Signal filtering method, device and equipment of ultra-high lead electroencephalogram machine and storage medium

Abstract

The application relates to the technical field of electroencephalogram signal processing, and discloses a super-high-lead electroencephalogram machine signal filtering method, device, equipment and storage medium.The method comprises the following steps: acquiring super-high-lead electroencephalogram signal data collected by an electroencephalogram machine system; processing the super-high-lead electroencephalogram signal data to obtain a reference signal and an input signal; synchronously inputting the reference signal and the input signal into an RBF neural network model for filtering; and optimizing parameters of the RBF neural network model according to an adaptive filtering algorithm to output filtered electroencephalogram signal data.In the application, the adaptive filtering algorithm is combined with the filtering technology of the RBF neural network to filter the massive data of the super-high-lead electroencephalogram machine; the adaptive filtering algorithm and the RBF neural network are complementary to each other; the overall performance of electroencephalogram signal processing can be improved; in particular, the adaptive filtering algorithm and the RBF neural network can further effectively filter high-throughput signals in electroencephalogram signals which are prone to be disturbed by noise environments, and the signal quality can be enhanced.

Description

Technical Field

[0001] This invention relates to the field of electroencephalogram (EEG) signal processing technology, and in particular to a signal filtering method, apparatus, device, and storage medium for ultra-high-lead electroencephalogram (EEG) machines. Background Technology

[0002] Current common filtering methods for EEG signals include median filtering, adaptive filtering, Butterworth filtering, wavelet transform, blind source separation algorithms, and deep learning algorithms. Collected data is cleaned using these methods to obtain a pure dataset. However, existing technologies suffer from limited feature extraction or pattern recognition performance after data filtering, and data preprocessing filtering accounts for a significant portion of the reason why the final recognition effect cannot be significantly improved. Furthermore, for massive datasets with ultra-high leads, existing filtering methods cannot simultaneously balance processing speed and signal quality, facing problems such as high computational complexity, slow processing speed, and the inability to achieve rapid convergence to obtain signals with lower noise interference. Summary of the Invention

[0003] The main objective of this invention is to provide a method, apparatus, device, and storage medium for filtering signals from ultra-high-lead electroencephalogram (EEG) machines, aiming to solve the problems of poor recognition effect, high computational complexity, and slow processing speed of existing commonly used EEG signal filtering methods for massive ultra-high-lead data.

[0004] To achieve the above objectives, the present invention provides a method for filtering signals from an ultra-high-lead electroencephalogram (EEG) machine, comprising:

[0005] Acquire ultra-high-lead EEG signal data from the EEG system;

[0006] The ultra-high-lead EEG signal data is processed to obtain a reference signal and an input signal;

[0007] The reference signal and the input signal are synchronously input into the RBF neural network model for filtering;

[0008] The parameters of the RBF neural network model are optimized using an adaptive filtering algorithm to output filtered EEG signal data.

[0009] In some embodiments, processing the ultra-high-lead EEG signal data to obtain a reference signal and an input signal includes:

[0010] Based on the ultra-high-conductance EEG signal data, EEG background activity signals and evoked potentials were determined;

[0011] The EEG background activity electrical signal is subjected to Butterworth bandpass filtering to obtain a noise-related signal, which is then used as a reference signal.

[0012] The evoked potentials are filtered based on Mallat wavelet transform to obtain an initial denoised signal, which is then used as the input signal.

[0013] In some embodiments, filtering the evoked potential based on Mallat wavelet transform to obtain an initial denoised signal and using the initial denoised signal as the input signal includes:

[0014] Based on Mallat wavelet transform, the evoked potentials are initially filtered using low-pass and high-pass filters to obtain denoised EEG signals.

[0015] The denoised EEG signal is reconstructed using a low-pass reconstruction filter and a high-pass reconstruction filter to obtain an initial denoised signal, which is then used as the input signal.

[0016] In some embodiments, optimizing the parameters of the RBF neural network model according to an adaptive filtering algorithm to output filtered EEG signal data includes:

[0017] The parameters of the RBF neural network model are optimized based on the normalized least mean square filtering algorithm and the affine projection filtering algorithm.

[0018] When the number of iterations of the RBF neural network model reaches a preset number of iterations, a response signal is output based on the RBF neural network model, and an error signal is obtained based on the response signal.

[0019] When the error signal meets the preset error threshold, the filtered EEG signal data is output.

[0020] In some embodiments, optimizing the parameters of the RBF neural network model according to the normalized least mean square filtering algorithm and the affine projection filtering algorithm includes:

[0021] An NLMS-AP adaptive filtering fusion algorithm is constructed based on the normalized least mean square filtering algorithm and the affine projection filtering algorithm; wherein, the fusion factor in the NLMS-AP adaptive filtering fusion algorithm includes stochastic gradient descent, and the fusion factor is updated through the expression of the activation function sigmoid;

[0022] The parameters of the RBF neural network model are adjusted online according to the NLMS-AP adaptive filtering fusion algorithm.

[0023] In some embodiments, the step of outputting a response signal based on the RBF neural network model and obtaining an error signal based on the response signal includes:

[0024] Based on the RBF neural network model, the input signal is activated and the output response signal is generated using m delayed sample values ​​and j hidden neuron Gaussian functions.

[0025] The response signal is compared with the expected signal to obtain an error signal; wherein the expected signal is the reference signal, and the error signal is the difference between the reference signal and the response signal.

