An AI voice and flexible sensing fusion-based medical care mattress control method

Abstract

The application discloses a medical care mattress control method based on AI voice and flexible sensing fusion, and relates to the technical field of intelligent medical care; the method synchronously collects the pressure distribution, body surface temperature and environmental audio stream of a bed occupant through a flexible pressure sensor array, a distributed temperature sensor group and an AI voice interaction module; a data processing module constructs a pressure and time integral model and a core and terminal temperature gradient model, predicts pressure sore risks and body temperature abnormal events, extracts respiratory periodic signals and voice prosody features from the audio stream, judges respiratory abnormalities and emotional abnormal events; a personalized benchmark model is established during the learning period to dynamically adjust the trigger threshold; a hierarchical control strategy is executed according to the judged abnormal events. The application replaces the traditional fixed threshold judgment with multi-modal perception data fusion and trend prediction, realizes early active discovery of pressure sores, body temperature, respiration and emotional abnormalities, and significantly improves the initiative and accuracy of nursing.

Description

Technical Field

[0001] This invention relates to the field of intelligent medical care technology, specifically to a method for controlling a medical care mattress based on the fusion of AI voice and flexible sensing. Background Technology

[0002] Currently, most call and monitoring devices for elderly care and postoperative bedridden patients are manually triggered, such as bedside call buttons and pull-cord call devices. These rely on the patient's manual operation ability, and some patients with limited mobility are unable to reach them due to stiffness or weakness, leading to call failure. To address this issue, some existing technologies utilize nursing mattresses with intelligent sensors. These mattresses typically incorporate a single type of sensor, such as a pressure sensor or a temperature sensor alone. In terms of data processing, existing devices often use fixed thresholds for anomaly detection, such as setting fixed turning intervals or fixed body temperature alarm values. Regarding voice interaction, while some devices incorporate voice recognition functionality, they are limited to simple command responses.

[0003] The existing technology mainly has the following specific problems:

[0004] 1. The sensor configuration is limited and cannot simultaneously acquire multimodal data such as pressure distribution, body surface temperature and ambient audio, resulting in incomplete patient status perception dimensions and making it difficult to achieve simultaneous monitoring and prediction of multiple events such as pressure ulcer risk, abnormal body temperature, abnormal breathing and emotional abnormalities.

[0005] 2. The data processing method is simple, only using fixed thresholds for judgment, without establishing a dynamic model to analyze the changing trends of monitoring data, lacking the ability to predict risks in advance from trends, and can only trigger alarms after an anomaly occurs and continues for a period of time, thus failing to achieve early intervention.

[0006] 3. Lack of personalized adaptive mechanism, unable to automatically adjust the judgment threshold according to the patient's long-term stress distribution habits, body temperature diurnal rhythm and behavioral patterns, resulting in different patients using the same set of standards, affecting the accuracy of judgment;

[0007] 4. The interaction method is simple and passive, only able to respond to patient-initiated calls, and unable to proactively trigger graded control strategies based on predicted abnormal events, resulting in insufficient nursing initiative and timeliness. Summary of the Invention

[0008] (a) Technical problems to be solved

[0009] To address the shortcomings of existing technologies, this invention provides a medical care mattress control method based on the fusion of AI voice and flexible sensing, which solves the problems mentioned in the background.

[0010] (II) Technical Solution

[0011] To achieve the above objectives, the present invention provides the following technical solution: a method for controlling a medical care mattress based on the fusion of AI voice and flexible sensing, comprising:

[0012] S1: Continuously collect pressure distribution data of the bedridden person through a flexible pressure sensor array; collect temperature data of different body surface locations of the bedridden person through a distributed temperature sensor group; capture ambient audio stream through the microphone array of the AI ​​voice interaction module;

[0013] S2: The data processing module inside the mattress establishes and updates a dynamic pressure distribution model of the pressure area based on the pressure distribution data; establishes a temperature gradient model between the core area and the distal area based on the temperature data of different body surface locations; and extracts respiratory-related periodic audio signals and speech prosody features based on the environmental audio stream.

[0014] S3: Based on the changing trend of the cumulative pressure value in the dynamic pressure distribution model, predict pressure ulcer risk events; based on the changing trend of temperature difference in the temperature gradient model, predict abnormal body temperature events; based on the respiratory-related periodic audio signal, determine abnormal respiratory events; based on the speech prosody features, determine abnormal emotional events.

