Anti-drowning alarm bracelet and method based on heart rate and water pressure identification

Abstract

This invention belongs to the field of anti-drowning alarm technology and discloses an anti-drowning alarm bracelet and method based on heart rate and water pressure recognition. In step S2, a dual calibration mechanism is established, consisting of a real-time environmental compensation internal circulation and a periodic individual and behavioral baseline external circulation. The real-time environmental compensation internal circulation is executed every 10 seconds, and the heart rate mutation threshold is lowered by 1.5% for every 1°C decrease in water temperature. When the vertical sinking acceleration detected in deep water is lower than -0.5g, the blood oxygen warning threshold is further lowered by 1%. The periodic individual and behavioral baseline external circulation is executed every 24 hours, and the physiological baseline is updated by combining the user's resting, exercise, and water activity data within 24 hours. If the user has 3 or more shallow water operation records in the past 7 days, the shallow water heart rate fluctuation threshold will be relaxed from 25% to 30% of the normal user baseline value. Through this mechanism, the system can dynamically adapt to environmental changes and individual behavioral characteristics, reducing false alarms and missed alarms.

Description

Technical Field

[0001] This invention belongs to the field of anti-drowning alarm technology, specifically an anti-drowning alarm bracelet and method based on heart rate and water pressure recognition. Background Technology

[0002] Drowning prevention equipment is an important tool for ensuring the safety of recreational swimming, professional water operations, and water activities for special groups. While conventional equipment in the industry can trigger alarms by monitoring physiological and environmental parameters such as heart rate and water pressure, the requirements for the reliability of such equipment are becoming increasingly stringent as water activities expand. Existing drowning prevention technologies generally suffer from the following technical problems: Most current devices use fixed physiological thresholds, such as a uniform setting that triggers an alarm when the heart rate increases by 30% from the resting value, without establishing a correlation and adjustment mechanism between environmental parameters and physiological benchmarks. Changes in water temperature can directly cause the heart rate benchmark to drift, and at this time, the fixed threshold can easily misjudge the sudden drop in heart rate in the early stages of drowning as normal fluctuations under low temperature. Moreover, the devices cannot distinguish between the slow decrease in heart rate caused by low temperature and the sudden drop in heart rate caused by drowning and suffocation, and there is no calibration for individual physiological differences, which ultimately leads to missed alarms and false alarms in actual use.

[0003] Most existing equipment is designed with a single alarm level. Once an alarm is triggered, resources are allocated according to the highest priority without considering the level of risk in the scenario. For example, when workers are performing maintenance in shallow water, the frequent immersion of their arms in water and the resulting heart rate fluctuations caused by brief exertion have the same alarm priority as those triggered by a child drowning in the deep sea. This leads to rescue platforms mis-dispatching vessels and wasting manpower. Furthermore, when a child is drowning in open water, if alarm information is queued with low-priority data such as equipment status notifications, the golden time for rescue may be missed. Summary of the Invention

[0004] The purpose of this invention is to provide an anti-drowning alarm bracelet and method based on heart rate and water pressure recognition, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a drowning prevention alarm method based on heart rate and water pressure recognition, comprising: Furthermore, step S1 is specifically as follows: The motion posture semantic recognition module is equipped with a six-axis motion sensor and a motion feature extraction chip to identify three types of semantic actions: active paddling, drowning struggle, and accidental sinking, which are used to distinguish between active movement and passive danger. The water temperature sensing extension module is equipped with a water temperature sensor and a water temperature and muscle movement correlation database. The correlation database pre-stores the correlation curves between water temperature and the intensity of struggling movements for three groups of people: children, adults and the elderly, within a water temperature range of 10 to 40°C. It also stores the user's group identifier. The physiological monitoring module integrates a heart rate sensor and a blood oxygen sensor for collecting heart rate and blood oxygen data; The step S1 also includes a power supply module equipped with a low-power rechargeable battery that supports a low-temperature insulation mode. The water pressure monitoring module is equipped with a water pressure sensor and a low-power electrochemical salinity sensor. The salinity sensor adopts a dual-principle design of "conductivity method + temperature compensation", with a detection range of 0‰~50‰, a detection accuracy of ±1‰, and a response time of ≤0.2 seconds. The sensor electrode uses a titanium alloy substrate (corrosion resistance ≥5000 hours salt spray test) and is covered with a polytetrafluoroethylene antifouling coating to resist interference from seawater impurities and microbial adhesion.

[0006] The salinity sensor and water pressure sensor adopt a "synchronous sampling + intermittent wake-up" collaborative working mode: under normal conditions, they are synchronized with the water pressure sensor to sample at 0.3 seconds / time, and in high-salinity environments (salinity ≥30‰), they automatically switch to encrypted sampling at 0.1 seconds / time; the sensor has a built-in fault self-checking mechanism. When the deviation of three consecutive sampling data exceeds ±3‰, the processing module is automatically triggered to call water temperature and water pressure data to estimate salinity (the estimation formula is shown in step S2) to ensure data continuity.

[0007] The water pressure sensor uses a ceramic piezoresistive chip, and the electrode surface is pre-coated with a nano-zirconia anti-corrosion layer. The salinity sensor and the water pressure sensor are linked via an I2C bus. When the salinity detection value exceeds 30‰ twice consecutively (with an interval of 0.1 seconds to avoid instantaneous interference and false triggering), the anti-corrosion coating activation command of the water pressure sensor is triggered. The collaborative communication module adopts a dual-mode communication architecture of 5G and Bluetooth, and has a built-in near-field rescue terminal collaborative protocol to establish closed-loop communication for alarm, response and feedback with smart lifebuoys within a 100-meter range, swimming pool lifeguard terminals within a 500-meter range and lifeguard terminals.

