin film sensor, including but not limited to PVDF flexible piezoelectric thin film and piezoelectric polymer / ceramic composite film; but not limited to these types of devices, any device that is sensitive to dynamic mechanical stress and outputs an electrical signal can be used.

[0061] For example, in this case, the carrier adopts a pet mat shape, preferably with external dimensions of 700 mm × 500 mm and a total thickness of 25–35 mm. The carrier includes, from top to bottom, a fabric layer, a mechanical coupling layer, a pressure sensor module, an insulating base layer, and a bottom buffer layer; an electronic compartment is set at the edge to accommodate the processor module, power module, and interfaces with the storage module and communication module. The overall structure is easy to disassemble, wash, replace, and maintain.

[0062] Fabric layer (1–2 mm): Waterproof and breathable fabric, preferably >150D polyester or nylon, with a water-repellent coating (DWR), breathable pore size 50–200 μm, hydrostatic pressure ≥3 kPa; the fabric has a detachable zippered cover that can be machine washed after removal.

[0063] Mechanical coupling layer (8–15 mm): Open-cell PU foam, density 25–32 kg / m³, compressive modulus 35–55 kPa, pore size 0.3–1.2 mm. This layer is used to diffuse and conduct chest and abdominal undulations and cardiac micro-vibrations in the plane downwards. A sealing border (EVA / ribbon) is preferably provided around the foam to limit lateral slippage.

[0064] Pressure sensor module: Preferably a PVDF flexible piezoelectric film with a thickness of 28–52 μm; polymer / ceramic composite films or other devices sensitive to dynamic stress can also be used. The shape can be a 40×40 mm square unit or a 40×300 mm strip unit. The preferred layout is a 4×6 regular matrix of 24 units with a center-to-center spacing of 70–90 mm; or 6 longitudinal strips with a strip spacing of 70–80 mm. Electrodes and leads are PVDF double-sided metallized (Al / Ag), with edges connected using silver paste + nickel-copper plated FPC pads; units are connected in series and parallel via FPC flexible flat cables to converge at the edge connector of the electronic compartment. Units are bonded to the underlying substrate with weakly adhesive acrylic dots, and soft corner positioning tabs are used for restraint to prevent hard spots.

[0065] Insulating substrate (0.2–0.5 mm): PET or PI sheet, which serves as electrical insulation and wiring support, and also forms a structural support surface; a thin layer of conductive coating can be sprayed on the back of the substrate to connect with the ground wire to suppress static electricity and electromagnetic interference.

[0066] Bottom cushioning layer (5–10 mm): EVA or rubber foam, Shore C 20–35, used to absorb ground-reflected vibrations and improve stability; bottom surface covered with anti-slip mesh or granules.

[0067] The electronic compartment is located in the middle or corner of the short side of the pad, with dimensions of approximately 95×60×18 mm. The outer shell is made of ABS / PC flame-retardant material, with double fixation via buckles and magnets for easy disassembly and assembly. The sensing layer is connected to the electronic compartment via a board-to-board connector (0.5 mm pitch, 10–40 pins); a Type-C charging / maintenance interface (waterproof plug seal) is reserved for external use. The sensing layer (3) is separate from the electronic compartment; the electronic compartment can be removed for cleaning, and the entire pad can be washed with water (the sensing layer and wiring are located in the waterproof inner liner). The electronic compartment shell is equipped with a silicone ring to achieve IPX4 splash resistance; foam pads are added inside the shell at high pressure points for stress relief.

[0068] The sensor leads use shielded flexible wires, with the shielding layer connected to system ground at one end to avoid loops; the sensor front-end and the subsequent digital section inside the electronic compartment are isolated by grounding zones. A TVS / varistor is connected in series between the housing grounding point and the system protective ground; the ESD design level meets the requirements of 4 kV for contact and 8 kV for air.

[0069] like Figure 2 As shown, the signal processing module specifically includes:

[0070] The preprocessing unit is used to select the corresponding filtering settings according to the target animal species and body size, and extract the respiratory cycle sequence and cardiac cycle sequence from the compression signal according to the corresponding filtering settings, and identify canine panting events or feline purring events, and perform noise reduction based on the identified events;

[0071] The basic feature extraction unit is used to calculate the time-domain features, frequency-domain features, time-frequency-domain features, and panting index / gurgling index of the target animal based on the respiratory cycle sequence and the cardiac cycle sequence.

[0072] The pose feature extraction unit is used to generate a local response map to mark the energy at each position based on the position of each pressure sensor in the pressure signal acquisition module and the energy of the acquired signal. Based on the local response map, the animal's thoracic and abdominal regions are identified as effective acquisition areas, as well as the center-periphery ratio and cross-unit coherence.

[0073] The physiological parameter detection unit is used to calculate the respiratory rate and heart rate of the target animal based on the time-domain and frequency-domain characteristics of the respiratory cycle sequence and cardiac cycle sequence of the effective acquisition area.

[0074] The sleep monitoring unit is used to divide the respiratory cycle sequence and heart cycle sequence into several short-term analysis windows, and extract the sleep state segment and depth based on the time domain and frequency domain characteristics of the short-term analysis windows, and determine the sleep posture based on the center-periphery ratio and cross-unit coherence.

