Signal noise reduction method and device based on short-time Fourier transform and sant transform

Abstract

The invention discloses a signal noise reduction method and device based on short-time Fourier transform and sant transform, and relates to the technical field of signal processing, and the method comprises the steps: obtaining an original signal, carrying out the sliding segmentation weighting of the original signal through a preset window function, and determining the length of a target original signal and a multi-frame weighted signal segment; respectively carrying out Slant transformation on each weighted signal segment to obtain corresponding transformation coefficients and form a transformation coefficient matrix; performing threshold processing on the transformation coefficient matrix to determine a target transformation coefficient matrix; slant inverse transformation is carried out frame by frame based on the target transformation coefficient matrix, windowing function reconstruction is carried out in combination with the target original signal length, and an initial noise reduction signal is output; and performing moving average smoothing processing on the initial noise reduction signal to determine a target noise reduction signal. A window function is introduced by adopting a short-time Fourier transform concept to carry out local observation on a time domain on a time sequence signal, the signal is divided into short frames, efficient separation and noise reduction are carried out in a Slant transform domain, and the signal noise reduction reliability is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of signal processing, and in particular, to a signal denoising method and device based on short-time Fourier transform and slant transform. Background Art

[0002] During the process from signal generation to signal acquisition by a device, the signal will be interfered by various noise sources, such as environmental noise, thermal noise, power frequency noise, etc. In the process of signal analysis, the presence of noise may cause misjudgments of true negatives or false positives, thereby affecting the quality of decision-making. Therefore, it is often necessary and important to perform denoising processing on the signal.

[0003] Since in reality, the frequency bandwidth of noise is often unknown and the composition is complex, it is usually impossible to solve the denoising of all signals through specific parameters. At the same time, the passband of the signal often overlaps with the noise passband. Traditional signal denoising methods perform low-pass filtering through filters such as Butterworth, Bessel, and Chebyshev. However, traditional low-pass filters are designed based on fixed frequency response characteristics and cannot be adjusted in real time according to changes in the noise environment. It is often difficult to effectively distinguish signals and noise in complex noise environments. In addition, it is easy to cause irreversible information loss and non-linear phase shift to the high-frequency useful information of the signal, resulting in low reliability of signal denoising. Summary of the Invention

[0004] The present invention provides a signal denoising method and device based on short-time Fourier transform and slant transform, which solves the technical problem of low reliability of signal denoising when traditional signal denoising methods perform low-pass filtering through filters.

[0005] A signal denoising method based on short-time Fourier transform and slant transform provided by the first aspect of the present invention includes:

[0006] Obtain the original signal, slide and segment the original signal by a preset window function to determine the target original signal length and multiple frames of weighted signal segments;

[0007] Perform Slant transform on each of the weighted signal segments to obtain corresponding transformation coefficients and form a transformation coefficient matrix;

[0008] Perform threshold processing on the transformation coefficient matrix to determine the target transformation coefficient matrix;

[0009] Perform inverse Slant transform frame by frame based on the target transformation coefficient matrix, and perform window function removal reconstruction in combination with the target original signal length to output an initial denoised signal;

[0010] Perform moving average smoothing processing on the initial denoised signal to determine the target denoised signal.

[0011] Optionally, the obtaining of the original signal includes slidingly dividing and weighting the original signal by a preset window function to determine the target original signal length and a plurality of weighted signal segments, including:

[0012] Obtain the original signal, divide the original signal according to a preset window function, and output a plurality of signal segments;

[0013] Multiply each of the signal segments element by element with the window function to determine the corresponding weighted signal segments;

[0014] Perform a difference operation on the initial original signal length of the original signal and the overlap length of the window function, then perform a ratio operation with the step size of the window function, and perform a floor operation to determine the number of frames;

[0015] Calculate the target original signal length of the original signal based on the number of frames;

[0016] Wherein, the determination process of the number of frames includes:

[0017] ;

[0018] The calculation process of the target original signal length includes:

[0019] ;

[0020] In the formula, is the number of frames, is the initial original signal length, is the window length, is the overlap ratio, is the step size, is the floor symbol; is the target original signal length.

