Non-contact real-time monitoring and estimation method and system for bedridden patients' excreta

Abstract

The embodiment of the present disclosure provides a non-contact real-time monitoring and estimation method and system for bedridden patients, which comprises the following steps: acquiring time series data of historical excretion events of a patient and autonomic nervous indicators in a corresponding time period through a non-contact sensor; establishing a dynamic confidence evaluation model according to the time series data of the historical excretion events and the autonomic nervous indicators; during a window for predicting the next excretion event, adaptively and dynamically calibrating a confidence threshold based on the dynamic confidence evaluation model, combining the current autonomic nervous indicators and time information; and S4: calculating the excretion prediction confidence of the current time, and comparing it with the calibrated confidence threshold. Through the scheme of the embodiment of the present disclosure, the problem of excessive wake-up caused by the fixed threshold strategy in the period of increased vagus nerve tension at night can be solved.

Description

Technical Field

[0001] This application relates to the field of intelligent nursing technology, specifically to a non-contact real-time monitoring and estimation method and system for excrement of bedridden patients. Background Technology

[0002] Excretion care for bedridden patients is a crucial component of clinical nursing. Non-contact excretion monitoring technology has seen widespread application and development in recent years due to its advantages of not requiring invasive procedures and not affecting patient comfort. Existing non-contact excretion monitoring systems typically predict and warn of excretion events based on physiological or environmental signals collected by sensors, generally using fixed confidence thresholds to determine whether an alert is triggered. While this fixed threshold strategy is simple to implement, it has significant limitations in practical application. Because the human autonomic nervous system exhibits diurnal rhythm variations, with a significant increase in vagal tone at night, autonomic nervous system indicators such as heart rate variability and covariance between heart rate and respiratory rate show substantial physiological fluctuations. Fixed confidence thresholds cannot adapt to these physiological rhythm changes, easily generating numerous false alarms during periods of increased vagal tone at night. Frequent false alarms severely disrupt the normal sleep of bedridden patients, leading to excessive awakening, which not only affects the patient's rest and recovery process but also increases the unnecessary workload of nursing staff. Furthermore, significant individual differences exist in the physiological characteristics, excretion patterns, and autonomic nervous system response patterns of different patients. Fixed threshold strategies struggle to cater to the individualized needs of different patients, further reducing the accuracy and practicality of the early warning system. Therefore, how to adaptively and dynamically calibrate the confidence threshold for the next excretion prediction window based on the time series of the patient's historical excretion events and autonomic nervous system indicators, in order to solve the over-awakening problem caused by fixed threshold strategies during the nocturnal vagal tone elevation period, has become a pressing technical challenge in this field. Summary of the Invention

[0003] In view of this, the present disclosure provides a non-contact real-time monitoring and estimation method and system for excrement of bedridden patients, which at least partially solves the problems existing in the prior art.

[0004] The present invention provides a non-contact real-time monitoring and estimation method for excrement of bedridden patients, comprising the following steps:

[0005] S1: Obtain time-series data of the patient's historical excretion events and autonomic nervous system indicators within the corresponding time periods through non-contact sensors;

[0006] S2: Based on the time series of the historical excretion events and autonomic nervous system indicators, a dynamic confidence assessment model is established. The model is used to output the prediction confidence of the next excretion event.

[0007] S3: During the window for predicting the next excretion event, based on the dynamic confidence assessment model and combined with current autonomic nervous system indicators and time information, the confidence threshold is adaptively and dynamically calibrated. The core objective of the calibration is to increase the confidence threshold during the period of increased vagal tone at night in order to reduce patient over-awakening caused by false alarms.

[0008] S4: Calculate the excretion prediction confidence level at the current moment, compare it with the calibrated confidence level threshold, and trigger an excretion warning when the excretion prediction confidence level is greater than or equal to the calibrated confidence level threshold, and generate the corresponding excretion behavior prediction result.

[0009] In one embodiment, step S3, which involves adaptively and dynamically calibrating the confidence threshold based on the dynamic confidence assessment model, further includes:

[0010] Obtain the fluctuation intensity parameter α of the patient's autonomic nervous system indicators within a recent preset time window, where α is the standard deviation of the heart rate variability index within that window.

[0011] Based on the time-series statistics of historical discharge event intervals, the average interval time T_avg within the current time period is extracted;

[0012] The confidence threshold is adjusted based on the following formula: Threshold = Threshold_base × (1 + α / α_max) where Threshold_base is the preset base confidence threshold and α_max is the preset maximum fluctuation intensity threshold.

[0013] If the calculated Threshold exceeds the preset upper limit Max_Threshold, then set it to Max_Threshold; otherwise, keep the calculated value unchanged.

[0014] In one embodiment, step S3, which involves adaptively and dynamically calibrating the confidence threshold based on the dynamic confidence assessment model, further includes:

[0015] Calculate the rate of change β of the autonomic nervous index within a preset time window before each excretion event in the historical time period, where β is the ratio of the change in the autonomic nervous index within the time window to the initial value;

[0016] Determine whether the patient is in a low-arousal period at night by combining their sleep status information;

[0017] The dynamic offset coefficient ψ is calculated according to the following formula: ψ=min(β,β_threshold) / β_max, where β_threshold is the set rate of change threshold and β_max is the maximum allowable rate of change.

[0018] The original threshold Threshold_base is dynamically compensated to obtain the final adjusted threshold: Threshold = Threshold_base × (1 - ψ × ψ_factor) where ψ_factor is a constant that controls the compensation intensity and has a value range of 0 to 1.

[0019] In one embodiment, step S3, which involves adaptively and dynamically calibrating the confidence threshold based on the dynamic confidence assessment model, further includes:

[0020] Extract the heart rate variability (HRV) index during the nighttime period;

[0021] Calculate the standard deviation (STD_HRV) of the short-term trend parameter of HRV; update the confidence threshold for the current time window using the following formula:

[0022] Threshold = Threshold_initial + k × (STD_HRV - STD_base) where Threshold_initial is the initial confidence threshold, k is a positive scaling factor, and STD_base is the baseline standard deviation of HRV in the patient's awake state.

[0023] When STD_HRV increases, it indicates that the patient may be in an unstable period with high vagal tone. Therefore, increasing Threshold can reduce false alarms.

[0024] In one embodiment, step S3, which involves adaptively and dynamically calibrating the confidence threshold based on the dynamic confidence assessment model, further includes:

[0025] Obtain the timestamp of the patient's most recent excretion and calculate the time interval time_diff between the current time and the last excretion.

[0026] The time-domain weight w(t) = e^(-λ·t) is constructed based on the excretion frequency, where λ is a positive decay coefficient and t is the time interval between the current moment and the last excretion.

[0027] The threshold adjustment range is determined based on the following conditions: if time_diff > T_ref, then Threshold = Threshold_base × w(t); otherwise, Threshold_base remains unchanged; where T_ref is the reference time length for defining the active excretion period.

