100 results about "Neuronal models" patented technology

The invention is applied to the field of mechanical fault diagnosis signal processing, and particularly provides a pulse neural network mechanical fault diagnosis method based on STFT dimension transformation, and the method comprises the steps: collecting a one-dimensional mechanical vibration signal, carrying out the wavelet decomposition, carrying out the wavelet reconstruction of a low-frequency component and a denoised high-frequency component, and carrying out the wavelet reconstruction of the low-frequency component and the denoised high-frequency component; obtaining a denoised one-dimensional vibration signal; performing short-time Fourier transform, and converting the time-frequency two-dimensional matrix into a time-frequency two-dimensional matrix; inputting the time-frequency two-dimensional matrix into an improved HH threshold neuron model, carrying out Poisson sparse coding on the time-frequency two-dimensional matrix, and only carrying out pulse response on signal significant features; constructing a suprathreshold coding convolutional network with residual connection, inputting a sparse coding matrix, training by adopting an unsupervised learning rule based on STDP, and adaptively adjusting a network synaptic weight; and inputting to a trained above-threshold coding convolutional network, and obtaining pulse emission activity of neurons of an output layer through network forward propagation to determine a fault diagnosis result.

The invention discloses a pulse neural network image classification method and system based on a space-time adaptive threshold, mainly solving the problems of poor nonlinear expression and limited time sequence and space feature processing ability in the prior art, and the scheme comprises the following steps: obtaining an image data set, and dividing the image data set into a training set and a test set; a spiking neural network main body structure comprising an input layer, a hidden layer and an output layer is selected, an existing neuron model is improved by introducing a space-time joint threshold adjustment mechanism, and improved neurons are placed in each neuron layer in the hidden layer to form a spiking neural network based on a space-time adaptive threshold. The training set is used to carry out iterative training; and inputting the test set into the trained pulse neural network to obtain an image classification result. According to the method, a space-time adaptive threshold mechanism is introduced, the threshold can be dynamically adjusted to adapt to time and space features, the processing capacity of the network on time sequence data and complex features and the classification accuracy of images are remarkably improved, and the method can be widely applied to dynamic visual tasks and event-driven scenes.

The invention provides a voice recognition method and device based on a brain-like model, electronic equipment and a storage medium, and the method comprises the steps: obtaining a whole-brain network topological structure according to a brain function network generated by human brain image data, and carrying out the recognition of a whole-brain network through employing a multi-class neuron model as a node and a synaptic plasticity model as an edge, constructing a multi-brain-region pulse neural network as a brain-like model; constructing a speech recognition framework of the brain-like model; electromagnetic intervention is applied to different brain areas of the brain-like model, optimal electromagnetic intervention parameters are determined by analyzing the voice recognition accuracy of the brain-like model before and after electromagnetic intervention, and brain-like model voice recognition is carried out according to the optimal electromagnetic intervention parameters. According to the invention, the speech recognition performance of the brain-like model can be effectively improved, the biological interpretability and neural information processing capability of the brain-like model are further improved, and the development of brain-like intelligence in the application of a mode recognition task is promoted.

The invention relates to the technical field of new energy system fault diagnosis, in particular to a solar cell coupling hydrogen production system fault diagnosis method based on big data, and the method comprises the steps: collecting photovoltaic, electrolytic cell and control unit multi-source heterogeneous data, and detecting the abnormality through an abnormality detection core model; matching and repairing by combining historical similar sections to obtain operation re-repairing data; a multi-scale expansion convolutional neural network is adopted to extract time sequence fault features, a backbone network is adopted to extract infrared image high-order features, and heterogeneous aggregation features are obtained through fusion of a spiking neuron model and virtual dendritic branches; constructing dynamic graph nodes and attributes according to features and components, and integrating multiple indexes to fit a causal association strength weight; and multi-link diagnosis: obtaining a fault type through a quantum annealing algorithm, positioning a fault through causal discovery and a virtual dry prediction algorithm, integrating and reporting to a server. The method can improve the fault diagnosis accuracy and real-time performance, and is suitable for fault monitoring of the solar hydrogen production system.

