Adaptive filtering-based blood oxygen signal denoising method and system

Abstract

The application discloses a blood oxygen signal denoising method and system based on adaptive filtering, comprising the following steps: synchronously collecting original blood oxygen signals and three-axis acceleration signals through a sensor module, obtaining time-aligned original blood oxygen signal sequences and three-axis acceleration signal sequences; fusing the three-axis acceleration signals into one-dimensional motion artifact reference noise signals for representing the user's limb motion state; based on multi-level energy analysis of the motion artifact reference noise signals, dynamically calculating adaptive step factor sequences through a nonlinear mapping function; constructing a time-varying step adaptive filter, calculating noise components in the original blood oxygen signals by using the motion artifact reference noise signals and the adaptive step factors; subtracting the calculated noise signals from the original blood oxygen signals to reconstruct pure blood oxygen signals free of motion artifacts. The application can realize accurate calculation and efficient suppression of motion artifacts, and significantly improve the purification quality and processing efficiency of blood oxygen signals.

Description

Technical Field

[0001] This invention relates to the technical field of signal denoising, and in particular to a method and system for denoising blood oxygen signals based on adaptive filtering. Background Technology

[0002] With the rapid development of modern medical technology and the continuous improvement of people's health awareness, respiratory monitoring devices are increasingly widely used in clinical medicine, health management, and sports rehabilitation. Wearable blood oxygen saturation monitoring devices, in particular, have become important tools for detecting physiological parameters due to their ability to monitor human blood oxygen levels in real time and continuously. However, in practical applications, the blood oxygen signals collected by respiratory devices are highly susceptible to interference from various motion artifacts, seriously affecting the accuracy and reliability of the monitoring data. This has become a key bottleneck restricting the further development of respiratory monitoring technology.

[0003] Traditional blood oxygenation signal processing methods primarily rely on fixed-parameter digital filtering techniques, such as bandpass filtering and notch filtering. While these methods can provide some noise suppression in static or slightly moving conditions, they often perform poorly when faced with complex and varied motion interference. Because motion artifacts overlap with blood oxygenation signals in the frequency domain, traditional linear filtering methods struggle to effectively separate useful signals from interference components, easily leading to signal distortion or loss of crucial physiological information. More seriously, fixed filtering parameters cannot adapt to dynamic changes in motion, lacking flexibility and adaptability when processing motion interference of varying intensities and patterns.

[0004] In recent years, adaptive filtering methods have improved upon the limitations of traditional methods to some extent by introducing reference signals and adaptive algorithms to achieve dynamic calculation and suppression of motion artifacts. However, existing adaptive filtering techniques generally suffer from slow convergence speed, large steady-state error, and poor adaptability to complex motion patterns. Especially when dealing with sudden motion changes and multimodal motion patterns, existing methods often require a long convergence time to reach a stable state, resulting in poor real-time performance. Furthermore, most existing methods employ fixed step size parameters or simple step size adjustment strategies, lacking a deep perception of motion intensity and trends, making it difficult to achieve truly intelligent adaptation.

[0005] Current mainstream respiratory equipment data processing technologies still suffer from technical shortcomings such as insufficient signal preprocessing, low accuracy in motion state recognition, and inadequate optimization of filter parameters. Many methods ignore the physiological constraints of blood oxygenation signals, potentially introducing non-physiological signal components during denoising, which affects the final monitoring accuracy. Furthermore, most existing technologies lack systematic quality assessment mechanisms, making it impossible to effectively determine the reliability of processed signals, which may pose potential safety risks in clinical applications.

[0006] Faced with the increasing demand for high-precision blood oxygen monitoring and the complex and ever-changing application environments, there is an urgent need to develop a data processing method for respiratory devices that can intelligently sense motion status, adaptively adjust processing parameters, and achieve efficient and accurate noise reduction. This method must not only possess excellent real-time performance and stability, but also maintain reliable processing results under various motion modes, providing accurate and reliable blood oxygen data support for clinical medicine and health monitoring. Summary of the Invention

[0007] In view of this, the present invention provides a blood oxygen signal denoising method based on adaptive filtering. The purpose is to achieve accurate calculation and efficient suppression of motion artifacts by constructing a multi-level energy analysis mechanism and a time-varying step-size adaptive filter, thereby significantly improving the purification quality and processing efficiency of blood oxygen signals and providing technical support for the stable and reliable operation of respiratory monitoring equipment in various motion scenarios.