[0015] S4: During the preset learning period, record the bedridden person's stress distribution habits, body temperature diurnal rhythm and voice interaction behavior patterns, establish a personalized benchmark model, and use the personalized benchmark model to dynamically adjust the trigger threshold of each event judgment in step S3.

[0016] S5: Based on any event determined in step S3, and combined with the trigger threshold adjusted by the personalized baseline model in step S4, actively trigger the call and interaction module in the mattress to execute the corresponding preset control strategy.

[0017] Furthermore, the establishment and updating of the dynamic pressure distribution model of the pressure-bearing area in step S2 further includes: constructing a pressure-time integral model for specific pressure-bearing areas of the bedridden person's body, and continuously calculating the cumulative pressure value of each pressure-bearing area according to the following formula. :

[0018] ;

[0019] In the formula, This represents the real-time pressure value of the pressurized area at time 10:00. This marks the starting point of the current pressure event. For the current moment, From to The cumulative integral of pressure over time within a given period;

[0020] The prediction of pressure ulcer risk events in step S3 further includes: monitoring the cumulative pressure value of each pressure-affected area. Within a preset time period, if the growth rate of any pressure area exceeds the personalized or default safety threshold, it is considered a predicted pressure ulcer risk event.

[0021] Furthermore, the establishment of the temperature gradient model between the core region and the terminal region in step S2 further includes: obtaining the first temperature value of the core region of the torso. and the second temperature value of the extremity region Calculate the temperature difference in real time using the following formula. :

[0022] ;

[0023] In the formula, This represents the real-time temperature value of the core region of the torso. This refers to the real-time temperature value of the extremity region. This refers to the real-time temperature difference between the core area of ​​the torso and the distal areas of the limbs.

[0024] The prediction of abnormal body temperature events in step S3 further includes: when the first temperature value is detected... The temperature difference rises simultaneously. When the temperature continues to decrease and exceeds the preset threshold, it is determined that an abnormal body temperature event has been predicted.

[0025] Furthermore, the separation and extraction of respiratory-related periodic audio signals in step S2 further includes: when the AI ​​voice interaction module is in a low-power monitoring state, using an adaptive filtering algorithm to separate periodic audio signals related to human breathing actions from the ambient audio stream; the judgment of abnormal breathing events in step S3 further includes: performing time-frequency analysis on the periodic audio signals to extract respiratory waveform feature parameters, including respiratory rate, the ratio of inspiratory to expiratory phase duration, and the presence or absence of intermittent wheezing patterns; when the respiratory rate exceeds a preset safety range, or a preset abnormal breathing sound pattern is identified, it is determined to be a respiratory abnormal event.

[0026] Furthermore, the extraction of speech prosodic features in step S2 further includes: identifying and extracting non-verbal vocal fragments emitted by the bedridden person from the environmental audio stream, or deconstructing the language instructions to extract their prosodic features, wherein the prosodic features include pitch, speech rate, voice intensity, and tremor; the judgment of abnormal emotional events in step S3 further includes: inputting the extracted prosodic features into a preset emotion classification model to perform multi-dimensional emotion dimension evaluation, generating a current emotion state label, and determining an abnormal emotional event when the current emotion state label matches a preset negative emotion condition.

[0027] Furthermore, the preset control strategy mentioned in step S5 includes at least the following: a first control strategy, which, when any event is predicted but has not yet reached the emergency alarm level, actively sends an interrogative voice to the bedridden person through the AI ​​voice interaction module, and after obtaining affirmative voice confirmation, sends a call signal to the care terminal; and a second control strategy, which, when an emergency event is predicted, directly generates alarm information containing the event type and real-time sensor data, sends it to the designated terminal, and simultaneously activates the local audible and visual alarm.

[0028] Furthermore, it also includes: the wireless communication modules of multiple medical care mattresses self-organize to form a local mesh network; when a medical care mattress detects an emergency and its connection to the wide area network fails, it forwards the alarm information to the target terminal through the local mesh network via other networked medical care mattress agents.

