Telecommunication signal detection method and system based on neural medium

Abstract

The present application relates to the technical field of electric signal detection, and discloses a method and system for detecting electric signals based on neural medium, which comprises the following steps: Step 1, preparing a biomimetic neural medium microelectrode array; Step 2, collecting multi-channel electric signal data through the biomimetic neural medium microelectrode array; Step 3, performing dynamic gain adjustment on the multi-channel electric signal data based on a synaptic plasticity model to generate electric signal data with optimized gain; and Step 4, constructing an adaptive noise suppression function according to the neurotransmitter concentration characteristics. The present application adopts a dynamic gain adjustment scheme simulating neurotransmitters and an adaptive noise suppression scheme based on neurotransmitter concentration, thereby achieving the technical effects of significantly improving signal fidelity and greatly improving anti-interference capability. Compared with the scheme in the prior art which relies on fixed gain amplification and single frequency domain filtering, the present application solves the defects of being unable to dynamically adapt to complex noise environments, serious signal distortion and insufficient noise suppression capability.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electrical signal detection, in particular to an electrical signal detection method and system based on neural medium. BACKGROUND

[0002] Electrical signal detection technology is one of the core technologies in biomedical engineering, neuroscience and brain-computer interface fields. The core challenge is how to accurately capture weak signals in a high-noise environment and meet the real-time processing needs of multi-channel and high-density data.

[0003] The current electrical signal detection technology faces core bottlenecks in practical application. The fundamental contradiction lies in the conflict between static parameterization design and dynamic complex environment, and the imbalance between serial architecture and high-throughput demand:

[0004] Traditional methods rely on fixed gain amplification circuits and single frequency domain filtering. Static parameters cannot adapt to the dynamic fluctuations of signal strength and the time-varying characteristics of noise spectrum. Specifically, fixed gain amplification is prone to cause weak signals to be masked by noise or strong signals to be saturated and distorted when the signal strength suddenly changes; pre-set frequency domain filtering can suppress fixed frequency band noise, but cannot distinguish non-stationary noise overlapping with signal spectrum, resulting in effective signals being misfiltered or noise remaining.

[0005] Traditional systems use serial data acquisition and processing architecture, which has low hardware resource utilization and cannot meet the needs of multi-channel synchronous high-density monitoring. Specifically, under the serial architecture, the number of channels increases linearly with the hardware cost, and the multi-channel synchronization accuracy decreases significantly due to clock drift, resulting in the loss of data correlation; the fixed parameter processing procedure lacks dynamic optimization capability and cannot be adjusted in real time according to signal characteristics, resulting in delayed processing of key signals.

[0006] Therefore, the present application proposes an electrical signal detection method and system based on neural medium to solve the above problems. SUMMARY

[0007] In view of the deficiencies of the prior art, the present application provides an electrical signal detection method and system based on neural medium to solve the problems raised in the background art.

[0008] To achieve the above purpose, the present application realizes the following technical scheme: an electrical signal detection method based on neural medium, comprising:

[0009] Step 1, preparing a biomimetic neural medium microelectrode array;

[0010] Step 2, collecting multi-channel electrical signal data through the biomimetic neural medium microelectrode array;

[0011] Step 3, dynamic gain adjustment of multi-channel electrical signal data based on synaptic plasticity model, to generate gain-optimized electrical signal data;

[0012] Step 4, constructing an adaptive noise suppression function according to neurotransmitter concentration characteristics, and performing frequency domain filtering processing on the gain-optimized electrical signal data to generate denoised electrical signal data;

[0013] Step 5, based on parallel computing architecture, multi-channel synchronous segmentation is performed on the denoised electrical signal data, and the segmented data blocks are distributed to a heterogeneous computing platform;

[0014] Step 6, on the heterogeneous computing platform, a dual-branch long short-term memory network is used to separate the signal and noise of the segmented data blocks, to generate reconstructed signal data;

[0015] Step 7, based on the reconstructed signal data, the signal gain, sampling rate and filtering parameters are dynamically adjusted through reinforcement learning strategy to form a closed-loop feedback mechanism.

[0016] Preferably, in step 1, the preparation of the biomimetic neural medium microelectrode array further comprises:

[0017] Substep 1.1, selecting a flexible polyimide substrate, and generating a graphene-carbon nanotube composite conductive layer on the surface of the substrate by chemical vapor deposition method, wherein the surface resistance of the composite conductive layer satisfies the formula:

[0018]

[0019] wherein R sheet is the surface resistance of the composite conductive layer, σ is the conductivity, and h is the thickness of the conductive layer.

[0020] Substep 1.2, etching a microelectrode array on the composite conductive layer by photolithography process, wherein the number of channels of the microelectrode array satisfies the formula:

[0021]

[0022] wherein N is the number of channels, A is the effective area of the substrate, and d is the electrode spacing.

