High-compression-rate electroencephalogram signal processing method easy to implement by hardware

Abstract

The invention provides an electroencephalogram signal processing method with a high compression ratio, which comprises the following steps of: firstly, acquiring an original electroencephalogram signal, then converting the original electroencephalogram signal into a sparse domain through sparse transformation such as discrete cosine transformation, and extracting a sparse coefficient; thirdly, an initial incidence matrix is constructed, and an expansion matrix is generated through vertical and horizontal expansion and nesting operation so as to meet the compressed sensing requirement; and finally, compressing the sparse coefficient by using the expansion matrix to obtain compressed signal representation. According to the method, the measurement matrix is expanded through the combination design based on the Hadamard matrix, the compression ratio is increased, the hardware power consumption is reduced, the problem that resources are consumed by online generation of the random sparse matrix in the prior art is solved, the method is particularly suitable for a scene with a low power consumption requirement, and the hardware performance of an electroencephalogram signal processing scheme is improved.

Description

Technical Field

[0001] The invention relates to electroencephalogram (EEG) signal processing technology, and in particular to an EEG signal processing method with high compression rate that is easy to implement with hardware. Background Art

[0002] Remote health monitoring systems are becoming increasingly important for doctors to perform long-term patient monitoring. They can help diagnose a variety of health problems that patients may suffer from. One of the signals that is often measured by remote health monitoring systems is the electroencephalogram (EEG), and the entire system usually consists of a set of lightweight, low-power sensors that transmit the acquired data to a master node or a remote medical provider. Due to the long-term monitoring nature of these systems, the sensors used generate a large amount of data, which poses a challenge for efficient storage and transmission. Therefore, compression of EEG signals plays a vital role in modern remote health monitoring systems to reduce the amount of stored data and transmission time, thereby improving efficiency and reducing the cost of these systems. The utilization of compression technology not only enables efficient data storage and transmission, but also simplifies the allocation of communication time slots, especially in cases involving multiple systems operating in the same frequency band or acquiring different physiological signals simultaneously. At the same time, there is a need to preserve the diagnostic information embedded in the measured ECG signal.

[0003] Standard compression is performed by sampling the signal according to Shannon's theory. This signal is then transformed using an appropriate orthogonal basis (e.g. discrete Fourier, wavelet, multiwavelet, etc.). Only some coefficients of the transformed signal need to be transmitted. This method shows good compression efficiency and low error after reconstruction. However, the main disadvantage of standard compression methods is the computational complexity, which is not conducive to implementation in hardware in low power demand scenarios.

[0004] Compressed Sensing (CS) is a novel compression method that avoids the high computational complexity associated with standard methods. It is a technique that allows signals to be acquired at a sampling rate lower than the Nyquist rate, provided that the signal is sparse or can be approximated as sparse in a certain transform domain. The core of CS is that if the signal has a certain structure (such as sparsity), then the signal can be accurately reconstructed with far fewer measurements than required by traditional sampling theory.

[0005] The signal compression process involves two important matrices:

[0006] Sparse representation matrix Ψ and measurement matrix Φ

[0007] In CS, the original signal x can be expressed as x=Ψs, where Ψ is the basis matrix that maps the signal from the time domain to another domain (usually a sparse domain), that is, the sparse basis selected in the CS method, and s is the obtained sparse representation vector.

[0008] The role of the measurement matrix Φ is to sample the signal at random positions. Φ is a matrix with a dimension of M*N, where M< <N。

[0009] The whole process of compressed sensing can be expressed as y = ΦΨs = As, where y is the result of signal compression and A is the reconstruction matrix. It can be observed that the reconstruction matrix A = ΦΨ. The compression ratio obtained by this method is M / N, which is closely related to the selection and dimension of the measurement matrix.

[0010] The calculation diagram of compressed sensing is as follows Figure 1 .

[0011] However, in practical applications, the measurement matrix needs to meet certain conditions so that the compressed results can be correctly restored. Therefore, some limitations in the process of selecting the measurement matrix will be discussed below, including the relationship with the sparse basis and the minimum limit on the number of measurements M.

[0012] In order to ensure that the signal can be recovered from a small number of measurements, the measurement matrix Φ must be sufficiently uncorrelated with the basis matrix Ψ. In other words, the rows of the measurement matrix cannot sparsely represent the columns of the basis matrix. The uncorrelation can be measured by calculating the maximum correlation μ(Φ,Ψ) between the two matrices, which is defined as the maximum modulus of the inner product between all basis vectors and the measurement vector. The coherence calculation formula is:

[0013]

[0014] Coherence μ(Φ,Ψ)∈(0,1). Through coherence, it can be determined whether there is a unique solution. If equation (2) holds, then according to (3), there is a high probability that there is a unique solution for the M samples obtained from the signal.

