Pulse neuron universal coding method based on electroencephalogram time-frequency representation

Abstract

The application discloses a general coding method of pulse neurons based on electroencephalogram time-frequency representation, and comprises the following steps: S1, pulse emission of neurons; after a neuron receives pulse stimulation, a postsynaptic potential gradually increases, and when the postsynaptic potential reaches a firing threshold, pulse emission is performed, and the potential is reset to zero. The general coding method of pulse neurons based on electroencephalogram time-frequency representation; the synchronous or asynchronous information transmission of tens of thousands of neurons in the brain constitutes a neural network of the brain. According to the representation of the discharge starting point of neurons and the brain waves formed after the brain waves, the causal representation is synchronously deduced in the paper, a neuron coding method with time and space characteristics and conforming to the biological process of neuron pulse emission is designed, the neuron coding method can express a single neuron coding sequence, express a time and space neuron pulse sequence matrix, and well represent the time and space information transmission state of brain activities.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of brain cognitive computing, and particularly relates to a general coding method of pulse neurons based on electroencephalogram time-frequency representation. BACKGROUND

[0002] In recent years, with the rise of brain-like intelligent technology, bionic modeling methods for brain neurons have been constantly emerging. Pulse neurons are a special kind of neuron, and the output thereof is a series of pulse sequences that encode information through time intervals.

[0003] The above-mentioned disadvantages are:

[0004] The cerebral cortex is composed of countless nerve cells, and the potential recorded from the surface of the cortex is the sum of the electric fields generated by countless neuron cells during activity. The rhythmic electroencephalogram generated by these devices is the process of simultaneous discharge and simultaneous stop of many homophasic neuron cells, and with such a synchronous process, the EEG device can record the brain waves. However, such brain waves are the total amount of electrical signals, and lack spatial resolution of brain electrical signals, so it is difficult to know whether the signals are generated in the cortex near the surface or in deeper areas. SUMMARY

[0005] The purpose of the present application is to provide a general coding method of pulse neurons based on electroencephalogram time-frequency representation to solve the problems raised in the background art.

[0006] To achieve the above-mentioned purpose, the present application provides the following technical solution: a general coding method of pulse neurons based on electroencephalogram time-frequency representation, the coding method comprising the following steps:

[0007] S1, pulse firing of neurons is performed, wherein the pulse firing of neurons is the transmission of biological information through synapses;

[0008] S2, the constitution of a brain neural network is performed, wherein the constitution of the brain neural network is the development of neural circuits and the formation of network connections between brain neurons according to the direction of pulse transmission of neurons, i.e. the neural network of the brain;

[0009] S3, electroencephalogram signal acquisition is performed, wherein the electroencephalogram signal acquisition is that different functional neuron groups corresponding to different brain regions can work synchronously or asynchronously, and the mixed signals of different functional neuron groups of each brain region are transmitted in the form of pulses;

[0010] S4, the transmission of spatiotemporal information is designed according to the discharge starting point of neurons and the characterization of brain waves after the formation of brain activity, a neuron encoding method with spatiotemporal characteristics and in line with the biological process of neuron pulse emission is generated, both single neuron encoding sequence and spatiotemporal neuron pulse sequence matrix are generated;

[0011] S5, the fitting brain wave form is derived by Fourier series to obtain the signal spectrum in the brain electrical signal.

[0012] Preferably, the neuron pulse emission includes neuron stimulation and pulse emission, the neuron stimulation gradually increases the post-synaptic potential by pulse stimulation of neurons, and reaches the emission threshold to achieve pulse emission.

[0013] Preferably, the collection of brain electrical signals includes signal collection, signal classification and signal removal, the signal collection collects the pulse signals emitted by neurons, the signal classification classifies different signals collected, and the signal removal removes noise signals obtained.

[0014] Preferably, the fitting brain wave form includes information collection and information display, the information collection collects and processes the brain electrical signals transmitted by the transmission of spatiotemporal information, and the information display displays the processed information on the display screen.

[0015] Preferably, the probability of activating neurons in a group of neuron pulse emission processes can be understood by the probability of whether the neuron discharges or not through sampling principle, and the sampled single column pulse sequence is generated by discharging neurons, and the unsampled neuron sequence is an inhibitory neuron, and the sampling frequency is set according to the result data, and the general expression function of the frequency domain emission probability of the discharge pulse sequence of a group of neurons is:

[0016] A′=PA (1);

[0017]

[0018] Wherein, A' is the sequence expression of multi-neuron discharge, A is the pulse emission model expression of neurons, P is the probability of emitting neurons in a group of neurons, n is the number of emitting neurons, and N is the number of a kind of neurons.