[0021] Optionally, the threshold processing of the transform coefficient matrix to determine the target transform coefficient matrix includes:

[0022] Perform a frame averaging operation on the same coefficient elements of each frame of the transform coefficient matrix to determine a plurality of average transform coefficients;

[0023] Determine a plurality of target average transform coefficients from each of the average transform coefficients according to a preset threshold strategy;

[0024] Set the coefficient elements in the transform coefficient matrix that are not associated with the target average transform coefficients to zero to determine the target transform coefficient matrix;

[0025] Wherein, the determination process of the average transform coefficient includes:

[0026] ;

[0027] In the formula, is the average transformation coefficient of the th coefficient element, is the th coefficient element of the transformation coefficient of the th frame, is the number of frames.

[0028] Optionally, determining multiple target average transformation coefficients from each of the average transformation coefficients according to a preset threshold strategy includes:

[0029] Sort the average transformation coefficients in descending order, and select multiple average transformation coefficients that meet the preset number of retained coefficients from largest to smallest as the target average transformation coefficients.

[0030] Optionally, determining multiple target average transformation coefficients from each of the average transformation coefficients according to a preset threshold strategy includes:

[0031] Compare each of the average transformation coefficients with a preset noise reduction threshold parameter, and use the average transformation coefficients greater than the noise reduction threshold parameter as the target average transformation coefficients.

[0032] Optionally, performing an inverse Slant transform frame by frame based on the target transform coefficient matrix, and performing window function removal reconstruction in combination with the target original signal length to output an initial noise reduction signal, including:

[0033] Perform an inverse Slant transform on the target transform coefficient matrix frame by frame to determine multiple frames of noise reduction signal segments;

[0034] Multiply each of the noise reduction signal segments element by element with the window function to determine corresponding weighted noise reduction signal segments;

[0035] Perform overlapping accumulation using each of the weighted noise reduction signal segments according to the target original signal length to output a reconstructed signal;

[0036] Perform a square operation on the window function to construct a square window function;

[0037] Perform overlapping accumulation on the square window function according to the target original signal length to determine a normalized window function;

[0038] Divide the reconstructed signal element by element with the normalized window function to output an initial noise reduction signal;

[0039] Among them, the process of the inverse Slant transform includes:

[0040] ;

[0041] In the formula, is the noise reduction signal segment of the th frame, is the transpose of the n - order Slant transform matrix, is the frame transform coefficient.

[0042] A signal denoising device based on short - time Fourier transform and Slant transform provided by the second aspect of the present invention includes:

[0043] A signal windowing module, configured to obtain the original signal, perform sliding segmentation and weighting on the original signal through a preset window function, and determine the target original signal length and multiple frames of weighted signal segments;

[0044] A signal transformation module, configured to perform Slant transform on each of the weighted signal segments respectively to obtain corresponding transformation coefficients and form a transformation coefficient matrix;

[0045] A threshold processing module, configured to perform threshold processing on the transformation coefficient matrix to determine the target transformation coefficient matrix;

[0046] A signal reconstruction module, configured to perform inverse Slant transform frame by frame based on the target transformation coefficient matrix, and perform window - function - removing reconstruction in combination with the target original signal length to output an initial denoised signal;

[0047] A signal smoothing module, configured to perform moving average smoothing processing on the initial denoised signal to determine the target denoised signal.

[0048] A computer device provided by the third aspect of the present invention includes a memory and a processor. When the computer program stored in the memory is executed by the processor, the processor executes the steps of the signal denoising method based on short - time Fourier transform and Slant transform as described in any one of the above.

[0049] A computer - readable storage medium provided by the fourth aspect of the present invention has a computer program stored thereon. When the computer program is executed, it implements the signal denoising method based on short - time Fourier transform and Slant transform as described in any one of the above.

[0050] A computer program product provided by the fifth aspect of the present invention includes a computer program / instructions. When the computer program / instructions are executed by a processor, they implement the signal denoising method based on short - time Fourier transform and Slant transform as described in any one of the above.