There is provided an improved neuron model for spiking neural networks. The neuron model comprises somatic synapses as well as dendritic synapses. The somatic synapses are used for receiving an input token, and the dendritic synapses are used for receiving contextual data. The contextual data is used to modulate the behavior of the spiking neuron. The neuron model may be used to implement an attention mechanism in SSNs and to implement a Mixture-of-Experts architecture.

The invention discloses a multi-frequency weak fault enhancement method based on fractional order resonance response atlas reconstruction, and the method comprises the steps: carrying out the noise reduction, trend term removal, phase compensation and modal decomposition of a mechanical vibration signal, determining a lag phase through a frequency sweep method for phase compensation, adjusting the phase of a same-frequency signal, reconstructing an effective component through the combination of a modal decomposition borrowing geometric method and a Pearson matrix, and carrying out the phase compensation. Constructing a nonlinear resonance model with historical memory; inputting the parameters into the same-direction coupling neuron model, dynamically updating and simulating transmission intensity change through synaptic weight, and solving response and step length; constructing a zero-energy coupling index based on a signal zero-crossing behavior, energy distribution and time-frequency analysis, wherein the time-frequency analysis comprises short-time Fourier transform and wavelet packet energy spectrum calculation to extract energy features; and iteratively optimizing parameters by using a quantum optimization algorithm, adjusting a rotation angle and preserving population diversity to prevent premature convergence, and outputting optimal parameters by taking a zero-energy coupling index as a target.

The invention relates to the technical field of pulse neural network computing hardware, in particular to a neural network computing circuit of a pulse self-attention mechanism. According to the method, invalid or inefficient pulse events are dynamically screened through the hardware mask module, the operation number is remarkably reduced, and calculation path delay and logic resource occupation are reduced; meanwhile, in order to adapt to novel networks with binary architecture such as QKFormer, calculation and storage of a V matrix are eliminated, and storage resources and calculation resources are further reduced; in addition, the event coding module only generates active neuron events, and input sparsity is achieved. A configurable IF neuron model is adopted, exponential operation is avoided, hardware implementation is facilitated, and the method is suitable for binary network deployment. The modular architecture can support function extension, assembly line and parallel work; and the parallelism degree of the design can be determined according to the actual data pulse distribution rate. Event driving and mask pruning are combined, so that the overall computing resources of the circuit are greatly reduced, and the power consumption is reduced.

This invention discloses an olfactory electrostimulation model for automatically generating current stimulation signals for an artificial nose, relating to the technical field of olfactory stimulation signal generation using biomimetic neural networks. The implementation steps include: First, constructing a biomimetic neural network framework. Based on the neuronal anatomical structure of the olfactory epithelium and olfactory bulb system, extracting the main neuron types, and organizing them according to their connection methods in biological systems; Second, using the Hindmarsh-Rose neuron model to model various neurons in the constructed network and simulate their dynamic firing behavior; Subsequently, setting the connection weights between neurons and the interlayer transmission current to realize the computational model of the entire biomimetic neural network; Finally, in the odor encoding and stimulation signal generation stage, the odor detection curve of the electronic nose is used as an external stimulus input to the biomimetic neural network. The network calculates and generates a series of peak signals, which are output as current stimulation signals to the olfactory bulb for stimulation, thereby realizing the olfactory induction function of the artificial nose.

This invention discloses a method for constructing a spiking neural network based on recurrent firing neurons, comprising: S1: constructing a recurrent firing spiking neuron model; S2: calculating positive and negative pulses based on the gradient function corresponding to each time step in the recurrent firing spiking neuron model to obtain cross-time-step positive and negative pulses; S3: backpropagating the cross-time-step positive and negative pulses back through the recurrent firing spiking neuron model to obtain a spiking neural network; S4: training the spiking neural network using a positive and negative membrane potential balance loss function to obtain a recurrent firing spiking neural network. This invention significantly improves the learning ability and information representation ability of the spiking neural network, while enhancing its robustness.

The application discloses a Spike-BP on-chip learning method and system based on a Ca-LIF neuron model and a processor, wherein a trainable linear leakage parameter and a positive and negative double-channel pulse transmitter mechanism based on calcium gating are adopted; for the Ca-LIF neuron model, a linear leakage operation can be realized only by subtraction, complex leakage compensation operations can be avoided when a gradient matrix is solved, hardware implementation complexity is reduced, a large amount of calculation resources of subsequent hardware design is saved, and the performance of pulse neural network training is improved.

