Multistage detection model for signal time window

Abstract

This invention discloses a multi-level detection model for signal time windows, comprising: a signal time window partitioning model, which partitions the signal time window at least two levels, i.e., a primary signal time window is composed of multiple secondary signal time windows; a signal time window detection model, which divides the secondary signal time windows into secondary interference time windows and secondary normal time windows using a time window detection method; and a classification model, which jointly judges the primary signal time window based on the detection results of the secondary signal time windows, thereby classifying the primary signal time window into sparse noise signal time windows, dense noise signal time windows, and normal signal time windows. This invention can improve the efficiency of signal processing and prevent missed detections.

Description

[0001] This application is a divisional application of Chinese patent application filed on June 6, 2023, with application number 202310658542.7, entitled "Method for detecting signal noise, signal detection model, and readable storage medium". Technical Field

[0002] This invention relates to the field of signal detection technology, and more specifically to a multi-level detection model for signal time windows. Background Technology

[0003] Since the acquisition of the original signal is often accompanied by other noise signals, resulting in a low signal-to-noise ratio and significantly impacting signal quality, noise reduction is a necessary preprocessing step before signal application. The noise reduction process generally includes locating and removing noise signals. Typically, methods such as ICA (Independent Component Analysis) and PCA (Principal Component Analysis) are first used to locate and analyze the noise signals, followed by filtering to remove them. However, these methods have two main shortcomings: first, the processing of the decomposed components is too coarse, often directly discarding entire sub-components containing noise signals, leading to the removal of some useful signals; second, they only address a single type of noise signal, applying the same processing method even when the signal contains multiple types of noise. Summary of the Invention

[0004] This invention aims to solve one of the technical problems existing in the prior art. To this end, this invention provides a multi-level detection model for signal time windows, which can improve the signal-to-noise ratio of the reference signal by measuring, classifying, evaluating, locating, and separating noise signals in the original signal.

[0005] The technical solution adopted by this invention to solve its technical problem is: a multi-level detection model for signal time windows, comprising:

[0006] The signal time window partitioning model requires at least two levels of signal time window partitioning, i.e., a primary signal time window is composed of multiple secondary signal time windows.

[0007] The signal time window detection model divides the secondary signal time window into a secondary interference time window and a secondary normal time window using a time window detection method.

[0008] The classification model jointly judges the primary signal time window based on the detection results of the secondary signal time window, so as to classify the primary signal time window into sparse noise signal time window, dense noise signal time window, and normal signal time window.

[0009] Furthermore, the time window detection method is at least one of the following: generalized feature decomposition algorithm, multi-feature classification algorithm, and multi-feature joint detection algorithm.

[0010] Further, the generalized eigenvalue decomposition method includes: calculating the feature matrices of the signal time window and the background signal respectively;

[0011] Calculating the maximum generalized eigenvalue between the feature matrix of the signal time window and the feature matrix of the background signal;

[0012] If the maximum generalized eigenvalue > the set threshold, mark this signal time window as a noise signal time window; otherwise, mark it as a normal signal time window.

[0013] Further, the multi-feature joint detection method includes:

[0014] Calculating multiple eigenvalues of the signal time window;

[0015] Performing threshold judgment on each of the multiple eigenvalues one by one. When all the multiple eigenvalues of the signal time window meet the threshold conditions, mark this signal time window as a noise signal time window.

[0016] Further, the multi-feature classification method includes:

[0017] Calculating multiple eigenvalues of the signal time window and inputting the eigenvalues into a classifier for classification to output noise signal time windows and normal signal time windows.

[0018] Further, the joint judgment of the first-level signal time window based on the detection results of the second-level signal time window includes:

[0019] Calculating the proportion p of the second-level interference time windows in the first-level signal time window;

[0020] When the proportion p = 0, mark the first-level signal time window as a normal signal time window;

[0021] When 0 < p < the proportion threshold, mark the first-level signal time window as a sparse noise signal time window;

[0022] When the proportion p ≥ the proportion threshold, mark the first-level signal time window as a dense noise signal time window.

[0023] Further, the window width of the first-level signal time window is greater than the window width of the second-level signal time window.

[0024] Further, when the first-level signal time window is a sparse noise signal time window, only construct different signal processing models for the second-level interference time windows in the first-level signal time window;

[0025] When the first-level signal time window is a dense noise signal time window, construct different signal processing models for the first-level signal time window.

[0026] The beneficial effects of this invention are that, for the first-level signal time window of sparse noise signals, this method only processes the second-level interference time window. This avoids processing non-noise signal segments, preventing the loss of effective signals, and also reduces the amount of data to be processed, improving processing efficiency. For the first-level signal time window of dense noise signals, processing the entire first-level signal time window results in higher computational efficiency and also prevents missed processing. Attached Figure Description

[0027] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0028] Figure 1 This is a flowchart of the signal noise detection method of the present invention.

[0029] Figure 2 This is a flowchart of the signal noise detection method of the present invention, which processes signals according to priority.

[0030] Figure 3 This is a flowchart of the low-frequency noise signal localization process of the present invention.

[0031] Figure 4 This is a comparison chart of the correlation coefficients of Example 1, Comparative Examples 1 and 2.

[0032] Figure 5 This is a comparison chart of the relative root mean square error of Example 1, Comparative Examples 1 and 2.

[0033] Figure 6 These are comparison charts showing the interference removal effects of Example 2, Comparative Examples 3 and 4.

[0034] Figure 7 This is a schematic diagram of the signal time window classification and detection method of the present invention.

[0035] Figure 8 This is a flowchart of the generalized feature decomposition method of the present invention.