[0051] From the above technical solutions, it can be seen that the present invention has the following advantages:

[0052] The above solution of the present invention provides a signal denoising method based on short-time Fourier transform and slant transform, including: obtaining an original signal, sliding and segmenting and weighting the original signal through a preset window function to determine the target original signal length and multiple weighted signal segments; performing slant transform on each weighted signal segment respectively to obtain corresponding transform coefficients and form a transform coefficient matrix; performing threshold processing on the transform coefficient matrix to determine the target transform coefficient matrix; performing inverse slant transform frame by frame based on the target transform coefficient matrix, and performing reconstruction without the window function in combination with the target original signal length to output an initial denoised signal; performing moving average smoothing processing on the initial denoised signal to determine the target denoised signal. Based on the above solution, the concept of short-time Fourier transform is introduced to perform local observation of the time-series signal in the time domain through the window function, and the boundary effect is reduced by the signal segmentation method of the window function. Dividing the signal into short frames for slant transform can provide the slant transform domain information of the signal at different time points, so as to utilize the characteristic differences between noise and clean signals, and concentrate the signal energy in the coefficients of a few slant transform domains. Therefore, the signal and noise components are efficiently separated in the slant transform domain. At the same time, since the transform coefficient matrix will show different distributions according to different noises in the environment, it helps to enhance the self-adaptability of signal denoising in a complex noise environment, achieve more refined denoising, and overall improve the reliability of signal denoising. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0054] Figure 1 It is a flowchart of the steps of a signal denoising method based on short-time Fourier transform and slant transform provided by an embodiment of the present invention;

[0055] Figure 2 It is a schematic diagram of the denoising experimental effect of the signal denoising method based on short-time Fourier transform and slant transform provided by an embodiment of the present invention;

[0056] Figure 3 It is a schematic diagram of the denoising experimental effect of the band-stop filter IIR digital notch filter provided by an embodiment of the present invention;

[0057] Figure 4 It is a structural block diagram of a signal denoising device based on short-time Fourier transform and slant transform provided by an embodiment of the present invention; DETAILED DESCRIPTION OF THE EMBODIMENTS

[0058] An embodiment of the present invention provides a signal denoising method and device based on short-time Fourier transform and slant transform, which are used to solve the technical problem that the traditional signal denoising method performs low-pass filtering through a filter, but causes irreversible information loss and non-linear phase shift to the high-frequency useful information of the signal, resulting in low reliability of signal denoising.

[0059] In order to make the invention purpose, features, and advantages of the present invention more obvious and understandable, the following will combine the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the embodiments described below are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative work belong to the scope of protection of the present invention.

[0060] Please refer to Figure 1 , Figure 1 , which is a step flowchart of a signal denoising method based on short-time Fourier transform and slant transform provided by an embodiment of the present invention.

[0061] A signal denoising method based on short-time Fourier transform and slant transform provided by the present invention includes:

[0062] Step 101: Obtain the original signal, perform sliding segmentation and weighting on the original signal through a preset window function, and determine the target original signal length and multiple weighted signal segments.

[0063] Step 101 includes the following sub-steps:

[0064] S11: Obtain the original signal, segment the original signal according to the preset window function, and output multiple signal segments;

[0065] S12: Multiply each signal segment with the window function element by element to determine the corresponding weighted signal segment;

[0066] S13: Perform a difference operation between the initial original signal length of the original signal and the overlap length, then perform a ratio operation with the step size of the window function, and perform a floor operation to determine the number of frames;

[0067] S14: Calculate the target original signal length of the original signal based on the number of frames.

[0068] It should be noted that the original signal refers to the original signal to be denoised. In this embodiment, the original signal is segmented by a sliding window according to different window functions selected according to the signal, such as Hamming window, Hanning window, etc. The total number of window signals is denoted as the number of frames , the length of the window signal is the window length , and the window length is set to It is a standardized processing method, is a non - negative integer and its value depends on the signal length, and the overlapping part shared between two adjacent frames is the overlapping length, that is, the window length and the overlapping ratio The product value of, the overlapping ratio mainly depends on the signal characteristics and the required slant transform domain coefficient resolution and time resolution. In signal processing, the overlapping length is usually set to half or three - quarters of the window length;

