A PPG signal enhancement method and system based on multi-stage processing and multi-channel fusion

Abstract

This invention provides a PPG signal enhancement method and system based on multi-level processing and multi-channel fusion. It acquires multi-channel raw PPG signals from a PPG sensor and synchronously acquired inertial measurement unit (IMU) data. For each channel's raw PPG signal, a three-level processing pipeline—real-time, mid-term, and long-term—is executed in parallel. The signal quality index (SQI) of each channel's PPG signal after the three-level processing is calculated in real-time. Dynamic weighted fusion is performed to output a high-quality fused PPG signal, dynamically optimizing key parameters in the three-level collaborative processing pipeline. This invention, based on a processing pipeline, a multi-channel dynamic weighted fusion mechanism based on real-time quality assessment, and a closed-loop dynamic parameter adjustment strategy, can reconstruct high-signal-to-noise ratio and high-fidelity high-quality PPG waveforms, significantly improving the feasibility of long-term, continuous, and stable monitoring in harsh operating environments.

Description

Technical Field

[0001] This invention belongs to the field of signal processing technology, and specifically addresses the cascaded processing and multi-source fusion enhancement method for weak signals with severe low-frequency noise and motion artifacts. It is applied to scenarios such as physiological signal preprocessing in wearable devices, and relates to a PPG signal enhancement method and system based on multi-level processing and multi-channel fusion. Background Technology

[0002] Photoplethysmography (PPG) is a physiological signal that non-invasively detects changes in blood volume in subcutaneous microvessels using optical methods. Key features such as heart rate (HR) and heart rate variability (HRV) can be extracted from PPG, making it an important objective indicator reflecting autonomic nervous system activity and assessing stress and emotional states. In recent years, wearable continuous health monitoring based on PPG, particularly for monitoring stress and response, has become a research hotspot.

[0003] In practical applications, especially in the harsh working environments mentioned above, the quality of PPG signals acquired through single-point wrist acquisition faces severe challenges, seriously limiting the reliability of subsequent analysis: Complex and continuous environmental interference: Equipment vibration, electromagnetic radiation, and variable lighting conditions (such as instrument indicator lights and screen light) in the workspace can introduce broadband environmental noise and light leakage interference. High-intensity and highly heterogeneous motion artifacts: The limb movements generated by workers during operations, emergency response, and daily activities have varying intensity, patterns, and frequencies, causing nonlinear and non-stationary distortions in the PPG signal, which are difficult to effectively address using traditional filtering methods.

[0004] Signal degradation under long-term monitoring: Long-term continuous wear can cause changes in the pressure of the sensor in contact with the skin, sweating, and skin temperature drift, resulting in a slow drift of the signal baseline and time-varying decay of the signal-to-noise ratio (SNR).

[0005] The inherent vulnerability of single-point acquisition: The wrist single-channel PPG signal is susceptible to local factors such as changes in local blood perfusion and instantaneous sensor displacement. Single-point failure will lead to monitoring interruption.

[0006] Existing technologies have proposed various solutions to address the PPG signal quality problem, but they still have significant limitations: Regarding motion artifact cancellation: most schemes use accelerometers as reference signals, combined with adaptive filtering (such as LMS and NLMS algorithms) for noise cancellation. However, these methods mainly rely on the assumption of a linear correlation between motion and PPG interference, and their effectiveness is limited in handling complex nonlinear coupled interference. Other schemes attempt to introduce multimodal signals such as electromyography (EMG) and extract relevant components through canonical correlation analysis (CCA) for artifact cancellation, but this increases the system's complexity and power consumption.

[0007] Regarding multi-signal utilization: Existing patents propose using multiple PPG sensors to collect signals and combine them to obtain better multi-sensor PPG signals. However, such methods mostly employ simple signal averaging or selection strategies, failing to dynamically and intelligently fuse signals based on the real-time quality of each channel. When some channels are subject to strong interference, the overall output signal quality will be reduced.

[0008] In terms of processing architecture: existing methods mostly focus on processing at a single time scale (such as real-time filtering), lacking a collaborative processing framework for transient noise, mid-term drift, and long-term trends. At the same time, processing parameters are usually fixed or manually preset, and cannot adapt to different individuals, different activity states, and changing environments.