[0019] Preferably, the group of pulse sequences emitted by the neurons is equivalent to a time domain series and a frequency domain series expression of Fourier function, and the pulse emission function of a group of neurons and the waveform expressed by the frequency domain series of Fourier may appear fitting, that is, the waveform of brain electrical waves, and the expression is:

[0020] A' = PA = X(k) (3).

[0021] Preferably, the electrical signal of each brain area is converted into the expression of Fourier series:

[0022]

[0023]

[0024] Preferably, the expression formula of the pulse firing process of a group of neurons is:

[0025] Suppose that N target data are periodically extended, then the function expression of the Fourier transform of the discrete signal in a period is:

[0026]

[0027] Through conversion, it is obtained that:

[0028]

[0029] Wherein, the sampling frequency is fs / N is the frequency resolution. n = 0, 1, 2,..., N-1, P is a probability, and w is a pulse firing weight;

[0030] The general formula of the neuron model expressed by the Fourier transform is:

[0031]

[0032]

[0033] Technical effects and advantages of the present application:

[0034] The general coding method of the pulse neuron based on the time-frequency representation of the electroencephalogram is as follows: the synchronous or asynchronous information transmission of the tens of thousands of neurons in the brain constitutes the neural network of the brain, so that according to the representation of the brain activity after the discharge starting point of the neuron and the electroencephalogram, the causal representation is synchronously deduced, a neuron coding method with time-space characteristics and in line with the biological process of the neuron pulse firing is designed, which can express the single neuron coding sequence, express the time-space neuron pulse sequence matrix, and well represent the time-space information transmission state of the brain activity. BRIEF DESCRIPTION OF DRAWINGS

[0035] Figure 1 The flowchart of the coding method of the present application. DETAILED DESCRIPTION

[0036] Clearly, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the protection scope of the present application.

[0037] The present application provides a general coding method of pulse neurons based on electroencephalogram time-frequency representation as shown in the formula (1), which comprises the following steps: Figure 1

[0038] S1, pulse firing of neurons is performed, and the pulse firing of neurons is to conduct biological information through synapses;

[0039] S2, the constitution of brain neural network is performed, and the constitution of brain neural network is that the brain develops neural circuits and forms network connection between brain neurons according to the direction of pulse conduction of neurons, i.e. the neural network of the brain;

[0040] S3, the collection of electroencephalogram signals is performed, and the collection of electroencephalogram signals is that different neuron groups corresponding to different functions of different brain regions can work synchronously or asynchronously, and the mixed signals of different neuron groups of different functions of each brain region are fired in the form of pulses;

[0041] S4, the transmission of space-time information is performed, and the transmission of space-time information is to design a neuron coding method with space-time characteristics and in line with the biological process of neuron pulse firing according to the representation of brain activity after the starting point of neuron firing and the formation of electroencephalogram, which generates not only single neuron coding sequence but also space-time neuron pulse sequence matrix;

[0042] S5, the fitting of electroencephalogram waveform is performed, and the fitting of electroencephalogram waveform is to obtain the signal spectrum in the electroencephalogram signal by deriving the Fourier transform formula through Fourier series.

[0043] It should be noted that the synchronous or asynchronous information conduction of tens of thousands of neurons in the brain constitutes the neural network of the brain, so that the present document synchronously deduces the causal representation according to the representation of brain activity after the starting point of neuron firing and the formation of electroencephalogram, and designs a neuron coding method with space-time characteristics and in line with the biological process of neuron pulse firing, which can express not only single neuron coding sequence but also space-time neuron pulse sequence matrix, and can well represent the space-time information transmission state of brain activity.

[0044] ​Further, the pulse output of the brain neurons is weighted, the brain electrical waveform is fitted by Fourier DFT, the weight matrix of the clinical guidance is obtained, the group coding of the pulse emission of the neurons is formed by the Fourier series, the real brain electrical formation mechanism of the human body is met, the biological interpretability is obtained, and the biological basis of the pulse brain algorithm coding method is revealed.

[0045] Specifically, the pulse emission of the neurons includes neuron stimulation and pulse emission, the neuron stimulation makes the post-synaptic potential gradually increase by pulse stimulation of the neurons, and the pulse emission is achieved after the emission threshold is reached.

[0046] Specifically, the collection of the brain electrical signals includes signal collection, signal classification and signal removal, the signal collection is to collect the pulse signals emitted by the neurons, the signal classification is to classify different signals collected, and the signal removal is to remove noise signals obtained.

