A smart watch-based intelligent early warning method for abnormal fluctuations in blood oxygen

Abstract

The application discloses a kind of based on smart watch's blood oxygen abnormal fluctuation intelligent early warning method, belong to intelligent wearable equipment health monitoring technical field.The method includes: by the photoplethysmography sensor of smart watch emission and reception two kinds of wavelength optical signal, conversion into electrical signal;The continuous waveform sequence of light attenuation amount in each pulsation cycle is constructed, and time domain morphological parameter is obtained by envelope fitting;Extract light attenuation rising branch curve segment, and the stored standard rising branch curve segment is waveform matched, and shape difference cumulative value is calculated;According to shape difference cumulative value, identify waveform distortion signal section, record its duration and occurrence frequency;Calculate fluctuation mutation coefficient, and trigger blood oxygen abnormal early warning signal when exceeding preset threshold value.The method can capture blood oxygen abnormal fluctuation characteristics from pulse waveform morphological difference, and dynamic early warning is carried out based on the time-frequency characteristics of waveform distortion event.

Description

Technical Field

[0001] This invention relates to the field of health monitoring technology for smart wearable devices, specifically to a smart watch-based method for intelligent early warning of abnormal blood oxygen fluctuations. Background Technology

[0002] Blood oxygen saturation is a key physiological indicator reflecting the body's oxygenation status. Continuous daily monitoring via wearable devices such as smartwatches is valuable for screening potential risks like sleep apnea and chronic obstructive pulmonary disease. Photoplethysmography (POP) utilizes two different wavelengths of light to penetrate biological tissue, calculating blood oxygen levels based on the difference in light absorption between oxyhemoglobin and deoxyhemoglobin. This is currently the mainstream non-invasive detection method. Existing blood oxygen anomaly detection technologies typically rely directly on the calculated blood oxygen saturation value, setting a fixed lower threshold for alarms. When the blood oxygen value at several consecutive sampling points falls below this threshold, a low blood oxygen alert is triggered. While this method is somewhat reliable under ideal, static conditions, it has significant limitations in everyday wear. When a user wears a smartwatch, limb movements, relative displacement between the watch and skin, and changes in ambient light can all introduce motion artifacts, increasing photoelectric signal noise and causing false drops in instantaneous blood oxygen saturation calculations, leading to frequent false alarms. Even with accelerometer-based motion compensation, it's difficult to completely eliminate signal singularities, and the fixed threshold method doesn't fully utilize the rich feature information contained in the waveform morphology. The occurrence of abnormal blood oxygenation is often accompanied by regular changes in the pulse wave waveform itself, such as subtle distortions in the slope of the rising limb, the sharpness of the peak, and the position of the diabetic notch. These changes occur before a significant drop in blood oxygen saturation or persist even after the saturation level recovers. Existing methods fail to extract morphological differences reflecting physiological fluctuations in blood oxygenation from the optical decay waveform profile within the pulse cycle. Therefore, they cannot identify implicit waveform distortion signals before the value falls below a threshold, nor can they determine the temporal clustering and sudden changes of distortion events. How to capture waveform matching differences caused by blood oxygen fluctuations from the rising limb of the optical decay envelope, and construct a mutation evaluation index that differs from conventional fluctuation patterns based on the persistence and frequency of the distortion signal segment, has become a key issue in suppressing false alarms and accurately alerting to abnormal blood oxygenation fluctuations. Summary of the Invention

[0003] The purpose of this invention is to provide an intelligent early warning method for abnormal blood oxygen fluctuations based on a smartwatch, in order to solve the problems of existing technologies that rely on fixed blood oxygen saturation thresholds, resulting in high false alarms, inability to effectively utilize pulse wave morphology changes to identify early abnormal waveform distortion, and lack of dynamic assessment of the temporal clustering and suddenness of waveform distortion events. This invention achieves intelligent early warning of abnormal blood oxygen fluctuations from two levels: matching the morphology of light attenuation waveforms and analyzing abrupt changes in the time-frequency characteristics of distortion.

[0004] To achieve the above objectives, the present invention provides the following technical solution: The present invention provides an intelligent early warning method for abnormal blood oxygen fluctuations based on a smartwatch. A photoplethysmography (PPG) sensor emits a first-wavelength light signal and a second-wavelength light signal to the skin of the wrist, and receives the light signal after tissue reflection, converting it into an electrical signal. Based on the electrical signal, a continuous waveform sequence of light attenuation within each pulsation cycle is constructed. The continuous waveform sequence is then fitted with an envelope to obtain the temporal morphological parameters of the light attenuation envelope. Based on the temporal morphological parameters, a rising branch curve segment of light attenuation is extracted within each pulsation cycle. This rising branch curve segment is then matched with a stored standard rising branch curve segment to calculate the cumulative value of shape difference. Based on the cumulative value of shape difference, waveform distortion signal segments in the blood oxygen signal are identified, and the duration and frequency of these waveform distortion signal segments are recorded. A fluctuation abrupt change coefficient is calculated based on the duration and frequency of occurrence. When the fluctuation abrupt change coefficient exceeds a preset threshold, an abnormal blood oxygen warning signal is triggered. This method utilizes subtle distortions in the pulsation wave morphology to capture the non-stationary characteristics of blood oxygen fluctuations, enabling early warnings to be issued before the blood oxygen level reaches a dangerously low value, reducing the risk of missed detections.

[0005] As a preferred embodiment of the present invention, the process of constructing a continuous waveform sequence and obtaining the time-domain morphological parameters of the optical attenuation envelope specifically involves: performing bandpass filtering on the electrical signal to remove DC components and high-frequency noise, and extracting the AC component waveform; detecting the time interval between two adjacent troughs in the AC component waveform, and defining a pulsation cycle by the time interval; within each pulsation cycle, acquiring the optical attenuation amplitude of each sampling point according to the sampling time sequence to form the continuous waveform sequence; performing cubic spline interpolation on all peaks in the continuous waveform sequence to generate an upper envelope, and performing cubic spline interpolation on all troughs to generate a lower envelope; calculating the difference between the upper and lower envelopes at each sampling time to obtain an instantaneous pulse amplitude sequence, which serves as the time-domain morphological parameter. By extracting the envelope difference to obtain the instantaneous pulse amplitude, the influence of baseline drift and individual differences on the absolute value of optical attenuation is effectively eliminated, allowing the morphological parameters to more purely reflect the blood flow pulsation characteristics and providing a reliable basis for subsequent waveform matching.

