118results about How to "Mitigate Vanishing Gradients" patented technology

The invention provides a meteorological monitoring method based on a GRU neural network and differential privacy federated learning. The meteorological monitoring method comprises the following steps: collecting local meteorological data to generate a meteorological time sequence data sequence; selecting a proper model structure according to the heterogeneity of a client, carrying out local meteorological monitoring model training based on a GRU neural network, and carrying out parameter updating on a local model through an LPDP-FedProx federal learning algorithm; the central server performs model aggregation and updating according to a global model weighted average aggregation algorithm after acquiring model parameters of the client; and the client receives the updated global model parameters and carries out the next round of local training until an optimal monitoring model is obtained. According to the method, the personalized differential privacy algorithm and the FedProx federated learning algorithm are combined, the LPDP-FedProx federated learning algorithm is provided, a federated learning framework capable of effectively processing heterogeneous data is constructed, the privacy of local data is effectively protected, and the security of federated learning is improved.

The invention relates to the technical field of artificial intelligence, and discloses a network security situation awareness method based on artificial intelligence, which comprises the following steps: acquiring network security data; performing feature extraction on the network security data by adopting a full-connection neural network based on dynamic group evolution optimization to obtain a first feature; performing feature dimension reduction on the first feature by using a self-encoding neural network based on contrast entropy to obtain a second feature; constructing a classifier model by using a fractional order neural network based on complexity perception, and training the classifier model by using the second feature; and performing network security situation awareness classification by using the trained classifier model. According to the method, the capability of distinguishing the complex network security situation is improved, a more accurate classification effect can be realized, and the applicability of the model in the network security data and the classification accuracy are enhanced.

The invention provides a rolling bearing fault diagnosis system and method based on a multi-source fusion residual network, and relates to the technical field of bearing fault diagnos.The rolling bearing fault diagnosis system and method adopt a one-dimensional convolution kernel and a residual block feature extraction module, the one-dimensional convolution kernel ensures that data of different sensors are not mixed in the feature extraction process, and the residual block feature extraction module is used for extracting the data of different sensors; the effectiveness of unique features of each type of signals is guaranteed, the gradient disappearance problem in a deep network is relieved through residual connection, a feature fusion module of a grouping cross attention mechanism is further adopted, sensor signals are grouped according to types, each group is weighted independently, and the sensor signals are subjected to clustering processing according to types. And local and global multi-scale features are further extracted by using multi-scale convolution, so that the uniqueness of different sensor signals is ensured, the expression ability of fusion features is improved, and the accuracy and robustness of rolling bearing fault diagnosis are improved.

The invention discloses pre-hospital emergency medical aid emergency help integrated equipment and system. The method comprises the following steps: S1, acquiring key vital sign data of a patient by using medical equipment; s2, calling basic information and medical history of the patient through equipment; s3, transmitting the collected data to a server of a remote data processing center in real time; s4, performing comprehensive analysis on the received vital sign data and the medical history information of the patient by applying a pre-trained deep learning model on the server, and generating a prediction report; s5, displaying a prediction result on a portable device or a remote monitoring station in real time by using a data visualization technology; s6, based on a visual prediction result and medical history analysis, guiding first-aid personnel to take targeted emergency medical measures; and S7, continuously monitoring the vital sign response of the patient and any health change related to the medical history, and updating the treatment strategy according to the further analysis of the deep learning model.

The invention discloses a speech enhancement method and system based on a time-frequency graph convolutional network in the technical field of speech signal processing, and the method comprises the steps: obtaining to-be-enhanced noisy speech data, inputting the to-be-enhanced noisy speech data into an improved metric generative adversarial network (CMGAN), and outputting an enhanced speech time domain signal; the improvement method of the metric generative adversarial network CMGAN comprises the following steps: replacing an encoder module of the metric generative adversarial network CMGAN with a time-frequency graph convolution encoder module; in the time-frequency graph convolution encoder module, a time convolution network and a graph convolution network are used for respectively extracting a long-time dependency relationship and global frequency characteristic information of noise-containing voice data to be enhanced, and dense residual error connection is introduced to fuse multi-scale characteristic information at the same time. The speech enhancement performance is improved under the condition of less model parameter quantity and calculation quantity.