[0036] Figure 9 This is a flowchart of the multi-feature joint detection of the present invention.

[0037] Figure 10 This is a flowchart of the multi-feature classification detection of the present invention.

[0038] Figure 11 This is a diagram showing the result of using the generalized eigenvalue decomposition method to remove noise signals according to the present invention.

[0039] Figure 12 This is a graph showing the result of denoising the signal using existing techniques.

[0040] Figure 13 This is a comparison chart of the results of joint feature detection and single feature detection in identifying noise signals. Detailed Implementation

[0041] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0042] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," "counterclockwise," "axial," "radial," and "circumferential," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, features defined with "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0043] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0044] like Figures 1 to 2 As shown, the signal noise detection method of the present invention includes the following steps: S1, dividing the signal into time windows; S2, classifying the signal time windows into noise signal time windows and normal signal time windows; S3, classifying the noise signal categories within the noise signal time windows; S4, constructing different signal processing models for the noise signal time windows according to the noise signal categories, so as to separate the corresponding noise signals. It should be noted that dividing the signal into multiple signal time windows and then classifying each signal time window can improve detection efficiency. After detecting the noise signal time windows, the noise signal categories contained within the noise signal time windows can be obtained. Different categories can be separated using different signal processing models, making the method more targeted.

[0045] For example, the signal processing model includes: obtaining a time decomposition matrix V and a spatial decomposition matrix U based on the noise signal time window, where the time decomposition matrix V is a K×K matrix, K is the number of time-domain sampling points; the spatial decomposition matrix U is an M×M matrix, M is the number of channels; extracting the first M columns from the time decomposition matrix V to form a source component matrix A, where M is the number of channels; locating the position of the noise signal in the source component matrix A; determining whether the position of the noise signal in the source component matrix A is null; if the position of the noise signal is not null, sequentially performing noise signal separation and source reconstruction to continue iterating; if the position of the noise signal is null, terminating the iteration. Source reconstruction includes: outputting a matrix V_hat after filtering the noise signal; performing source decomposition inverse operations on the spatial decomposition matrix U and the matrix V_hat to obtain the signal S_hat. Of course, the signal processing model in this case can also be non-iterative, i.e., single-step noise signal separation.

[0046] It should be noted that the time decomposition matrix V is a decomposition of the signal in the time dimension, and each column of the time decomposition matrix V reflects the time-domain distribution information of each sub-component. The spatial decomposition matrix U is a decomposition of the signal in the spatial distribution dimension of the electrodes, and each column reflects the spatial distribution information of each sub-component. The time decomposition matrix V and spatial decomposition matrix U for obtaining the noise signal time window can be obtained using either Singular Value Decomposition (SVD) or Eigenvalue Decomposition (ED), without restriction here. Singular Value Decomposition (SVD) involves decomposing the multi-channel signal matrix S into: S = U∑V T , S∈R M×K It is a signal matrix with M channels and K time samples, U∈R M×M It is a left singular matrix, ∑∈R M×K It is a singular value matrix, V∈R K×K It is a right singular matrix. The left singular matrix U is a spatial decomposition of the signal matrix S, where each column reflects the spatial distribution of its components. The right singular matrix V is a temporal decomposition of the signal matrix S, where each column reflects the temporal distribution of its components. Eigenvalue decomposition (ED) involves: for a signal matrix S ∈ R... M×K Performing SVD to obtain the left singular matrix U and the right singular matrix V can be equivalently obtained through the following ED: Calculate the spatial covariance matrix Rs = S*S of S. T / K, perform ED on Rs: Rs=U∑V T U represents the characteristic matrix of Rs, which is the left singular matrix of S; calculate the time-domain covariance matrix Rt = S T *S / M, perform ED on Rt: Rt=V∑V T V represents the eigenma matrix of Rt, which is the right singular matrix of S. Before locating the noise signal, this invention first constructs the source signal (i.e., the source component matrix A), A = [v1, v2, ..., v...].M ], v1~v M This represents the first M columns of the time decomposition matrix V.

[0047] It should be noted that the noise signal categories within the noise signal measurement window include: non-biological noise signals and biological noise signals. To improve the accuracy of noise signal removal, this invention employs different measurement and separation methods for different types of noise signals. Since biological and non-biological noise signals have different sources and varying degrees of influence on the signal, the presence of multiple types of noise signals can affect signal detection, especially when high-energy noise signals are present. Therefore, this invention prioritizes or simultaneously removes a certain type of noise signal during noise signal separation, which is beneficial for accurate separation between the signal and the interference source. Non-biological and biological noise signals are determined, for example but not limited to, based on energy (signal-to-noise ratio). Noise signals with a definite source are separated first, i.e., high-energy noise signals are processed first. When the energy of a non-biological noise signal significantly exceeds that of a biological noise signal, the non-biological noise signal is processed first, followed by the biological noise signal. The priority order can be set as: non-biological noise signal > biological noise signal. When the noise energy of non-biological noise signals is much lower than that of biological noise signals, biological noise signals should be processed first, followed by non-biological noise signals. The priority order can be set as: non-biological noise signal < biological noise signal. Then, different signal processing models are run sequentially according to the priority order to separate noise signals of different levels step by step. When the noise energy of non-biological noise signals is similar to that of biological noise signals, any type of noise signal can be removed simultaneously or randomly.

[0048] As an alternative implementation method for non-biological noise signals.