[0008] To achieve the above objectives, the present invention provides a blood oxygen signal denoising method based on adaptive filtering, comprising the following steps: A1: The raw blood oxygen signal and triaxial acceleration signal are acquired synchronously through the sensor module to obtain the time-aligned raw blood oxygen signal sequence and triaxial acceleration signal sequence; A2: The triaxial acceleration signal sequence is fused into a one-dimensional motion artifact reference noise signal to characterize the user's limb movement state; A3: Multi-level energy analysis based on motion artifact reference noise signal, dynamically calculating adaptive step size factor sequence through nonlinear mapping function; A4: Construct a time-varying step-size adaptive filter, and use the one-dimensional motion artifact reference noise signal and the adaptive step-size factor sequence to calculate the noise component in the original blood oxygen signal; A5: Subtract the computational noise signal from the original blood oxygen signal to reconstruct a pure blood oxygen signal free of motion artifacts.

[0009] As a further improvement of the present invention: Optionally, in step A1, the raw blood oxygenation signal and triaxial acceleration signal are synchronously acquired through a sensor module to obtain a time-aligned raw blood oxygenation signal sequence and triaxial acceleration signal sequence, including: A sensor module data acquisition system is established, using photoelectric sensors to acquire raw blood oxygen signals containing real pulse wave information and noise to be filtered out on the same time axis. Simultaneously, a triaxial accelerometer, physically located adjacent to the photoelectric sensors, synchronously acquires the user's limb movement status. axis, shaft and Acceleration data in three dimensions (axis, sigma, and sigma) are sampled at a uniform frequency to ensure time synchronization between the two signals, thus obtaining the raw blood oxygen signal sequence. and triaxial acceleration signal sequence , , ,in Indicates the sampling time index. , and They represent the sampling time n respectively axis, shaft and Axis acceleration signal value.

[0010] Optionally, in step A2, the triaxial acceleration signals are fused into a one-dimensional motion artifact reference noise signal to characterize the user's limb movement state, including: The triaxial acceleration signal sequence obtained in step A1 is subjected to vector synthesis processing. The magnitude of the vector sum of the three axial acceleration values ​​at each sampling point is calculated, and the three-dimensional motion information is fused into a one-dimensional signal sequence. The motion artifact reference noise signal is calculated as follows: ; in, Indicates the sampling time The motion artifact reference noise signal value.

[0011] Optionally, the multi-level energy analysis based on the motion artifact reference noise signal in step A3, which dynamically calculates the adaptive step size factor sequence through a nonlinear mapping function, includes: An adaptive step size factor dynamic calculation mechanism based on motion intensity is established. First, the time-frequency domain characteristics of the motion artifact reference noise signal are calculated through multi-level energy analysis. Then, by combining motion state prediction and step size stability constraints, an adaptive step size factor sequence that dynamically changes with time is generated. The multi-level energy analysis includes two time scales: short-term energy and medium-term energy. Short-term energy is used to capture instantaneous motion changes, while medium-term energy is used to identify motion pattern trends. The calculation formula is: ; in, Indicates the sampling time Short-term energy value, Indicates the length of the short-term energy analysis window. Indicates the sampling time The motion artifact reference noise signal value; Mid-term energy The calculation formula is: ; in, Indicates the sampling time Mid-term energy value, Indicates the length of the medium-term energy analysis window and ; The comprehensive index of exercise intensity is obtained by weighted fusion of short-term and medium-term energy levels, and the calculation formula is as follows: ; in, Indicates the sampling time The comprehensive index of exercise intensity This represents the short-term energy weighting coefficient. Indicates the medium-term energy weighting coefficient and , Indicates the energy difference sensitivity coefficient. Represents positive numbers that should not be divided by zero; The motion state prediction is achieved by analyzing the changing trends of comprehensive motion intensity indicators, including motion trend factors. Specifically: ; in, Indicates the sampling time The movement trend factor, Indicates the length of the trend analysis window. Indicates a historical moment index. , Sampling times are represented sequentially. , The comprehensive index of exercise intensity; The adaptive step size factor is calculated using an improved hyperbolic tangent nonlinear mapping function. Combining motion intensity and trend prediction information, the calculation formula is as follows: ; in, Indicates the sampling time Adaptive step size factor, Represents the basic step size factor. Indicates the step size adjustment range. Indicates the motion intensity mapping parameters. The parameter represents the trend influence, and tanh represents the hyperbolic tangent function.