The invention relates to the technical field of monitoring data processing, in particular to a dam monitoring data processing method and system. The method comprises the following steps: firstly, obtaining and preprocessing dam historical environment data and dam historical deformation data, and constructing a monitoring data set; secondly, reading the monitoring data set by using a deep FAR-LSTM network, and performing network training; then, reading real-time monitoring data by using a deep FAR-LSTM network, predicting dam deformation parameters, generating a report, and sending the report to a worker for confirmation; and if the dam deformation parameter is abnormal, adding an early warning signal in the report to prompt a worker. And finally, the deep FAR-LSTM network performs real-time updating of the model by using continuously updated monitoring data and deformation data to ensure that a change rule of parameters of an environment where the dam is located is met. According to the invention, the processing efficiency and accuracy of dam monitoring data can be improved.

The invention discloses an underground water level prediction method and system based on a space-time attention gating neural network. The method comprises the following steps: constructing a historical underground water level space-time sequence data set; performing principal component analysis on the space-time sequence data set to obtain a dimension-reduced space-time sequence data set; for the spatio-temporal sequence data set after dimension reduction, constructing a plurality of spatio-temporal characteristic matrixes through a sliding window method; training an underground water level prediction model by using the plurality of spatial-temporal characteristic matrixes to obtain a trained underground water level prediction model; obtaining time-space sequence data of historical underground water level of a to-be-measured position, constructing a time-space characteristic matrix of the underground water level, and predicting the underground water level of the to-be-measured position by using the trained underground water level prediction model; according to the method, rapid and accurate prediction of the underground water level is realized, and the generalization ability, the prediction precision and the interpretability of an underground water level prediction model are greatly improved while the modeling cost is reduced.

The invention discloses a KAN-based photovoltaic power prediction method and system, and the method comprises the following steps: collecting historical data, preprocessing the historical data to form a data set, and dividing the data set into a training set and a test set; constructing a photovoltaic output day-ahead prediction model, and training the photovoltaic output day-ahead prediction model; testing and evaluating the trained photovoltaic output day-ahead prediction model; using the photovoltaic output day-ahead prediction model to predict photovoltaic power; the photovoltaic output day-ahead prediction model comprises a spatial feature extraction module, a time sequence feature extraction module and a spatial-temporal feature fusion and prediction power output module; the spatial feature extraction module extracts variable features at the same time point, learns the relation of the input variable features, and outputs spatial features; the time sequence feature extraction module extracts features between feature variables at different time points, learns a variable feature interaction relationship at different time points, and outputs time sequence features; and finally, the output module fuses the time sequence features and the spatial features and outputs a photovoltaic output prediction result.

The invention discloses an image restoration method based on a multi-scale residual module and feature fusion, and belongs to the technical field of image processing. The multi-scale residual module designed by the invention can effectively capture feature information of different scales of the image, the quality of the repaired image is improved from a fine texture to a macroscopic structure, a feature fusion strategy fusing detail branches and semantic branches can fully integrate features of different levels, the detail expression and overall coordination of the image are enhanced, and the image quality is improved. Through fusion of multi-scale features, limitation possibly brought by single-scale features is avoided, and the repaired image is more real and fine in color, texture and structure. And the multi-scale residual error module can relieve the gradient disappearance problem, the calculation efficiency is high, the image restoration task can be completed in a short time, and the introduced and designed dual-channel global attention module has good robustness, so that the method can stably play a role in various complex image damage conditions.

The invention relates to a DNA methylation detection method and device and electronic equipment, and the method comprises the following steps: obtaining a DNA sequence to be detected; obtaining original electric signal data and a base sequence corresponding to the DNA sequence to be detected through a nanopore sequencing technology; obtaining a methylation detection result corresponding to the DNA sequence to be detected according to the original electric signal data, the base sequence and a deep learning model; wherein the deep learning model comprises a sequence feature extraction module, a signal feature extraction module, a feature fusion module and a classification module; the sequence feature extraction module comprises a Transform encoder; the signal feature extraction module comprises a Conformer encoder; the feature fusion module comprises a Transform decoder; the classification module comprises a full connection layer; according to the method provided by the invention, the identification accuracy of the methylation site can be improved, high-precision methylation detection is realized, and the problems of gradient disappearance and gradient explosion can be effectively relieved.