[0015]

[0016] where k represents the sparsity of s after the sparse transformation by x = Ψs. Equation (3) shows a clear relationship: as the coherence decreases, the number of samples M required for sparse reconstruction also decreases.

[0017]

[0018] In addition to incoherence, more general conditions for a unique solution are based on the restricted isometry property (RIP) and the null space property (NSP). Based on the NSP, it can be said that if a matrix A is suitable for CS, then it needs to obtain at least twice as many samples as its sparsity, i.e.

[0019] M ≥ 2k (4)

[0020] The RIP condition determines whether the matrix A is suitable for CS. The RIP condition is as follows

[0021]

[0022] Where δk∈R is the RIP constant. When δK<1 holds, the matrix A satisfies the RIP condition.

[0023] The comparison results of the compression effects of the four commonly used measurement matrices: random Gaussian matrix, random Bernoulli matrix, partial Hadamard matrix, and sparse binary matrix are shown in Table 1 below.

[0024] Table 1: Comparison of the compression effects of four measurement matrices

[0025]

[0026] It can be seen from the table that the performance of the four measurement matrices is at the same level. In the actual acquisition and compression process, it takes a lot of computing resources to generate the four matrices online. The preferred solution is to consider generating a certain measurement matrix offline and store it in the acquisition and compression device.

[0027] Table 2: Measurement matrix hardware implementation features

[0028]

[0029] By further comparing the common measurement matrix selection schemes, the random Gaussian matrix requires large storage resources, and the resulting matrix multiplication operation is a floating-point operation, which consumes a lot of computing resources. Although the random Bernoulli matrix has only two values ​​for its elements, like the random Gaussian matrix, the resulting matrix multiplication operation is a floating-point operation. The elements of some Hadamard matrices are all -1 or 1, and the calculation type in the matrix multiplication calculation is addition or subtraction. For sparse random binary matrices, the number of matrix elements 1 is limited. During the matrix multiplication process, it is not necessary to perform calculations every time. It is only necessary to add the data when the matrix element is 1, and no data operation is performed when the matrix element is 0. Compared with the random Bernoulli matrix and some Hadamard matrices, the amount of calculation and computational complexity are lower. However, in the scenario of hardware implementation, it is still necessary to generate random matrices every time the signal is compressed, which is not conducive to reducing the power consumption of the hardware.

[0030] It should be noted that the information disclosed in the above background technology section is only used for understanding the background of the present application, and therefore may include information that does not constitute prior art known to ordinary technicians in the field. Summary of the invention

[0031] The main purpose of the present invention is to solve the problems existing in the above-mentioned background technology and to provide a high-compression EEG signal processing method that is easy to implement in hardware.

[0032] To achieve the above object, the present invention adopts the following technical solutions:

[0033] In a first aspect of the present invention, a high compression rate electroencephalogram signal processing method that is easy to implement by hardware comprises the following steps:

[0034] S1. Data acquisition: obtaining raw EEG signals;

[0035] S2, sparse transformation: convert the EEG signal into a sparse domain through sparse transformation and extract sparse coefficients;

[0036] S3, measurement matrix generation: constructing an initial correlation matrix, vertically and horizontally expanding the initial correlation matrix to reduce coherence and meet the requirements of compressed sensing, and generating an expanded matrix for compressed sensing through nested operations;

[0037] S4, compressed sensing: Use the generated expansion matrix to compress the sparse coefficients to obtain a compressed signal representation.

[0038] Furthermore, in step S2, the sparse transform includes discrete Fourier transform DFT, discrete cosine transform DCT and discrete wavelet transform, preferably discrete cosine transform DCT.

[0039] Furthermore, in step S2, a sparse domain for performing sparse transformation on the EEG signal is selected to satisfy the condition M≥2k, where M is the number of rows of the measurement matrix and k is the sparsity.

[0040] Furthermore, step S2 also includes: after discrete cosine transform, determining the maximum value among the sparse coefficients corresponding to the original EEG signal; setting the sparse coefficients of a given proportion a that is less than the maximum sparse coefficient to 0, wherein a is selected so that it has no effect on the recovery performance of the compressed signal or the effect is negligible.

[0041] Furthermore, in step S3, the initial correlation matrix is ​​expanded vertically and horizontally, which specifically includes the following sub-steps:

[0042] S3.1. Construct the initial incidence matrix: define a finite set X with v elements, select all possible ways of v-1 elements from X to form a subset family P, and construct an incidence matrix H. The dimension of the incidence matrix H is v×v, and the elements are binary values, indicating the relationship between the elements.

