Electroencephalogram motion artifact removal method, device and storage device under lower limb motion state

Abstract

The application provides a lower limb movement state electroencephalogram movement artifact removal method, equipment and a storage device, the corresponding electromyogram reference noise is estimated according to the movement speed and speed change of the lower limb; the movement artifact is preliminarily eliminated through adaptive noise cancellation to the electromyogram reference noise; the signal processed through the adaptive noise cancellation is subjected to variational mode decomposition, and the intrinsic mode function of each order variable is obtained; the intrinsic mode function of each order is subjected to independent component analysis, and the independent component is obtained; the time domain feature, the frequency spectrum feature and the intersequence similarity of the independent component are calculated, then the artifact is identified and removed through hierarchical clustering identification, and the remaining components are reconstructed to obtain the processed independent component, i.e. the reconstructed electroencephalogram signal. The application has the beneficial effect that the interference caused by electromyogram noise when the lower limb speed changes can be quickly responded.

Description

Technical Field

[0001] This invention relates to the field of electroencephalogram (EEG) signals, and more particularly to a method, apparatus, and storage device for removing EEG motion artifacts during lower limb movement. Background Technology

[0002] Electroencephalography (EEG) signals are the sum of extracellular potentials caused by postsynaptic potentials during the firing of numerous neurons in the brain. It is a method of recording brain activity using electrophysiological indicators. EEG signals can be broadly divided into two types: (a) spontaneous EEG: brain potential changes generated spontaneously by the nervous system without any specific external stimulation; and (b) evoked EEG: potential changes induced in corresponding parts of the brain when a sensory stimulus (such as sound, light, image, or somatosensory sensation) is applied. The generation mechanism of EEG signals is complex, yet it contains rich information. EEG is a spatially discrete, non-stationary, time-varying signal with an irregular time-domain waveform, making it difficult to summarize patterns. From a frequency domain perspective, it exhibits rhythmicity. During EEG data acquisition, interference noise is introduced due to factors such as the external environment, eye movements, and muscle movements. Data preprocessing aims to remove interference noise from the raw EEG signals to obtain relatively pure EEG signals suitable for emotion recognition. Currently, commonly used preprocessing methods mainly include filtering, principal component analysis, and independent component analysis.

[0003] Filtering can remove some interference with relatively fixed frequency bands. For example, power frequency interference can be filtered out using bandpass or low-pass filtering. For EEG noise that is difficult to remove by filtering, such as electrooculography (EOG) noise and electromyography (EMG) noise, principal component analysis (PCA) and independent component analysis (ICA) can be used for processing.

[0004] Principal component analysis (PCA) decomposes the EEG signal into independent components based on the distribution of EEG leads, removes unwanted interference, and then reconstructs the EEG to remove artifacts. Because PCA uses the orthogonality principle to decompose the original EEG signal into independent components, it cannot effectively separate artifacts that are similar to the EEG waveform.

[0005] Independent component analysis (ICA) is a blind source signal separation method. Since artifacts such as electrocardiogram (ECG) and electrooculogram (EOG) signals in EEG signals are generated by independent signal sources, ICA can demix the raw data, removing interfering signals such as ECG and EOG signals, thus obtaining a clean EEG signal. ICA does not require a dedicated reference electrode to record artifacts during artifact removal, and the components after decomposition are independent of each other, resulting in high accuracy in artifact removal. It has been widely used in the preprocessing of EEG signals. MATLAB toolboxes such as EEGLAB and Python toolkits such as MNE can be used to perform ICA on EEG data.

[0006] In addition to the above, mode decomposition is also involved in data processing, commonly including empirical mode decomposition, ensemble empirical mode decomposition, and variational mode decomposition. Variational mode decomposition is an adaptive, fully non-recursive method for variational mode and signal processing. This technique has the advantage of being able to determine the number of mode decompositions. Its adaptability is manifested in determining the number of mode decompositions for a given sequence based on the actual situation. In the subsequent search and solution process, it can adaptively match the optimal center frequency and finite bandwidth for each mode, and can achieve effective separation of intrinsic mode components, frequency domain partitioning of the signal, and thus obtain the effective decomposition components of the given signal, ultimately obtaining the optimal solution to the variational problem. It overcomes the end-point effect and mode component aliasing problems of empirical mode decomposition methods, and has a more solid mathematical theoretical foundation. It can reduce the non-stationarity of time series with high complexity and strong nonlinearity, decomposing to obtain relatively stationary subsequences containing multiple different frequency scales, making it suitable for non-stationary sequences.