[0047] It should be noted that the brain electrical signals are easily disturbed, the original brain electrical signals collected by the sensor include not only the brain electrical signals, but also noise signals such as electromyogram and electrooculogram, and the unnecessary noise signals such as electromyogram and electrooculogram are removed in the collection process.

[0048] Specifically, the fitting of the brain electrical waveform includes information collection and information display, the information collection is to collect and process the brain electrical signals transmitted by the space-time information, and the information display is to display the processed information on the display screen.

[0049] Specifically, the probability of activating the neurons in the pulse emission process of a group of neurons is understood by the probability of whether the neurons discharge or not through the sampling principle, and the single column pulse sequence obtained by sampling is equivalent to the discharging neurons, the neuron sequence not sampled is equivalent to the inhibited neurons, and the sampling frequency is set according to the result data, and the general expression function of the frequency domain emission probability of the discharge pulse sequence of a group of neurons is:

[0050] A' = PA (1) ;

[0051]

[0052] Wherein, A' is a sequence expression of multi-neuron discharge, A is a pulse emission model expression of the neurons, P is the probability of the discharging neurons in a group of neurons, n is the number of the discharging neurons, and N is the number of a kind of neurons.

[0053] It should be noted that the cell membrane inside and outside through ion exchange transmission of biological information, when only one output, the voltage will automatically leak gradually back to the resting state. At the same time, since a neuron can receive external stimulation, can also receive synaptic input current stimulation produced by presynaptic neuron, set the neuron pulse emission model expression for A.

[0054] Specifically, the neuron firing a group of pulse sequence is equivalent to the Fourier function of a time series and frequency series expression, a group of neuron firing function and Fourier frequency series expression waveform may appear fitting, i.e. the brain wave waveform, its expression is:

[0055] A' = PA = X(k) (3).

[0056] Specifically, the electrical signal of each brain area is converted into Fourier series expression:

[0057]

[0058]

[0059] It should be noted that the Fourier function of time domain and frequency domain expression:

[0060]

[0061]

[0062] Specifically, the expression formula of a group of neuron pulse emission process:

[0063] Let N target data do periodic extension, then the Fourier transform of the function expression of the discrete signal in a cycle is:

[0064]

[0065] Conversion gives:

[0066]

[0067] Where the sampling frequency is fs / N is the frequency resolution, n = 0, 1, 2, …, N-1, P is the probability, w is the pulse emission weight;

[0068] The Fourier transform expression of neuron model general formula is:

[0069]

[0070]

[0071] The formula is arranged and the screening property of Dirac comb function is used to get:

[0072]

[0073] When T is replaced by N, i.e. only when t = nTs has a value, the above formula can be simplified as:

[0074]

[0075] Where:

[0076] X[k] = X(kω)NTs, x[n] = x(nTs) (14);

[0077] It can be seen that the Fourier transform expression at this time is:

[0078]

[0079] Thus the Fourier series expression is:

[0080]

[0081] Or:

[0082]

[0083] Since the last n in formula (13) is the corresponding point number rather than the frequency, it can be seen from formula (13) that:

[0084]

[0085] That is

[0086] Where n is the point number of FFT, and fs is the sampling frequency. Therefore, the frequency corresponding to the kth number after the discrete periodic Fourier transform is fs / N is the frequency resolution.

[0087] At this time, the Fourier series expression is:

[0088]

[0089] Or:

[0090]

[0091] Since the Fourier series is derived from the impulse signal as the original signal, i.e. the impulse response function is used to simulate the electrical discharge process of neurons, the function expression after sampling the original impulse signal is the same as the biological information contained in the firing neurons of this group of neurons, i.e.:

[0092]

[0093] A' = X[k] (22);

[0094] The original pulse sampling signal can be replaced by a representation signal expressing the biological electric signal of neuron firing by Fourier frequency domain series formula, and the signal function can be obtained by substituting the signal function:

[0095]

[0096] The general expression of Fourier series in frequency domain is:

[0097]

[0098] The general expression in time domain is:

[0099]

[0100] The general expression of Fourier series in frequency domain is:

[0101]

[0102] The general expression of Fourier series in time domain is:

[0103]

[0104] The outer two loops are combined and simplified to obtain:

[0105]

[0106] It should be noted that through the pulse firing expression of a single neuron, it can be inferred that the pulse firing expression of a group of neurons with the same function, and then by using the analytical characteristics of Fourier series in time domain, the pulse firing state of a group of neurons with the same function is deduced and converted by Fourier series, the general expression of a group of neurons with the same function is obtained, and then the Fourier transform formula is deduced by using Fourier series, the common signal spectrum in electroencephalogram is obtained, and thus the general group firing coding mode of brain neurons can be obtained.

