Multi-mode sensing ward monitoring system and method
By using multi-source sensing modules and deep neural networks to fuse multi-modal features, the problem of patient identification and data binding in multi-modal sensing ward monitoring systems has been solved, achieving high-precision, privacy-protected, and non-intrusive monitoring, thus improving the identification accuracy and clinical practical value of ward monitoring systems.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-01-04
- Publication Date
- 2026-04-14
AI Technical Summary
Existing multimodal sensing ward monitoring systems suffer from privacy protection issues and data confusion when identifying and distinguishing multiple patients, leading to incorrect status judgments.
It employs a multi-source sensing module, including a wearable wristband, millimeter-wave radar, and smart insoles. It binds patient data through a spatiotemporal consistency verification algorithm and uses a deep neural network for multimodal feature fusion and judgment. Combined with a hierarchical configuration module and a notification terminal, it achieves alarm linkage.
It achieves accurate binding of patient identity with multi-source data, improves the accuracy of data association and privacy protection, enhances the identification accuracy of complex risk scenarios, and ensures a balance between high-precision monitoring and non-intrusive monitoring.
Smart Images
Figure CN121845522A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of information technology, and in particular to a multimodal sensing ward monitoring system and method. Background Technology
[0002] With the continuous evolution of smart healthcare systems, ward monitoring, as a core component of clinical nursing and patient safety, directly impacts the quality of medical services and the efficiency of emergency response. Modern ward monitoring systems typically involve multiple dimensions, including physiological parameter monitoring, behavioral status recognition, and environmental perception, aiming to achieve 24 / 7, non-invasive health monitoring of hospitalized patients. In this context, multimodal sensing technology, due to its ability to integrate heterogeneous information such as vision, audio, physiological signals, and spatial location, has become a key pathway to improve the accuracy and robustness of monitoring systems.
[0003] Among them, multimodal sensing ward monitoring systems focus on collecting patients' vital signs, activity trajectories, posture changes, and surrounding environmental data through multiple sensors in a coordinated manner, and performing fusion analysis based on a unified data framework to achieve early identification and rapid response to abnormal conditions. The core objective of this type of system is to overcome the limitations of single-modal information and build a comprehensive, accurate, and timely intelligent monitoring closed loop.
[0004] Since multiple patients may be present in a ward, existing multimodal sensing technologies include visual sensors, such as cameras, to better identify and differentiate patients. However, the use of cameras raises privacy concerns and impacts the patient's hospital experience. Furthermore, multiple patients may cause data confusion, leading to incorrect assessments of the patient's condition. Summary of the Invention
[0005] In view of this, the purpose of the present invention is to provide a multimodal sensing ward monitoring system to improve the above-mentioned problems.
[0006] This invention discloses a multimodal sensing ward monitoring system, the system comprising: The multi-source sensing module utilizes multiple sensors to collect relevant patient data; wherein, the multi-source sensing module includes a wearable wristband worn by the patient for collecting the patient's primary identity data, location data, and primary vital sign data; a millimeter-wave radar installed in the ward for collecting the patient's human target coordinate data, posture data, and secondary vital sign data; and a smart insole placed in the patient's shoe for collecting the patient's secondary identity recognition data and motion data; The data processing center receives relevant data collected by the multi-source sensing module to determine whether an abnormal state has occurred, and triggers an alarm when an abnormal state occurs. The data processing center includes a signal transceiver, a hierarchical configuration module, a positioning and identification module, and an alarm logic module. The hierarchical configuration module is used to set the monitoring strategy and configuration parameters of the various sensors, and the configuration parameters are sent to each sensor through the signal transceiver. The positioning and identification module receives relevant data collected by the multi-source sensing module through the signal transceiver, and uses a spatiotemporal consistency verification algorithm to bind the data of the same patient collected by multiple sensors, generating collection data corresponding to each patient. The alarm logic module, based on the monitoring strategy, obtains the patient's location and status from the collection data, determines whether an abnormal state has occurred, and triggers an alarm if an abnormal state occurs, sending alarm-related information to the notification terminal through the signal transceiver. The terminal is notified to receive alarm-related information and execute alarm linkage measures.
[0007] Preferably, the wearable bracelet has a built-in ultra-wideband positioning chip, a heart rate sensor, and a blood oxygen saturation sensor; the first identity data is used to identify the patient by reading the wristband information through the NFC chip built into the bracelet.