[0026] In some embodiments, the method further includes:

[0027] The collected ultra-high-conductance EEG signal data is divided into multiple small blocks;

[0028] The data corresponding to each of the aforementioned small blocks is assigned to different GPU threads;

[0029] The steps of the ultra-high-lead electroencephalogram (EEG) signal filtering method are performed by processing the data corresponding to each small block in parallel using GPU or CUDA.

[0030] Furthermore, to achieve the above objectives, the present invention also proposes a signal filtering device for an ultra-high-lead electroencephalogram (EEG) machine, comprising:

[0031] The signal acquisition module is used to acquire ultra-high-lead EEG signal data collected by the EEG machine system;

[0032] The wavelet transform module is used to process the ultra-high-lead EEG signal data to obtain a reference signal and an input signal;

[0033] The signal filtering module is used to synchronously input the reference signal and the input signal into the RBF neural network model for filtering;

[0034] An adaptive optimization module is used to optimize the parameters of the RBF neural network model according to an adaptive filtering algorithm, so as to output filtered EEG signal data.

[0035] Furthermore, to achieve the above objectives, the present invention also proposes an electronic device comprising: a memory, a processor, and an ultra-high-lead electroencephalogram (EEG) signal filtering program stored in the memory and executable on the processor, wherein the ultra-high-lead EEG signal filtering program is configured to implement the ultra-high-lead EEG signal filtering method described above.

[0036] Furthermore, to achieve the above objectives, the present invention also proposes a storage medium storing a high-lead EEG signal filtering program, which is used to enable the processor to implement the high-lead EEG signal filtering method as described above when executed.

[0037] This invention provides a signal filtering method for ultra-high-lead electroencephalography (EEG) machines, comprising: acquiring ultra-high-lead EEG signal data collected by an EEG machine system; processing the ultra-high-lead EEG signal data to obtain a reference signal and an input signal; synchronously inputting the reference signal and the input signal into an RBF neural network model for filtering; and optimizing the parameters of the RBF neural network model according to an adaptive filtering algorithm to output filtered EEG signal data. This invention employs an adaptive filtering algorithm combined with RBF neural network filtering technology to filter massive amounts of data from ultra-high-lead EEG machines. The adaptive filtering algorithm and RBF neural network complement each other, helping to improve the overall performance of EEG signal processing, especially in EEG signals susceptible to noise interference, further effectively filtering high-throughput signals and enhancing signal quality. This solves the problems of poor recognition effect, high computational complexity, and slow processing speed of existing commonly used EEG signal filtering methods for massive ultra-high-lead data. Attached Figure Description

[0038] Figure 1 This is a schematic diagram of the structure of an electronic device in the hardware operating environment involved in the embodiments of the present invention;

[0039] Figure 2 This is a flowchart illustrating an embodiment of the high-lead electroencephalogram (EEG) signal filtering method of the present invention;

[0040] Figure 3 This is a flowchart of the high-performance parallel computing processing filtering method involved in the embodiments of the present invention;

[0041] Figure 4 The flowchart shows the filtering method based on Mallat wavelet and NLMS-AP adaptive filtering combined with RBFNN neural network algorithm involved in the embodiment of the present invention.

[0042] Figure 5 This is a schematic diagram of the Mallat wavelet decomposition process involved in the embodiment of the present invention;

[0043] Figure 6 This is a schematic diagram of the Mallat wavelet reconstruction process involved in the embodiment of the present invention;

[0044] Figure 7 This is a schematic diagram of the NLMS adaptive filter principle structure involved in the embodiment of the present invention;

[0045] Figure 8 This is a diagram of the adaptive filtering structure of the AP-NLMS algorithm involved in the embodiments of the present invention;

[0046] Figure 9 This is a diagram of the RBFNN network structure involved in the embodiments of the present invention;

[0047] Figure 10 This is a framework diagram of the filtering method combining the NLMS-AP adaptive filtering algorithm and the RBFNN neural network involved in the embodiments of the present invention;

[0048] Figure 11 This is a structural block diagram of an embodiment of the signal filtering device for an ultra-high-lead electroencephalogram (EEG) machine according to the present invention.

[0049] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0050] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0051] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indication will also change accordingly.

[0052] Furthermore, the use of terms such as "first" and "second" in this invention is for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of that feature. Additionally, the technical solutions of the various embodiments can be combined with each other, but only on the basis of being achievable by those skilled in the art. If the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by this invention. It should be understood that the specific embodiments described herein are only for explaining the invention and are not intended to limit the invention.

[0053] Reference Figure 1 , Figure 1 This is a schematic diagram of the electronic device structure of the hardware operating environment involved in the embodiments of the present invention.

[0054] like Figure 1As shown, the electronic device may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display screen or an input unit such as a keyboard; optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wireless-Fidelity (Wi-Fi) interface). The memory 1005 may be high-speed random access memory (RAM) or stable non-volatile memory (NVM), such as a disk drive. The memory 1005 may also optionally be a storage device independent of the aforementioned processor 1001.

[0055] Those skilled in the art will understand that Figure 1 The structure shown does not constitute a limitation on the electronic device and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0056] like Figure 1 As shown, the memory 1005, which serves as a storage medium, may include an operating system, a network communication module, a user interface module, and a high-lead electroencephalogram (EEG) signal filtering program.