[0069] In specific implementation, represents the original signal, where, is the target original signal length of the original signal, that is, the original signal length consistent with the signal length divided by the window function in integers. Segment the original signal with a window function of to obtain a series of signal segments, and there is an overlapping part between each signal segment. Then the th frame signal segment is , represents the rd signal of the th signal segment, with a step size of , then the th frame signal segment is ; In actual application, the length of the last signal segment may not be the same as the window length of the set window function. To ensure that the length of each frame of signal is the same during subsequent signal processing, it is necessary to determine the target original signal length of the original signal according to the division of the window function. First, determine the number of frames , is the floor symbol. Assume is the initial original signal length of the original signal. Then, truncate at the last frame of the original signal, that is, , and delete the signal segments exceeding this position. The remaining signal length is denoted as . Therefore, the calculation process of the target original signal length includes: ;

[0070] Express the window function as , taking the form of a column vector as an example:

[0071] ;

[0072] In the formula, is the th window element; then the rd frame windowed signal segment, that is, the weighted signal segment, is , is the th weighted element of the th frame weighted signal segment, denotes element-wise multiplication, is the window function transpose.

[0073] Step 102: Perform the Slant transform on each weighted signal segment respectively to obtain the corresponding transform coefficients and form a transform coefficient matrix.

[0074] It should be noted that for each weighted signal segment perform the Slant transform using an n-order Slant transform matrix to obtain the transform coefficients , is the transpose of the weighted signal segment. Within each short frame, the Slant transform provides the Slant transform domain information of the signal and can distinguish different transform domain coefficient components in the signal;

[0075] where ;

[0076] In the formula, 0 is a matrix of all zeros, is the identity matrix of order ( ), The matrix dimension of is , and the Slant transform matrix is generated iteratively with its initial condition being . According to its property of satisfying a real-valued orthogonal matrix, it can be solved that:

[0077] ;

[0078] Express the transform coefficients as:

[0079] ;

[0080] In the formula, is the th coefficient element of the transform coefficients of the th frame; then, by arranging the transform coefficients in matrix form, a transform coefficient matrix can be obtained:

[0081] .

[0082] Step 103: Perform threshold processing on the transform coefficient matrix to determine the target transform coefficient matrix.

[0083] Step 103 includes the following sub-steps:

[0084] S21: Perform frame averaging operation on the same coefficient elements of each frame of the transform coefficient matrix to determine multiple average transform coefficients;

[0085] S22: Determine multiple target average transform coefficients from each average transform coefficient according to a preset threshold strategy;

[0086] S23. Set the coefficient elements in the transform coefficient matrix that are not associated with each target average transform coefficient to zero to determine the target transform coefficient matrix.

[0087] In a specific embodiment, sub-step S22 includes: arranging the average transform coefficients in descending order, and selecting multiple average transform coefficients that meet the preset number of retained coefficients from largest to smallest as the target average transform coefficients.

[0088] In another specific embodiment, sub-step S22 includes: comparing each average transform coefficient with a preset noise reduction threshold parameter, and taking the average transform coefficients greater than the noise reduction threshold parameter as the target average transform coefficients.

[0089] It should be noted that in this embodiment, the same coefficient element refers to the coefficient element at the same position in each frame. For example and , first, calculate the average of each coefficient element over all frames as a representative noise evaluation index. Of course, the median of the same coefficient elements in each frame can also be used as a representative noise evaluation index; the determination process of the average transform coefficient includes: , where is the average transform coefficient of the th coefficient element, is the th coefficient element of the transform coefficient of the th frame. Taking the average transform coefficient of the first coefficient element as an example, , and so on, an average transform coefficient matrix of can be obtained:

[0090] ;

