WO2023178737A1

WO2023178737A1 - Spiking neural network-based data enhancement method and apparatus

Info

Publication number: WO2023178737A1
Application number: PCT/CN2022/085668
Authority: WO
Inventors: 郑胜杰; 李文艺; 李骁健
Original assignee: 中国科学院深圳先进技术研究院
Priority date: 2022-03-24
Filing date: 2022-04-08
Publication date: 2023-09-28
Also published as: CN116861967A

Abstract

A spiking neural network-based data enhancement method and apparatus. The method is applied to implantable brain-computer interface data, and can extract information distribution characteristics under a condition of limited neural information data; on the basis of biological attributes of a spiking neural network, the method is suitable for learning the distribution characteristics of neural information, so that a neural signal conforming to the information distribution characteristics is generated, and the neural signal serves as brain information data enhancement; the method considers cluster activity characteristics of biological neurons, and uses the cluster activity characteristics as a data enhancement basis, and the spiking neural network has biological properties, and is suitable for directly generating neural information, so that brain-computer interface information is enhanced, and the method has a great significance for research and application of brain-computer interfaces.

Description

A data enhancement method and device based on impulse neural network

Technical field

The present invention relates to the technical field of brain-computer interface, and specifically, to a data enhancement method, device, storage medium and equipment based on impulse neural network.

Background technique

The implantable brain-computer interface is a system that allows the brain to directly interact with the outside world, thereby helping patients restore, adjust, and enhance functions. The key technology of the implantable brain-computer interface is that it needs to be located in the cerebral cortex area of the brain's skull. Sensors are implanted to directly extract the information conveyed by neurons inside the brain; at the same time, the brain-computer interface system needs to be equipped with algorithms for interpreting neural information, so that it can interpret neural information, such as decoding the subject's movement intention, visual Information and so on.

After neural electrodes (sensors) are implanted in the brain, the neural information collection ability of the neural electrodes will gradually decrease over time. The main reasons are as follows: (1) The electrodes will cause some damage to the brain, causing glial scar tissue to grow around the electrodes, blocking The electrode contacts and neurons are separated; (2) The electrode contacts are gradually damaged in the electrolyte solution environment in the brain, reducing the electrical sensing performance of the electrode. In addition, for decoding algorithms, the ability to decode neural information will gradually decline over time; the main reasons are as follows: (1) The neural electrodes will shift due to brain shaking, resulting in changes in recorded information, and most decoding algorithms cannot Adapt to this change in neural information; (2) The brain's own neuroplasticity (the learning process of neurons) leads to the reconnection of neurons and changes in connection strength; therefore, the brain's neural information will change with the learning process.

Neural manifold refers to the neural activity pattern of a specific neuron group. The internal functions of the brain are derived from the collaborative information communication of neuron groups. A manifold is a space with local Euclidean space properties. It is used in mathematics. It is suitable for describing geometric shapes; and the activity of neuron clusters can be represented in a low-dimensional manifold space; therefore, under the constraints of the manifold space, it can assist the brain-computer interface to use a small amount of neural signals to complete the decoding of neural information.

The spiking neural network is often hailed as the third generation artificial neural network because it simulates the information transmission method of biological neurons; within organisms, communication between neurons is through the transmission of pulse signals, and information is contained in the transmitted pulses. in the sequence; while traditional neural networks use continuous values as the transmission information between artificial neurons, and the model architecture is of artificial design type and generally does not have biological functional characteristics; therefore, the biologically inspired spiking neural network uses spiking neurons and synapses. Therefore, the spiking neural network model based on biological properties is very suitable for studying brain information modeling and intelligent reasoning tasks.

Data augmentation is a technology that artificially expands the training data set by generating more equivalent data from limited data. Building a brain information decoding model through a brain-computer interface requires a large amount of brain information data, but the cost of obtaining a large amount of data is too high, and Data is an important factor in improving brain-computer interface decoding. Data enhancement is an effective means to overcome the lack of training data and is also widely used in various fields of artificial intelligence.

Therefore, existing technology still has shortcomings, and it is necessary to further develop data enhancement technology.

Contents of the invention

Embodiments of the present invention provide a data enhancement method and device based on impulse neural networks to solve the problem of data enhancement when training data is insufficient.