The invention discloses an improved YOLOv8 chip surface defect detection model construction method and application, and the method comprises the steps: constructing a network structure of an improved YOLOv8 target detection model, which comprises a Backbone module, a Neck module and a Head module; the Backbone module comprises a multi-layer DCNv3 network, a C2f network and an SPPF network, and outputs a first effective feature layer, a second effective feature layer and a third effective feature layer after performing multi-layer convolution processing on the picture; the Neck module adopts a BiFPN bidirectional feature pyramid feature fusion structure, inputs three effective feature layers output by the Backbone module into a ConcatBiFPN network for bidirectional adaptive fusion, and then outputs the three effective feature layers through a CBAM attention mechanism network; the Head module comprises a Deect network structure, and the Deect network structure is connected with a CBAM attention mechanism network and outputs a target detection result; the improved YOLOv8 target detection model structure is used for accurately recognizing the surface defects of the chip, and the adaptability to different types of defects is improved.

The invention discloses a zero-sample image super-resolution reconstruction method and system based on a dense Unet diffusion model. The zero-sample image super-resolution reconstruction method comprises the steps that the trained dense Unet diffusion model is adopted to carry out zero-sample image super-resolution reconstruction; the dense Unet diffusion model is obtained by introducing a dense Unet network structure in a forward diffusion process of a diffusion model without classifier guidance; according to the dense Unet network structure, on the basis of a Unet network structure, two DRB modules are additionally arranged in each layer of encoder and each layer of decoder; each DRB module comprises a time step coding layer, a label coding layer, a dense residual block and a time label direct embedding standardization layer; the dense residual block comprises four convolution layers and a parameter correction linear unit; the feature maps are densely connected; and the time label direct embedding standardization layer is used for firstly embedding a time sequence coding feature from the time step coding layer and then summing the time sequence coding feature with a label coding feature from the label coding layer so as to directly embed information into the dense residual block.

The invention discloses a Chinese text grammar error correction method combining syntactic information and a pre-training language model, and the Chinese grammar error correction capability can be effectively improved by fusing the syntactic information into the pre-training language model Chine BART. The method comprises the following steps: importing a data set and converting the data set into a target format; syntactic information is generated by utilizing HanLP, part-of-speech tags and dependency relationships are converted into vectors, and the vectors are fused with default word embedding; in the encoding stage, a syntax-enhanced multi-head self-attention mechanism and a residual network are used; in the decoding stage, a self-attention mechanism and an encoder-decoder attention mechanism are combined with syntactic information for generation, and a decoder is gradually optimized in combination with a residual network; a plurality of candidate sentences are generated by using Beam Search, and the optimal candidate sentence is selected according to syntactic consistency and semantic rationality. According to the method, the syntactic information and the pre-training language model are combined, so that the model learns semantic information of sentences and pays attention to the syntactic structure, and the Chinese grammar error correction capability of the model can be improved.

The invention discloses a privatized large model construction method and system based on generative artificial intelligence, and relates to the technical field of privatized large models, and the method comprises the steps: obtaining a data source through an enterprise internal system or a financial database, carrying out the data preprocessing, and building a privatized large model. Feedforward network parameters are adjusted and optimized based on an ecological niche dynamic optimization algorithm, and a network training process is optimized based on a self-encoding neural network algorithm of unbalanced dynamic optimization. According to the method, the problem that a traditional optimization algorithm is likely to fall into local optimum in high-dimensional financial data is effectively solved, rapid changes and complex characteristics of data in the financial market can be better dealt with, the robustness of the model in the face of abnormal values, noise and extreme events in the financial data is improved, and the stability and prediction accuracy of the model are enhanced.

The invention discloses a non-invasive blood pressure prediction method based on signal dimension extension and a lightweight network, which comprises the following steps: firstly collecting PPG blood pressure data and ABP data, then denoising the collected PPG blood pressure data and ABP data, and cleaning the denoised data. The processed data is used for data signal dimension expansion, a PPG blood pressure measurement model is constructed, and the PPG blood pressure measurement model comprises a basic convolution layer and a residual grouping convolution module Group ResidualBock; and then deploying the constructed PPG blood pressure measurement model into the wearable device to obtain the wearable blood pressure monitoring device. And finally, measuring the processed PPG signal by using wearable blood pressure monitoring equipment to obtain a blood pressure value. According to the method, the PPG signal dimension is expanded, the data processing link is optimized, and the model blood pressure prediction accuracy is improved; and meanwhile, a lightweight algorithm architecture is adopted, so that resource occupation is reduced, the system can be deployed in wearable embedded equipment, and the security of personal privacy data is guaranteed.