[0075] The instantaneous health event detection unit is used to identify instantaneous health events based on time-frequency domain features;

[0076] An emotion assessment unit is used to identify the current emotion of a target animal based on time-domain features, panting index / gurgling index, and heart rate.

[0077] The safety monitoring unit is used to monitor dynamic evidence in real time to indicate whether the target animal is present or not, and to calculate the duration of presence based on the dynamic evidence.

[0078] The aggregation unit is used to send the monitoring results calculated by each unit to the processor module, so that the processor module can send the data to an external terminal and / or cloud server through the communication module for data interaction.

[0079] Each unit will be described in detail below.

[0080] The preprocessing unit specifically includes:

[0081] The canine signal preprocessing subunit, when the user-preset target animal species is canine, searches for the corresponding heart rate range and median window duration (preferably 5–15 seconds, with the latter shortened for smaller dogs) from a size-filter parameter mapping table (where the heart rate range for large dogs is preferably 5–15 Hz, and for small dogs it is preferably 8–28 Hz). Then, it extracts the respiratory cycle sequence from the pressure signal using a bandpass filter within the canine respiratory rate range (0.05–1.0 Hz), and extracts the heart rate cycle sequence from the pressure signal using a bandpass filter within the searched heart rate range. Next, it performs baseline suppression on the respiratory and heart rate cycle sequences, applying a sliding median subtraction using the searched median window duration and a first-order high-pass filter (preferably 0.05 Hz). The process involves calculating the short-time energy of each segment in the respiratory and cardiac cycle sequences (using a 1-second window and a 0.25-second step), and then using an adaptive threshold (e.g., mean plus 3 times the standard deviation) to identify and remove noise. Segments with short-time energy greater than a preset threshold in the surrounding segments are marked as tail / limb impact events rather than noise. Segments with short-time energy increased to a preset energy threshold and whose main energy is concentrated in 2–8 Hz and dominated by the central region are marked as canine panting events rather than noise. This method can reduce the impact of gross movements, jumping, or external impacts on the estimation of physiological parameters.

[0082] The feline signal preprocessing subunit, when the user-preset target animal species is feline, searches for the corresponding median window duration (preferably 5–15 seconds, with the window length shortened to accommodate its smaller size and higher heart rate) from the body size-filter parameter mapping table based on the user-preset target animal body size. Then, it extracts the respiratory cycle sequence from the pressure signal using a bandpass filter within the feline respiratory frequency range (0.05–1.2 Hz) and the cardiac cycle sequence from the pressure signal using a bandpass filter within the feline cardiac frequency range (8–28 Hz). Next, it performs baseline suppression on the respiratory and cardiac cycle sequences by sliding median subtraction based on the found median window duration and a first-order high-pass filter (preferably 0.05 Hz). Finally, it extracts the cardiac cycle sequence from the range of 25–50 Hz. The segment where the signal energy in the Hz band continuously increases and exceeds the preset threshold is identified as a feline purring event, and the feline purring event is filtered out from the cardiac cycle sequence (specifically, it can be implemented as a dynamic notch filter); finally, the short-time energy of each segment of the respiratory cycle sequence and cardiac cycle sequence is calculated, and the noise in each component is identified and removed using an adaptive threshold.

[0083] The feature extraction unit extracts multiple features from the compressive signal for physiological parameter estimation and in-situ determination to form feature data; the features take into account differences in animal species and body size, and set specific measures for high-frequency panting in dogs and purring vibrations in felines, specifically including:

[0084] The temporal feature extraction subunit is used to extract the corresponding main peak position, coefficient of variation, stability, zero crossover rate, mean and variance of the envelope, and activity index based on short-time energy from the respiratory cycle sequence and the cardiac cycle sequence, respectively.

[0085] The frequency domain feature extraction subunit is used to extract the corresponding spectral peak frequency, bandwidth, and harmonic energy ratio from the respiratory cycle sequence and the cardiac cycle sequence, respectively.

[0086] The time-frequency domain feature extraction subunit is used to calculate the wavelet (or short-time Fourier) energy distribution, instantaneous frequency, and Teager energy operator for respiratory cycle sequences and cardiac cycle sequences, respectively.

[0087] The panting index extraction subunit is used to calculate the ratio of the energy of the panting event frequency band to the total energy of the respiratory cycle sequence when the target animal is a canine, and use it as the panting index.

[0088] The Gurgle Index Extraction Subunit is used to calculate the ratio of the energy of the feline gurgle event frequency band to the total energy of the heartbeat cycle sequence when the target animal is a feline, and this ratio is used as the Gurgle Index.

[0089] The pose feature extraction unit is used to calculate spatial features to determine the animal's position and posture, specifically including:

[0090] The response map generation subunit is used to generate a local response map that marks the energy at each location by mapping the pressure signal energy of each pressure sensor in the pressure signal acquisition module to its physical location.

[0091] The response map parameter calculation subunit is used to calculate the sum of energy of the pressure sensors in the predefined central region divided by the sum of energy in the surrounding region, as the center-periphery ratio, based on the local response map; calculate the coherence function of different pressure sensors in the respiratory or cardiac frequency band, as the cross-unit coherence; calculate the average point of the position-weighted calculation with the energy of each pressure sensor as the weight, as the response centroid; and calculate the principal component direction of the energy distribution, as the principal axis.