According to an embodiment of the present invention, a data enhancement method based on spiking neural network is provided, including the following steps:

Obtain the raw data of the implanted brain-computer interface, the raw data includes the neuron population, and convert the spike sequence of the neuron population into spike firing rate data;

Dimensionality reduction is performed on the pulse firing rate data to obtain the neuron population activity rules in the low-dimensional manifold space;

Establish a spiking neural network model and use the manifold spatial activity patterns of real neuron groups as the objective function to perform supervised learning;

In the learning iteration process of the spiking neural network model, the dimensionality of the spike sequence information generated by the spiking neural network model is reduced, and a neuron group activity pattern that is similar to the real neuron group activity pattern is obtained;

After the iteration of the spiking neural network model is completed, perturbation neurons are set up in the spiking neural network model. Based on the noise signals generated by the perturbing neurons, nerve impulse information data with noise and in line with the laws of real biological nerve activity is created and output.

Furthermore, the original data also includes the movement position information of the neuron group, and Gaussian smoothing is performed on the movement position information to remove part of the noise data.

Furthermore, a spiking neural network model is established, using the manifold spatial activity patterns of real neuron groups as the objective function, and performing supervised learning including:

The spiking neural network model uses the leakage integration firing model as the basis of the model. The leakage integration firing model is as follows:

u _i ＝u _rest ,u _i >u _threshold

Among them, u _i represents the membrane potential of the neuron cell, which is related to the cell pulse emission; τ _m represents the time constant of the differential equation, which is used to control the change of the membrane potential with time; u _rest is a constant parameter , expressed as the resting potential of the cell membrane, that is, the size of the membrane potential of the cell in the resting state; I _i (t) is the input current, which will affect the membrane potential of the cell as an external input; R is the cell membrane impedance; τ _s is Synaptic time constant; u _threshold represents the pulse firing threshold. When the neuron's membrane potential u _i exceeds u _threshold , a pulse is fired, and the membrane potential u _i is reset to u _rest .

Furthermore, a spiking neural network model is established, and the manifold spatial activity pattern of the real neuron group is used as the objective function. Supervised learning includes:

Supervised learning is performed by replacing the gradient learning algorithm to update the neural network model weights; the replacing gradient learning algorithm is described as follows:

S _i [n]∝Θ(u _i [n]-u _threshold )

Among them, S _i [n] is represented as a pulse sequence, and Θ(x) is represented as a Heavisitter step function. Since the pulse signal has non-differentiable properties, during back propagation, the Θ(x) function is represented by σ(x) By substitution, we have S _i [n]∝σ(u _i [n]-u _threshold ), where

Further, the original data of the implanted brain-computer interface is obtained. The original data includes the neuron group, and the spike sequence of the neuron group is converted into the spike firing rate data as follows:

The spike train of each neuron in the neuron population is processed by sliding window and converted into spike firing rate data.

Furthermore, in the learning iteration process of the spiking neural network model, the dimensionality of the spike sequence information generated by the spiking neural network model is reduced, and the neuron group activity rules that are similar to the real neuron group activity rules are obtained:

Use the mean square error as the loss function. The mean square error is the mean of the square sum of the errors corresponding to the predicted data and the original data. The mean square error formula is:

Among them, _yi is the dimensionality reduction data output of the impulse neural network,

is the objective function, n is the number of samples.

Further, after the iteration of the spiking neural network model is completed, perturbation neurons are set up in the spiking neural network model, and based on the noise signals generated by the perturbing neurons, nerve impulse information data with noise and in line with the laws of real biological nerve activity is created and output. Specifically:

Some neurons in the spiking neural network model are based on Poisson neurons and set as perturbation neurons. The model of Poisson neurons is:

Using the LIF neuron model:

I _i is the input current, which is converted from a random pulse signal. The random pulse signal conforms to the Poisson distribution law;

r represents the spike firing rate of the neuron, and P _T [n] represents the probability that the nth spike of the neuron is fired within the duration T;

Implemented in the program, the calculation can be simplified. The probability of pulse emission within the time interval Δt can be rΔt, where x _rand is a random variable with a value between 0 and 1;

A data enhancement device based on spiking neural network, including:

The pulse conversion module is used to obtain the original data of the implanted brain-computer interface. The original data includes the neuron population and converts the pulse sequence of the neuron population into pulse firing rate data;

The first dimensionality reduction module is used to reduce the dimensionality of the pulse firing rate data and obtain the neuron group activity rules in the low-dimensional manifold space;

The supervised learning module is used to establish a spiking neural network model and perform supervised learning using the manifold spatial activity patterns of real neuron groups as the objective function;

The second dimensionality reduction module is used to reduce the dimensionality of the spike sequence information generated by the spike neural network model during the learning iteration process of the spike neural network model, and obtain neuron group activity patterns that are similar to the real neuron group activity patterns;

The data output module is used to set perturbation neurons in the impulse neural network model after the iteration of the impulse neural network model. Based on the noise signals generated by the perturbation neurons, create nerve impulse information data with noise and in line with the laws of real biological nerve activity. Output it.