[0057] exist Figure 1 In the illustrated electronic device, the network interface 1004 is mainly used for data communication with the network server; the user interface 1003 is mainly used for data interaction with the user; the processor 1001 and the memory 1005 in the electronic device of the present invention can be installed in the electronic device, and the electronic device calls the ultra-high-lead electroencephalogram (EEG) signal filtering program stored in the memory 1005 through the processor 1001, and executes the ultra-high-lead EEG signal filtering method provided in the embodiment of the present invention.

[0058] This invention proposes a method, device, equipment, and storage medium for filtering signals from an ultra-high-lead electroencephalogram (EEG) machine.

[0059] This invention provides a method for filtering signals from an ultra-high-lead electroencephalogram (EEG) machine, referring to... Figure 2 , Figure 2 This is a flowchart illustrating an embodiment of the high-lead electroencephalography (EEG) signal filtering method of the present invention.

[0060] like Figure 2As shown, the ultra-high-lead electroencephalogram (EEG) signal filtering method includes:

[0061] Step S100: Acquire EEG signal data from the ultra-high-lead EEG system;

[0062] Step S200: Process the ultra-high-lead EEG signal data to obtain a reference signal and an input signal;

[0063] Step S300: Synchronously input the reference signal and the input signal into the RBF neural network model for filtering;

[0064] Step S400: Optimize the parameters of the RBF neural network model according to the adaptive filtering algorithm to output filtered EEG signal data.

[0065] It should be noted that the execution subject in this embodiment can be an electronic device, which can be a computer device with data processing function, or other devices that can achieve the same or similar functions. This embodiment does not limit this. In this embodiment, a computer device is used as an example for explanation.

[0066] In one embodiment, the method further includes: dividing the acquired ultra-high-lead EEG signal data into multiple small blocks; allocating the data corresponding to each small block to different GPU threads; performing parallel processing on the data corresponding to each small block according to GPU or CUDA, and executing the ultra-high-lead EEG signal filtering method.

[0067] like Figure 3 As shown, ultra-high-lead EEG data (i.e., ultra-high-lead EEG signal data) is acquired through an electroencephalography (EEG) system. The ultra-high-lead EEG data is then processed by a high-performance parallel computing filter to output filtered EEG data. It can be understood that the high-performance parallel computing filter utilizes GPU / CUDA for high-performance parallel computing during data processing. The ultra-high-lead data is parallelized and divided into multiple smaller blocks (e.g., ...). Figure 3 X shown EEG1 X EEG2 , ..., X EEGN Each small block can be processed independently, and multiple small blocks of data can be allocated to different GPU threads. During parallel computing, memory is effectively managed on the GPU, with reasonable allocation of global and shared memory. Considering the use of CUDA kernel functions, the number of threads and blocks needs to be configured appropriately based on the GPU architecture and data characteristics; this embodiment does not impose any limitations on this.

[0068] For example, refer to Figure 3The steps for implementing the ultra-high-lead EEG signal filtering method (e.g., steps S200 to S400) are as follows: the high-performance parallel computing processing filter uses Mallat wavelet transform for initial simple filtering, the reconstructed signal is then subjected to adaptive filtering through an adaptive filter, and finally the filtered EEG data is output through the RBFNN algorithm.

[0069] It should be noted that in this embodiment, the technique of using Mallat wavelet transform combined with adaptive filtering and neural network algorithm to reduce noise is intended to effectively improve the quality of EEG signals and the processing speed. The following describes in detail the steps of the ultra-high-lead EEG signal filtering method based on Mallat wavelet and NLMS-AP adaptive filtering combined with RBFNN neural network algorithm.

[0070] In one embodiment, the ultra-high-lead EEG signal data is processed to obtain a reference signal and an input signal, including: determining the EEG background activity signal and evoked potentials based on the ultra-high-lead EEG signal data; performing Butterworth bandpass filtering on the EEG background activity signal to obtain a noise-related signal and using the noise-related signal as the reference signal; filtering the evoked potentials based on Mallat wavelet transform to obtain an initial denoised signal and using the initial denoised signal as the input signal.

[0071] Specifically, the background electrical activity signal of EEG is spontaneous ultra-high-lead EEG signal data, and the evoked potentials (EPs) in the EEG signal can be ultra-high-lead EEG signal data collected through the motor imagery experimental paradigm. For example... Figure 4 As shown, the EEG background active electrical signal and evoked potentials (EPs) are processed separately. The EEG background active electrical signal is subjected to Butterworth bandpass filtering to obtain a noise-correlated signal, which is used as the reference signal. The evoked potentials are processed based on Mallat wavelet decomposition and reconstruction to obtain a noisy useful signal (i.e., the initial denoised signal), which is used as the input signal.

[0072] In one embodiment, filtering the evoked potentials based on Mallat wavelet transform to obtain an initial denoised signal and using the initial denoised signal as the input signal includes: initially filtering the evoked potentials based on Mallat wavelet transform using a low-pass filter and a high-pass filter to obtain a denoised EEG signal; reconstructing the denoised EEG signal using a low-pass reconstruction filter and a high-pass reconstruction filter to obtain the initial denoised signal and using the initial denoised signal as the input signal.