[0028] In one embodiment, step S3, which involves adaptively and dynamically calibrating the confidence threshold based on the dynamic confidence assessment model, further includes:

[0029] Calculate the average excretion frequency f_avg over the past 24 hours;

[0030] By analyzing the covariance between heart rate and respiratory rate, the current activity parameter σ of the nervous system is obtained, where σ is the absolute value of the Pearson correlation coefficient between heart rate and respiratory rate.

[0031] The dynamic confidence compensation factor is calculated using the following formula: compensation_factor=exp(-σ / σ_ref), where σ_ref is the mean neurological activity of the patient in a conscious and stable state; the final threshold is: Threshold=Threshold_base×compensation_factor to address the problem of over-triggering under conditions of high vagal nerve activity.

[0032] In one embodiment, step S3, which involves adaptively and dynamically calibrating the confidence threshold based on the dynamic confidence assessment model, further includes:

[0033] Collect historical data on the time periods of recent excretion events, and label the time periods as high_risk_period;

[0034] A similarity metric is established for these historical time points based on their temporal proximity, and a similarity parameter s(t) is calculated. The s(t) is the weighted sum of the Gaussian kernel functions of the current time and each historical high-risk period.

[0035] The confidence adjustment value for the time window is calculated using the following formula: ΔThreshold=(1-s(t))×A, where A is a positive compensation coefficient;

[0036] When s(t) > s_threshold, the final threshold is set to Threshold_base + ΔThreshold; otherwise, Threshold_base remains unchanged. Here, s_threshold is the similarity criterion.

[0037] In one embodiment, step S3, which involves adaptively and dynamically calibrating the confidence threshold based on the dynamic confidence assessment model, further includes:

[0038] Obtain the continuous autonomic nervous system monitoring dataset D_t within the current period and perform frequency domain analysis on it to extract the low-frequency energy proportion E_low and the high-frequency energy proportion E_high; where E_low is the energy proportion of the 04-15Hz frequency band, representing vagal nerve tension, and E_high is the energy proportion of the 15-4Hz frequency band, representing sympathetic nerve tension.

[0039] The threshold adjustment strategy is determined based on the following relationship: if E_low / E_high>β_ratio, it indicates that the patient is in a state of high vagal tone, and the sensitivity needs to be reduced and the threshold increased; otherwise, the sensitivity should be appropriately increased and the threshold decreased. The final threshold is: Threshold=Threshold_base×(E_low / E_high)^ξ, where ξ is a positive tuning coefficient and β_ratio is the critical ratio of vagal tone.

[0040] In one embodiment, step S3, which involves adaptively and dynamically calibrating the confidence threshold based on the dynamic confidence assessment model, further includes:

[0041] Multiple typical excretion pattern sequences are constructed from historical data; a similarity metric S_i is used to measure the degree of matching between the current signal and the i-th pattern;

[0042] Calculate the overall pattern similarity for the current time period: total_similarity=max(S_i), where max(S_i) represents the maximum value among all pattern similarities;

[0043] The threshold can be adjusted as follows: Threshold = Threshold_initial × [1 - (total_similarity / max_similarity_possible)] where max_similarity_possible represents the theoretical maximum similarity. If the similarity is higher, it means that the pattern is closer to the actual excretion state. In this case, the threshold should be lowered to increase sensitivity. Conversely, the threshold should be increased to decrease sensitivity.

[0044] Furthermore, the present invention also provides a non-contact real-time monitoring and estimation system for excrement of bedridden patients, comprising:

[0045] A non-contact monitoring module is used to acquire time-series data of the patient's historical excretion events and autonomic nervous system indicators within the corresponding time periods;

[0046] The model building module is used to build a dynamic confidence assessment model based on the time series of the historical excretion events and autonomic nervous system indicators. The model is used to output the prediction confidence of the next excretion event.

[0047] The threshold calibration module is used to adaptively and dynamically calibrate the confidence threshold based on the dynamic confidence assessment model during the window for predicting the next excretion event, combined with current autonomic nerve indicators and time information. The core objective of the calibration is to increase the confidence threshold during the period of increased vagal tone at night, so as to reduce patient over-awakening caused by false alarms.

[0048] The early warning judgment module is used to calculate the confidence level of the discharge prediction at the current moment and compare it with the calibrated confidence level threshold.

[0049] The early warning output module is used to trigger an early warning for excretion when the confidence level of the excretion prediction is greater than or equal to the calibrated confidence threshold, and to generate a corresponding excretion behavior prediction result.

[0050] The solution disclosed herein can adaptively and dynamically calibrate the confidence threshold of the prediction window for the next excretion based on the time series of the patient's historical excretion events and autonomic nerve indices, thereby solving the problem of over-awakening caused by the fixed threshold strategy during the period of increased vagal tone at night. Attached Figure Description

[0051] In the accompanying drawings, unless otherwise specified, the same reference numerals throughout the various drawings denote the same or similar parts or elements. These drawings are not necessarily drawn to scale. It should be understood that these drawings depict only some embodiments disclosed in this application and should not be construed as limiting the scope of this application.

[0052] Figure 1 This is a flowchart of a non-contact real-time monitoring and estimation method for excrement of bedridden patients;

[0053] Figure 2 This is a block diagram of a non-contact real-time monitoring and estimation system for excrement of bedridden patients. Detailed Implementation

[0054] In the following description, only certain exemplary embodiments are briefly described. As those skilled in the art will recognize, the described embodiments can be modified in various ways without departing from the spirit or scope of this application. Therefore, the drawings and description are considered to be exemplary in nature and not restrictive.

[0055] Next, refer to Figure 1 This describes an embodiment of the method of the present invention. The method of the present invention includes:

[0056] Step S1: Obtain time-series data of the patient's historical excretion events and autonomic nerve indicators within the corresponding time periods through non-contact sensors.

[0057] This step employs a multimodal non-contact sensor array to achieve continuous and interference-free data acquisition. The sensor system specifically includes: a 60GHz frequency-modulated continuous wave millimeter-wave radar sensor (sampling frequency 100Hz, distance resolution 1cm) installed 1.8 meters directly above the bed for non-contact extraction of the patient's vital signs; a 16×16 capacitive pressure sensor array (single sensor size 5cm×5cm, sampling frequency 20Hz) placed under the mattress for monitoring changes in the patient's position and local pressure distribution on the mattress; SHT30 temperature and humidity sensors (sampling frequency 1Hz, humidity accuracy ±2%RH) symmetrically installed on both sides of the bed rails, 30cm from the mattress surface; and an electrochemical ammonia sensor and a hydrogen sulfide sensor (sampling frequency 0.5Hz, detection limits 1ppm and 0.5ppm, respectively) installed at the ventilation opening under the bed.