A computer-readable medium. The computer-readable storage medium stores one or more programs. The one or more programs can be executed by one or more processors to implement data enhancement based on spiking neural networks as any of the above. steps in the method.

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 realizes the connection and communication between the processor and the memory;

When the processor executes a computer-readable program, the steps in any one of the above spiking neural network-based data enhancement methods are implemented.

The embodiment of the present invention proposes a data enhancement method and device based on spiking neural network. The method of this application is used for implanted brain-computer interface data, and can extract information distribution characteristics under the condition of limited neural information data; based on spiking neural network The biological attributes of the network are suitable for learning the distribution characteristics of neural information, thereby generating neural signals that conform to the information distribution characteristics, and use this as brain information data enhancement; the present invention takes into account the cluster activity characteristics of biological neurons as the basis for data enhancement. , Impulsive 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 of the drawings

Figure 1 is a flow chart of the data enhancement method based on spiking neural network according to the present invention;

Figure 2 shows the neural signals collected by the implantable brain-computer interface of the present invention and the corresponding motion control;

Figure 3 is a representation in low-dimensional manifold space of the neural signals collected by the implantable brain-computer interface of the present invention corresponding to the test;

Figure 4 is a schematic diagram of the impulse neural network architecture designed by the present invention;

Figure 5 is a schematic diagram of brain-computer interface data generated by the spiking neural network designed in the present invention;

Figure 6 is a schematic diagram of the data enhancement device based on the impulse neural network of the present invention;

Figure 7 shows the terminal equipment of the present invention.

Detailed ways

In order to make the purpose, technical solutions and advantages of the present application more clear, the present application will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the present application and are not used to limit the present application.

It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It is to be understood that the data so used are interchangeable under appropriate circumstances so that the embodiments of the invention described herein are capable of being practiced in sequences other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, e.g., a process, method, system, product, or apparatus that encompasses a series of steps or units and need not be limited to those explicitly listed. Those steps or elements may instead include other steps or elements not expressly listed or inherent to the process, method, product or apparatus.

Referring to Figure 1, according to an embodiment of the present invention, a data enhancement method based on a spiking neural network is provided, including the following steps:

S100 obtains the raw data from the implanted brain-computer interface. The raw data includes the neuron population, and converts the pulse sequence of the neuron population into pulse firing rate data.

The original data includes information from the neuron group extracted from the motor cortex, movement position and movement speed information corresponding to the subject's movement control, and markers corresponding to the type of movement control test; each neuron group in the brain-computer interface-related movement The neuron pulse sequence is processed by a sliding window and converted into neuron pulse firing rate data; the movement position information is Gaussian smoothed to remove part of the noise data; the moving average filter can convert the neuron population signal of the original signal into Downsampling improves the signal-to-noise ratio; the Gaussian filter is a filter used to remove noise. It is also used for digital signal processing and removing the noise of the motion position of the original information; as shown in Figure 2, it is an implantable brain-computer interface Collected neural signals and corresponding motor control.

The following defines a moving average filter for the vector x for the difference equation:

The moving average filter moves windows along the data with a length of window size (window Size) and calculates the average of the data contained in each window.

The following is a one-dimensional Gaussian filter smoothing the motion positions in the original information:

x is the signal input and σ is the standard deviation.

S200 reduces the dimensionality of the pulse firing rate data and obtains the neuron group activity rules in low-dimensional manifold space.

For the same task experiment, the brain-computer interface collects neural impulse data and performs dimensionality reduction to obtain the neuron activity rules of low-dimensional manifold space. Specifically:

For the data of the same experiment, first perform Principal Component Analysis (PCA), and also use the cumulative ratio (The cumulative explained variance ratio) to determine the amount of variance explained by the principal components, thereby obtaining statistically significant principal components. number, as the dimension representing the activity of the neuron group, thus obtaining the activity rules of the neuron group in the low-dimensional manifold space; as shown in Figure 3, the neural signals of the corresponding experiments collected for the implanted brain-computer interface are in the low-dimensional manifold space. dimensional manifold space representation.