[0091] Then, select and determine the transform coefficients in the transform domain according to a certain threshold strategy as the target average transform coefficients. The threshold strategy refers to a strategy for screening or modifying the transform coefficients according to preset rules or conditions, which defines how to apply the threshold to decide which coefficients to retain, which coefficients to set to zero or adjust, such as hard threshold processing, soft threshold processing in wavelet transform, etc. The target average transform coefficient is understood to be the average transform coefficient corresponding to the clean signal part; in one implementation, the threshold strategy is to set the number of retained coefficients, that is, the number of retained average transform coefficients. In specific implementation, first arrange the average transform coefficients in descending order and retain their indices I, and set the number of retained coefficients to , then select The average transform coefficients are taken as the target average transform coefficients, and it is considered that the target average transform coefficients correspond to the signal part, and the remaining average transform coefficients are considered to correspond to the noise part; in another implementation, the threshold strategy is to set a noise reduction threshold parameter, that is, to determine whether it belongs to the quantitative standard of noise. In the specific implementation, it is considered that the average transform coefficients greater than the noise reduction threshold parameter are the target average transform coefficients corresponding to the signal, and conversely, the average transform coefficients less than the noise reduction threshold parameter correspond to the noise;

[0092] Finally, the coefficient elements that are not associated with the target average transform coefficients, that is, the coefficient elements associated with the average transform coefficients of the noise part, are set to zero in the transform coefficient matrix to achieve the purpose of noise reduction, thereby obtaining the target transform coefficient matrix:

[0093] .

[0094] Step 104: perform Slant inverse transform frame by frame based on the target transform coefficient matrix, reconstruct the window function in combination with the target original signal length, and output the initial noise reduction signal.

[0095] Step 104 includes the following sub-steps:

[0096] S31, performing Slant inverse transformation on the target transformation coefficient matrix frame by frame to determine multiple frames of noise reduction signal segments;

[0097] S32, multiplying each denoised signal segment by the window function element by element to determine a corresponding weighted denoised signal segment;

[0098] S33, overlapping and accumulating each weighted noise reduction signal segment according to the target original signal length, and outputting a reconstructed signal;

[0099] S34, using a window function to perform a square operation to construct a square window function;

[0100] S35, overlapping and accumulating the square window function according to the target original signal length to determine the normalized window function;

[0101] S36, dividing the reconstructed signal by the normalized window function element by element, and outputting an initial denoised signal.

[0102] It should be noted that, in this embodiment, after noise reduction is completed, the target transformation coefficient matrix is ​​subjected to Slant inverse transformation frame by frame, and the process of Slant inverse transformation includes: , where is the transpose of the n-order Slant transformation matrix, For the Frame noise reduction signal segment, exemplarily, the first frame noise reduction signal segment , the second frame of noise reduction signal fragment , the third-frame noise reduction signal segment is as follows:

[0103] , , ;

[0104] Then, perform window function reconstruction by means of accumulation and normalization to obtain the complete noise reduction signal: Multiply each frame of the noise reduction signal segment by the window function element by element to determine the weighted noise reduction signal segment . It can be understood that if, as in the previous example, the original signal is in matrix row form, while the noise reduction signal segment and the window function are in matrix column form, then transpose the two and multiply element by element, that is , is the transpose of the window function, is the transpose of the noise reduction signal segment. Just try to keep the arrangement of the weighted noise reduction signal segment consistent with the original signal. The specific arrangement form can be adaptively set. Exemplarily, the first-frame weighted noise reduction signal segment is , the second-frame weighted noise reduction signal segment is , and the third-frame weighted noise reduction signal segment is ; According to the target original signal length let the reconstructed signal be matrix of , and then add each weighted noise reduction signal segment into in the form of overlap and add for signal reconstruction to obtain the reconstructed signal , is the th weighted noise reduction element of the th frame of the weighted noise reduction signal segment, is the th weighted noise reduction element of the th frame of the weighted noise reduction signal segment; To achieve normalization, first find the square of the window function to obtain the squared window function . If the aforementioned weighted noise reduction signal segment uses the transpose of the window function, then the squared window function is . Exemplarily is matrix. Let the normalized window function be matrix of . Obtain the normalized window function in the same accumulation manner. Divide each element of by each element of one by one to achieve normalization, eliminate the influence of the window function, and obtain matrix which is the initial noise reduction signal . The determination process of the initial noise reduction signal includes:

[0105] 。

[0106] Step 105: Perform moving average smoothing on the initial noise-reduced signal to determine the target noise-reduced signal.

[0107] It should be noted that the reconstructed initial noise-reduced signal may exhibit a sawtooth phenomenon. Therefore, the smoothdata function is used to smooth the initial noise-reduced signal to obtain the smoothed target noise-reduced signal. In a preferred implementation, moving average smoothing is adopted. In another preferred implementation, ssa smoothing is adopted.

