A data augmentation method and apparatus based on spiking neural networks

Abstract

This invention relates to a data augmentation method and apparatus based on spiking neural networks. The method of this application is used for implantable brain-computer interface data and can extract information distribution characteristics in the case of limited neural information data. Based on the biological properties of spiking neural networks, they are suitable for learning the distribution characteristics of neural information, thereby generating neural signals that conform to the information distribution characteristics, and thus using them as brain information data augmentation. This invention considers the cluster activity characteristics of biological neurons as the basis for data augmentation. Spiking neural networks have biological properties and are suitable for directly generating neural information, thereby enhancing brain-computer interface information, which is of great significance for the research and application of brain-computer interfaces.

Description

Technical Field

[0001] This invention relates to the field of brain-computer interface technology, and more specifically, to a data augmentation method, apparatus, storage medium, and device based on a spiking neural network. Background Technology

[0002] An implantable brain-computer interface (BCI) is a system that allows the brain to interact directly with the outside world, thereby helping patients restore, regulate, and enhance their functions. The key technology of an implantable BCI is to implant sensors in the cerebral cortex within the skull, which can directly extract information transmitted by neurons inside the brain. At the same time, the BCI system needs to be paired with algorithms to interpret neural information, such as decoding the subject's motor intentions, visual information, etc.

[0003] After the implantation of neural electrodes (sensors) in the brain, the neural electrodes' ability to collect neural information gradually decreases over time. The main reasons are as follows: (1) The electrodes can cause some damage to the brain, leading to the growth of glial scar tissue around the electrodes, which blocks the electrode contacts from neurons; (2) The electrode contacts gradually break down in the electrolyte solution environment in the brain, reducing the electrical sensing performance of the electrodes. In addition, for decoding algorithms, the ability to decode neural information gradually decreases over time. The main reasons are as follows: (1) The neural electrodes can shift due to brain shaking, causing changes in the recorded information, and most decoding algorithms cannot adapt to such changes in neural information; (2) The brain's own neural plasticity (the learning process of neurons) leads to the reconnection of neurons and changes in connection strength; therefore, the neural information of the brain will change with the learning process.

[0004] A neural manifold refers to the neural activity pattern of a specific group of neurons. The functions within the brain originate from the collaborative information communication of neuronal groups. A manifold is a space that has Euclidean space properties and is used in mathematics to describe geometric shapes. The activity of neuronal clusters can be represented in a low-dimensional manifold space. Therefore, under the constraints of the manifold space, brain-computer interfaces can use a small number of neural signals to complete the decoding of neural information.

[0005] Spiking neural networks are often hailed as the third generation of artificial neural networks because they simulate the information transmission method of biological neurons. In biological systems, communication between neurons is achieved by transmitting pulse signals, with information contained in the transmitted pulse sequence. In contrast, traditional neural networks use continuous values ​​as the information transmitted between artificial neurons, and their model architecture is artificially designed and generally lacks biological functional characteristics. Therefore, biologically inspired spiking neural networks employ spiking neurons and synaptic models, and can use biologically-based learning methods to update synaptic weights. As a result, biologically-based spiking neural network models are very suitable for studying brain information modeling and intelligent reasoning tasks.

[0006] Data augmentation is a technique that artificially expands a training dataset by generating more equivalent data from a limited dataset. Brain-computer interfaces (BCIs) require a large amount of brain information data to build brain information decoding models, but acquiring such large amounts of data is too costly. Data is a crucial factor in improving BCI decoding, and data augmentation is an effective means of overcoming insufficient training data and is widely used in various fields of artificial intelligence.

[0007] Therefore, existing technologies still have shortcomings, and it is necessary to further develop data augmentation technologies. Summary of the Invention

[0008] This invention provides a data augmentation method and apparatus based on a spiking neural network to address the problem of data augmentation when training data is insufficient.