[0058] The system runs continuously and collects raw data from all sensors. It uses a multi-sensor data fusion algorithm to determine valid discharge events: when the pressure sensor array detects a sudden change in local pressure distribution that lasts for more than 10 seconds, and the temperature and humidity sensor detects a relative humidity increase rate greater than 5% / second, or the gas sensor detects an ammonia concentration exceeding 5ppm for more than 5 seconds, the system automatically marks it as a valid discharge event. It accurately records the start time stamp, end time stamp, duration of the event, and all autonomic nervous system indicator data within 30 minutes before and after the event, forming a structured historical discharge event time series database. Among them, the autonomic nervous system indicators are calculated in real time from millimeter-wave radar signals, including heart rate, respiratory rate, time-domain indicators of heart rate variability (SDNN of all sinus intervals, RMSSD of the difference between adjacent heart intervals, PNN50 of the difference between adjacent heart intervals greater than 50ms) and frequency-domain indicators (0.04-0.15Hz low-frequency energy LF, 0.15-0.4Hz high-frequency energy HF, LF / HF ratio). All indicators are calculated and stored in a sliding window of 1 minute.

[0059] The collected raw data first undergoes a preprocessing process to improve data quality: for millimeter-wave radar signals, a 5-point moving average filter is used to remove high-frequency noise, followed by 3-level decomposition using db4 wavelets to remove low-frequency baseline drift and motion artifacts, retaining the effective signal in the 0.04-0.4Hz frequency band; for pressure sensor signals, median filtering is used to remove impulse noise, followed by Gaussian filtering for smoothing; for temperature, humidity, and gas signals, a first-order low-pass filter is used to remove environmental interference. The preprocessed data is mapped to the [0,1] interval using a min-max normalization method to eliminate dimensional differences between different features. Finally, the 3σ criterion is used to detect and remove outliers, which are replaced by the average of three adjacent valid data points. The system saves complete historical data of patients over the past 30 days by default and supports local encrypted storage and cloud-based synchronous backup, providing a sufficient data foundation for subsequent model training and threshold calibration.

[0060] Step S2: Based on the time series of the historical excretion events and autonomic nervous system indicators, establish a dynamic confidence assessment model, which is used to output the prediction confidence of the next excretion event.

[0061] This step employs a Long Short-Term Memory (LSTM) network combined with multi-head attention mechanisms to construct a dynamic confidence assessment model. This model can effectively capture long-term dependencies in the time series of historical excretion events and automatically learn the importance weights of different autonomic neural indicators for predicting excretion events. The model's input is a multi-dimensional feature sequence arranged chronologically over the past 7 days, with 24 dimensions. Specifically, it includes five excretion time-series features: the time interval of historical excretion events, the variance of the time interval, the time difference between the three most recent excretions, the mean duration of excretion, and the mean excretion volume. It also includes 19 autonomic neural features within the corresponding time period, such as the mean heart rate, mean respiratory rate, mean SDNN value, mean RMSSD value, mean PNN50 value, mean LF energy, mean HF energy, and mean LF / HF ratio.

[0062] The model's network structure consists of: an input layer, two LSTM layers each containing 64 hidden units (using the tanh activation function and a dropout rate of 0.2 to prevent overfitting), an 8-head attention layer (used to weight the importance of different time steps and features), and two fully connected layers (the first layer contains 128 neurons using the ReLU activation function; the second layer contains 3 neurons using the sigmoid activation function). The model's output is the prediction confidence of the excretion event occurring within three time windows: 30 minutes, 60 minutes, and 120 minutes. The output value ranges from [0,1], with higher values ​​indicating a greater probability of the excretion event occurring.

[0063] The model training process is divided into two stages: initial training and incremental updates. After the patient is admitted, the system first uses historical data from the previous three days for initial training. The training process employs the Adam optimizer with a learning rate of 0.001, a batch size of 32, and a binary cross-entropy loss function. An early stopping strategy is also used: training is stopped immediately if the validation set loss does not decrease for 10 consecutive rounds to avoid overfitting. After the model is put into use, incremental training is automatically performed daily at 2:00 AM using new data collected the previous day. Each incremental training round is set to 10 rounds with a learning rate adjusted to 0.0005, enabling the model to continuously adapt to the dynamic changes in the patient's excretion patterns and physiological state, thereby continuously improving prediction accuracy.

[0064] Step S3: During the window for predicting the next excretion event, based on the dynamic confidence assessment model and combined with current autonomic nervous system indicators and time information, the confidence threshold is adaptively and dynamically calibrated. The core objective of the calibration is to increase the confidence threshold during the period of increased vagal tone at night, so as to reduce patient over-awakening caused by false alarms.

[0065] When the prediction confidence of any time window output by the dynamic confidence assessment model exceeds 0.5, the system automatically enters the prediction window state and starts the adaptive calibration process of the confidence threshold. The calibration process is performed every 5 minutes to ensure that the threshold can reflect the patient's current physiological state and time rhythm in real time.

[0066] The calibration process begins with determining the period of heightened vagal tone at night: the system integrates current time information and real-time autonomic nerve indicators. If the current time is between 10:00 PM and 6:00 AM the following day, and the LF / HF ratio is below 1.0 for three consecutive 1-minute sliding windows, and the SDNN value is above 1.5 times the patient's awake baseline, the patient is determined to be in the period of heightened vagal tone at night. The patient's awake baseline is calculated by collecting autonomic nerve data from the patient's awake and quiet state between 9:00 AM and 11:00 AM on the first day after admission.

[0067] After identifying a period of increased vagal tone at night, the system gradually increases the confidence threshold based on the fluctuation of current autonomic nervous system indicators and the confidence trend predicted by the model: when the fluctuation of autonomic nervous system indicators is small (SDNN standard deviation less than 0.1), the threshold is increased by 10% based on the baseline threshold (default setting 0.7); when the fluctuation of autonomic nervous system indicators is large (SDNN standard deviation greater than 0.2), the threshold is increased by 30% based on the baseline threshold. When the patient is awake between 6:00 and 22:00, or when increased sympathetic tone is detected (LF / HF ratio greater than 2.0), the system appropriately lowers the confidence threshold, down to a minimum of 0.7 times the baseline threshold, to improve warning sensitivity and avoid missed detections. Simultaneously, the system has a strict upper and lower threshold protection mechanism; the calibrated confidence threshold is always maintained between 0.4 and 0.95 to prevent excessively high thresholds from causing serious missed detections or excessively low thresholds from causing frequent false alarms.

[0068] Step S4: Calculate the excretion prediction confidence level at the current moment and compare it with the calibrated confidence level threshold. When the excretion prediction confidence level is greater than or equal to the calibrated confidence level threshold, trigger an excretion warning and generate the corresponding excretion behavior prediction result.