[0108] To verify the effectiveness of this solution, taking electroencephalogram (EEG) signals as an example, this solution is used to perform noise reduction on EEG signals. Among them, moving average smoothing is adopted. After completing the preprocessing, steps such as feature extraction and classification are carried out. The experimental results are as Figure 2 shown, and experiments are carried out using traditional EEG signal noise reduction methods, that is, using a band-stop filter IIR digital notch filter to achieve signal noise reduction. After completing the preprocessing, the same steps of feature extraction and classification are carried out as above. The experimental results are as Figure 3 shown; that is, except for the different noise reduction methods adopted in the two experimental schemes, the other steps are the same. According to Figure 2 and Figure 3 it can be seen that the accuracy rate of this solution is 0.7359, and the accuracy rate of the traditional method is 0.6620. Since 0.7359 > 0.6620, it can be obtained that the signal noise reduction method proposed in this embodiment has certain excellent effects.

[0109] In the embodiment of the present invention, the concept of short-time Fourier transform is adopted to introduce a window function to perform local observation of the time-series signal in the time domain. By segmenting the signal through the window function, the boundary effect is reduced. And splitting the signal into short frames for slant transform can provide the slant transform domain information of the signal at different time points. Thus, by utilizing the characteristic differences between noise and clean signals, the signal energy is concentrated in a few slant transform domain coefficients. Therefore, the signal and noise components are efficiently separated in the slant transform domain. At the same time, since the transform coefficient matrix will exhibit different distributions according to different noises in the environment, it helps to enhance the self-adaptability of signal noise reduction in a complex noise environment, achieve more refined noise reduction, and overall improve the reliability of signal noise reduction.

[0110] Please refer to Figure 4 , Figure 4 which is the structural block diagram of a signal noise reduction device based on short-time Fourier transform and slant transform provided by the embodiment of the present invention.

[0111] A signal denoising device based on short-time Fourier transform and slant transform provided by the present invention includes:

[0112] A signal windowing module 401, configured to obtain an original signal, perform sliding segmentation and weighting on the original signal through a preset window function, and determine a target original signal length and multiple weighted signal segments;

[0113] A signal transformation module 402, configured to perform slant transform on each weighted signal segment respectively to obtain corresponding transformation coefficients and form a transformation coefficient matrix;

[0114] A threshold processing module 403, configured to perform threshold processing on the transformation coefficient matrix to determine a target transformation coefficient matrix;

[0115] A signal reconstruction module 404, configured to perform inverse slant transform frame by frame based on the target transformation coefficient matrix, and perform reconstruction without the window function in combination with the target original signal length, and output an initial denoised signal;

[0116] A signal smoothing module 405, configured to perform sliding average smoothing on the initial denoised signal to determine a target denoised signal.

[0117] Optionally, the signal windowing module 401 is specifically configured to:

[0118] Obtain the original signal, segment the original signal according to the preset window function, and output multiple signal segments;

[0119] Multiply each signal segment by the window function element by element to determine the corresponding weighted signal segment;

[0120] Perform a difference operation on the initial original signal length of the original signal and the overlap length of the window function, then perform a ratio operation with the step size of the window function, and perform a floor operation to determine the number of frames;

[0121] Calculate the target original signal length of the original signal based on the number of frames;

[0122] Among them, the determination process of the number of frames includes:

[0123] ;

[0124] The calculation process of the target original signal length includes:

[0125] ;

[0126] In the formula, is the number of frames, is the initial original signal length, is the window length, is the overlap ratio, is the step size, is the floor symbol; is the length of the target original signal.

[0127] Optionally, the threshold processing module 403 is specifically configured to:

[0128] Perform frame averaging operations on the same coefficient elements of each frame of the transform coefficient matrix to determine multiple average transform coefficients;

[0129] Determine multiple target average transform coefficients from each average transform coefficient according to a preset threshold strategy;

[0130] Set the coefficient elements in the transform coefficient matrix that are not associated with each target average transform coefficient to zero to determine the target transform coefficient matrix;

[0131] Among them, the determination process of the average transform coefficient includes:

[0132] ;

[0133] In the formula, is the average transform coefficient of the th coefficient element, is the th coefficient element of the th frame transform coefficient, is the number of frames.