[0008] Furthermore, step S2 is specifically as follows: Based on the water temperature, water pressure, movement posture and physiological parameter data collected in step S1, a dual calibration mechanism is established, consisting of real-time environmental compensation internal circulation and periodic individual and behavioral baseline external circulation. A dynamic calibration database is also provided to store the water temperature, water pressure, salinity, physiological parameters and subsequent feedback data required for calibration. The real-time environmental compensation internal circulation is executed once every 10 seconds. On the one hand, it calculates and dynamically adjusts the heart rate mutation threshold according to the rule that the heart rate mutation threshold is lowered by 1.5% for every 1°C decrease in water temperature. On the other hand, combined with the water pressure and vertical acceleration data detected in step S1, if the vertical sinking acceleration is detected to be lower than -0.5g in the deep water environment, the blood oxygen warning threshold is further lowered by 1%. The water depth ≥ 1 meter is defined as the deep water area and the water depth < 1 meter is defined as the shallow water area. In addition to adjusting the heart rate mutation threshold and blood oxygen warning threshold, salinity-water temperature co-calibration is performed simultaneously: when the water temperature is below 10℃ or above 35℃, the temperature drift error of the salinity detection value increases. At this time, the associated database of the water temperature sensing extension module is called, and the salinity value is corrected a second time based on the user's population (child / adult / elderly) and water temperature data. The correction coefficient is ±0.03 for children, ±0.02 for adults, and ±0.04 for the elderly to ensure detection accuracy in high water temperature or low water temperature and high salinity environments. The periodic individual and behavioral baseline external cycle is executed once every 24 hours, and the physiological baseline is adjusted by combining the user's resting, exercise and water activity data collected in step S1 within 24 hours; at the same time, for the user's work scenario identified in step S1, if the user has 3 or more shallow water work records in the past 7 days, the heart rate fluctuation threshold in shallow water area is relaxed from 25% to 30% of the normal user baseline value.

[0009] The salinity data employs a dual calibration mechanism: ① Real-time calibration: The salinity detection value is corrected every 10 seconds based on water temperature data. For every 5°C change in water temperature, the salinity correction coefficient is adjusted by ±0.02. Formula: Sal adj =Sal raw ×(1+0.004×(T w -25)), T w This is the real-time water temperature, with 25℃ as the reference temperature; ② Periodic calibration: Every 24 hours, the system retrieves historical salinity data from the user's dynamic calibration database (such as the average salinity of the same water area over the past 7 days) to correct sensor drift errors (drift compensation range ±0.5‰). When the user switches water areas, such as from a freshwater swimming pool to a seawater bathing beach, the processing module automatically identifies sudden changes in salinity (change ≥10‰ within 5 seconds) and triggers immediate calibration to ensure detection accuracy.

[0010] Furthermore, step S3 is specifically as follows: Based on the threshold calibrated in step S2, at the data acquisition level, the monitoring state is divided according to the normal or suspected danger threshold determined in step S2. In the normal monitoring state, data is collected at a specific frequency through step S1, wherein the water temperature and water pressure data are collected every 0.3 seconds, the heart rate and blood oxygen data are collected every 1 second, and the movement posture data are collected every 0.2 seconds. When entering a suspected dangerous state, the frequency of motion posture acquisition is increased to once every 0.05 seconds, while simultaneously recording the duration and intensity of the movement. At the scene labeling level, combining the action semantics identified in step S1, the user's group identifier, salinity environment label, and the data features calibrated in step S2, the five types of data—water temperature, water pressure, salinity, heart rate, and blood oxygen—are treated as five-parameter time-series data and continuously recorded according to the collection timestamp. A scene labeling system with five dimensions—identity, water depth, movement state, action semantics, and salinity environment—is established. For example, children—deep water area—drowning and struggling—high salinity (35‰). The labels are bound to the monitoring data according to the timestamp, providing a more accurate environmental context for the drowning determination in step S4.

[0011] Furthermore, step S4 is specifically as follows: Combining the five-parameter time-series data and semantic scene tags from step S3, and referring to the dynamic threshold from step S2, a time-series correlation judgment logic is established. The judgment logic is considered to be suspected drowning if it meets two or more of the following conditions: the first is the basic judgment condition, namely, water depth ≥ 1 meter, heart rate ≥ 25% increase compared to the threshold calibrated in step S2, and blood oxygen ≥ 100% of the warning threshold calibrated in step S2. The second is the action semantic condition, that is, step S3 identifies drowning struggle or accidental sinking action, and the duration exceeds 1 second; the third is the water temperature correlation condition, that is, when the water temperature collected in step S3 is below 20℃, if the heart rate drops suddenly and is accompanied by passive sinking action, even if the blood oxygen does not reach the warning threshold calibrated in step S2, a suspected judgment will be triggered. If step S3 detects active paddling motions, it is determined to be normal exercise even if the heart rate exceeds the threshold calibrated in step S2.

[0012] Furthermore, step S5 is specifically as follows: Based on the determination of suspected drowning in step S4, five dynamic risk features are first extracted from the five-parameter time-series data collected in step S3, including the duration of struggling, the rate of decrease in blood oxygen, the change in water depth, the action semantic risk coefficient combined with the action semantic recognition in step S1, and the water temperature correlation risk coefficient combined with the water temperature data in step S3. The initial weight of the water temperature correlation risk coefficient is 0.3. Subsequently, a three-level risk classification is determined based on the dynamic risk characteristics. Level 1 is emergency drowning, which requires the simultaneous fulfillment of three key requirements, as follows: In terms of movement: it recognizes two types of movements: drowning struggle and accidental sinking, and the drowning struggle lasts for more than 3 seconds; In terms of physiological indicators: blood oxygen levels drop rapidly, with a decrease of more than 5% every 5 minutes; In terms of environmental conditions: it is located in deep water and the water temperature and corresponding risk coefficients meet the requirements: risk coefficient of 1.3 for 10-15℃, risk coefficient of 1.1 for 15-20℃, and risk coefficient of 1.0 for 20-25℃. Level 2 is a non-urgent anomaly and must meet any of the following conditions: Combination of action and environmental / physiological indicators: Drowning struggle lasts for 1-3 seconds and blood oxygen does not decrease or the person is in shallow water; Combination of environmental and physiological indicators: water temperature below 15℃, sudden drop in heart rate and no drowning struggles; Level 3 is a false trigger. If the action semantic tag corresponding to active paddling is detected, or if the user completes dual confirmation by touch button and voice within the preset confirmation time, it is determined to be a false trigger and the alarm process is terminated.

[0013] Furthermore, step S6 is specifically as follows: Based on the three-level risk level in step S5, the corresponding rescue response mechanism is activated. For level one emergency drowning, a level one alarm is triggered. Locally, the buzzer sounds via the hardware activation in step S1, accompanied by alternating red and blue flashing and vibration alerts. In the near field, the target location and predicted movement trajectory are sent to the smart lifebuoy within 100 meters via the collaborative communication module in step S1, triggering the lifebuoy's automatic propulsion function. Simultaneously, the user's identity, physiological data collected in step S3, and water temperature environment are pushed to the rescue pavilion terminal within 500 meters, enabling the terminal to automatically generate a rescue route combined with the ocean current direction. Remotely, the alarm signal, real-time location, and the five-parameter time-series curve from step S3 are transmitted to the rescue platform via the 5G network with the highest priority. At the same time, three preset emergency contact numbers are dialed, and the user's name and drowning location are automatically played upon connection. For non-emergency level 2 anomalies, a level 2 alarm is activated, with a short beep from the local buzzer accompanied by a yellow flashing light; the anomaly notification and the current location collected in step S3 are sent only to the emergency contact, while a notification is pushed to the near-field Bluetooth terminal. For a false trigger of level three, a level three alarm is activated, with only the local green light flashing for 10 seconds. At the same time, the cause of the false trigger is recorded, and the data is fed back to the dynamic calibration database in step S2.