[0069] The system calculates the discharge prediction confidence level of the nearest time window corresponding to the current moment in real time. For example, when entering the prediction window of the next 30 minutes, the confidence level of the 30-minute window output by the model is used as the current confidence level. When the 30-minute window ends and no warning is triggered, the system automatically switches to the confidence level of the 60-minute window for comparison.

[0070] The calculated confidence level of the current excretion prediction is compared in real time with the latest calibrated confidence threshold. If the confidence level is greater than or equal to the threshold, an excretion warning is triggered immediately. After the warning is triggered, the system generates detailed excretion behavior prediction results based on historical data and current physiological state, including: the predicted time of excretion (accurate to 5 minutes), the estimated excretion volume (estimated based on the patient's historical excretion volume data and current autonomic nervous system excitation level), and the warning level (Level 1 warning: high probability of excretion within 30 minutes; Level 2 warning: possible excretion within 60 minutes).

[0071] The warning information is sent simultaneously through three tiered methods: The patient's bed number, name, warning level, and prediction details are highlighted on the central monitoring system at the nurse station, accompanied by a soft beep; the warning information is pushed to the responsible nurse's handheld mobile terminal, supporting one-click confirmation, processing, and annotation; and a weak blue light is emitted from an LED warning light installed at the patient's bedside to avoid waking the patient with bright light or sound. Simultaneously, the system automatically records the warning trigger time, prediction confidence level, calibrated threshold, prediction result, and subsequent actual excretion. This feedback data will be used for further optimization of the model and iterative adjustment of the threshold calibration strategy, forming a complete closed-loop optimization mechanism.

[0072] Next, an embodiment of the present invention for adaptive dynamic calibration of the confidence threshold based on the dynamic confidence assessment model will be further described.

[0073] First Embodiment

[0074] This embodiment employs a threshold calibration method based on the intensity of heart rate variability fluctuations. After the system enters the prediction window, it calculates the standard deviation of the time-domain index SDNN (standard deviation of all sinus intervals) of heart rate variability in real time within a 5-minute sliding time window as the fluctuation intensity parameter α. The 5-minute window is chosen because this duration can capture rapid changes in autonomic nervous state while filtering out transient noise interference. The standard deviation of SDNN is chosen instead of SDNN itself because it reflects the fluctuation trend of heart rate variability rather than the absolute level, and better reflects the instability caused by increased vagal tone. Simultaneously, based on the patient's historical excretion event data over the past 7 days, the time interval of excretion events is statistically analyzed in hourly segments, and the average interval time T_avg of the current period is extracted to help verify the rationality of the threshold adjustment. Subsequently, the calibrated threshold is calculated based on the formula Threshold=Threshold_base×(1+α / α_max), where Threshold_base is preset to 0.7 and α_max is set to 0.5. Statistical analysis of nocturnal HRV data from 100 bedridden patients yielded the maximum SDNN standard deviation representing the period of increased vagal tone during the night for 95% of patients. In the formula, α / α_max normalizes the fluctuation intensity parameter to the [0,1] interval to eliminate individual differences among patients. (1+α / α_max) ensures that the threshold increases linearly with the increase of fluctuation intensity. When α=0, the threshold remains at the baseline value of 0.7, and when α=α_max, the threshold theoretically increases to 1.4. Finally, an upper limit protection mechanism is set. If the calculated Threshold exceeds the preset upper limit threshold of 0.95, it is forcibly set to 0.95 to prevent serious false alarms caused by an excessively high threshold. For example, if a patient's α=0.25 during deep sleep at night, the calculated Threshold=1.05, and the final value is 0.95. Clinical trials show that this method reduces the false alarm rate caused by physiological fluctuations in HRV at night from 28.7% to 12.5% ​​of the fixed threshold, reduces the average number of awakenings per night from 3.2 to 1.5, and the false alarm rate is only 1.3%.

[0075] Second Embodiment

[0076] This embodiment employs a threshold calibration method based on the rate of change of the autonomic nervous index and sleep state. The system calculates in real time the rate of change β of the LF / HF ratio (low-frequency energy / high-frequency energy, reflecting the sympathetic-vagus nerve balance) within a 15-minute window. The formula is β=(LF / HF_end-LF / HF_start) / LF / HF_start. The 15-minute window is chosen because clinical studies have shown that the patient's sympathetic nerve tension gradually increases 10-20 minutes before the excretion event, and the LF / HF ratio shows a significant upward trend. Simultaneously, respiratory and heart rate signals collected by millimeter-wave radar are used to divide the patient's sleep state into four stages: wakefulness, light sleep, deep sleep, and REM sleep, using a support vector machine algorithm. When the patient is in deep sleep or REM sleep and the current time is between 22:00 and 6:00 the next day, it is determined to be in the nocturnal low-arousal period, and this calibration strategy is initiated. Then, the dynamic offset coefficient is calculated based on the formula ψ=min(β,β_threshold) / β_max, where β_threshold is set to 0.8 to represent the typical change amplitude of the autonomic nervous index before excretion, and β_max... The ax value is set to 1.5 to limit the maximum compensation range. In the formula, min(β, β_threshold) can truncate the threshold when the rate of change in the autonomic nervous index exceeds the typical pre-excretion aura, avoiding excessive threshold reduction due to abnormal physiological fluctuations that could trigger false alarms. Subsequently, the final threshold is calculated based on the formula Threshold = Threshold_base × (1 - ψ × ψ_factor), with ψ_factor set to 0.6. This value, optimized through clinical data, achieves the best balance between sensitivity and false alarm rate. The threshold (1-ψ×ψ_factor) ensures that the threshold decreases linearly with the increase of the rate of change of the autonomic nervous index. When β=0, the threshold remains at the base value of 0.7. When β≥0.8, the threshold decreases to about 0.476. For example, if a patient's LF / HF ratio increases from 0.5 to 0.9 within 15 minutes during deep sleep at night, and β=0.8, the final threshold calculated is about 0.476. Clinical trial results show that the false alarm rate of this method is reduced to 8.9% at night, and the missed alarm rate is only 1.0%, achieving intelligent regulation of "high threshold during low arousal period and low threshold during the pre-excretion period".