The application discloses a three-dimensional Hindmarsh-Rose neuron model circuit under periodic excitation, comprising three neuron circuits and a matching circuit, and is composed of nine operational amplifiers and two multiplier circuits. The three-dimensional Hindmarsh-Rose neuron model circuit under periodic excitation can provide hardware support for studying neuron dynamic behaviors under periodic excitation, and has important theoretical research significance and practical application value.

The present invention relates to a method for enhancing the fault symptom of a dual-target optimized co-directional linear coupled biological neuron. A co-directional linear coupled biological neuron module is constructed by a co-directional linear coupled hyperbolic tangent biological neuron and a Gaussian biological neuron. Then, a characteristic frequency amplitude ratio measurement index and a dwell time distribution index are designed based on the model response as the dual-target synchronous optimization function of the genetic algorithm to optimize the parameters of the co-directional linear coupled biological neuron model, realizing the extraction of weak symptom features of strong noise and strong coupling machinery under complex working conditions and timely diagnosing the early weak faults of mechanical equipment.

This application discloses a method, apparatus, electronic device, and storage medium for recognizing facial expression images, relating to the field of image recognition technology. It includes: determining the original vector of the expression image based on its pixel values; performing principal component analysis on the original vector to obtain the target vector of the expression image; inputting the target vector into a feature extraction network to obtain feature information of the expression image; encoding the feature information using pulse frequency based on a Poisson distribution to obtain a pulse sequence of the expression image; and inputting the pulse sequence into an expression recognition model to obtain the recognition result of the expression image. The expression recognition model is a multilayer spiking neural network, where the spiking neuron model is an improvement based on the leaky integral discharge neuron (LIF) model. This application's technical solution offers better interpretability and flexibility, significantly reducing network energy consumption while improving computational power, making it suitable for more intelligent small robots such as customer service robots.

The invention discloses a novel discrete memristor coupling heterogeneous neuron model construction method and application, and designs a novel discrete memristor model to simulate two different neuron models of synaptic coupling to obtain a final novel discrete memristor coupling heterogeneous neuron model. A phase diagram, an iterative sequence, a bifurcation diagram, a Lyapunov exponential spectrum and other methods are used for analyzing the dynamic characteristics of the model, and it is found that the model presents rich dynamic behaviors such as hyper-chaos, chaos and periods and a neuron discharge mode. The high complexity shows that the novel discrete memristor coupling heterogeneous neuron model can be applied to the security field of image encryption and the like. The constructed coupled neuron model is a discrete system, so that the digital circuit of the model is easy to realize through a DSP (Digital Signal Processor). The coupling neuron model-based application to image encryption is good in effect and high in security.

The invention discloses a low-power-consumption buffer circuit for neuron stimulation, which belongs to the technical field of microelectronics and comprises a neuron model, a current driving module, a buffer stage operational amplifier and a feedback resistance network. The neuron model is used for simulating electrical characteristics of biological neurons, the input end of the neuron model is connected with a node WE of the current driving module, and the output end is connected with a node RE; the current driving module is used for generating a current stimulation pulse under the control of the enable signal and providing the current stimulation pulse to the input end of the neuron model for electrical stimulation; the buffer stage operational amplifier is used for stabilizing the port voltage of the neuron model at a node RE, and comprises an amplification stage, a bias stage, a current buffer stage, an output stage and an auxiliary current stimulation stage; and the feedback resistance network is used for sampling voltage from a node RE and providing feedback voltage for the buffer stage operational amplifier after voltage division. The high-performance buffer circuit under low power consumption is realized, the standby time is prolonged, the chip area cost is saved, and integration is facilitated.