[0014] Furthermore, step S7 is specifically as follows: Based on the feedback data from the rescue process in step S6, the collaborative communication module in step S1 first receives feedback data from the rescue platform and lifeguard terminal, including rescue arrival time, rescue effectiveness, and reasons for false alarms. The feedback data is then correlated with the judgment results in step S4 and the grading results in step S5. Subsequently, the core model was optimized based on the associated data. On the one hand, for the dynamic calibration system in step two, if the accuracy rate of the first-level alarm in step S5 in a certain scenario is less than 90%, the weight of the water temperature associated risk coefficient will be automatically increased. On the other hand, for the graded model in step S5, if the false alarm rate in a certain area exceeds 20%, the action semantic recognition threshold in step S3 will be adjusted for that area. Federated learning technology is used so that the user physiological data collected in step S3 is used for model training only locally, and the cloud only receives the desensitized feature weights.

[0015] Furthermore, step S8 is specifically as follows: The design environment adaptation module is designed so that, in low-temperature environments, the hardware level activates the low-temperature insulation mode through the power supply module in step S1, and uses the built-in heating element to maintain the battery temperature above 5°C; the algorithm level adjusts the heart rate benchmark and blood oxygen warning threshold in step S2. In high-salinity seawater environments, the hardware achieves dual functions of corrosion prevention activation and durability assurance: ① Activation mechanism: First, the salinity sensor of the water pressure monitoring module in step S1 collects data in real time. When the salinity is detected to be ≥30‰ twice in a row, it is determined to be a high salinity environment. The processing module outputs a pulse microcurrent of 1.2V and 50mA (lasting for 3 seconds) to the water pressure sensor to drive the oxidation reaction on the electrode surface: Ti+O2→TiO2, generating a dense titanium dioxide oxide film with a thickness of 50~80nm. The titanium dioxide oxide film has a hardness of HV≥500 and a salt corrosion resistance grade of ≥C5-M, covering the exposed area of ​​the electrode surface. ② Durability design: After the oxide film is formed, the sensor automatically detects the electrode impedance every 30 minutes. The reference impedance is 1kΩ~5kΩ. When the impedance exceeds 10kΩ (indicating oxide film wear), microcurrent recoating is triggered again, and the current is reduced to 30mA to avoid excessive oxidation. At the same time, the dynamic calibration database records the activation time and salinity environment for each activation. When the cumulative activation count reaches 10, an electrode maintenance reminder is pushed to the user.

[0016] Algorithm level: ① The salinity correction coefficient is derived based on the seawater density-salinity-depth correlation model, specifically as follows: Salinity 30‰~40‰: Seawater density is 1.02~1.03 times higher than freshwater, water depth calculation correction factor 1.05, compensating for the influence of density on water pressure detection; Salinity 40‰~50‰: Seawater density is 1.03~1.04 times higher than freshwater density, correction factor 1.1; ② When the salinity sensor fails, the backup algorithm is activated: based on water temperature (T) w Salinity is estimated using water pressure (P) and water pressure. est =0.02×P-0.1×T w +5, applicable to water depths of 0~10 meters, ensuring that the water depth calculation error is ≤±0.1 meters and does not affect the drowning determination logic in step S4; In high-altitude waters, the blood oxygen warning threshold and heart rate mutation threshold in step S2 are adjusted at the algorithm level to ensure that the judgment logic in step S4 conforms to the physiological data characteristics of the high-altitude environment.

[0017] This invention also provides a drowning prevention alarm bracelet based on heart rate and water pressure recognition, which, based on the above method, includes: The physiological monitoring module integrates a heart rate sensor for collecting heart rate data; The water pressure monitoring module, equipped with a water pressure sensor, is used to detect water depth and immersion status; The motion posture semantic recognition module integrates a six-axis motion sensor and a motion feature extraction chip to identify active paddling, drowning struggle, and accidental sinking movements. The water temperature sensing extension module is equipped with a water temperature sensor and a database linking water temperature and muscle movement, used to collect water temperature data. The collaborative communication module adopts a 5G and Bluetooth dual-mode communication architecture to push alarm information and location data; The processing module, electrically connected to the above module, is used to establish individual physiological benchmarks and combine them with dynamic water temperature calibration to determine the threshold and jointly determine the risk of drowning. The alarm module, controlled by the processing module, can trigger tiered audible and visual alarms. The power supply and carrier module uses a flexible waterproof carrier to adapt to different wrist sizes and is equipped with a low-power rechargeable battery; The environment adaptation module is electrically connected to the processing module. Through hardware insulation, anti-corrosion activation, and algorithm threshold adjustment, it optimizes the device's adaptability in complex environments such as low temperature, high salinity, and high altitude.

[0018] The beneficial effects of this invention are as follows: 1. In step S2 of this invention, a dual calibration mechanism is established, consisting of a real-time environmental compensation internal circulation and a periodic individual and behavioral baseline external circulation. The real-time environmental compensation internal circulation is executed every 10 seconds, and the heart rate mutation threshold is lowered by 1.5% for every 1°C decrease in water temperature. When the vertical sinking acceleration detected in deep water is below -0.5g, the blood oxygen warning threshold is further lowered by 1%. The periodic individual and behavioral baseline external circulation is executed every 24 hours, and the physiological baseline is updated by combining the user's resting, exercise, and water activity data within 24 hours. If the user has 3 or more shallow water operation records in the past 7 days, the shallow water heart rate fluctuation threshold will be relaxed from 25% to 30% of the normal user baseline value. Through this mechanism, the system can dynamically adapt to environmental changes and individual behavioral characteristics, reducing false alarms and missed alarms.

[0019] 2. In step S5, the present invention refines the risk levels into Level 1 emergency drowning, Level 2 non-emergency anomaly, and Level 3 false trigger. Step S6 corresponds to a differentiated response mechanism: Level 1 alarms link local, near-field, and remote resources; Level 2 alarms only notify relevant personnel and push attention reminders to near-field Bluetooth terminals; Level 3 false triggers only cause the local green light to flash for 10 seconds and feed back the false trigger data to step S2. Thus, resources can be reasonably allocated according to the risk level of the scenario, avoiding waste of resources in non-emergency scenarios and ensuring the priority of rescue in emergency drowning situations.