[0043] S3.2, vertical expansion: vertically expand the correlation matrix H to reduce the coherence. The specific operation is: calculate the maximum inner product value of each column in the matrix H, determine the number of rows to be expanded based on the maximum value, and add "1" to the corresponding position of the matrix to generate a new matrix U;

[0044] S3.3, horizontal expansion: Horizontally expand the matrix U to further reduce the coherence while keeping the coherence value unchanged. The specific operation is: symmetrically apply the operation on the columns in the vertical expansion to the rows, calculate the maximum inner product value of each row in the matrix U, and adjust the rows according to the maximum value to generate a new matrix V. The dimension of the matrix V is m×r, where the number of columns r of the matrix has an upper bound, and the upper bound is related to the initial parameter v defined in S3.1.

[0045] Furthermore, in step S3, an extended matrix for compressed sensing is generated through nested operations, specifically including: using a binary matrix V, which is composed of multiple column vectors, each column vector represents a subset; for each column vector, its element 1 is replaced by a different row of another matrix B, and the element 0 is replaced by an all-zero row vector, thereby generating an extended matrix; the matrix B is a partial Hadamard matrix, which is used to ensure that the extended matrix meets the requirements of compressed sensing.

[0046] Further, in step S3, the partial Hadamard matrix is ​​generated by the following method, specifically including:

[0047] Determine the order: Choose an order n that is not less than the required number of rows d and is a multiple of 1, 2, or 4 to determine the size of the Hadamard matrix;

[0048] Construct the basic matrix: create a Hadamard matrix of order 1, and then recursively construct Hadamard matrices of higher order according to the rules;

[0049] Expanding the matrix: Recursively construct a Hadamard matrix of order n by copying and modifying the existing Hadamard matrix of order;

[0050] Extract Partial Matrix: Extract the first d rows from the order-n Hadamard matrix to generate the required partial Hadamard matrix.

[0051] Furthermore, by selecting the dimension parameter v of the correlation matrix H, the generated extended matrix is ​​controlled to have different matrix dimension sizes and coherence values.

[0052] In a second aspect of the present invention, a computer-readable storage medium stores a computer program, which, when executed by a processor, implements the high-compression EEG signal processing method that is easy to implement in hardware.

[0053] In a third aspect of the present invention, a computer program product includes a computer program, and when the computer program is executed by a processor, the high-compression EEG signal processing method that is easy to implement in hardware is implemented.

[0054] The present invention has the following beneficial effects:

[0055] The present invention proposes an EEG signal processing solution suitable for scenarios with actual hardware requirements. By adopting an extended measurement matrix based on a combination design of the Hadamard matrix in the EEG compression process, it can be applied to EEG signal compression scenarios while meeting the basic restrictions of compressed sensing, thereby improving the compression rate of EEG signals and reducing power consumption in hardware implementation scenarios.

[0056] The present invention applies a new theory of compressed sensing measurement matrix to EEG signal compression in actual scenarios. Under the basic constraints of compressed sensing, a higher compression rate can be achieved compared to existing technical solutions. In addition, the random sparse matrix involved in the existing mainstream technical solutions needs to be generated online during each signal processing process. This process consumes a lot of hardware resources, making it difficult to apply it to hardware in low-power demand scenarios to bring practical benefits. The EEG signal processing solution proposed in the present invention helps to solve the balance between high compression rate and hardware friendliness, and improves the performance level of existing EEG signal processing solutions on hardware.

[0057] Compared with the existing technical solutions, the high-compression EEG signal processing solution suitable for hardware implementation proposed in the present invention can improve the compression rate of EEG signals and reduce the consumption of hardware resources in the hardware implementation scenario.

[0058] First, by applying a new type of measurement matrix that has been proven feasible in the field of mathematical theory to the actual scenario of EEG signal compression, the present invention solves the limitation problem that the number of measurements M of the original measurement matrix is ​​too small compared to the sparsity K of the EEG signal itself, which makes it impossible to meet the theoretical conditions of compressed sensing application. This improves the compression rate in the process of EEG signal compression processing.