[0049] In this case, non-biological noise signals include, but are not limited to, power frequency noise, environmental electromagnetic interference, and step noise. Optionally, any type of noise signal can be removed simultaneously, randomly, or in order of priority. Preferably, based on the spatial decomposition matrix, power frequency noise is usually unavoidable and distributed across all channels, so it is usually removed first. Environmental electromagnetic interference is also distributed across all channels, so it is removed secondarily. Step noise usually occurs in a few channels with large amplitude variations, so it is processed last. Therefore, the priority order of non-biological noise signals can be set as follows based on the prevalence of noise signals: power frequency noise > environmental electromagnetic interference > step noise.

[0050] (1) The following measures are taken to address power frequency noise and environmental electromagnetic interference:

[0051] For example, the measurement of power frequency noise and environmental electromagnetic interference includes locating the power frequency noise and environmental electromagnetic interference signals using spatial and time-frequency characteristics. The separation of power frequency noise and environmental electromagnetic interference includes filtering the source components containing power frequency noise and environmental electromagnetic interference at specific frequencies to separate the signals. Both power frequency noise and environmental electromagnetic interference affect the signals in all channels; therefore, the variance of the spatial decomposition vector corresponding to power frequency noise and environmental electromagnetic interference is almost zero. Furthermore, the spectrum of the time decomposition matrix corresponding to power frequency noise and environmental electromagnetic interference exhibits peaks at specific frequencies; therefore, the location of power frequency noise and environmental electromagnetic interference can be located using spatial and time-frequency characteristics. Simultaneously, filtering at specific frequencies can separate the power frequency noise and environmental electromagnetic interference signals. For example, the specific frequency of power frequency noise might be 50Hz or 60Hz, while the specific frequency of environmental electromagnetic interference is generally set based on the electromagnetic frequencies in the acquisition environment.

[0052] (2) The step noise is handled as follows:

[0053] For example, measuring step noise involves performing first-order differencing on each column of the source component matrix A, and determining the location of the step noise signal based on the absolute value of the difference. Separating step noise involves performing wavelet threshold filtering on the components corresponding to the step noise signal to isolate it. Step noise typically causes instantaneous amplitude changes in signals of certain channels; therefore, performing first-order differencing on each column of the source component matrix A and finding the position where the absolute value of the difference significantly exceeds that of other columns indicates the location of the step noise.

[0054] This is one possible implementation method for biological noise signals.

[0055] In this case, the biological noise signal includes at least one of the following: high-frequency noise signal, low-frequency noise signal, and periodic noise signal. Biological signals generally include physiological signals such as electromyography (EMG), electrooculography (EOG), electrokinetic electroencephalography (EMG), electrocardiography (ECG), and electroencephalography (EEG). Since EMG is a high-frequency signal, EMG and EMG are low-frequency signals, and ECG is a periodic signal, the types of biological noise signals included will differ depending on the selected biological signal used as the reference signal. If EMG is used as the reference signal, its biological noise signal includes at least one of the following: other high-frequency noise signals besides EMG, low-frequency noise signals, and periodic noise signals; and so on. For non-EEG reference signals, the corresponding signal can be directly obtained based on the measurement or separation method of a single biological noise signal. Alternatively, biological noise signals other than the reference signal can be removed using biological noise signal localization or separation methods to indirectly obtain the reference signal, achieving a denoising effect.

[0056] If EEG is used as the reference signal, high-frequency noise signals (e.g., but not limited to electromyography), low-frequency noise signals (e.g., but not limited to electrooculography and electrokinetic signals), and periodic noise signals (e.g., but not limited to electrocardiogram) are processed. The corresponding signal processing models are run simultaneously, randomly, or according to priority to remove any type of noise signal. For example, because high-frequency noise signals have a wide frequency and spatial distribution, they are more likely to cause abnormal identification of other noise signals, so high-frequency noise signals are processed first. Low-frequency noise signals have larger amplitudes and obvious spatial characteristics, making them relatively easy to locate, so they are prioritized after high-frequency noise signals. Finally, periodic noise signals are removed. Therefore, the priority order of biological noise signals can be: high-frequency noise signals > low-frequency noise signals > periodic noise signals. The specific processing of biological noise signals in EEG includes:

[0057] (1) The processing of high-frequency noise signals is as follows:

[0058] The measurement of high-frequency noise signals includes: calculating the energy characteristics of each column of the signal in the source component matrix A, and calculating the high-frequency to low-frequency energy ratio r of each column of the signal. hl If the high-frequency energy ratio r hl If the set energy ratio threshold r0 is set, the corresponding column signal is marked as a high-frequency noise signal sub-component, and the set of all high-frequency noise signal sub-component positions is set as the high-frequency noise signal position L. h Energy characteristics, for example, include the power spectral density psd(m) = pwelch(v). m The high-frequency energy ratio of each signal column:

[0059] Where m = 1, 2, ..., M, h1 and h2 represent the lower and upper bounds of the high-frequency band, respectively, and l1 and l2 represent the lower and upper bounds of the low-frequency band, respectively. If the high-frequency energy ratio of a certain column is r hl If r > 0, then this column of signals is first marked as a high-frequency noise signal sub-component, and the position of the high-frequency noise signal is L. h This is the set of locations of all high-frequency noise signal sub-components. The location L of the high-frequency noise signal is obtained. h Subsequently, the separation of high-frequency noise signals includes: separation of position L h The high-frequency interference components are filtered or subjected to CCA (canonical correlation analysis) filtering to separate the high-frequency noise signal.