[0006] Preferably, the process of extracting the optical attenuation rising branch curve segment and calculating the cumulative shape difference value is as follows: In the instantaneous pulse amplitude sequence, all sampling points corresponding to the rise from the trough amplitude to the peak amplitude within each pulsation cycle are determined and connected in chronological order to form the optical attenuation rising branch curve segment; a standard rising branch curve segment generated by averaging the rising branch sampling points within historical normal pulsation cycles is retrieved from the storage unit; the optical attenuation rising branch curve segment and the standard rising branch curve segment are aligned to the same starting point on the time axis, and the absolute value of the amplitude difference is calculated for each sampling point to obtain the instantaneous difference sequence; the absolute values ​​of the amplitude differences of all sampling points in the instantaneous difference sequence are summed to obtain the cumulative shape difference value. This method focuses on the rising branch, a waveform locality sensitive to changes in blood oxygenation. The cumulative shape difference value can sensitively quantify the degree of abnormality in the pulsation waveform while suppressing the interference of the slowly changing portion during diastole, thus enhancing the specificity of abnormality detection.

[0007] Furthermore, before waveform matching, the standard rising branch curve segment is stored after being normalized according to the duration of the pulsation cycle. The light attenuation rising branch curve segment is then subjected to the same duration normalization process before alignment and matching. Through duration normalization, the scale difference caused by waveform stretching under different heart rates is eliminated, making the morphological comparison between different cycles more objective. The cumulative value of shape difference can more accurately reflect the waveform distortion itself rather than the deformation caused by heart rate.

[0008] Preferably, the process of identifying waveform distortion signal segments and recording their duration and frequency of occurrence specifically involves: comparing the cumulative value of shape difference corresponding to each pulsation cycle with a first distortion threshold; when the cumulative value of shape difference is greater than the first distortion threshold, marking the pulsation cycle as a distortion cycle; continuously monitoring multiple pulsation cycles, merging multiple pulsation cycles consecutively marked as distortion cycles into one waveform distortion signal segment; recording the start and end timestamps of the waveform distortion signal segment, calculating the duration, and counting the number of occurrences of the waveform distortion signal segment per unit time as the frequency of occurrence. This method aggregates isolated distortion cycles into continuous abnormal events, avoiding frequent instantaneous false alarms caused by sporadic interference. Simultaneously, the recorded duration and frequency of occurrence objectively reflect the scale and concentration of abnormal events, providing structured features for subsequent mutation determination.

[0009] As a further improvement, the first distortion threshold is dynamically set based on the mean and standard deviation of the cumulative shape differences of the current user's historical normal pulsation cycles. This allows the distortion criterion to adapt to the user's own physiological waveform characteristics, maintaining reasonable judgment sensitivity in different states of the same individual and among different individuals, reducing missed detections or false alarms caused by individual differences.

[0010] During the merging of distortion cycles, if the number of normal cycles between two adjacent distortion cycles is less than a preset interval threshold, they are merged into the same waveform distortion signal segment. This measure can tolerate brief normal fluctuations and completely merges consecutive abnormal events belonging to the same pathological process, making the calculated duration and frequency of occurrence more consistent with the actual physiological abnormal process.

[0011] Preferably, the process of calculating the fluctuation mutation coefficient and triggering an early warning is as follows: multiply the duration and frequency of the waveform distortion signal segment by the corresponding first weighting coefficient and second weighting coefficient, respectively, and add the products to obtain a weighted cumulative value; obtain the historical weighted cumulative values ​​of multiple historical pulsation cycles within the previous time window, and perform an arithmetic mean on the historical weighted cumulative values ​​to obtain the historical mean; divide the weighted cumulative value by the historical mean to obtain the ratio, which is used as the fluctuation mutation coefficient; when the fluctuation mutation coefficient is greater than the mutation judgment threshold, the blood oxygen abnormality early warning signal is triggered. By using weighted accumulation to jointly evaluate the duration and frequency, and then comparing it with the historical baseline to obtain the mutation coefficient, it is possible to keenly capture sudden deterioration of blood oxygen fluctuation characteristics, while avoiding overreaction to slow trend changes, thus balancing the timeliness and accuracy of the early warning.

[0012] To further enhance robustness, the length of the preceding time window is preferably set to 30 beat cycles. Furthermore, before calculating the historical mean, outliers deviating from the mean by more than three standard deviations are removed before the arithmetic mean is calculated. This process eliminates potential occasional strong interference or artifacts within the window, making the historical baseline more robust and ensuring that the fluctuation mutation coefficient is only sensitive to truly anomalous sudden changes.

[0013] Furthermore, when determining the pulsation cycle by detecting the time interval between two adjacent troughs, if the detected time interval exceeds the preset normal pulsation interval range, the current pulsation cycle is discarded and re-detected. This mechanism can effectively filter out erroneous cycle segmentation caused by motion artifacts or occasional arrhythmias, ensuring that only reliable pulsation cycles with complete physiological information are entered into subsequent waveform analysis, thus improving the anti-interference capability of the overall early warning link.