[0023] Substep 1.3, modifying a dopamine oxidoreductase layer on the surface of the microelectrode by electrochemical deposition method, wherein the current density of the electrochemical deposition satisfies the Butler-Volmer kinetics equation:

[0024]

[0025] wherein j is the deposition current density, j0 is the exchange current density, α is the charge transfer coefficient, F is the Faraday constant, η is the voltage difference, R is the gas constant, and T is the temperature.

[0026] Preferably, in step 2, the multi-channel electrical signal data collected by the biomimetic neural medium microelectrode array further comprises:

[0027] Sub-step 2.1, using the dopamine oxidoreductase layer on the surface of the biomimetic neural medium microelectrode array, the neural electrical signal is converted into an electrochemical current signal, and the current signal intensity satisfies Faraday's law:

[0028] I = nF·A·k·[DA],

[0029] Where I is the electrochemical current, n is the number of electron transfer, F is the Faraday constant, A is the effective area of the electrode, k is the enzyme activity constant, and [DA] is the local concentration of dopamine;

[0030] Sub-step 2.2, the electrochemical current signal is converted into a voltage signal by a transimpedance amplifier, and the output voltage of the transimpedance amplifier satisfies the formula:

[0031] V out = -R f ·I,

[0032] Where V out is the output voltage, R f is the feedback resistance, and I is the electrochemical current;

[0033] Sub-step 2.3, the multi-channel voltage signals are synchronously collected by using a timestamp alignment protocol, and the synchronization error satisfies the formula:

[0034]

[0035] Where Δt is the clock correction, t master and t slave are the master device clock values, D link is the physical distance between devices, and c is the signal propagation speed.

[0036] Preferably, in step 3, the multi-channel electrical signal data is dynamically gain-adjusted based on a synaptic plasticity model to generate gain-optimized electrical signal data, further comprising:

[0037] Sub-step 3.1, detecting the action potential peak in the multi-channel electrical signal data and recording the triggering time t i , and the detection condition of the action potential peak satisfies the formula:

[0038] V out ≥ V th ,

[0039] Where V out is the output voltage, and V th is the threshold voltage.

[0040] Sub-step 3.2, according to the action potential trigger time sequence, based on the synaptic long-term potentiation model, the dynamic gain coefficient is calculated, the gain coefficient satisfies the formula:

[0041]

[0042] Wherein, G(t) is the time-varying gain coefficient, G0 is the baseline gain, a is the synaptic plasticity intensity factor, t i is the action potential trigger time, t is the current time point, τ is the synaptic effect time constant;

[0043] Sub-step 3.3, the dynamic gain coefficient is multiplied by the multi-channel electrical signal data point by point, and the gain-optimized electrical signal data is generated:

[0044] V optimized (t) = G(t) · V(t),

[0045] Wherein, V optimized (t) is the gain-optimized voltage signal, G(t) is the time-varying gain coefficient, and V(t) is the voltage signal.

[0046] Preferably, in step 4, according to the neurotransmitter concentration characteristics, an adaptive noise suppression function is constructed, and the gain-optimized electrical signal data is processed in frequency domain to generate denoising electrical signal data, further comprising:

[0047] Sub-step 4.1, the extracellular γ-aminobutyric acid concentration [GABA] is detected in real time by microdialysis probe, and the γ-aminobutyric acid concentration is positively correlated with the noise suppression intensity;

[0048] Sub-step 4.2, the gain-optimized voltage signal generated in step 3 is subjected to fast Fourier transform, and the noise main frequency band f0 is extracted, which satisfies the formula:

[0049]

[0050] Wherein, is the Fourier transform, f min and f max are the noise frequency band range, f0 is the center frequency of the noise main frequency band, and V optimized (t) is the gain-optimized voltage signal;

[0051] Sub-step 4.3, based on the γ-aminobutyric acid concentration [GABA] and the noise main frequency band f0, a frequency domain weighting function is constructed, and the gain-optimized electrical signal data is processed by filtering:

[0052]

[0053] wherein W(f) is the frequency domain weighting function value, β=k·[GABA] is the inhibition strength factor, σ is the inhibition bandwidth, f is the frequency component, f0 is the center frequency of the noise dominant band;

[0054] Sub-step 4.4, multiplying the frequency domain weighting function W(f) with the gain-optimized frequency spectrum data, generating the denoised electrical signal data through inverse Fourier transform:

[0055]

[0056] wherein W(f) is the frequency domain weighting function value, V optimized (t) is the gain-optimized voltage signal, is the Fourier transform.

[0057] Preferably, in the step 5, the denoised electrical signal data is divided into multiple channels synchronously based on the parallel computing architecture, and the divided data blocks are distributed to the heterogeneous computing platform, further comprising:

[0058] Sub-step 5.1, determining the data block duration T block according to the number of channels N and the real-time requirement, wherein the data block duration satisfies the formula:

[0059]

[0060] wherein L buffer is the buffer capacity, f s is the number of channels for acquisition, and f s is the single-channel sampling rate.

