31 results about "Neuron membrane" patented technology

The invention discloses an image classification method based on a pulse neural network. The method comprises a step of encoding an externally input image analog quantity into a pulse time series, a step of adding time delay information to the pulse time series and storing the pulse time sequence with the added time delay information into an FIFO memory for buffering, a step of inputting the pulsetime sequence with the currently added time delay information into an IF pulse neuron model to generate a neuron membrane voltage signal, wherein a pipeline architecture is adopted by the IF pulse neuron model and a neuron model calculation method is optimized, and a step of generating a neural pulse sequence based on a Poisson distribution by comparing the membrane voltage signal and a thresholdand carrying out classification and judgment on the membrane voltage signal. According to the image classification method, the pulse neural network has powerful computing power and can simulate various neuron signals and arbitrary continuous functions, the operation efficiency is high, and so the pulse neural network has a hardware realization value.

Described is system for simulating spiking neural networks for image and video processing. The system processes an image with a spiking neural network simulator having a plurality of inter-connected modules. Each module comprises a plurality of neuron elements. Processing the image further comprises performing a neuron state update for each module, that includes aggregating input spikes and updating neuron membrane potentials, and performing spike propagation for each module, which includes transferring spikes generated in a current time step. Finally, an analysis result is output.

The invention discloses an image classification method based on a multilayer spring convolutional neural network, and relates to the field of image processing, and the method comprises the steps: converting images in a training set into pulse sequences; setting a spring neuron parameter, and constructing a convolutional neural network by using a spring neuron; taking the pulse sequence as input, training the convolutional neural network layer by layer, obtaining visual features of the pulse sequence, and obtaining a classification result, the training method being an unsupervised learning algorithm based on a Hebbian rule; converting the to-be-identified image into a pulse sequence, and inputting the pulse sequence into the trained convolutional neural network to obtain a classification result of the to-be-identified image. The technical problem of sparking neuron membrane voltage redundancy calculation caused by increased neuron scale is solved, and meanwhile, the technical problem oflearning non-convergence caused by the fact that an SNN activation function cannot be derived and a back-propagation calculation residual error cannot be used is solved.

A method and system are provided for updating a neuron membrane potential in a spike time dependent plasticity model in a Neuromorphic system. The method includes approximating a shape of an analog spike signal from an axon input using a hardware-based digital axon timer. The method further includes generating a first intermediately updated neuron membrane potential value from a current axon timer value, a current synapse weight value and a current neuron membrane potential value using a first look-up table and an accumulator. The method also includes generating a second intermediately updated neuron membrane potential value with a leak decay effect using a second look-up table and the first intermediately updated neuron membrane potential value. The method additionally includes generating a final updated neuron membrane potential value based on a comparison of the second intermediately updated neuron membrane potential value with a neuron fire threshold level using a comparator.

The invention provides a neuron electrical activity simulator under electromagnetic radiation. The neuron electrical activity simulator includes a neuron circuit, and is characterized in that the simulator also includes an electromagnetic radiation effect circuit and an adder, an input end of the electromagnetic radiation effect circuit is connected with an output end of the neuron circuit and equivalent outside magnetic flux, so as to input neuron response, i.e., membrane potential x and equivalent outside magnetic flux phi<ext>, an output end of the electromagnetic radiation effect circuit is connected with the adder, after an output electromagnetic radiation effect i of the electromagnetic radiation effect circuit to a neuron and an ionic current I<ext> that flows into a neuron cell membrane are summed through the adder, an input end of the neuron circuit is accessed, and after entering a stable operation state, the neuron electrical activity simulator outputs neuron response, i.e., membrane potential, which contains an electromagnetic memory effect. The neuron electrical activity simulator under electromagnetic radiation considers the electromagnetic memory effect of neuron membrane potential discharge, and guarantees the neuron electrical activity law.