[0014] The technical effects and advantages provided by the present invention in the above technical solution are as follows: By receiving the electrical signal output from a photoplethysmography (PPG) sensor, a continuous waveform sequence of optical attenuation within each pulse cycle is constructed. Cubic spline interpolation is then performed on the peaks and troughs of the sequence to obtain upper and lower envelopes. Using the instantaneous pulse amplitude sequence as the time-domain morphological parameter, the rising limb segment of optical attenuation corresponding to the rise from the trough amplitude to the peak amplitude is extracted within each pulse cycle. This segment is aligned with the standard rising limb segment generated by averaging historical normal pulse cycles to the same starting point on the time axis. The absolute value of the amplitude difference is calculated for each sampling point, and the cumulative value of the shape difference is obtained. This method directly characterizes the local morphological deviation of the optical attenuation waveform contour during pulse wave formation, rather than relying solely on calculated blood oxygen saturation values. It can sensitively capture subtle distortion features such as changes in the steepness of the rising limb and blunting of the peak shape caused by abnormal blood oxygenation. When motion or light interference causes instantaneous baseline fluctuations, the interference signal is unlikely to regularly alter the envelope trend of the entire rising limb. Therefore, the cumulative value of shape difference will not increase significantly, effectively suppressing false differences caused by daily wear interference, making the waveform distortion judgment result more consistent with the true physiological fluctuations of blood oxygen. The cumulative value of shape difference is compared with a dynamically set first distortion threshold. Pulse cycles exceeding the threshold are marked as distortion cycles. After continuous monitoring of multiple pulse cycles, temporally consecutive distortion cycles are merged into waveform distortion signal segments. Their duration and frequency per unit time are recorded, multiplied by the corresponding weighting coefficients, and summed to obtain a weighted cumulative value. The arithmetic mean of the historical weighted cumulative values ​​of multiple historical pulse cycles within the previous time window is then obtained as the historical mean. The current weighted cumulative value is divided by the historical mean to obtain the fluctuation mutation coefficient. This coefficient exceeding the mutation judgment threshold is used as the condition for triggering an abnormal blood oxygenation warning. This method extends the waveform morphology differences within isolated cycles into a joint representation of the persistence and frequency of distortion events over time. It establishes a dynamic baseline by statistically comparing an individual's historical fluctuation levels, avoiding the impact of fundamental differences in pulse wave morphology among different users and occasional physiological fluctuations in a calm state on the warning threshold setting. A waveform distortion signal segment is only identified as an abnormal fluctuation requiring an alarm when it simultaneously possesses a relatively long duration and a high frequency, and exhibits a sudden amplification exceeding a preset proportion compared to the weighted level of the historical window. This achieves a balance between suppressing sporadic spikes and timely capturing persistent blood oxygen deterioration events, resulting in a lower false alarm rate and more timely detection of genuine anomalies. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0016] Figure 1 This is a flowchart of a smartwatch-based intelligent early warning method for abnormal blood oxygen fluctuations; Figure 2 This is a flowchart of photoplethysmography (PPG) signal processing and instantaneous pulse amplitude acquisition. Figure 3 This is a schematic diagram of the continuous waveform and envelope of light attenuation within the pulsation cycle; Figure 4 This is a schematic diagram illustrating the validity determination of the time interval between pulsations; Figure 5 This is a schematic diagram showing the normalized time-domain comparison and instantaneous absolute value of the difference between the optical attenuation rising branch curve segment and the standard rising branch curve segment. Figure 6 This is a schematic diagram of the cumulative value of the difference in the shape of the pulsation cycle and the waveform distortion signal segment; Figure 7 This is a schematic diagram illustrating the changes in the fluctuation mutation coefficient and early warning signals of abnormal blood oxygen fluctuations. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, 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.

[0018] See Figure 1 This invention provides a smartwatch-based intelligent early warning method for abnormal blood oxygen fluctuations. The method utilizes a photoplethysmography (PPG) sensor built into the smartwatch to emit a first-wavelength light signal and a second-wavelength light signal to the skin of the wrist, and receives the light signal reflected by the tissue, converting it into an electrical signal. Based on this electrical signal, a continuous waveform sequence of light attenuation within each pulsation cycle is constructed, and envelope fitting is performed on the continuous waveform sequence to obtain the temporal morphological parameters of the light attenuation envelope. Based on the temporal morphological parameters, a rising branch curve segment of light attenuation is extracted within each pulsation cycle. This rising branch curve segment is then waveform-matched with a pre-stored standard rising branch curve segment, and the cumulative value of the shape difference between the two is calculated. Based on the cumulative value of the shape difference, signal segments in the blood oxygen signal exhibiting waveform distortion are identified, and the duration and frequency of each waveform distortion signal segment are recorded. The fluctuation abrupt change coefficient is calculated using the duration and frequency. When the fluctuation abrupt change coefficient exceeds a preset threshold, a blood oxygen abnormality early warning signal is triggered. Example 1:

[0019] In specific implementation, please refer to Figure 2The electrical signal output from the photoplethysmography (PPG) sensor of the smartwatch was bandpass filtered. The lower cutoff frequency of the bandpass filter was set to 0.5 Hz to filter out the DC component and baseline drift interference in the electrical signal, and the upper cutoff frequency was set to 10 Hz to filter out high-frequency noise and electromyographic interference in the electrical signal. After bandpass filtering, the AC component waveform reflecting the changes in pulse pulsation was extracted from the original electrical signal.

[0020] In some embodiments, a trough detection algorithm is used to identify all trough locations on the AC component waveform. The trough detection algorithm calculates the first-order difference sequence of the AC component waveform, finds the zero-crossing points in the first-order difference sequence corresponding to changes from negative to positive values, marks the sampling points corresponding to these zero-crossing points on the AC component waveform as candidate troughs, and selects the point with the smallest amplitude from the candidate troughs as the trough. The time interval between two adjacent troughs is detected and compared with a preset normal heart rate interval range. The normal heart rate interval range is set according to the heart rate range under resting conditions, with a lower limit corresponding to a heart rate interval of 0.6 seconds and an upper limit corresponding to a heart rate interval of 2.0 seconds. When the detected time interval is within the normal heart rate interval range, a heart rate cycle is defined by this time interval, and the heart rate cycle consists of all sampling points between the previous trough and the next trough.

[0021] Optionally, within each pulsation cycle, the optical attenuation amplitude corresponding to each sampling point is sequentially acquired from the AC component waveform according to the sampling time sequence of the photoplethysmography (PPG) sensor. The optical attenuation amplitude refers to the degree of attenuation of the light signal intensity after tissue reflection relative to the incident light signal intensity, expressed in absorbance units. The optical attenuation amplitudes of all sampling points within the pulsation cycle are arranged in chronological order of sampling time to form a continuous waveform sequence. Each sampling point in the continuous waveform sequence has two attributes: a sampling time index and an optical attenuation amplitude.

[0022] In some implementations, when fitting the envelope of a continuous waveform sequence, a peak detection algorithm is first used to identify all peaks in the sequence. The peak detection algorithm calculates the first-order difference sequence of the continuous waveform sequence, finds the zero-crossing points where the values ​​change from positive to negative, and marks the corresponding sampling points on the continuous waveform sequence as peaks. Cubic spline interpolation is then performed on all identified peaks. This interpolation constructs a cubic polynomial within the interval between each adjacent peak, satisfying the conditions of continuous function values, continuous first derivative, and continuous second derivative at the peaks. Cubic spline interpolation generates an upper envelope covering the entire duration of the pulsation cycle. The value of the upper envelope at each sampling time represents the upper limit of the light attenuation trend within the pulsation cycle.

[0023] Simultaneously, cubic spline interpolation is performed on all identified trough points in the continuous waveform sequence. Trough points are local minima identified in the continuous waveform sequence using a trough detection algorithm. Cubic spline interpolation constructs a cubic polynomial within the interval between each adjacent trough point, satisfying the conditions of continuous function values, continuous first derivative, and continuous second derivative at the trough point. A lower envelope covering the entire pulsation cycle duration is generated through cubic spline interpolation. The value of the lower envelope at each sampling time represents the lower limit of optical attenuation within the pulsation cycle.