[0059] By further comparing the common measurement matrix selection schemes, the random Gaussian matrix requires large storage resources, and the resulting matrix multiplication operation is a floating-point operation, which consumes a lot of computing resources. Although the random Bernoulli matrix has only two values ​​for its elements, like the random Gaussian matrix, the resulting matrix multiplication operation is a floating-point operation. The elements of some Hadamard matrices are all -1 or 1, and the calculation type in the matrix multiplication calculation is addition or subtraction. For sparse random binary matrices, the number of matrix elements 1 is limited. During the matrix multiplication process, it is not necessary to perform calculations every time. It is only necessary to add the data when the matrix element is 1, and no data operation is performed when the matrix element is 0. Compared with the random Bernoulli matrix and some Hadamard matrices, the amount of calculation and computational complexity are lower. However, in the scenario of hardware implementation, it is still necessary to generate random matrices every time the signal is compressed, which is not conducive to reducing the power consumption of the hardware. The EEG signal compression processing flow designed in the present invention can store the determined measurement matrix in the hardware processing module in advance, thereby eliminating the need to consume hardware resources for matrix generation during actual operation, and facilitating the implementation of the EEG signal compression algorithm in the hardware low-power consumption scenario required by actual needs.

[0060] Other beneficial effects of the embodiments of the present invention will be further described below. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 Schematic diagram of compressed sensing calculation.

[0062] Figure 2 This is a flow chart of the method for processing EEG signals according to an embodiment of the present invention.

[0063] Figure 3 The variation of sparsity k with respect to X is shown. DETAILED DESCRIPTION

[0064] The following is a detailed description of the embodiments of the present invention. It should be emphasized that the following description is only exemplary and is not intended to limit the scope and application of the present invention.

[0065] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of the embodiments of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0066] In the existing methods of processing EEG signals using compressed sensing, the measurement matrix used is difficult to simultaneously meet the requirements of high compression rate and low power consumption hardware implementation. As a result, in actual EEG monitoring scenarios, the monitoring equipment has high energy consumption and heat generation, which is not conducive to the monitored person carrying it with them, and the signal transmission efficiency is poor due to storage space limitations. In view of the above shortcomings, the present invention proposes an EEG signal processing solution suitable for actual hardware demand scenarios. By adopting a measurement matrix based on a combination design and expansion of the Hadamard matrix in the EEG compression process, it can be applied to EEG signal compression scenarios while meeting the basic restrictions of compressed sensing, thereby improving the compression rate of EEG signals and reducing power consumption in hardware implementation scenarios.

[0067] See also Figure 2 The embodiment of the present invention provides a high compression rate electroencephalogram signal processing method that is easy to implement by hardware, comprising the following steps:

[0068] Step S1, data acquisition: obtaining original EEG signals.

[0069] Step S2, sparse transformation: convert the EEG signal into a sparse domain through sparse transformation and extract sparse coefficients.

[0070] In some embodiments, in step S2, the sparse transform includes discrete Fourier transform DFT, discrete cosine transform DCT and discrete wavelet transform, and a preferred embodiment uses discrete cosine transform DCT.

[0071] In some embodiments, in step S2, a sparse domain for performing sparse transformation on the EEG signal is selected to satisfy the condition M≥2k, where M is the number of rows of the measurement matrix and k is the sparsity.

[0072] In a preferred embodiment, step S2 also includes: after discrete cosine transform, determining the maximum value among the sparse coefficients corresponding to the original EEG signal; setting the sparse coefficients that are less than a given proportion a of the maximum sparse coefficient to 0, wherein a is selected so that it has no effect on the recovery performance of the compressed signal or the effect is negligible.

[0073] Step S3, measurement matrix generation: construct an initial correlation matrix, vertically and horizontally expand the initial correlation matrix to reduce coherence and meet compressed sensing requirements, and generate an expanded matrix for compressed sensing through nested operations.

[0074] In some embodiments, in step S3, the initial correlation matrix is ​​vertically and horizontally expanded, specifically including the following sub-steps:

[0075] S3.1. Construct the initial incidence matrix: define a finite set X with v elements, select all possible ways of v-1 elements from X to form a subset family P, and construct an incidence matrix H. The dimension of the incidence matrix H is v×v, and the elements are binary values, indicating the relationship between the elements.

[0076] S3.2, vertical expansion: vertically expand the correlation matrix H to reduce the coherence. The specific operation is: calculate the maximum inner product value of each column in the matrix H, determine the number of rows to be expanded based on the maximum value, and add "1" to the corresponding position of the matrix to generate a new matrix U;

[0077] S3.3, horizontal expansion: Horizontally expand the matrix U to further reduce the coherence while keeping the coherence value unchanged. The specific operation is: symmetrically apply the operation on the columns in the vertical expansion to the rows, calculate the maximum inner product value of each row in the matrix U, and adjust the rows according to the maximum value to generate a new matrix V. The dimension of the matrix V is m×r, where the number of columns r of the matrix has an upper bound, and the upper bound is related to the initial parameter v defined in S3.1.