[0009] Therefore, how to provide a PPG signal enhancement method and system based on multi-level processing and multi-channel fusion to overcome the shortcomings of existing technologies, and to reconstruct high-signal-to-noise ratio and high-fidelity high-quality PPG waveforms from multi-channel wrist PPG signals subject to complex interference, thereby significantly improving the feasibility of long-term, continuous and stable monitoring in harsh operating environments, has become an urgent technical problem to be solved. Summary of the Invention

[0010] This invention provides a PPG signal enhancement method and system based on multi-level processing and multi-channel fusion. Based on a three-level collaborative processing pipeline, a multi-channel dynamic weighted fusion mechanism based on real-time quality assessment, and a closed-loop dynamic parameter adjustment strategy, it can reconstruct high-signal-to-noise ratio and high-fidelity high-quality PPG waveforms from multi-channel wrist PPG signals subjected to complex interference, significantly improving the feasibility of long-term, continuous, and stable monitoring in harsh operating environments.

[0011] In one embodiment of the present invention, a PPG signal enhancement method based on multi-level processing and multi-channel fusion is provided, applied to a data processing device, comprising: S101, Synchronous signal acquisition: Acquire multi-channel raw PPG signals from the PPG sensor on the wrist wearable device, as well as synchronously acquired inertial measurement unit data; S102, Three-level collaborative processing: For the raw PPG signal of each channel, the following three-level processing is performed in parallel to form a processing pipeline: S21, Real-time processing stage: Adaptive bandpass filtering is performed on the raw PPG signal to suppress high-frequency environmental noise and some baseline drift; at the same time, based on the inertial measurement unit data, a pattern recognition algorithm is used to detect motion artifact events and their intensity levels in real time. S22, Intermediate Processing Stage: For the motion artifact periods detected in S21, nonlinear adaptive filtering is initiated, and noise cancellation is performed using the inertial measurement unit data and / or preprocessed electromyographic signals as reference inputs; for non-intense motion periods, wavelet transform is used for multi-resolution analysis to separate and correct low-frequency baseline drift caused by sensor micro-movements and slow body position changes. S23, Long-term processing level: For the signal processed by S22, trend analysis and stripping are performed in minutes to hours to remove ultra-low frequency trend noise caused by slow changes in skin physiological state and long-term wear effect of sensor, and stabilize the signal baseline. S103, Channel Quality Assessment and Dynamic Fusion: The Signal Quality Index (SQI) of each channel after three-level processing is calculated in real time. The SQI is obtained by comprehensively calculating at least two of the following features: signal-to-noise ratio, waveform period consistency based on the coefficient of variation of adjacent pulse intervals, significance of pulse peak points, and purity of signal power spectrum within a preset physiological frequency band. Dynamic weighted fusion is performed based on the SQI values ​​of each channel. The fusion weight _i of the i-th channel is calculated as: weight _i = (SQI_i)^k / Σ(SQI_j)^k, where k is an adjustable gain factor greater than 1, used to enhance the contribution of high-quality channels, and finally outputting a fused high-quality PPG signal. S104. Closed-loop parameter optimization: Based on the real-time quality assessment results of the fused high-quality PPG signal, the SQI statistical characteristics of each channel, and the motion artifact detection results, the key parameters in the three-stage collaborative processing pipeline are dynamically optimized, including: the cutoff frequency of the adaptive bandpass filter in S21, the step size and structural parameters of the nonlinear adaptive filter in S22, and the window length of the trend term analysis in S23.

[0012] Furthermore, the nonlinear adaptive filtering uses the inertial measurement data and / or preprocessed electromyographic signals as reference inputs, and processes them using a neural network-based filter or a kernel adaptive filter.

[0013] Furthermore, the trend analysis and stripping specifically involves removing ultra-low frequency trend components from the signal using a sliding window polynomial fitting or empirical mode decomposition method.

[0014] Furthermore, the signal quality index (SQI) is obtained by comprehensively calculating at least two of the following characteristics: signal-to-noise ratio, periodic consistency based on the coefficient of variation of adjacent pulse wave intervals, significance measure of pulse wave peak points, and purity of signal power spectrum within a preset physiological frequency band.

[0015] Furthermore, the processing parameters for dynamic optimization include: the cutoff frequency of the adaptive bandpass filter in step S21, the step size parameter or network structure parameter of the nonlinear adaptive filter in step S22, and the window length of the trend term analysis in step S23.

[0016] Furthermore, the adjustable gain factor k ranges from 2 to 3.

[0017] Furthermore, the initial cutoff frequency of the adaptive bandpass filter is from 0.5 Hz to 8 Hz, and it can be dynamically fine-tuned according to the estimated heart rate.