[0061] Sub-step 5.2, dividing the denoised electrical signal data V denoised (t) generated in step 4 into data blocks according to the time window T block , generating a data block set wherein the number of samples contained in each data block is:

[0062] M=T block ·f s ,

[0063] wherein M is the number of samples in a single data block, T block is the data block duration, and f s is the single-channel sampling rate

[0064] Sub-step 5.3, distributing the data blocks X k to the multiple computing units of the heterogeneous computing platform based on the load balancing strategy, wherein the distribution weight satisfies the formula:

[0065]

[0066] wherein wj Cj is the assigned weight of the computing unit j. j Rj is the real-time residual computing power of the computing unit j.

[0067] Preferably, in step 6, the signal and noise separation of the segmented data block is performed on the heterogeneous computing platform by using a double-branch long short-term memory network to generate reconstructed signal data, further comprising:

[0068] Sub-step 6.1, for the data block X assigned in step 5 k perform time window clipping to generate an input sequence x t , the length of the time window is T w , the step size is S, and the formula is satisfied:

[0069]

[0070] where x t is the voltage value of the t-th sampling point in the data block X k ;

[0071] Sub-step 6.2, extract signal features and noise features respectively through the double-branch long short-term memory network, and the feature updating process satisfies the LSTM gating mechanism:

[0072] f t = σ(W f · [h t-1 , x t ] + b f ),

[0073] i t = σ(W i · [h t-1 , x t ] + b i ),

[0074] o t = σ(W o · [h t-1 , x t ] + b o ),

[0075] c t = f t ⊙ c t-1 + i t ⊙ tanh(W c · [h t-1 , x t ] + b c ),

[0076] h t = o ttanh(c t ),

[0077] where f t , i t , o t are forget gate, input gate, output gate, W f , W i , W o , W c and b f , b i , b o , b c are trainable parameters, σ is Sigmoid function, ⊙ is element-wise multiplication, h t is current hidden state, h t-1 is previous hidden state, c t is current cell state, c t-1 is previous cell state.

[0078] Sub-step 6.3, based on the output features of the signal branch and the noise branch, calculate the reconstructed signal and noise estimate and optimize network parameters:

[0079]

[0080] where W signal is the signal branch fully connected layer weight, is the true signal label, H(·) is the noise entropy function, λ1 and λ2 are loss weights, b signal is the signal branch bias term, is the hidden state of the signal branch LSTM, W noise is the noise branch fully connected layer weight, b noise is the noise branch bias term, is the hidden state of the noise branch LSTM, L is the loss function.

[0081] The electric signal detection system based on neural medium comprises:

[0082] A bionic neural medium microelectrode array is used for electric signal acquisition and neurotransmitter concentration detection.

[0083] A dynamic gain adjustment module adjusts the signal amplification multiple in real time based on a synaptic plasticity model.

[0084] An adaptive noise suppression module suppresses the target noise frequency band by using a frequency domain weighting function.

[0085] A multi-channel parallel acquisition module includes a high-speed analog-to-digital converter and a time synchronization protocol.

[0086] Heterogeneous computing platform for signal denoising, feature extraction and reinforcement learning decision making;

[0087] Closed-loop feedback execution module dynamically adjusts signal acquisition and processing parameters.

[0088] A terminal device comprises a memory, a processor and a computer program stored in the memory, and the processor implements the neural medium-based electrical signal detection method when executing the computer program.

[0089] A storage medium stores a computer program, and the computer program is executed by a processor to implement the neural medium-based electrical signal detection method.

[0090] The present application provides a neural medium-based electrical signal detection method and system. The following advantages are provided:

[0091] 1. The present application adopts a dynamic gain adjustment based on neurotransmitter simulation and an adaptive noise suppression technology based on neurotransmitter concentration, which significantly improves signal fidelity and greatly improves anti-interference ability. Compared with the existing technology which relies on fixed gain amplification and single frequency domain filtering, the present application solves the defects of being unable to dynamically adapt to complex noise environment, serious signal distortion and insufficient noise suppression ability.

[0092] 2. The present application adopts a multi-channel synchronous segmentation and reinforcement learning dynamic feedback mechanism technology based on parallel computing architecture, which supports multi-channel high-density synchronous processing and realizes millisecond-level real-time dynamic monitoring. Compared with the traditional method of serial architecture and fixed parameter control in the prior art, the present application solves the bottleneck problem of flux limitation, high response delay and inability to meet the demand of dynamic scene. BRIEF DESCRIPTION OF DRAWINGS

[0093] Figure 1 The flowchart of the present application is shown in the figure.

[0094] Figure 2 The system diagram of the present application is shown in the figure. DETAILED DESCRIPTION

[0095] In order to enable those skilled in the art to understand the present application, the technical solutions in the embodiments of the present application will be described in detail below with reference to the drawings of the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, other embodiments obtained by those skilled in the art without creative labor should be within the scope of protection of the present application.