[0078] In some embodiments, further, in step S3, an extended matrix for compressed sensing is generated through nested operations, specifically including: using a binary matrix V, which is composed of multiple column vectors, each column vector represents a subset; for each column vector, its element 1 is replaced by a different row of another matrix B, and the element 0 is replaced by an all-zero row vector, thereby generating an extended matrix; the matrix B is a partial Hadamard matrix, which is used to ensure that the extended matrix meets the requirements of compressed sensing.

[0079] In some embodiments, further, in step S3, the partial Hadamard matrix is ​​generated by the following method, specifically including:

[0080] Determine the order: Choose an order n that is not less than the required number of rows d and is a multiple of 1, 2, or 4 to determine the size of the Hadamard matrix;

[0081] Construct the basic matrix: create a Hadamard matrix of order 1, and then recursively construct Hadamard matrices of higher order according to the rules;

[0082] Expanding the matrix: Recursively construct a Hadamard matrix of order n by copying and modifying the existing Hadamard matrix of order;

[0083] Extract Partial Matrix: Extract the first d rows from the order-n Hadamard matrix to generate the required partial Hadamard matrix.

[0084] In some embodiments, by selecting the dimension parameter v of the correlation matrix H, the generated expansion matrix is ​​controlled to have different matrix dimension sizes and coherence values.

[0085] Step S4, compressed sensing: compress the sparse coefficients using the generated expansion matrix to obtain a compressed signal representation.

[0086] The present invention provides an electroencephalogram signal processing method with high compression rate and easy hardware implementation. By adopting a measurement matrix extended by a combination design based on a Hadamard matrix, not only the basic restriction constraints of compressed sensing are met, but also the compression rate of electroencephalogram signals is significantly improved, and the power consumption in the hardware implementation scenario is reduced. The method solves the limitation problem of the original measurement matrix in the process of theory-practice migration by applying a new theory of compressed sensing measurement matrix to practical scenarios, and achieves a higher compression rate. Compared with the prior art, the present invention reduces the consumption of hardware resources, and avoids resource consumption in the signal processing process by storing the determined measurement matrix in the hardware processing module in advance, thereby reducing power consumption. In addition, the measurement matrix design of the present invention reduces storage requirements and calculation amount, optimizes calculation efficiency, and reduces calculation complexity by using addition or subtraction instead of floating-point operations. This method is more friendly in hardware implementation, particularly suitable for low-power demand scenarios, and improves the performance level of electroencephalogram signal processing solutions in hardware.

[0087] The specific embodiments of the present invention are further described below.

[0088] The present invention proposes a selection scheme for the measurement matrix in the EEG signal compression processing scheme. The present invention proposes an effective solution for the extended matrix involved in the EEG signal compression scenario in order to simultaneously meet the measurement matrix condition restrictions and measurement number restrictions required by the compressed sensing technology. The embodiment of the present invention also proposes an evaluation and processing method for EEG signals using discrete cosine transform (DCT).

[0089] The processing flow of the EEG signal in the embodiment of the present invention is summarized as follows:

[0090] 1. Perform sparse transformation on EEG signals under discrete cosine transform (DCT).

[0091] 2. Set the coefficients that are less than a certain proportion of the maximum value of the sparse coefficient to 0. This operation does not affect the recovery performance of the compressed signal or the impact can be ignored.

[0092] 3. The sparse basis obtained in the process is operated with the generated extended matrix. Due to the discussion and improvement of the limitations of the extended matrix in actual application scenarios in the present invention, it can be considered that the signal meets the recovery conditions after compression and can be restored normally.

[0093] 4. Obtain the compressed EEG signal. The involved operation process is convenient to implement in the scenario of low hardware power consumption, and the obtained compressed signal has a high compression ratio.

[0094] The present invention applies the measurement matrix extended from the combination design based on the Hadamard matrix (hereinafter simply referred to as the extended matrix) to the scenario of EEG signal compression for implementation.

[0095] The existing construction method of the measurement matrix and the matrix satisfy the above constraints, but the problem that still needs to be solved in the actual application scenario of EEG signals is the selection problem of the measurement number M. It has been introduced in the background technology that M needs to satisfy M≥2k, where k represents the sparsity of the EEG signal, that is, the number of non-zero elements of the signal, and M represents the measurement number M in the dimension M*N of the measurement matrix φ. The actual EEG signals are usually continuously collected and have poor sparsity, so the extended matrix cannot be directly used for compression implementation.

[0096] The high-compression-ratio EEG signal processing method that is easy to implement in hardware proposed by the present invention has the overall flowchart as Figure 2 shown.