The invention discloses an unmanned aerial vehicle image transmission signal modulation mode identification system and method based on multi-modal fusion. The method comprises the following steps: (S100) collecting and preprocessing unmanned aerial vehicle communication signal data; (S200) converting the one-dimensional time-domain signal into a two-dimensional Gramer angle field graph and estimating a two-dimensional time-frequency graph of the image transmission signal; (S300) carrying out feature merging; (S400) generating a training set and a test set; (S500) constructing a VGG-ResNet hybrid architecture deep learning network and carrying out training; and (S600) identifying the image transmission signal modulation mode of the unmanned aerial vehicle to be identified. The method disclosed by the invention is based on a multi-modal fusion technology and shows remarkable advantages, and data features of different modals can be comprehensively utilized by fusing the one-dimensional time domain signal and the two-dimensional time frequency graph information to realize more comprehensive description of data.

The invention relates to the technical field of medical image analysis, in particular to a training method and system for a cataract recognition model and a storage medium. The method comprises the following steps: acquiring cataract eye image samples and non-cataract eye image samples in different stages; lesion region segmentation and image synchronous simulation enhancement are carried out on the cataract eye image samples and the non-cataract eye image samples in different stages to obtain cataract lesion enhanced image samples and non-cataract enhanced image samples; extracting local detail features of the lesion and extracting structural features of a mask pair, and constructing a multi-task convolutional neural network learning framework to train a cataract recognition model and optimize focus weighted loss; and obtaining a cataract related key feature set and carrying out online incremental learning updating to generate a cataract recognition online incremental updating model. According to the invention, the accuracy and robustness of cataract recognition model training can be improved.

The invention discloses a landslide surface inspection method based on RTDETR, and the method comprises the steps: collecting and obtaining an existing open-source landslide type data set, dividing the existing open-source landslide inspection data set into a training set, a verification set and a test set, and carrying out the specific scene data enhancement and conventional data enhancement operation of the data of the training set; constructing a landslide surface inspection model which is jointly formed by a ResNet-based multi-scale adaptive convolutional feature extraction backbone network, a CNN + Transform-based multi-scale feature fusion layer, and a DETR decoding layer and an adaptive high-order matching cost function-based detection head frame; learning weight parameters by using the training set, further optimizing and adjusting the verification set, and finally testing the landslide surface inspection model by using the test set to determine a final detection result; the accuracy of landslide type detection is greatly improved.

The invention discloses an improved YOLOv7 human body key point detection method. According to the method, manual labeling is carried out through LabelImage to obtain abundant training samples; replacing an original MP module in the YOLOv7 with an L-MP module to improve feature extraction, and adding a LiftDown Pool branch and residual connection; an EDH-Conv module is newly added between a Backbone module and a Head module of the YOLOv7 network, and the output of the EDH-Conv module is used as spatial weight distribution to correct an activated region, so that the detection capability of the network on a small target is further improved; an improved RSwin Transformer encoder is added to improve the detection effect of small targets and overlapped targets in a complex scene; a loss function is changed into an improved key point detection loss based on L2 loss by referring to Wise-IoU LOSS, and a dynamic focusing mechanism used for adjusting a key point error weight is introduced to accelerate convergence of the model. According to the method, the human body key point detection and identification capability in a complex scene is enhanced, and the method has higher robustness, accuracy and efficiency for human body key point detection, identification and counting tasks in sports.

The invention relates to the field of artificial intelligence, in particular to a Crohn disease mucosa healing prediction method and system, and aims to improve the prediction accuracy. The prediction system comprises a preprocessing module and a neural network model. The preprocessing module is used for preprocessing the pre-treatment original ultrasonic image of the to-be-predicted lesion intestinal segment and the corresponding original clinical data; the neural network model comprises a feature extraction module, a first preset number of residual modules, a fusion module and a judgment module. The feature extraction module is used for acquiring a first feature map from the preprocessed ultrasonic image; the first preset number of residual modules are connected in series and are used for extracting a second feature map with deep features from the first feature map; the fusion module fuses the second feature map and the preprocessed clinical data; and the judgment module judges whether the to-be-predicted diseased intestinal segment can achieve mucosa healing according to the fusion features. The attention information is added for the image features by using the clinical data, so that the prediction accuracy can be improved.

