PPG signal noise reduction method combining discrete wavelet transform and optimized SG filtering

Abstract

The invention discloses a PPG signal noise reduction method combining discrete wavelet transform and optimized SG filtering. The method comprises the following steps: carrying out signal decomposition on an original PPG signal by adopting DWT, and carrying out soft threshold processing on the decomposed signal by utilizing a soft threshold to obtain a preliminary de-noised signal; the parameter selection of an SG filter is carried out in advance by adopting grid optimization, so that the smoothing processing of the PPG signal is realized; and the final de-noised PPG signal is obtained from the output port of the SG filter.

Description

Technical Field

[0001] The present invention belongs to the technical field of PPG signal noise reduction, and specifically relates to a PPG signal noise reduction method combining discrete wavelet transform and optimized SG filtering. Background Art

[0002] PPG is an important biometric technology designed to capture physiological signals by detecting light reflected or transmitted through the skin's surface. Due to its application in wearable devices, PPG signal detection is subject to noise caused by external factors such as motion and light, which can affect the calculated physiological parameters.

[0003] PPG signal noise reduction involves extracting the useful signal from the noise, reducing interference, after the signal is affected by factors such as motion and the environment. Numerous algorithms are available for noise reduction in PPG signals, including recursive inversion of discrete wavelet transforms, wavelet transforms and decomposition and reconstruction, SG denoising methods, and averaging filtering. While averaging filtering is commonly used, it can oversmooth the signal, resulting in loss of critical original information. A method is needed that can smooth the signal without losing critical information while also reducing noise. Summary of the Invention

[0004] The present invention aims to overcome the shortcomings of existing technologies by proposing a PPG signal denoising method that combines discrete wavelet transform (DWT) and optimized SG filtering. In the initial denoising phase, the wavelet coefficients are decomposed into three layers, using the heuristic threshold sqtwolog. The threshold scaling method uses sln (noise estimation based on the first or first-order wavelet coefficients) to adjust the threshold. Other parameter configurations include the use of the Daubechies 8-point wavelet ('db8'), which is suitable for capturing the multi-scale characteristics of biosignals. During the smoothing phase, grid optimization is added to pre-select the SG parameters. This method effectively achieves smoothing of the signal without losing the original information, building on the first denoising step, thereby improving the accuracy of the PPG signal.

[0005] The technical solution of the present invention is to provide a PPG signal denoising method combining discrete wavelet transform and optimized SG filtering, comprising the following steps:

[0006] Step 1: Use DWT to decompose the original PPG signal, and use soft thresholding to perform soft thresholding on these decomposed signals to obtain a preliminary denoised signal.

[0007] Step 2: Grid optimization is used to select the parameters of the SG filter in advance to achieve smoothing of the PPG signal. The final denoised PPG signal is obtained from the output port of the SG filter. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 is a system block diagram of the present invention;

[0009] Figure 2 This is a discrete wavelet transform flow chart in the present invention;

[0010] Figure 3 This is the optimized SG filtering flow chart of the present invention;

[0011] Figure 4 This is a signal waveform diagram before noise reduction in the present invention;

[0012] Figure 5 This is a waveform diagram of the signal after noise reduction in the present invention;

[0013] Figure 6 This is a spectrum comparison diagram before and after noise reduction in the present invention;

[0014] Figure 7 This is a visualization diagram of the evaluation index after noise reduction in the present invention; DETAILED DESCRIPTION

[0015] The following is a detailed description of an embodiment of the present invention in conjunction with the accompanying drawings, but the present invention is not limited to this embodiment. Based on the embodiment of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0016] The present invention conducted simulation experiments in the MATLAB environment; the data set used the wrist pulse wave database (Wrist PPG dataset) from the Massachusetts Institute of Technology Arrhythmia Database (MIT-BIH); these signals were collected during exercise at a sampling rate of 256 data points per second, recording approximately 10 minutes of data, and the database contains raw PPG signals. The present invention selected two exercise states, walking and running, for a total of 11 test subjects, and selected the raw PPG signal of walking test subject 3 for verification.

[0017] like Figure 1 As shown, the PPG signal denoising method of the present invention combined with discrete wavelet transform and optimized SG filtering includes the following steps:

[0018] Step 1: After signal preprocessing, discrete wavelet transform is performed to perform preliminary noise reduction;

[0019] Step 2: The signal after preliminary noise reduction is grid optimized, and the SG filter uses the optimized parameters to perform a second filtering to obtain the final noise-reduced signal.