[0097] The generation process of the extended matrix is briefly described as follows:

[0098] 1. Let X=(x1,x2,...,x v ) be a finite set containing v (v≥2), and all subsets of X formed by arbitrarily selecting v - 1 elements from the set X are P={P1,P2,...,P v )}, where P1,P2,...,P v represent v blocks. Define its incidence matrix as a v*v-dimensional binary matrix H=(h i,j ), and the specific composition rule is as follows: if x i ∈P j , then the corresponding h i,j = 1, otherwise the corresponding h i,j = 0. The obtained matrix is defined as the initial incidence matrix H, and its coherence is 1 - 1 / (v - 1). Since the coherence approaches 1 when v increases, the incidence matrix H cannot be used as the measurement matrix, and the matrix H needs to be extended.

[0099] 2. Vertical extension: If v≥4, for 1≤j≤v, let λj = max 1≤j<j1≤n 〈h j ,h j1 〉, then H should be extended by λ j-1 rows. For 1≤i<i1≤v, if H i,j = H i,j1 = H i1,j = Hi1,j1 =1, then let H i1,j1 =0. If <h j ,h j1 〉=λ j , then λ should be expanded j -1 hour j1 The last row is "1"; if <h j ,h j1 〉<λ j , then it should be in h j1 The end of 〈hj,hj1〉 is extended to 〈hj,hj1〉-1 rows of "1"; the rest of the elements are "0". This process continues until j=v-1. The operator <>d mentioned in the process represents the inner product operation of the elements. The resulting new matrix is ​​represented by U. The dimension of the resulting matrix U is m*v (m=v 2 -3v+3)-dimensional binary matrix with a coherence value of 1 / (v-1).

[0100] 3. Horizontal expansion:

[0101] Following the same idea as vertical expansion, horizontal transformation can be completed by only performing symmetric transformation on the operations of row and column elements. During the transformation, the coherence value remains unchanged at 1 / (v-1). The transformed matrix is ​​denoted as V, with a dimension of m*r, where r has an upper bound, i.e. r≤mv / (v-1).

[0102] 4. Nested operations:

[0103] Let V = (v1, v2, ..., vr) be an m × r binary matrix with fixed column weights d = v-1, where vj represents the jth column of the matrix V, j = 1, 2, ..., r. Let B be a d*n1-dimensional matrix, for a certain column vj, the element 1 in vj is replaced by a different row in the matrix B. The element 0 is replaced by a 1*n1-dimensional all-zero row vector (0, 0, ..., 0). The nested matrix is ​​used as the measurement matrix for compressed sensing.

[0104] The selected matrix B is a partial Hadamard matrix of d*n-1, where d=v-1, and v is the initial input parameter. The generation process of the required partial Hadamard matrix is ​​as follows:

[0105] 5. Partial Hadamard Matrix Generation

[0106] 5.1 Determine the order:

[0107] First, find a suitable order n, which should be greater than or equal to the required number of rows d, and n must be 1, 2, or a multiple of 4 (i.e., 4k form). For example, if you need to generate a partial Hadamard matrix with 5 rows, then n should be 8, because 8 is the smallest 4k form number larger than 5.

[0108] 5.2 Construct the basic matrix:

[0109] Let's start with the simplest Hadamard matrix. A rank 1 matrix is ​​a matrix consisting of a single element, whose value is +1.

[0110] A rank 2 matrix is ​​a 2x2 matrix whose top left and top right elements are +1, while the bottom right element is -1 and the bottom left is also +1.

[0111] 5.3 Expand the matrix using recursive construction:

[0112] If you already have a Hadamard matrix of order n, you can construct a matrix of order 2n by copying it. Specifically, copy the original matrix once and place it at the upper left of the new matrix, then copy it again and place it at the upper right of the new matrix. Next, place the original matrix at the lower left of the new matrix and the original matrix at the lower right, but change all +1s to -1s and all -1s to +1s.

[0113] Construct progressively larger matrices:

[0114] For example, if you already have a matrix of order 2, you can use the above method to construct a matrix of order 4. Then, use this 4x4 matrix to construct an 8x8 matrix, and so on, until you reach the required order n.

[0115] Extract part of the matrix:

[0116] Once the complete n×n Hadamard matrix is ​​obtained, it is only necessary to take out the first d rows to obtain a partial Hadamard matrix of dimension d*n.

[0117] Through the above matrix construction steps, a series of extended matrices are obtained. The obtained extended matrices will have different dimensions due to different selections of the initial parameter v. Table 3 shows the dimensions and coherence values ​​of the matrices constructed under different parameter selection conditions.