[0009] According to an embodiment of the present invention, a data augmentation method based on a spiking neural network is provided, comprising the following steps:

[0010] Acquire raw data from the implantable brain-computer interface, including neuronal populations, and convert the pulse sequences of the neuronal populations into pulse firing rate data;

[0011] Dimensionality reduction was performed on the pulse firing rate data to obtain the neuronal population activity patterns in the low-dimensional manifold space;

[0012] A spiking neural network model was established, and the manifold spatial activity patterns of real neuron populations were used as the objective function for supervised learning.

[0013] During the learning and iteration process of the spiking neural network model, the dimensionality of the pulse sequence information generated by the spiking neural network model is reduced to obtain the neuron population activity law that is similar to the real neuron population activity law.

[0014] After the spiking neural network model has completed its iterations, perturbation neurons are set in the spiking neural network model. Based on the noise signals generated by the perturbation neurons, neural pulse information data with noise that conforms to the laws of real biological neural activity is created and output.

[0015] Furthermore, the raw data also includes neuron population motion location information, which is then Gaussian smoothed to remove some noise data.

[0016] Furthermore, a spiking neural network model is established, using the manifold spatial activity patterns of a real neuron population as the objective function, and supervised learning is performed, including:

[0017] The spiking neural network model is based on the leak-integrated release model, which is as follows:

[0018]

[0019]

[0020] u i =u rest u i >u threshold

[0021] Among them, u i τ represents the magnitude of the membrane potential of a neuron, which is related to the firing of cell pulses; m The time constant, expressed as a differential equation, is used to control the magnitude of the change in membrane potential over time; u rest I is a constant parameter representing the resting potential of the cell membrane, that is, the magnitude of the membrane potential of the cell in its resting state; i (t) represents the input current, which, as an external input, affects the cell membrane potential; R is the cell membrane impedance; τ s u is the synaptic time constant; threshold This represents the pulse firing threshold, when the neuron's membrane potential u... i More than u threshold A pulse is emitted at regular intervals, and the membrane potential u is... i Reset to u rest .

[0022] Furthermore, a spiking neural network model is established, using the manifold spatial activity patterns of a real neuron population as the objective function, and supervised learning is performed, including:

[0023] Supervised learning is performed instead of gradient learning to update the weights of the neural network model; the alternative to gradient learning is described below:

[0024] S i [n]∝Θ(ui [n]-u threshold )

[0025] Among them, S i [n] represents the pulse sequence, and Θ(x) represents the Hevichite step function. Due to the non-differentiable nature of pulse signals, during back propagation, the Θ(x) function is replaced by σ(x), thus S i [n]∝σ(u i [n]-u threshold ),in

[0026] Furthermore, the raw data of the implanted brain-computer interface is acquired, including neuronal populations. The pulse sequence of the neuronal population is converted into pulse firing rate data, specifically as follows:

[0027] The pulse sequence of each neuron in the neuronal population is processed by a sliding window to convert it into pulse firing rate data.

[0028] Furthermore, during the learning iteration process of the spiking neural network model, the dimensionality of the pulse sequence information generated by the spiking neural network model is reduced to obtain the specific neuron population activity law that approximates the real neuron population activity law:

[0029] The mean squared error is used as the loss function. The mean squared error is the sum of the squares of the errors between corresponding points in the predicted data and the original data. The formula for the mean squared error is:

[0030]

[0031] Among them, y i For the dimensionality reduction data output of a spiking neural network, Let n be the objective function and n be the number of samples.

[0032] Furthermore, after the spiking neural network model has completed its iterations, perturbation neurons are added to the model. Based on the noise signals generated by these perturbation neurons, neural pulse information data that is noisy yet conforms to the laws of real biological neural activity is created and output as follows:

[0033] In the spiking neural network model, some neurons are based on Poisson neurons and are set as perturbation neurons. The model of the Poisson neuron is as follows:

[0034] Using the LIF neuron model:

[0035]

[0036] I i The input current is converted from a random pulse signal, which follows a Poisson distribution.