The invention discloses a neuron circuit which comprises a pulse generation circuit, an adjustment circuit, and a comparison circuit. The pulse generation circuit is configured to simulate nerve impulse oscillation through a first Tau-cell circuit structure and a second Tau-cell circuit structure, the first Tau-cell circuit structure includes a first capacitor Cv used for simulating a neuron membrane potential v, and the second Tau-cell circuit structure includes a second capacitor Cu used for simulating a neuron membrane potential adjustment variable u. The adjustment circuit is connected with the pulse generation circuit, and is used for reassigning a value to the neuron membrane potential v. The comparison circuit is connected with the pulse generation circuit, and is used for reassigning a value to the neuron membrane potential adjustment variable u. The power consumption of the neuron circuit can be lowered, and the occupied area of the neuron circuit can be reduced.

The invention provides a method for calculating neuron membrane voltage of a spiking neural network, and relates to the technical field of spiking neural networks, and the method comprises the steps: firstly, converting an input signal into a pulse sequence through employing a time coding method, selecting a Spike Response Model neuron model as a membrane voltage calculation model of a neuron non-nonstress time period of the spiking neural network, and carrying out the calculation of the membrane voltage of the spiking neural network; then establishing a membrane voltage model generated by pulse accumulation of the pre-synaptic neurons in the post-synaptic neuron non-stress period, and finally establishing a pulse neural network neuron membrane voltage calculation model to realize calculation of the post-synaptic neuron membrane voltage; according to the method, the accumulation of neuron input pulses before the non-stress period highlights is calculated, the loss of input information is made up, the method is applied to image processing, the loss of information is reduced, and the utilization rate of the input information is improved.

The invention provides a reconfigurable autonomous learning spiking neural network processor. The reconfigurable autonomous learning spiking neural network processor comprises a processing unit arraycomposed of a plurality of processing units and channels in the east, south, west and north directions. Each processing unit comprises an external routing module, a sequencing module, a pulse queue module, a controller module, a search module, a memory module, a client module, a server module, an internal routing module, a plurality of exclusive computing resources and a plurality of borrowable computing resources. The processor adopts a pulse packet composed of pulse generation time and a source neuron ID to transmit signals; the computing executing modes of the computing resources include aninference mode and a learning mode, and a target neuron membrane potential, a synaptic weight related variable or a synaptic weight is updated by adopting a reconfigurable circuit; the computing resources comprise a self-adaptive clock-driven and event-driven computing mechanism module, and according to the updating time interval, the computing mode that the computing unit executes updating computing is changed in a self-adaptive mode.

The invention belongs to the field of data processing, particularly relates to a saliency map generation system, method and device based on a dynamic vision sensor, and aims to solve the problems. Themethod comprises the following steps: acquiring event data through a dynamic vision sensor, accumulating and caching DVS event data and information acquired from mutually associated neurons through each neuron in a target detection network, acquiring a neuron membrane potential through an activation function every other preset time, and outputting a pulse signal according to the size relationshipbetween the neuron membrane potential and a preset saturation threshold; accumulating the membrane potential u (t0) at the previous moment with the currently received excitation through attenuation,and using the result for updating and judging the pulse signal and the membrane potential. According to the invention, redundant background information of the saliency map generated by the target detection network array is eliminated, the credibility and accuracy of target detection are improved, and misjudgment is avoided.

The invention discloses an FMRI signal noise processing method, in particular to a thermal noise processing method. The method has the advantages that a noise value of Gaussian white noise in nerve cells, affecting neuron membrane potential and energy function, is measured, and the signal-to-noise ratio of neural signals to Gaussian white noise in nerve cells is further calculated; accordingly, FMRI signals can be subjected to spectral analysis, the influence of white noise upon the FMRI signals is eliminated, and more accurate results are obtained.

The invention discloses a GPU-based spiking neural network simulation method. The method comprises the steps of initializing a neural network structure and a network weight; loading a data set and carrying out pulse coding by adopting a pulse coding module; calling a GPU calculation module to select a proper GPU to calculate the pulse neuron membrane voltage according to the data volume, the calculation amount and the priority of the calculation task, comparing whether the pulse neuron membrane voltage exceeds a threshold value or not, and issuing a pulse; creating a pulse queue, and adding neurons of a trigger pulse into the pulse queue; if the pulse queue is not empty, finding out a post-synaptic neuron corresponding to the next layer according to the network structure, and repeating the steps S3-S5 until an output layer is reached; and calculating a loss function according to the result of the output layer and the actual pulse result, and updating the neural network by adopting a gradient descent mode until the iteration is completed. According to the invention, the training speed of the spiking neural network is accelerated, the advantages of the spiking neural network in the aspects of low power consumption and low time delay are exerted, and the conditions of overlarge data set, insufficient video memory and incapability of training are avoided.