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

First, there are m rows and n columns of data;

Step 1: Organize the original data into a matrix X with n rows and m columns;

Step 2: Zero-mean each row of X;

Step 3: Find the covariance matrix

Step 4: Find the eigenvalues and corresponding eigenvectors of the covariance matrix;

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

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

S300 establishes a spiking neural network model and uses the manifold spatial activity patterns of real neuron groups as the objective function to perform supervised learning.

The present 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 a supervised learning algorithm and uses a method called substitution. Gradient Learning (Surrogate Gradient Learning) algorithm, thereby updating the weights of the neural network model; as shown in Figure 4, it is a schematic diagram of the impulse neural network architecture designed by the present invention.

The leaked integration distribution model is as follows:

u _i ＝u _rest ,u _i >u _threshold

Instead the gradient learning algorithm is described as follows:

S _i [n]∝Θ(u _i [n]-u _threshold )

S400: During the learning iteration process of the spiking neural network model, reduce the dimensionality of the spike sequence information generated by the spiking neural network model, and obtain a neuron group activity pattern that is similar to the real neuron group activity pattern.

In the learning iteration process of the neural network model, the pulse sequence generated by the output layer of the spiking neural network needs to be dimensionally reduced, and the dimensionality reduction data is used as the output. The dimensionality reduction activity pattern of the real neuron group is used as the objective function, and the mean square error (MSE, Mean-Square Loss) as the loss function; the training results need to meet the requirements, and the neuron population activity rules of the output layer need to be similar to the real neuron population activity rules; the mean square error loss is also called quadratic loss, L2 loss, and is often used for regression prediction In the task, 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 the two is.

The mean square error is the mean of the square sum of the errors corresponding to the predicted data and the original data. The formula is as follows:

is the objective function, n is the number of samples.

S500: After the iteration of the spiking neural network model is completed, perturbation neurons are set in the spiking neural network model. Based on the noise signals generated by the perturbing neurons, nerve impulse information data with noise and in line with the laws of real biological nerve activity is created and output.

After the iteration of the spiking neural network is completed, some neurons in the output layer of the spiking neural network will be started and set as perturbation neurons based on Poisson Neuron. The nerve impulse information of biological neurons is based on Poisson distribution. , in this application, the LIF neuron generates a Poisson distributed signal; a random pulse signal is input into the perturbation neuron, so that the perturbation neuron generates a noise signal; thus, the output layer can generate noise and conform to the laws of real biological neural activity. Nerve impulse information data; as shown in Figure 5, it is a schematic diagram of the brain-computer interface data generated by the impulse neural network designed by the present invention.

The model of Poisson neuron is:

Using the LIF neuron model:

The data generated according to the present invention can be further used in the fields of brain-computer interface information decoding and brain-like intelligence research; the method of this application is specially used for implanted brain-computer interface data, and can extract information under the condition of limited neural information data. distribution characteristics. Based on the biological properties of the spiking neural network, it is suitable for learning the distribution characteristics of neural information, thereby generating neural signals that conform to the information distribution characteristics, as brain information data enhancement.

The present invention takes into account the cluster activity characteristics of biological neurons as the basis for data enhancement. At the same time, 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.

This invention takes into account the collaborative nature of brain-computer interfaces and brain-like intelligent algorithms. The enhanced data is more in line with the internal laws of neural information, and the generated data is more biologically interpretable.

Referring to Figure 6, according to another embodiment of the present invention, a data enhancement device based on a spiking neural network is provided, including:

The pulse conversion module 100 is used to obtain the original data of the implanted brain-computer interface. The original data includes the neuron population, and converts the pulse sequence of the neuron population into pulse firing rate data;

The first dimensionality reduction module 200 is used to reduce the dimensionality of the pulse firing rate data and obtain the neuron group activity rules in the low-dimensional manifold space;

The supervised learning module 300 is used to establish a spiking neural network model and perform supervised learning using the manifold spatial activity patterns of real neuron groups as the objective function;

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, and obtain a neuron group activity pattern that is similar to the real neuron group activity pattern. .