[0029] Furthermore, it also includes: in postoperative monitoring mode, the data processing module performs multimodal data fusion based on the heart rate variability features extracted from the pressure distribution data, the judgment results of the respiratory abnormal events, and the judgment results of the emotional abnormal events, to generate a comprehensive assessment report on postoperative pain and recovery status.

[0030] (III) Beneficial Effects

[0031] (1) This invention uses a flexible pressure sensor array, a distributed temperature sensor group and an AI voice interaction module to simultaneously collect three modal data: pressure distribution, body surface temperature and environmental audio stream. This solves the problem of single sensor configuration and inability to achieve multimodal physiological data collaborative collection in the prior art, and provides a data foundation for multidimensional perception and comprehensive judgment of the state of bedridden patients.

[0032] (2) By establishing a dynamic pressure distribution model and a core and end temperature gradient model, this invention predicts pressure ulcer risk events and abnormal body temperature events based on the changing trends of pressure accumulation and temperature difference. This replaces the fixed threshold judgment method of the prior art and can make early predictions when the risk has not yet occurred but the trend is forming. This realizes the shift from passive response to active prediction and significantly improves the timeliness and accuracy of nursing care.

[0033] (3) By separating the respiratory-related periodic audio signal from the ambient audio stream and extracting the speech prosody features, the present invention can simultaneously realize the judgment of respiratory abnormality and emotional abnormality using existing microphone array hardware without the need to add a special sensor. This expands the functional dimensions of the system and solves the problem that the existing technology lacks the ability to actively perceive the respiratory and emotional states of bedridden patients.

[0034] (4) This invention records the behavior patterns and physiological rhythms of bedridden patients during a preset learning period, establishes a personalized benchmark model, and dynamically adjusts the trigger thresholds for each event judgment. This solves the problems of poor adaptability and high false alarm rate caused by the use of fixed judgment standards for all patients in the prior art, enabling the system to adapt to the individual differences of different patients and become more accurate with use. Attached Figure Description

[0035] Figure 1 This is a flowchart illustrating the process of this invention;

[0036] Figure 2 This is a schematic diagram of the workflow of step S2 in this invention;

[0037] Figure 3 This is a schematic diagram of the installation of the flexible pressure sensor and distributed temperature sensor of the present invention;

[0038] Figure 4 This is a schematic diagram of the feedback interface of the AI ​​voice interaction module of the present invention. Detailed Implementation

[0039] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0040] Please see Figures 1 to 4 As shown, the embodiments of the present invention provide the following technical solutions:

[0041] A method for controlling a medical care mattress based on the fusion of AI voice and flexible sensing includes the following steps, please refer to [link / reference]. Figure 1 and Figure 3 :

[0042] S1: Continuously collect pressure distribution data of the bedridden person through a flexible pressure sensor array; collect temperature data of different body surface locations of the bedridden person through a distributed temperature sensor group; and capture ambient audio stream through the microphone array of the AI ​​voice interaction module.

[0043] The flexible pressure sensor array uses liquid metal or conductive polymer as the substrate material, designed with a 256-point array layout and a thickness controlled between 0.1mm and 0.5mm. It is laid flat inside the mattress, so there is no foreign body sensation when in contact with the human body. The sensor array has a sensitivity of 2.15 kPa⁻¹ (in the range of 0 to 10 kPa), a response time of no more than 30ms, and a stretchability of no less than 200%, enabling it to accurately capture the subtle body movements and pressure distribution changes of the bedridden person.

[0044] The distributed temperature sensor array employs high-precision thermistor sensors integrated into a flexible pressure sensor array, with 3 to 5 detection points evenly distributed. It achieves a measurement accuracy of ±0.05℃, a measurement range of 35℃ to 42℃, and a response time of no more than 3 seconds, enabling real-time acquisition of temperature data from different body surface locations of bedridden individuals. For example, at least one temperature measurement point is located in the core trunk region (e.g., the center of the back), and at least one temperature measurement point is located in the extremity region (e.g., the lower leg) to support subsequent temperature gradient analysis.

[0045] The AI ​​voice interaction module adopts a lightweight AI voice model customized for medical scenarios, which can be optimized based on Qwen3-ASR-1.7B and integrates the basic diagnosis and popular science capabilities of a large medical model. The module integrates a microphone array as a voice acquisition unit, supports omnidirectional sound pickup and noise filtering, and continuously captures ambient audio streams in low-power monitoring mode. The module has a preset dedicated wake word, such as "Xiao Yi", with a wake-up response time of no more than 0.5 seconds. It supports the recognition of at least 22 Chinese dialects, and the accuracy rate of core medical term recognition is no less than 94.2%, with an accuracy rate of no less than 90% in noisy environments.