[0008] Preferably, the millimeter-wave radar adopts a frequency-modulated continuous wave system, which transmits electromagnetic waves and receives human body reflection signals to obtain the coordinate data of the human target in three-dimensional space; the posture data is obtained by analyzing the spatial position change trajectory of human joints, and can identify four basic postures: standing, walking, sitting, lying down, and falling.
[0009] Preferably, the smart insole has a built-in pressure sensor array and an inertial measurement unit. The pressure sensor array consists of 32 pressure sensing points with a distribution density of 2 sensing points per square centimeter. The motion data includes gait cycle, plantar pressure distribution, and gait speed parameters. The second identification data is realized through communication between the insole and the ward reader via an RFID tag embedded in the insole.
[0010] Preferably, the hierarchical configuration module stores three standard monitoring strategies, corresponding to three clinical scenarios: intensive care, general nursing, and rehabilitation observation. Each monitoring strategy defines configuration parameters for sensor sampling frequency, data reporting cycle, and alarm threshold. The configuration parameters are sent to each sensor node via a signal transceiver in JSON format.
[0011] Preferably, the spatiotemporal consistency verification algorithm defines the data association function as follows:
[0012] Where t represents the timestamp, t i The time for data acquisition by the i-th sensor; p represents the spatial coordinates, p iLet w be the spatial coordinates of the i-th sensor. i Let σ be the weighting coefficient for the i-th sensor. t and σ p These are time and space tolerance parameters, respectively; when the function value exceeds the set threshold, it is determined to be data from the same patient.
[0013] Preferably, the alarm logic module realizes multimodal feature fusion judgment and constructs a risk assessment model based on deep neural network. The input layer of the model receives a total of 12-dimensional feature vectors from vital sign data from wearable wristbands, attitude data from millimeter-wave radar, and motion data from smart insoles. The model contains three hidden layers with 64, 32, and 16 neurons respectively, and uses the ReLU activation function. The output layer calculates the risk probability value through the Sigmoid function, and triggers an alarm when the probability value exceeds 0.95.
[0014] Preferably, the alarm logic module supports composite abnormal scene recognition. When the millimeter-wave radar outputs a fall posture signal and one of the following two conditions is met simultaneously, a high-level alarm is immediately triggered: Condition 1 is that the wearable wristband detects a heart rate drop of more than 30% within 3 seconds; Condition 2 is that the smart insole detects an abnormal foot pressure distribution that lasts for more than 5 seconds.
[0015] Preferably, the notification terminal includes three types of devices: bedside terminal, central display screen at nurse station, and mobile nursing terminal, supporting three alarm methods: audible and visual alarm, SMS push, and voice broadcast.
[0016] This invention also provides a multimodal sensing ward monitoring method, based on the multimodal sensing ward monitoring system described above, comprising: The multi-source sensing module uses multiple sensors to collect relevant patient data; The data processing center receives relevant data collected by the multi-source sensing module to determine whether an abnormal state has occurred, and triggers an alarm when an abnormal state occurs. The terminal is notified to receive alarm-related information and execute alarm linkage measures.
[0017] Compared with the prior art, the present invention has the following beneficial effects: 1. Through the collaborative work of the multi-source sensing modules, the accurate binding of patient identity with multi-source data was achieved, solving the problem of inconsistent data association; 2. The hierarchical configuration module supports dynamic monitoring strategy adjustment, and can flexibly configure sensing parameters according to the patient's condition level; 3. By using multimodal feature fusion judgment in the alarm logic module, the accuracy of identification in complex risk scenarios is significantly improved; 4. The entire system ensures high-precision monitoring while balancing privacy protection and non-intrusive monitoring, thus improving its clinical practical value. Attached Figure Description
[0018] Figure 1 This is a schematic diagram of the structure of the multimodal sensing ward monitoring system provided in the first embodiment of the present invention.
[0019] Figure 2 This is a schematic diagram of the structure of the multi-source sensing module provided in the first embodiment of the present invention.
[0020] Figure 3 This is a schematic diagram of the structure of a data processing center provided in the first embodiment of the present invention.
[0021] Figure 4 This is a flowchart illustrating the multimodal sensing ward monitoring method provided in the second embodiment of the present invention. Detailed Implementation
[0022] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. 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.