[0092] The effective sampling region extraction sub-unit is used to obtain the position of the pressure sensor with cross-unit coherence higher than the threshold along the main axis with the response centroid as the center of the animal's chest and abdomen as the center. This position is then used as the effective sampling region (ROI). When a pressure sensor with energy significantly higher than that of the effective sampling region but located in the edge region appears in the local response map, and the coherence of the pressure sensor with the pressure signal of the pressure sensor in the effective sampling region is lower than the threshold, it is identified as an artifact and edge interference from the limbs or tail and is not used as an effective sampling region.

[0093] Furthermore, to balance the immediacy of the response with the stability of the state assessment, the aforementioned features employ a dual-scale time window analysis strategy. A short window (preferably 10–30 s) is used to calculate features that require rapid updates to support the estimation of instantaneous heart rate and respiratory rate; while a long window (preferably 1–5 min) is used to calculate long-term statistics of features to support robust determination of slow-changing trends such as sleep stages, in-situ status, and emotional arousal levels.

[0094] The physiological parameter detection unit, based on the complete feature dataset generated in the preceding steps and covering the time domain, frequency domain, time-frequency domain, and spatial dimensions, performs a comprehensive and in-depth estimation of heart rate and respiration, specifically including:

[0095] The respiratory rate estimation subunit is used to apply Fast Fourier Transform (FFT) or autocorrelation analysis to the envelope signal of the respiratory cycle sequence based on the effective sampling area to calculate the respiratory cycle and then the respiratory rate by finding the main peak position within the search interval of 0.6–6 s.

[0096] The heart rate estimation subunit is used to apply Fast Fourier Transform (FFT) or autocorrelation analysis to the envelope signal of the heart rate cycle sequence based on the effective sampling area, so as to calculate the heart rate by finding the main peak position within the search interval of 0.6–6 s;

[0097] The signal verification subunit is used to determine whether the respiratory rate or heart rate simultaneously meets the following criteria: the activity index is lower than the exercise threshold, the coefficient of variation and stability are better than the set standard, the bandwidth of the main peak of the signal spectrum is lower than the threshold and the harmonic energy ratio fluctuation value is less than the threshold. If so, the estimation is determined to be accurate; otherwise, the estimation is determined to be incorrect and the estimation is re-performed, thereby effectively filtering out misjudgments caused by shallow breathing or exercise interference.

[0098] The sleep detection unit performs a comprehensive and in-depth estimation of sleep states based on the complete feature dataset generated in the preceding steps, covering the time domain, frequency domain, time-frequency domain, and spatial dimensions. To ensure the accuracy of all analyses, this unit prioritizes signal processing within the effective sampling area (ROI) determined by spatial features. This aims to suppress motion artifacts introduced from non-thoracic and abdominal regions such as limbs or tails at the source, thereby providing high signal-to-noise ratio input for all subsequent analyses. Specifically, this includes:

[0099] Signal segmentation subunits are used to divide the respiratory cycle sequence and cardiac cycle sequence into continuous short-time analysis windows (Epoch, e.g., 30 seconds).

[0100] The sleep determination subunit checks whether each short-term analysis window simultaneously meets the following criteria: the activity indices of the respiratory cycle sequence and the cardiac cycle sequence of the corresponding window are both below the movement threshold; the zero-crossing rate is below the threshold (to filter out "quiet wakefulness" states where the animal is macroscopically still but exhibits muscle tension or minor postural adjustments, as signals in the sleep state are mainly dominated by slow breathing, and its zero-crossing rate is significantly lower); and the stability is above the threshold (to confirm that the physiological rhythm has entered the regular state unique to sleep). If so, the short-term analysis window is determined to be a potential sleep window. When the number of consecutive potential sleep windows exceeds a preset minimum duration threshold (e.g., 3 minutes for dogs and 1 minute for felines), the total events of the potential sleep window are defined as a continuous sleep state segment; the remaining potential sleep windows are classified as wakefulness state segments.

[0101] The sleep level assessment subunit is used to evaluate the depth and stability of breathing based on the mean and variance of the respiratory cycle sequence envelope. A continuous sleep state segment with a mean higher than the threshold and a variance lower than the threshold is a deep sleep stage, while the opposite is a light sleep stage or a restless stage.

[0102] The sleep posture determination subunit is used to determine that the animal is curled up when the center-to-periphery ratio is greater than a threshold (e.g., 2.5) and stretched out when it is less than a threshold (e.g., 1.5). When the cross-unit coherence is greater than a threshold (e.g., 0.8), it is determined to be rolling over as a whole, and when it is less than a threshold (e.g., 0.3), it is determined to be localized limb twitching. This allows for a more accurate quantification of the type and severity of sleep disturbance events.

[0103] The instantaneous health event detection unit specifically includes:

[0104] The transient health event identification subunit is used to mark the current moment as the occurrence time of a transient health event when the output value of the Teager energy operator in the time frequency domain features exceeds a threshold (e.g., 5-10 times the recent baseline energy).