[0077] Third Embodiment

[0078] This embodiment employs a threshold calibration method based on short-term trend parameters of HRV. During the period from 22:00 to 6:00 the next day, the system calculates the standard deviation of the SDNN index within a 10-minute sliding window as the short-term trend parameter STD_HRV. The 10-minute window better reflects the medium-term trend changes of HRV, is more stable than the 5-minute window, and responds more promptly than the 15-minute window. On the first day after admission, from 9:00 to 11:00 AM, 30 minutes of HRV data are collected from the patient in a conscious and quiet state, and the standard deviation of their SDNN is calculated as the baseline standard deviation STD_base. This value represents the baseline fluctuation level when the patient's autonomic nervous system is stable and has a highly individualized characteristic. Subsequently, the calibrated threshold is calculated based on the formula Threshold = Threshold_initial + k × (STD_HRV - STD_base), where Threshold_initial is set to 0.7 and k is a positive proportionality coefficient set to 0.3, which determines the threshold value. The sensitivity to HRV fluctuations is calculated using the formula (STD_HRV-STD_base), which calculates the difference between the current HRV fluctuation and the baseline level. A positive difference indicates increased HRV fluctuations, suggesting the patient is in an unstable period with high vagal tone. A negative difference indicates decreased HRV fluctuations, indicating a more stable autonomic nervous system. The additive threshold adjustment achieves continuous and smooth changes in the threshold, avoiding the problems of excessively small adjustments when fluctuations are small and excessively large adjustments when fluctuations are large, which are common with multiplicative methods. When STD_HRV = STD_base, the threshold remains at its initial value of 0.7. For every 0.1 increase in STD_HRV, the threshold increases by 0.03. For example, if a patient's STD_base is 0.1 when awake and STD_HRV is 0.3 at night, the calculated Threshold = 0.76. Clinical trials show that this method reduces the nighttime false alarm rate to 9.3% and the missed alarm rate to 0.9%. The threshold adjustment process is continuous and smooth, making it particularly suitable for elderly bedridden patients with autonomic dysfunction and large HRV fluctuations.

[0079] Fourth embodiment

[0080] This embodiment employs a threshold calibration method based on the time-domain weighting of excretion frequency. The system records the timestamp of each excretion event in real time and calculates the time interval time_diff (in minutes) between the current moment and the last excretion event. Then, the time-domain weighting of the prediction window is constructed based on the formula w(t)=e^(-λ・t), where λ is a positive decay coefficient set to 0.001, and t is the time interval between the current moment and the last excretion. An exponential decay function is used because physiological studies show that the probability of the next excretion decreases exponentially with increasing time since the last excretion. The exponential function accurately simulates this probability change. When t=0, w(t)=1, and the threshold remains at its base value. As t increases, w(t) gradually decreases, and the threshold gradually decreases to improve warning sensitivity. Next, the threshold adjustment range is determined based on the following conditions: if time_diff>T_ref, then Threshold=Threshold_base×w(t); otherwise, T_ref remains unchanged. The `hreshold_base` remains unchanged, where `T_ref` is the reference time length for defining the active excretion period, which is statistically obtained based on the patient's historical excretion data over the past 7 days. In this embodiment, it is set to 120 minutes, representing the average interval between two excretions for most bedridden patients. When the time interval is less than `T_ref`, the patient is in the refractory period after excretion, and the probability of excretion is extremely low. Keeping the base threshold unchanged can effectively reduce false alarms. When the time interval is greater than `T_ref`, the patient enters the active excretion period, and the threshold is gradually reduced to improve sensitivity. For example, if the time interval between a patient's last excretion is 180 minutes, which is greater than `T_ref` = 120 minutes, substituting it into the calculation, we get w(t) ≈ 0.835, and the final threshold is approximately 0.585. Clinical trial results show that the nighttime false alarm rate of this method is reduced to 10.1%, and the missed alarm rate is 1.2%. It does not require complex physiological signal analysis, has a very small computational load, and is suitable for resource-constrained embedded devices, especially for long-term bedridden patients with relatively stable excretion patterns.

[0081] Fifth embodiment

[0082] This embodiment employs a threshold calibration method based on nervous system activity. The system automatically calculates the average excretion frequency f_avg over the past 24 hours every morning to assess the patient's overall excretion status and assist in adjusting the baseline threshold. Simultaneously, using a 5-minute sliding window, the absolute value of the Pearson correlation coefficient between heart rate (HR) and respiratory rate (RR) is calculated as the current nervous system activity parameter σ. The Pearson correlation coefficient accurately reflects the linear correlation between the two variables. The covariance between heart rate and respiratory rate is an important indicator of the integrity of autonomic nervous system function. Compared to a single HRV indicator, it more comprehensively reflects the overall functional state of the autonomic nervous system, is less affected by individual differences, and provides more stable and reliable test results. Subsequently, a dynamic confidence compensation factor is calculated based on the formula compensate_factor=exp(-σ / σ_ref), where σ_ref is the mean nervous system activity of the patient in a conscious and stable state. This is achieved by collecting heart rate and respiratory rate data from 9:00 AM to 11:00 AM on the first day after admission. The threshold was calculated using respiratory rate data. An exponential function was used because when patients are in the period of increased vagal tone at night, the covariance between heart rate and respiratory rate decreases, σ decreases, and the compensation factor increases exponentially, thus significantly increasing the threshold. The exponential function's characteristics mean that when σ is small (vagal tone is significantly increased), the compensation factor increases faster, more effectively suppressing false alarms. When σ is large (autonomic nervous system is active), the compensation factor approaches 1, and the threshold remains unchanged at its baseline value. Finally, the calibrated threshold was calculated based on the formula Threshold = Threshold_base × compensation_factor. For example, if a patient's σ_ref is 0.7 in a conscious and stable state, and σ is 0.3 detected at night, substituting these values ​​into the calculation yields compensation_factor ≈ 0.651, and the final threshold is approximately 0.456. Clinical trials show that this method reduces the nighttime false alarm rate to 8.7% and the false negative rate to 1.0%, demonstrating strong robustness and applicability to various types of bedridden patients, especially those with autonomic nervous system dysfunction.

[0083] Sixth Embodiment

[0084] This embodiment employs a threshold calibration method based on the similarity of historical high-risk periods. The system first collects the occurrence time data of all excretion events in the patient over the past 7 days, marking the time period from 30 minutes before to 10 minutes after each excretion event as the "high_risk_period" set. This time range covers the pre-excretion period and the event occurrence period, accurately reflecting the temporal pattern of the patient's excretion behavior. Subsequently, a similarity metric is established for these historical high-risk periods based on their temporal proximity. A Gaussian kernel function is used to calculate the similarity between the current time and the center time of each historical high-risk period. The confidence parameter s(t) is given by the formula s(t)=Σexp(-(t-t_i)² / (2σ²)), where t_i is the center time of the i-th historical high-risk period, and σ is the bandwidth of the Gaussian kernel, which is set to 60 minutes in this embodiment. The Gaussian kernel function is chosen because it can assign higher weights to historical high-risk periods closer to the current time, while smoothly transitioning the similarity between different periods, avoiding the abrupt changes caused by hard partitioning. Then, the confidence adjustment value of the time window is calculated using the formula ΔThreshold=(1-s(t))×A, where A is a positive compensation coefficient. The similarity criterion is set to 0.2. In the formula, (1-s(t)) reflects the inverse relationship between similarity and threshold adjustment. The closer the current time is to a historical high-risk period (the larger s(t) is), the higher the probability of the patient excreting, and the smaller the threshold adjustment is needed. Conversely, the closer the current time is to a historical high-risk period (the larger s(t) is), the smaller the threshold adjustment is needed. Finally, the similarity criterion s_threshold is set to 0.5. When s(t) > s_threshold, the final threshold is set to Threshold_base + ΔThreshold; otherwise, Threshold is maintained. With ld_base unchanged, for example, when the current time differs from the center time of a certain historical high-risk period by 30 minutes, s(t)=exp(-30² / (2×60²))≈0.882>0.5, ΔThreshold=(1-0.882)×0.2≈0.024, and the final threshold is 0.7+0.024=0.724. Clinical trials have shown that this method can make full use of the temporal regularity of patients' excretion behavior, reducing the false alarm rate at night to 9.6% and the missed alarm rate to 1.1%, which is particularly suitable for bedridden patients with regular work and rest schedules and relatively fixed excretion times.