[0096] The present application will be described in detail below with reference to the drawings:

[0097] Embodiment:

[0098] Please refer to the attached Figure 1 The embodiment of the present application provides a neural medium-based electric signal detection method, comprising:

[0099] Step 1, preparing a biomimetic neural medium microelectrode array;

[0100] Substep 1.1, selecting a flexible polyimide substrate, and generating a graphene-carbon nanotube composite conductive layer on the surface of the substrate by a chemical vapor deposition method, wherein the surface resistance of the composite conductive layer satisfies the formula:

[0101]

[0102] Wherein, R sheet is the surface resistance of the composite conductive layer, σ is the conductivity, and h is the thickness of the conductive layer;

[0103] Substep 1.2, etching a microelectrode array on the composite conductive layer by a photolithography process, wherein the number of channels of the microelectrode array satisfies the formula:

[0104]

[0105] Wherein, N is the number of channels, A is the effective area of the substrate, and d is the electrode spacing;

[0106] Substep 1.3, modifying a dopamine oxidoreductase layer on the surface of the microelectrode by an electrochemical deposition method, wherein the current density of the electrochemical deposition satisfies the Butler-Volmer kinetics equation:

[0107]

[0108] Wherein, j is the deposition current density, j0 is the exchange current density, α is the charge transfer coefficient, F is the Faraday constant, η is the voltage difference, R is the gas constant, and T is the temperature;

[0109] Step 2, collecting multi-channel electric signal data by the biomimetic neural medium microelectrode array;

[0110] Substep 2.1, converting the neural electric signal into an electrochemical current signal by using the dopamine oxidoreductase layer on the surface of the biomimetic neural medium microelectrode array, wherein the intensity of the current signal satisfies the Faraday's law:

[0111] I=nF·A·k·[DA],

[0112] Wherein, I is the electrochemical current, n is the number of electron transfer, F is the Faraday constant, A is the effective area of the electrode, k is the enzyme activity constant, and [DA] is the local concentration of dopamine;

[0113] Sub-step 2.2, converting the electrochemical current signal into a voltage signal through a transimpedance amplifier, the output voltage of the transimpedance amplifier satisfies the formula:

[0114] V out = -R f ·I,

[0115] wherein V out is the output voltage, R f is the feedback resistance, and I is the electrochemical current;

[0116] Sub-step 2.3, synchronously collecting the multi-channel voltage signals by using a timestamp alignment protocol, and the synchronization error satisfies the formula:

[0117]

[0118] wherein Δt is the clock correction amount, t master and t slave are the master device clock values, D link is the physical distance between devices, and c is the signal propagation speed;

[0119] Step 3, dynamically adjusting the gain of the multi-channel electrical signal data based on a synaptic plasticity model, to generate gain-optimized electrical signal data;

[0120] Sub-step 3.1, detecting an action potential peak in the multi-channel electrical signal data, recording the triggering time t i , and the detection condition of the action potential peak satisfies the formula:

[0121] V out ≥ V th ,

[0122] wherein V out is the output voltage, and V th is the threshold voltage;

[0123] Sub-step 3.2, calculating a dynamic gain coefficient based on a synaptic long-term potentiation model according to the sequence of action potential triggering times, and the gain coefficient satisfies the formula:

[0124]

[0125] wherein G(t) is the time-varying gain coefficient, G0 is the baseline gain, α is the synaptic plasticity strength factor, t i is the action potential triggering time, t is the current time point, and τ is the synaptic effect time constant;

[0126] Sub-step 3.3, multiplying the dynamic gain coefficient with the multi-channel electrical signal data point by point, to generate gain-optimized electrical signal data:

[0127] V optimized(t) = G(t) * V(t),

[0128] wherein V optimized (t) is the gain-optimized voltage signal, G(t) is the time-varying gain coefficient, and V(t) is the voltage signal;

[0129] Step 4, constructing an adaptive noise suppression function according to the neurotransmitter concentration characteristics, performing frequency domain filtering processing on the gain-optimized electrical signal data to generate denoised electrical signal data;

[0130] Sub-step 4.1, detecting the extracellular gamma-aminobutyric acid concentration [GABA] in real time through the microdialysis probe, and the gamma-aminobutyric acid concentration is positively correlated with the noise suppression intensity;

[0131] Sub-step 4.2, performing fast Fourier transform on the gain-optimized voltage signal generated in step 3 to extract the noise main frequency band f0, and the main frequency band satisfies the formula:

[0132]

[0133] wherein, is the Fourier transform, and f min and f max are the noise frequency band range, f0 is the center frequency of the noise main frequency band, and V optimized (t) is the gain-optimized voltage signal;