[0012] This step significantly improves the filter's response and adaptability to complex motion states by establishing an adaptive step-size factor calculation mechanism that combines multi-level energy analysis with nonlinear mapping. Compared to traditional fixed-step or simple adaptive methods, the core advantage of this step lies in its ability to accurately perceive the instantaneous changes and long-term trends of motion artifacts, thereby achieving more intelligent parameter adjustment.

[0013] This step first employs dual-timescale analysis of short-term and medium-term energy to simultaneously capture both instantaneous impacts and persistent patterns of motion. Short-term energy analysis offers high temporal resolution and is highly sensitive to sudden changes in motion, while medium-term energy analysis provides stable motion background information. The combination of both enables the system to accurately distinguish between different types of motion disturbances. This multi-layered energy analysis architecture effectively avoids the misjudgment problems that can easily arise from single-timescale analysis, and maintains stable performance, especially during rapid switching of motion patterns.

[0014] This step intelligently fuses multiple energy characteristics through the calculation of a comprehensive motion intensity index. It considers not only the absolute magnitude of energy but also introduces an energy difference term to reflect the complexity of the motion. This design allows the algorithm to more accurately quantify motion intensity, providing a reliable basis for subsequent step size adjustments. In particular, the introduction of an energy difference sensitivity coefficient enhances the system's ability to identify irregular motion.

[0015] Optionally, in step A4, a time-varying step-size adaptive filter is constructed, and the noise component in the original blood oxygenation signal is calculated using the motion artifact reference noise signal and the adaptive step-size factor, including: A time-varying step-size adaptive filter based on the least mean square algorithm is constructed. The motion artifact reference noise signal obtained in step A2 is used as the reference input, and the original blood oxygen signal obtained in step A1 is used as the desired signal. The adaptive step-size factor calculated in step A3 is used to dynamically update the filter coefficients. The calculation formula for the computational noise signal output by the time-varying step-size adaptive filter is as follows: ; in, Indicates the sampling time Calculate the noise signal value. Indicates the sampling time The filter coefficient vector, Indicates the sampling time The reference signal vector consists of the current and past The motion artifact reference noise signal values ​​at each sampling time are constituted, i.e. , The superscript indicates the filter order. Indicates vector transpose; The update formula for the coefficient vector of the time-varying step-size adaptive filter is: ; in, Indicates the sampling time The filter coefficient vector, Indicates the sampling time The error signal is calculated using the following formula: .

[0016] This step, by constructing a time-varying step-size adaptive filter, achieves accurate calculation and dynamic tracking of motion artifacts in blood oxygenation signals, demonstrating significant technical advantages compared to traditional fixed-parameter filtering methods. The core innovation of this step lies in directly applying the adaptive step-size factor calculated in the preceding steps to the real-time update process of the filter coefficients, enabling the filter to dynamically adjust its learning rate and convergence characteristics according to changes in motion state.

[0017] This step uses the least mean square algorithm as its basic framework, but achieves a fundamental improvement in algorithm performance by introducing a time-varying step size mechanism. The traditional least mean square algorithm uses a fixed step size parameter, which inherently contradicts the convergence speed and steady-state error. However, this step dynamically adjusts the step size factor, enabling the use of a larger step size to quickly track signal changes during rapid motion, and a smaller step size to ensure computational accuracy during smooth motion, thus simultaneously achieving excellent performance with fast convergence and low steady-state error.

[0018] Optionally, in step A5, the computational noise signal is subtracted from the original blood oxygen signal to reconstruct a clean blood oxygen signal free of motion artifacts, including: The time-varying step-size adaptive filter in step A4 is applied at each sampling time. Stable output error signal This was determined to be the final pure PPG signal. : ; in, Indicates the sampling time The pure PPG signal value.