[0037]

[0038] r represents the neuron's firing rate, P T [n] represents the probability that the nth pulse of a neuron fires within a duration T;

[0039] Implemented within the program, the calculation can be simplified, and the firing probability of the pulse within the time interval Δt can be rΔt, where x rand It is a random variable with values ​​between 0 and 1;

[0040]

[0041] A data augmentation device based on a spiking neural network includes:

[0042] The pulse conversion module is used to acquire the raw data of the implantable brain-computer interface, including the neuron population, and convert the pulse sequence of the neuron population into pulse firing rate data.

[0043] The first dimensionality reduction module is used to reduce the dimensionality of the pulse firing rate data to obtain the neuron population activity patterns in the low-dimensional manifold space.

[0044] The supervised learning module is used to build a spiking neural network model, using the manifold spatial activity patterns of real neuron populations as the objective function for supervised learning.

[0045] The second dimensionality reduction module is used to reduce the dimensionality of the pulse sequence information generated by the spiking neural network model during the learning iteration process, so as to obtain the neuron population activity law that is similar to the real neuron population activity law.

[0046] The data output module is used to set up perturbation neurons in the spiking neural network model after the model has completed its iteration. Based on the noise signals generated by the perturbation neurons, it creates and outputs neural pulse information data that is noisy but conforms to the laws of real biological neural activity.

[0047] A computer-readable medium, a computer-readable storage medium storing one or more programs, which can be executed by one or more processors to implement the steps in the data augmentation method based on a spiking neural network as described in any of the preceding claims.

[0048] A terminal device includes: a processor, a memory, and a communication bus; the memory stores a computer-readable program that can be executed by the processor;

[0049] The communication bus enables communication between the processor and memory;

[0050] When the processor executes a computer-readable program, it implements the steps in any of the above-described data augmentation methods based on spiking neural networks.

[0051] This invention proposes a data augmentation method and apparatus based on spiking neural networks. The method is used for implantable brain-computer interface (BCI) data and can extract information distribution features even with limited neural information data. Based on the biological properties of spiking neural networks, they are suitable for learning the distribution features of neural information, thereby generating neural signals that conform to these distribution features, which serve as brain information data augmentation. This invention considers the cluster activity characteristics of biological neurons as the basis for data augmentation. Spiking neural networks possess biological properties and are suitable for directly generating neural information, thereby enhancing BCI information, which is of great significance for the research and application of BCIs. Attached Figure Description

[0052] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:

[0053] Figure 1 This is a flowchart of the data augmentation method based on spiking neural networks according to the present invention;

[0054] Figure 2 This refers to the neural signals and corresponding motor control acquired by the implantable brain-computer interface of this invention.

[0055] Figure 3 The neural signals acquired by the implantable brain-computer interface of this invention in the corresponding experiment are represented in a low-dimensional manifold space;

[0056] Figure 4 This is a schematic diagram of the spiking neural network architecture designed in this invention;

[0057] Figure 5 A schematic diagram of brain-computer interface data generated by the spiking neural network designed in this invention;

[0058] Figure 6 This is a schematic diagram of the data augmentation device based on a spiking neural network according to the present invention;

[0059] Figure 7 This is the terminal device of the present invention. Detailed Implementation

[0060] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0061] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0062] See Figure 1 According to an embodiment of the present invention, a data augmentation method based on a spiking neural network is provided, comprising the following steps:

[0063] The S100 acquires raw data from the implantable brain-computer interface, including neuronal populations, and converts the pulse sequences of the neuronal populations into pulse firing rate data.

[0064] The raw data includes neuronal firing information extracted from the motor cortex, motion position and speed information corresponding to the subject's motor control, and labels corresponding to the type of motor control test; the pulse sequence of each neuron in the neuronal population related to brain-computer interface movement is processed by sliding window to convert it into neuronal pulse firing rate data; Gaussian smoothing is applied to the motion position information to remove some noise data; a moving average filter can downsample the neuronal population signal of the raw signal to improve the signal-to-noise ratio; a Gaussian filter is used to remove noise, also used in digital signal processing to remove noise from the motion position of the raw information; such as Figure 2 The image shows the neural signals and corresponding motor control acquired by the implanted brain-computer interface.