[0018] In another embodiment of the present invention, a PPG signal enhancement system based on multi-level processing and multi-channel fusion is provided, which is based on the PPG signal enhancement method based on multi-level processing and multi-channel fusion described in any one of the above embodiments and deployed in a data processing device. The system includes: The signal acquisition interface module is used to acquire multi-channel raw PPG signals and inertial measurement unit data from the wrist-worn wearable device. A three-level collaborative processing engine is connected to the signal acquisition interface module, and the engine includes: The real-time processing unit is used to perform adaptive bandpass filtering on the raw PPG signal of each channel and to detect motion artifact events and intensity levels in real time based on the data from the inertial measurement unit. An intermediate processing unit, connected to the real-time processing unit, is used to initiate nonlinear adaptive filtering for noise cancellation during motion artifact periods identified by the real-time processing unit, and / or perform wavelet transform on non-violent motion periods to correct low-frequency baseline drift. A long-term processing unit, connected to the intermediate-term processing unit, is used to perform trend analysis and stripping of the signal in windows ranging from minutes to hours. The signal quality assessment and fusion module is connected to the three-level collaborative processing engine. It is used to calculate the signal quality index (SQI) of the PPG signal after processing each channel in real time, and to perform dynamic weighted fusion based on the SQI to output a high-quality PPG signal. The fusion weight of the i-th channel is proportional to (SQI_i)^k, where k is an adjustable gain factor greater than 1. The closed-loop parameter optimization module is connected to the signal quality assessment and fusion module and the three-level collaborative processing engine. It is used to dynamically adjust the processing parameters of each unit in the three-level collaborative processing engine according to the output signal quality, the SQI statistical characteristics of each channel and the motion detection results.

[0019] Furthermore, the intermediate processing unit includes a first processing branch and a second processing branch; the first processing branch is configured to be activated when the real-time processing unit detects that the motion intensity exceeds a preset threshold, and to perform nonlinear adaptive filtering with reference to inertial measurement unit data; the second processing branch is configured to be activated when the motion intensity is lower than a preset threshold, and to perform wavelet transform for baseline correction.

[0020] In another embodiment of the present invention, a data processing device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements a PPG signal enhancement method based on multi-level processing and multi-channel fusion as described in any of the preceding embodiments.

[0021] In another embodiment of the present invention, a non-intrusive emotion monitoring system is provided, based on the PPG signal enhancement method based on multi-level processing and multi-channel fusion described in any of the above claims, the system comprising: The smart wristband integrates a PPG sensor and an inertial measurement unit for acquiring multi-channel raw PPG signals and inertial measurement data. A PPG signal enhancement system based on multi-level processing and multi-channel fusion is used to process the original signal and output a high-quality PPG signal. The emotion state analysis module is used to extract heart rate variability features from the high-quality PPG signal and output the corresponding emotion or stress state classification results based on the pre-trained physiological-psychological state mapping model.

[0022] In another embodiment of the present invention, a computer-readable storage medium is provided, on which a computer program is stored, wherein the computer program, when executed by a processor, implements the PPG signal enhancement method based on multi-level processing and multi-channel fusion as described above.

[0023] The beneficial effects of this invention are as follows: As can be seen from the above scheme, the embodiments of the present invention provide a PPG signal enhancement method and system based on multi-level processing and multi-channel fusion. This method acquires multi-channel raw PPG signals collected by a PPG sensor, as well as synchronously acquired inertial measurement unit (IMU) data. For each channel's raw PPG signal, a three-level processing stage (real-time, mid-term, and long-term) is performed in parallel to form a processing pipeline. The signal quality index (SQI) of each channel's PPG signal after the three-level processing is calculated in real time. Dynamic weighted fusion is performed to output a high-quality fused PPG signal, dynamically optimizing key parameters in the three-level collaborative processing pipeline. The technical solution of the present invention, based on the processing pipeline, a multi-channel dynamic weighted fusion mechanism based on real-time quality assessment, and a closed-loop dynamic parameter adjustment strategy, can reconstruct high-signal-to-noise ratio, high-fidelity high-quality PPG waveforms from multi-channel wrist PPG signals subjected to complex interference, significantly improving the feasibility of long-term, continuous, and stable monitoring in harsh operating environments. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the architecture of a PPG signal enhancement system based on multi-level processing and multi-channel fusion according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating a PPG signal enhancement method based on multi-level processing and multi-channel fusion provided in an embodiment of the present invention. Figure 3 This is a timing comparison diagram of the processing effect of a three-level collaborative processing pipeline on a single-channel PPG signal provided by an embodiment of the present invention; Figure 4 This is a flowchart illustrating the system verification method in an embodiment of the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0026] This invention provides a PPG signal enhancement method, system, and data processing device based on multi-level processing and multi-channel fusion. This method, through an innovative three-level collaborative processing pipeline, a multi-channel dynamic weighted fusion mechanism based on real-time quality assessment, and a closed-loop dynamic parameter adjustment strategy, aims to reconstruct high-signal-to-noise ratio, high-fidelity high-quality PPG waveforms from multi-channel wrist PPG signals subjected to complex interference, significantly improving the feasibility of long-term, continuous, and stable monitoring in harsh operating environments.