[0023] The first embodiment of this invention provides a multimodal sensing ward monitoring system for comprehensive monitoring of patients' vital signs and ensuring their safety. See also... Figure 1 This system comprises three core components: a multi-source sensing module 10, a data processing center 20, and a notification terminal 30. The multi-source sensing module 10 is deployed at key parts of the patient's body and within the ward space, collecting the patient's physiological parameters, movement status, and spatial location information through a heterogeneous sensor network. The data processing center 20, located in the server room at the nurses' station, is responsible for fusing and analyzing multi-source data and identifying anomalies. The notification terminal 30 is distributed throughout the ward area, the nurses' station, and on the mobile devices of medical staff, enabling the immediate transmission of alarm information and coordinated response.
[0024] In this embodiment, as Figure 2As shown, the multi-source sensing module 10 may include a wearable wristband 11, a millimeter-wave radar 12, and a smart insole 13. The wearable wristband 11 is worn by the patient on their wrist and incorporates an ultra-wideband positioning chip, a heart rate sensor, and a blood oxygen saturation sensor. The ultra-wideband positioning chip calculates the distance between the patient and the positioning base station in the ward by measuring the time-of-flight of the radio frequency signal, achieving real-time positioning with an accuracy of 0.1 meters. The heart rate sensor uses photoplethysmography (PPG), illuminating skin tissue with green LED light and detecting changes in light absorption caused by blood flow, continuously collecting heart rate data at a sampling frequency of 10Hz. The blood oxygen saturation sensor utilizes a dual-wavelength measurement principle of 660nm red light and 940nm infrared light, calculating the light absorption ratio of oxyhemoglobin to deoxyhemoglobin in arterial blood to obtain the blood oxygen saturation value. The identification function is achieved through the near-field communication chip built into the wearable wristband 11. When the patient is near the wristband, the chip automatically reads the patient's ID number, name, medical record number, and other identification information stored in the wristband, completing the initial identification data collection.
[0025] In this embodiment, the millimeter-wave radar 12 can be installed above the head of the hospital bed. It employs a frequency-modulated continuous wave (FM-CW) system with a working frequency of 60 GHz. It transmits FM-CW waves and receives micro-Doppler signals reflected from the human body. The radar signal processing unit mixes, filters, and converts the received signal to digital. It then extracts target distance information using a fast Fourier transform and obtains the precise coordinates of the human target in three-dimensional space using a radar direction estimation algorithm. The posture recognition function analyzes the spatial positional changes of 15 major joints of the human body (including the head, neck, shoulder, elbow, wrist, hip, knee, and ankle). It employs a temporal pattern recognition algorithm based on a long short-term memory network, accurately distinguishing between four basic postures: standing, walking, sitting / lying down, and falling. The vital signs monitoring function detects the micro-Doppler effect caused by chest displacement. It uses an empirical mode decomposition algorithm to separate respiratory and heart rate signals, obtaining respiratory rate and heart rate data as secondary vital signs data.
[0026] In this embodiment, the smart insole 13 placed inside the patient's shoe incorporates a pressure sensor array. This array consists of 32 piezoresistive sensing points distributed according to the anatomical structure of the foot, with a density of two sensing points per square centimeter, covering key pressure areas such as the heel, arch, metatarsals, and toes. The pressure sensors utilize the Wheatstone bridge principle to convert the foot pressure distribution into voltage signal changes, which are then collected by a 12-bit analog-to-digital converter. The inertial measurement unit integrates a three-axis accelerometer, a three-axis gyroscope, and a three-axis magnetometer, calculating motion parameters such as gait cycle, gait speed, and stride length using a sensor fusion algorithm. Identification is achieved through an RFID tag embedded in the insole. When the patient enters the ward area, an ultra-high frequency reader installed at the bottom of the bed automatically reads the patient's unique identification code stored in the tag, completing the acquisition of secondary identity data.
[0027] In this embodiment, the signal transceiver of the data processing center 20 adopts a dual-band communication design, supporting an adaptive switching mechanism between the 2.4GHz and 5GHz bands. When severe interference is detected in the 2.4GHz band, it automatically switches to the 5GHz band to transmit data, ensuring a stable data transmission rate of 150 megabits per second and strictly controlling the communication latency to within 100 milliseconds. Data transmission employs a time-division multiple access-based scheduling mechanism, allocating fixed transmission time slots to different types of sensors to avoid transmission failures caused by data packet collisions.