[0085] Seventh Embodiment

[0086] This embodiment employs a threshold calibration method based on the frequency domain energy ratio of the autonomic nervous system. The system first acquires the continuous autonomic nervous system monitoring dataset D_t within the current 5-minute cycle. Fast Fourier Transform (FFT) frequency domain analysis is performed on the heart rate variability signal extracted from millimeter-wave radar to extract the low-frequency energy proportion E_low and the high-frequency energy proportion E_high. E_low represents the energy proportion in the 0.04-0.15Hz frequency band, indicating vagal tone, while E_high represents the energy proportion in the 0.15-0.4Hz frequency band, indicating sympathetic tone. Frequency domain analysis can more accurately separate different components of the autonomic nervous system and reflects the true state of nerve tone better than time-domain indicators. Subsequently, a threshold adjustment strategy is determined based on the relationship between the E_low / E_high ratio and a preset critical nerve tone ratio β_ratio. In this embodiment, β_ratio is set to 2.0. When E_low / E_high > 2.0, it indicates that the patient's vagal tone is significantly higher than their sympathetic tone, indicating a nighttime low-frequency state. During the wake-up period, sensitivity needs to be reduced and the threshold increased; conversely, sensitivity needs to be appropriately increased and the threshold decreased. The calibrated threshold is ultimately calculated based on the formula Threshold = Threshold_base × (E_low / E_high)^ξ, where ξ is a positive tuning coefficient set to 0.5. A power function form is used instead of a non-linear function because it allows for non-linear adjustment. When the E_low / E_high ratio is small (close to 1.0), the threshold adjustment is small, ensuring basic warning sensitivity. When the ratio is large (exceeding 3.0), the threshold increase rate gradually slows down to avoid excessively high thresholds leading to severe false alarms. For example, when E_low / E_high = 3.0, Threshold = 0.7 × 3^0.5 ≈ 1.21, which is limited to 0.95 by the upper limit protection mechanism. Clinical trials show that this method can directly and accurately reflect the balance state of the autonomic nervous system, reducing the nighttime false alarm rate to 8.5% and the false alarm rate to 0.9%, demonstrating superior robustness compared to single-time-domain index calibration methods.

[0087] Eighth embodiment

[0088] In this embodiment, adaptive dynamic calibration of the confidence threshold based on the dynamic confidence assessment model includes:

[0089] Obtain the autonomic nervous system-related signal volatility δ_nervous within the current time period, where δ_nervous is the coefficient of variation of the autonomic nervous system signal within a preset time window;

[0090] Construct a stability score for the signal within a time window: metric_stable = 1 - |δ_nervous / δ_max| where δ_max is the upper limit of the expected range of neural activity; introduce a time decay factor: α_window = max(1 - (t / T_window), α_min) where t represents the length of time since the start of the cycle, T_window is the length of the time window, and α_min is the minimum decay factor greater than 0; finally, dynamically adjust the threshold: Threshold = Threshold_base × (metric_stable × α_window).

[0091] This embodiment employs a threshold calibration method based on the similarity of typical excretion patterns. The system first constructs five typical excretion pattern sequences from the patient's historical data over the past 14 days using the K-means clustering algorithm. Each pattern sequence includes the changing trends of 12 autonomic nervous system indicators, such as heart rate, respiratory rate, SDNN, and LF / HF ratio, within 30 minutes prior to the excretion event, along with corresponding temporal characteristics. These typical patterns cover different scenarios, including regular morning defecation, random afternoon defecation, and emergency nighttime defecation. Then, the Dynamic Time Warping (DTW) algorithm is used to calculate the matching degree S_i between the current 15-minute signal sequence and the i-th typical pattern sequence. The DTW algorithm effectively handles the scaling and offset issues of time series and is more suitable for physiological signal similarity matching than the traditional Euclidean distance. Next, the overall pattern similarity of the current time period, total_similarity=max(S_i), is calculated. The maximum value among all pattern similarities is taken because if the current signal highly matches any typical excretion pattern, it indicates that the patient has a high probability of about to excrete. Finally, the threshold is adjusted using the formula Threshold = Threshold_initial × [1 - (total_similarity / max_similarity_possible)], where max_similarity_possible represents the theoretical maximum similarity, set to 1.0. In the formula, (total_similarity / max_similarity_possible) normalizes the similarity to the [0,1] interval, and (1 - normalized similarity) ensures that the threshold and similarity are inversely proportional. That is, the higher the similarity, the closer the pattern is to the actual excretion state, and the lower the threshold, the higher the warning sensitivity. For example, when total_similarity = 0.6, Threshold = 0.7 × (1 - 0.6 / 1) = 0.28. Clinical trials show that this method can identify the personalized excretion patterns of different patients, reducing the false alarm rate at night to 9.1% and the missed alarm rate to 1.0%, making it suitable for patients with complex and diverse excretion patterns.

[0092] Ninth Embodiment

[0093] In this embodiment, adaptive dynamic calibration of the confidence threshold based on the dynamic confidence assessment model includes:

[0094] Obtain the autonomic nervous system-related signal volatility δ_nervous within the current time period, where δ_nervous is the coefficient of variation of the autonomic nervous system signal within a preset time window;

[0095] Construct a stability score for the signal within a time window: metric_stable = 1 - |δ_nervous / δ_max| where δ_max is the upper limit of the expected range of neural activity; introduce a time decay factor: α_window = max(1 - (t / T_window), α_min) where t represents the length of time since the start of the cycle, T_window is the length of the time window, and α_min is the minimum decay factor greater than 0; finally, dynamically adjust the threshold: Threshold = Threshold_base × (metric_stable × α_window).