The application discloses a simulation system supporting fast large-scale brain simulation, and belongs to the technical field of brain simulation.The simulation system comprises a hardware device layer, which provides a plurality of hardware device resources; a data communication layer, which provides a plurality of types of communication modes between the computing nodes; an operation abstraction layer, which provides a plurality of types of neuron models, synapse models, connection rules and learning rules; an API layer, which provides an API interface and receives user requirements through the API interface; a hardware abstraction layer, which calls corresponding hardware kernels; and a network abstraction layer, which firstly records the topological structure of a brain simulation network, performs resource allocation and neuron cluster mapping, then creates the brain simulation network on the hardware device, and finally performs training or execution of the brain simulation network.The application fully utilizes cluster hardware resources to realize faster and larger-scale brain simulation, and solves the problems of lack of resource scheduling and resource allocation, insufficient utilization of device parallelism, too long communication time consumption, and limited support of hardware and interfaces.

The invention relates to the field of gas concentration detection and intelligent traceability, in particular to a gas traceability trajectory decision-making method based on a pulse neural network, and aims to solve the problems of low gas source traceability precision, slow response and poor adaptability in the prior art. According to the method, a plurality of gas sensors and wind direction sensors are arranged, gas concentration and wind direction data are collected in real time, pulse coding processing is adopted, and the data are converted into pulse signals to be input into a pulse neural network. The spiking neural network adopts a leakage integral and distribution neuron model, and can dynamically adjust the moving direction of the equipment according to the change of gas concentration and wind direction data. By training the neural network, the network can learn and make decisions in real time, and the equipment is controlled to accurately track the gas source along the gas concentration gradient. According to the method, through multi-sensor data fusion and real-time learning, the unmanned equipment can quickly adapt to and accurately execute a gas source tracing task in a complex environment, and the precision and efficiency of gas tracing are remarkably improved.

The invention discloses an image encryption method based on fractional order heterogeneous neurons and DNA dynamic coding, and the method comprises the steps: 1) dividing an original color image into a two-dimensional matrix, and carrying out the partitioning processing; 2) constructing a fractional order heterogeneous neuron model; 3) generating a plurality of groups of chaotic sequences and dynamic keys; 4) dynamically selecting a DNA coding rule and an operation rule; 5) continuing to dynamically select the DNA decoding operation rule, and performing DNA decoding on the block matrix; 6) performing replacement operation on the three two-dimensional matrixes; and 7) generating a ciphertext image, combining the two-dimensional matrixes of the three channels after the operation in the steps, and generating an encrypted ciphertext image. The invention further discloses an image decryption method based on the fractional order heterogeneous neurons and DNA dynamic coding. The invention belongs to the technical field of information encryption, and solves the problems of poor security, low processing efficiency and insufficient dynamic adaptability of the existing image encryption technology.

The invention discloses a brain simulation-oriented high-performance numerical differential solver method and system. The method comprises the following steps of: receiving a neuron cluster model and simulation parameters thereof; the differential equation definition of the model is analyzed, and when the input is in a character string form, the input is converted into a function form; setting a current moment t, a simulation step length dt and simulation time simt; the simulation parameters are analyzed and classified; judging whether the model contains noise or not, if yes, initializing a stochastic differential equation solver, and otherwise, initializing an ordinary differential equation solver; adaptive operators of different neuron models are called in each step length to calculate numerical solutions, and variables are subjected to parallel calculation and updating in the calculation process; updating the variable yn and the current moment t when tlt; and when t is greater than or equal to simt, entering the next step of length calculation until t is greater than or equal to simt. According to the method, the high-performance differential solver supporting parallel solving of the whole neuron cluster is provided, the parallel acceleration characteristic of hardware such as a GPU is fully utilized, the simulation speed is increased, and large-scale brain simulation is achieved.

The invention relates to the technical field of neuron circuits, and provides a configurable neuron circuit and a control method, and the circuit comprises a mode selection module which selects a corresponding target working mode according to a received neuron configuration signal; the calculation module is connected to the mode selection module, and performs corresponding membrane potential updating according to the target working mode selected by the mode selection module and the received pulse information of the presynaptic neurons to obtain a membrane potential updating result; the comparison module is connected to the calculation module and is used for comparing the membrane potential updating result with a preset threshold value and determining whether the current neuron can generate a pulse or not according to a comparison result; and the pulse generation module is connected to the comparison module, and is used for generating a pulse signal and an AER (Advanced Encryption Register) coding group package and outputting updated membrane potential information when the current neuron is determined to generate the pulse. According to the technical scheme, selection of two neuron models can be achieved, the circuit structure is simplified, and meanwhile more complex behavior modes are achieved.