[0046] S2: The data processing module within the mattress establishes and updates a dynamic pressure distribution model of the pressure-bearing area based on the pressure distribution data; establishes a temperature gradient model between the core and distal regions based on temperature data from different body surface locations; and extracts respiratory-related periodic audio signals and speech prosody features based on the ambient audio stream. Please refer to [link / reference] for details. Figure 2 .

[0047] The data processing module uses a low-power MCU chip (such as the GD32VF103 series) and is electrically connected to the flexible pressure sensor array, the distributed temperature sensor group and the AI ​​voice interaction module, respectively, and is responsible for receiving and processing the data collected by each sensor.

[0048] Step S2, which involves establishing and updating the dynamic pressure distribution model of the pressure-bearing area, further includes: constructing a pressure-time integral model for specific pressure-bearing areas of the bedridden person's body, and continuously calculating the cumulative pressure value of each pressure-bearing area according to the following formula. :

[0049] ;

[0050] In the formula, This represents the real-time pressure value of the pressurized area at time 10:00. This marks the starting point of the current pressure event. For the current moment, From to The cumulative integral of pressure over time within a given period.

[0051] The specific pressure-affected areas refer to key areas on the body of bedridden patients that are prone to pressure sores, such as the sacral region, heel region, and scapular region. The data processing module automatically identifies and marks these key pressure-affected areas by analyzing the spatial distribution characteristics of pressure values ​​at each point in the 256-point pressure array, and independently establishes a pressure-time integral model for each area. For example, when the bedridden patient maintains a supine position, the pressure value in the sacral region remains at a high level, and the data processing module records the time from the start of the pressure increase in this area. From now time All pressure data up to date are used to calculate the cumulative pressure value through real-time integration. .

[0052] Step S2, which establishes a temperature gradient model between the core region and the terminal region, further includes: obtaining the first temperature value of the core region of the torso. and the second temperature value of the extremity region Calculate the temperature difference in real time using the following formula. :

[0053] ;

[0054] In the formula, This represents the real-time temperature value of the core region of the torso. This refers to the real-time temperature value of the extremity region. This represents the real-time temperature difference between the core area of ​​the torso and the extremity areas.

[0055] For example, Take the temperature value at the central measurement point on the back, which represents the core body temperature; The temperature value measured at the calf represents the peripheral body temperature; under normal physiological conditions, there is a certain temperature gradient between the core and the extremities. The core temperature usually remains within a stable range. When a bedridden patient is in the pre-febrile stage, the core temperature... It begins to rise, while peripheral vasoconstriction causes the temperature at the extremities to increase. The decline or rise is lagging, making The temperature difference is showing a continuous shrinking trend. By monitoring this dynamic change in temperature difference, abnormal body temperature trends can be identified before the core temperature reaches the traditional fever threshold.

[0056] The step S2 of separating and extracting the respiratory-related periodic audio signal further includes: when the AI ​​voice interaction module is in a low-power monitoring state, using an adaptive filtering algorithm to separate the periodic audio signal related to human breathing from the ambient audio stream.

[0057] Specifically, the microphone array of the AI ​​voice interaction module continuously captures the ambient audio stream in a low-power monitoring state. The data processing module employs an adaptive filtering algorithm to first model and filter out steady-state background noise in the ambient audio stream. Then, based on the quasi-periodic characteristics of breathing sounds in the time-frequency domain, it extracts signal components with regular repetition features from the filtered audio stream, i.e., breathing-related periodic audio signals. The adaptive filtering algorithm can be the least mean square algorithm or the recursive least squares algorithm, etc.

[0058] Step S2, which involves extracting speech prosodic features, further includes: identifying and extracting non-verbal vocal fragments from the bedridden person in the ambient audio stream, or deconstructing the language instructions to extract their prosodic features, which include pitch, speech rate, voice intensity, and tremor.