[0105] The transient health event analysis subunit acquires signals within a preset time window before and after the event's occurrence. It generates a wavelet energy distribution time-spectrum using continuous wavelet transform. Event classification is performed by matching the wavelet energy distribution time-spectrum with a preset event energy fingerprint, and the instantaneous frequency of the signal within the preset time window is calculated to determine the accuracy of the transient health event classification. For example, a typical cough event fingerprint is defined as a broadband vertical energy concentration pattern with a duration of less than 0.3 seconds on the time axis and energy mainly concentrated in the 2-15 Hz frequency band on the frequency axis. The instantaneous frequency of the signal is acquired; an explosive cough or sneeze event is usually accompanied by a sharp, non-physical spike in the instantaneous frequency. Ultimately, an event is identified as a "cough" if and only if a TEO is triggered, the wavelet energy spectrum matches the cough fingerprint, and the instantaneous frequency shows a corresponding sharp fluctuation. This multi-feature cross-validation mechanism can capture these fleeting events with high sensitivity and a low false alarm rate, and effectively classify them.

[0106] The emotion assessment unit uses the spatial posture features and species-specific behavioral indices unique to this invention as core situational modulators to dynamically interpret traditional physiological parameters, thereby overcoming the shortcomings of existing general models that cannot be accurately applied to different species and scenarios. Specifically, it includes:

[0107] The dynamic baseline calculation subunit is used to calculate a dynamic baseline based on heart rate variability (HRV) entropy.

[0108] The emotion recognition subunit determines the current emotion as excitement / exercise if both the panting index and activity index are above a threshold when the target animal species is canine, and heat stress if both are above a threshold, adding a "potential discomfort" score. When the target animal species is feline, satisfaction / relaxation is determined if both the gurgling index and heart rate variability entropy are above a threshold, and self-soothing if both are below a threshold, adding a "potential discomfort" score. The final state determination is a decision-making process that integrates the above information. The model compares real-time physiological parameters with its dynamic baseline to obtain the deviation. This deviation is then weighted and fused with postural priors and contextualized behavioral indices. This ensures that a small physiological parameter abnormality is significantly amplified in situations identified as "heat stress" or "self-soothing," while conversely, even a slight increase in heart rate will suppress the "alertness" score in a "relaxed" posture and behavior context. Through this unique fusion method, the present invention achieves a highly personalized and context-aware animal state assessment that far surpasses existing technologies, and can output state labels or continuous "awake-stress" indices.

[0109] The safety monitoring unit specifically includes:

[0110] The dynamic evidence monitoring subunit is used to monitor dynamic evidence in real time. The dynamic evidence is of three types: first, upper / lower boundary events, i.e., energy transitions that occur within a very short period of time; second, periodic physiological wave evidence, i.e., the stable existence of relevant narrow band spectral lines or envelopes of respiratory cycle sequences or cardiac cycle sequences over a continuous period of time; and third, low-frequency energy duty cycle, i.e., the continuous overthreshold of low-frequency band energy relative to the environmental baseline.

[0111] The in-service duration calculation subunit is used to update the in-service probability with a fixed step size. When any dynamic evidence is established, the in-service probability is increased, and when dynamic evidence is missing, it is decreased at a set rate. When the in-service probability reaches the upper threshold, it is determined to be in-service, and when it is below the lower threshold, it is determined to be out of service. The in-service duration is accumulated to obtain the total in-service duration.

[0112] The aggregation unit submits the output results of the above units to the processor module for aggregation. The processor module can cache the monitoring results and key features locally at a preset granularity, and at the same time interact with external terminals and / or cloud servers through the communication module to display heart rate, respiration, sleep and presence information in real time, and trigger prompts or alarms when abnormal events are detected (such as apnea, heart rate exceeding the limit, abnormal prolonged lying down or prolonged absence).

[0113] The usage process of this invention is as follows:

[0114] Place the carrier in a frequently used resting area for your pet (such as a kennel, a corner of the living room, or a flat area inside the cage); connect it to a power source or power it after installing a power module. Upon first use, complete device discovery and initial binding via an external terminal (mobile application), enter animal information (species, size range, nickname, etc.), and select or automatically match the default monitoring template; after binding, the system enters standby mode.

[0115] The system automatically initiates data acquisition upon detecting events or evidence of presence at the edge of the upper pad. The external terminal displays real-time heart rate (HR), respiratory rate (RR), sleep / wake status, presence / absence status, and cumulative presence duration; a simplified contact distribution diagram can also be displayed when multiple units are deployed. Upon completion of the status assessment, the external terminal simultaneously provides animal emotion / status labels (e.g., relaxed, drowsy, alert / potentially stressed) and an emotion index. During nighttime or user-defined quiet periods, the system continuously monitors in a low-power mode; the next day, it automatically generates a sleep report (including total sleep duration, sleep efficiency, number and duration of long sleep segments, arousal index, number of turnings, respiratory rhythm stability, etc.) and an animal emotion / status report (overall nighttime arousal level trend, abnormal periods marked). Reports can be viewed or exported from the external terminal.