[0027] like Figures 1 to 4 As shown, Figure 1 This is a schematic diagram of the architecture of a PPG signal enhancement system based on multi-level processing and multi-channel fusion according to an embodiment of the present invention. Figure 2 This is a flowchart illustrating a PPG signal enhancement method based on multi-level processing and multi-channel fusion provided in an embodiment of the present invention. Figure 3 This is a timing comparison diagram of the processing effect of a three-level collaborative processing pipeline on a single-channel PPG signal, provided by an embodiment of the present invention. Figure 4 This is a flowchart illustrating the system verification method in an embodiment of the present invention.

[0028] Figure 1 and Figure 2 A PPG signal enhancement method based on multi-level processing and multi-channel fusion is proposed and applied to a data processing device, comprising: S101, Synchronous signal acquisition: Acquire multi-channel raw PPG signals from the PPG sensor on the wrist wearable device, as well as synchronously acquired inertial measurement unit data; S102, Three-level collaborative processing: For the raw PPG signal of each channel, the following three-level processing is performed in parallel to form a processing pipeline: S21, Real-time processing stage: Adaptive bandpass filtering is performed on the raw PPG signal to suppress high-frequency environmental noise and some baseline drift; at the same time, based on the inertial measurement unit data, a pattern recognition algorithm is used to detect motion artifact events and their intensity levels in real time. S22, Intermediate Processing Stage: For the motion artifact periods detected in S21, nonlinear adaptive filtering is initiated, and noise cancellation is performed using the inertial measurement unit data and / or preprocessed electromyographic signals as reference inputs; for non-intense motion periods, wavelet transform is used for multi-resolution analysis to separate and correct low-frequency baseline drift caused by sensor micro-movements and slow body position changes. S23, Long-term processing level: For the signal processed by S22, trend analysis and stripping are performed in minutes to hours to remove ultra-low frequency trend noise caused by slow changes in skin physiological state and long-term wear effect of sensor, and stabilize the signal baseline. S103, Channel Quality Assessment and Dynamic Fusion: The Signal Quality Index (SQI) of each channel after three-level processing is calculated in real time. The SQI is obtained by comprehensively calculating at least two of the following features: signal-to-noise ratio, waveform period consistency based on the coefficient of variation of adjacent pulse intervals, significance of pulse peak points, and purity of signal power spectrum within a preset physiological frequency band. Dynamic weighted fusion is performed based on the SQI values ​​of each channel. The fusion weight _i of the i-th channel is calculated as: weight _i = (SQI_i)^k / Σ(SQI_j)^k, where k is an adjustable gain factor greater than 1, used to enhance the contribution of high-quality channels, and finally outputting a fused high-quality PPG signal. S104. Closed-loop parameter optimization: Based on the real-time quality assessment results of the fused high-quality PPG signal, the SQI statistical characteristics of each channel, and the motion artifact detection results, the key parameters in the three-stage collaborative processing pipeline are dynamically optimized, including: the cutoff frequency of the adaptive bandpass filter in S21, the step size and structural parameters of the nonlinear adaptive filter in S22, and the window length of the trend term analysis in S23.

[0029] In this embodiment of the invention, a PPG signal enhancement method based on multi-level processing and multi-channel fusion is disclosed. This method acquires multi-channel raw PPG signals from a PPG sensor on a wrist-worn wearable device, using either one sensor with four channels, one sensor with eight channels, or multiple sensors. Specifically, for the long-term processing stage, the signal processed by S22 is subjected to trend analysis and stripping within a 5-minute time window.

[0030] In this embodiment of the invention, the following features are present: Significantly improved processing efficiency and robustness: The proposed three-level collaborative processing architecture of "real-time-medium-long-term" systematically covers all scale interferences from millisecond-level transient noise to hour-level trend drift, and has stronger environmental adaptability and signal recovery capability than single processing methods.