[0020] 3. In step S8, this invention designs an environment adaptation module. In low-temperature environments, the hardware activates a low-temperature insulation mode via the power supply module, using a built-in heating element to maintain the battery temperature above 5°C. The algorithm adjusts the heart rate baseline and blood oxygen warning threshold in step S2. In high-salinity seawater environments, the hardware activates the anti-corrosion coating activation mode of the water pressure sensor, causing an oxide film to form on the electrode surface. The algorithm introduces a salinity correction coefficient in the water depth calculation in step S3. In high-altitude waters, the algorithm adjusts the blood oxygen warning threshold and heart rate mutation threshold in step S2. Through the collaborative optimization of hardware and algorithms, the applicability of the device in various complex environments is improved, covering scenarios such as recreational swimming and professional water operations. Attached Figure Description

[0021] Figure 1 This is a flowchart of the drowning prevention alarm method based on heart rate and water pressure recognition of the present invention; Figure 2 This is a flowchart illustrating the execution of the dual calibration mechanism of the present invention. Figure 3 This is a flowchart of the suspected drowning determination and risk classification process of the present invention. Detailed Implementation

[0022] 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.

[0023] like Figures 1 to 3 As shown, this embodiment of the invention provides a drowning prevention alarm method based on heart rate and water pressure recognition, including: Specifically, step S1 is as follows: A multi-dimensional perception and communication hardware system is established. The motion posture semantic recognition module is equipped with a six-axis motion sensor and a motion feature extraction chip to identify three types of semantic actions: active paddling, drowning struggle, and accidental sinking, which are used to distinguish between active movement and passive danger. Active paddling is characterized by a peak acceleration of 0.3 to 0.8g and an angular velocity variation frequency of 1 to 2Hz. The characteristics of drowning struggle are that the peak acceleration exceeds 1g and fluctuates irregularly, and the frequency of angular velocity change exceeds 3Hz. The characteristics of an unexpected sinking are a vertical acceleration of less than -0.5g lasting for 0.5 seconds.

[0024] The water temperature sensing extension module is equipped with a water temperature sensor and a water temperature and muscle movement correlation database. The correlation database pre-stores the correlation curves between water temperature and the intensity of struggling movements for three groups of people: children, adults, and the elderly, within a water temperature range of 10 to 40°C. Taking a 15°C environment as an example, the peak value of the struggling acceleration of children at this temperature will be 20% lower than that at 25°C. At the same time, it stores the user's group (child / adult / elderly) identifier. The physiological monitoring module integrates a heart rate sensor and a blood oxygen sensor for collecting heart rate and blood oxygen data; The step S1 also includes a power supply module equipped with a low-power rechargeable battery, which supports a low-temperature insulation mode and automatically starts when the water temperature is below 10°C. The water pressure monitoring module is equipped with a water pressure sensor to detect water depth and immersion status; The collaborative communication module adopts a dual-mode communication architecture of 5G and Bluetooth, and has a built-in near-field rescue terminal collaborative protocol. It establishes a closed-loop communication for alarm, response and feedback with smart lifebuoys within a 100-meter range, swimming pool lifeguard terminals within a 500-meter range and lifeguard terminals. At the same time, it realizes data interaction with the remote rescue platform through the 5G network.

[0025] Specifically, step S2 is as follows: Based on the water temperature, water pressure, movement posture and physiological parameter data collected in step S1, a dual calibration mechanism is established, consisting of real-time environmental compensation internal circulation and periodic individual and behavioral baseline external circulation. A dynamic calibration database is also provided to store the water temperature, water pressure, physiological parameters and subsequent feedback data required for calibration. The real-time environmental compensation internal circulation is executed once every 10 seconds. If the heart rate drops more than 1.5 times the heart rate mutation threshold after calibration in step S2 within 5 seconds, it is judged as a sudden drop in heart rate. On the one hand, the heart rate mutation threshold is calculated and dynamically adjusted according to the rule that the heart rate mutation threshold is lowered by 1.5% for every 1°C decrease in water temperature. On the other hand, combined with the water pressure and vertical acceleration data detected in step S1, if the vertical sinking acceleration detected is lower than -0.5g in deep water environment, the blood oxygen warning threshold is further lowered by 1%, for example, from 92% to 91% in seawater. The blood oxygen warning benchmark threshold is 93% in freshwater and 92% in seawater due to the difference in dissolved oxygen. Water depth ≥ 1 meter is defined as deep water area and water depth < 1 meter is defined as shallow water area. The periodic individual and behavioral baseline external cycle is executed once every 24 hours. It adjusts the physiological baseline by combining the user's resting, movement, and water activity data collected in step S1 within 24 hours. For example, if an adult swims between 18:00 and 19:00 for 3 consecutive days, the heart rate baseline for that period will be automatically increased by 5 beats / minute. It also needs to combine the user's movement posture data to calibrate the normal threshold for acceleration fluctuation. The vertical acceleration fluctuation range for ordinary users is -0.3g to 0.6g, while for children, due to their smaller movement amplitude, the threshold is narrowed to -0.2g to 0.4g. At the same time, for the user's work scenario identified in step S1, if the user has 3 or more shallow water work records in the past 7 days, it is defined as a scenario with a water depth of <1 meter and mainly fixed work actions, such as water equipment maintenance and shallow water aquaculture, excluding dynamic scenarios such as children playing in the water and swimming. The heart rate fluctuation threshold in shallow water is relaxed from 25% to 30% of the ordinary user baseline value to reduce false alarms in work scenarios.

[0026] Formula for dynamic adjustment of heart rate mutation threshold:

[0027] In the formula: This represents the adjusted real-time heart rate mutation threshold, which is the critical value for determining whether the heart rate has undergone an abnormal mutation. The unit is %; the value is updated in real time as the water temperature changes and is used in step S4 for suspected drowning determination. This represents the baseline threshold for heart rate mutations, expressed as a percentage (%). The baseline value for ordinary users is 25%, determined by periodic individual and behavioral baseline extracorporeal circulation. This indicates the fixed downward adjustment ratio of the heart rate mutation threshold for every 1°C decrease in water temperature, adapting to the effect of water temperature on heart rate; Indicates the change in water temperature. ,in It is the reference water temperature calibrated in step S2, which is the average water temperature of the user's daily water area; This is the real-time value collected by the water temperature sensor in step S1, only when... The formula takes effect at time 0.