[0024] In practice, for each pulsation cycle, the difference between the upper and lower envelopes at each same sampling time is calculated to obtain the instantaneous pulse amplitude sequence. The instantaneous pulse amplitude at the nth sampling time in the instantaneous pulse amplitude sequence is then calculated. The calculation method is as follows:

[0025] in, This represents the instantaneous pulse amplitude at the nth sampling time, expressed in absorbance units. This represents the amplitude of the upper envelope at the nth sampling time. This represents the amplitude of the lower envelope at the nth sampling time. The instantaneous pulse amplitudes at all sampling times within the pulsation period are arranged in chronological order to obtain the instantaneous pulse amplitude sequence, which serves as the temporal morphological parameter of the optical attenuation envelope.

[0026] See Figure 3 In the figure, the horizontal axis represents time in seconds, the left side of the vertical axis represents the light attenuation amplitude in absorbance units, and the right side represents the instantaneous pulse amplitude in absorbance units. The solid blue curve represents a continuous waveform sequence, depicting the continuous change in light attenuation within the pulsation cycle. The red dashed line represents the upper envelope, obtained by cubic spline interpolation fitting the peaks of the continuous waveform sequence; the upper envelope reflects the upper limit trend of light attenuation within the pulsation cycle. The green dotted line represents the lower envelope, also obtained by cubic spline interpolation fitting the troughs, representing the lower limit trend of light attenuation within the pulsation cycle. The purple dashed line represents the instantaneous pulse amplitude sequence, whose value is the difference between the upper and lower envelopes at the corresponding sampling time, reflecting the change in instantaneous pulse amplitude within the pulsation cycle.

[0027] As can be observed from the figure, the continuous waveform sequence exhibits a rapid upward trend from 0 seconds to approximately 0.15 seconds, reaching a peak before gradually declining. The overall curve lies between the upper and lower envelopes, reflecting the natural fluctuation process of the light attenuation signal within the pulsation cycle. The upper envelope shows a smooth decreasing trend, linearly decreasing from approximately 1.05 absorbance units initially to approximately 0.18 absorbance units at 1 second, indicating that the maximum value of light attenuation decreases uniformly over time within the pulsation cycle. The lower envelope exhibits an arc-shaped change, first rising and then falling, with the peak appearing at approximately 0.5 seconds, reflecting the lower limit of light attenuation change within the pulsation cycle.

[0028] The instantaneous pulse amplitude sequence gradually decreases from an initial absorbance of approximately 0.85 units to near zero over time, with the curve exhibiting a monotonically decreasing trend, indicating the decline in pulse amplitude from peak to trough within the pulsation cycle. This sequence is calculated using the difference between the upper and lower envelopes, representing the pulsation amplitude of the pulsation cycle and reflecting the time-domain morphological parameters of the waveform. Example 2:

[0029] In practical implementation, when detecting the time interval between two adjacent troughs on the AC component waveform, the sampling time positions of all troughs in the AC component waveform are first obtained through a trough detection algorithm. The trough detection algorithm traverses the first-order difference sequence of the AC component waveform, marking the sampling points corresponding to the zero-crossing points where the first-order difference value changes from negative to positive as candidate troughs. From the set of candidate troughs, the sampling point with the lowest amplitude within each local interval of continuous decline followed by rise is selected as the trough. The sampling index number of each trough in the AC component waveform is recorded.

[0030] In some embodiments, for any two adjacent troughs in the sampling time sequence, the time interval between the two troughs is calculated. The time interval is calculated by subtracting the sampling index number of the previous trough from the sampling index number of the later trough, obtaining the difference in the number of sampling points. This difference is then multiplied by the sampling period of the photoplethysmography (PPG) sensor to obtain the time interval value in seconds. The preset normal pulsation interval range is defined by a lower threshold and an upper threshold. The lower threshold is set to 0.6 seconds, and the upper threshold is set to 2.0 seconds. The 0.6-second lower threshold corresponds to an upper limit of 100 heart rates per minute, and the 2.0-second upper threshold corresponds to a lower limit of 30 heart rates per minute. This range covers the possible pulsation interval variations that may occur in a person at rest and during mild activity.

[0031] Optionally, the detected time interval is compared with a preset normal pulsation interval range. If the time interval is less than the lower threshold of 0.6 seconds, it indicates that the interval between the two troughs is too short, possibly caused by spurious troughs due to high-frequency noise or signal jitter; the current pulsation cycle does not possess effective physiological pulsation characteristics. If the time interval is greater than the upper threshold of 2.0 seconds, it indicates that the interval between the two troughs is too long, possibly caused by signal loss, poor sensor contact, or missed detection due to motion artifacts; the current pulsation cycle also does not possess effective physiological pulsation characteristics. In both cases, the time interval is determined to exceed the preset normal pulsation interval range.

[0032] In practice, when the time interval exceeds the preset normal pulsation interval range, the current pulsation cycle is removed. The removal process involves removing the previous trough from the list of valid troughs, keeping the record position of the next trough unchanged, and repositioning the detection starting point to the previous valid trough before the removed trough. Starting from the repositioned detection starting point, the adjacent trough detection step is repeated to find the previous valid trough adjacent to the retained next trough in the sampling time sequence. The time interval between the newly detected adjacent troughs is recalculated, and the recalculated time interval is compared again with the preset normal pulsation interval range.

[0033] In some implementations, a pulse cycle validity determination formula is introduced during the re-detection process to generate a judgment result on whether to retain the current pulse cycle after each obtained time interval. The pulse cycle validity determination formula is expressed as follows:

[0034] in, This indicates the validity status flag of the m-th detected pulsation cycle. A value of 1 indicates that the pulsation cycle is valid and should be retained. A value of 0 indicates that the pulsation cycle is invalid, and it will be removed and a re-detection will be triggered. This represents the time interval between two adjacent valley points detected at the m-th time, in seconds. This represents the lower limit threshold of the preset normal heart rate interval range, with a fixed value of 0.6 seconds. The setting is based on the heart rate interval corresponding to the upper limit of sinus heart rate in a resting state. This represents the upper limit threshold of the preset normal heart rate interval range, with a fixed value of 2.0 seconds. The setting is based on the heart rate interval corresponding to the lower limit of sinus heart rate in a resting state.