[0031] Intelligent fusion strategy: Based on the dynamic weighted fusion mechanism of real-time SQI, it can automatically select and trust the highest quality signal channel, effectively resist single-point local failure or instantaneous strong interference, and ensure the continuous stability and reliability of the system output, which is better than simple signal averaging or selection strategies.

[0032] Strong adaptability: The closed-loop parameter optimization mechanism enables the system to have self-learning and adaptive capabilities, and can automatically adjust according to the physiological characteristics of different users, different activity patterns and changing environmental noise levels, so as to achieve personalized optimal processing and improve the universality of the method.

[0033] This lays a solid foundation for advanced applications: the high-fidelity PPG output significantly improves the accuracy of subsequent extraction of heart rate and HRV time-frequency domain features (such as SDNN and LF / HF ratio). This provides an unprecedented high-quality data source for advanced analysis modules such as real-time stress emotional state grading and cognitive load assessment based on physiological signals, which is particularly beneficial for building reliable personnel status monitoring and early warning systems in special environments such as long-term isolation and confinement.

[0034] In another embodiment of the present invention, the nonlinear adaptive filtering uses the inertial measurement data and / or the preprocessed electromyographic signal as reference input, and is processed using a neural network-based filter or a kernel adaptive filter.

[0035] In another embodiment of the present invention, the trend term analysis and stripping specifically involves: using a sliding window polynomial fitting or empirical mode decomposition method to remove ultra-low frequency trend components from the signal.

[0036] In another embodiment of the present invention, the signal quality index (SQI) is obtained by comprehensively calculating at least two of the following characteristics: signal-to-noise ratio, periodic consistency based on the coefficient of variation of adjacent pulse wave intervals, significance measure of pulse wave peak points, and purity of signal power spectrum within a preset physiological frequency band.

[0037] In another embodiment of the present invention, the processing parameters for dynamic optimization include: the cutoff frequency of the adaptive bandpass filter in step S21, the step size parameter or network structure parameter of the nonlinear adaptive filter in step S22, and the window length of the trend term analysis in step S23.

[0038] In another embodiment of the present invention, the adjustable gain factor k ranges from 2 to 3.

[0039] In another embodiment of the present invention, the initial cutoff frequency of the adaptive bandpass filter is 0.5 Hz to 8 Hz, and can be dynamically fine-tuned according to the estimated heart rate.

[0040] Figure 3 middle, Figure 3 This is a timing comparison diagram of the processing effect of a three-level collaborative processing pipeline on a single-channel PPG signal provided in an embodiment of the present invention.

[0041] In this embodiment of the invention, a PPG signal enhancement method based on multi-level processing and multi-channel fusion is applied to a prototype system for monitoring the status of personnel in long-term isolated and confined environments. This embodiment constructs a prototype system integrating the method of this invention to simulate continuous monitoring of the physiological status of workers in long-term isolated and confined environments.

[0042] Hardware configuration: A custom-designed smart wristband is used, which integrates a PPG acquisition module with three independent LED-photodiode pairs (located at the radial artery, ulnar artery, and back of the hand, respectively) and a nine-axis IMU. The wristband transmits raw data to an edge computing gateway (data processing device) via Bluetooth Low Energy.

[0043] Software implementation: Deploy the algorithm of this invention on the edge computing gateway.

[0044] Real-time processing stage: A Butterworth bandpass filter with a cutoff frequency of [0.5, 8] Hz is used, and the parameters can be dynamically fine-tuned according to the estimated heart rate. Motion detection uses a lightweight convolutional neural network model based on the variance of the IMU signal and frequency domain features.

[0045] Mid-processing stage: For time periods detected as "high-intensity motion," a pre-trained two-layer LSTM network is used as a nonlinear adaptive filter, taking the acceleration signal as input, to predict and eliminate motion artifacts in the PPG. For other time periods, a 5-level decomposition using the Db4 wavelet is employed, and the approximation coefficients are thresholded to correct the baseline.

[0046] Long-term processing stage: Empirical mode decomposition (EMD) is performed on the signal every 5 minutes, removing the first two intrinsic mode functions (IMF) to eliminate ultra-low frequency trends.

[0047] Fusion module: Calculates the SQI of each channel once per second, where waveform period consistency is evaluated using the coefficient of variation of adjacent pulse intervals. Set k=2 for weighted fusion.

[0048] Parameter optimization: Every 10 minutes, parameters at each level are adjusted by a rule engine based on the average SQI and motion event frequency of the fused signal over the past period. For example, the bandwidth of the bandpass filter is tightened during quiet periods to further suppress noise.