[0028] In this embodiment, as Figure 3 As shown, the data processing center 20 includes a signal transceiver 21, a hierarchical configuration module 22, a positioning and identification module 23, and an alarm logic module 24. The hierarchical configuration module 22 stores three standard monitoring strategies, corresponding to three clinical scenarios: intensive care, routine care, and rehabilitation observation. The intensive care strategy targets critically ill postoperative patients, setting sensor sampling frequencies of 10Hz for the wearable wristband, 20Hz for the millimeter-wave radar, and 50Hz for the smart insole, with a data reporting cycle of 1 second. Alarm thresholds include combinations of parameters such as heart rate below 50 beats / min or above 120 beats / min, blood oxygen saturation below 90%, and immediate alarm upon fall detection. The routine care strategy targets patients with stable conditions, setting sensor sampling frequencies of 2Hz for the wearable wristband, 10Hz for the millimeter-wave radar, and 20Hz for the smart insole, with a data reporting cycle of 5 seconds and appropriately relaxed alarm thresholds. The rehabilitation observation strategy targets patients about to be discharged, primarily monitoring motor function and daily activity abilities. All configuration parameters are sent to each sensor node via a signal transceiver in JSON format. The instructions include fields such as device address, parameter type, parameter value, and effective time.
[0029] In this embodiment, when the positioning and identification module 23 performs multi-source data binding operations, it employs a spatiotemporal consistency verification algorithm. This algorithm defines the data association function as follows:
[0030] Where t represents the timestamp, t i The time for data acquisition by the i-th sensor; p represents the spatial coordinates, p i Let w be the spatial coordinates of the i-th sensor. i Let σ be the weighting coefficient for the i-th sensor. t and σ p These are time and space tolerance parameters, respectively; when the function value exceeds the set threshold, it is determined to be data from the same patient.
[0031] In this embodiment, the spatiotemporal consistency verification algorithm can aggregate data collected by multiple sensors, avoiding confusion between data collected from different patients in the same ward, thereby enabling accurate identification of patients without the use of visual sensors.
[0032] Among them, the time tolerance parameter σ t The sampling frequency is dynamically adjusted based on the sensor sampling frequency, such as setting it to 0.5 seconds, with a spatial tolerance parameter σ. p The positioning distance is dynamically adjusted according to the ward layout, and can be set to 0.1 meters. The weighting coefficient is allocated based on the sensor accuracy. For example, the positioning weight of wearable wristband 11 is 0.4, the positioning weight of millimeter-wave radar 12 is 0.35, and the positioning weight of smart insole 13 is 0.25. When the function value exceeds the set threshold, such as 0.8, it is determined to be data from the same patient, and a complete data collection record for each patient is generated.
[0033] The alarm logic module 24 implements multimodal feature fusion judgment. Specifically, a risk assessment model based on a deep neural network is constructed. The input layer of the risk assessment model receives 12-dimensional feature vectors from wearable wristband 11 (heart rate, blood oxygen saturation), millimeter-wave radar 12 (respiratory rate, posture encoding), and smart insole 13 (gait cycle, plantar pressure distribution entropy, gait speed, etc.). The risk assessment model contains three hidden layers: the first hidden layer with 64 neurons uses a modified linear unit activation function, and batch normalization is used to accelerate the training process; the second hidden layer with 32 neurons uses an exponential linear unit activation function, and dropout regularization is introduced to prevent overfitting; the third hidden layer with 16 neurons uses a hyperbolic tangent activation function, and L2 weight decay constraint is added. The output layer calculates the risk probability value using an S-shaped growth curve function, and triggers an alarm when the probability value exceeds 0.95.
[0034] In this embodiment, the alarm logic module 24 supports composite abnormal scene recognition. A high-level alarm is immediately triggered when a fall posture signal is detected from the millimeter-wave radar and one of the following conditions is met simultaneously: Condition 1: The wearable wristband detects a heart rate drop of more than 30% from a baseline value within 3 seconds. The baseline value is obtained by calculating the average heart rate over the past 30 seconds using a sliding window method. Condition 2: The smart insole detects an abnormal foot pressure distribution that lasts for more than 5 seconds. The abnormal pressure distribution is determined by comprehensively calculating the pressure center trajectory offset and the rate of change of the foot contact area.
[0035] In this embodiment, the notification terminal 30 includes three types of devices: a bedside terminal, a central display screen at the nurse station, and a mobile nursing terminal. The bedside terminal uses a 10-inch touchscreen, which displays a red warning icon and emits an 85-decibel buzzer when an alarm is triggered; the central display screen at the nurse station is a 55-inch LCD video wall, which displays the location of the alarmed bed and the patient's basic information in the form of an electronic map; the mobile nursing terminal is a tablet device provided to medical staff, which supports both SMS push and voice broadcast alarm methods to ensure timely delivery of alarm information.