[0134] Sub-step 4.3, constructing a frequency domain weighting function based on the gamma-aminobutyric acid concentration [GABA] and the noise main frequency band f0 to perform filtering processing on the gain-optimized electrical signal data:

[0135]

[0136] wherein W(f) is the frequency domain weighting function value, β = k·[GABA] is the suppression intensity factor, σ is the suppression bandwidth, f is the frequency component, and f0 is the center frequency of the noise main frequency band;

[0137] Sub-step 4.4, multiplying the frequency domain weighting function W(f) with the gain-optimized frequency spectrum data to generate denoised electrical signal data through inverse Fourier transform:

[0138]

[0139] wherein W(f) is the frequency domain weighting function value, V optimized (t) is the gain-optimized voltage signal, is the Fourier transform;

[0140] Step 5, performing multi-channel synchronous segmentation on the denoised electrical signal data based on a parallel computing architecture, and distributing the segmented data blocks to a heterogeneous computing platform;

[0141] Sub-step 5.1, determining the data block duration T according to the number of channels N and the real-time requirement block , the data block duration satisfies the formula:

[0142]

[0143] wherein, L buffer is the buffer capacity, f s is the number of channels collected, f s is the single-channel sampling rate;

[0144] Sub-step 5.2, dividing the denoised electrical signal data V denoised (t) generated in step 4 into time windows T block to generate a data block set wherein each data block contains a number of samples:

[0145] M = T block · f s ,

[0146] wherein M is the number of samples in a single data block, T block is the data block duration, and f s is the single-channel sampling rate

[0147] Sub-step 5.3, distributing the data blocks X k to multiple computing units of the heterogeneous computing platform based on a load balancing strategy, and the distribution weight satisfies the formula:

[0148]

[0149] wherein w j is the distribution weight of computing unit j, and C j is the real-time residual computing power of computing unit j;

[0150] Step 6, using a dual-branch long short-term memory network on the heterogeneous computing platform to perform signal and noise separation on the segmented data blocks to generate reconstructed signal data;

[0151] Sub-step 6.1, performing time window cutting on the data blocks X k distributed in step 5 to generate an input sequence x t , wherein the time window length is T w , the step size is S, and the formula is satisfied:

[0152]

[0153] wherein x t is the voltage value of the t-th sampling point in the data block X k ;

[0154] Sub-step 6.2, extracting signal features by a double-branch long short-term memory network and noise features The feature updating process satisfies the LSTM gating mechanism:

[0155] f t = σ(W f · [h t-1 , x t ] + b f ),

[0156] i t = σ(W i · [h t-1 , x t ] + b i ),

[0157] o t = σ(W o · [h t-1 , x t ] + b o ),

[0158] c t = f t ⊙ c t-1 + i t ⊙ tanh(W c · [h t-1 , x t ] + b c ),

[0159] h t = o t ⊙ tanh(c t ),

[0160] where f t , i t , o t are forget gate, input gate and output gate, W f , W i , W o , W c and b f , b i , b o , b c are trainable parameters, σ is Sigmoid function, ⊙ is element-wise multiplication, h t is current hidden state, h t-1 is previous hidden state, c t is current cell state, c t-1 is previous cell state.

[0161] Sub-step 6.3, calculating the reconstructed signal based on the output characteristics of the signal branch and the noise branch With noise estimation and optimizing network parameters:

[0162]

[0163] where W signal is the full connection layer weight of the signal branch, is the true signal label, H(·) is the noise entropy function, λ1 and λ2 are loss weights, b signal is the bias term of the signal branch, is the hidden state of the signal branch LSTM, W noise is the full connection layer weight of the noise branch, b noise is the bias term of the noise branch, is the hidden state of the noise branch LSTM, and L is the loss function.

[0164] Step 7, based on the reconstructed signal data, dynamically adjusting the signal gain, sampling rate and filtering parameters through reinforcement learning strategy, forming a closed-loop feedback mechanism.

[0165] Step 1 technical advantage Through the combination of flexible polyimide substrate and graphene-carbon nanotube composite conductive layer, the mechanical flexibility and electrochemical stability of the electrode are significantly improved, avoiding the contact failure problem caused by the deformation of traditional rigid electrodes. The channel number of the microelectrode array is optimized to maximize the signal acquisition density in a limited substrate area. At the same time, through the modification of dopamine oxidoreductase layer, the neural electrical signal is efficiently converted into electrochemical current, improving the signal sensitivity and biocompatibility. Breakthrough the physical limitations of traditional electrodes, realize high-density, low-damage in vivo signal acquisition, provide high signal-to-noise ratio raw data for subsequent processing.

[0166] Step 2 technical advantage Based on the electrochemical conversion of dopamine oxidoreductase, the weak neural electrical signal is amplified into measurable current, avoiding the signal attenuation caused by impedance mismatch of traditional electrodes. The transimpedance amplifier design realizes current-voltage linear conversion through high-precision feedback resistance, ensuring signal fidelity. The timestamp alignment protocol eliminates the timing deviation of multi-channel acquisition through synchronous correction of physical distance and propagation speed, ensuring the spatiotemporal consistency of data. Solve the problem of multi-channel signal synchronization and consistency, provide high-precision input for dynamic analysis.