[0073] Specifically, in this embodiment, Mallat wavelet transform is used to perform preliminary filtering on the original data (EPs in EEG signals). Mallat wavelet transform, also known as multi-resolution analysis, decomposes a signal into components of different frequencies while preserving time information. The core is to perform multi-scale decomposition of the signal by iteratively using low-pass and high-pass filters. It mainly includes the following parts: (1) using low-pass filter H and high-pass filter G to extract the approximate (coarse) part and the detailed (fine) part, respectively; (2) according to the target frequency range of EEG, after filtering out the high-frequency signal, the low-pass reconstruction filter h and high-pass reconstruction filter g are used to reconstruct the processed signal, where H and h correspond to the scaling function, and G and g correspond to the wavelet function, as shown in the following formula:

[0074] A m [f(t)]=∑H(2t-n)A m-1 [f(t)]

[0075] D m [f(t)]=∑ n G(2t-n)A m-1 [f(t)]

[0076] Among them, A m D represents the near-form coefficients at scale m. m This represents the detail coefficient at scale m.

[0077] Specifically, the Mallat wavelet decomposition process and the reconstruction process are as follows: Figure 5 and Figure 6 As shown: Figure 5 The Mallat wavelet decomposition process involves decomposing the signal into waves to remove high-frequency noise and preliminarily selecting the approximate and detail frequency components of the target EEG signal, thus obtaining the denoised EEG signal. Figure 6 The Mallat wavelet reconstruction process reconstructs the denoised EEG signal to obtain a preliminarily denoised signal, which is then used as the input signal.

[0078] In one embodiment, optimizing the parameters of the RBF neural network model according to an adaptive filtering algorithm to output filtered EEG signal data includes: optimizing the parameters of the RBF neural network model according to a normalized least mean square filtering algorithm and an affine projection filtering algorithm; when the number of iterations of the RBF neural network model reaches a preset number of iterations, outputting a response signal based on the RBF neural network model and obtaining an error signal based on the response signal; and when the error signal meets a preset error threshold, outputting the filtered EEG signal data.

[0079] The optimization of the RBF neural network model parameters based on the Normalized Least Mean Square (NMS) filtering algorithm and the Affine Projection (AP) filtering algorithm includes: constructing an NLMS-AP adaptive filtering fusion algorithm based on the NMS-AP and AP algorithms; wherein the fusion factor in the NLMS-AP adaptive filtering fusion algorithm includes stochastic gradient descent, and the fusion factor is updated through the sigmoid expression of the activation function; and adjusting the parameters of the RBF neural network model online according to the NLMS-AP adaptive filtering fusion algorithm.

[0080] It is understood that in this embodiment, adaptive filtering includes the Normalized Least Mean Square (NLMS) filtering algorithm, which has a slow convergence speed but high data quality, and the Fast Convergence Affine Projection (AP) filtering algorithm, which has a fast convergence speed but relatively low signal quality. The two algorithms, NLMS and AP, can complement each other and, combined with the gradient descent idea, can balance convergence speed and data quality. This embodiment improves the performance of the filter (high-performance parallel computing processing filter) by improving the core algorithm of the adaptive filter.

[0081] Specifically, the NLMS algorithm principle structure is as follows: The NLMS adaptive filter is an improvement on the LMS filter. The NLMS filtering algorithm normalizes the step size based on the original LMS, which can adaptively adjust to a suitable size according to the coefficient update rule of the tap vector, thereby effectively removing interference noise and more clearly identifying coarse brain activity patterns.

[0082] like Figure 7 As shown, Figure 7 This is a schematic diagram of the NLMS adaptive filter principle. Figure 7 In this context, x(n) represents the input noise. It is an estimator for offline modeling of the secondary channel S(z):

[0083]

[0084] When the input noise x(n) is an estimate after passing through the secondary channel At this time, the signal can be represented as:

[0085]

[0086] in, This represents the i-th impulse response coefficient estimated by the secondary channel at time n.

[0087] refer to Figure 7 The filter tap vector is w(n) = [w(n), w(n-1), ..., w(n-L+1)].

[0088] refer to Figure 7 d(n) is the reference signal (e.g., background EEG activity), which is the spontaneous activity of the brain in a resting state or when it is not affected by specific stimuli. It is usually more complex and random, and is expressed as d(n) = x(n) * p(n) + m(n); where p(n) represents the impulse response of the main channel transfer function P(z), and m(n) is the added white noise interference.

[0089] refer to Figure 7 The noise signal y(n) = w T (n)x(n), denoised signal y′(n)=y(n)*s(n); error signal e(n)=d(n)-y′(n)=d(n)-w T (n)x f (n); where xf(n) = [xf(n), xf(n-1), ..., xf(n-L+1)], in this embodiment, x f (n) represents evoked potentials (EPs) in the electroencephalogram (EEG) signal. Evoked potentials (EPs) are EEG signals generated by specific stimuli through a specific experimental paradigm.

[0090] It should be noted that the gradient relationship between the objective function and the number of taps in the LMS algorithm is as follows:

[0091]

[0092] Where p=E[d(n)x f [n] represents the input signal x f The cross-correlation vector between (n) and the reference signal d(n), It is x f The autocorrelation vector of (n). The estimate of the tap vector can be expressed as w(n+1) = w(n) + 1 / 2ρg w By analyzing g w The estimate can be derived as follows:

[0093]

[0094] Therefore, the tap vector update expression is w(n+1)=w(n)-ρe(n)x f (n), where ρ is the step size factor. A larger step size results in faster algorithm convergence, which improves the overall data processing speed, but worsens the signal filtering effect. Conversely, a smaller step size results in slower algorithm convergence, but significantly improves the signal filtering effect. The constraint on the step size factor ρ is:

[0095]

[0096] Where, λ max It is the largest eigenvalue of the autocorrelation matrix.