[0060] (2) The processing of low-frequency noise signals is as follows:

[0061] The measurement of low-frequency noise signals includes: standardizing the source component matrix A to obtain the source component matrix B; performing peak count on each column of the signal in the source component matrix B to obtain the number of peaks in each column; and calculating the peak frequency p of each column of the signal per unit time. m1 If the peak frequency p m1 >If a frequency threshold p0 is set, the corresponding column signal is marked as a low-frequency noise signal sub-component, and the time position L of the low-frequency noise signal sub-component is recorded. v The spatial decomposition matrix U is obtained based on the noise signal time window, where U is an M×M matrix and M is the number of channels. Each column of the spatial decomposition matrix U is mapped and matched with a spatial location template to obtain matching coefficients. If the matching coefficients > a set matching threshold, the spatial location L of the low-frequency noise signal sub-component is output. u According to formula L l =merge(L v ,L u Determine the location L of the low-frequency noise signal. l .

[0062] It should be noted that the standardization of the source component matrix A is based on the sample mean μ. m =mean(v m ) and variance σ m =std(v m The process is performed, where m = 1, 2, ..., M, and M represents the number of channels. The formula for normalization is: K represents the number of time samples, A represents the source component matrix before standardization, and B represents the source component matrix after standardization. The standardized method has high generalization ability, and the fixed threshold adapts to different individuals. Peak detection is performed on each column of the source component matrix B, based on three features: peak height, half-peak width, and peak-to-peak distance. After obtaining the number of peaks in each column, the frequency p of the peaks occurring per unit time is calculated. m1 If the peak frequency p of a certain signal m1 If p > 0, then the signal column is considered to contain low-frequency noise, and the signal column is marked as a low-frequency noise signal sub-component. The time position L of the low-frequency noise signal sub-component is recorded. v Measurements of low-frequency noise signals require consideration of spatial location for final confirmation.

[0063] Separation of low-frequency noise signals includes: separation of position L lThe low-frequency noise signal sub-components are subjected to wavelet threshold filtering to separate the low-frequency noise signal. The wavelet threshold filtering process includes: wavelet decomposition of the low-frequency noise signal sub-components to obtain wavelet signals q; setting a threshold for each layer of wavelet signals; retaining the portion of each layer of wavelet signals exceeding the threshold (denoted as p, which is the noise signal); then subtracting the retained portion p from the wavelet signal q to obtain the clean wavelet signal; finally, reconstructing the clean wavelet signal to obtain the sub-components after removing the low-frequency noise signal.

[0064] like Figure 3 As shown, taking the electrooculogram (EOG) noise signal as an example, the temporal position L of the EOG noise signal sub-components is obtained. v After that, it is also necessary to obtain the spatial location L of the electrooculogram noise signal. u First, it is necessary to obtain spatial location templates related to electrooculography (EOG). These templates include blink templates, vertical eye movement templates, and horizontal eye movement templates. The blink templates, vertical eye movement templates, and horizontal eye movement templates can be obtained by performing spatial blink position detection, spatial vertical eye movement position detection, and spatial horizontal eye movement position detection on the signals, respectively. Each column u of the spatial decomposition matrix U... m The channel locations are mapped onto the brain topography space, and then matched with three spatial location templates to obtain three matching coefficients: BL(m), VE(m), and HE(m). Each matching coefficient has M values ​​(the same as the number of channels). A matching threshold is set for each matching coefficient, and the u values ​​corresponding to matching coefficients greater than the threshold are recorded. m The spatial location of the final output consists of three points: the spatial blink position L. u1 Vertical eye movement position L u2 Horizontal eye movement position L u3 Finally, according to formula L... l =merge(L v ,L u1 ,L u2 ,L u3 ) = L v ∩(L u1 ∪L u2 ∪L u3 The location of the final electrooculogram noise signal is determined. The method of determining the location of low-frequency noise signals by combining time and spatial domains can constrain the number of low-frequency noise signal locations and prevent signal loss due to excessive subsequent processing components.

[0065] (3) The processing of periodic noise signals is as follows:

[0066] The measurement of periodic noise signals includes: standardizing the source component matrix A to obtain the source component matrix C; performing peak counts on each column of the signal in the source component matrix C to obtain the number of peaks in each column; and calculating the peak frequency p of each column of the signal per unit time. m2 If the peak frequency p m2 >If a frequency threshold p0 is set, the corresponding column signal is marked as a periodic noise signal sub-component, and the position of the periodic noise signal sub-component is set as the position L of the periodic noise signal. z The measurement method for periodic noise signals is the same as that for measuring the time position of low-frequency noise signals, and will not be repeated here.

[0067] Separation of periodic noise signals includes: using template matching to separate the signal at position L. z The periodic interference sub-components are processed to separate the periodic noise signal. The periodic noise signal is, for example, an electrocardiogram (ECG) noise signal. The template matching method involves: detecting the periodic interference sub-components using a QRS wave detection algorithm to determine the R-wave position within the sub-component; extracting all QRS waves from the signal based on the R-wave position; iteratively clustering all QRS wave signals to obtain several periodic noise signal templates; then matching each periodic interference sub-component with one of the periodic noise signal templates; recording the periodic noise signal template X with the highest matching degree; and subtracting this periodic noise signal template X from the sub-component to obtain the clean sub-component.

[0068] The following comparison will be illustrated through specific examples.

[0069] Example 1: This method is used to remove noise signals from simulation data. The simulation data consists of electrooculography noise signals with different signal-to-noise ratios and electromyography artifacts.

[0070] Comparative Example 1: The existing MARA (Multiple Artifact Rejection Algorithm) method is used to remove noise from the simulation data.

[0071] Comparative Example 2: The existing ASR (Artifact Subspace Reconstruction) method is used to remove noise from the simulation data.