The invention relates to the technical field of artificial intelligence, and discloses a time series prediction method based on long and short term features, which comprises the following steps: inputting multi-dimensional time series data with the length of T; the method comprises the following steps: processing input time sequence data through a long and short feature layer in an encoder to generate long-term features and short-term features, and performing feature fusion on the extracted long-term features and short-term features; the long and short feature layer comprises a convolutional neural network and a long and short-term memory neural network, the convolutional neural network is used for extracting multi-dimensional data features of the input data, and the long and short-term memory neural network is used for extracting long-term and short-term features of the input data. By combining the long and short feature layers of the convolutional neural network and the long and short-term memory neural network, the short-term fluctuation and the long-term trend in the time sequence data can be effectively extracted, and the multi-scale feature extraction technology enables the model to better capture the complex time sequence dependency relationship in the data, so that the accuracy of time sequence prediction is remarkably improved.

The invention belongs to the related technical field of intelligent fault diagnosis of equipment, and discloses an equipment fault diagnosis method and system based on a Gramb angle field and a VAE-GAN, and the method comprises the steps: an offline stage: collecting equipment vibration signals, and building an initial signal sample set; a GAF coding technology is used to code the signal sample into a Gramer angle field picture; establishing a sample augmentation model based on the GAN, and training by using the initial picture sample set; acquiring an extended picture sample set by using the trained sample augmentation model; establishing a picture classification model, and training based on the extended picture sample set; an on-line diagnosis stage: encoding a signal sample to be detected into a Gramer angle field picture; and inputting the trained picture classification model to obtain a fault mode classification result. According to the method, the equipment fault vibration signal is converted into the two-dimensional image, the advantages of the GAN in image generation can be fully played, and the problems of few fault samples and unbalanced sample distribution in the prior art are solved.

The present invention relates to the field of artificial intelligence technology, and discloses a network security situation awareness method based on artificial intelligence, including obtaining network security data; using a fully connected neural network based on dynamic population evolution optimization to extract features from network security data to obtain a first feature; using an autoencoder neural network based on contrast entropy to reduce the first feature to obtain a second feature; using a fractional-order neural network based on complexity perception to construct a classifier model, and using the second feature to train the classifier model; and using the trained classifier model to classify network security situation awareness. The present invention improves the ability to distinguish complex network security situations, can achieve more accurate classification effects, and enhances the applicability of the model in network security data and the accuracy of classification.

The invention discloses an SAR intelligent refocusing method based on deep learning. The method mainly solves the problems that in traditional SAR image refocusing, phase error modeling is complex, the image resolution is low, and details are lost. According to the implementation scheme, the method comprises the following steps: acquiring a data set of an SAR pure image and a corresponding defocused image, and dividing the data set into a training set, a verification set and a test set; constructing a deep learning network comprising an encoder and a decoder; using the training set to carry out forward and back propagation iteration to train the network, and evaluating and updating model parameters on the verification set to obtain a trained network model; and inputting a defocused image in a test set into the trained network model, extracting image features through an encoder, and reconstructing a high-resolution image through a decoder to obtain a restored image result. According to the method, the deep features of the SAR image can be effectively extracted, gradient disappearance is relieved, features of different levels are effectively fused, the reconstruction quality and resolution of the SAR image are improved, the detail performance of the restored image is enhanced, and the method can be used for radar imaging and evaluation.

The invention relates to the technical field of power load prediction, in particular to a short-term power load prediction method based on a hybrid cavity convolutional neural network, which comprises the following steps: acquiring original power load data, and constructing a load sequence data set; converting a load sequence in the load sequence data set into a Grubm angle field to obtain preprocessed load input data; constructing a load prediction model based on the hybrid cavity convolutional neural network, and training the load prediction model by using the preprocessed load input data; and collecting power load data in real time, and performing power load prediction by using the trained load prediction model. According to the method, the load sequence is converted into the Grubrum angle field to extract the fine-grained features of the load sequence, the fine-grained features are used as the input of the load prediction model, complex time sequence features can be captured, abnormal changes in the load sequence can be identified, the fine changes of the load in a short time can be effectively reflected, and the accuracy of short-term power load prediction can be improved.