The invention provides a spiking neural network target identification method. The loss of neuron membrane potential is effectively avoided by constructing the proportional attenuation characteristic of a spiking neuron model, the spiking emission rate of neurons is ensured, and the problem that neuron spiking emission is not activated enough is solved; the problem of neuron spiking emission rate over-activation is effectively reduced by constructing the self-adaptive threshold characteristics of the spiking neuron model.

The invention discloses a burn area segmentation system based on a pulse neural network U-shaped model, which comprises an image acquisition device and a neural network, and is characterized in that the image acquisition device is used for acquiring a burn image of a burn part of a patient, and the neural network comprises the pulse neural network U-shaped model and a pulse coupling neural network; the pulse neural network U-shaped model comprises a neuron membrane potential updating and pulse emitting layer, a lower sampling layer, an upper sampling layer, a feature fusion layer and a loss optimization module, training of burn area segmentation is carried out through the collected image, a burn area is segmented through the trained pulse neural network U-shaped model, and a burn area segmentation result is obtained. And the pulse coupling neural network calculates an adaptive threshold to binarize the segmentation result of the burn region to obtain a visualization result.

The invention provides a bionic neuron memristor and a preparation method thereof, and relates to the technical field of bionic neurons, the bionic neuron memristor comprises a substrate, parallel counter electrodes and a dielectric layer; the parallel counter electrode and the dielectric layer are arranged on the substrate; the parallel counter electrode comprises a positive electrode layer and a negative electrode layer, and the dielectric layer is arranged between the positive electrode layer and the negative electrode layer; the dielectric layer is made of a semiconductor material of which crystal lattices contain low-activation-energy ions; two kinds of low-activation-energy ions exist in crystal lattices in the dielectric layer, are regulated and controlled by an electric field under driving of the electric field, move towards the positive electrode and the negative electrode respectively and are used for simulating the membrane potential change process generated by transportation change of sodium and potassium ions inside and outside a membrane when neurons are stimulated. According to the invention, simulation of neuron membrane potential change, neuron cumulative emission phenomenon and neuron refractory behavior can be realized without building a peripheral circuit.

The invention provides an optical flow-based target detection method for compound eye visual inspiration, which comprises the following steps of: determining synaptic input conductance matrixes for a first neural network and a second neural network according to an optical flow matrix output by an EMD (Empirical Mode Decomposition) array; obtaining output matrixes of the first neural network and the second neural network according to the membrane potential matrixes of the first neural network and the second neural network, and further determining a synaptic input conductance matrix of a third neural network; and according to the synaptic input conductance matrix of the third neural network, obtaining a neuron membrane potential matrix of the third neural network, establishing a nonlinear relationship between a single neuron membrane potential and a single neuron output in the third neural network, and obtaining a motion detection result of the target object. According to the method, the working principle of a perfectly evolved compound eye vision system is used as a starting point for reference, and parallel point-by-point space-time smoothing and nonlinear dichotomy transformation are performed on the primary motion detector array estimation optical flow, so that motion detection and target-background separation based on a biological perception optical flow mechanism are realized.

The invention discloses a pulse neural network simulation strategy based on a GPU. The pulse neural network simulation strategy comprises the following steps: initializing a neural network structure and a network weight; loading parameters and a network structure in the GPU, and creating a pulse queue; calculating a neuron membrane voltage according to the pulse distribution condition in the pulse queue; according to the membrane voltage value of the neuron and a threshold value, whether a pulse is emitted is judged; and the process from S3 to S5 is repeated until iteration is completed. According to the method, the simulation speed of the spiking neural network is accelerated, the advantages of GPU parallel computing and the characteristics of sparsity and concurrency of the spiking neural network are brought into full play, and meanwhile simulation of a larger-scale spiking neural network model is supported.