[0019] This invention also discloses a blood oxygen signal denoising system based on adaptive filtering, comprising: Data acquisition module: Synchronously acquires raw blood oxygenation signals and triaxial acceleration signals through sensor modules to obtain time-aligned raw blood oxygenation signal sequences and triaxial acceleration signal sequences; Signal synthesis module: fuses triaxial acceleration signals into a one-dimensional motion artifact reference noise signal; Energy Analysis Module: Based on multi-level energy analysis of motion artifact reference noise signals, it dynamically calculates the adaptive step size factor sequence through a nonlinear mapping function; Noise Calculation Module: Constructs a time-varying step-size adaptive filter and uses motion artifact reference noise signal and adaptive step-size factor to calculate the noise component in the original blood oxygen signal; Denoising module: Subtracts the calculated noise signal from the original blood oxygen signal to reconstruct a pure blood oxygen signal free of motion artifacts.

[0020] Compared with the prior art, the present invention has at least the following beneficial effects: This invention establishes an adaptive step-size calculation mechanism that combines multi-level energy analysis with nonlinear mapping. Compared to traditional fixed-step-size methods, this mechanism dynamically adjusts filtering parameters based on real-time changes in motion intensity. During intense motion, a larger step size is used to quickly track signal changes, while a smaller step size is used during gentle motion to ensure computational accuracy, thus achieving an optimal balance between convergence speed and steady-state error. Through dual-timescale analysis of short-term and medium-term energy, the system can simultaneously capture both instantaneous impacts and sustained patterns of motion, effectively avoiding the misjudgment problems that can easily arise from single-timescale analysis.

[0021] This invention, through the synergistic effect of a motion trend prediction mechanism and a time-varying step-size adaptive filter, can not only accurately identify the current motion state but also predict future motion trends based on historical data. This gives the filter a forward-looking parameter adjustment capability rather than a passive response. This predictive adjustment mechanism, combined with an improved hyperbolic tangent nonlinear mapping function, ensures the numerical stability and convergence characteristics of the step-size factor, avoiding the risk of algorithm divergence.

[0022] This invention constructs a multi-level quality assessment and physiological constraint mechanism, which not only optimizes the denoising effect from a signal processing perspective but also verifies the rationality of the processing results from a physiological perspective. By establishing a signal correction mechanism based on physiological constraints, it can effectively identify and correct abnormal signal components, ensuring that the output signal conforms to human physiological characteristics. The comprehensive application of multiple quality assessment indicators enables the system to monitor the processing effect in real time and adjust the processing strategy in a timely manner when a quality degradation is detected, ensuring the robustness and adaptability of the algorithm. Attached Figure Description

[0023] Figure 1 This is a flowchart illustrating a blood oxygen signal denoising method based on adaptive filtering according to an embodiment of the present invention. Figure 2 The diagrams show the effects of multilevel energy measurement analysis and adaptive step factor dynamic adjustment: (a) diagram of multilevel energy analysis, and (b) diagram of adaptive step size dynamic adjustment. Figure 3 This is a schematic diagram comparing the noise calculation and signal reconstruction of an adaptive filter. Detailed Implementation

[0024] The present invention will be further described below with reference to the accompanying drawings, but this is not intended to limit the present invention in any way. Any modifications or substitutions made based on the teachings of the present invention shall fall within the protection scope of the present invention.

[0025] Example 1: A blood oxygen signal denoising method based on adaptive filtering, such as Figure 1 As shown, it includes the following steps: A1: The raw blood oxygenation signal and triaxial acceleration signal are acquired synchronously through the sensor module to obtain a time-aligned raw blood oxygenation signal sequence and triaxial acceleration signal sequence. A sensor module data acquisition system is established, using photoelectric sensors to acquire raw blood oxygen signals containing real pulse wave information and noise to be filtered out on the same time axis. Simultaneously, a triaxial accelerometer, physically adjacent to the photoelectric sensors, synchronously acquires the user's limb movement status. axis, shaft and Acceleration data in three dimensions (axis, sigma, and sigma) are sampled at a uniform frequency to ensure time synchronization between the two signals, thus obtaining the raw blood oxygen signal sequence. and triaxial acceleration signal sequence , , ,in Indicates the sampling time index. , and They represent the sampling time n respectively axis, shaft and Axis acceleration signal value. In this embodiment, the sampling frequency is set to 125Hz, and the sensor module achieves synchronous data acquisition through the SPI communication interface, with a timestamp accuracy of 8ms.