[0020] like Figure 2As shown in the figure, discrete wavelet transform is used for preliminary noise reduction. The specific process is as follows:

[0021] Step 1: After signal preprocessing, the wavelet function db8 is selected for decomposition of the signal, and the scaling parameter sln is selected. The signal is decomposed through low-pass and high-pass filters to obtain approximate coefficients (representing the low-frequency part of the signal) and detail coefficients (representing the high-frequency part of the signal) respectively;

[0022] Step 2: Downsample the decomposed signal each time and decompose the approximate coefficients therein to the next level until the required number of decomposition layers is reached or specific analysis requirements are met. The number of decomposition layers in the present invention is three, and three cycles are performed during the decomposition;

[0023] Step 3: Based on the set threshold, the soft threshold heuristic threshold sqtwolog is used for processing. The processing formula is:

[0024]

[0025]

[0026] λ is the threshold parameter; x is the unprocessed wavelet coefficient; x λ are the processed wavelet coefficients.

[0027] Step 4: Perform inverse downsampling on the decomposed signal and reconstruct the signal.

[0028] like Figure 3 As shown, the optimized SG filter

[0029] Step 1: Determine the optimization target as the SG parameter, define the upper and lower bounds of the parameter, initialize it, evaluate the optimized parameters, and select the optimal parameters;

[0030] Step 2: According to the given optimal parameters, the top window size and order are removed, and a fitting matrix is ​​constructed to fit the data. The fitting formula used is as follows:

[0031]

[0032] The polynomial coefficients are determined based on the residuals. The residual formula is as follows:

[0033]

[0034]

[0035]

[0036] In the above formula, the partial derivative is calculated to determine the minimum value of the residual;

[0037] Write the above into matrix form, solve it, and bring it into the fitting equation to get the fitting curve. The specific matrix form is as follows:

[0038] X·A=Y

[0039] The final least squares solution equation is as follows:

[0040]

[0041] Step 3: Move the window and repeat the data fitting calculation to obtain the final curve and perform boundary processing.

[0042] like Figure 4 As shown in Figure 1, the original curve of the PPG signal contains a lot of noise.

[0043] like Figure 5 As shown, after the denoising method of the present invention is performed, the noise of the curve is basically removed, the curve is smooth, and the key information of the detailed inflection points is retained.

[0044] like Figure 6 As shown, by comparing the spectrum before and after signal denoising, the present invention removes a large amount of noise.

[0045] like Figure 7 As shown in the figure, the results of the present invention are compared with those of discrete wavelet transform plus average filtering. The higher the SNR, the better the noise removal. The smaller the RMSE value, the better. The higher the PSNR, the better. Generally, above 40dB is considered excellent. Figure 7 The present invention can better reduce noise and retain key information; Figure 7 WT stands for walking tester, and RT stands for running tester.

[0046] The above-described embodiments do not constitute a limitation on the scope of protection of this technical solution. Any modifications, equivalent replacements, and improvements made within the spirit and principles of the above-described embodiments shall be included in the scope of protection of this technical solution.

Claims

1. A PPG signal denoising method combining discrete wavelet transform and optimized SG filtering is characterized by The following steps are involved: Step 1: Use DWT to decompose the original PPG signal, and use soft thresholding to perform soft thresholding on these decomposed signals to obtain a preliminary denoised signal. Step 2: Use the grid search algorithm to select the parameters of the SG filter in advance to achieve smoothing of the PPG signal. The final denoised PPG signal is obtained from the output port of the SG filter.

2. The PPG signal denoising method combining discrete wavelet transform and optimized SG filtering according to claim 1 is characterized in that The step 1 specifically includes the following steps: Step 1.1: After signal preprocessing, the signal is decomposed using the wavelet function db8 and the scaling parameter sln. The signal is then decomposed using a low-pass filter and a high-pass filter to obtain the approximate coefficients (representing the low-frequency part of the signal) and the detail coefficients (representing the high-frequency part of the signal), respectively. Step 1.2: Downsample the decomposed signal each time and decompose the approximate coefficients therein to the next level until the required number of decomposition layers is reached or the specific analysis requirements are met. The number of decomposition layers in the present invention is three, and three cycles are performed during the decomposition; Step 1.3: Based on the set threshold, use the soft threshold heuristic threshold sqtwolog processing, the processing formula is: λ is the threshold parameter; x is the unprocessed wavelet coefficient; x λ are the processed wavelet coefficients. Step 1.4: Perform inverse downsampling on the decomposed signal and reconstruct the signal.

3. The PPG signal denoising method combining discrete wavelet transform and optimized SG filtering according to claim 1 is characterized in that The step 2 specifically includes the following steps: Step 2.1: Determine the optimization target as the SG parameter, define the upper and lower bounds of the parameter, initialize it, evaluate the optimized parameter, and select the optimal parameter; Step 2.2: According to the given optimal parameters, the top window size and order are removed, and a fitting matrix is ​​constructed to fit the data. The fitting formula used is as follows: The polynomial coefficients are determined based on the residuals. The residual formula is as follows: In the above formula, the partial derivative is calculated to determine the minimum value of the residual; Write the above into matrix form, solve it, and bring it into the fitting equation to get the fitting curve. The specific matrix form is as follows: X·A=Y The final least squares solution equation is as follows: Step 2.3: Move the window and repeat the data fitting calculation to obtain the final curve and perform boundary processing.

Images