[0116] Specifically, the system employs a multi-layer state machine to ensure robustness and low power consumption:

[0117] Occupancy State Machine (Safety Monitoring): States can be divided into out-of-position (S0), candidate in-position (S1), and in-position (S2). S0→S1: Detection of an upper pad edge event or a low-frequency energy duty cycle continuously exceeding the threshold (≥30–90 s). S1→S2: Evidence of respiratory / cardiac cycle persists for ≥20–60 s. S2→S1 / S0: If evidence is missing, the in-position probability decreases according to the hold-attenuation mechanism; if it falls below the lower threshold, out-of-position is determined. The output is the in-position probability PPP (0–1) and the cumulative in-position duration (seconds).

[0118] Health estimation state machine (health monitoring): The state can be divided into unstable segment (H0), stable breathing / heart rate segment (H1), and sleep segment (H2). H0→H1: RR / HR peak SNR>6 dB, coefficient of variation below the threshold; H1→H2: Low body movement + stable breathing + minimum duration (2–3 min); H2→H1 / H0: Frequent turning over or increased body movement, drastic fluctuations in RR / HR. Outputs include HR, RR, sleep / wake label, sleep efficiency, long sleep segment, arousal index, and turning over count.

[0119] With user authorization, the communication module synchronizes monitoring results and key characteristics to the cloud. The cloud performs long-term trend analysis, abnormal pattern clustering, and health report generation, supporting daily / weekly / monthly retrieval of pets' HR / RR intervals, sleep quality changes, and in-situ behavioral patterns. When the cloud detects a trend that significantly deviates from the individual baseline (such as a long-term decline in sleep efficiency or increased nighttime awakenings), it can push "attention suggestions" to external terminals.

[0120] Typical user path (example):

[0121] 1) Initial binding → Select "Medium-sized dog" template → Place pad → Automatic on-site monitoring;

[0122] 2) Daytime: View HR / RR / position information in real time on the APP;

[0123] 3) Nighttime: The system silently monitors sleep and generates a sleep and mood report the following day;

[0124] 4) If "breathing apnea" or "long-term high emotional index" occurs, the APP will immediately issue an alert and provide non-diagnostic suggestions such as "observation suggestion / seeking medical attention nearby / soothing suggestion";

[0125] 5) Monthly health trend reports are generated in the cloud for users or veterinarians to refer to.

[0126] The scenario-based event flow (example) is as follows:

[0127] (A) Short daytime rest

[0128] Pet on pad → Trigger strong pad edge, S0 → S1;

[0129] Stable respiratory narrow band spectral lines were detected within 30 seconds → S1 → S2 (in situ), indicating RR;

[0130] If you remain resting for 1–2 minutes and your heart rate / respiratory rate is stable, mark it as “nap / rest” and do not include it in your nighttime sleep report.

[0131] When the device leaves the pad, trigger the edge event again → S2 → S0, record the duration in place and upload it.

[0132] (B) After the owner leaves home (possible separation anxiety / stress tendency)

[0133] Before and after leaving home, the system continuously monitored the frequency of on-site / off-site switching and the increase in body movement index;

[0134] If, after taking over, the patient exhibits an increase in RR with increased volatility, a higher HR, frequent turning over, and a lack of long sleep segments, the status engine will output a "alert / possibly stressed" label, and the Emotion Index (EI) will rise (0–1).

[0135] If the symptoms do not subside after 30–60 minutes, a "Possible anxiety / discomfort" alert can be sent, suggesting that the owner observe remotely or play a soothing sound.

[0136] (C) Nighttime sleep

[0137] If the patient is in place and activity decreases in the evening, they enter H1; if they meet the criteria of "low body movement + stable breathing + continuous ≥3 min", they enter H2 (sleep).

[0138] Generate sleep / wake sequences: identify long sleep segments (≥5–10 min), number of awakenings, and sleep efficiency;

[0139] If apnea (e.g., no effective respiratory cycle ≥10–15 s, adaptive to body type) or HR exceeding the limit for >30 s is detected, an abnormal alert will be sent immediately;

[0140] The following day, a nighttime sleep report is generated, including total sleep duration, efficiency, long sleep period, arousal index, respiratory rhythm stability, number of times the patient turned over, etc.

[0141] (D) Recovery after exercise / play

[0142] In the initial stage of padding, body movement is large and RR / HR is high → H0;

[0143] Within 5–10 minutes, RR / HR decreases and body movement decreases → H1;

[0144] If the patient quickly transitions to H2 and a long sleep phase is established, the outer end of the sleep cycle may indicate "good recovery"; if the patient is unable to enter H2 for a long time, it may indicate "poor rest quality".

[0145] Anomaly Detection and Alarms: When abnormal indicators are detected, the external terminal will issue a prompt or alarm. Anomalies include, but are not limited to: apnea (absence of a respiratory cycle for a preset duration while in place), heart rate exceeding limits (sustained time exceeding the upper / lower limit relative to the individual baseline), abnormally prolonged bed rest (prolonged time in place with sparse body movement during the day), prolonged disposition (prolonged disposition at night), and persistently high mood index accompanied by unstable breathing / frequent turning over, etc. Alarms can be sent via in-app notifications, pop-ups, or optional SMS / email; thresholds and triggering strategies can be configured by the user on the external terminal.

[0146] The present invention has the following beneficial effects:

[0147] 1) Non-restraint, low-invasiveness, and high compliance: Animals do not need to wear collars / breast straps or have electrodes attached to their bodies, making long-term continuous monitoring in rest / sleep settings more natural and less likely to be rejected.