Embodiments of the present application provide a kind of electric appliance fault prediction method and device based on neuron model, electronic equipment and readable storage medium, the method comprises: constructing neuron model and weight regression model, determine the business weight mark and nerve node mark of the neuron model;Obtain the fault diagnosis sample data of specified electric appliance, and according to the fault diagnosis sample data of the specified electric appliance, the business weight mark and the nerve node mark, the neuron model is trained;Using the weight regression model determines the fitting value of the neuron model after training;When the fitting value meets preset threshold value, the neuron model after training is determined as the neuron model of training completion, to carry out fault prediction according to the specified electric appliance of the neuron model of training completion.This embodiment of the present application can avoid the information collected by the electric appliance of fault to cause intelligent perception error, improve the accuracy of intelligent perception.

The embodiment of the invention relates to the field of artificial intelligence, and provides an epilepsy electroencephalogram signal detection method and device based on frequency attention and a pulse recurrent neural network and a storage medium, and the method comprises the steps: obtaining an epilepsy electroencephalogram signal to be detected; performing datum line following coding on the epilepsy electroencephalogram signal based on a grid search algorithm to obtain a coded epilepsy electroencephalogram signal; constructing a bidirectional pulse recurrent neural network model based on a preset adaptive pulse neuron model; the bidirectional pulse recurrent neural network model is used as a reference model, the reference model is adjusted based on a frequency attention mechanism and a pulse full-connection linear coding mechanism, and an improved model is constructed; and inputting the epilepsy electroencephalogram signal into the trained improved model to output a detection result associated with the epilepsy electroencephalogram signal. By adopting the method, the model calculation performance can be improved on the basis of reducing the energy consumption, and the detection effect on the epilepsy electroencephalogram signals is improved.

The invention relates to the technical field of educational informatization, and discloses a smart classroom interactive behavior real-time supervision method fused with attention recognition, and the method comprises the steps: collecting classroom video streams, carrying out the time domain differential coding through an integral issuing neuron model, and generating teacher and student end pulse streams; constructing a self-adaptive confidence mask based on the visual pulse stream of the student side, performing gating repair and interpolation on the original video signal, and reconstructing a clean physiological blood volume pulse wave signal; extracting a student physiological feature vector, and calculating a dynamic lag order parameter according to the teacher end pulse stream; and performing time-varying Granger causality test on the teacher end pulse stream and the student physiological feature vector by using the parameters, and outputting an interactive behavior judgment result. By quantifying the causal association between the teaching stimulation and the physiological response of the student, the problems that the false attention state is difficult to distinguish and the non-contact physiological signal is susceptible to action interference in the prior art are solved, and the accuracy and robustness of classroom supervision are improved.

The invention discloses a quantum neuromorphic attention alignment algorithm based on AR and VR, and belongs to the technical field of quantum, and the algorithm comprises the steps: obtaining pulse data; the amplitude and the phase are coded, mapping to the Hilbert space is realized through a parameterized quantum gate, and a trainable attention weight is obtained; a pulse neural network is used for processing quantum state characteristics, an IZHXY neuron model is used for realizing a dynamic threshold mechanism, and synaptic weight updating follows an STDXY rule; according to the method, adversarial training is carried out through a quantum encoding QELXY device Q (x) and a classic generator G (z), modal alignment is realized through a WASSXY distance, the problems of modal splitting, intention misjudgment and low energy efficiency faced by an existing VR / AR algorithm can be solved, intelligent alignment and prediction of multi-sensory information in a virtual environment are realized, and user experience and system performance are improved.

This invention discloses an adaptive frequency-coordinated support method and system for energy storage and renewable energy grid-connected devices in weak power grids. First, the available active power is determined, and the grid connection point frequency is locked and input into a neural network model. Then, a fuzzy rule inference table is constructed, and the fuzziness is resolved to obtain the adaptive proportional coefficients of the neural network. The adaptive correction amount of the frequency reference value is determined, and the reference value of the total active power is calculated. Simultaneously, considering the safe operation of the system and equipment protection, the power adjustment range and power change rate of the units are limited. An economic-frequency coordinated optimization objective function is established, the power reference value is calculated, and a power command is issued. Finally, the actual output power of the power station is monitored, and an early warning is issued when the deviation exceeds a threshold. This invention achieves synergy, economy, and reliability of energy storage devices and renewable energy power station outputs by determining the correction amount of the frequency reference value from the perspective of eliminating the steady-state error of the weak power grid and allocating active power from the perspective of coordinated support.