[0036] The following is a further description of some preferred embodiments of the present invention.
[0037] In this embodiment, customized monitoring needs can be achieved for different wards by configuring various parameters.
[0038] For example, in the long-term care ward of a geriatric department, a multimodal sensing ward monitoring system conducts behavioral monitoring and safety alerts for Alzheimer's patients. The system configuration is optimized for the characteristics of elderly patients; the hierarchical configuration module 21 adds monitoring strategies such as wandering detection, prolonged stillness, and abnormal toileting. The millimeter-wave radar 22 adds wandering behavior recognition functionality, triggering a mild alarm when it detects a patient walking back and forth in a fixed area more than 10 times by analyzing the repetition and regularity of the patient's movement trajectory. The smart insole 13 pressure sensor array adds foot pressure asymmetry analysis, detecting walking function abnormalities early by comparing the pressure distribution differences between the left and right feet.
[0039] The multi-source data binding algorithm is adjusted to address the slow movement characteristics of elderly patients, with a time tolerance parameter σ. t The spatial tolerance parameter σ is relaxed from 0.5 seconds to 2 seconds. pThe measurement range has been expanded from 0.1 meters to 0.2 meters. Weighting coefficients have been redistributed: the weight of millimeter-wave radar has been increased to 0.5, the weight of smart insoles has been adjusted to 0.3, and the weight of wearable wristbands has been reduced to 0.2, thus placing greater emphasis on spatial behavior monitoring. The risk assessment model's feature vector now includes behavioral indicators specific to elderly patients, such as wandering frequency, duration of stillness, and number of times they get out of bed at night. The output layer alarm thresholds are set tiered according to nursing care levels: 0.9 for top-level care, 0.85 for level 1 care, and 0.8 for level 2 care.
[0040] The alarm logic module 24 adds a behavioral pattern anomaly detection function. By establishing a baseline model of the patient's daily behavior, an alert is triggered when a significant deviation from the baseline is detected. The baseline model uses a Gaussian mixture model and includes 16 behavioral characteristics such as wake-up time, activity area, walking distance, and sitting / lying time. The model parameters are automatically updated at 2:00 AM daily. The notification terminal adds a family member notification function. When a high-level alarm is triggered, in addition to medical staff, an alarm SMS is sent to the mobile phones of designated family members, including the patient's name, type of abnormality, time of occurrence, and handling suggestions.
[0041] In summary, compared with the prior art, the present invention has at least the following beneficial effects: 1. Through the collaborative work of the multi-source sensing modules, the accurate binding of patient identity with multi-source data was achieved, solving the problem of inconsistent data association; 2. The hierarchical configuration module supports dynamic monitoring strategy adjustment, and can flexibly configure sensing parameters according to the patient's condition level; 3. By using multimodal feature fusion judgment in the alarm logic module, the accuracy of identification in complex risk scenarios is significantly improved; 4. The entire system ensures high-precision monitoring while balancing privacy protection and non-intrusive monitoring, thus improving its clinical practical value.
[0042] Please participate Figure 4 The second embodiment of the present invention also provides a multimodal sensing ward monitoring method, based on the multimodal sensing ward monitoring system described above, comprising: S201, the multi-source sensing module uses multiple sensors to collect relevant patient data; S202, The data processing center receives relevant data collected by the multi-source sensing module to determine whether an abnormal state has occurred, and triggers an alarm when an abnormal state occurs; S203, notify the terminal to receive alarm-related information and execute alarm linkage measures.
[0043] The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the concept of this application. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this application.
Claims
1. A multimodal sensing ward monitoring system, characterized in that, The system includes: The multi-source sensing module utilizes multiple sensors to collect relevant patient data; wherein, the multi-source sensing module includes a wearable wristband worn by the patient for collecting the patient's primary identity data, location data, and primary vital sign data; a millimeter-wave radar installed in the ward for collecting the patient's human target coordinate data, posture data, and secondary vital sign data; and a smart insole placed in the patient's shoe for collecting the patient's secondary identity recognition data and motion data; The data processing center determines whether a patient is in an abnormal state based on relevant data collected by the multi-source sensing module, and triggers an alarm when an abnormal state is detected. The data processing center includes a signal transceiver, a hierarchical configuration module, a location identification module, and an alarm logic module. The hierarchical configuration module sets the monitoring strategy and configuration parameters for the various sensors, which are then sent to each sensor via the signal transceiver. The location identification module receives relevant data collected by the multi-source sensing module via the signal transceiver and uses a spatiotemporal consistency verification algorithm to bind data from the same patient collected by multiple sensors, generating data corresponding to each patient. The alarm logic module, based on the monitoring strategy, obtains the patient's location and state from the collected data, determines whether an abnormal state is detected, and triggers an alarm if an abnormal state is detected, sending alarm-related information to a notification terminal via the signal transceiver. The terminal is notified to receive alarm-related information and execute alarm linkage measures.