[0049] Verification Method: To objectively evaluate the effectiveness of this system, a 30-day simulated environment verification scheme was designed and implemented. The process is as follows: Figure 4As shown in the figure. This scheme aims to quantitatively compare the system of this invention with the traditional single-channel adaptive filtering method. In a controlled simulation chamber, subjects simultaneously wore the experimental wristband of this system and the control wristband of the traditional method, and performed two standard tasks: (a) routine instrument operation and (b) sudden alarm stress. Evaluation indicators included: signal-to-noise ratio (SNR) of output PPG, heart rate detection error rate with medical-grade ECG as a reference, system robustness under simulated sensor occlusion, and the correlation (Pearson correlation coefficient) between heart rate variability (HRV) features extracted from PPG and the subjects' daily psychological stress scale (PSS) scores. Statistical analysis showed that in routine exercise and sudden stress task scenarios, the average SNR of the output PPG of the system of this invention was improved by more than 15 dB compared with the traditional method, and the heart rate detection error rate was reduced by more than 70% (p<0.01). The fusion mechanism effectively avoided signal interruption caused by single-point occlusion. Based on the HRV features extracted from the high-quality PPG signals of this system, the average correlation coefficient between the HRV features and the PSS scores is consistently above 0.85, which is significantly higher than that of traditional methods, demonstrating its ability to provide reliable data for tracing physiological states under harsh conditions.

[0050] Figure 4 middle, Figure 4 This is a flowchart illustrating the system verification method in an embodiment of the present invention.

[0051] In another embodiment of the present invention, an example of dynamic parameter adjustment is provided: After system initialization, during the user's nighttime sleep phase (low movement, stable environment), the closed-loop optimization module detects consistently excellent signal quality and infrequent motion events. At this time, it automatically performs the following adjustments: (a) narrowing the passband of the real-time processing stage's bandpass filter to [0.8, 3.5] Hz to maximize suppression of potential power frequency harmonics and wider-band environmental noise; (b) lowering the sensitivity threshold of the motion detection model to reduce false triggers; and (c) extending the trend analysis window of the long-term processing stage to 15 minutes to more smoothly track physiological rhythm changes during sleep. When the user enters a high-intensity daytime work phase, the system senses frequent motion events and a broadened signal spectrum, and then reverses the above parameters to a more relaxed default state or work optimization state to ensure that effective physiological information can still be captured in dynamic environments.

[0052] In another embodiment of the present invention, a PPG signal enhancement system based on multi-level processing and multi-channel fusion is provided, which is based on the PPG signal enhancement method based on multi-level processing and multi-channel fusion described in any one of the above embodiments and deployed in a data processing device. The system includes: The signal acquisition interface module is used to acquire multi-channel raw PPG signals and inertial measurement unit data from the wrist-worn wearable device. A three-level collaborative processing engine is connected to the signal acquisition interface module, and the engine includes: The real-time processing unit is used to perform adaptive bandpass filtering on the raw PPG signal of each channel and to detect motion artifact events and intensity levels in real time based on the data from the inertial measurement unit. An intermediate processing unit, connected to the real-time processing unit, is used to initiate nonlinear adaptive filtering for noise cancellation during motion artifact periods identified by the real-time processing unit, and / or perform wavelet transform on non-violent motion periods to correct low-frequency baseline drift. A long-term processing unit, connected to the intermediate-term processing unit, is used to perform trend analysis and stripping of the signal in windows ranging from minutes to hours. The signal quality assessment and fusion module is connected to the three-level collaborative processing engine. It is used to calculate the signal quality index (SQI) of the PPG signal after processing each channel in real time, and to perform dynamic weighted fusion based on the SQI to output a high-quality PPG signal. The fusion weight of the i-th channel is proportional to (SQI_i)^k, where k is an adjustable gain factor greater than 1. The closed-loop parameter optimization module is connected to the signal quality assessment and fusion module and the three-level collaborative processing engine. It is used to dynamically adjust the processing parameters of each unit in the three-level collaborative processing engine according to the output signal quality, the SQI statistical characteristics of each channel and the motion detection results.

[0053] In another embodiment of the present invention, the intermediate processing unit includes a first processing branch and a second processing branch; the first processing branch is configured to be activated when the real-time processing unit detects that the motion intensity exceeds a preset threshold, and to perform nonlinear adaptive filtering with reference to inertial measurement unit data; the second processing branch is configured to be activated when the motion intensity is lower than a preset threshold, and to perform wavelet transform for baseline correction.

[0054] In another embodiment of the present invention, a data processing device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements a PPG signal enhancement method based on multi-level processing and multi-channel fusion as described in any of the preceding embodiments.