[0118] Table 3: Relationship between the dimension and coherence of the expansion matrix and the initial parameter v

[0119]

[0120] Under the technical solution of the present invention, when v=18 is selected, that is, the measurement matrix dimension is 273x5460, and the coherence is 3 / 17, then M=273, N=5460, and the compression rate achieved is (NM) / N=95%, which is higher than the compression rate that can be achieved by the prior art. Since the constructed matrix is ​​a deterministic matrix, it can be directly stored in the storage module of the hardware system, reducing the re-consumption of hardware resources during actual operation.

[0121] Selection of sparse domain:

[0122] In order to satisfy the requirement of M≥2k for the number of measurements M, a suitable sparse domain should be selected to perform sparse transformation on the EEG signal, and whether the sparsity k after the transformation meets the requirement should be evaluated.

[0123] Commonly used sparse observation methods include DFT, DCT, discrete wavelet transform, etc., wherein discrete wavelet transform may result in different sparsities due to different sparse basis selections. The present invention evaluates the sparse domain in Table 4.

[0124] Table 4: Comparison of the evaluation of EEG signal sparsity under different sparse bases

[0125]

[0126] The results show that DCT has the best sparse transform effect in the selection of sparse domain of EEG signals, that is, EEG signals have the best sparsity after discrete cosine transform.

[0127] Therefore, the present invention selects discrete cosine transform as the sparse domain used in the EEG signal compression process. In actual operation, discrete cosine transform will generate sparse coefficients corresponding to the original EEG signal, and the obtained sparse coefficients have a maximum value. It can be considered that in actual scenarios, the sparse coefficients that are less than a certain proportion a of the maximum value (in the experimental image, a = X*0.0001) are set to 0. This operation has no effect on the recovery performance of the compressed signal or can be ignored.

[0128] Figure 3 The variation of sparsity k with respect to X is shown.

[0129] Through calculation, we can know that the dimension of the selected measurement matrix is ​​273x5460, and the ratio of the measurement number M is 273 / 5460=0.05. Therefore, if M≥2k is to be satisfied, k≤0.025 is required. From the experimental image, we can see that the selected sparse transformation can meet the above conditions under a certain ratio a.

[0130] The present invention applies a new theory of compressed sensing measurement matrix to EEG signal compression in actual scenarios. Under the basic constraints of compressed sensing, a higher compression rate can be achieved compared to existing technical solutions. In addition, the random sparse matrix involved in the existing mainstream technical solutions needs to be generated online during each signal processing process. This process consumes a lot of hardware resources, making it difficult to apply it to hardware in low-power demand scenarios to bring practical benefits. The EEG signal processing solution proposed in the present invention helps to solve the balance between high compression rate and hardware friendliness, and improves the performance level of existing EEG signal processing solutions on hardware.

[0131] An embodiment of the present invention further provides a storage medium for storing a computer program, which at least performs the above method when executed.

[0132] An embodiment of the present invention further provides a control device, comprising a processor and a storage medium for storing a computer program; wherein the processor is configured to execute at least the method described above when executing the computer program.

[0133] An embodiment of the present invention further provides a processor, wherein the processor executes a computer program and at least executes the method described above.

[0134] The storage medium can be implemented by any type of non-volatile storage device, or a combination thereof. Among them, the non-volatile memory can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a magnetic random access memory (FRAM), a ferromagnetic random access memory (FRAM), a flash memory, a magnetic surface memory, an optical disc, or a compact disc read-only memory (CD-ROM); the magnetic surface memory can be a disk memory or a tape memory. The storage medium described in the embodiments of the present invention is intended to include, but is not limited to, these and any other suitable types of memory.

[0135] In the several embodiments provided by the present invention, it should be understood that the disclosed systems and methods can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0136] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0137] In addition, all functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may be separately used as a unit, or two or more units may be integrated into one unit; the above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0138] Those skilled in the art can understand that: all or part of the steps of implementing the above method embodiments can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiments; and the aforementioned storage medium includes: mobile storage devices, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), disks or optical disks, etc. Various media that can store program codes.

[0139] Alternatively, if the above-mentioned integrated unit of the present invention is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present invention can be essentially or partly reflected in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROM, RAM, magnetic disks or optical disks.

[0140] The methods disclosed in the several method embodiments provided by the present invention can be arbitrarily combined without conflict to obtain new method embodiments.

[0141] The features disclosed in several product embodiments provided by the present invention can be arbitrarily combined without conflict to obtain new product embodiments.

[0142] The features disclosed in several method or device embodiments provided by the present invention can be arbitrarily combined without conflict to obtain new method embodiments or device embodiments.

[0143] The above contents are further detailed descriptions of the present invention in combination with specific preferred embodiments, and it cannot be determined that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art of the present invention, several equivalent substitutions or obvious variations can be made without departing from the concept of the present invention, and the performance or use is the same, which should be regarded as belonging to the protection scope of the present invention.