[0059] For example, when a bedridden person utters nonverbal sounds such as "sigh" or "hmm," or verbal commands such as "I feel uncomfortable" or "I feel unwell," the AI ​​voice interaction module, while recognizing the semantic content, extracts prosodic feature parameters such as the fundamental frequency, the range of fundamental frequency variation, the speech rate, the energy envelope, and the frequency tremor of the speech segment. The fundamental frequency reflects the pitch, the speech rate reflects how fast or slow one speaks, the energy envelope reflects the changes in the intensity of the sound, and the frequency tremor reflects the stability of the sound.

[0060] S3: Based on the changing trend of the cumulative pressure value in the dynamic pressure distribution model, predict pressure ulcer risk events; based on the changing trend of temperature difference in the temperature gradient model, predict abnormal body temperature events; based on the respiratory-related periodic audio signal, determine abnormal respiratory events; based on the speech prosody features, determine abnormal emotional events.

[0061] The prediction of pressure ulcer risk events in step S3 further includes: monitoring the cumulative pressure value of each pressure-affected area. Within a preset time period, if the growth rate of any pressure area exceeds the personalized or default safety threshold, it is considered a predicted pressure ulcer risk event.

[0062] Taking the sacral region as an example, assuming a default safety threshold of a certain empirical value, while a bedridden person maintains a supine position, the sacral region... The value continues to increase. If within a certain monitoring window, If the growth rate of the pressure value exceeds the safety threshold, it indicates that the pressure accumulation in the tissue of that area has reached a dangerous level, and the data processing module will predict that there is a risk of pressure sores in that area. Compared with the traditional approach that only judges whether it has exceeded 2 hours, this method uses the integral cumulative value of pressure and time and the growth rate to predict the trend, which can issue an early warning when the risk is accumulating rapidly before tissue damage occurs, and achieve earlier intervention.

[0063] The prediction of abnormal body temperature events in step S3 further includes: when the first temperature value is detected... The temperature difference rises simultaneously. When the temperature continues to decrease and exceeds the preset threshold, it is determined that an abnormal body temperature event has been predicted.

[0064] For example, the preset temperature difference threshold is 0.5℃. When The temperature rose from 36.5℃ to 37.0℃ within 30 minutes, while... The temperature dropped from 34.5℃ to 34.0℃. If the temperature drops from 2.0℃ to 3.0℃ (this is an example value; the actual threshold needs to be calibrated based on clinical data), and exceeds the preset threshold, the system will determine it as an abnormal body temperature event, indicating that the bedridden person may be in the early stage of fever. At this time, the core temperature has not yet reached the traditional fever standard of 37.3℃, but the abnormality can be identified in advance through temperature difference trend analysis.

[0065] The judgment of abnormal breathing events in step S3 further includes: performing time-frequency analysis on the periodic audio signal to extract respiratory waveform feature parameters, including respiratory rate, the ratio of the duration of the inspiratory phase to the expiratory phase, and whether there is an intermittent wheezing pattern. When the respiratory rate exceeds a preset safe range, or a preset abnormal breathing sound pattern is identified, it is judged as an abnormal breathing event.

[0066] For example, the normal resting respiratory rate of an adult is typically 12 to 20 breaths per minute. If the extracted respiratory rate consistently exceeds 24 breaths per minute or falls below 8 breaths per minute, the system identifies it as an abnormal respiratory event. Similarly, if the ratio of inspiratory to expiratory phase duration deviates significantly, or if intermittent wheezing patterns are detected—characterized by irregular, brief interruptions in the respiratory waveform followed by rapid recovery—the system also identifies these as abnormal respiratory events. This method utilizes existing hardware such as a microphone array to achieve non-contact respiratory monitoring without requiring additional sensors.

[0067] The step S3 of judging abnormal emotional events further includes: inputting the extracted prosodic features into a preset emotion classification model, performing multi-dimensional emotion dimension evaluation, generating a current emotion state label, and judging it as an abnormal emotional event when the current emotion state label matches a preset negative emotion condition.

[0068] The pre-defined emotion classification model can be built based on a deep learning network and pre-trained using speech samples labeled with emotion tags. It can output emotion state classification results based on the input prosodic feature vector, such as labels like calm, anxiety, frustration, and pain. When the generated emotion tag is a negative emotion such as anxiety, frustration, or pain, and the duration exceeds a preset duration, the system determines it as an abnormal emotion event and triggers subsequent reassuring interaction strategies. For example, if a bedridden patient utters a trembling and slow "I feel uncomfortable" voice for a long time due to postoperative pain, the system can identify it as a painful emotion and proactively initiate a reassuring voice interaction or call a nurse.