[0072] Example 2: This method is used to remove noise signals from real EEG data containing electrooculography (EOG) noise signals.

[0073] Comparative Example 3: Noise removal was performed on real EEG data using the existing MARA method.

[0074] Comparative Example 4: Noise removal was performed on real EEG data using existing ASR methods.

[0075] The evaluation indicators for the processing effect of simulation data are: the correlation coefficient and the relative root mean square error between the data after removing noise signals and the data without noise signals. The higher the correlation coefficient, the better the effect of removing artifacts. The smaller the relative root mean square error, the better the effect of removing noise signals.

[0076] The evaluation metric for the processing effect of real data is the waveform comparison between the raw EEG data and the EEG data after removing noise signals.

[0077] Figure 4 (a) is a comparison of the correlation coefficients of Example 1, Comparative Example 1, and Comparative Example 2 in the strong ocular electroacoustic channel. Figure 4 (b) is a comparison of the correlation coefficients of Example 1, Comparative Example 1, and Comparative Example 2 in the weak electrooculography channel. Figure 4 (c) is a comparison of the correlation coefficients of the electromyography channels in Example 1, Comparative Example 1, and Comparative Example 2. Figure 5 (a) is a comparison of the relative root mean square error of Example 1, Comparative Example 1, and Comparative Example 2 in the strong electrooculography channel. Figure 5 (b) is a comparison of the relative root mean square error of Example 1, Comparative Example 1, and Comparative Example 2 in the weak electrooculography channel. Figure 5 (c) is a comparison graph of the relative root mean square error of the electromyography channels in Example 1, Comparative Example 1, and Comparative Example 2. The numbers on the horizontal axis, "1" represents Example 1, "2" represents Comparative Example 1, and "3" represents Comparative Example 2. From... Figure 4 and Figure 5 The results show that the correlation coefficient of this method is higher than that of existing technologies, while the relative root mean square error is lower. This demonstrates that this method, when removing various types of noise signals, can not only remove different types of noise signals more accurately, but also reduce the loss of effective signal during noise removal.

[0078] Figure 6 The graph compares the waveforms of the original EEG signal, Example 2, Comparative Example 3, and Comparative Example 4. The horizontal axis represents time, and the vertical axis represents amplitude. As can be seen from the graph, the original signal contains multiple peaks (i.e., noise signals). After processing using this method and the two existing techniques, these peaks are filtered out. Compared with the MARA method, this method results in less loss of signal details and less impact on the effective signal when removing noise. Compared with the ASR method, the waveform processed by the ASR method still retains some noise signals, showing poor noise removal performance, while the waveform after noise removal by this method is more stable.

[0079] For example, signal time windows can be detected using time window detection methods, which include at least one of the following: generalized feature decomposition, multi-feature classification, and multi-feature joint detection. Time window detection methods are applicable to either primary or secondary signal time windows. When secondary signal time windows exist, they are typically marked using time window detection methods. Then, the classification of the primary signal time window is determined by jointly analyzing the marking results of multiple secondary signal time windows, classifying them as sparse noise signal time windows, dense noise signal time windows, or normal signal time windows.

[0080] like Figure 8 The generalized eigenvalue decomposition method shown includes: calculating the feature matrices of the signal time window and the background signal respectively; calculating the maximum generalized eigenvalue between the covariance matrix of the signal time window and the covariance matrix of the background signal; if the maximum generalized eigenvalue is greater than a set threshold, the signal time window is marked as a noise signal time window; otherwise, it is marked as a normal signal time window. It should be noted that the background signal is the EEG baseline or a full-band EEG signal. The generalized eigenvalue can reflect the degree of difference between the two matrices, and the category of the signal time window is identified by the magnitude of the generalized eigenvalue. For example, let R1 be the covariance matrix of the current signal time window, and R2 be the covariance matrix after high-pass filtering of the current signal time window. b R1 is the covariance matrix of the background signal. The largest generalized eigenvalue α1 between R2 and R1 represents the electromyographic noise signal intensity of the current signal window. R1 and R2 are related. b The maximum generalized eigenvalue α2 represents the noise signal intensity of the current signal window (regardless of noise signal type). If α1 is greater than a set threshold, the signal window is marked as an electromyographic noise signal window; otherwise, it is a normal signal window. If α2 is greater than a set threshold, the signal window is marked as a noise signal window; otherwise, it is a normal signal window.

[0081] For example, Figure 11 This is a diagram showing the denoising effect after identifying the noise signal using the generalized feature decomposition method (processing only the noise signal portion). Figure 12 This is a diagram showing the effect of directly denoising the entire signal using existing technology. (Comparison) Figure 11 and Figure 12 It can be seen that both the proposed method and the prior art suppress noise signals; however, compared with the prior art, the proposed method results in less loss of the effective portion of the signal.

[0082] like Figure 9As shown, the multi-feature joint detection method includes: calculating multiple feature values ​​of a signal time window; performing threshold judgment on each feature value; and marking the signal time window as a noise signal time window when all feature values ​​of the signal time window meet the threshold conditions. For example, feature values ​​can be line length, zero-crossing rate, mean, etc. K feature values ​​of the signal time window are calculated, and each feature value is judged one by one, with each feature value corresponding to a set threshold. When all K feature values ​​meet their respective set thresholds, the signal time window is judged as a noise signal time window; otherwise, it is a normal signal time window. For example, if a signal time window calculates two feature values, line length and zero-crossing rate, if the line length feature meets the set threshold, the signal time window is first marked as a candidate interference; if the zero-crossing rate feature also meets the set threshold, the signal time window is finally marked as a noise signal time window. This can improve the accuracy of noise signal judgment and prevent false judgments.