[0065] The following is a moving average filter for vector x defined by the difference equation:

[0066]

[0067] The moving average filter moves the data along a window of length equal to the window size and calculates the average value of the data contained in each window.

[0068] The following is a one-dimensional Gaussian filter used to smooth the motion positions in the original information:

[0069]

[0070] x is the signal input, and σ is the standard deviation.

[0071] S200 performs dimensionality reduction on the pulse firing rate data to derive the neuronal population activity patterns in the low-dimensional manifold space.

[0072] Dimensionality reduction was performed on neural impulse data collected by brain-computer interfaces for the same task experiment to derive the neuronal activity patterns in the low-dimensional manifold space. Specifically:

[0073] For data from the same experiment, principal component analysis (PCA) is first performed. Simultaneously, the cumulative explained variance ratio is used to determine the variance explained by each principal component, thus deriving the number of statistically significant principal components. This number serves as the dimension representing the neuronal population activity, thereby revealing the patterns of neuronal population activity in the low-dimensional manifold space. For example... Figure 3 As shown, the neural signals acquired by the implanted brain-computer interface in the corresponding experiment are represented in a low-dimensional manifold space.

[0074] The specific steps of the PCA algorithm are explained in detail below:

[0075] We first have m rows and n columns of data;

[0076] Step 1: Arrange the original data into an n x m matrix X;

[0077] Step 2: Zero-mean normalize each row of X;

[0078] Step 3: Calculate the covariance matrix

[0079] Step 4: Find the eigenvalues ​​and corresponding eigenvectors of the covariance matrix;

[0080] Step 5: Arrange the eigenvectors into a matrix from top to bottom according to the size of their corresponding eigenvalues, and take the first K rows to form matrix P;

[0081] Step 6: Y = PX is the data after dimensionality reduction to K dimensions.

[0082] The S300 establishes a spiking neural network model, using the manifold spatial activity patterns of a real neuron population as the objective function for supervised learning.

[0083] This invention uses a spiking neural network as the model for generating this data. The neuron model is based on the Leaky Integrate and Fire Model (LIF). The learning algorithm is based on supervised learning, using an algorithm called Surrogate Gradient Learning to update the weights of the neural network model; for example... Figure 4 The diagram shown is a schematic of the spiking neural network architecture designed in this invention.

[0084] The leaked integration and distribution model is as follows:

[0085]

[0086]

[0087] u i =u rest u i >u threshold

[0088] Among them, u i τ represents the magnitude of the membrane potential of a neuron, which is related to the firing of cell pulses; m The time constant, expressed as a differential equation, is used to control the magnitude of the change in membrane potential over time; u rest I is a constant parameter representing the resting potential of the cell membrane, that is, the magnitude of the membrane potential of the cell in its resting state; i (t) represents the input current, which, as an external input, affects the cell membrane potential; R is the cell membrane impedance; τ s u is the synaptic time constant; threshold This represents the pulse firing threshold, when the neuron's membrane potential u... i More than u threshold A pulse is emitted at regular intervals, and the membrane potential u is... i Reset to u rest .

[0089] The alternative gradient learning algorithm is described as follows:

[0090] S i [n]∝Θ(u i [n]-u threshold )

[0091] Among them, S i[n] represents the pulse sequence, and Θ(x) represents the Hevichite step function. Due to the non-differentiable nature of pulse signals, during back propagation, the Θ(x) function is replaced by σ(x), thus S i [n]∝σ(u i [n]-u threshold ),in

[0092] S400: During the learning iteration of the spiking neural network model, the dimensionality of the pulse sequence information generated by the spiking neural network model is reduced to obtain the neuron population activity law that is similar to the real neuron population activity law.