[0035] Optionally, the re-detection process continues until the time interval between two adjacent troughs falls within the preset normal pulsation interval range, i.e., the time interval meets the condition of being greater than or equal to 0.6 seconds and less than or equal to 2.0 seconds. At this point, the current pulsation cycle is marked as a valid pulsation cycle and retained. Then, the next set of adjacent troughs is detected along the sampling time sequence, and the time interval comparison and validity judgment process is repeated.

[0036] See Figure 4 In the graph, the horizontal axis represents the pulse cycle number, ranging from 0 to 600, and the vertical axis represents the time interval between two adjacent troughs, in seconds. Blue solid dots represent the time interval data of valid pulse cycles, and red crosses represent the time interval data of invalid pulse cycles. The green dashed line represents the preset lower threshold of 0.6 seconds for normal pulse intervals, and the orange dashed line represents the preset upper threshold of 2.0 seconds for normal pulse intervals.

[0037] As shown in the figure, most of the blue solid dots are distributed within the range of 0.6 seconds to 1.8 seconds, indicating that the time intervals of these pulsation cycles are within the preset normal pulsation interval range and meet the valid pulsation cycle determination criteria. The red crosses are mainly distributed in the areas below 0.6 seconds and above 1.8 seconds, indicating that the time intervals of these pulsation cycles exceed the normal pulsation interval range and are considered invalid pulsation cycles, which may be caused by high-frequency noise, signal jitter, poor sensor contact, or motion artifacts.

[0038] Specifically, the figure shows several obvious clusters of invalid pulsation cycles, concentrated in areas with time intervals far below 0.6 seconds and significantly above 1.8 seconds. This indicates that abnormal time intervals occurred within these pulsation cycles. The system promptly removes these invalid pulsation cycles based on preset thresholds to ensure that subsequent analysis is based on accurate and valid pulsation cycle data. Example 3:

[0039] In practice, for each effective pulsation cycle, the optical attenuation rising branch curve segment is determined from the instantaneous pulse amplitude sequence. The instantaneous pulse amplitude sequence is composed of the instantaneous pulse amplitudes at each sampling moment within the pulsation cycle arranged in chronological order. Within the instantaneous pulse amplitude sequence, the starting and ending positions of the rising branch are determined by locating the sampling points corresponding to the trough amplitude and the peak amplitude within the pulsation cycle. The trough amplitude sampling point is the sampling point with the smallest amplitude in the instantaneous pulse amplitude sequence within the pulsation cycle, and the peak amplitude sampling point is the sampling point with the largest amplitude in the instantaneous pulse amplitude sequence within the pulsation cycle. Using the trough amplitude sampling point as the starting point of the rising branch and the peak amplitude sampling point as the ending point, all sampling points from the starting point to the ending point, including both endpoints, are extracted and connected in chronological order of sampling time to form the optical attenuation rising branch curve segment. Each sampling point in the optical attenuation rising branch curve segment contains two attributes: the relative position of the sampling time and the instantaneous pulse amplitude.

[0040] In some embodiments, a pre-established standard rising branch curve segment is retrieved from the smartwatch's storage unit. The standard rising branch curve segment is established by accumulating multiple normal pulsation cycles of the user in a resting stable state during the smartwatch's wearing history. A normal pulsation cycle is defined as a pulsation cycle where the cumulative value of shape difference does not exceed a first distortion threshold. For each normal pulsation cycle, a rising branch sampling point sequence is extracted according to the above-described method for extracting the light attenuation rising branch curve segment. Since the duration of different pulsation cycles varies, the number of rising branch sampling points in each normal pulsation cycle differs. Therefore, each rising branch sampling point sequence undergoes duration normalization processing, mapping the rising branch duration to a unified normalized time axis. The normalized time axis evenly divides the rising branch duration into 100 equidistant intervals, and linear interpolation resampling is performed on the amplitude of the sampling points within each interval to obtain 100 normalized sampling points. The normalized ascending branch sampling point sequence of all normal pulsation cycles is arithmetically averaged point-by-point at the same normalized time axis position to obtain a sequence consisting of 100 average amplitudes, which serves as the standard ascending branch curve segment. The standard ascending branch curve segment is stored in a memory unit in the form of a normalized time axis index and the corresponding average amplitude.

[0041] Optionally, after obtaining the optical attenuation rising branch curve segment, the optical attenuation rising branch curve segment undergoes the same duration normalization processing as the standard rising branch curve segment. Specifically, the duration normalization process involves determining the number of original sampling points contained in the optical attenuation rising branch curve segment, and uniformly dividing the duration of the optical attenuation rising branch curve segment into 100 equidistant intervals. The number of original sampling points spanned by each equidistant interval on the original time axis is determined by dividing the total number of original sampling points by 100. Within each equidistant interval, the instantaneous pulse amplitude of the original sampling points contained within that interval is linearly interpolated to obtain the amplitude corresponding to the normalized time at the center of that interval. After normalization processing, the optical attenuation rising branch curve segment is converted into a sequence composed of amplitudes corresponding to 100 normalized times, with the normalized time index ranging from 1 to 100.

[0042] In practice, the normalized optical attenuation rising branch curve segment is aligned with the standard rising branch curve segment to the same starting point on the time axis. The alignment method is as follows: the first normalized time of the optical attenuation rising branch curve segment corresponds to the first normalized time of the standard rising branch curve segment, and the 100th normalized time of the optical attenuation rising branch curve segment corresponds to the 100th normalized time of the standard rising branch curve segment, ensuring that the amplitude comparison of the two curve segments is performed under the same normalized time index. At each normalized time index position, the absolute value of the difference between the amplitude of the optical attenuation rising branch curve segment and the amplitude of the standard rising branch curve segment is calculated, resulting in an instantaneous difference sequence containing 100 elements.

[0043] In some implementations, the absolute values ​​of the amplitude differences of all elements in the instantaneous difference sequence are summed to calculate the cumulative shape difference value. The formula for calculating the cumulative shape difference value is expressed as:

[0044] in, This represents the cumulative shape difference corresponding to one pulsation cycle, expressed in absorbance units. Represents the normalized time index. The value of is an integer ranging from 1 to 100; This indicates the rising branch of the normalized optical attenuation curve segment at the [number]th [time]. The instantaneous pulse amplitude at each normalized time point, expressed in absorbance units; This indicates the standard ascending branch curve segment at the 1st... The average amplitude at each normalized time point, in absorbance units, is obtained by normalizing and averaging sampling points from the rising limb of historical normal pulsation cycles. Cumulative value of shape difference. It reflects the overall deviation between the light attenuation rising branch curve segment of the current pulsation cycle and the normal rising branch waveform template.