The invention belongs to the technical field of stereoscopic vision, and discloses a stereoscopic matching method based on unsupervised online learning. The method is realized through a stereo matching module based on a multi-level cost body, a non-occlusion region confidence estimation module and a Mann-Kendall trend detection module. Compared with the prior art, the method has a better stereo matching effect. The model provided by the invention has the best result in both the initial stage and the later stage, and due to the good generalization of the model, the model can always keep a relatively accurate prediction result in a sheltered region. The model has a good prediction result in the occlusion area due to high generalization, and the tendency of parameter error update is relieved due to the fact that the self-adaptive adjustment strategy provided by the invention eliminates the influence of error loss of most of the occlusion areas, so that the network can maintain the optimal performance all the time.

The present invention discloses a bearing fault prediction method based on improved VMD-CNN-BiLSTM, introduces SMP chaotic mapping to initialize the distribution of eagle groups, introduces search space for global optimization in the first stage of the northern goshawk algorithm, and introduces Levy flight strategy improvement algorithm in the second stage of the northern goshawk algorithm; uses the improved northern goshawk algorithm to confirm the secondary penalty term in the VMD model; uses the effective weighted index MBVH to screen IMFs, and reconstructs the signal, and divides the reconstructed signal into a training set and a test set; uses a convolutional neural network to extract features, and performs residual connection on the output sequence of CNN and the output sequence of BiLSTM to construct an improved CNN-BiLSTM prediction model; inputs the test set, recombines and superimposes the model prediction values, obtains the output sequence, and completes the bearing fault prediction. Compared with the prior art, the present invention improves the time and accuracy of bearing fault prediction by using the UNGO algorithm to optimize the VMD denoising model, MBVH to screen the signal, and optimizes the CNN-BiLSTM prediction model.

The invention discloses a three-stage self-adaptive signal waveform image recognition method, which comprises the following steps of: acquiring a single-channel signal waveform display screenshot of an oscilloscope as a signal waveform image, preprocessing and marking the signal waveform image to form a required data set, and dividing the data set into a training set and a test set; performing data enhancement on the images in the training set, and inputting the enhanced images and corresponding labels into a signal waveform image classification model for training to obtain an optimal model; intercepting the optimal model, wherein the connection of the optimal model from the initial convolution unit to the second full-connection neural network layer is reserved as a signal waveform image feature extraction model; and after the signal waveform image feature extraction model is obtained, signal waveform image recognition is completed for the test set. According to the invention, the problem of low detection precision caused by the complex conditions of non-uniform line colors and positions of signal waveform display by an oscilloscope, small signal amplitude and the like in the signal waveform identification process is solved.

A blind source separation method for a multi-source fault map of an oil-immersed power transformer comprises the following steps: inputting a multi-source partial discharge fault map into a CNN encoder, adding a ResNet network in a generative network, and introducing a gradient penalty term on the basis of a traditional CGAN network antagonism loss function; a deep learning dedicated server is used for multiple times of training and verification, model training parameters are continuously optimized and adjusted, and a network weight data file meeting oil-immersed power transformer partial discharge fault diagnosis task requirements is obtained; a trained YOLOv8 network is adopted as a diagnosis network, the fault type of each generated single-source atlas is identified, the fault type of a multi-source atlas is determined, and the fault state of the oil-immersed power transformer is deduced and diagnosed; and deploying the obtained Encoder-CGAN network and YOLOv8 network weight data files to power inspection equipment and a power inspection platform to complete fault monitoring and deduction diagnosis tasks of the oil-immersed power transformer. And the problem of deducing and diagnosing the fault type of the partial discharge fault map of the oil-immersed power transformer is solved.

The invention discloses an adaptive gradient-based spiking neural network training method, which comprises the following steps of: firstly, constructing a to-be-trained spiking neural network adopting LIF model neurons, and inputting data set samples into the to-be-trained spiking neural network in batches; then obtaining pre-synaptic inputs of all neurons in each layer and membrane potentials of all neurons in each layer in the spiking neural network to be trained, associating the width of the proxy gradient with statistical characteristics of the membrane potentials through a correlation function, and adaptively adjusting the width of the proxy gradient in each layer; and finally, calculating cross entropy loss of an output layer of the pulse neural network and a sample label, performing gradient back propagation, and updating a synaptic weight. According to the method, the problem of gradient disappearance or mismatching caused by fixed proxy gradient function width is effectively relieved, the energy consumed by single reasoning of the network is less, and the trained network has excellent classification performance.