2. The multimodal sensing ward monitoring system according to claim 1, characterized in that, The wearable bracelet has a built-in ultra-wideband positioning chip, heart rate sensor, and blood oxygen saturation sensor; the first identity data is used to identify the patient by reading the wristband information through the built-in NFC chip.
3. The multimodal sensing ward monitoring system according to claim 1, characterized in that, The millimeter-wave radar adopts a frequency-modulated continuous wave system. It transmits electromagnetic waves and receives reflected signals from the human body to obtain the coordinate data of the human target in three-dimensional space. The posture data is obtained by analyzing the spatial position change trajectory of the human joints and can identify four basic postures: standing, walking, sitting, lying down, and falling.
4. The multimodal sensing ward monitoring system according to claim 1, characterized in that, The smart insole has a built-in pressure sensor array and an inertial measurement unit. The pressure sensor array consists of 32 pressure sensing points, with a distribution density of 2 sensing points per square centimeter. The motion data includes gait cycle, plantar pressure distribution, and gait speed parameters. The second identification data is realized through communication between the insole and the ward reader via an RFID tag embedded in the insole.
5. The multimodal sensing ward monitoring system according to claim 1, characterized in that, The hierarchical configuration module stores three standard monitoring strategies, corresponding to three clinical scenarios: intensive care, general nursing, and rehabilitation observation. Each monitoring strategy defines configuration parameters for sensor sampling frequency, data reporting cycle, and alarm threshold. These configuration parameters are sent to each sensor node via a signal transceiver in JSON format.
6. The multimodal sensing ward monitoring system according to claim 1, characterized in that, The spatiotemporal consistency verification algorithm defines the data association function as follows: Where t represents the timestamp, t i The time for data acquisition by the i-th sensor; p represents the spatial coordinates, p i Let w be the spatial coordinates of the i-th sensor. i Let σ be the weighting coefficient for the i-th sensor. t and σ p These are time and space tolerance parameters, respectively; when the function value exceeds the set threshold, it is determined to be data from the same patient.
7. The multimodal sensing ward monitoring system according to claim 1, characterized in that, The alarm logic module implements multimodal feature fusion judgment and constructs a risk assessment model based on deep neural networks. The input layer of the risk assessment model receives a total of 12-dimensional feature vectors from vital sign data from wearable wristbands, attitude data from millimeter-wave radar, and motion data from smart insoles. The model contains three hidden layers with 64, 32, and 16 neurons respectively, and uses the ReLU activation function. The output layer calculates the risk probability value through the Sigmoid function, and triggers an alarm when the probability value exceeds 0.
95.
8. The multimodal sensing ward monitoring system according to claim 1, characterized in that, The alarm logic module supports composite abnormal scene recognition. When the millimeter-wave radar outputs a fall posture signal and meets one of the following two conditions, a high-level alarm is immediately triggered: Condition 1 is that the wearable wristband detects a heart rate drop of more than 30% within 3 seconds; Condition 2 is that the smart insole detects an abnormal foot pressure distribution that lasts for more than 5 seconds.
9. The multimodal sensing ward monitoring system according to claim 1, characterized in that, The notification terminals include three types of devices: bedside terminals, central display screens at nurse stations, and mobile nursing terminals, supporting three alarm methods: audible and visual alarms, SMS push notifications, and voice broadcasts.
10. A multimodal sensing ward monitoring method, based on the multimodal sensing ward monitoring system as described in any one of claims 1 to 9, characterized in that, include: The multi-source sensing module uses multiple sensors to collect relevant patient data; The data processing center receives relevant data collected by the multi-source sensing module to determine whether an abnormal state has occurred, and triggers an alarm when an abnormal state occurs; The terminal is notified to receive alarm-related information and execute alarm linkage measures.