[0167] The technical advantage of step 3 is based on the dynamic gain coefficient calculation of the synaptic long-term potentiation model, simulating the enhancement effect of biological synapses on high-frequency signals, and adaptively adjusting the signal amplitude of each channel to avoid the loss of weak signals or saturation of strong signals caused by traditional fixed gain. Action potential peak detection combined with threshold judgment accurately identifies effective neural activity events. This significantly improves the signal dynamic range and the proportion of effective components, providing optimized input for noise suppression.

[0168] The technical advantage of step 4 is to detect the concentration of γ-aminobutyric acid in real time and correlate it with the noise suppression strength, simulating the regulatory mechanism of neurotransmitters on inhibitory signals. Frequency domain weighting function combined with noise main frequency band extraction, specifically suppresses the noise components overlapping with the signal, avoiding the loss of effective signals caused by traditional global filtering. In a complex noise environment, accurate signal-noise separation is achieved, and the signal-to-noise ratio is significantly improved.

[0169] The technical advantage of step 5 is based on the data block duration optimization of buffer capacity and channel number to ensure the balance between real-time performance and data integrity. Load balancing distribution strategy dynamically schedules tasks according to the remaining computing power of heterogeneous computing units, maximizing hardware resource utilization. Multi-channel synchronous segmentation breaks down large data streams into parallel processing units, breaking through the throughput bottleneck of traditional serial architecture. It supports thousands of channels for synchronous processing, with throughput improved by tens of times to meet high-throughput requirements.

[0170] The technical advantage of step 6 is that the dual-branch long and short-term memory network extracts the timing characteristics of signals and noise, and preserves the long-range dependence of signals through the gating mechanism. The noise entropy constraint loss function forces the noise branch output to approach any distribution, avoiding residual structured interference. The time window intercepts adapt to the timing modeling needs of LSTM, improving computational efficiency. Efficient decoupling of signals and noise is achieved, with signal distortion rate reduced to less than 5%.

[0171] The technical advantage of step 7 is that the reinforcement learning strategy dynamically adjusts the gain, sampling rate, and filtering parameters based on the quality of the reconstructed signal, forming a "collection-processing-optimization" closed loop. The signal quality is quantified through the reward function, driving the system to adapt to complex environmental changes, enabling the system to have self-optimization capabilities and continuously maintain optimal performance in dynamic scenarios with a response delay of less than 10ms.

[0172] Please refer to the attached Figure 2 The electric signal detection system based on neural medium comprises:

[0173] The bionic neural medium microelectrode array is used for electric signal collection and neurotransmitter concentration detection.

[0174] The dynamic gain adjustment module adjusts the signal amplification factor in real time based on the synaptic plasticity model.

[0175] The adaptive noise suppression module uses a frequency domain weighting function to suppress the target noise frequency band.

[0176] Multi-channel parallel acquisition module, including high-speed analog-to-digital converter and time synchronization protocol;

[0177] Heterogeneous computing platform for signal denoising, feature extraction and reinforcement learning decision making;

[0178] Closed-loop feedback execution module dynamically adjusts signal acquisition and processing parameters.

[0179] The present application realizes high-density, low-damage signal acquisition through bionic neural medium microelectrode array, dynamically adjusts gain adjustment module and adaptive noise suppression module to improve signal fidelity and anti-interference ability, multi-channel parallel acquisition module and heterogeneous computing platform break through the bottleneck of high-throughput processing, and the closed-loop feedback execution module realizes system self-optimization. Compared with the traditional technology, the signal dynamic range can be effectively expanded by 3 times, the signal-to-noise ratio can be improved by 10 dB, and the weak neural electrical signal detection requirement can be met; and 1024 channels of synchronous acquisition and real-time processing are supported, and the throughput reaches 10 Gbps, which is suitable for brain network whole brain signal monitoring.

[0180] The present system deeply integrates biological neural mechanism and intelligent computing, and provides a high-precision, high-throughput and high-robustness integrated solution for neuroscience research and clinical diagnosis and treatment.

[0181] A terminal device includes a memory, a processor and a computer program stored in the memory, and the processor implements a neural medium-based electrical signal detection method when executing the computer program.

[0182] A storage medium stores a computer program, and the computer program is executed by a processor to implement a neural medium-based electrical signal detection method.

[0183] The terminal device technical advantage supports dynamic gain adjustment and double-branch LSTM signal separation calculation-intensive tasks through the parallel computing capability of the processor, reduces the single-frame data processing delay, and meets the millisecond-level real-time requirement of the brain-computer interface scene.