[0045] See Figure 5In the graph, the horizontal axis represents time in seconds, ranging from 0 seconds to approximately 0.27 seconds, and the vertical axis represents the instantaneous pulse amplitude in absorbance units. The solid blue line represents the light attenuation rising branch curve segment extracted during the current pulse cycle, while the dashed red line represents the pre-stored standard rising branch curve segment. The purple shaded area indicates the absolute value of the difference between the amplitude of the light attenuation rising branch curve segment and the standard rising branch curve segment at each sampling time, reflecting the instantaneous difference.

[0046] As shown in the figure, both the rising branch of the optical attenuation curve and the standard rising branch curve exhibit an upward trend as time progresses, with the amplitude gradually increasing from approximately 0.23 to approximately 0.79, indicating the increasing process of optical attenuation within the pulsating cycle. In the first half (0~0.1 seconds), the amplitudes of the two curves are relatively close, the absolute value of the instantaneous difference is small, and the purple shading is narrow and close to zero. In the middle and later stages (approximately 0.1~0.2 seconds), the blue solid line is slightly higher than the red dashed line, showing a more significant deviation, the absolute value of the instantaneous difference increases, the purple shading gradually widens, reaches its peak, and then gradually decreases to close to zero after approximately 0.2 seconds, indicating that the shape of the rising branch of optical attenuation differs somewhat from the standard shape during this period. At the end (after 0.2 seconds), the amplitudes of the two curves tend to be consistent again, and the absolute value of the difference decreases. Example 4:

[0047] In practice, the cumulative shape difference value calculated for each pulsation cycle is compared with a first distortion threshold. The first distortion threshold is set by extracting the set of cumulative shape difference values ​​corresponding to the current user's historical normal pulsation cycles accumulated over a past period from the smartwatch's storage unit. Historical normal pulsation cycles refer to pulsation cycles previously determined to be non-distorted. The arithmetic mean of all values ​​in the cumulative shape difference value set is calculated to obtain the historical shape difference mean, and the standard deviation of all values ​​in the cumulative shape difference value set is calculated to obtain the historical shape difference standard deviation. The formula for calculating the first distortion threshold is expressed as:

[0048] in, This indicates the first distortion threshold, expressed in absorbance units. The historical average shape difference is obtained by calculating the arithmetic mean of the cumulative shape differences corresponding to all historical normal pulsation cycles of the current user, and the unit is absorbance units. The standard deviation of historical shape differences is calculated from the cumulative value of shape differences corresponding to all historical normal pulsation cycles of the current user, and the unit is absorbance units. This represents the threshold adjustment coefficient. The value of is set to 2, based on the assumption that, under the normal distribution, the probability of a value exceeding the mean plus twice the standard deviation is less than 5%, thus serving as the statistical boundary for distinguishing between normal fluctuations and abnormal distortions. The first distortion threshold is dynamically updated as historical normal pulsation cycle data accumulates. Each time a new pulsation cycle is determined to be a non-distorted cycle, its cumulative shape difference value is included in the historical cumulative shape difference value set, the historical shape difference mean and historical shape difference standard deviation are recalculated, and the value of the first distortion threshold is updated.

[0049] In some embodiments, for any given pulsation cycle, if the cumulative shape difference value corresponding to that pulsation cycle is greater than the current value of the first distortion threshold, that pulsation cycle is marked as a distortion cycle. The distortion cycle is marked by setting a Boolean distortion flag in the attribute record of the pulsation cycle, and setting the distortion flag to a true value. If the cumulative shape difference value corresponding to that pulsation cycle is less than or equal to the current value of the first distortion threshold, the distortion flag is set to a false value, indicating that the pulsation cycle is not marked as a distortion cycle.

[0050] Optionally, multiple consecutive pulsation cycles are monitored one by one, and the distortion flag status of each pulsation cycle is acquired sequentially according to the sampling time sequence. When a pulsation cycle with a true distortion flag is detected, this pulsation cycle is taken as the candidate starting point of the current waveform distortion signal segment, and the detection continues to the next adjacent pulsation cycle. If the distortion flag of the next pulsation cycle is also true, the next pulsation cycle is included in the candidate range of the current waveform distortion signal segment, and the detection continues. If the distortion flag of the next pulsation cycle is false, it is determined whether the candidate range of the current waveform distortion signal segment is valid. If the candidate range contains only one distortion cycle, the single distortion cycle is taken as an independent waveform distortion signal segment. If the candidate range contains multiple consecutive distortion cycles, all distortion cycles in the candidate range are merged into one waveform distortion signal segment.

[0051] In some implementations, during the process of merging multiple pulsation cycles consecutively marked as distortion cycles into a single waveform distortion signal segment, a determination of the number of normal cycles between adjacent distortion cycles is introduced. When there is at least one normal cycle between two distortion cycles that are adjacent in sampling time, the number of normal cycles is counted. A preset interval threshold is set to two normal cycles. This is based on the assumption that when the number of normal pulsation cycles between two distortion events does not exceed two, it indicates that the two distortion events are closely correlated in time and may be caused by the continuous influence of the same interference source or the same physiological abnormality; therefore, they should be merged into the same waveform distortion signal segment for overall analysis. When the number of normal cycles is less than the preset interval threshold of 2, two adjacent distortion cycles and the normal cycle between them are merged into the same waveform distortion signal segment. When the number of normal cycles is greater than or equal to the preset interval threshold of 2, two adjacent distortion cycles are divided into two independent waveform distortion signal segments.

[0052] In practical implementation, for each identified waveform distortion signal segment, the start and end timestamps of the waveform distortion signal segment are recorded. The start timestamp is the start time of the first distortion cycle within the waveform distortion signal segment, and the end timestamp is the end time of the last distortion cycle within the waveform distortion signal segment. The duration of the waveform distortion signal segment is calculated, determined by the time difference between the end and start timestamps, in seconds. The number of times the waveform distortion signal segment appears per unit time is counted, which is used as the occurrence frequency. The unit time is set to 60 seconds. The occurrence frequency is calculated by counting the number of waveform distortion signal segments within the most recent sliding time window, dividing the count by 60 seconds, and obtaining the occurrence frequency value in times per second. The length of the sliding time window is set to 300 seconds. The sliding time window slides forward once every 60 seconds, and after each slide, the number of occurrences of waveform distortion signal segments within the window is recounted and the occurrence frequency is updated.