[0134] Optionally, determining multiple target average transform coefficients from each average transform coefficient according to a preset threshold strategy includes:

[0135] Arrange each average transform coefficient in descending order, and select multiple average transform coefficients that meet the preset number of coefficients to be retained from largest to smallest as the target average transform coefficients.

[0136] Optionally, determining multiple target average transform coefficients from each average transform coefficient according to a preset threshold strategy includes:

[0137] Compare each average transform coefficient with a preset noise reduction threshold parameter, and use the average transform coefficients greater than the noise reduction threshold parameter as the target average transform coefficients.

[0138] Optionally, the signal reconstruction module 404 is specifically configured to:

[0139] Perform Slant inverse transformation on each frame of the target transform coefficient matrix to determine multiple frames of noise-reduced signal segments;

[0140] Multiply each noise-reduced signal segment element by element with a window function to determine the corresponding weighted noise-reduced signal segment;

[0141] Perform overlapping accumulation on each weighted noise-reduced signal segment according to the length of the target original signal, and output the reconstructed signal;

[0142] Perform a square operation on the window function to construct a square window function;

[0143] Overlap and add the square window function according to the length of the target original signal to determine the normalized window function;

[0144] Divide the reconstructed signal element by element with the normalized window function and output the initial noise-reduced signal;

[0145] Among them, the process of the Slant inverse transform includes:

[0146] ;

[0147] In the formula, is the noise-reduced signal segment of the th frame, is the transpose of the nth-order Slant transform matrix, is the transform coefficient of the th frame.

[0148] An embodiment of the present invention also provides a computer device, including a memory and a processor, and a computer program is stored in the memory; when the computer program is executed by the processor, the processor executes the steps of the signal noise reduction method based on the short-time Fourier transform and the slant transform in any of the above embodiments.

[0149] An embodiment of the present invention also provides a computer-readable storage medium, on which a computer program / instruction is stored, and when the computer program / instruction is executed by the processor, the steps of the signal noise reduction method based on the short-time Fourier transform and the slant transform in any of the above embodiments are implemented.

[0150] An embodiment of the present invention also provides a computer program product, including a computer program / instruction, and when the computer program / instruction is executed by the processor, the steps of the signal noise reduction method based on the short-time Fourier transform and the slant transform in any of the above embodiments are implemented.

[0151] Those skilled in the art can clearly understand that for the convenience and conciseness of description, the specific working processes of the above-described devices and modules can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0152] In several embodiments provided by the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections between each other can be through some interfaces. The indirect couplings or communication connections of the devices or units can be in electrical, mechanical or other forms.

[0153] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0154] In addition, in each embodiment of the present invention, the functional units can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.

[0155] If the above-mentioned integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in each embodiment of the present invention. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0156] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of each embodiment of the present invention.

Claims

1. A signal denoising method based on short-time Fourier transform and slant transform, characterized in that: include: Acquire the original signal, perform sliding segmentation weighting on the original signal through a preset window function, and determine the target original signal length and multi-frame weighted signal segments; Performing Slant transformation on each of the weighted signal segments to obtain corresponding transformation coefficients and form a transformation coefficient matrix; Performing threshold processing on the transform coefficient matrix to determine a target transform coefficient matrix; Based on the target transformation coefficient matrix, Slant inverse transformation is performed frame by frame, and window removal function is reconstructed in combination with the target original signal length to output an initial noise reduction signal; The initial noise reduction signal is subjected to sliding average smoothing processing to determine a target noise reduction signal.

2. The signal denoising method based on short-time Fourier transform and slant transform according to claim 1, characterized in that: The obtaining of the original signal, sliding segmentation and weighting of the original signal by a preset window function, and determining a target original signal length and a plurality of weighted signal segments include: Acquire an original signal, segment the original signal according to a preset window function, and output a plurality of signal segments; Multiplying each of the signal segments by the window function element by element to determine a corresponding weighted signal segment; Performing a difference operation on the initial original signal length of the original signal and the overlapping length of the window function, performing a ratio operation with the step length of the window function, and rounding down to determine the number of frames; Calculating a target original signal length of the original signal based on the number of frames; The process of determining the number of frames includes: ； The calculation process of the target original signal length includes: ； In the formula, is the number of frames, is the initial original signal length, is the window length, is the overlap ratio, is the step length, is the floor rounding symbol; is the target original signal length.