[0097] It should be noted that, since the step size factor cannot simultaneously take into account both convergence speed and signal filtering effect, this embodiment normalizes the step size of the above LMS algorithm to obtain an NLMS algorithm. This NLMS algorithm can improve the convergence speed while significantly improving the signal filtering effect.

[0098] Specifically, the NLMS algorithm can be defined as:

[0099]

[0100] Then its unconstrained cost function can be expressed as:

[0101]

[0102] in, Let λ represent the square of the Euclidean second normal form, and λ be the Lagrange factor. Taking the derivative of the cost function, we can obtain the update formula for the tap vector:

[0103]

[0104] Because the input noise may have too little energy, R x Since incomplete rank occurs, a regularization factor δ is introduced to adjust w(n+1). δ is a positive number close to 0, and its formula can be expressed as:

[0105]

[0106] Therefore, the range of the step size can be expressed as:

[0107]

[0108] In practical use, the step size is exemplarily taken to be [0, 2].

[0109] Understandably, since the convergence speed slows down during input enhancement in the NLMS algorithm, this embodiment introduces a fast-convergence affine projection (AP) filtering algorithm to further accelerate the convergence speed. The AP algorithm can process multiple high-channel EEG data simultaneously and achieves fast convergence, which is undoubtedly beneficial for filtering ultra-high-lead data. The derivation principle of the AP filtering algorithm is as follows:

[0110] Input noise is represented as x f (n)=[x f (n), x f (n-1), ..., x f [(n-M+1)], where M represents the projection order. The output vector y′(n) is represented as d(n) is then expressed as d(n) = [d(n), d(n-1), ..., d(n-M+1)] T ;

[0111] To make w(n) and w(n+1) as close as possible, then:

[0112]

[0113] Therefore, the cost function can be expressed as:

[0114]

[0115] Differentiation yields:

[0116]

[0117] make Then w(n+1) = w(n) + X(n)λ. This leads to... From the above formula, the weights of the AP algorithm can be derived as follows:

[0118]

[0119] Among them, I K It is a K*K identity matrix, used to adjust the matrix size.

[0120] It should be noted that the reference Figure 8 , Figure 8 This is a diagram of the adaptive filtering structure of the AP-NLMS algorithm. Introducing the fusion factor λ(n) from the AP-NLMS fusion algorithm into the stochastic gradient descent approach ensures filtering performance while reducing computational complexity. Since both the AP and NLMS algorithms require a filter for parameter updates in this embodiment, theoretically, two filters would need to be iteratively updated after fusion. In this embodiment, after optimization and improvement, only one filter needs to be updated, which significantly reduces computation and results in faster convergence.

[0121] Specifically, the fusion factor is set:

[0122]

[0123] Updated using the sigmoid activation function expression and stochastic gradient descent.

[0124] Update the variable as follows:

[0125]

[0126] Then update the weight coefficient vector of the AP-NLMS algorithm:

[0127]

[0128] Where λ(n) is the fusion factor, a(n) is the variable updated by the fusion factor, w(n) is the weight coefficient of the fusion algorithm, u1 is the input noise signal of the AP algorithm after filtering, and U2 is the input noise signal of the NLMS algorithm after filtering. For ease of derivation, we denote it as u1 = x f (n), U2(n)=X f (n). ρ is a constant representing the step size of the combined algorithm, namely the AP-NLMS algorithm. e1 and e2 represent the noise reduction errors of the AP algorithm and the NLMS algorithm, respectively. e1 and e2 share the overall error system e(n).

[0129] As can be understood, RBFNN is an artificial neural network that uses radial basis functions as activation functions. It consists of three layers: an input layer, hidden layers, and an output layer. The non-linear RBFNN neural network structure is shown in the diagram below. Figure 9 As shown.

[0130] like Figure 9 As shown, in the input layer, x(n) is the sampled value of the noisy EPs signal at time n, and the total input vector is X(n) = [x(n), x(n-1), ..., x(n-m+1)]. T Where m represents the length of the input signal, j hidden layers perform nonlinear mapping on the input signal, and x(n-m+1) represents the sampled value after a delay of m.

[0131] like Figure 9 As shown, in the hidden layer, k1(n), k2(n), ..., k j (n) represents the response function of the j-th hidden layer neuron at time n to the input signal:

[0132]

[0133] Where, ||X(n)-C i (n)|| represents the Euclidean distance, σ i (n) represents the expansion coefficient of the i-th neuron, C i (n) is the center position vector of the i-th hidden neuron at time n.

[0134] like Figure 9 As shown, in the output layer, the output of the RBFNN with j hidden layer neurons at time n is:

[0135]

[0136] Among them, w i (n) represents the weights of the RBFNN, k i (n) is the corresponding response vector.

[0137] In one embodiment, the process of outputting a response signal based on the RBF neural network model and obtaining an error signal based on the response signal includes: activating the input signal using m delayed sample values ​​and j hidden neuron Gaussian functions based on the RBF neural network model and outputting a response signal; comparing the response signal with a desired signal to obtain an error signal; wherein the desired signal is the reference signal, and the error signal is the difference between the reference signal and the response signal.