[0053] See Figure 6 In the graph, the horizontal axis represents the pulsation cycle number, and the vertical axis represents the cumulative shape difference, in absorbance units. The blue solid line is the cumulative shape difference curve for each pulsation cycle, and the red dashed line is the dynamically calculated first distortion threshold. The cumulative shape difference reflects the overall deviation of the current pulsation cycle's light attenuation rising branch curve segment from the standard rising branch curve segment. The first distortion threshold is dynamically updated based on the mean and standard deviation of shape differences from historical normal pulsation cycles; in the current graph, the threshold shows a slow decreasing trend with the pulsation cycle number.

[0054] The orange dots in the diagram indicate pulsation cycles identified as distorted cycles, i.e., cycles whose cumulative shape difference exceeds the first distortion threshold at that time. It can be seen that multiple distorted cycles are concentrated in four intervals: approximately 100-150, 250-280, 420-450, and 530-550. These intervals are marked with light purple rectangles as waveform distortion signal segments, representing abnormal waveform intervals formed by the merging of multiple consecutive distorted cycles. Example 5:

[0055] In practice, the duration of each waveform distortion signal segment is multiplied by a first weighting coefficient, and the frequency of occurrence of the same waveform distortion signal segment is multiplied by a second weighting coefficient. The two products are then added together to obtain a weighted cumulative value for that waveform distortion signal segment. The first weighting coefficient is set to 0.6, and the second weighting coefficient is set to 0.4. The basis for setting the first and second weighting coefficients is that the duration reflects the severity of a single distortion event, while the frequency of occurrence reflects the temporal density of distortion events. In the early warning of abnormal blood oxygen fluctuations, the duration of a single distortion event has a slightly stronger characterizing effect on abnormal physiological states than the frequency of occurrence. Therefore, the first weighting coefficient corresponding to the duration is set higher than the second weighting coefficient corresponding to the frequency of occurrence, and the sum of the two is 1, forming a weighted combination.

[0056] In some embodiments, the historical weighted cumulative values ​​of multiple historical pulsation cycles within the previous time window are obtained. The length of the previous time window is 30 pulsation cycles. Based on the current detection time, 30 completed pulsation cycles are traced back, and the weighted cumulative value of the waveform distortion signal segment to which each pulsation cycle belongs is extracted. If a pulsation cycle is not marked as a distortion cycle, its corresponding weighted cumulative value is 0. If a waveform distortion signal segment covers multiple pulsation cycles, then the weighted cumulative value of the entire waveform distortion signal segment is used as the weighted cumulative value of that pulsation cycle for each of these multiple pulsation cycles. The weighted cumulative values ​​of the 30 pulsation cycles are combined into a historical weighted cumulative value set.

[0057] Optionally, outlier removal can be performed on 30 values ​​in the historical weighted cumulative value set. First, calculate the arithmetic mean of the 30 values ​​to obtain the historical weighted cumulative value mean. Then, calculate the standard deviation of the 30 values ​​to obtain the historical weighted cumulative value standard deviation. Each value is examined individually. If the absolute value of the deviation from the historical weighted cumulative value mean exceeds three times the historical weighted cumulative value standard deviation, that value is considered an outlier. The three-standard-deviation rule is based on the assumption of a normal distribution, where the probability of a data point falling outside the range of the mean plus or minus three standard deviations is less than 0.3%, representing a low-probability event. Such values ​​are highly likely to be generated by occasional, drastic interference and are not representative, thus they should be removed. The values ​​identified as outliers are deleted from the historical weighted cumulative value set, and the arithmetic mean of all remaining values ​​is recalculated; the result is used as the historical mean.

[0058] In some implementations, the fluctuation abrupt change coefficient is obtained by calculating the ratio of the weighted cumulative value of the current waveform distortion signal segment to the historical mean. The formula for calculating the fluctuation abrupt change coefficient is expressed as:

[0059] in, This represents the fluctuation abrupt change coefficient, which is dimensionless. This represents the weighted cumulative value of the current waveform distortion signal segment. It is obtained by multiplying the duration of the current waveform distortion signal segment by the first weighting coefficient 0.6, and adding the occurrence frequency of the current waveform distortion signal segment by the second weighting coefficient 0.4; The historical mean is represented by the arithmetic mean of the weighted cumulative historical values ​​of 30 historical pulsations within the previous time window, after removing outliers that deviate from the historical weighted cumulative mean by more than three times the standard deviation of the historical weighted cumulative values.

[0060] In practical implementation, when the fluctuation mutation coefficient When the value exceeds the mutation threshold, a blood oxygen abnormality warning signal is triggered. The mutation threshold is set to 1.8. The basis for setting the mutation threshold to 1.8 is that when the weighted cumulative value exceeds 1.8 times the historical average, it indicates that the abnormal fluctuation of the blood oxygen signal reflected by the current waveform distortion signal segment has reached more than 1.8 times the historical normal fluctuation level, constituting a statistically significant deviation, which can effectively distinguish between physiological fluctuations and pathological abnormal fluctuations. The blood oxygen abnormality warning signal is generated by pushing a warning prompt message on the smartwatch's display interface, and simultaneously generating a vibration reminder for 2 seconds via a vibration motor.

[0061] See Figure 7In the graph, the horizontal axis represents time (seconds), and the vertical axis represents the fluctuation and mutation coefficient. The solid curve represents the fluctuation and mutation coefficient value changing over time, and the red dashed line represents the preset mutation judgment threshold of 1.8. The orange star mark in the legend represents the blood oxygen abnormality warning signal triggered at the corresponding time point.

[0062] As shown in the graph, the fluctuation abrupt change coefficient remained below 1.8 for most of the time, indicating a relatively stable blood oxygenation signal with small fluctuation amplitude. Between approximately 150 and 200 seconds, the fluctuation abrupt change coefficient exhibited two significant peaks, both exceeding the threshold of 1.8, with the highest peak approaching 3.3. Warning signals were marked near these peaks, indicating abnormal fluctuations in the blood oxygenation signal during this period, and the system successfully triggered an anomaly warning. Subsequently, the fluctuation abrupt change coefficient returned to below the threshold, maintaining a relatively stable state for a period.