[0148] 2) Single physical quantity covers multiple indicators: With the pressure signal as the only input, heart rate, respiration, sleep-related parameters and in-situ / out-of-situ status can be inferred simultaneously, simplifying the system structure and reducing costs and maintenance burden.

[0149] 3) Both the structure and the algorithm are open: there are no restrictions on sensor materials and arrangements, communication methods and carrier shapes. The implementation path of health detection and safety monitoring is scalable, which facilitates cross-species / cross-body-size deployment and subsequent functional iteration.

[0150] 4) Robustness and scalability: By managing the storage / communication / power supply and signal processing modules through the processor, it is easy to achieve low-power operation, edge caching and cloud analysis; a mechanical coupling layer can be added as needed to optimize the transmission of small vibrations and improve the detectability of weak physiological signals.

[0151] 5) Easy to scale up: The carrier has flexible form (mat, nest, floor mat, transport box liner, etc.), which can be easily implemented in multiple scenarios such as home, hospital (pet hospital / laboratory animal facility) and boarding / transportation, and supports remote care and abnormal alarm.

[0152] 6) Animal-specific frequency bands and feature adaptation: Covering canine panting at 2–8 Hz and higher cardiac micro-vibration frequencies (approximately 6–20 Hz) for small animals, and identifying / suppressing feline purring vibrations (~25–150 Hz) to avoid missed detections / aliasing.

[0153] 7) Structural solution for contact / coupling and posture differences: Through a mechanical coupling layer (optional) with central enhancement and peripheral vibration suppression and a non-uniform layout with central density and peripheral sparseness, combined with automatic selection of effective sampling area (ROI) of chest and abdomen, under the common animal lying / curling / side lying and soft isolation of hair, the SNR of chest and abdomen micro-vibration is improved and tail / limb artifacts are suppressed.

[0154] 8) Timing adaptation of polyphasic sleep / short abortion: Using technical methods to correctly handle short sleep segments and short abortions in animals, reducing the risk of misjudging normal patrols as fragmented sleep, thus distinguishing it from the assumption of long, stable sleep segments in humans.

[0155] 9) Hygiene and durability engineering adaptation: The structure, such as the detachable electronic compartment, meets the requirements of animal cleaning / urine / biting and scratching environment; unlike human mattress monitoring, which is mostly an internal structure that cannot be removed and washed.

[0156] 10) Parameter grading and hardware coordination: PVDF thickness, unit area / density, front-end feedback capacitance, bandwidth and sampling rate are configured in grades according to species / body size to ensure detection sensitivity and dynamic range for different animals in the 0.05–20 Hz micro-vibration frequency band; a unified parameter set for humans is difficult to be compatible in animal scenarios.

[0157] 11) From physiological indicators to contextualized output of state / emotion: Based on HR / RR / sleep / posture, combined with characteristics such as panting ratio, respiratory stability, body movement sparseness / turning frequency, etc., output management-level assessment of animal state / emotion (arousal level); different from the conventional reporting system of human monophasic sleep scenario.

[0158] In summary, this invention not only surpasses existing animal monitoring solutions in terms of "non-wearable, multi-indicator, and remote" capabilities, but also addresses the fundamental differences between animals and humans in physiological frequency bands, posture / coupling, sleep structure, and environmental interference by proposing an integrated structure-algorithm adaptation and enhancement. Thus, it achieves stable, long-term, and scalable continuous monitoring of vital signs without increasing the burden on animals, demonstrating significant practical value and industrialization prospects.

[0159] It is worth noting that in the embodiments of the above system, the various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy distinction between each other and are not used to limit the scope of protection of the present invention.

[0160] The embodiments described above are merely illustrative. The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art will clearly understand that each implementation can be achieved using software plus necessary general-purpose hardware platforms, or it can be implemented solely through hardware, as long as the function or purpose can be achieved.

[0161] It should be understood that the embodiments and descriptions above are only the principles, main features and advantages of the present invention. Various changes and modifications can be made to the present invention without departing from the spirit and scope of the invention, and all such changes and modifications fall within the protection scope of the present invention.

Claims

1. An animal non-restraint physiological monitoring system based on pressure signals, characterized in that, It includes a pressure signal acquisition module, a signal processing module, a storage module, a communication module, and a power supply module, all connected to the processor module. The signal processing module specifically includes: The preprocessing unit is used to select the corresponding filtering settings according to the target animal species and body size, and extract the respiratory cycle sequence and cardiac cycle sequence from the compression signal according to the corresponding filtering settings, and identify canine panting events or feline purring events, and perform noise reduction based on the identified events; The basic feature extraction unit is used to calculate the time-domain features, frequency-domain features, time-frequency-domain features, and panting index / gurgling index of the target animal based on the respiratory cycle sequence and the cardiac cycle sequence. The pose feature extraction unit is used to generate a local response map to mark the energy at each position based on the position of each pressure sensor in the pressure signal acquisition module and the energy of the acquired signal. Based on the local response map, the animal's thoracic and abdominal regions are identified as effective acquisition areas, as well as the center-periphery ratio and cross-unit coherence. The physiological parameter detection unit is used to calculate the respiratory rate and heart rate of the target animal based on the time-domain and frequency-domain characteristics of the respiratory cycle sequence and cardiac cycle sequence of the effective acquisition area. The sleep monitoring unit is used to divide the respiratory cycle sequence and heart cycle sequence into several short-term analysis windows, and extract the sleep state segment and depth based on the time domain and frequency domain characteristics of the short-term analysis windows, and determine the sleep posture based on the center-periphery ratio and cross-unit coherence. The instantaneous health event detection unit is used to identify instantaneous health events based on time-frequency domain features; An emotion assessment unit is used to identify the current emotion of a target animal based on time-domain features, panting index / gurgling index, and heart rate. The safety monitoring unit is used to monitor dynamic evidence in real time to indicate whether the target animal is present or not, and to calculate the duration of presence based on the dynamic evidence. The aggregation unit is used to send the results calculated by each unit to the processor module, so that the processor module can send the data to an external terminal and / or cloud server through the communication module for data interaction.