[0026] Optionally, in step A2, the triaxial acceleration signal sequence is fused into a one-dimensional motion artifact reference noise signal to characterize the user's limb movement state, including: The triaxial acceleration signal sequence obtained in step A1 is subjected to vector synthesis processing. The magnitude of the vector sum of the three axial acceleration values ​​at each sampling point is calculated, and the three-dimensional motion information is fused into a one-dimensional signal sequence. The motion artifact reference noise signal is calculated as follows: ; in, Indicates the sampling time The motion artifact reference noise signal value; as an alternative technical solution, a triaxial signal fusion method based on principal component analysis is adopted, which extracts the first principal component containing the main motion information as the motion artifact reference signal by performing principal component decomposition on the triaxial acceleration signal.

[0027] Optionally, for extreme cases where the motion signal is too weak in static or slightly moving states, when the motion artifact reference noise signal R(n) is continuous The number of sampling points is less than the threshold. In this embodiment , A signal enhancement mechanism based on historical motion patterns is enabled. By analyzing users' historical motion data, a personalized motion baseline model is established. The enhanced reference signal is calculated using the following formula: ,in The enhancement factor is set to 0.3. To ensure that the filter can still work effectively under extremely low motion conditions, a motion baseline signal is constructed based on user historical data.

[0028] Optionally, the multi-level energy analysis based on the motion artifact reference noise signal in step A3, which dynamically calculates the adaptive step size factor sequence through a nonlinear mapping function, includes: An adaptive step size factor dynamic calculation mechanism based on motion intensity is established. First, the time-frequency domain characteristics of the motion artifact reference noise signal are calculated through multi-level energy analysis. Then, by combining motion state prediction and step size stability constraints, an adaptive step size factor sequence that dynamically changes with time is generated. The multi-level energy analysis includes two time scales: short-term energy and medium-term energy. Short-term energy is used to capture instantaneous motion changes, while medium-term energy is used to identify motion pattern trends, such as... Figure 2 As shown in (a), short-term energy The calculation formula is: ; in, Indicates the sampling time Short-term energy value, This indicates the length of the short-term energy analysis window. In this embodiment, Set to 16 sampling points, Indicates the sampling time The motion artifact reference noise signal value; Mid-term energy The calculation formula is: ; in, Indicates the sampling time Mid-term energy value, Indicates the length of the medium-term energy analysis window and In this embodiment, Set to 64 sampling points; The comprehensive index of exercise intensity is obtained by weighted fusion of short-term and medium-term energy levels, such as... Figure 2 As shown in (b), the calculation formula is: ; in, Indicates the sampling time The comprehensive index of exercise intensity This represents the short-term energy weighting coefficient. Indicates the medium-term energy weighting coefficient and In this embodiment, and Set them to 0.7 and 0.3 respectively. In this embodiment, the energy difference sensitivity coefficient is represented. Set to 0.5. In this embodiment, the positive constant to avoid division by zero is represented. Set to 1e-6; The motion state prediction is achieved by analyzing the changing trends of comprehensive motion intensity indicators, including motion trend factors. Specifically: ; in, Indicates the sampling time The movement trend factor, Indicates the length of the trend analysis window. In this embodiment, Set to 8 sampling points Indicates a historical moment index. , Sampling times are represented sequentially. , The comprehensive index of exercise intensity; The adaptive step size factor is calculated using an improved hyperbolic tangent nonlinear mapping function. Combining motion intensity and trend prediction information, the calculation formula is as follows: ; in, Indicates the sampling time Adaptive step size factor, This represents the basic step size factor, in this embodiment, Set to 0.01, Indicates the step size adjustment range. In this embodiment, Set to 0.05, In this embodiment, the motion intensity mapping parameters are represented. Set to 2.0. In this embodiment, the parameter representing the influence of the trend is: Set to 0.2, where tanh represents the hyperbolic tangent function.