[0069] S4: During the preset learning period, record the bedridden person's stress distribution habits, body temperature diurnal rhythm, and voice interaction behavior patterns to establish a personalized baseline model, and use the personalized baseline model to dynamically adjust the trigger thresholds for each event judgment in step S3.

[0070] During the preset learning period, the data processing module continuously records: the bedridden person's habitual sleeping positions and corresponding pressure distribution characteristics at different times; the daily fluctuation pattern of body temperature; and the active periods and habitual language patterns of daily voice interaction. Based on the above data, the system generates a personalized baseline model for the bedridden person. For example, if a bedridden person habitually maintains a right-side lying position for extended periods, the baseline value of pressure accumulation rate in their right hip area is higher than that of the general population. Based on this, the system automatically and appropriately raises the pressure ulcer risk trigger threshold for that area to reduce false positive alarms. As another example, if a bedridden person has a low basal body temperature, with normal axillary body temperature maintained at around 36.0℃, the system automatically lowers the absolute threshold for abnormal body temperature from the default 37.3℃ to 36.8℃, while also considering their individual temperature difference. The baseline fluctuation range is adjusted to determine the temperature gradient threshold.

[0071] S5: Based on any event determined in step S3, and combined with the trigger threshold adjusted by the personalized baseline model in step S4, actively trigger the call and interaction module in the mattress to execute the corresponding preset control strategy.

[0072] The preset control strategy mentioned in step S5 includes at least the following:

[0073] The first control strategy involves proactively issuing an inquiry-based voice message to the bedridden patient via the AI ​​voice interaction module when any event is predicted but has not yet reached the emergency alarm level. After receiving affirmative voice confirmation, the system then sends a call signal to the caregiver's terminal. For example, if the system predicts that the risk of pressure sores in the sacral region is accumulating but has not yet reached the emergency threshold, it proactively announces, "Do you feel some discomfort in your sacral region? Do you need help turning over?" If the bedridden patient replies with affirmative words such as "yes" or "okay," the system sends a call signal to the nurse station terminal or doctor's mobile app via the wireless communication module, while simultaneously activating the audio-visual prompts of the feedback module. If the bedridden patient replies "no" or does not respond, the system continues monitoring.

[0074] The second control strategy, when an emergency is predicted, directly generates an alarm message containing the event type and real-time sensor data, sends it to a designated terminal, and simultaneously activates a local audible and visual alarm. For example, if a respiratory rate exceeding 30 breaths per minute is detected, accompanied by intermittent wheezing, or if a pressure sensor detects a pressure surge exceeding 50 kPa, indicating a possible fall, the system classifies it as an emergency. The alarm message includes key information such as the event type, trigger time, and real-time respiratory rate or pressure data, enabling medical staff to quickly understand the situation and respond. The local audible and visual alarm is activated by flashing LED indicators on the edge of the mattress and a buzzer, ensuring that nurses notice the alarm immediately even if they are not checking the terminal.

[0075] Please refer to details. Figure 4 After the call and interaction module executes the preset control strategy, the nurse or doctor sends a response message through the terminal. The response message is transmitted to the mattress via the wireless communication module, and the feedback module broadcasts the response content to the bedridden person through AI voice synthesis, such as "The nurse has received the call and is on her way, please wait a moment," to alleviate the bedridden person's anxiety. In addition, the AI ​​voice interaction module also supports interactive companionship. The bedridden person can wake up the AI ​​voice module with the wake word "Xiao Yi" and then make interactive requests, such as "Tell me a story," "Explain my condition," or "Remind me to take my medicine in half an hour." The AI ​​voice module responds in real time based on a preset medical science database and interactive content database, realizing functions such as story broadcasting, explanation of symptoms, and medication reminders. For example, if the bedridden person asks, "What should I do if I have a fever?" the AI ​​voice module can broadcast based on the built-in medical knowledge base: "Fever may be caused by infection. Please keep hydrated. The nurse has received your abnormal temperature alert and will check you as soon as possible."