The data output module 500 is used to set perturbation neurons in the impulse neural network model after the iteration of the impulse neural network model is completed, and based on the noise signals generated by the perturbation neurons, create nerve impulse information data that is noisy and conforms to the laws of real biological nerve activity. and output it.

The embodiment of the present invention proposes a data enhancement method and device based on a spiking neural network. This application is used for implanted brain-computer interface data, which can extract information distribution characteristics under the condition of limited neural information data; based on the spiking neural network The biological attributes of the brain are suitable for learning the distribution characteristics of neural information, thereby generating neural signals that conform to the information distribution characteristics, and use this as brain information data enhancement; the present invention takes into account the cluster activity characteristics of biological neurons as the basis for data enhancement. Impulsive 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.

Based on the above microbubble detection method, this embodiment provides a computer-readable storage medium. The computer-readable storage medium stores one or more programs. The one or more programs can be executed by one or more processors to implement The steps in the data enhancement method based on spiking neural network in the above embodiment.

A terminal device, including: 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 realizes connection and communication between the processor and the memory; the processor executes the computer-readable program When implementing the above steps in the data enhancement method based on spiking neural network.

Based on the above data enhancement method based on impulse neural network, this application provides a terminal device, as shown in Figure 7, which includes at least one processor (processor) 20; a display screen 21; and a memory (memory) 22. It can also Including communications interface (Communications Interface) 23 and bus 24. Among them, the processor 20, the display screen 21, the memory 22 and the communication interface 23 can complete communication with each other through the bus 24. The display screen 21 is configured to display a user guidance interface preset in the initial setting mode. Communication interface 23 can transmit information. The processor 20 can call logical instructions in the memory 22 to execute the methods in the above embodiments.

In addition, the above-mentioned logical instructions in the memory 22 can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product.

As a computer-readable storage medium, the memory 22 can be configured to store software programs, computer-executable programs, such as program instructions or modules corresponding to the methods in the embodiments of the present disclosure. The processor 20 executes software programs, instructions or modules stored in the memory 22 to execute functional applications and data processing, that is, to implement the methods in the above embodiments.

The memory 22 may include a stored program area and a stored data area, where the stored program area may store an operating system and an application program required for at least one function; the stored data area may store data created according to the use of the terminal device, etc. In addition, the memory 22 may include high-speed random access memory, and may also include non-volatile memory. For example, there are many media that can store program code, such as U disk, mobile hard disk, read-only memory (ROM), random access memory (Random Access Memory, RAM), magnetic disk or optical disk, or they can also be temporary state storage media.

In addition, the specific process of loading and executing the multiple instruction processors in the above storage medium and terminal device has been described in detail in the above method, and will not be described one by one here.

The specific embodiments of the invention have been described in detail above, but they are only used as examples, and the invention is not limited to the specific embodiments described above. For those skilled in the art, any equivalent modifications or substitutions to the invention are also within the scope of the invention. Therefore, equivalent transformations, modifications and improvements can be made without departing from the spirit and principles of the invention. etc., should all be included in the scope of the present invention.

Claims

A data enhancement method based on spiking neural network, characterized by including the following steps:

Obtaining raw data from the implanted brain-computer interface, the raw data including a population of neurons, and converting the pulse sequence of the population of neurons into pulse firing rate data;

Perform dimensionality reduction on the pulse firing rate data to obtain the neuron group activity rules in the low-dimensional manifold space;

Establish a spiking neural network model and use the manifold spatial activity patterns of real neuron groups as the objective function to perform supervised learning;

During the learning iteration process of the spiking neural network model, dimensionality reduction is performed on the pulse sequence information generated by the spiking neural network model, and a neuron group activity pattern that is similar to the real neuron group activity pattern is obtained;

After the iteration of the impulse neural network model is completed, perturbation neurons are set in the impulse neural network model, and based on the noise signals generated by the perturbation neurons, nerve impulse information data with noise and in line with the laws of real biological nerve activity is created and Output it.
The data enhancement method based on impulse neural network according to claim 1, characterized in that the original data also includes the movement position information of the neuron group, and Gaussian smoothing is performed on the movement position information to remove part of the noise data. .
The data enhancement method based on spiking neural network according to claim 2, characterized in that establishing the spiking neural network model, using the manifold space activity pattern of the real neuron group as the objective function, and performing supervised learning includes:

The spiking neural network model is based on the leakage integrated firing model. The leakage integrated firing model is as follows:

u i ＝u rest ,u i >u threshold

Among them, u i represents the membrane potential of the neuron cell, which is related to the cell pulse emission; τ m represents the time constant of the differential equation, which is used to control the change of the membrane potential with time; u rest is a constant parameter , expressed as the resting potential of the cell membrane, that is, the size of the membrane potential of the cell in the resting state; I i (t) is the input current, which will affect the membrane potential of the cell as an external input; R is the cell membrane impedance; τ s is Synaptic time constant; u threshold represents the pulse firing threshold. When the neuron's membrane potential u i exceeds u threshold , a pulse is fired, and the membrane potential u i is reset to u rest .
The data enhancement method based on spiking neural network according to claim 2, characterized in that, establishing the spiking neural network model, using the manifold space activity pattern of the real neuron group as the objective function, and performing supervised learning includes:

Supervised learning is performed by replacing the gradient learning algorithm so that the weights of the neural network model are updated; the replacing gradient learning algorithm is described as follows:

S i [n]∝Θ(u i [n]-u threshold )

Among them, S i [n] is represented as a pulse sequence, and Θ(x) is represented as a Heavisitter step function. Since the pulse signal has non-differentiable properties, during back propagation, the Θ(x) function is represented by σ(x) By substitution, we have S i [n]∝σ(u i [n]-u threshold ), where
The data enhancement method based on spiking neural network according to claim 1, characterized in that: the original data of the implanted brain-computer interface is obtained, the original data includes a neuron group, and the pulse of the neuron group is The sequence converted into pulse firing rate data is specifically:

The spike train of each neuron in the neuron population is subjected to sliding window processing and converted into spike firing rate data.
The data enhancement method based on spiking neural network according to claim 1, characterized in that, in the learning iteration process of the spiking neural network model, dimensionality reduction is performed on the pulse sequence information generated by the spiking neural network model. , it can be concluded that the neuron population activity law that is similar to the real neuron population activity law is specifically:

The mean square error is used as the loss function. The mean square error is the mean of the square sum of the errors corresponding to the predicted data and the original data. The mean square error formula is:

Among them, yi is the dimensionality reduction data output of the impulse neural network,
is the objective function, n is the number of samples.
The data enhancement method based on spiking neural network according to claim 1, characterized in that, after the iteration of the spiking neural network model is completed, a perturbation neuron is set in the spiking neural network model, and based on the perturbation neuron, The noise signal is used to create nerve impulse information data with noise and in line with the laws of real biological nerve activity, and its output is specifically as follows:

Some neurons in the spiking neural network model are based on Poisson neurons and set as the perturbation neurons. The model of the Poisson neurons is:

Using the LIF neuron model:

I i is the input current, which is converted from a random pulse signal. The random pulse signal conforms to the Poisson distribution law;

r represents the spike firing rate of the neuron, and P T [n] represents the probability that the nth spike of the neuron is fired within the duration T;

Implemented in the program, the calculation can be simplified. The probability of pulse emission within the time interval Δt can be rΔt, where x rand is a random variable with a value between 0 and 1;
A data enhancement device based on impulse neural network, which is characterized by including:

A pulse conversion module, used to obtain raw data of the implanted brain-computer interface, where the raw data includes a neuron population, and convert the pulse sequence of the neuron population into pulse firing rate data;

The first dimensionality reduction module is used to reduce the dimensionality of the pulse firing rate data and obtain the neuron group activity rules in the low-dimensional manifold space;

The supervised learning module is used to establish a spiking neural network model and perform supervised learning using the manifold spatial activity patterns of real neuron groups as the objective function;

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, and obtain a model that is similar to the activity pattern of the real neuron group. The activity patterns of neuron groups;

A data output module is used to set perturbation neurons in the spiking neural network model after the iteration of the spiking neural network model is completed, and based on the noise signal generated by the perturbing neurons, create a noisy and consistent biological neural activity pattern. nerve impulse information data and output it.
A computer-readable medium, characterized in that the computer-readable storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement claims 1- Steps in the data enhancement method based on spiking neural network according to any one of 7.
A terminal device, characterized in that it 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 implements connection communication between the processor and the memory;

When the processor executes the computer readable program, the steps in the data enhancement method based on spiking neural network according to any one of claims 1-7 are implemented.