[0029] Optionally, in step A4, a time-varying step-size adaptive filter is constructed, and the noise component in the original blood oxygenation signal is calculated using the one-dimensional motion artifact reference noise signal and the adaptive step-size factor sequence, including: A time-varying step-size adaptive filter based on the least mean square algorithm is constructed. The motion artifact reference noise signal obtained in step A2 is used as the reference input, and the original blood oxygen signal obtained in step A1 is used as the desired signal. The adaptive step-size factor calculated in step A3 is used to dynamically update the filter coefficients. The calculation formula for the computational noise signal output by the time-varying step-size adaptive filter is as follows: ; in, Indicates the sampling time Calculate the noise signal value. Indicates the sampling time The filter coefficient vector, Indicates the sampling time The reference signal vector consists of the current and past The motion artifact reference noise signal values ​​at each sampling time are constituted, i.e. , The superscript indicates the filter order. Indicates vector transpose; The update formula for the coefficient vector of the time-varying step-size adaptive filter is: ; in, Indicates the sampling time The filter coefficient vector, Indicates the sampling time The error signal is calculated using the following formula: .

[0030] Optionally, in step A5, the computational noise signal is subtracted from the original blood oxygen signal to reconstruct a clean blood oxygen signal free of motion artifacts, including: The time-varying step-size adaptive filter in step A4 is applied at each sampling time. Stable output error signal This was determined to be the final pure PPG signal. ,like Figure 3 As shown: ; in, Indicates the sampling time The pure PPG signal value.

[0031] Example 2: This invention also discloses a blood oxygen signal denoising system based on adaptive filtering, comprising the following five modules: Data acquisition module: Synchronously acquires raw blood oxygenation signals and triaxial acceleration signals through sensor modules to obtain time-aligned raw blood oxygenation signal sequences and triaxial acceleration signal sequences; Signal synthesis module: fuses triaxial acceleration signals into a one-dimensional motion artifact reference noise signal; Energy Analysis Module: Based on multi-level energy analysis of motion artifact reference noise signals, it dynamically calculates the adaptive step size factor sequence through a nonlinear mapping function; Noise Calculation Module: Constructs a time-varying step-size adaptive filter and uses motion artifact reference noise signal and adaptive step-size factor to calculate the noise component in the original blood oxygen signal; Denoising module: Subtracts the calculated noise signal from the original blood oxygen signal to reconstruct a pure blood oxygen signal free of motion artifacts.

[0032] It should be noted that the sequence numbers of the above embodiments of the present invention are merely for descriptive purposes and do not represent the superiority or inferiority of the embodiments. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, apparatus, article, or method. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0033] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0034] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A blood oxygen signal denoising method based on adaptive filtering, characterized in that, Includes the following steps: A1: The raw blood oxygen signal and triaxial acceleration signal are acquired synchronously through the sensor module to obtain the time-aligned raw blood oxygen signal sequence and triaxial acceleration signal sequence; A2: The triaxial acceleration signal sequence is fused into a one-dimensional motion artifact reference noise signal to characterize the user's limb movement state; A3: Multi-level energy analysis is performed based on motion artifact reference noise signal. The adaptive step size factor sequence is dynamically calculated through nonlinear mapping function. A dynamic calculation mechanism for the adaptive step size factor based on motion intensity is established. First, the time-frequency domain characteristics of motion artifact reference noise signal are calculated through multi-level energy analysis. Then, combined with motion state prediction and step size stability constraint, the adaptive step size factor sequence that changes dynamically with time is generated. The multi-level energy analysis includes two time scales: short-term energy and medium-term energy. Short-term energy is used to capture instantaneous motion changes, while medium-term energy is used to identify motion pattern trends. The adaptive step size factor is calculated by combining the improved hyperbolic tangent nonlinear mapping function with the comprehensive motion intensity index and the motion trend factor. A4: Construct a time-varying step-size adaptive filter, and use the one-dimensional motion artifact reference noise signal and the adaptive step-size factor sequence to calculate the noise component in the original blood oxygen signal; A5: Subtract the calculated noise component from the original blood oxygen signal to reconstruct a pure blood oxygen signal free of motion artifacts.

2. The blood oxygen signal denoising method based on adaptive filtering according to claim 1, characterized in that, Step A1 includes: A sensor module data acquisition system is established, using photoelectric sensors to acquire raw blood oxygen signals containing real pulse wave information and noise to be filtered out on the same time axis. Simultaneously, a triaxial accelerometer, physically adjacent to the photoelectric sensors, synchronously acquires the user's limb movement status. axis, shaft and Acceleration data in three dimensions (axis, sigma, and sigma) are sampled at a uniform frequency to ensure time synchronization between the two signals, thus obtaining the raw blood oxygen signal sequence. and triaxial acceleration signal sequence , , ,in Indicates the sampling time index. , and They represent the sampling time n respectively axis, shaft and Axis acceleration signal value.