[0083] like Figure 13 As shown, the figure compares the detection results using a multi-feature joint detection method and a single-feature detection method. The horizontal axis represents the time window, and the vertical axis represents the channels, arranged from top to bottom. Figure 13 The first small image in the diagram represents the original signal. The areas selected by the black box (from left to right) represent electromyographic noise signals, normal EEG, epileptic EEG, and electromyographic noise signals, respectively. The second small image shows the retrieval results obtained using only the line length feature. The third small image shows the detection results obtained by combining the line length and zero-crossing rate features. The white highlights in the second and third small images indicate the locations identified as electromyographic noise signals. This demonstrates that the original signal contains electromyographic noise signals, normal signals, and epileptic signals. Using a single feature detection method, useful epileptic signals are also classified as noise signals, leading to inaccurate subsequent epilepsy warnings. Our proposed method, however, can accurately identify electromyographic noise signals and will not misclassify epileptic signals as electromyographic signals.

[0084] like Figure 11 As shown, the multi-feature classification method includes: calculating multiple feature values ​​for the signal time window, inputting these feature values ​​into a classifier for classification, and outputting noise signal time windows and normal signal time windows. Feature values ​​include, for example, line length, zero-crossing rate, and mean value. After calculating the feature values ​​for the signal time window, these feature values ​​are input into the classifier, which can output the classification result (noise signal or normal signal). By classifying the data from each of the M channels within the signal time window, the spatiotemporal domain classification result of the signal time window can be obtained.

[0085] like Figure 7As shown, the noise signal categories in the measurement noise signal time window also include: The noise signal time window is divided into a sparse noise signal time window and a dense noise signal time window according to the spatio-temporal sparsity of the noise signal. The division of the signal time window includes at least two levels of division, that is, a first-level signal time window is composed of multiple second-level signal time windows. The classification detection of the signal time window also includes: jointly judging the first-level signal time window based on the detection results of the second-level signal time windows, so as to divide the first-level signal time window into a sparse noise signal time window, a dense noise signal time window, and a normal signal time window. Jointly judging the first-level signal time window based on the detection results of the second-level signal time windows includes: detecting the second-level signal time windows, and dividing them into second-level interference time windows and second-level normal time windows; calculating the proportion p of the second-level interference time windows in the first-level signal time window, that is, when the proportion p = 0, marking the first-level signal time window as a normal signal time window; when 0 < p < the proportion threshold, marking the first-level signal time window as a sparse noise signal time window; when the proportion p ≥ the proportion threshold, marking the first-level signal time window as a dense noise signal time window.

[0086] In other words, in addition to classifying the noise signal according to the type of the noise signal, the present invention can also classify the noise signal time window according to the sparsity of the noise signal. First, the signal time window is divided into two levels, namely the second-level signal time window and the first-level signal time window. The first-level signal time window is composed of multiple second-level signal time windows. That is, the window width of the first-level signal time window is greater than that of the second-level signal time window, and the amount of data contained in the first-level signal time window is also more than that of the second-level signal time window. Therefore, when the present invention performs data processing, it first detects the second-level signal time windows to obtain the detection results of each second-level signal time window (divided into second-level interference time windows and second-level normal time windows). In this way, it can also reflect how many second-level interference time windows are included in the first-level signal time window, and the proportion p = the number of second-level interference time windows / the number of second-level signal time windows. If p = 0, it means that there is no second-level interference time window, and this first-level signal time window is a normal signal time window and does not need further processing. If 0 < p < the proportion threshold, it means that there are a small number of second-level interference time windows, and this first-level signal time window is a sparse noise signal time window. At this time, when removing the noise signal, only the second-level interference time windows need to be processed, and there is no need to process the entire first-level signal time window. If the proportion p ≥ the proportion threshold, it means that there are more second-level interference time windows, and this first-level signal time window is a dense noise signal time window. At this time, when removing the noise signal, the entire first-level signal time window needs to be processed.

[0087] When the primary signal window is a sparse noise signal window, different signal processing models are constructed only for the secondary interference windows within the primary signal window. When the primary signal window is a dense noise signal window, different signal processing models are constructed for the primary signal window. That is, for the primary signal window of a "sparse noise signal," this method only processes the secondary interference windows. This avoids processing non-noise signal segments, preventing the loss of effective signals, and also reduces the amount of data to be processed, improving processing efficiency. For the primary signal window of a "dense noise signal," processing the entire primary signal window is performed, resulting in higher computational efficiency and preventing missed processing.

[0088] It should be noted that the thresholds set in this case can be set according to needs, historical data, or technical experience, and are not specifically limited in this case.

[0089] The signal noise detection method of this invention employs a combined spatiotemporal approach to measure and separate multiple types of noise signals, thereby improving the accuracy of noise removal and reducing the loss of valid signals. Furthermore, by using different measurement and separation methods for different types of noise signals, this method is more targeted and yields more accurate identification results compared to using a single processing method.

[0090] This invention also provides a signal detection model for a signal noise detection method, comprising: a signal time window segmentation model for segmenting signal time windows; a signal time window detection model for classifying and detecting signal time windows to obtain noise signal time windows and normal signal time windows; a classification model for measuring the noise signal category within the noise signal time window; and a signal processing model for separating different categories of noise signals based on their categories. For a description of the parts related to the signal noise detection method, please refer to the section on signal noise detection method; it will not be repeated here.