[0138] Specifically, the RBFNN (RBF Neural Network Model) filtering method combined with an adaptive algorithm mainly consists of two steps: The first step is to activate the input signal and output a response using m delayed sample values ​​and j hidden neurons' Gaussian functions, and compare the response signal with the expected signal to obtain the error signal; The second step is to perform adaptive adjustment, and based on the error signal result, use the NLMS-AP adaptive filtering algorithm to optimize and adjust the parameters in the RBFNN network until the number of iterations is reached, and output the filtered EEG signal y(n).

[0139] For example, the framework diagram of the filtering method combining the NLMS-AP adaptive filtering algorithm and the RBFNN neural network is as follows: Figure 10 As shown, x(n) represents the EPs signal in the noisy input EEG, y(n) represents the EPs signal in the filtered output EEG, d(n) is the background electrical activity signal of the EEG, and e(n) is the error signal, expressed as e(n) = d(n) - y(n). By optimizing the error e(n), the RBFNN parameters are adjusted online to minimize the mean square error between the RBFNN outputs y(n) and d(n), thus obtaining the final filtered EEG signal.

[0140] It should be noted that when filtering high-throughput EEG signals, Mallat wavelets are first used for preliminary screening to obtain the approximate target signal in the frequency domain. Then, the AP-NLMS algorithm is used for adaptive filtering, balancing signal quality and data convergence speed. However, since the AP-NLMS algorithm is an adaptive linear filtering method that excels at handling the linear components of signals, this embodiment employs the RBF neural network algorithm to further improve the nonlinear mapping capability of EEG signals and enhance data generalization ability. This combines the advantages of the AP-NLMS adaptive filtering algorithm and the RBF neural network to effectively filter high-throughput signals and enhance signal quality. AP-NLMS provides the RBF neural network with a cleaner and more informative input signal, while the RBF neural network can utilize these high-quality signals for complex nonlinear mapping and pattern recognition tasks. This combined approach complements the respective methods, contributing to improved overall EEG signal processing performance, especially in EEG signals susceptible to noise interference.

[0141] In this embodiment, a filtering technique based on Mallat wavelet and NLMS-AP adaptive filtering combined with the RBFNN neural network algorithm is used to filter the massive data from an ultra-high-lead EEG machine. First, a preliminary simple filtering using Mallat wavelet transform is applied to remove high-frequency noise and preliminarily screen out the approximate and detail frequency components of the target EEG signal. Then, the denoised EEG signal is reconstructed to obtain the preliminarily denoised signal. This technique can determine the target frequency range. Next, NLMS-AP adaptive filtering is used to optimize the RBFNN neural network algorithm parameters. The NLMS algorithm further improves the signal filtering effect and reduces noise interference, while AP can quickly converge the network model, allowing the filtering method to simultaneously consider signal quality and processing speed. During network model training, GPU / CUDA is used for high-performance parallel computing. The ultra-high-lead data is parallelized, divided into multiple small blocks, each of which can be processed independently. These small blocks are allocated to different GPU threads, further improving the convergence speed. In this embodiment, by using wavelet transform combined with adaptive filtering and neural network algorithms to reduce noise, the quality of the EEG signal and the processing speed are effectively improved.

[0142] This embodiment provides a signal filtering method for ultra-high-lead electroencephalography (EEG) machines, comprising: acquiring ultra-high-lead EEG signal data collected by an EEG machine system; processing the ultra-high-lead EEG signal data to obtain a reference signal and an input signal; synchronously inputting the reference signal and the input signal into an RBF neural network model; filtering according to an adaptive filtering algorithm and the RBF neural network model, and outputting filtered EEG signal data. This invention employs an adaptive filtering algorithm combined with RBF neural network filtering technology to filter massive amounts of data from ultra-high-lead EEG machines. The adaptive filtering algorithm and RBF neural network complement each other, helping to improve the overall performance of EEG signal processing, especially in EEG signals susceptible to noise interference, further effectively filtering high-throughput signals and enhancing signal quality. This solves the problems of poor recognition effect, high computational complexity, and slow processing speed of existing commonly used EEG signal filtering methods for massive ultra-high-lead data.

[0143] Furthermore, this embodiment of the invention also proposes a storage medium storing a high-lead EEG signal filtering program, which, when executed by a processor, implements the steps of the high-lead EEG signal filtering method described above.

[0144] Reference Figure 11 , Figure 11 This is a structural block diagram of an embodiment of the signal filtering device for an ultra-high-lead electroencephalogram (EEG) machine according to the present invention.

[0145] like Figure 11 As shown, the ultra-high-lead electroencephalogram (EEG) signal filtering device includes:

[0146] Signal acquisition module 10 is used to acquire ultra-high-lead EEG signal data acquired by the EEG machine system;

[0147] Wavelet transform module 20 is used to process the ultra-high-lead EEG signal data to obtain a reference signal and an input signal;

[0148] Signal filtering module 30 is used to synchronously input the reference signal and the input signal into the RBF neural network model for filtering;

[0149] The adaptive optimization module 40 is used to optimize the parameters of the RBF neural network model according to the adaptive filtering algorithm, so as to output filtered EEG signal data.

[0150] In one embodiment, the wavelet transform module 20 is specifically used to determine the EEG background activity signal and evoked potentials based on the ultra-high-lead EEG signal data; perform Butterworth bandpass filtering on the EEG background activity signal to obtain a noise-related signal and use the noise-related signal as a reference signal; filter the evoked potentials based on Mallat wavelet transform to obtain an initial denoised signal and use the initial denoised signal as an input signal.