[0007] Existing EEG preprocessing methods primarily target EEG data acquired under static conditions, with limited research on EEG data during human movement. Furthermore, these methods struggle to quickly process changing noise when movement conditions change, i.e., speed changes. This invention proposes a method for removing motion artifacts from lower limb EEG data during human movement based on fuzzy noise processing. It employs a combination of processing techniques to specifically address motion-induced EEG artifacts, effectively removing them. Summary of the Invention

[0008] To address the aforementioned problems, this invention provides a method, apparatus, and storage device for removing EEG motion artifacts during lower limb movement. The method for removing EEG motion artifacts during lower limb movement mainly includes:

[0009] S1: Estimate the corresponding electromyographic reference noise based on the movement speed and speed change of the lower limbs;

[0010] S2: Initially eliminate motion artifacts from electromyographic reference noise using adaptive noise cancellation technology;

[0011] S3: Perform variational mode decomposition on the signal after adaptive noise cancellation processing to obtain the eigenmode functions of each order of variables;

[0012] S4: Perform independent component analysis on the eigenmode functions of each order to obtain independent components;

[0013] S5: Calculate the temporal features, spectral features, and inter-order similarity of the independent components. Then, through hierarchical clustering identification and artifact removal, reconstruct the remaining components to obtain the processed independent components, i.e., reconstruct the EEG signal.

[0014] Further, in step S1, the corresponding electromyographic reference noise is estimated using a fuzzy noise processing method. The specific process is as follows:

[0015] S1.1: Based on the different movement states of the human lower limbs, the movement speed is classified and fuzzified. Before this, the electromyographic noise of the human lower limbs at different movement speeds needs to be measured in order to formulate fuzzy rules.

[0016] S1.2: After the speed is set, the speed classifier will compare the real-time speed with the preset speed and divide the speed gradient. Each speed level has its corresponding fuzzy reference noise. Through digital-to-analog conversion, the change in speed is used to represent the change in noise. The speed classifier will output the first speed of the grade each time.

[0017] S1.3: Perform digital-to-analog conversion on the obtained velocity output to obtain the required fuzzy reference noise, that is, estimate the corresponding electromyographic reference noise.

[0018] Furthermore, in step S2, the preliminary process of eliminating motion artifacts is as follows:

[0019] Let the original EEG signal be s(n), but the acquired signal contains noise z0(n). The original input signal is h(n) = s(n) + z0(n). To eliminate the noise z0(n) without affecting the original signal s(n), a reference noise source signal z1(n) is introduced. z1(n) is only related to z0(n) and not to the original EEG signal s(n). The reference noise consists of fuzzy reference noise and conventional reference noise. The fuzzy reference noise is mainly electromyographic noise during movement, while the conventional reference noise is mainly noise with a relatively fixed frequency. The initial input signal h(n) is applied to the P terminal of the adaptive filter, and the reference noise source signal z1(n) is input to the Q terminal of the adaptive filter. After being filtered by the adaptive filter, z1(n) generates a reference signal y(n). The difference between h(n) and y(n) generates an error reference output e(n). The error reference output e(n) provides a control signal to the adaptive filter in a form similar to negative feedback to adjust e(n), so that e(n) tends to s(n). The final e(n) is the original signal required for the initial elimination of motion artifacts.

[0020] Furthermore, in step S3, the solution process for the variational mode decomposition is as follows:

[0021]

[0022]

[0023] In the formula, K represents the number of modes to be decomposed, u k ω k δ(t) represents the k-th modal component and center frequency after decomposition, respectively. δ(t) is the Dirac function, * is the convolution operator, and f is the EEG signal after preliminary processing by adaptive noise cancellation.

[0024] Further, in step S4, K eigenmode functions are obtained through variational mode decomposition, forming a vector x, which follows the formula x = As, where A is the unknown mixing matrix, and s is the unknown source signal, representing the independent components to be determined. Given a vector x with K members, each of which is a random variable, then:

[0025]

[0026]

[0027] Let H = A -1 If s = Hx, and the values ​​of vector x are recorded J times, then dataset B is formed.