[0091] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the aforementioned signal and noise detection method. The computer-readable storage medium may be located at at least one of multiple network servers in a computer network. The aforementioned storage medium may include, but is not limited to, various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0092] This invention also provides a signal time window detection model based on the generalized feature decomposition method, comprising: calculating the feature matrices of the signal time window and the background signal respectively; calculating the maximum generalized eigenvalue between the feature matrix of the signal time window and the feature matrix of the background signal; if the maximum generalized eigenvalue is greater than a set threshold, then the signal time window is marked as a noise signal time window; otherwise, it is marked as a normal signal time window. For a detailed explanation, please refer to the relevant section on signal noise detection methods, which will not be repeated here.

[0093] This invention also provides a signal time window detection model based on a multi-feature classification method, comprising: calculating multiple feature values ​​of a signal time window; performing threshold judgment on each of the multiple feature values; and marking the signal time window as a noise signal time window when all multiple feature values ​​of the signal time window meet the threshold condition. For a detailed explanation, please refer to the relevant section on signal noise detection methods, which will not be repeated here.

[0094] This invention also provides a signal time window detection model based on a multi-feature joint detection method, comprising: calculating multiple feature values ​​of the signal time window, inputting the feature values ​​into a classifier for classification, and outputting noise signal time windows and normal signal time windows. For a detailed explanation, please refer to the relevant section on signal and noise detection methods, which will not be repeated here.

[0095] This invention also provides a high-frequency noise signal detection model, including: dividing the signal time window; classifying the signal time window into noise signal time windows and normal signal time windows; and constructing a signal processing model for the noise signal time windows to separate high-frequency noise signals. The signal processing model includes: obtaining a time decomposition matrix V based on the noise signal time window, where V is a K×K matrix and K is the number of time-domain sampling points; extracting the first M columns from the time decomposition matrix V to form a source component matrix A, where M is the number of channels; locating the position of the high-frequency noise signal in the source component matrix A; determining whether the position of the high-frequency noise signal in the source component matrix A is null; when the position of the high-frequency noise signal is not null, sequentially separating the high-frequency noise signal and reconstructing the source to continue iterating; when the position of the high-frequency noise signal is null, terminating the iteration. Locating the position of the high-frequency noise signal in the source component matrix A includes: calculating the energy characteristics of each column of the signal in the source component matrix A; calculating the high-low frequency energy ratio r of each column of the signal. hl If the high-frequency energy ratio r hl If the set energy ratio threshold r0 is set, the corresponding column signal is marked as a high-frequency noise signal sub-component, and the set of all high-frequency noise signal sub-component positions is set as the high-frequency noise signal position L. h Separating high-frequency noise signals includes: for position L h The high-frequency interference components are subjected to low-pass filtering or CCA filtering to separate the high-frequency noise signal. For details, please refer to the relevant section on signal and noise detection methods, which will not be repeated here.

[0096] This invention also provides a low-frequency noise signal detection model, comprising: dividing the signal time window; classifying and detecting the signal time window into noise signal time windows and normal signal time windows; and constructing a signal processing model for the noise signal time windows to separate low-frequency noise signals. The signal processing model includes: obtaining a spatial decomposition matrix U and a temporal decomposition matrix V based on the noise signal time window, wherein the spatial decomposition matrix U is an M×M matrix, M is the number of channels, and the temporal decomposition matrix V is a K×K matrix, K is the number of time-domain sampling points; extracting the first M columns from the temporal decomposition matrix V to form a source component matrix A, wherein M is the number of channels; locating the position of the low-frequency noise signal in the source component matrix A; determining whether the position of the low-frequency noise signal in the source component matrix A is a null value; when the position of the low-frequency noise signal is not a null value, sequentially performing low-frequency noise signal separation and source reconstruction to continue iterating; when the position of the low-frequency noise signal is a null value, terminating the iteration. The location of low-frequency noise signals in source component matrix A includes: standardizing source component matrix A to obtain source component matrix B; performing peak counts on each column of signals in source component matrix B to obtain the number of peaks in each column; and calculating the peak frequency p of each column of signals per unit time. m1 If the peak frequency p m1 >If a frequency threshold p0 is set, the corresponding column signal is marked as a low-frequency noise signal sub-component, and the time position L of the low-frequency noise signal sub-component is recorded. v The signal in each column of the spatial decomposition matrix U is mapped and then matched using the remaining spatial position template to obtain the matching coefficients. If the matching coefficients are greater than the set matching threshold, the spatial position L of the low-frequency noise signal sub-component is output. u According to formula L l =merge(L v ,L u Determine the location L of the low-frequency noise signal. l Separating low-frequency noise signals includes: for position L l The low-frequency interference components are subjected to wavelet threshold filtering to separate the low-frequency noise signal. For details, please refer to the relevant section on signal and noise detection methods, which will not be repeated here.

[0097] This invention also provides a periodic noise signal detection model, including: dividing the signal into time windows; classifying the signal time windows into noise signal time windows and normal signal time windows; and constructing a signal processing model for the noise signal time windows to separate periodic noise signals. The signal processing model includes: obtaining a time decomposition matrix V based on the noise signal time windows, where V is a K×K matrix and K is the number of time-domain sampling points; extracting the first M columns from the time decomposition matrix V to form a source component matrix A, where M is the number of channels; locating the position of the periodic noise signal in the source component matrix A; determining whether the position of the periodic noise signal in the source component matrix A is null; when the position of the periodic noise signal is not null, sequentially separating the periodic noise signal and reconstructing the source to continue iterating; when the position of the periodic noise signal is null, terminating the iteration. Locating the position of the periodic noise signal in the source component matrix A includes: standardizing the source component matrix A to obtain a source component matrix C; performing peak value statistics on each column of signals in the source component matrix C to obtain the number of peak values ​​in each column; and calculating the peak frequency p of each column of signals per unit time. m2 If the peak frequency p m2 >If a frequency threshold p0 is set, the corresponding column signal is marked as a periodic noise signal sub-component, and the position of the periodic noise signal sub-component is set as the position L of the periodic noise signal. z Separation of periodic noise signals includes: using template matching to separate the noise signal at position L. z The periodic interference components are processed to separate the periodic noise signal. For details, please refer to the relevant section on signal and noise detection methods, which will not be repeated here.