3. The signal denoising method based on short-time Fourier transform and slant transform according to claim 1, characterized in that: The step of performing threshold processing on the transform coefficient matrix to determine a target transform coefficient matrix includes: Performing frame averaging operation on the same coefficient element of each frame of the transform coefficient matrix to determine a plurality of average transform coefficients; Determining a plurality of target average transform coefficients from the average transform coefficients according to a preset threshold strategy; Setting coefficient elements in the transform coefficient matrix that are not associated with the target average transform coefficients to zero to determine a target transform coefficient matrix; The process of determining the average transform coefficient includes: ； In the formula, For the The average transform coefficient of coefficient elements, For the The frame transform coefficients coefficient elements, is the frame number.

4. The signal denoising method based on short-time Fourier transform and slant transform according to claim 3 is characterized in that: The determining a plurality of target average transform coefficients from the average transform coefficients according to a preset threshold strategy includes: The average transform coefficients are arranged in descending order, and a plurality of average transform coefficients that meet a preset coefficient retention number are selected from large to small as target average transform coefficients.

5. The signal denoising method based on short-time Fourier transform and slant transform according to claim 3, characterized in that: The determining a plurality of target average transform coefficients from the average transform coefficients according to a preset threshold strategy includes: The average transform coefficients are compared with a preset noise reduction threshold parameter, and the average transform coefficients greater than the noise reduction threshold parameter are used as target average transform coefficients.

6. The signal noise reduction method based on short-time Fourier transform and slant transform according to claim 1, characterized in that: The method of performing Slant inverse transformation frame by frame based on the target transformation coefficient matrix, reconstructing the window function in combination with the target original signal length, and outputting an initial noise reduction signal includes: Performing Slant inverse transformation on the target transformation coefficient matrix frame by frame to determine multiple frames of noise reduction signal segments; Multiplying each of the noise reduction signal segments by the window function element by element to determine a corresponding weighted noise reduction signal segment; According to the target original signal length, each of the weighted noise reduction signal segments is overlapped and accumulated, and a reconstructed signal is output; Using the window function to perform a square operation to construct a square window function; Performing overlapping accumulation on the square window function according to the target original signal length to determine a normalized window function; Dividing the reconstructed signal by the normalized window function element by element, and outputting an initial noise reduction signal; The process of the Slant inverse transformation includes: ； In the formula, For the Frame noise reduction signal fragment, is the transpose of the n-order Slant transformation matrix, For the Frame transform coefficients.

7. A signal noise reduction device based on short-time Fourier transform and slant transform, characterized in that: include: A signal windowing module is used to obtain an original signal, perform sliding segmentation weighting on the original signal through a preset window function, and determine a target original signal length and a multi-frame weighted signal segment; A signal transformation module, used to perform Slant transformation on each of the weighted signal segments to obtain corresponding transformation coefficients and form a transformation coefficient matrix; A threshold processing module, used for performing threshold processing on the transform coefficient matrix to determine a target transform coefficient matrix; A signal reconstruction module, used to perform Slant inverse transformation frame by frame based on the target transformation coefficient matrix, and to perform window removal function reconstruction in combination with the target original signal length, and output an initial noise reduction signal; The signal smoothing module is used to perform sliding average smoothing processing on the initial noise reduction signal to determine a target noise reduction signal.

8. A computer device, characterized in that: It comprises a memory and a processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the processor executes the steps of the signal denoising method based on short-time Fourier transform and slant transform as described in any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that: When the computer program / instructions are executed by the processor, the steps of the signal noise reduction method based on short-time Fourier transform and slant transform as described in any one of claims 1 to 6 are implemented.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by the processor, the steps of the signal noise reduction method based on short-time Fourier transform and slant transform as described in any one of claims 1 to 6 are implemented.

Images