This invention discloses a rapid incremental learning technique based on a spiking neuromorphic model, belonging to the field of computer vision technology. The invention selects the YOLOv3-tiny model as the base model for the SpikingYoLOv3 spiking neural network, and improves its adaptability. It uses the IF neuron model as the neuron model for SpikingYoLOv3. An incremental learning training method based on network expansion is employed for YOLOv3-tiny. The weights of the trained model are assigned to the newly constructed SpikingYoLOv3 model using an ANN-to-SNN conversion method. Poisson coding is used to encode the test images, converting them into asynchronous pulse sequences, which are then fed into the SpikingYoLOv3 model for testing. Comparative analysis with the corresponding YOLOv3-tiny model reveals that the SpikingYoLOv3 model has superior detection performance. Applying SpikingYoLOv3 to a battlefield environment enables accurate detection of different types of targets. This invention solves the problem of traditional network models needing retraining when encountering new tasks, providing a new solution for complex target detection tasks.

The invention discloses a construction method of an improved pulse recurrent neural network, which comprises the steps of neuron model improvement, connection weight optimization, recurrent structure improvement and training algorithm optimization, and is characterized in that neurons can dynamically adjust thresholds according to own activity states through a self-adaptive threshold adjustment mechanism; the response capability of neurons to different input signals is improved, and excessive or insufficient pulse distribution is avoided. According to the connection weight distribution method based on the attention mechanism, the network can automatically focus on important information connection, the information transmission efficiency is improved, and the key information processing capacity of the network is enhanced. The hierarchical recursive structure is combined with jump connection, so that the capturing capability of the network on the long-time dependency relationship is effectively improved, the gradient problem is avoided, and the stability and convergence speed of model training are improved. The improved stochastic gradient descent algorithm is combined with the adaptive learning rate adjustment strategy, so that the convergence speed of the model is accelerated, the training efficiency is improved, and the model can reach the optimal solution more quickly.

The application is applied to the field of mechanical fault diagnosis signal processing, and particularly discloses a mechanical fault diagnosis method based on STFT dimension transformation, which comprises the following steps: collecting a one-dimensional mechanical vibration signal, and performing wavelet decomposition; performing wavelet reconstruction on low-frequency components and high-frequency components after denoising processing, so as to obtain a one-dimensional vibration signal after denoising; performing short-time Fourier transform, so as to convert the signal into a time-frequency two-dimensional matrix; inputting the time-frequency two-dimensional matrix into an improved HH threshold neuron model, performing Poisson sparse coding on the time-frequency two-dimensional matrix, and only performing pulse response on signal significant features; constructing a threshold coding convolution network with residual connection, inputting a sparse coding matrix, training by using an unsupervised learning rule based on STDP, and adaptively adjusting network synaptic weights; inputting into the trained threshold coding convolution network, and determining a fault diagnosis result by means of pulse firing activities of output layer neurons obtained through network forward propagation.

The invention discloses a lightweight mixed pulse neural network method for hyperspectral image classification. The method comprises the following steps: carrying out preprocessing and block extraction on an input hyperspectral image; constructing a hybrid pulse neural network composed of a pulse standard convolutional layer, two pulse depth separable convolutional blocks and a pulse-driven self-attention mechanism module; using a leakage integration issuing neuron model as a network basic unit; training the network by adopting a proxy gradient method based on a Gaussian function derivative; and accumulating the membrane potentials of the full connection layer of T time steps and taking time average to obtain a final classification result. According to the method, the parameter quantity and the calculation complexity are remarkably reduced through the pulse depth separable convolution, the space-spectrum remote dependency relationship in the hyperspectral image is captured with lower calculation and energy cost through the pulse self-attention mechanism, and the global modeling capability of the network on the space-spectrum characteristics is enhanced; and the method is particularly suitable for resource-limited edge equipment.