[0151] The process of filtering the evoked potentials based on Mallat wavelet transform to obtain an initial denoised signal and using the initial denoised signal as the input signal includes: initially filtering the evoked potentials using a low-pass filter and a high-pass filter based on Mallat wavelet transform to obtain a denoised EEG signal; reconstructing the denoised EEG signal using a low-pass reconstruction filter and a high-pass reconstruction filter to obtain an initial denoised signal and using the initial denoised signal as the input signal.

[0152] In one embodiment, the adaptive optimization module 40 is specifically used to obtain an error signal based on the reference signal and the input signal synchronously input to the RBF neural network model; optimize the parameters of the RBF neural network model based on the error signal, the normalized least mean square filtering algorithm, and the affine projection filtering algorithm; and output the filtered EEG signal data when the number of iterations of the RBF neural network model reaches a preset number of iterations.

[0153] The process of obtaining an error signal based on the synchronous input reference signal and the input signal in response to the RBF neural network model includes: activating the input signal and outputting a response signal using m delayed sample values ​​and j hidden neuron Gaussian functions based on the RBF neural network model; comparing the response signal with the desired signal to obtain the error signal; wherein the desired signal is the reference signal, and the error signal is the difference between the reference signal and the response signal.

[0154] The optimization of the parameters of the RBF neural network model based on the error signal, the normalized minimum mean square filtering algorithm, and the affine projection filtering algorithm includes: constructing an NLMS-AP adaptive filtering fusion algorithm based on the normalized minimum mean square filtering algorithm and the affine projection filtering algorithm; and adjusting the parameters of the RBF neural network model online according to the NLMS-AP adaptive filtering fusion algorithm to minimize the mean square error between the output signal of the RBF neural network model and the EEG background active electrical signal.

[0155] In one embodiment, the device further includes: dividing the acquired ultra-high-lead EEG signal data into multiple small blocks; allocating the data corresponding to each small block to different GPU threads; performing parallel processing on the data corresponding to each small block according to GPU or CUDA; and executing the ultra-high-lead EEG signal filtering method through wavelet transform module 20, signal filtering module 30 and adaptive optimization module 40.

[0156] like Figure 3 As shown, ultra-high-lead EEG data (i.e., ultra-high-lead EEG signal data) is acquired through an electroencephalography (EEG) system. The ultra-high-lead EEG data is then processed by a high-performance parallel computing filter to output filtered EEG data. It can be understood that the high-performance parallel computing filter includes a wavelet transform module 20, a signal filtering module 30, and an adaptive optimization module 40. GPU / CUDA is used for high-performance parallel computing during data processing. The ultra-high-lead data is parallelized and divided into multiple small blocks (e.g., Figure 3 X shown EEG1 X EEG2 , ..., X EEGN Each small block can be processed independently, and multiple small blocks of data can be allocated to different GPU threads. During parallel computing, memory is effectively managed on the GPU, with reasonable allocation of global and shared memory. Considering the use of CUDA kernel functions, the number of threads and blocks needs to be configured appropriately based on the GPU architecture and data characteristics; this embodiment does not impose any limitations on this.

[0157] In this embodiment, a filtering technique based on Mallat wavelet and NLMS-AP adaptive filtering combined with the RBFNN neural network algorithm is used to filter the massive data from an ultra-high-lead EEG machine. First, a preliminary simple filtering using Mallat wavelet transform is applied to remove high-frequency noise and preliminarily screen out the approximate and detail frequency components of the target EEG signal. Then, the denoised EEG signal is reconstructed to obtain the preliminarily denoised signal. This technique can determine the target frequency range. Next, NLMS-AP adaptive filtering is used to optimize the RBFNN neural network algorithm parameters. The NLMS algorithm further improves the signal filtering effect and reduces noise interference, while AP can quickly converge the network model, allowing the filtering method to simultaneously consider signal quality and processing speed. During network model training, GPU / CUDA is used for high-performance parallel computing. The ultra-high-lead data is parallelized into multiple small blocks, each of which can be processed independently. These small blocks are allocated to different GPU threads, further improving the convergence speed. In this embodiment, by using wavelet transform combined with adaptive filtering and neural network algorithms to reduce noise, the quality of the EEG signal and the processing speed are effectively improved.

[0158] This embodiment provides a signal filtering device for an ultra-high-lead electroencephalogram (EEG) machine, comprising: a signal acquisition module 10 for acquiring ultra-high-lead EEG signal data acquired by the EEG machine system; a wavelet transform module 20 for processing the ultra-high-lead EEG signal data to obtain a reference signal and an input signal; a signal filtering module 30 for synchronously inputting the reference signal and the input signal into an RBF neural network model for filtering; and an adaptive optimization module 40 for optimizing the parameters of the RBF neural network model according to an adaptive filtering algorithm to output filtered EEG signal data. This invention employs an adaptive filtering algorithm combined with RBF neural network filtering technology to filter massive amounts of data from an ultra-high-lead EEG machine. The adaptive filtering algorithm and RBF neural network complement each other, helping to improve the overall performance of EEG signal processing, especially in EEG signals susceptible to noise interference, further effectively filtering high-throughput signals and enhancing signal quality. This solves the problems of poor recognition effect, high computational complexity, and slow processing speed of existing commonly used EEG signal filtering methods for massive ultra-high-lead data.

[0159] It should be noted that technical details not described in detail in this embodiment of the ultra-high-lead EEG signal filtering device can be found in any embodiment of the present invention applied to the ultra-high-lead EEG signal filtering method as described above, and will not be repeated here.