[0028] Formula for additional adjustment of blood oxygen warning threshold in deep water area:

[0029] Applicable conditions: All conditions must be met simultaneously. 1m and The amount must be less than -0.5g; otherwise, the formula will not work. In the formula: This represents the additional blood oxygen warning threshold for deep water areas, a critical value for determining whether blood oxygen levels are abnormally low, expressed in %; it is only effective in high-risk scenarios and applies when the value is below the baseline threshold to avoid missed detections. This represents the baseline threshold for blood oxygen warning, expressed in %; the baseline value in the seawater scenario is 92%, determined by the periodic individual and behavioral baseline external circulation in step S2. 1% represents the fixed additional reduction percentage of the blood oxygen warning threshold in deep water areas and accelerated sinking scenarios; it is a compensation coefficient set for high-risk scenarios to adapt to the characteristics of accelerated hypoxia. This indicates the real-time water depth, in meters (m); it is calculated from the water pressure sensor data collected in step S1. 1m is the deep water area; This represents the vertical downward acceleration, with the unit being g, where 1g ≈ 9.8 m / s². 2 The data is collected by the six-axis motion sensor in step S1, with upward movement being positive and downward movement being negative.

[0030] Specifically, step S3 is as follows: Based on the threshold calibrated in step S2, at the data acquisition level, the monitoring state is divided according to the normal or suspected danger threshold determined in step S2. In the normal monitoring state, data is collected at a specific frequency through step S1, wherein the water temperature and water pressure data are collected every 0.3 seconds, the heart rate and blood oxygen data are collected every 1 second, and the movement posture data are collected every 0.2 seconds. When entering a suspected dangerous state, for example, if irregular acceleration fluctuations are detected (such as vertical acceleration exceeding the normal threshold ±0.2g calibrated in step S2), and the fluctuation range exceeds the normal threshold calibrated in step S2, the motion posture acquisition frequency is increased to 0.05 seconds per time, and the duration and intensity of the action are recorded simultaneously, such as the dynamic process of the struggling action increasing from 0.8g to 1.2g. At the scene labeling level, combining the action semantics and user group identifiers identified in step S1, and the data features calibrated in step S2, five types of data—water temperature, water pressure, salinity, heart rate, and blood oxygen—are treated as five-parameter time-series data and continuously recorded according to the collection timestamp. A scene labeling system is established with five dimensions: identity, water depth, movement state, action semantics, and salinity environment, generating scene labels such as child-deep water area-disorderly struggling-vertical sinking, and adult-shallow water area-active paddling-horizontal movement. The labels are bound to the monitoring data at the millisecond-level timestamp.

[0031] Specifically, step S4 is as follows: Combining the five-parameter time-series data and semantic scene tags from step S3, and referring to the dynamic threshold from step S2, a time-series correlation judgment logic is established to achieve preliminary screening of suspected drowning. The judgment logic is considered to be suspected drowning if it meets two or more of the following conditions: the first is the basic judgment condition, namely, water depth ≥ 1 meter, heart rate ≥ 25% increase compared to the threshold calibrated in step S2, and blood oxygen ≥ 1 meter. The second is the action semantic condition, that is, step S3 identifies drowning struggle or accidental sinking action, and the duration exceeds 1 second; the third is the water temperature correlation condition, that is, when the water temperature collected in step S3 is below 20℃, if the heart rate drops suddenly, that is, drops by 10 beats / minute or more within 5 seconds, and is accompanied by passive sinking action, even if the blood oxygen does not reach the warning threshold calibrated in step S2, a suspected judgment will be triggered. To avoid misjudgment, if step S3 detects active paddling motions, even if the heart rate exceeds the threshold calibrated in step S2, it will be judged as normal exercise and will not proceed to the subsequent alarm process.

[0032] Specifically, step S5 is as follows: Based on the determination of suspected drowning in step S4, five dynamic risk features are first extracted from the five-parameter time-series data collected in step S3, including the duration of struggling, the rate of decrease in blood oxygen, the change in water depth, the action semantic risk coefficient combined with the action semantic recognition in step S1, and the water temperature correlation risk coefficient combined with the water temperature data in step S3. The initial weight of the water temperature correlation risk coefficient is 0.3. The standard for setting the action semantic risk coefficient is as follows: 1.2 corresponds to drowning and struggling, 1.0 corresponds to accidental sinking, and 0.8 corresponds to sinking without action. The higher the value, the higher the risk level. The standard for setting the risk coefficient related to water temperature is as follows: 1.3 for water temperature of 10 to 15℃, 1.1 for 15 to 20℃, and 1.0 for 20 to 25℃. The higher the value, the higher the risk level.

[0033] Subsequently, a three-level risk classification is determined based on the dynamic risk characteristics. The first level is emergency drowning, which is determined by the first-level emergency drowning risk assessment formula. Level 2 is a non-emergency abnormality and must meet any of the following conditions: ① Drowning struggle is detected (action semantic risk coefficient 1.2), and the struggle lasts for 1-3 seconds, blood oxygen shows no downward trend, and water depth is <1 meter; ② Water temperature is below 15℃, heart rate drops by 10 beats / minute or more within 5 seconds, and no drowning struggle is detected. Level 3 is for false triggering exclusion. If the action semantic tag is detected to correspond to active paddling, or if the user completes dual confirmation by touch button and voice within the preset confirmation time, it is determined to be a false trigger and the alarm process is terminated to avoid ineffective scheduling.

[0034] Formula for Level 1 Emergency Drowning Risk Assessment:

[0035] Judgment rule: When Furthermore, when the action semantic tags are drowning struggle and accidental sinking, it is judged as a first-level emergency drowning; In the formula: This represents the Level 1 emergency drowning risk assessment value, which comprehensively considers the risks associated with movement, blood oxygen, water depth, and water temperature. The higher the value, the more urgent the risk, and it is used in step S6 to determine the rescue level. The duration of the struggle is indicated in seconds (s). The duration of the drowning struggle is collected in step S3 and clearly defined. A time greater than 3 seconds indicates a high risk. When the value is 0, no risk is contributed; This indicates the semantic risk coefficient of the action, such as drowning and struggling. =1.2, Unexpected subsidence =1.0, Active stroke / No movement =0, only high-risk actions are included in the assessment; This indicates the rate of decrease in blood oxygen saturation, expressed as % / 5min; the decrease in blood oxygen saturation per unit time is calculated using step S3. 5 minutes is considered high risk. The value is 0 at 5 minutes, indicating no risk contribution. Indicates the blood oxygen risk factor, only when 5 minutes (high risk of hypoxia) =1, otherwise =0; This represents the real-time water depth; it is a logical variable, not an actual numerical value; file definition. Take 1 for 1m (deep water area). 1m (shallow water area) is taken as 0, only deep water area is included in the assessment; This indicates the risk factor associated with water temperature, 10℃. When the real-time water temperature is <15℃, =1.3; 15℃ When the real-time water temperature is <20℃, =1.1; 20℃ When the real-time water temperature is <25℃, =1.0, the lower the water temperature, the higher the coefficient; This indicates the threshold for determining a Level 1 emergency drowning risk, which must meet the following conditions. 3.6 (3s × 1.2) (5% / 5minx1) (1×1), that is 5.6; This value is the dimensionless risk assessment threshold. The calculation logic is to directly sum the results of each high-risk item, and the weight of each risk item is 1.