[0063] Between approximately 380 and 440 seconds, the fluctuation abrupt change coefficient exhibited multiple consecutive peaks with high amplitudes, the highest peak approaching 4.0, all significantly exceeding the abrupt change threshold of 1.8, and warning signals were triggered at multiple time points. The continuous occurrence of fluctuation abrupt change coefficient peaks during this phase indicates that the blood oxygen signal was in a state of continuous abnormal fluctuation, and the system's weighted cumulative value for the duration and frequency of this waveform distortion signal segment was high, meeting the abnormal warning conditions. The fluctuation abrupt change coefficient rapidly decreased after 420 seconds, returning to the normal range.

[0064] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for intelligent early warning of abnormal blood oxygen fluctuations based on a smartwatch, characterized in that, The method includes: The smartwatch uses a photoplethysmography sensor to emit a first wavelength light signal and a second wavelength light signal to the skin of the wrist, and receives the light signal reflected by the tissue and converts it into an electrical signal. Based on the electrical signal, a continuous waveform sequence of light attenuation within each pulsation cycle is constructed, and envelope fitting is performed on the continuous waveform sequence to obtain the time-domain morphological parameters of the light attenuation envelope. Specifically, this includes: performing bandpass filtering on the electrical signal to filter out DC components and high-frequency noise, extracting the AC component waveform; detecting the time interval between two adjacent troughs in the AC component waveform, defining a pulsation cycle by the time interval; collecting the light attenuation amplitude at each sampling point according to the sampling time sequence within each pulsation cycle to form the continuous waveform sequence; performing cubic spline interpolation on all peaks in the continuous waveform sequence to generate an upper envelope; performing cubic spline interpolation on all troughs to generate a lower envelope; calculating the difference between the upper and lower envelopes at each sampling time to obtain an instantaneous pulse amplitude sequence, which serves as the time-domain morphological parameters. Based on the time-domain morphological parameters, a light attenuation rising branch curve segment is extracted in each pulsation cycle. The rising branch curve segment is then matched with the stored standard rising branch curve segment to calculate the cumulative value of shape difference. Based on the cumulative value of the shape difference, identify waveform distortion signal segments in the blood oxygen signal, and record the duration and frequency of occurrence of the waveform distortion signal segments; Based on the duration and frequency of occurrence, a fluctuation mutation coefficient is calculated. When the fluctuation mutation coefficient exceeds a preset threshold, a blood oxygen abnormality warning signal is triggered.

2. The intelligent early warning method for abnormal blood oxygen fluctuations based on a smartwatch according to claim 1, characterized in that, Based on the time-domain morphological parameters, a light attenuation rising branch curve segment is extracted within each pulsation cycle. This rising branch curve segment is then waveform-matched with a stored standard rising branch curve segment, and the cumulative shape difference value is calculated. Specifically, this includes: In the instantaneous pulse amplitude sequence, all sampling points corresponding to the rise from the trough amplitude to the peak amplitude within each pulsation cycle are determined and connected in chronological order to form the rising branch curve segment of the optical attenuation; Retrieve a pre-established standard ascending branch curve segment from the storage unit. The standard ascending branch curve segment is generated by averaging the ascending branch sampling points within a historical normal pulsation cycle. Align the optical attenuation rising branch curve segment with the standard rising branch curve segment to the same starting point of the time axis, calculate the absolute value of the amplitude difference point by point, and obtain the instantaneous difference sequence. The cumulative value of the shape difference is obtained by summing the absolute values ​​of the amplitude differences of all sampling points in the instantaneous difference sequence.

3. The intelligent early warning method for abnormal blood oxygen fluctuations based on a smartwatch according to claim 2, characterized in that, Based on the cumulative value of the shape difference, waveform distortion signal segments in the blood oxygenation signal are identified, and the duration and frequency of occurrence of the waveform distortion signal segments are recorded, specifically including: The cumulative value of the shape difference corresponding to each pulsation cycle is compared with a first distortion threshold. When the cumulative value of the shape difference is greater than the first distortion threshold, the pulsation cycle is marked as a distortion cycle. Continuously monitor multiple pulsation cycles, and merge multiple pulsation cycles that are continuously marked as the distortion cycle into one waveform distortion signal segment; Record the start and end timestamps of the waveform distortion signal segment, calculate the duration, and count the number of times the waveform distortion signal segment appears per unit time, which is taken as the occurrence frequency.

4. The intelligent early warning method for abnormal blood oxygen fluctuations based on a smartwatch according to claim 3, characterized in that, Based on the duration and frequency of occurrence, a fluctuation mutation coefficient is calculated. When the fluctuation mutation coefficient exceeds a preset threshold, a blood oxygen abnormality warning signal is triggered, specifically including: The duration and frequency of the waveform distortion signal segment are multiplied by the corresponding first weighting coefficient and second weighting coefficient, respectively, and the products are added together to obtain the weighted cumulative value. Obtain the historical weighted cumulative value of multiple historical pulsation cycles within the previous time window, and perform an arithmetic average on the historical weighted cumulative value to obtain the historical mean; Divide the weighted cumulative value by the historical mean to obtain the ratio, and use the ratio as the fluctuation mutation coefficient; When the fluctuation mutation coefficient is greater than the mutation determination threshold, the blood oxygen abnormality warning signal is triggered.

5. The intelligent early warning method for abnormal blood oxygen fluctuations based on a smartwatch according to claim 1, characterized in that, When detecting the time interval between two adjacent troughs, if the time interval exceeds the preset normal pulsation interval range, the current pulsation cycle is discarded and the detection is repeated.

6. The intelligent early warning method for abnormal blood oxygen fluctuations based on a smartwatch according to claim 2, characterized in that, The standard rising branch curve segment is stored after being normalized according to the pulsation period duration. The optical attenuation rising branch curve segment is then normalized to the same duration before alignment and matching.

7. The intelligent early warning method for abnormal blood oxygen fluctuations based on a smartwatch according to claim 3, characterized in that, The first distortion threshold is dynamically set based on the mean and standard deviation of the cumulative value of the shape difference of the current user's historical normal pulsation cycle.

8. The intelligent early warning method for abnormal blood oxygen fluctuations based on a smartwatch according to claim 3, characterized in that, When multiple pulsation cycles consecutively marked as the distortion cycle are combined into one waveform distortion signal segment, if the number of normal cycles between two adjacent distortion cycles is less than a preset interval threshold, then the two adjacent distortion cycles are combined into the same waveform distortion signal segment.

9. A method for intelligent early warning of abnormal blood oxygen fluctuations based on a smartwatch according to claim 4, characterized in that, The length of the previous time window is 30 pulsation cycles, and the historical weighted cumulative value is arithmetically averaged after removing outliers that deviate from the mean by more than 3 times the standard deviation.

Images