2. The animal non-restraint physiological monitoring system based on pressure signals according to claim 1, characterized in that, The pressure signal acquisition module specifically includes: A carrier for the target animal to contact or lie on; Several pressure sensors are installed at different locations within the carrier to collect pressure signals generated by the interaction between the target animal and the carrier when in-situ evidence is detected. The pressure sensors are indirectly or directly contacted with the carrier through a mechanical coupling structure, which is a mechanical coupling layer. This mechanical coupling layer forms at least two different mechanical impedances in the central region and the peripheral region of the carrier, wherein the compressive modulus and / or thickness and / or porosity of the central region are different from those of the peripheral region.

3. The animal non-restraint physiological monitoring system based on pressure signals according to claim 1, characterized in that, The preprocessing unit specifically includes: The canine signal preprocessing subunit, when the user-preset target animal species is canine, searches for the corresponding cardiac frequency range and median window duration from the body size-filter parameter mapping table based on the user-preset target animal body size. Then, it extracts the respiratory cycle sequence from the compression signal using bandpass filtering within the canine respiratory frequency range, and extracts the cardiac cycle sequence from the compression signal using bandpass filtering within the searched cardiac frequency range. Next, it performs baseline suppression on the respiratory and cardiac cycle sequences, achieved through sliding median subtraction using the searched median window duration and first-order high-pass filtering. Finally, it calculates the short-time energy of each segment in the respiratory and cardiac cycle sequences, and uses adaptive thresholding to identify and remove noise. Segments with short-time energy greater than a preset threshold in the surrounding segments are marked as tail / limb impact events rather than noise. Segments with short-time energy rising to a preset energy threshold, with the main energy concentrated in 2–8 Hz and dominated by the central region, are marked as canine panting events rather than noise. The feline signal preprocessing subunit, when the user-preset target animal species is feline, finds the corresponding median window duration from the body size-filter parameter mapping table based on the user-preset target animal body size. Then, it extracts the respiratory cycle sequence from the pressure signal using bandpass filtering within the feline respiratory frequency range, and extracts the cardiac cycle sequence from the pressure signal using bandpass filtering within the feline cardiac frequency range. Next, it performs baseline suppression on the respiratory and cardiac cycle sequences by sliding median subtraction based on the found median window duration and first-order high-pass filtering. Then, it identifies segments in the cardiac cycle sequence where the signal energy in the 25–50 Hz frequency band continuously increases and exceeds a preset threshold as feline purring events, and filters out these events from the cardiac cycle sequence. Finally, it calculates the short-time energy of each segment of the respiratory and cardiac cycle sequences and uses adaptive thresholding to identify and remove noise from each component.

4. The animal non-restraint physiological monitoring system based on pressure signals according to claim 1, characterized in that, The feature extraction unit specifically includes: The temporal feature extraction subunit is used to extract the corresponding main peak position, coefficient of variation, stability, zero crossover rate, mean and variance of the envelope, and activity index based on short-time energy from the respiratory cycle sequence and the cardiac cycle sequence, respectively. The frequency domain feature extraction subunit is used to extract the corresponding spectral peak frequency, bandwidth, and harmonic energy ratio from the respiratory cycle sequence and the cardiac cycle sequence, respectively. The time-frequency domain feature extraction subunit is used to calculate the energy distribution, instantaneous frequency, and Teager energy operator for respiratory cycle sequences and cardiac cycle sequences, respectively. The panting index extraction subunit is used to calculate the ratio of the energy of the panting event frequency band to the total energy of the respiratory cycle sequence when the target animal is a canine, and use it as the panting index. The Gurgle Index Extraction Subunit is used to calculate the ratio of the energy of the feline gurgle event frequency band to the total energy of the heartbeat cycle sequence when the target animal is a feline, and this ratio is used as the Gurgle Index.