[0093] In the learning and iteration process of a neural network model, the pulse sequence generated by the output layer of a spiking neural network needs to be dimensionality reduced. The dimensionality-reduced data is used as the output, the dimensionality-reduced activity pattern of the real neuron population is used as the objective function, and the mean square error (MSE) is used as the loss function. The training results must satisfy the requirement that the activity pattern of the neuron population in the output layer is approximately similar to that of the real neuron population. The mean square error loss, also known as the quadratic loss or L2 loss, is often used in regression prediction tasks. The mean square error function measures the quality of the model by calculating the square of the distance (i.e., error) between the predicted value and the actual value. That is, the closer the predicted value and the actual value are, the smaller the mean square error between them.

[0094] The mean square error is the average of the sum of squares of the errors between corresponding points in the predicted data and the original data, as shown in the following formula:

[0095]

[0096] Among them, y i For the dimensionality reduction data output of a spiking neural network, Let n be the objective function and n be the number of samples.

[0097] S500: After the spiking neural network model has completed its iteration, perturbation neurons are set in the spiking neural network model. Based on the noise signals generated by the perturbation neurons, neural pulse information data with noise that conforms to the laws of real biological neural activity is created and output.

[0098] After the spiking neural network completes its iterations, some neurons in the output layer are activated and set as perturbation neurons, based on Poisson neurons. The neural impulse information of biological neurons is based on a Poisson distribution; in this application, LIF neurons generate Poisson-distributed signals. Random impulse signals are input into the perturbation neurons, thereby causing them to generate noise signals. Thus, the output layer can generate neural impulse information data with noise that conforms to the laws of real biological neural activity. Figure 5The diagram shown is a schematic of brain-computer interface data generated by the spiking neural network designed in this invention.

[0099] The model of the Poisson neuron is as follows:

[0100] Using the LIF neuron model:

[0101]

[0102] I i The input current is converted from a random pulse signal, which follows a Poisson distribution.

[0103]

[0104] r represents the neuron's firing rate, P T [n] represents the probability that the nth pulse of a neuron fires within a duration T;

[0105] Implemented within the program, the calculation can be simplified, and the firing probability of the pulse within the time interval Δt can be rΔt, where x rand It is a random variable with values ​​between 0 and 1;

[0106]

[0107] The data generated according to this invention can be further used in fields such as brain-computer interface information decoding and brain-like intelligence research. The method of this application is specifically designed for implantable brain-computer interface data, and can extract information distribution characteristics even with limited neural information data. Based on the biological properties of spiking neural networks, it is suitable for learning the distribution characteristics of neural information, thereby generating neural signals that conform to the information distribution characteristics, thus serving as brain information data enhancement.

[0108] This invention considers the cluster activity characteristics of biological neurons as the basis for data augmentation. Simultaneously, spiking neural networks possess biological properties, making them suitable for directly generating neural information, thereby enhancing brain-computer interface (BCI) information, which is of great significance for the research and application of BCIs.

[0109] This invention takes into account the synergistic nature of brain-computer interfaces and brain-like intelligent algorithms, and the enhanced data is more in line with the internal laws of neural information, resulting in data that is more biologically interpretable.

[0110] See Figure 6 According to another embodiment of the present invention, a data augmentation device based on a spiking neural network is provided, comprising:

[0111] The pulse conversion module 100 is used to acquire the raw data of the implantable brain-computer interface, including the neuron population, and convert the pulse sequence of the neuron population into pulse firing rate data.

[0112] The first dimensionality reduction module 200 is used to reduce the dimensionality of the pulse firing rate data to obtain the neuron population activity patterns in the low-dimensional manifold space.

[0113] The supervised learning module 300 is used to build a spiking neural network model, using the manifold spatial activity patterns of the real neuron population as the objective function for supervised learning.