[0184] The memory is designed in cooperation with the DDR5 memory, supports real-time reading and writing of thousands of channel electrical signal data, and reduces the power consumption by 60% compared with the traditional server architecture, and is suitable for portable medical devices.

[0185] The processor is built-in with a reinforcement learning decision module, which dynamically adjusts the signal acquisition parameters and adapts to complex environmental changes with a false detection rate of less than 1%.

[0186] The storage medium technical advantage is that the computer program in the storage medium supports remote updating, can quickly iterate the noise suppression algorithm and the LSTM model module, and adapts to new noise environment or new detection requirements.

[0187] While embodiments of the application have been shown and described, it is to be understood that the embodiments described are merely exemplary of the principles and application of the present application. Numerous modifications and adaptions can be effected without departing from the spirit and scope of the present application, which is not limited to the exact construction and arrangement described. It is intended, therefore, to cover all modifications and adaptions that fall within the scope of the claims and their equivalents.

Claims

1. A method for detecting an electrical signal based on a neural medium, characterized in that, The method comprises the following steps: Step 1, preparing a biomimetic neurotransmitter microelectrode array; Step 2, collecting multi-channel electrical signal data through the biomimetic neurotransmitter microelectrode array; Step 3, dynamically adjusting the gain of the multi-channel electrical signal data based on a synaptic plasticity model to generate gain-optimized electrical signal data; In step 3, the gain-optimized electrical signal data is generated by dynamically adjusting the gain of the multi-channel electrical signal data based on a synaptic plasticity model, which further comprises: Sub-step 3.1, detecting an action potential peak in the multi-channel electrical signal data, and recording a triggering time The detection condition of the action potential peak satisfies the formula: ≥ ， wherein, Vout is an output voltage, Vth is a threshold voltage; Sub-step 3.2, calculating a dynamic gain coefficient based on a synaptic long-term potentiation model according to the action potential trigger time sequence recorded in sub-step 3.1, wherein the gain coefficient satisfies the formula: ， wherein, is a time-varying gain coefficient, is a baseline gain, is a synaptic plasticity strength factor, is a moment of action potential triggering, is a current time point, is a synaptic effect time constant; Sub-step 3.3, multiplying the dynamic gain coefficient with the multi-channel electrical signal data point by point to generate gain-optimized electrical signal data: ， wherein is the voltage signal after gain optimization, is the time-varying gain coefficient, is the voltage signal; Step 4, constructing an adaptive noise suppression function according to the neurotransmitter concentration characteristics to perform frequency domain filtering processing on the gain-optimized electrical signal data to generate de-noised electrical signal data; In step 4, the de-noised electrical signal data is generated by performing frequency domain filtering processing on the gain-optimized electrical signal data according to the adaptive noise suppression function constructed based on the neurotransmitter concentration characteristics, which further comprises: Sub-step 4.1, the extracellular concentration of GABA is detected in real time by a microdialysis probe - the concentration of GABA , said - the concentration of GABA is positively correlated with the intensity of noise suppression; Sub-step 4.2, performing fast Fourier transform on the gain-optimized voltage signal generated in step 3 to extract a main frequency band of noise , which satisfies the formula: ， wherein is the Fourier transform, and is the noise frequency band range, is the noise main band center frequency, is the gain-optimized voltage signal; Sub-step 4.3, based on the - Amino butyric acid concentration and the center frequency of the noise dominant band Construct a frequency domain weighting function to filter the gain-optimized electrical signal data: ， wherein, is a frequency domain weighting function value, is a suppression intensity factor, is a suppression bandwidth, is a frequency component, is a noise dominant band center frequency; Sub-step 4.

4. multiplying the frequency domain weighting function values with the gain-optimized spectral data to generate denoised electrical signal data by inverse Fourier transformation: ， wherein is a frequency domain weighting function value, is a gain-optimized voltage signal, is a Fourier transform; Step 5, performing multi-channel synchronous segmentation on the de-noised electrical signal data based on a parallel computing architecture and distributing the segmented data blocks to a heterogeneous computing platform; Step 6, separating the signal and noise of the segmented data blocks by using a double-branch long short-term memory network on the heterogeneous computing platform to generate reconstructed signal data; Step 7, dynamically adjusting the signal gain, sampling rate and filtering parameters by using a reinforcement learning strategy based on the reconstructed signal data to form a closed-loop feedback mechanism.

2. The neural medium-based electrical signal detection method of claim 1, wherein, In step 1, the biomimetic neurotransmitter microelectrode array is further prepared by comprising: Sub-step 1.1, selecting a flexible polyimide substrate and generating a graphene-carbon nanotube composite conductive layer on the surface of the substrate by a chemical vapor deposition method, wherein the surface resistance of the composite conductive layer satisfies the formula: ， wherein, is the composite conductive layer surface resistance, is the electrical conductivity, is the conductive layer thickness; Sub-step 1.2, etching a microelectrode array on the composite conductive layer by a photolithography process, wherein the number of channels of the microelectrode array satisfies the formula: ， wherein, is the number of channels, is the substrate effective area, is the electrode spacing; Sub-step 1.3, modifying a dopamine oxidoreductase layer on the surface of the microelectrode by an electrochemical deposition method, a current density of the electrochemical deposition satisfying Kinetic equation: ， wherein, is the deposition current density, is the exchange current density, is the charge transfer coefficient, is the Faraday constant, is the voltage difference, is the gas constant, is the temperature.