[0028]

[0029] After obtaining dataset B, the FastICA method is used to unmix the data and estimate the values ​​of A, H, and s to obtain the required m independent components.

[0030] A storage device that stores instructions and data for implementing a method for removing brainwave motion artifacts during lower limb movement.

[0031] A device for removing EEG motion artifacts during lower limb movement includes: a processor and a storage device; the processor loads and executes instructions and data in the storage device to implement a method for removing EEG motion artifacts during lower limb movement.

[0032] The beneficial effect of the technical solution provided by this invention is that it can quickly respond to the interference caused by electromyographic noise when the speed of the lower limb changes. Attached Figure Description

[0033] The present invention will be further described below with reference to the accompanying drawings and embodiments. In the accompanying drawings:

[0034] Figure 1 This is a flowchart of a method for removing EEG motion artifacts during lower limb movement in an embodiment of the present invention.

[0035] Figure 2 This is a schematic diagram of fuzzy reference noise acquisition in an embodiment of the present invention.

[0036] Figure 3 This is a schematic diagram of a fuzzy neural network in an embodiment of the present invention.

[0037] Figure 4 This is a schematic diagram of the adaptive noise cancellation process for preliminary elimination of motion artifacts in the fuzzy noise processing of an embodiment of the present invention.

[0038] Figure 5 This is a schematic diagram of the hardware device working in an embodiment of the present invention. Detailed Implementation

[0039] To provide a clearer understanding of the technical features, objectives, and effects of the present invention, specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0040] In daily life, lower limb exercise refers to the training of the lower limbs using appropriate movement methods. Lower limb rehabilitation training using rehabilitation robots involves lower limb movement. It can simulate the walking pattern of a normal person, bear a portion of the body weight, and walk at a controlled speed, providing effective rehabilitation training for patients with lower limb movement disorders. By helping patients simulate the walking patterns of normal people, it trains lower limb muscle strength, restores the nervous system's control over walking function, and ultimately achieves the goal of restoring lower limb motor function and walking ability.

[0041] EEG motion artifacts are artifacts in EEG signals during motion. They include electrooculography (EOG) interference, electromyography (EMG) interference, electrocardiography (ECG) interference, power line interference, and EEG signal interference caused by motion.

[0042] The movement of the lower limbs inevitably introduces electromyographic (EMG) noise interference into EEG measurements. To minimize this interference, a method combining fuzzy noise processing and adaptive noise cancellation is proposed for EEG artifact removal. This method focuses on the velocity variation and the acquisition of adaptive noise cancellation reference noise to specifically remove EMG noise artifacts. Please refer to [reference needed]. Figure 1 , Figure 1 This is a flowchart of a method for removing EEG motion artifacts during lower limb movement in an embodiment of the present invention, specifically including:

[0043] S1: Fuzzy noise processing. The corresponding electromyographic reference noise is estimated based on the movement speed and speed changes of the lower limbs; that is, the fuzzy reference noise is obtained as follows:

[0044] Different lower limb movement states produce different levels of electromyographic (EMG) noise interference. To quickly estimate the reference noise when speed changes, this invention proposes a noise processing method—fuzzy noise processing. Fuzzy noise processing primarily targets EMG noise interference caused by lower limb movement. Based on different lower limb movement states, the movement speed is graded and fuzzified. Prior to this, EMG noise at different lower limb movement speeds needs to be measured to formulate fuzzy rules. The movement speed of the lower limbs can be represented by the angular velocity of the foot pedal rotation. During subsequent rehabilitation training, the noise frequency to be used as the reference noise can be determined based on the movement speed for filtering. It is important to note that the change in lower limb movement state mentioned here refers to a change from one uniform speed state to another, and the movement of the lower limbs is driven by a machine. Only in this way is it meaningful to study the impact of lower limb speed and speed changes on EMG noise. Figure 2 As shown, after the speed is set, the speed classifier compares the real-time speed with the preset speed and divides the speed into gradients. For example, if the speed setting is b and the real-time speed is a, then the speed classifier will divide the speed into k levels, namely [a, v1, v2, ..., v k [,b], where each speed level in this classification is either a, b, or from v1 to v kEach speed has its corresponding fuzzy reference noise, which is measured before rehabilitation training. This allows the speed change to represent the noise change through digital-to-analog conversion. The speed classifier outputs the first speed of each class, v1, which is between [a, b]. It's important to note that v1 is a variable until it equals the set speed b. The deviation e between v1 and a, and the rate of change of deviation ec, are then used as inputs for fuzzy control. In addition, the deviation e also controls the fuzzy inference method. When the deviation is greater than the preset value p, a lookup table method is used for fuzzy inference; when the deviation is less than the preset value p, a fuzzy neural network is used for fuzzy inference, resulting in the K value for PI control. P K I The speed output is controlled by a value. A lookup table method is used for fuzzy inference to quickly respond to changes in error, while fuzzy neural network PI control aims to eliminate steady-state errors. Finally, the obtained speed output is converted from digital to analog to obtain the required fuzzy reference noise. The structure of the fuzzy neural network PI controller is as follows: Figure 3 As shown, the first layer is the input layer, with the deviation e and the rate of change of deviation ec as inputs, x1 = e, x2 = ec.

[0045]

[0046] In the formula, X (1) ,Y (1) These are the input and output of the first network layer, respectively.

[0047] The second layer is the fuzzification layer, which calculates the membership values ​​of each input component. This layer first performs fuzzy quantization on the input variables using the discrete precise quantization method. The actual value range of e is [a, b], and its fuzzy universe of discourse is [a1, b1] (a1 < b1). The quantization process is shown in Equation 1. Since the calculated E is usually not an integer, it needs to be rounded down, as shown in Equation 2.

[0048]

[0049]

[0050] There are h linguistic variables, and the absolute value of the fuzzy universe of discourse width is z. The fuzzy universe of discourse for the output variables and the linguistic variable partitioning are consistent with those for the input. After partitioning, the membership function F is used to determine the membership value F corresponding to each linguistic variable. ij (x i The number of nodes used in this layer is 2h.

[0051]

[0052] In the formula, d ij It is the center of the membership function, l ijIt is the width of the membership function.

[0053] The third layer is the fuzzy inference layer, which has a total of h layers. 2 Each node represents h 2 A fuzzy rule. β n (n = 1, 2, 3, ..., h) 2 ) represents the applicability of each fuzzy rule, used to match fuzzy rules.

[0054]

[0055] The fourth layer is the normalization layer, h 2 This layer, consisting of several nodes, primarily performs a normalization operation on the overall network structure. The formula is as follows:

[0056]

[0057] The fifth layer is the output layer, which has two nodes. It performs defuzzification calculations on the fuzzy quantities obtained from fuzzy inference as shown in formula (7), corresponding to the parameters K of the PI controller. P K I .

[0058]

[0059] In the formula, y m For network output, α mn It is the connection weight between the return layer and the output layer.

[0060] When using rule tables for fuzzy inference, the weighted average method is used to defuzzify the fuzzy output obtained from the fuzzy inference, as shown in formula (8):

[0061]

[0062] In the formula, μ 1i (e), μ 2i (ec) is the membership function of the input variable, u i This is the i-th output.

[0063] Regarding PI, an incremental PI algorithm is used, and the formula is as follows:

[0064]

[0065] S2: The fuzzy reference noise is used for adaptive noise cancellation, specifically:

[0066] The obtained blurred reference noise is combined with conventional interference reference noise, and adaptive noise cancellation techniques are used to initially eliminate motion artifacts. For example... Figure 4As shown: the original EEG signal is s(n), but the acquired signal contains noise z0(n), so the original input signal is h(n) = s(n) + z0(n). To eliminate the noise z0(n) without affecting the original signal s(n), a reference noise source signal z1(n) is introduced. The noise z1(n) is only related to the noise z0(n) and not to the signal source s(n). The reference noise consists of fuzzy reference noise and conventional reference noise. The fuzzy reference noise is obtained by... Figure 2 The main noise source is electromyographic noise during exercise, while conventional reference noise is mainly fixed-frequency noise such as power frequency interference. The original input h(n) is applied to the P terminal of the adaptive filter, and the reference noise source signal z1(n) is applied to the Q terminal of the adaptive filter. After the noise z1(n) is filtered by the adaptive filter AF, a reference signal y(n) is generated. The difference between h(n) and y(n) generates an error reference output e(n). This error reference output e(n) provides a control signal to the adaptive filter AF in a form similar to negative feedback, adjusting e(n) so that the value of h(n) - y(n), i.e., e(n), tends to s(n). Finally, the output e(n) is obtained, which is the desired original signal.