[0114] The second dimensionality reduction module 400 is used to reduce the dimensionality of the pulse sequence information generated by the spiking neural network model during the learning iteration process of the spiking neural network model, so as to obtain the neuron population activity law that is similar to the real neuron population activity law.

[0115] The data output module 500 is used to set perturbation neurons in the spiking neural network model after the model iteration is completed, and to create and output neural pulse information data with noise that conforms to the laws of real biological neural activity based on the noise signals generated by the perturbation neurons.

[0116] This invention proposes a data augmentation method and apparatus based on spiking neural networks. This application is used for implantable brain-computer interface (BCI) data and can extract information distribution characteristics even with limited neural information data. Based on the biological properties of spiking neural networks, they are suitable for learning the distribution characteristics of neural information, thereby generating neural signals that conform to these distribution characteristics, thus serving as brain information data augmentation. This invention considers the cluster activity characteristics of biological neurons as the basis for data augmentation. Spiking neural networks possess biological properties and are suitable for directly generating neural information, thereby enhancing BCI information, which is of great significance for the research and application of BCIs.

[0117] Based on the microbubble detection method described above, this embodiment provides a computer-readable storage medium storing one or more programs that can be executed by one or more processors to implement the steps in the data augmentation method based on spiking neural networks as described above.

[0118] A terminal device includes: a processor, a memory, and a communication bus; the memory stores a computer-readable program that can be executed by the processor; the communication bus enables communication between the processor and the memory; when the processor executes the computer-readable program, it implements the steps in the above-described data augmentation method based on a spiking neural network.

[0119] Based on the above-described data augmentation method based on spiking neural networks, this application provides a terminal device, such as... Figure 7As shown, it includes at least one processor 20; a display screen 21; and a memory 22, and may also include a communications interface 23 and a bus 24. The processor 20, display screen 21, memory 22, and communications interface 23 can communicate with each other via the bus 24. The display screen 21 is configured to display a preset user guide interface in the initial setup mode. The communications interface 23 can transmit information. The processor 20 can invoke logical instructions in the memory 22 to execute the methods described in the above embodiments.

[0120] Furthermore, the logical instructions in the aforementioned memory 22 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium.

[0121] The memory 22, as a computer-readable storage medium, can be configured to store software programs, computer-executable programs, such as program instructions or modules corresponding to the methods in the embodiments of this disclosure. The processor 20 executes functional applications and data processing by running the software programs, instructions, or modules stored in the memory 22, thereby implementing the methods in the above embodiments.

[0122] The memory 22 may include a program storage area and a data storage area. The program storage area may store the operating system and application programs required for at least one function; the data storage area may store data created based on the use of the terminal device. Furthermore, the memory 22 may include high-speed random access memory (RAM) and non-volatile memory. Examples include various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks, as well as transient storage media.

[0123] Furthermore, the specific process of loading and executing multiple instruction processors in the aforementioned storage medium and terminal device has been described in detail in the above method, and will not be repeated here.

[0124] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A data augmentation method based on a spiking neural network, characterized in that, Includes the following steps: Acquire raw data from the implantable brain-computer interface, the raw data including a neuronal population, and convert the pulse sequence of the neuronal population into pulse firing rate data; The pulse firing rate data is dimensionality reduced to obtain the neuron population activity patterns in the low-dimensional manifold space; A spiking neural network model was established, and the manifold spatial activity patterns of real neuron populations were used as the objective function for supervised learning. During the learning iteration process of the spiking neural network model, the dimensionality of the spiking neural network model-generated spiking sequence information is reduced to obtain a neuron population activity pattern that is similar to the real neuron population activity pattern. After the spiking neural network model has completed its iterations, perturbation neurons are added to the model. Based on the noise signals generated by these perturbation neurons, neural pulse information data that is noisy yet conforms to the laws of real biological neural activity is created and output. Wherein: The establishment of the spiking neural network model, using the manifold spatial activity patterns of a real neuron population as the objective function, and the supervised learning process includes: The spiking neural network model is based on the leak-integrated release model, which is as follows: in, This represents the magnitude of the membrane potential of a neuron, which is related to the firing of cell pulses. The time constant, expressed as a differential equation, is used to control the magnitude of the change in membrane potential over time. It is a constant parameter, representing the resting potential of the cell membrane, that is, the magnitude of the membrane potential of the cell in the resting state; As an input current, it affects the cell membrane potential when used as an external input. Cell membrane impedance; The synaptic time constant; This represents the pulse firing threshold, when the neuron's membrane potential... Exceed Pulses are emitted periodically, and the membrane potential is... Reset to .