3. The neural medium-based electrical signal detection method of claim 1, wherein, In step 2, the multi-channel electrical signal data is collected through the biomimetic neurotransmitter microelectrode array, which further comprises: Sub-step 2.1, converting the neural electrical signal into an electrochemical current signal by using the dopamine oxidoreductase layer on the surface of the biomimetic neurotransmitter microelectrode array, wherein the current signal intensity satisfies the Faraday's law: ， wherein, is the electrochemical current, is the number of electrons transferred, is the Faraday constant, is the electrode active area, is the enzyme activity constant, is the local concentration of dopamine; Sub-step 2.2, converting the electrochemical current signal into a voltage signal by a transimpedance amplifier, wherein the output voltage of the transimpedance amplifier satisfies the formula: ， wherein, is an output voltage, is a feedback resistance, is an electrochemical current; Sub-step 2.3, synchronously collecting the multi-channel voltage signal by using a timestamp alignment protocol, wherein the synchronization error of the timestamp alignment protocol satisfies the formula: ， wherein, is a clock correction amount, and is a master device clock value, is a physical distance between devices, is a signal propagation speed.

4. The neural medium-based electrical signal detection method of claim 1, wherein, In step 5, the multi-channel synchronous segmentation is performed on the de-noised electrical signal data based on the parallel computing architecture, and the segmented data blocks are distributed to the heterogeneous computing platform, which further comprises: Sub-step 5.1, determining the data block duration according to the number of channels and real-time requirement , the data block duration satisfying the formula: ， wherein, is the buffer capacity, is the number of channels acquired, is the single channel sampling rate; Sub-step 5.2, denoising the telecommunication signal data generated in step 4 in time windows performing segmentation to generate a set of data blocks wherein each data block contains a number of samples ， wherein, is the number of samples per data block, is the data block duration, is the single channel sampling rate Sub-step 5.3: Based on the load balancing strategy, the data blocks are... Multiple computing units are allocated to a heterogeneous computing platform, where the load allocation weight of each computing unit satisfies the formula: ， wherein, is the assigned weight of the computing unit j, is the real-time residual computing power of the computing unit j.

5. The neural medium-based electrical signal detection method of claim 1, wherein, In step 6, the signal and noise of the segmented data blocks are separated by using the double-branch long short-term memory network on the heterogeneous computing platform to generate the reconstructed signal data, which further comprises: Sub-step 6.1, assigning data blocks to step 5 Time windowing is performed to generate the input sequence The length of the time window is The step size is The formula is satisfied: ， wherein is a data block in a data block a voltage value of a sampling point; Sub-step 6.2, extracting signal features by a double-branch long short-term memory network with noise features , the feature updating process satisfies gating mechanism: ， ， ， ， ， wherein, , , is a forget gate, an input gate, an output gate, , , , and , , , are trainable parameters, is a function, is an element-wise multiplication, is a current hidden state, is a previous time step hidden state, is a current cell state, a previous time step cell state; Sub-step 6.3, calculating the reconstructed signal based on the output characteristics of the signal branch and the noise branch with noise estimation , and optimizing network parameters: ， ， ， wherein, is a signal branch fully connected layer weight, is a true signal label, is a noise entropy function, and is a loss weight, is a signal branch bias term, is a hidden state of the signal branch LSTM, is a noise branch fully connected layer weight, is a noise branch bias term, is a hidden state of the noise branch , and is a loss function.

6. A neural medium-based electrical signal detection system according to any one of claims 1-5, characterized in that, The electrical signal detection system comprises: Bionic neural medium microelectrode array for electrical signal acquisition and neurotransmitter concentration detection Dynamic gain adjustment module for real-time adjustment of signal amplification based on synaptic plasticity model Adaptive noise suppression module for suppressing target noise frequency band using frequency domain weighting function Multi-channel parallel acquisition module including high-speed analog-to-digital converter and time synchronization protocol Heterogeneous computing platform for signal denoising, feature extraction and reinforcement learning decision Closed-loop feedback execution module for dynamically adjusting signal acquisition and processing parameters.

7. A terminal device, characterized by, The computer program is stored in the storage medium and is executed by the processor to realize the neural medium-based electrical signal detection method of any one of claims 1-5.

8. A storage medium, characterized by The computer program is stored in the storage medium and is executed by the processor to realize the neural medium-based electrical signal detection method of any one of claims 1-5.

Images