[0160] It should be understood that the above are merely illustrative examples and do not constitute any limitation on the technical solutions of the present invention. In specific applications, those skilled in the art can make settings as needed, and the present invention does not impose any restrictions on this.

[0161] It should be noted that the workflow described above is merely illustrative and does not limit the scope of protection of this invention. In practical applications, those skilled in the art can select some or all of the workflow to achieve the purpose of this embodiment according to actual needs, and no restrictions are imposed here.

[0162] Furthermore, it should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0163] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0164] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as read-only memory (ROM) / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0165] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A signal filtering method for ultra-high-lead electroencephalography (EEG) machines, characterized in that, include: Acquire ultra-high-lead EEG signal data from the EEG system; The ultra-high-lead EEG signal data is processed to obtain a reference signal and an input signal; The reference signal and the input signal are synchronously input into the RBF neural network model for filtering; The parameters of the RBF neural network model are optimized according to the adaptive filtering algorithm to output filtered EEG signal data; The parameters of the RBF neural network model are optimized according to an adaptive filtering algorithm to output filtered EEG signal data, including: optimizing the parameters of the RBF neural network model according to a normalized least mean square filtering algorithm and an affine projection filtering algorithm; When the number of iterations of the RBF neural network model reaches a preset number of iterations, a response signal is output based on the RBF neural network model and an error signal is obtained based on the response signal; when the error signal meets a preset error threshold, filtered EEG signal data is output.

2. The method as described in claim 1, characterized in that, The process of processing the ultra-high-lead EEG signal data to obtain a reference signal and an input signal includes: Based on the ultra-high-conductance EEG signal data, EEG background activity signals and evoked potentials were determined; The EEG background activity electrical signal is subjected to Butterworth bandpass filtering to obtain a noise-related signal, which is then used as a reference signal. The evoked potentials are filtered based on Mallat wavelet transform to obtain an initial denoised signal, which is then used as the input signal.

3. The method as described in claim 2, characterized in that, The step of filtering the evoked potential based on Mallat wavelet transform to obtain an initial denoised signal and using the initial denoised signal as the input signal includes: Based on Mallat wavelet transform, the evoked potentials are initially filtered using low-pass and high-pass filters to obtain denoised EEG signals. The denoised EEG signal is reconstructed using a low-pass reconstruction filter and a high-pass reconstruction filter to obtain an initial denoised signal, which is then used as the input signal.

4. The method as described in claim 1, characterized in that, The optimization of the parameters of the RBF neural network model based on the normalized least mean square filtering algorithm and the affine projection filtering algorithm includes: An NLMS-AP adaptive filtering fusion algorithm is constructed based on the normalized least mean square filtering algorithm and the affine projection filtering algorithm; wherein, the fusion factor in the NLMS-AP adaptive filtering fusion algorithm includes stochastic gradient descent, and the fusion factor is updated through the expression of the activation function sigmoid; The parameters of the RBF neural network model are adjusted online according to the NLMS-AP adaptive filtering fusion algorithm.

5. The method as described in claim 1, characterized in that, The step of outputting a response signal based on the RBF neural network model and obtaining an error signal based on the response signal includes: Based on the RBF neural network model, the input signal is activated and the output response signal is generated using m delayed sample values ​​and j hidden neuron Gaussian functions. The response signal is compared with the expected signal to obtain an error signal; wherein the expected signal is the reference signal, and the error signal is the difference between the reference signal and the response signal.

6. The method as described in claim 1, characterized in that, The method further includes: The collected ultra-high-conductance EEG signal data is divided into multiple small blocks; The data corresponding to each of the aforementioned small blocks is assigned to different GPU threads; The steps of the ultra-high-lead electroencephalogram (EEG) signal filtering method are performed by processing the data corresponding to each small block in parallel using GPU or CUDA.

7. A signal filtering device for an ultra-high-lead electroencephalogram (EEG) machine, characterized in that, include: The signal acquisition module is used to acquire ultra-high-lead EEG signal data collected by the EEG machine system; The wavelet transform module is used to process the ultra-high-lead EEG signal data to obtain a reference signal and an input signal; The signal filtering module is used to synchronously input the reference signal and the input signal into the RBF neural network model for filtering; An adaptive optimization module is used to optimize the parameters of the RBF neural network model according to an adaptive filtering algorithm, so as to output filtered EEG signal data. The parameters of the RBF neural network model are optimized according to an adaptive filtering algorithm to output filtered EEG signal data, including: optimizing the parameters of the RBF neural network model according to a normalized least mean square filtering algorithm and an affine projection filtering algorithm; When the number of iterations of the RBF neural network model reaches a preset number of iterations, a response signal is output based on the RBF neural network model and an error signal is obtained based on the response signal; when the error signal meets a preset error threshold, filtered EEG signal data is output.

8. An electronic device, characterized in that, The electronic device includes: a memory, a processor, and an ultra-high-lead electroencephalogram (EEG) signal filtering program stored in the memory and executable on the processor, wherein the ultra-high-lead EEG signal filtering program is configured to implement the ultra-high-lead EEG signal filtering method as described in any one of claims 1 to 6.

9. A storage medium, characterized in that, The storage medium stores a high-lead EEG signal filtering program, which is used to enable the processor to implement the high-lead EEG signal filtering method as described in any one of claims 1 to 6 when executed.

Images