[0055] In another embodiment of the present invention, a non-intrusive emotion monitoring system is provided, based on the PPG signal enhancement method based on multi-level processing and multi-channel fusion described in any of the above claims, the system comprising: The smart wristband integrates a PPG sensor and an inertial measurement unit for acquiring multi-channel raw PPG signals and inertial measurement data. A PPG signal enhancement system based on multi-level processing and multi-channel fusion is used to process the original signal and output a high-quality PPG signal. The emotion state analysis module is used to extract heart rate variability features from the high-quality PPG signal and output the corresponding emotion or stress state classification results based on the pre-trained physiological-psychological state mapping model.

[0056] In another embodiment of the present invention, a computer-readable storage medium is provided, on which a computer program is stored, wherein the computer program, when executed by a processor, implements the PPG signal enhancement method based on multi-level processing and multi-channel fusion as described above.

[0057] This invention provides a PPG signal enhancement method and system based on multi-level processing and multi-channel fusion. It acquires multi-channel raw PPG signals from a PPG sensor and synchronously acquired inertial measurement unit (IMU) data. For each channel's raw PPG signal, a three-level processing pipeline—real-time, mid-term, and long-term—is executed in parallel. The signal quality index (SQI) of each channel's PPG signal after the three-level processing is calculated in real-time. Dynamic weighted fusion is performed to output a high-quality fused PPG signal, dynamically optimizing key parameters in the three-level collaborative processing pipeline. This invention, based on a processing pipeline, a multi-channel dynamic weighted fusion mechanism based on real-time quality assessment, and a closed-loop dynamic parameter adjustment strategy, can reconstruct high-signal-to-noise ratio, high-fidelity high-quality PPG waveforms from multi-channel wrist PPG signals subjected to complex interference, significantly improving the feasibility of long-term, continuous, and stable monitoring in harsh operating environments. By systematically addressing full-scale interference through a multi-level architecture and enhancing system robustness through intelligent fusion, high-fidelity PPG signals can be stably acquired from wrist-worn wearable devices in high-interference environments such as long-term isolation and closed environments, providing a reliable data foundation for accurate monitoring of emotions and stress states.

[0058] The above are preferred embodiments of the present invention. It should be noted that, for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A PPG signal enhancement method based on multi-level processing and multi-channel fusion, applied to data processing equipment, characterized in that, The method includes: S101, Synchronous signal acquisition: Acquire multi-channel raw PPG signals from the PPG sensor on the wrist wearable device, as well as synchronously acquired inertial measurement unit data; S102, Three-level collaborative processing: For the raw PPG signal of each channel, the following three-level processing is performed in parallel to form a processing pipeline: S21, Real-time processing stage: Adaptive bandpass filtering is performed on the raw PPG signal to suppress high-frequency environmental noise and some baseline drift; at the same time, based on the inertial measurement unit data, a pattern recognition algorithm is used to detect motion artifact events and their intensity levels in real time. S22, Intermediate Processing Stage: For the motion artifact periods detected in S21, nonlinear adaptive filtering is initiated, and noise cancellation is performed using the inertial measurement unit data and / or preprocessed electromyographic signals as reference inputs; for non-intense motion periods, wavelet transform is used for multi-resolution analysis to separate and correct low-frequency baseline drift caused by sensor micro-movements and slow body position changes. S23, Long-term processing level: For the signal processed by S22, trend analysis and stripping are performed in minutes to hours to remove ultra-low frequency trend noise caused by slow changes in skin physiological state and long-term wear effect of sensor, and stabilize the signal baseline. S103, Channel Quality Assessment and Dynamic Fusion: The Signal Quality Index (SQI) of each channel after three-level processing is calculated in real time. The SQI is obtained by comprehensively calculating at least two of the following features: signal-to-noise ratio, waveform period consistency based on the coefficient of variation of adjacent pulse intervals, significance of pulse peak points, and purity of signal power spectrum within a preset physiological frequency band. Dynamic weighted fusion is performed based on the SQI values ​​of each channel. The fusion weight _i of the i-th channel is calculated as: weight _i = (SQI_i)^k / Σ(SQI_j)^k, where k is an adjustable gain factor greater than 1, used to enhance the contribution of high-quality channels, and finally outputting a fused high-quality PPG signal. S104. Closed-loop parameter optimization: Based on the real-time quality assessment results of the fused high-quality PPG signal, the SQI statistical characteristics of each channel, and the motion artifact detection results, the key parameters in the processing pipeline are dynamically optimized, including: the cutoff frequency of the adaptive bandpass filter in S21, the step size and structural parameters of the nonlinear adaptive filter in S22, and the window length of the trend term analysis in S23.