[0036] Specifically, step S6 is as follows: Based on the three-level risk level in step S5, the corresponding rescue response mechanism is activated to achieve precise allocation of rescue resources; for Level 1 emergency drowning, a Level 1 alarm is activated, and the local system activates an 85dB buzzer continuously via hardware activation in step S1, accompanied by alternating red and blue flashing (1Hz frequency) and a 500Hz vibration alert; in the near field, the collaborative communication module in step S1 sends the target position and predicted trajectory (based on the sinking speed collected in step S3 over the past 3 seconds) to the smart lifebuoy within 100 meters, triggering the lifebuoy's automatic propulsion function, while simultaneously moving towards the 500-meter range. The rescue pavilion terminal inside pushes the user's identity, physiological data collected in step S3, and water temperature environment, enabling the terminal to automatically generate a rescue route combined with ocean current direction; remotely, it transmits alarm signals, real-time positioning (including drift prediction within 10 minutes, based on ocean current reference data related to water depth changes, vertical sinking speed, and water temperature collected in step S3) and the five-parameter time series curve of step S3 to the rescue platform with the highest priority through the 5G network. At the same time, it dials 3 preset emergency contact numbers, and automatically plays the user's name and drowning location (latitude and longitude) after the call is connected. For non-emergency level 2 anomalies, a level 2 alarm is activated, with a local buzzer sounding briefly for 1 second, accompanied by a yellow flash (0.8Hz frequency); only an anomaly notification and the current location collected in step S3 are sent to emergency contacts, and a notification is pushed to near-field Bluetooth terminals (such as the Walker Bracelet), without triggering the rescue platform dispatch. For a false trigger of level three, a level three alarm is activated, with only the local green light flashing for 10 seconds. At the same time, the cause of the false trigger is recorded, such as active paddling or user confirmation identified in step S3. This data is then fed back to the dynamic calibration database in step S2 for subsequent threshold optimization.

[0037] Specifically, step S7 is as follows: Based on the feedback data from the rescue process in step S6, the collaborative communication module in step S1 first receives feedback data from the rescue platform and the lifeguard terminal, including the rescue arrival time (time from alarm to contact with the user), rescue effectiveness (such as whether the rescue was successful), and reasons for false alarms (such as user accidental touch or environmental interference). The feedback data is then correlated with the judgment result in step S4 and the grading result in step S5. Subsequently, the core model was optimized based on the associated data. On the one hand, for the dynamic calibration system in step two, if the accuracy rate of the first-level alarm in step S5 is less than 90% in a certain scenario (such as "children in a -15℃ deep water area"), the weight of the water temperature associated risk coefficient will be automatically increased from 0.3 to 0.5. On the other hand, for the graded model in step S5, if the false alarm rate in a certain area (such as a specific lake) exceeds 20%, the action semantic recognition threshold in step S3 will be adjusted for that area, for example, the peak value of struggling acceleration will be increased from 1g to 1.2g. To protect user privacy, federated learning technology is used so that the user physiological data collected in step S3 is only used for model training locally, and the cloud only receives the desensitized feature weights, such as the heart rate threshold adjustment coefficient for children, thus avoiding the leakage of raw data.

[0038] Specifically, step S8 is as follows: The design environment adaptation module is designed so that in a low-temperature environment where the water temperature is below 10℃, the hardware level activates the low-temperature insulation mode through the power supply module in step S1, and uses the built-in heating element to maintain the battery temperature above 5℃; the algorithm level adjusts the heart rate benchmark in step S2 (reduced by 10 beats / minute to adapt to the natural decrease in heart rate under low temperature) and the blood oxygen warning threshold (reduced to 90% to match the low blood oxygen saturation characteristic under low temperature). In high-salinity seawater environments with salinity exceeding 30‰, the hardware activates the anti-corrosion coating activation mechanism through the water pressure sensor in step S1, causing an oxide film to form on the electrode surface. In terms of algorithms, a salinity correction coefficient is introduced in the water depth calculation in step S3. The correction coefficient is 1.05 when the salinity is 30‰~40‰ and 1.1 when the salinity is 40‰~50‰. Through coefficient correction, the water depth data error is controlled within ±0.05 meters, thereby improving the accuracy of the determination in step S4. In waters at altitudes exceeding 1500 meters, the algorithm adjusts the blood oxygen warning threshold and heart rate mutation threshold in step S2. The blood oxygen warning threshold is lowered to 89%, and the heart rate mutation threshold is lowered by 2% from the baseline value for ordinary users. This is to adapt to the characteristic that the thin air at high altitudes causes a 3% to 5% reduction in normal blood oxygen levels, ensuring that the judgment logic in step S4 conforms to the physiological data characteristics of high-altitude environments.

[0039] This invention also provides a drowning prevention alarm bracelet based on heart rate and water pressure recognition, which, based on the above method, includes: The physiological monitoring module integrates a heart rate sensor for collecting heart rate data; The water pressure monitoring module, equipped with a water pressure sensor, is used to detect water depth and immersion status; The motion posture semantic recognition module integrates a six-axis motion sensor and a motion feature extraction chip to identify active paddling, drowning struggle, and accidental sinking movements. The water temperature sensing extension module is equipped with a water temperature sensor and a database linking water temperature and muscle movement, used to collect water temperature data. The collaborative communication module adopts a 5G and Bluetooth dual-mode communication architecture to push alarm information and location data; The processing module, electrically connected to the above module, is used to establish individual physiological benchmarks and combine them with dynamic water temperature calibration to determine the threshold and jointly determine the risk of drowning. The alarm module, controlled by the processing module, can trigger tiered audible and visual alarms. The power supply and carrier module uses a flexible waterproof carrier to adapt to different wrist sizes and is equipped with a low-power rechargeable battery; The environment adaptation module is electrically connected to the processing module. It is used to optimize the device's adaptability in complex environments such as low temperature, high salinity, and high altitude through hardware insulation, anti-corrosion activation, and algorithm threshold adjustment.

[0040] 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.

[0041] 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.