[0076] S6: A local mesh network is self-organized among the wireless communication modules of multiple medical care mattresses; when a medical care mattress detects an emergency and its connection to the wide area network fails, it forwards the alarm information to the target terminal through the local mesh network via other networked medical care mattress agents.

[0077] The wireless communication module adopts a dual-mode design of Bluetooth 5.0 and WiFi. The Bluetooth module supports establishing a mesh network connection with neighboring mattresses, while the WiFi module supports wide area network (WAN) connections with the nurse station terminal or cloud server. Multiple mattresses in the same or adjacent wards can self-organize into a local mesh network via the Bluetooth module. When a mattress fails to connect to the WAN due to a WiFi signal interruption or router malfunction, the generated emergency alarm information is transmitted via the local mesh network through a multi-hop forwarding method to neighboring mattresses that still maintain WAN connectivity. These neighboring mattresses then relay the alarm to the nurse station terminal via WiFi, ensuring that emergency calls are delivered under all circumstances.

[0078] Meanwhile, the wireless communication module has a watchdog retry mechanism. When a WiFi connection is detected to be lost, it automatically attempts to reconnect, with a retrieval count of 3. If the connection cannot be restored after 3 retries, it switches to the aforementioned proxy forwarding mode. In addition, a local redundant buzzer alarm is triggered simultaneously with all alarm information sent, ensuring that even in the extreme case of a complete network outage, caregivers can still be notified of alarms through audible and visual alerts.

[0079] S7: In postoperative monitoring mode, the data processing module performs multimodal data fusion based on the heart rate variability features extracted from the pressure distribution data, the judgment results of the respiratory abnormality event, and the judgment results of the emotional abnormality event, and generates a comprehensive assessment report on postoperative pain and recovery status.

[0080] Postoperative monitoring is a key application scenario for this solution. In this mode, the data processing module extracts minute pressure fluctuation signals caused by heartbeats from high-frequency sampling of pressure distribution data. After filtering and peak detection algorithms, it calculates heart rate variability characteristics, which reflect the activity state of the autonomic nervous system and are correlated with pain stress. Simultaneously, combining the respiratory and emotional abnormality assessment results from step S3, the data processing module uses a weighted fusion algorithm to comprehensively assess the current pain level and postoperative recovery trend of the bedridden patient. For example, when heart rate variability decreases, respiratory rate increases, and the emotional state label is "painful," the system comprehensively judges that the bedridden patient may be in a moderate to severe pain state, generating a comprehensive postoperative pain and recovery status assessment report, including pain score, recovery trend, and recommended intervention measures, which is sent to the doctor's terminal to assist clinical decision-making.

[0081] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0082] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

[0083] The various embodiments of this application have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or improvement of the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A method for controlling a medical care mattress based on the fusion of AI voice and flexible sensing, characterized in that, Includes the following steps: S1: Continuously collect pressure distribution data of the bedridden person through a flexible pressure sensor array; collect temperature data of different body surface locations of the bedridden person through a distributed temperature sensor group; capture ambient audio stream through the microphone array of the AI ​​voice interaction module; S2: The data processing module inside the mattress establishes and updates a dynamic pressure distribution model of the pressure area based on the pressure distribution data; establishes a temperature gradient model between the core area and the distal area based on the temperature data of different body surface locations; and extracts respiratory-related periodic audio signals and speech prosody features based on the environmental audio stream. S3: Based on the changing trend of the cumulative pressure value in the dynamic pressure distribution model, predict pressure ulcer risk events; Based on the temperature difference change trend in the temperature gradient model, predict abnormal body temperature events; Based on the respiratory-related periodic audio signals, respiratory abnormalities are identified; Based on the aforementioned speech prosody features, determine abnormal emotional events; S4: During the preset learning period, record the bedridden person's stress distribution habits, body temperature diurnal rhythm and voice interaction behavior patterns, establish a personalized benchmark model, and use the personalized benchmark model to dynamically adjust the trigger threshold of each event judgment in step S3. S5: Based on any event determined in step S3, and combined with the trigger threshold adjusted by the personalized baseline model in step S4, actively trigger the call and interaction module in the mattress to execute the corresponding preset control strategy.