[0067] S3: Perform variational mode decomposition on the signal after adaptive noise cancellation processing to obtain the eigenmode functions of each order variable.

[0068] Assuming the EEG signal after initial adaptive noise cancellation processing is denoted as f, and it is decomposed into k components, the eigenmode functions of each order of function variable are obtained. The solution process for variational mode decomposition is as follows:

[0069] In the formula, K represents the number of modes to be decomposed (a positive integer), u k ω k δ(t) represents the k-th modal component and center frequency after decomposition, respectively, δ(t) is the Dirac function, and * is the convolution operator.

[0070] To solve equation (10), we introduce the Lagrange multiplication operator λ, and obtain:

[0071]

[0072] In the formula, α is a quadratic penalty factor, which reduces Gaussian noise interference. By using an alternating direction multiplier iterative algorithm combined with Passevar's theorem and Fourier isometric transform, the modal components and center frequencies are optimized, and the saddle point of the augmented Lagrange function is searched. The iteratively optimized u is then used to find the optimal modal components and center frequencies. k ω k The expressions for λ are as follows:

[0073]

[0074]

[0075]

[0076] In the formula, γ represents the noise tolerance, which is used to meet the fidelity requirements of signal decomposition. Corresponding to u i Fourier transforms of f(t), λ(t), and f(t).

[0077] S4: Perform independent component analysis on the eigenmode functions of each order to obtain independent components. The specific operation is as follows:

[0078] After variational mode decomposition, K eigenmode functions are obtained, forming a vector x, which follows the formula x = As, where A is the unknown mixing matrix, and s is the unknown source signal, representing the independent components to be determined. Given a vector x with K members, each of which is a random variable, then:

[0079]

[0080]

[0081] Let H = A -1 If s = Hx, and the values ​​of vector x are recorded J times, then dataset B is formed.

[0082]

[0083] After obtaining dataset B, the values ​​of A, H, and s are estimated by unmixing the data using the Fast Independent Component Analysis (FastICA) method, thus obtaining the required m independent components.

[0084] S5: For the obtained m independent components, calculate their temporal features, spectral features, and inter-order similarity. Then, identify and remove artifacts through hierarchical clustering, reconstruct the remaining components to obtain the processed independent components. Calculate the kurtosis value of each independent component as its temporal feature. The calculation method is as follows:

[0085]

[0086] In the formula s i Let E() represent the i-th independent component, and let E() be the expectation function.

[0087]

[0088] In the formula, L is an independent component s i The amount of data, Represents independent components si The nth power of the first data point.

[0089] Please see Figure 5 , Figure 5 This is a schematic diagram of the hardware device in operation according to an embodiment of the present invention. The hardware device specifically includes: a brainwave motion artifact removal device 401 for lower limb movement, a processor 402, and a storage device 403.

[0090] A device 401 for removing EEG motion artifacts during lower limb movement: The device 401 for removing EEG motion artifacts during lower limb movement implements the method for removing EEG motion artifacts during lower limb movement.

[0091] Processor 402: The processor 402 loads and executes the instructions and data in the storage device 403 to implement the method for removing EEG motion artifacts in a lower limb movement state.

[0092] Storage device 403: The storage device 403 stores instructions and data; the storage device 403 is used to implement the method for removing EEG motion artifacts in a lower limb movement state.

[0093] The beneficial effects of this invention are: when the speed of lower limb movement changes, the corresponding electromyographic noise will also change, which will lead to serious EEG artifacts. In order to deal with the electromyographic noise interference caused by the change of lower limb speed, a method combining fuzzy noise processing and adaptive noise cancellation is proposed, which can quickly deal with the interference caused by electromyographic noise when the lower limb speed changes.