[0107] Finally, it should be noted that the above only describes the preferred embodiments of the present application and is not used to limit the present application, although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent replacement for part of the technical features, and any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A universal encoding method for spiking neurons based on EEG time-frequency representation, characterized in that, The encoding method includes the following steps: S1, the neurons pulse-fire, wherein the pulse-fire of the neurons is the transmission of biological information through synapses; S2, constructing the brain's neural network, which is constructed by the brain developing neural circuits according to the direction of the pulse transmission of neurons, forming a network connection between brain neurons, i.e., the brain's neural network. S3, collect brain signals. The brain signal collection means that the neuronal groups with different functions corresponding to different brain regions can work synchronously or asynchronously, and the mixed signals of the neuronal groups with different functions in each brain region are emitted in the form of pulses. S4, transmitting spatiotemporal information, which is based on the neuronal firing initiation point and the brain activity representation after the formation of brain waves. Through a neuronal encoding method that has spatiotemporal characteristics and conforms to the biological process of neuronal pulse firing, both a single neuronal encoding sequence and a spatiotemporal neuronal pulse sequence matrix are generated. S5, fit the EEG waveform, wherein the fitted EEG waveform is obtained by deriving the Fourier transform formula through Fourier series to obtain the signal spectrum in the EEG signal; A formula representing the impulse firing process of a group of neurons: If we assume N target data points are periodically extended, then the Fourier transform function of the discrete signal within one period is expressed as: （1）； The result obtained through conversion is: （2）； Where the sampling frequency is ω= fs / N is the frequency resolution, n=0, 1, 2, ..., N-1, P is the probability, w is the pulse firing weight, T represents the sampling period, k represents the frequency index, s represents the bioelectric signal fired by the neuron; ω is a continuous frequency domain variable. This represents the spectral characteristics of the neuron's impulse firing signal at frequency ω, where t is the time variable and A is the expression for the neuron's impulse firing model. The general formula for the Fourier transform expression neuron model is: （3）； （4）。 2. The universal coding method for spiking neurons based on EEG time-frequency representation according to claim 1, characterized in that, The pulse firing of the neuron includes the stimulation of the neuron and the firing of pulses. The stimulation of the neuron causes the postsynaptic potential to gradually increase by stimulating the neuron with pulses. Once the firing threshold is reached, the pulse is fired.

3. The universal coding method for spiking neurons based on EEG time-frequency representation according to claim 1, characterized in that, The acquisition of the EEG signals includes signal collection, signal classification, and signal removal. Signal collection involves collecting the pulse signals emitted by neurons, signal classification involves classifying the different collected signals, and signal removal involves removing the noise signals obtained.

4. The universal encoding method for spiking neurons based on EEG time-frequency representation according to claim 1, characterized in that, The fitted EEG waveform includes information collection and information display; the information collection involves combining spatiotemporal information. The transmitted brainwaves are collected and processed, and the information display is the display of the processed information on a screen.

5. The universal encoding method for spiking neurons based on EEG time-frequency representation according to claim 1, characterized in that, The probability of activating neurons during the firing process of a group of neurons can be understood through a sampling principle. This is equivalent to the sampled single-pulse sequence representing the firing neurons, and the unsampled neuron sequences representing the inhibiting neurons. The sampling frequency is set based on the result data. The general expression function for the firing probability in the frequency domain of a group of neuron firing pulse sequences is: （5）； （6）； Where A' represents the sequence expression of multi-neuron firing, n represents the number of firing neurons, and N represents the number of neurons of one type.

6. The universal coding method for spiking neurons based on EEG time-frequency representation according to claim 2, characterized in that, The sequence of pulses fired by the neurons corresponds to a time-domain and frequency-domain expression of a Fourier function. The waveform of the pulse firing function of a group of neurons and the frequency-domain expression of the Fourier function may fit, i.e., the waveform of the electroencephalogram (EEG), which is expressed as: （7）； in This is the waveform of brainwaves.

7. The universal encoding method for spiking neurons based on EEG time-frequency representation according to claim 3, characterized in that, The electrical signals of each of the aforementioned brain regions are converted into Fourier series representations: （8）； （9）； in It is represented as the time-domain spectrum of the firing of a group of neurons; It is represented as the frequency domain spectrum of a group of neuron firing.

Images