Claims

1. A drowning prevention alarm method based on heart rate and water pressure recognition, characterized in that: The specific steps of this method are as follows: Step S1: Set up the motion posture semantic recognition module, water temperature sensing extension module, physiological monitoring module, water pressure monitoring module, and collaborative communication module; Step S2: Real-time environmental compensation internal circulation periodically adjusts thresholds according to water temperature and water pressure, while periodic individual and behavioral baseline external circulation periodically updates physiological benchmarks based on user behavior; Step S3: Based on the data collection frequency adapted to the monitoring status, establish a four-dimensional scene tag system and bind the data to the scene; Step S4: Set basic judgment conditions, action semantic conditions, and water temperature correlation conditions. When two or more conditions are met, a suspected drowning judgment is triggered. Step S5: After determining suspected drowning, extract five risk features from the data in step S3. Based on the risk features, refine the risk level: Level 1 is emergency drowning, Level 2 is non-emergency anomaly, and Level 3 is accidental triggering. Step S6: Based on the risk level, a Level 1 alarm will link local, near-field and remote resources, a Level 2 alarm will only notify relevant personnel, and a Level 3 false trigger will only provide a local notification and will also feed back the false trigger data to Step S2; Step S7: Optimize the calibration mechanism in step S2 and the hierarchical model in step S5; Step S8: Design an environment adaptation module to optimize the device's adaptability in complex environments from both hardware and algorithm perspectives.

2. The drowning prevention alarm method based on heart rate and water pressure recognition according to claim 1, characterized in that: The specific steps of S1 are as follows: The motion posture semantic recognition module is equipped with a six-axis motion sensor and a motion feature extraction chip to identify three types of semantic actions: active paddling, drowning struggle, and accidental sinking, which are used to distinguish between active movement and passive danger. The water temperature sensing extension module is equipped with a water temperature sensor and a water temperature and muscle movement correlation database. The correlation database pre-stores the correlation curves between water temperature and the intensity of struggling movements for three groups of people: children, adults and the elderly, within a water temperature range of 10 to 40°C. It also stores the user's group identifier. The physiological monitoring module integrates a heart rate sensor and a blood oxygen sensor for collecting heart rate and blood oxygen data; The step S1 also includes a power supply module equipped with a rechargeable battery that supports a low-temperature insulation mode. The water pressure monitoring module is equipped with a water pressure sensor and a salinity sensor. The salinity sensor has a detection range of 0‰ to 50‰ and a detection accuracy of ±1‰, and is used to detect the salinity of the water in real time. The water pressure sensor is used to detect the water depth and immersion status. The salinity sensor and the water pressure sensor are linked. When the salinity detection value exceeds 30‰, the anti-corrosion coating activation command of the water pressure sensor is triggered. The collaborative communication module adopts a dual-mode communication architecture of 5G and Bluetooth, and has a built-in near-field rescue terminal collaborative protocol to establish closed-loop communication for alarm, response and feedback with smart lifebuoys within a 100-meter range, swimming pool lifeguard terminals within a 500-meter range and lifeguard terminals.

3. The drowning prevention alarm method based on heart rate and water pressure recognition according to claim 2, characterized in that: Step S2 is as follows: Based on the water temperature, water pressure, movement posture and physiological parameter data collected in step S1, a dual calibration mechanism is established, consisting of real-time environmental compensation internal circulation and periodic individual and behavioral baseline external circulation. A dynamic calibration database is also provided to store the water temperature, water pressure, salinity, physiological parameters and subsequent feedback data required for calibration. The real-time environmental compensation internal circulation is executed once every 10 seconds. On the one hand, it calculates and dynamically adjusts the heart rate mutation threshold according to the rule that the heart rate mutation threshold is lowered by 1.5% for every 1°C decrease in water temperature. On the other hand, combined with the water pressure and vertical acceleration data detected in step S1, if the vertical sinking acceleration is detected to be lower than -0.5g in the deep water environment, the blood oxygen warning threshold is further lowered by 1%. The water depth ≥ 1 meter is defined as the deep water area and the water depth < 1 meter is defined as the shallow water area. The periodic individual and behavioral baseline external cycle is executed once every 24 hours, and the physiological baseline is adjusted by combining the user's resting, movement and water activity data collected in step S1 within 24 hours; Meanwhile, for the user's work scenarios identified in step S1, if the user has 3 or more shallow water work records in the past 7 days, the heart rate fluctuation threshold in shallow water will be relaxed from the normal user benchmark value of 25% to 30%.

4. The drowning prevention alarm method based on heart rate and water pressure recognition according to claim 3, characterized in that: Step S3 is as follows: Based on the threshold calibrated in step S2, at the data acquisition level, the monitoring state is divided according to the normal or suspected danger threshold determined in step S2. In the normal monitoring state, data is collected at a specific frequency through step S1, wherein the water temperature and water pressure data are collected every 0.3 seconds, the heart rate and blood oxygen data are collected every 1 second, and the movement posture data are collected every 0.2 seconds. When entering a suspected dangerous state, the frequency of motion posture acquisition is increased to once every 0.05 seconds, while simultaneously recording the duration and intensity of the movement. At the scene labeling level, combining the action semantics and user group identifiers identified in step S1, and the data features calibrated in step S2, five types of data—water temperature, water pressure, salinity, heart rate, and blood oxygen—are treated as five-parameter time-series data and continuously recorded according to the collection timestamp; a scene labeling system is established with five dimensions: identity, water depth, movement state, action semantics, and salinity environment, and the labels are bound to the monitoring data according to the timestamp.

5. The drowning prevention alarm method based on heart rate and water pressure recognition according to claim 4, characterized in that: Step S4 is as follows: Combining the five-parameter time-series data and semantic scene tags from step S3, and referring to the dynamic threshold from step S2, a time-series correlation judgment logic is established. The judgment logic is considered to be suspected drowning if it meets two or more of the following conditions: the first is the basic judgment condition, namely, water depth ≥ 1 meter, heart rate ≥ 25% increase compared to the threshold calibrated in step S2, and blood oxygen ≥ 100% of the warning threshold calibrated in step S2. The second is the action semantic condition, that is, step S3 identifies drowning struggle or accidental sinking action, and the duration exceeds 1 second; the third is the water temperature correlation condition, that is, when the water temperature collected in step S3 is below 20℃, if the heart rate drops suddenly and is accompanied by passive sinking action, even if the blood oxygen does not reach the warning threshold calibrated in step S2, a suspected judgment will be triggered. If step S3 detects active paddling motions, it is determined to be normal exercise even if the heart rate exceeds the threshold calibrated in step S2.