3. The blood oxygen signal denoising method based on adaptive filtering according to claim 2, characterized in that, Step A2 includes: The triaxial acceleration signal sequence obtained in step A1 is subjected to vector synthesis processing. The magnitude of the vector sum of the three axial acceleration values ​​at each sampling point is calculated, and the three-dimensional motion information is fused into a one-dimensional signal sequence. The motion artifact reference noise signal is calculated as follows: ； in, Indicates the sampling time The motion artifact reference noise signal value.

4. The blood oxygen signal denoising method based on adaptive filtering according to claim 3, characterized in that, Step A3 includes: Short-term energy The calculation formula is: ； in, Indicates the sampling time Short-term energy value, Indicates the length of the short-term energy analysis window. Indicates the sampling time The motion artifact reference noise signal value; Mid-term energy The calculation formula is: ； in, Indicates the sampling time The mid-term energy value, Indicates the length of the medium-term energy analysis window and ; The comprehensive index of exercise intensity is obtained by weighted fusion of short-term and medium-term energy levels, and the calculation formula is as follows: ； in, Indicates the sampling time The comprehensive index of exercise intensity This represents the short-term energy weighting coefficient. Indicates the medium-term energy weighting coefficient and , Indicates the energy difference sensitivity coefficient; The motion state prediction is achieved by analyzing the changing trends of comprehensive motion intensity indicators, including motion trend factors. The specific calculation formula is as follows: ； in, Indicates the sampling time The movement trend factor, Indicates the length of the trend analysis window. Indicates a historical moment index. , Sampling times are represented sequentially. , The comprehensive index of exercise intensity Represents positive numbers that should not be divided by zero; The formula for calculating the adaptive step size factor is: ； in, Indicates the sampling time Adaptive step size factor, Represents the basic step size factor. Indicates the step size adjustment range. Represents the motion intensity mapping parameters. The parameter represents the trend influence, and tanh represents the hyperbolic tangent function.

5. The blood oxygen signal denoising method based on adaptive filtering according to claim 4, characterized in that, Step A4 includes: A time-varying step-size adaptive filter based on the least mean square algorithm is constructed. The motion artifact reference noise signal obtained in step A2 is used as the reference input, and the original blood oxygen signal obtained in step A1 is used as the desired signal. The adaptive step-size factor calculated in step A3 is used to dynamically update the filter coefficients. The calculation formula for the computational noise signal output by the time-varying step-size adaptive filter is as follows: ； in, Indicates the sampling time Calculate the noise signal value. Indicates the sampling time The filter coefficient vector, Indicates the sampling time The reference signal vector consists of the current and past The motion artifact reference noise signal values ​​at each sampling time are constituted, i.e. , The superscript indicates the filter order. Indicates vector transpose; The update formula for the coefficient vector of the time-varying step-size adaptive filter is: ； in, Indicates the sampling time The filter coefficient vector, Indicates the sampling time The error signal is calculated using the following formula: .

6. The blood oxygen signal denoising method based on adaptive filtering according to claim 5, characterized in that, Step A5 includes: The time-varying step-size adaptive filter in step A4 is applied at each sampling time. Stable output error signal This was determined to be the final pure PPG signal. : ； in, Indicates the sampling time The pure PPG signal value.

7. A blood oxygen signal denoising system based on adaptive filtering, characterized in that, include: Data acquisition module: Synchronously acquires raw blood oxygenation signals and triaxial acceleration signals through sensor modules to obtain time-aligned raw blood oxygenation signal sequences and triaxial acceleration signal sequences; Signal synthesis module: fuses triaxial acceleration signals into a one-dimensional motion artifact reference noise signal; Energy Analysis Module: Performs multi-level energy analysis based on motion artifact reference noise signal, and dynamically calculates adaptive step size factor sequence through nonlinear mapping function; Noise Calculation Module: Constructs a time-varying step-size adaptive filter and uses motion artifact reference noise signal and adaptive step-size factor to calculate the noise component in the original blood oxygen signal; Denoising module: Subtracts the calculated noise component from the original blood oxygen signal to reconstruct a clean blood oxygen signal free of motion artifacts; To achieve the blood oxygen signal denoising method based on adaptive filtering as described in any one of claims 1-6.

Images