Online neuron classification method based on neuromorphic computing
The present invention discloses an online neuron spike sorting method based on neuromorphic computing, which converts neuron spike signals collected from the cerebral cortex into spike signals through field coding, classifies different waveforms and corresponding time stamps by means of spiking neural networks, and realizes online neuron spike sorting; at the same time, the online update method of spiking neural network is used to adapt to the online changes of neuronal spike waveform and improve the accuracy of long-term online neuronal spike sorting. This method has fast computational speed, which can improve the speed of spike sorting process, maintain high consistency in classification on different datasets, and facilitate the deployment of implanted chips.
1. An online neuron spike sorting method based on neuromorphic computing, comprising the following steps:
(1) obtaining original multi-channel neural signals, removing low-frequency local field potentials through a bandpass filter, and performing whitening preprocessing and artifact removal on each channel's neural signals;
(2) detecting and aligning candidate spikes on each signal channel, specifically by using a nonlinear energy operator to calculate the energy intensity of each position in the discrete signal, a time window exceeding the threshold is determined as a candidate spike, and then aligning the candidate spike based on the spike position;
(3) constructing a spiking neural network, with the first layer being the perception layer and the second layer being the cognitive layer, each neuron on the cognitive layer connects to the perception layer neurons in a fully connected manner and dynamically updates the connecting synapses;
(4) the perception layer of the spiking neural network is used to spike code the candidate spike, and the discrete signal of each time point of the input candidate spike is mapped to a group of spike sequences in the form of Gaussian Receptive field coding;
neurons on the cognitive layer responding to different spike sequence inputs, and updating the connecting synapses between the activated neurons with the corresponding neurons in the perception layer based on the winner-take-all mechanism; when the cumulative voltage of neurons in the cognitive layer exceeds the voltage threshold, the neurons in the cognitive layer output the spike sequence as a time stamp sequence in response to the action potential of different cells;
the form of Gaussian Receptive field coding is as follows:
I (x,y,t) =P μ,δ ( S t )
wherein, μ is the central position of neurons in the Receptive field, δ is the width of neurons in the Receptive field, S t is the signal sequence at time t, I (x,y,t) is the spike firing of the neurons (x, y) in the perception layer at time t, and P is the Poisson process of the Gaussian Receptive field;
the winner-take-all mechanism is: when a neuron is activated, other neurons are suppressed and not updated, only the weight of connecting synapses between the activated neuron with the neurons in the perception layer is enhanced or reduced;
updating the connecting synapses between the activated neurons with the corresponding neurons in the perception layer, the neuron selection method is as follows:
ϵ
.
=
max
ϵ
z
(
ϵ
,
t
)
wherein, {dot over (∈)} is the neurons in the cognitive layer for selected execution updates, z(∈,t) is the voltage value of neurons in the cognitive layer at time t;
in the initial state, all weight values are initialized, utilizing the Hebb learning rule to force neurons to find waveforms of interest, each neuron in the cognitive layer is fully connected to the perception layer, and the weights of these synapses are initialized to zero, when outputting neuron triggers, the Hebb learning rules are applied to the input synapses;
when updating the connecting synapses between the activated neurons with the corresponding neurons in the perception layer, the weight update method of the connecting synapses between the two layers is as follows:
ω
ˆ
t
+
1
=
{
min
(
ω
t
+
τ
stdp
+
,
ω
max
)
,
if
presynapses
firing
max
(
ω
t
-
τ
stdp
-
,
ω
min
)
,
if
postsynapses
firing
wherein, {circumflex over (ω)} t+1 is the synaptic weight at t+1 time after update, ω t is the synaptic weight at time t before the update, τ stdp+ is the constant for postsynapses firing, τ stdp− is the constant for presynapses firing, ω max is the maximum value of synaptic weight, ω min is the minimum value of synaptic weight;
(5) for the original neural signal corresponding to the time stamp sequence, the spike and noise are divided according to a pre-set threshold, and each channel reconstructs waveforms from different cells based on the time stamp sequence output by the spiking neural network.
2. The online neuron spike sorting method based on neuromorphic computing according to claim 1 , wherein, in the step (1), the bandpass filter adopts a 3rd order Butterworth filter with a bandpass frequency of 300-3000 Hz.
3. The online neuron spike sorting method based on neuromorphic computing according to claim 1 , wherein, in the step (2), using a nonlinear energy operator to calculate the energy intensity of each position in the discrete signal, the formula is:
ψ[ x ( n )]= x 2 ( n )− x ( n+ 1)· x ( n− 1)
wherein, x(n) is the sampling point of the n time waveform.
4. The online neuron spike sorting method based on neuromorphic computing according to claim 1 , wherein, in step (2), when aligning the candidate spike based on the spike position, the spike position is first interpolated through upsampling, and after realignment, the waveform is downsampled to its original length.