Claims

1. A high compression rate EEG signal processing method that is easy to implement in hardware, characterized in that: The following steps are involved: S1. Data acquisition: obtaining raw EEG signals; S2, sparse transformation: convert the EEG signal into a sparse domain through sparse transformation and extract sparse coefficients; S3, measurement matrix generation: constructing an initial correlation matrix, vertically and horizontally expanding the initial correlation matrix to reduce coherence and meet the requirements of compressed sensing, and generating an expanded matrix for compressed sensing through nested operations; S4, compressed sensing: Use the generated expansion matrix to compress the sparse coefficients to obtain a compressed signal representation.

2. The high compression rate EEG signal processing method that is easy to implement with hardware as claimed in claim 1, characterized in that: In step S2, the sparse transform includes discrete Fourier transform DFT, discrete cosine transform DCT and discrete wavelet transform, preferably discrete cosine transform DCT.

3. The high compression rate EEG signal processing method that is easy to implement with hardware as claimed in claim 1 or 2, characterized in that: In step S2, a sparse domain for performing sparse transformation on the EEG signal is selected, satisfying the condition M≥2k, where M is the number of rows of the measurement matrix and k is the sparsity.

4. The high compression rate EEG signal processing method that is easy to implement with hardware as claimed in any one of claims 1 to 3, characterized in that: Step S2 also includes: after discrete cosine transform, determining the maximum value among the sparse coefficients corresponding to the original EEG signal; setting the sparse coefficients that are less than a given proportion a of the maximum sparse coefficient to 0, where a is selected so that it has no effect on the recovery performance of the compressed signal or the effect is negligible.

5. The high compression rate EEG signal processing method that is easy to implement by hardware according to any one of claims 1 to 4, characterized in that: In step S3, the initial correlation matrix is ​​expanded vertically and horizontally, which specifically includes the following sub-steps: S3.

1. Construct the initial incidence matrix: define a finite set X with v elements, select all possible ways of v-1 elements from X to form a subset family P, and construct an incidence matrix H. The dimension of the incidence matrix H is v×v, and the elements are binary values, indicating the relationship between the elements. S3.2, vertical expansion: vertically expand the correlation matrix H to reduce the coherence. The specific operation is: calculate the maximum inner product value of each column in the matrix H, determine the number of rows to be expanded based on the maximum value, and add "1" to the corresponding position of the matrix to generate a new matrix U; S3.3, horizontal expansion: Horizontally expand the matrix U to further reduce the coherence while keeping the coherence value unchanged. The specific operation is: symmetrically apply the operation on the columns in the vertical expansion to the rows, calculate the maximum inner product value of each row in the matrix U, and adjust the rows according to the maximum value to generate a new matrix V. The dimension of the matrix V is m×r, where the number of columns r of the matrix has an upper bound, and the upper bound is related to the initial parameter v defined in S3.

1.

6. The high compression rate EEG signal processing method that is easy to implement with hardware as claimed in claim 5, characterized in that: In step S3, an extended matrix for compressed sensing is generated through nested operations, specifically including: using a binary matrix V, which is composed of multiple column vectors, each column vector represents a subset; for each column vector, its element 1 is replaced by a different row of another matrix B, and the element 0 is replaced by an all-zero row vector, thereby generating an extended matrix; the matrix B is a partial Hadamard matrix, which is used to ensure that the extended matrix meets the requirements of compressed sensing.

7. The high compression rate EEG signal processing method that is easy to implement with hardware as claimed in claim 6, characterized in that: In step S3, the partial Hadamard matrix is ​​generated by the following method, specifically including: Determine the order: Choose an order n that is not less than the required number of rows d and is a multiple of 1, 2, or 4 to determine the size of the Hadamard matrix; Construct the basic matrix: create a Hadamard matrix of order 1, and then recursively construct Hadamard matrices of higher order according to the rules; Expanding the matrix: Recursively construct a Hadamard matrix of order n by copying and modifying the existing Hadamard matrix of order; Extract Partial Matrix: Extract the first d rows from the order-n Hadamard matrix to generate the required partial Hadamard matrix.

8. The high compression rate EEG signal processing method that is easy to implement by hardware as claimed in any one of claims 5 to 7, characterized in that: By selecting the dimension parameter v of the correlation matrix H, the generated expansion matrix is ​​controlled to have different matrix dimension sizes and coherence values.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the high-compression EEG signal processing method that is easy to implement with hardware as described in any one of claims 1 to 8 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the high-compression EEG signal processing method that is easy to implement with hardware as described in any one of claims 1 to 8 is implemented.

Images