6. The drowning prevention alarm method based on heart rate and water pressure recognition according to claim 5, characterized in that: Step S5 is as follows: Based on the determination of suspected drowning in step S4, five dynamic risk features are first extracted from the five-parameter time-series data collected in step S3, including the duration of struggling, the rate of decrease in blood oxygen, the change in water depth, the action semantic risk coefficient combined with the action semantic recognition in step S1, and the water temperature correlation risk coefficient combined with the water temperature data in step S3. The initial weight of the water temperature correlation risk coefficient is 0.

3. Subsequently, a three-level risk classification is determined based on the dynamic risk characteristics. Level 1 is emergency drowning, which requires the simultaneous fulfillment of three key requirements, as follows: In terms of movement: it recognizes two types of movements: drowning struggle and accidental sinking, and the drowning struggle lasts for more than 3 seconds; In terms of physiological indicators: blood oxygen levels drop rapidly, with a decrease of more than 5% every 5 minutes; In terms of environmental conditions: it is located in deep water and the water temperature and corresponding risk coefficients meet the requirements: risk coefficient of 1.3 for 10-15℃, risk coefficient of 1.1 for 15-20℃, and risk coefficient of 1.0 for 20-25℃. Level 2 is a non-urgent anomaly and must meet any of the following conditions: Combination of action and environmental / physiological indicators: Drowning struggle lasts for 1-3 seconds and blood oxygen does not decrease or the person is in shallow water; Combination of environmental and physiological indicators: water temperature below 15℃, sudden drop in heart rate and no drowning struggles; Level 3 is a false trigger. If the action semantic tag corresponding to active paddling is detected, or if the user completes dual confirmation by touch button and voice within the preset confirmation time, it is determined to be a false trigger and the alarm process is terminated.

7. The drowning prevention alarm method based on heart rate and water pressure recognition according to claim 6, characterized in that: Step S6 is as follows: Based on the three-level risk level in step S5, the corresponding rescue response mechanism is activated. For level one emergency drowning, a level one alarm is activated, and the local area activates the buzzer through the hardware in step S1, while simultaneously flashing red and blue lights and vibrating to remind the user. In the near field, the collaborative communication module in step S1 sends the target location and trajectory prediction to the smart lifebuoy within 100 meters, triggering the lifebuoy's automatic propulsion function. At the same time, it pushes the user's identity, physiological data collected in step S3, and water temperature environment to the rescue pavilion terminal within 500 meters, enabling the terminal to automatically generate a rescue route combined with the ocean current direction. In the remote field, the alarm signal, real-time positioning, and the five-parameter time-series curve from step S3 are transmitted to the rescue platform with the highest priority via the 5G network. At the same time, it dials three preset emergency contact numbers, and automatically plays the user's name and drowning location after the call is connected. For non-emergency level 2 anomalies, a level 2 alarm is activated, with a short beep from the local buzzer accompanied by a yellow flashing light; the anomaly notification and the current location collected in step S3 are sent only to the emergency contact, while a notification is pushed to the near-field Bluetooth terminal. For a false trigger of level three, a level three alarm is activated, with only the local green light flashing for 10 seconds. At the same time, the cause of the false trigger is recorded, and the data is fed back to the dynamic calibration database in step S2.

8. The drowning prevention alarm method based on heart rate and water pressure recognition according to claim 7, characterized in that: Step S7 is as follows: Based on the feedback data from the rescue process in step S6, the collaborative communication module in step S1 first receives feedback data from the rescue platform and lifeguard terminal, including rescue arrival time, rescue effectiveness, and reasons for false alarms. The feedback data is then correlated with the judgment results in step S4 and the grading results in step S5. Subsequently, the core model was optimized based on the associated data. On the one hand, for the dynamic calibration system in step S2, if the accuracy rate of the first-level alarm in step S5 is less than 90% in a certain scenario, the weight of the water temperature associated risk coefficient will be automatically increased. On the other hand, for the graded model in step S5, if the false alarm rate in a certain area exceeds 20%, the action semantic recognition threshold in step S3 will be adjusted for that area. Federated learning technology is used so that the user physiological data collected in step S3 is used for model training only locally, and the cloud only receives the desensitized feature weights.

9. The drowning prevention alarm method based on heart rate and water pressure recognition according to claim 8, characterized in that: Step S8 is as follows: The design environment adaptation module is designed so that, in low-temperature environments, the hardware level activates the low-temperature insulation mode through the power supply module in step S1, and uses the built-in heating element to maintain the battery temperature above 5°C; the algorithm level adjusts the heart rate benchmark and blood oxygen warning threshold in step S2. In a high-salinity seawater environment, at the hardware level, the salinity sensor of the water pressure monitoring module in step S1 first collects the water salinity data in real time. When the salinity detection value exceeds 30‰, it is determined to be a high-salinity seawater environment, and then the anti-corrosion coating activation mode of the water pressure sensor is triggered. A dense oxide film is generated through the micro-current oxidation reaction on the electrode surface. At the algorithm level, based on the real-time salinity value collected by the salinity sensor, a dynamic salinity correction coefficient is introduced in the water depth calculation in step S3: the correction coefficient is 1.05 when the salinity is 30‰~40‰, and the correction coefficient is 1.1 when the salinity is 40‰~50‰. The coefficient correction ensures that the water depth data error is controlled within ±0.05 meters. In high-altitude waters, the blood oxygen warning threshold and heart rate mutation threshold in step S2 are adjusted at the algorithm level to ensure that the judgment logic in step S4 conforms to the physiological data characteristics of the high-altitude environment.

10. A drowning prevention alarm bracelet based on heart rate and water pressure recognition, wherein the method described in claim 9 is characterized in that: include: Physiological monitoring module, integrating a heart rate sensor; Water pressure monitoring module, equipped with a water pressure sensor; The motion posture semantic recognition module integrates a six-axis motion sensor and a motion feature extraction chip; Water temperature sensing expansion module, configured with water temperature sensor and database linking water temperature and muscle movement; Power supply module, equipped with a rechargeable battery; The collaborative communication module adopts a dual-mode communication architecture of 5G and Bluetooth; The processing module, electrically connected to the above module, is used to establish individual physiological benchmarks and combine them with dynamic water temperature calibration to determine the threshold and jointly determine the risk of drowning. The alarm module, controlled by the processing module, is used to trigger tiered audible and visual alarms. The carrier module uses a flexible, waterproof carrier to adapt to different wrist sizes; The environment adaptation module is electrically connected to the processing module. Through hardware insulation, anti-corrosion activation, and algorithm threshold adjustment, it optimizes the device's adaptability in complex environments.

Images