5. The animal non-restraint physiological monitoring system based on pressure signals according to claim 1, characterized in that, The pose feature extraction unit specifically includes: The response map generation subunit is used to generate a local response map that marks the energy at each location by mapping the pressure signal energy of each pressure sensor in the pressure signal acquisition module to its physical location. The response map parameter calculation subunit is used to calculate the sum of energy of the pressure sensors in the predefined central region divided by the sum of energy in the surrounding region, as the center-periphery ratio, based on the local response map; calculate the coherence function of different pressure sensors in the respiratory or cardiac frequency band, as the cross-unit coherence; calculate the average point of the position-weighted calculation with the energy of each pressure sensor as the weight, as the response centroid; and calculate the principal component direction of the energy distribution, as the principal axis. The effective sampling region extraction sub-unit is used to obtain the position of the pressure sensor with cross-unit coherence higher than the threshold along the main axis with the response centroid as the center of the animal's chest and abdomen as the center, and this position is taken as the effective sampling region. When a pressure sensor with energy significantly higher than the effective sampling region but located in the edge region appears in the local response map, and the coherence of the pressure sensor with the pressure signal of the pressure sensor in the effective sampling region is lower than the threshold, it is identified as an artifact and edge interference from the limbs or tail and is not taken as the effective sampling region.

6. The animal non-restraint physiological monitoring system based on pressure signals according to claim 1, characterized in that, The physiological parameter detection unit specifically includes: The respiratory rate estimation subunit is used to calculate the respiratory cycle and then the respiratory rate based on the envelope signal and main peak position of the respiratory cycle sequence in the effective sampling area. The heart rate estimation subunit is used to calculate the heart rate based on the envelope signal and main peak position of the heart rate cycle sequence in the effective sampling area. The signal verification subunit is used to determine whether the respiratory rate or heart rate simultaneously meets the following criteria: the activity index is lower than the exercise threshold, the coefficient of variation and stability are better than the set standard, the bandwidth of the main peak of the signal spectrum is lower than the threshold and the harmonic energy ratio fluctuation value is less than the threshold. If so, the estimation is determined to be accurate; otherwise, the estimation is determined to be incorrect and the estimation is repeated.

7. The animal non-restraint physiological monitoring system based on pressure signals according to claim 1, characterized in that, The sleep detection unit specifically includes: The signal division subunit is used to divide the respiratory cycle sequence and the cardiac cycle sequence into continuous short-time analysis windows, respectively. The sleep judgment subunit checks whether each short-term analysis window simultaneously meets the following criteria: the activity index of the respiratory cycle sequence and the cardiac cycle sequence of the corresponding window are both below the motion threshold, the zero crossover rate is below the threshold, and the stability is above the threshold. If so, the short-term analysis window is determined to be a potential sleep window. When the number of consecutive potential sleep windows exceeds a preset minimum duration threshold, the total events of the potential sleep window are defined as a continuous sleep state segment. The remaining potential sleep windows are classified as wakefulness state segments. The sleep level assessment subunit is used to evaluate the depth and stability of breathing based on the mean and variance of the respiratory cycle sequence envelope. A continuous sleep state segment with a mean higher than the threshold and a variance lower than the threshold is a deep sleep stage, while the opposite is a light sleep stage or a restless stage. The sleep posture determination subunit is used to determine that the animal is curled up when the center-periphery ratio is greater than the threshold, and otherwise is stretched out. When the cross-unit coherence is greater than the threshold, it is determined that the animal is rolling over as a whole, and otherwise is determined to be local limb twitching.

8. The animal non-restraint physiological monitoring system based on pressure signals according to claim 1, characterized in that, The instantaneous health event detection unit specifically includes: The instantaneous health event identification subunit is used to mark the current time as the occurrence time of an instantaneous health event when the output value of the Teager energy operator in the time-frequency domain features exceeds a threshold. The instantaneous health event analysis subunit is used to acquire signals within a preset time window before and after the occurrence time, generate a wavelet energy distribution time spectrum using continuous wavelet transform, classify events by matching the wavelet energy distribution time spectrum with a preset event energy fingerprint, and calculate the instantaneous frequency of the signal within the preset time window to determine the accuracy of the instantaneous health event classification.

9. The animal non-restraint physiological monitoring system based on pressure signals according to claim 1, characterized in that, The emotion assessment unit specifically includes: The dynamic baseline calculation subunit is used to calculate the dynamic baseline based on heart rate variability entropy. The emotion recognition subunit determines the current emotion as excitement / exercise if both the panting index and activity index are above a threshold when the target animal species is canine, and as heat stress when both are below a threshold.

10. The animal non-restraint physiological monitoring system based on pressure signals according to claim 1, characterized in that, The safety monitoring unit specifically includes: The dynamic evidence monitoring subunit is used to monitor dynamic evidence in real time. The dynamic evidence is of three types: first, upper / lower boundary events, i.e., energy transitions that occur within a very short period of time; second, periodic physiological wave evidence, i.e., the stable existence of relevant narrow band spectral lines or envelopes of respiratory cycle sequences or cardiac cycle sequences over a continuous period of time; and third, low-frequency energy duty cycle, i.e., the continuous overthreshold of low-frequency band energy relative to the environmental baseline. The in-service duration calculation subunit is used to update the in-service probability with a fixed step size. When any dynamic evidence is established, the in-service probability is increased, and when dynamic evidence is missing, it is decreased at a set rate. When the in-service probability reaches the upper threshold, it is determined to be in-service, and when it is below the lower threshold, it is determined to be out of service. The in-service duration is accumulated to obtain the total in-service duration.

Images