2. The medical care mattress control method based on the fusion of AI voice and flexible sensing according to claim 1, characterized in that, Step S2, which involves establishing and updating the dynamic pressure distribution model of the pressure-bearing area, further includes: constructing a pressure-time integral model for specific pressure-bearing areas of the bedridden person's body, and continuously calculating the cumulative pressure value of each pressure-bearing area according to the following formula. : ； In the formula, This represents the real-time pressure value of the pressurized area at time 10:

00. This marks the starting point of the current pressure event. For the current moment, From to The cumulative integral of pressure over time within a given period; The prediction of pressure ulcer risk events in step S3 further includes: monitoring the cumulative pressure value of each pressure-affected area. Within a preset time period, if the growth rate of any pressure area exceeds the personalized or default safety threshold, it is considered a predicted pressure ulcer risk event.

3. The medical care mattress control method based on the fusion of AI voice and flexible sensing according to claim 1, characterized in that, Step S2, which establishes a temperature gradient model between the core region and the terminal region, further includes: obtaining the first temperature value of the core region of the torso. and the second temperature value of the extremity region Calculate the temperature difference in real time using the following formula. : ； In the formula, This represents the real-time temperature value of the core region of the torso. This refers to the real-time temperature value of the extremity region. This refers to the real-time temperature difference between the core area of ​​the torso and the distal areas of the limbs. The prediction of abnormal body temperature events in step S3 further includes: when the first temperature value is detected... The temperature difference rises, and at the same time... When the temperature continues to decrease and exceeds the preset threshold, it is determined that an abnormal body temperature event has been predicted.

4. The medical care mattress control method based on the fusion of AI voice and flexible sensing according to claim 1, characterized in that, Step S2, which involves separating and extracting the periodic audio signal related to breathing, further includes: when the AI ​​voice interaction module is in a low-power monitoring state, using an adaptive filtering algorithm to separate the periodic audio signal related to human breathing from the ambient audio stream; Step S3, which involves judging abnormal breathing events, further includes: performing time-frequency analysis on the periodic audio signal to extract respiratory waveform feature parameters, including respiratory rate, the ratio of the duration of the inspiratory phase to the expiratory phase, and whether there is an intermittent panting pattern. When the respiratory rate exceeds a preset safe range, or a preset abnormal breathing sound pattern is identified, it is judged as an abnormal breathing event.

5. The medical care mattress control method based on the fusion of AI voice and flexible sensing according to claim 1, characterized in that, Step S2, which involves extracting speech prosodic features, further includes: identifying and extracting non-verbal vocal fragments emitted by the bedridden person from the environmental audio stream, or deconstructing the language instructions to extract their prosodic features, which include pitch, speech rate, voice intensity, and tremor. Step S3, which involves judging abnormal emotional events, further includes: inputting the extracted prosodic features into a preset emotion classification model to perform multi-dimensional emotion assessment and generate a current emotion state label. When the current emotion state label matches a preset negative emotion condition, it is judged as an abnormal emotional event.

6. The medical care mattress control method based on the fusion of AI voice and flexible sensing according to claim 1, characterized in that, The preset control strategies mentioned in step S5 include at least the following: a first control strategy, which, when any event is predicted but has not yet reached the emergency alarm level, actively sends an interrogative voice to the bedridden person through the AI ​​voice interaction module, and after obtaining affirmative voice confirmation, sends a call signal to the care terminal; and a second control strategy, which, when an emergency event is predicted, directly generates alarm information containing the event type and real-time sensor data, sends it to the designated terminal, and simultaneously activates the local audible and visual alarm.

7. The medical care mattress control method based on the fusion of AI voice and flexible sensing according to claim 1, characterized in that, Also includes: The wireless communication modules of multiple medical care mattresses self-organize to form a local mesh network; When a medical care mattress detects an emergency and its connection to the wide area network fails, it forwards the alarm information to the target terminal via the local mesh network through other networked medical care mattress agents.

8. The medical care mattress control method based on the fusion of AI voice and flexible sensing according to claim 1, characterized in that, Also includes: In postoperative monitoring mode, the data processing module performs multimodal data fusion based on the heart rate variability features extracted from the pressure distribution data, the judgment results of the respiratory abnormality events, and the judgment results of the emotional abnormality events, and generates a comprehensive assessment report on postoperative pain and recovery status.

Images