[0096] Specifically, this embodiment employs a threshold calibration method based on the volatility and time decay of autonomic neural signals. The system first acquires the autonomic neural-related signals within the current 5-minute time window (using the SDNN index in this embodiment), and calculates the coefficient of variation (COP) of the signal within this window as the volatility δ_nervous. The COP is the ratio of the signal's standard deviation to its mean. Compared to the standard deviation alone, it eliminates the influence of the absolute signal level and more accurately reflects the relative volatility of the signal. Subsequently, a signal stability score is constructed: metric_stable = 1 - |δ_nervous / δ_max|, where δ_max is the upper limit of the expected neural activity range, set to 0.6. The stability score ranges from [0,1]. A higher score indicates a more stable autonomic neural signal and a lower probability of false alarms; therefore, the threshold can be appropriately lowered to improve sensitivity. Next, a time decay factor α_window = max(1 - (t / T_window), α_min) is introduced, where t represents the current time as the prediction period. The initial time period is defined as follows: T_window is the prediction window length, set to 120 minutes; α_min is the minimum decay factor greater than 0, set to 0.3. The time decay factor reflects the increasing probability of patient excretion as the prediction window progresses, therefore the threshold should be gradually reduced. The dynamically adjusted threshold is calculated using the formula Threshold = Threshold_base × (metric_stable × α_window). For example, when δ_nervous = 0.2, metric_stable = 1 - 0.2 / 0.6 ≈ 0.667; when t = 60 minutes, α_window = 1 - 60 / 120 = 0.5; and the final threshold = 0.7 × 0.667 × 0.5 ≈ 0.233. Clinical trials show that this method considers both signal stability and time factors, enabling finer threshold adjustments. The false alarm rate at night is reduced to 9.8%, the missed alarm rate to 1.2%, and the computational complexity is low, making it suitable for real-time operation.

[0097] Tenth Embodiment

[0098] In this embodiment, adaptive dynamic calibration of the confidence threshold based on the dynamic confidence assessment model includes:

[0099] Calculate the correlation ρ between the time of excretion and the corresponding neural activation level in historical data, where ρ is the Pearson correlation coefficient;

[0100] A user-individualized difference parameter γ is introduced to adjust for individual physiological fluctuation characteristics, which are obtained through training with patient historical data;

[0101] The confidence threshold is adjusted according to the following formula: Adjust_Threshold=Threshold_base×(1-γ×|ρ|). This strategy can effectively distinguish the neurological characteristics of different patients and prevent over-triggered problems caused by fixed values.

[0102] Specifically, this embodiment employs a threshold calibration method based on the correlation between neural activation levels and individualized parameters. The system first calculates the Pearson correlation coefficient ρ between the time of each excretion event and the average neural activation level within 15 minutes prior to the event, based on historical data from the past 7 days. Neural activation levels are represented by the LF / HF ratio. The Pearson correlation coefficient ρ ranges from -1 to 1; a larger absolute value indicates a stronger correlation between changes in the patient's neural activation level and the excretion event, and a higher predictive ability of the neural indicators for the excretion event. Subsequently, an individualized difference parameter γ is introduced. This parameter is obtained by training the patient's historical data from the first 3 days after admission using a gradient descent method, and its value ranges from 0 to 1. It is used to adjust for individualized physiological fluctuations. Different patients exhibit significant differences in their autonomic nervous system response patterns; some patients show significant changes in neural activation before excretion, while others show only slight changes. The γ parameter can be adjusted according to each patient's specific situation. The strength of the threshold correction is determined. Finally, the confidence boundary is corrected according to the formula Adjust_Threshold=Threshold_base×(1-γ×|ρ|). In the formula, |ρ| is taken as the absolute value of the correlation coefficient because regardless of whether the neural activation level is increased or decreased, as long as there is a strong correlation with the excretion event, it can be used as a basis for prediction. (1-γ×|ρ|) ensures that the stronger the correlation, the lower the threshold and the higher the warning sensitivity. For example, when ρ=0.7 and γ=0.5, Adjust_Threshold=0.7×(1-0.5×0.7)=0.7×0.65=0.455. Clinical trials show that this method achieves true individualized threshold calibration, can effectively distinguish the characteristics of the neural state of different patients, reduces the false alarm rate at night to 8.3%, and the false negative rate to 0.8%, with the best overall performance. It is especially suitable for bedridden patients with large individual differences and poor results from conventional methods.

[0103] In addition, such as Figure 2 As shown, the present invention also provides a non-contact real-time monitoring and estimation system 200 for bedridden patients' excrement, comprising:

[0104] The non-contact monitoring module 201 is used to acquire time-series data of the patient's historical excretion events and autonomic nerve indicators within the corresponding time period;

[0105] The model building module 202 is used to build a dynamic confidence assessment model based on the time series of the historical excretion events and autonomic nervous system indicators. The model is used to output the prediction confidence of the next excretion event.

[0106] The threshold calibration module 203 is used to adaptively and dynamically calibrate the confidence threshold based on the dynamic confidence assessment model during the window for predicting the next excretion event, combined with the current autonomic nervous system indicators and time information. The core objective of the calibration is to increase the confidence threshold during the period of increased vagal tone at night, so as to reduce the patient over-awakening caused by false alarms.

[0107] The early warning judgment module 204 is used to calculate the discharge prediction confidence level at the current moment and compare it with the calibrated confidence level threshold.

[0108] The early warning output module 205 is used to trigger an early warning for excretion when the excretion prediction confidence level is greater than or equal to the calibrated confidence level threshold, and to generate a corresponding excretion behavior prediction result.

[0109] The functions and implementation methods of each module of the monitoring and estimation system 200 of the present invention have been described above with reference to the accompanying drawings, and will not be repeated here.

[0110] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various variations or substitutions within the technical scope disclosed in this application, and these should all be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A non-contact real-time monitoring and quantification method of excreta for bedridden patients, characterized in that, Includes the following steps: S1: Obtain time-series data of the patient's historical excretion events and autonomic nervous system indicators within the corresponding time periods through non-contact sensors; S2: Based on the time series of the historical excretion events and autonomic nervous system indicators, a dynamic confidence assessment model is established. The model is used to output the prediction confidence of the next excretion event. S3: During the window period for predicting the next excretion event, the confidence threshold is adaptively and dynamically calibrated based on the dynamic confidence assessment model, combined with the current autonomic nervous system indicators and time information. S4: Calculate the excretion prediction confidence level at the current moment, compare it with the calibrated confidence level threshold, and trigger an excretion warning when the excretion prediction confidence level is greater than or equal to the calibrated confidence level threshold, and generate the corresponding excretion behavior prediction result.

2. The method of claim 1, wherein, Step S3, which involves adaptive dynamic calibration of the confidence threshold based on the dynamic confidence assessment model, further includes: Obtain the fluctuation intensity parameter α of the patient's autonomic nervous system indicators within a recent preset time window, where α is the standard deviation of the heart rate variability index within that window. Based on the time-series statistics of historical discharge event intervals, the average interval time T_avg within the current time period is extracted; The confidence threshold is adjusted based on the following formula: Threshold = Threshold_base × (1 + α / α_max) where Threshold_base is the preset base confidence threshold and α_max is the preset maximum fluctuation intensity threshold. If the calculated Threshold exceeds the preset upper limit Max_Threshold, then set it to Max_Threshold; otherwise, keep the calculated value unchanged.