[0098] The present invention also provides a multi-level detection model for signal time windows, including: at least two levels of dividing signal time windows, that is, a first-level signal time window is composed of multiple second-level signal time windows; using the time window detection method to detect the second-level signal time windows, which are divided into second-level interference time windows and second-level normal time windows; based on the detection results of the second-level signal time windows, jointly judge the first-level signal time windows to divide the first-level signal time windows into sparse noise signal time windows, dense noise signal time windows, and normal signal time windows. The time window detection method is at least one of the generalized eigenvalue decomposition algorithm, multi-feature classification algorithm, and multi-feature joint detection algorithm. The generalized eigenvalue decomposition method includes: respectively calculating the feature matrices of the signal time window and the background signal; calculating the maximum generalized eigenvalue between the feature matrix of the signal time window and the feature matrix of the background signal; if the maximum generalized eigenvalue > the set threshold, mark the signal time window as a noise signal time window; otherwise, mark it as a normal signal time window. The multi-feature joint detection method includes: calculating multiple eigenvalues of the signal time window; performing threshold judgment on each of the multiple eigenvalues one by one, and when all the multiple eigenvalues of the signal time window meet the threshold conditions, mark the signal time window as a noise signal time window. The multi-feature classification method includes: calculating multiple eigenvalues of the signal time window and inputting the eigenvalues into a classifier for classification, and outputting noise signal time windows and normal signal time windows. Jointly judging the first-level signal time windows based on the detection results of the second-level signal time windows includes: calculating the proportion p of the second-level interference time windows in the first-level signal time window; when the proportion p = 0, mark the first-level signal time window as a normal signal time window; when 0 < p < the proportion threshold, mark the first-level signal time window as a sparse noise signal time window; when the proportion p ≥ the proportion threshold, mark the first-level signal time window as a dense noise signal time window. For specific elaboration, please refer to the relevant part of the signal noise detection method, which will not be elaborated here.

[0099] In summary, the signal noise detection method, signal detection model, and readable storage medium of the present invention use the spatio-temporal domain joint method to measure and separate multiple types of noise signals from signals, which can improve the accuracy of noise signal removal and reduce the loss of effective signals. Moreover, for different types of noise signals, different measurement methods and separation methods are adopted. Compared with using the same processing method, this method is more targeted and the recognition result will be more accurate.

[0100] Inspired by the above ideal embodiments according to the present invention, through the above description, relevant staff can make various changes and modifications completely within the scope of not deviating from the technical idea of this invention. The technical scope of this invention is not limited to the content in the specification, and its technical scope must be determined according to the scope of the claims.

Claims

1. A multi-level detection model for signal time windows, characterized in that, It includes: A signal time window division model that divides the signal time window into at least two levels, that is, a first-level signal time window is composed of multiple second-level signal time windows; A signal time window detection model that divides the second-level signal time window into a second-level interference time window and a second-level normal time window by the time window detection method; A classification model that jointly judges the first-level signal time window based on the detection results of the second-level signal time window to classify the first-level signal time window into a sparse noise signal time window, a dense noise signal time window, and a normal signal time window; When the first-level signal time window is a sparse noise signal time window, only a different signal processing model is constructed for the second-level interference time window in the first-level signal time window; When the first-level signal time window is a dense noise signal time window, a different signal processing model is constructed for the first-level signal time window.

2. The multi-level detection model according to claim 1, wherein The time window detection method is at least one of a generalized eigenvalue decomposition algorithm, a multi-feature classification algorithm, and a multi-feature joint detection algorithm.

3. The multi-level detection model according to claim 2, wherein The generalized eigenvalue decomposition method includes: calculating the feature matrices of the signal time window and the background signal respectively; Calculating the maximum generalized eigenvalue between the feature matrix of the signal time window and the feature matrix of the background signal; If the maximum generalized eigenvalue > the set threshold, mark the signal time window as a noise signal time window; otherwise, mark it as a normal signal time window.

4. The multi-level detection model according to claim 2, wherein The multi-feature joint detection method includes: Calculating multiple eigenvalues of the signal time window; Performing threshold judgment on each of the multiple eigenvalues one by one. When all the multiple eigenvalues of the signal time window meet the threshold conditions, mark the signal time window as a noise signal time window.

5. The multi-level detection model according to claim 2, wherein The multi-feature classification method includes: Calculating multiple eigenvalues of the signal time window and inputting the eigenvalues into a classifier for classification, and outputting a noise signal time window and a normal signal time window.

6. The multi-level detection model according to claim 1, wherein The joint judgment of the first-level signal time window based on the detection results of the second-level signal time window includes: Calculating the proportion p of the second-level interference time window in the first-level signal time window; When the proportion p = 0, mark the first-level signal time window as a normal signal time window; When 0 < p < the proportion threshold, mark the first-level signal time window as a sparse noise signal time window; When the proportion p ≥ the proportion threshold, mark the first-level signal time window as a dense noise signal time window.

7. The multi-level detection model according to claim 1, wherein The window width of the first-level signal time window is greater than the window width of the second-level signal time window.

Images