[0094] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for removing EEG motion artifacts during lower limb movement, characterized in that: include: S1: Estimate the corresponding electromyographic reference noise based on the movement speed and speed change of the lower limbs; In step S1, the corresponding electromyographic reference noise is estimated using a fuzzy noise processing method. The specific process is as follows: S1.1: Based on the different movement states of the human lower limbs, the movement speed is classified and fuzzified. Before this, the electromyographic noise of the human lower limbs at different movement speeds needs to be measured in order to formulate fuzzy rules. S1.2: After the speed is set, the speed classifier will compare the real-time speed with the preset speed and divide the speed gradient. Each speed level has its corresponding fuzzy reference noise. Through digital-to-analog conversion, the speed change represents the noise change. The speed classifier will output the first speed of the grade each time. S1.3: Perform digital-to-analog conversion on the obtained velocity output to obtain the required fuzzy reference noise, that is, estimate the corresponding electromyographic reference noise; S2: Initially eliminate motion artifacts from electromyographic reference noise using adaptive noise cancellation technology; S3: Perform variational mode decomposition on the signal after adaptive noise cancellation processing to obtain the eigenmode functions of each order of variables; S4: Perform independent component analysis on the eigenmode functions of each order to obtain independent components; S5: Calculate the temporal features, spectral features, and inter-order similarity of the independent components. Then, through hierarchical clustering identification and artifact removal, reconstruct the remaining components to obtain the processed independent components, i.e., reconstruct the EEG signal.

2. The method for removing EEG motion artifacts during lower limb movement as described in claim 1, characterized in that: In step S2, the preliminary process of eliminating motion artifacts is as follows: Let the original EEG signal be However, the collected signal contains noise. The original input signal is In order to eliminate noise Without affecting the original brainwave signals A reference noise source signal was introduced. , Only with Related to the original brain electrical signals Irrelevant; The reference noise consists of fuzzy reference noise and conventional reference noise. Fuzzy reference noise mainly consists of electromyographic noise during movement, while conventional reference noise mainly consists of noise with a relatively fixed frequency; the original input signal The signal is applied to the P terminal of the adaptive filter, referencing the noise source signal. The input is fed into the Q terminal of the adaptive filter; After filtering by the adaptive filter, a reference signal is generated. , and Difference to generate error reference output Error reference output A control signal is applied to the adaptive filter in a manner similar to negative feedback to adjust it. , making tending to The final result That is, the original signal required for the initial elimination of motion artifacts.

3. The method for removing EEG motion artifacts during lower limb movement as described in claim 1, characterized in that: In step S3, the solution process for variational mode decomposition is as follows: In the formula This indicates the number of modes that need to be decomposed. , They represent the corresponding first and second parts after decomposition. Each modal component and center frequency, It is the Dirac function, and * is the convolution operator. The EEG signal is after initial processing with adaptive noise cancellation.

4. The method for removing EEG motion artifacts during lower limb movement as described in claim 1, characterized in that: In step S4, variational mode decomposition yields... There are 1 eigenmode functions, and the vector formed by them is: Then obey ,in For an unknown mixture matrix, It is an unknown source signal, and its independent components need to be determined; the known vector... , It contains If there are n members, and each member is a random variable, then: make ,but Record vector value J This results in dataset B: After obtaining dataset B, the FastICA method is used to unmix the data and estimate the... The value of is used to obtain the required m independent components.

5. The method for removing EEG motion artifacts during lower limb movement as described in claim 1, characterized in that: In step S5, the kurtosis value of each independent component is calculated as the time-domain characteristic of each independent component. The calculation method is as follows: In the formula Indicates the first i Each independent component It is the expectation function: In the formula L Independent components The amount of data, Represents independent components The first data point n Power of 1.

6. A storage device, characterized in that: The storage device stores instructions and data for implementing the method for removing EEG motion artifacts in lower limb movement states as described in any one of claims 1 to 5.

7. A device for removing EEG motion artifacts during lower limb movement, characterized in that: include: A processor and a storage device; the processor loads and executes instructions and data in the storage device to implement the method for removing EEG motion artifacts in lower limb movement states as described in any one of claims 1 to 5.

Images