3. The method according to claim 1, characterized in that, Step S3, which involves adaptive dynamic calibration of the confidence threshold based on the dynamic confidence assessment model, further includes: Calculate the rate of change β of the autonomic nervous index within a preset time window before each excretion event in the historical time period, where β is the ratio of the change in the autonomic nervous index within the time window to the initial value; Determine whether the patient is in a low-arousal period at night by combining their sleep status information; The dynamic offset coefficient ψ is calculated according to the following formula: ψ=min(β,β_threshold) / β_max, where β_threshold is the set rate of change threshold and β_max is the maximum allowable rate of change. The original threshold Threshold_base is dynamically compensated to obtain the final adjusted threshold: Threshold = Threshold_base × (1 - ψ × ψ_factor) where ψ_factor is a constant that controls the compensation intensity and has a value range of 0 to 1.

4. The method according to claim 1, characterized in that, Step S3, which involves adaptive dynamic calibration of the confidence threshold based on the dynamic confidence assessment model, further includes: Extract the heart rate variability (HRV) index during the nighttime period; Calculate the standard deviation (STD_HRV) of the short-term trend parameter of HRV; update the confidence threshold for the current time window using the following formula: Threshold = Threshold_initial + k × (STD_HRV - STD_base) where Threshold_initial is the initial confidence threshold, k is a positive scaling factor, and STD_base is the baseline standard deviation of HRV in the patient's awake state. When STD_HRV increases, it indicates that the patient may be in an unstable period with high vagal tone. Therefore, increasing Threshold can reduce false alarms.

5. The method according to claim 1, characterized in that, Step S3, which involves adaptive dynamic calibration of the confidence threshold based on the dynamic confidence assessment model, further includes: Obtain the timestamp of the patient's most recent excretion and calculate the time interval time_diff between the current time and the last excretion. The time-domain weight w(t) = e^(-λ·t) is constructed based on the excretion frequency, where λ is a positive decay coefficient and t is the time interval between the current moment and the last excretion. The threshold adjustment range is determined based on the following conditions: if time_diff > T_ref, then Threshold = Threshold_base × w(t); otherwise, Threshold_base remains unchanged; where T_ref is the reference time length for defining the active excretion period.

6. The method according to claim 1, characterized in that, Step S3, which involves adaptive dynamic calibration of the confidence threshold based on the dynamic confidence assessment model, further includes: Calculate the average excretion frequency f_avg over the past 24 hours; By analyzing the covariance between heart rate and respiratory rate, the current activity parameter σ of the nervous system is obtained, where σ is the absolute value of the Pearson correlation coefficient between heart rate and respiratory rate. The dynamic confidence compensation factor is calculated using the following formula: compensation_factor=exp(-σ / σ_ref), where σ_ref is the mean neurological activity of the patient in a conscious and stable state; the final threshold is: Threshold=Threshold_base×compensation_factor to address the problem of over-triggering under conditions of high vagal nerve activity.

7. The method according to claim 1, characterized in that, Step S3, which involves adaptive dynamic calibration of the confidence threshold based on the dynamic confidence assessment model, further includes: Collect historical data on the time periods of recent excretion events, and label the time periods as high_risk_period; A similarity metric is established for these historical time points based on their temporal proximity, and a similarity parameter s(t) is calculated. The s(t) is the weighted sum of the Gaussian kernel functions of the current time and each historical high-risk period. The confidence adjustment value for the time window is calculated using the following formula: ΔThreshold=(1-s(t))×A, where A is a positive compensation coefficient; When s(t) > s_threshold, the final threshold is set to Threshold_base + ΔThreshold; otherwise, Threshold_base remains unchanged. Here, s_threshold is the similarity criterion.

8. The method according to claim 1, characterized in that, Step S3, which involves adaptive dynamic calibration of the confidence threshold based on the dynamic confidence assessment model, further includes: Obtain the continuous autonomic nervous system monitoring dataset D_t within the current period and perform frequency domain analysis on it to extract the low-frequency energy proportion E_low and the high-frequency energy proportion E_high; where E_low is the energy proportion in the 0.04-0.15Hz frequency band, representing vagal nerve tension, and E_high is the energy proportion in the 0.15-0.4Hz frequency band, representing sympathetic nerve tension. The threshold adjustment strategy is determined based on the following relationship: if E_low / E_high>β_ratio, it indicates that the patient is in a state of high vagal tone, and the sensitivity needs to be reduced and the threshold increased; otherwise, the sensitivity should be appropriately increased and the threshold decreased. The final threshold is: Threshold=Threshold_base×(E_low / E_high)^ξ, where ξ is a positive tuning coefficient and β_ratio is the critical ratio of vagal tone.

9. The method according to claim 1, characterized in that, Step S3, which involves adaptive dynamic calibration of the confidence threshold based on the dynamic confidence assessment model, further includes: Multiple typical excretion pattern sequences are constructed from historical data; a similarity metric S_i is used to measure the degree of matching between the current signal and the i-th pattern; Calculate the overall pattern similarity for the current time period: total_similarity=max(S_i), where max(S_i) represents the maximum value among all pattern similarities; The threshold can be adjusted as follows: Threshold = Threshold_initial × [1 - (total_similarity / max_similarity_possible)] where max_similarity_possible represents the theoretical maximum similarity. If the similarity is higher, it means that the pattern is closer to the actual excretion state. In this case, the threshold should be lowered to increase sensitivity. Conversely, the threshold should be increased to decrease sensitivity.

10. A non-contact real-time monitoring and estimation system for excrement of bedridden patients, characterized in that, include: A non-contact monitoring module is used to acquire time-series data of the patient's historical excretion events and autonomic nervous system indicators within the corresponding time periods; The model building module is used to build a dynamic confidence assessment model based on the time series of the historical excretion events and autonomic nervous system indicators. The model is used to output the prediction confidence of the next excretion event. The threshold calibration module is used to adaptively and dynamically calibrate the confidence threshold based on the dynamic confidence assessment model during the window for predicting the next excretion event, combined with current autonomic nerve indicators and time information. The core objective of the calibration is to increase the confidence threshold during the period of increased vagal tone at night, so as to reduce patient over-awakening caused by false alarms. The early warning judgment module is used to calculate the confidence level of the discharge prediction at the current moment and compare it with the calibrated confidence level threshold. The early warning output module is used to trigger an early warning for excretion when the confidence level of the excretion prediction is greater than or equal to the calibrated confidence level threshold, and to generate a corresponding excretion behavior prediction result.

Images