2. The data augmentation method based on a spiking neural network according to claim 1, characterized in that, The raw data also includes the neuron population motion location information, which is then Gaussian smoothed to remove some noise data.

3. The data augmentation method based on a spiking neural network according to claim 1, characterized in that, The acquisition of raw data from the implantable brain-computer interface, including neuronal populations, and the conversion of the pulse sequence of the neuronal population into pulse firing rate data specifically involves: The pulse sequence of each neuron in the neuron population is processed by a sliding window to convert it into pulse firing rate data.

4. The data augmentation method based on a spiking neural network according to claim 1, characterized in that, During the learning iteration process of the spiking neural network model, the dimensionality reduction of the pulse sequence information generated by the spiking neural network model to obtain a neuron population activity pattern that approximates the real neuron population activity pattern is specifically as follows: The mean squared error is used as the loss function, where the mean squared error is the sum of the squares of the errors between corresponding points in the predicted data and the original data.

5. The data augmentation method based on a spiking neural network according to claim 1, characterized in that, After the spiking neural network model has completed its iterations, perturbation neurons are added to the model. Based on the noise signals generated by these perturbation neurons, neural pulse information data that is noisy yet conforms to the laws of real biological neural activity is created and output. Specifically, this is as follows: In the spiking neural network model, some neurons are based on Poisson neurons and are set as perturbation neurons. The model of the Poisson neuron is as follows: Using the LIF neuron model: The input current is converted from a random pulse signal, which follows a Poisson distribution. This represents the impulse firing rate of the neuron. The first neuron represents the neuron's... Each pulse in duration The probability of internal distribution; Implemented within the program, calculations are simplified, and pulses occur at time intervals. The probability of distribution within is , in It is a random variable with values ​​between 0 and 1; 。 6. A data augmentation device based on a spiking neural network, utilizing the data augmentation method based on a spiking neural network as described in claim 1, characterized in that, include: The pulse conversion module is used to acquire the raw data of the implantable brain-computer interface, the raw data including neuronal populations, and convert the pulse sequence of the neuronal populations into pulse firing rate data; The first dimensionality reduction module is used to reduce the dimensionality of the pulse firing rate data to obtain the neuron population activity patterns in the low-dimensional manifold space. The supervised learning module is used to build a spiking neural network model, using the manifold spatial activity patterns of real neuron populations as the objective function for supervised learning. The second dimensionality reduction module is used to reduce the dimensionality of the pulse sequence information generated by the spiking neural network model during the learning iteration process of the spiking neural network model, so as to obtain the neuron population activity law that is similar to the real neuron population activity law. The data output module is used to set perturbation neurons in the spiking neural network model after the iteration is completed, and to create and output neural pulse information data with noise that conforms to the laws of real biological neural activity based on the noise signals generated by the perturbation neurons.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores one or more programs, which can be executed by one or more processors to implement the steps in the data augmentation method based on a spiking neural network as described in any one of claims 1-5.

8. A terminal device, characterized in that, include: Processor, memory, and communication bus; The memory stores a computer-readable program that can be executed by the processor; The communication bus enables communication between the processor and the memory; When the processor executes the computer-readable program, it implements the steps in the data augmentation method based on a spiking neural network as described in any one of claims 1-5.

Images