2. The PPG signal enhancement method based on multi-level processing and multi-channel fusion according to claim 1, characterized in that, The nonlinear adaptive filtering uses inertial measurement data and / or preprocessed electromyographic signals as reference inputs, and is processed using neural network-based filters or kernel adaptive filters.

3. The PPG signal enhancement method based on multi-level processing and multi-channel fusion according to claim 1, characterized in that, The trend analysis and stripping specifically involves removing ultra-low frequency trend components from the signal using sliding window polynomial fitting or empirical mode decomposition methods.

4. The PPG signal enhancement method based on multi-level processing and multi-channel fusion according to claim 1, characterized in that, The processing parameters for dynamic optimization include: the cutoff frequency of the adaptive bandpass filter in step S21, the step size parameter or network structure parameter of the nonlinear adaptive filter in step S22, and the window length of the trend term analysis in step S23.

5. The PPG signal enhancement method based on multi-level processing and multi-channel fusion according to claim 1, characterized in that, The adjustable gain factor k ranges from 2 to 3.

6. The PPG signal enhancement method based on multi-level processing and multi-channel fusion according to claim 1, characterized in that, The adaptive bandpass filter has an initial cutoff frequency of 0.5 Hz to 8 Hz and can be dynamically fine-tuned according to the estimated heart rate.

7. A PPG signal enhancement system based on multi-level processing and multi-channel fusion, based on the PPG signal enhancement method based on multi-level processing and multi-channel fusion as described in any one of claims 1 to 7, characterized in that, Deployed on a data processing device, the system includes: The signal acquisition interface module is used to acquire multi-channel raw PPG signals and inertial measurement unit data from the wrist-worn wearable device. A three-level collaborative processing engine is connected to the signal acquisition interface module, and the engine includes: The real-time processing unit is used to perform adaptive bandpass filtering on the raw PPG signal of each channel and to detect motion artifact events and intensity levels in real time based on the data from the inertial measurement unit. An intermediate processing unit, connected to the real-time processing unit, is used to initiate nonlinear adaptive filtering for noise cancellation during motion artifact periods identified by the real-time processing unit, and / or perform wavelet transform on non-violent motion periods to correct low-frequency baseline drift. A long-term processing unit, connected to the intermediate-term processing unit, is used to perform trend analysis and stripping of the signal in windows ranging from minutes to hours. The signal quality assessment and fusion module is connected to the three-level collaborative processing engine. It is used to calculate the signal quality index (SQI) of the PPG signal after processing each channel in real time, and to perform dynamic weighted fusion based on the SQI to output a high-quality PPG signal. The fusion weight of the i-th channel is proportional to (SQI_i)^k, where k is an adjustable gain factor greater than 1. The closed-loop parameter optimization module is connected to the signal quality assessment and fusion module and the three-level collaborative processing engine. It is used to dynamically adjust the processing parameters of each unit in the three-level collaborative processing engine according to the output signal quality, the SQI statistical characteristics of each channel and the motion detection results.

8. A PPG signal enhancement system based on multi-level processing and multi-channel fusion according to claim 7, characterized in that, The intermediate processing unit includes a first processing branch and a second processing branch; the first processing branch is configured to be activated when the real-time processing unit detects that the motion intensity exceeds a preset threshold, and to perform nonlinear adaptive filtering with reference to inertial measurement unit data; the second processing branch is configured to be activated when the motion intensity is lower than a preset threshold, and to perform wavelet transform for baseline correction.

9. A data processing apparatus, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements a PPG signal enhancement method based on multi-level processing and multi-channel fusion as described in any one of claims 1 to 6.

10. A non-intrusive emotion monitoring system, based on the PPG signal enhancement method according to any one of claims 1 to 6, characterized in that, The system includes: The smart wristband integrates a PPG sensor and an inertial measurement unit for acquiring multi-channel raw PPG signals and inertial measurement data. A PPG signal enhancement system based on multi-level processing and multi-channel fusion is used to process the original PPG signal and output a high-quality PPG signal. The emotion state analysis module is used to extract heart rate variability features from the high-quality PPG signal and output the corresponding emotion or stress state classification results based on the pre-trained physiological-psychological state mapping model.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements a PPG signal enhancement method based on multi-level processing and multi-channel fusion as described in any one of claims 1 to 6.

Images