212 results about "Softmax function" patented technology

The invention discloses a rotating machine fault diagnosis method based on a multi-view space-time diagram attention network, and belongs to the technical field of rotating machine intelligent monitoring and fault diagnosis. A multi-view space-time diagram attention network sequentially comprises a topological graph construction layer, a two-channel parallel diagram attention network, a multi-view feature fusion module, a gating recursive unit time feature extraction module, a full connection layer and a Softmax function layer, and during training, firstly, a distance topological graph and a similarity topological graph of input data are constructed; through a two-channel parallel graph attention network, spatial feature extraction is carried out from two perspectives of geometric distance and feature similarity; through a multi-view feature fusion module, dynamic fusion of dual-channel features is realized through a learnable attention weight matrix; and finally, the fused spatial features are input into the gated loop unit network, fault feature information in the time dimension is deeply mined, and the accuracy and robustness of fault diagnosis of the rotating machine can be effectively improved.

The invention discloses a strip steel surface defect detection and classification method based on a graph neural network, and the method comprises the steps: segmenting a strip steel surface image into different scale sub-blocks, and extracting the node features of each scale sub-block; constructing graph structures with different scales according to the similarity of the node features; constructing a stacked multi-scale image attention network model, taking the image structures of different scales as input for training, inputting the image structures of different scales into the trained stacked multi-scale image attention network model, performing feature extraction on nodes in the image structures of different scales, and performing feature extraction on the nodes of the image structures of different scales; aggregating and updating node features in the different scale graph structures through a graph attention mechanism, and obtaining a feature matrix according to the updated node features in the different scale graph structures; and paving the obtained feature matrix into a one-dimensional vector, inputting the one-dimensional vector into a full connection layer, carrying out nonlinear operation, outputting the probability of each category through a softmax function, and taking the category with the maximum output probability as the category of the strip steel surface defects.

The invention discloses a reservoir level prediction method based on a neural network and uncertainty perception and a medium. The method comprises the following steps: performing multi-scale characteristic decomposition on multivariable hydrological time series data; then, the macroscopic state and the microcosmic anomaly degree of system operation are subjected to combined recognition in parallel; then, constructing a hybrid expert (MoE) neural network architecture comprising a plurality of expert sub-networks and a gating network; a dynamic weight distribution mechanism of uncertainty perception is designed, and a temperature parameter of a Softmax function in a gated network is regulated and controlled in real time by using previously identified abnormality, so that a more conservative decision is forcibly made when a system enters an unknown mode; and finally, performing weighted fusion on a model prediction result, and performing end-to-end optimization on the whole system. According to the method, the problem that the generalization ability of a traditional model is insufficient when the traditional model faces a data distribution external mode is effectively solved, and the prediction robustness and precision are remarkably improved.

The invention relates to a garbage classification multi-stage verification method based on a dynamic confidence threshold. The method comprises the following steps: acquiring image information to be identified; performing image preprocessing on the to-be-identified image information; performing garbage classification identification on the processed to-be-identified image through the trained classification model, and outputting an identification result; after an identification result output by the classification model is converted into probability distribution through a Softmax function, the identified garbage type confidence coefficient and the probability thereof are obtained; judging whether the highest confidence degree in the obtained garbage types exceeds a preset threshold value or not; if so, outputting the garbage type corresponding to the highest confidence coefficient; and if not, outputting a preset number of garbage types with the confidence ranked before and the probability of the garbage types. The result is directly output when the confidence coefficient is high, the user experience is improved, when the confidence coefficient is low, the preset number of garbage types ranked before and the probability of the garbage types are displayed, and the misjudgment risk is avoided.

The invention discloses a mechanical equipment fault intelligent diagnosis method and system based on a bidirectional time sequence convolutional network and an attention mechanism, and the method comprises the steps: collecting a vibration signal of mechanical equipment, and carrying out the preprocessing of the vibration signal through employing a variational mode decomposition algorithm optimized by an improved Hemma optimization algorithm; training a mechanical equipment fault diagnosis model by adopting the sample data set with the label to obtain a trained mechanical equipment fault diagnosis model; the model comprises a bidirectional time sequence convolution module fusing multi-head self-attention, a feature fusion module, a channel attention module, a global average pooling module, a full connection module and a Softmax function which are connected in sequence. And inputting the preprocessed vibration signals into the trained mechanical equipment fault diagnosis model to obtain mechanical equipment fault category probability distribution. According to the invention, noise reduction, feature enhancement and equipment fault mode accurate identification of the vibration signal under a high-noise background can be realized.

The invention discloses a metallurgy multi-modal image classification method based on adaptive marginal learning and cross-modal enhancement, and the method comprises the steps: collecting a splash image data set, and carrying out the classification and marking; the category label is expanded into a BOF steelmaking process description text; preprocessing the multi-modal data according to small sample learning requirements; the ResNet50 is utilized to extract image features; a semantic feature representation is extracted by using a Transform; text features and image features are enhanced, and a cross-modal attention mechanism is introduced to improve the semantic discrimination capability of visual representation; constructing a self-adaptive marginal generator, and dynamically adjusting a classification boundary; an adaptive marginal loss function is adopted to replace a traditional cross entropy loss function; calculating an image-text similarity score, converting the score into probability distribution by adopting a softmax function, and then converting the probability distribution into a prediction category; performing training optimization and performance evaluation on the model; and carrying out online classification on the spattering images. The method can realize accurate and intelligent identification and classification of the splashing state of the converter, and is suitable for small sample industrial scenes.

The invention provides an image data processing method and system based on a lightweight sparse neural network, and relates to the technical field of machine learning, and the method comprises the steps: obtaining to-be-processed image data; performing data enhancement and tensor standardization preprocessing of random flipping and random cutting on the image data in sequence to obtain standard tensor data meeting model input requirements; inputting the standard tensor data into the lightweight sparse neural network model; according to the model, local edge and texture features and global semantic features are extracted step by step from an image through a plurality of sparse convolution layers; and mapping the extracted physical features into category confidence through a full connection layer, and finally outputting a classification result through a Softmax function. Through a gradient-guided differential variation strategy and GPU parallel optimization, efficient image processing is realized on a mobile terminal, an embedded device and an edge computing device with limited resources, computing overhead and memory occupation are reduced, and meanwhile, high-precision classification performance is kept.

The invention relates to the technical field of natural language processing, in particular to a foreign trade letter intelligent semantic error correction and business term optimization method and device, and the method comprises the steps: obtaining a target foreign trade letter, and constructing a target corpus; performing Chinese word segmentation and part-of-speech tagging on the target foreign trade letter; performing term optimization based on the word segmentation result and the knowledge graph; identifying the letter title by using a conditional random field model, and converting the letter title into structured data; a Bi-LSTM-CRF model is adopted to carry out risk point detection, including Bi-LSTM coding, feature engineering and Max-pooling technologies, a part-of-speech sequence is obtained through a softmax function and a Viterbi path, and sequence labeling is carried out to obtain a risk point detection result; and finally, performing Chinese error correction based on the word segmentation result after part-of-speech tagging and the knowledge graph. The recognition and correction accuracy of foreign trade terminologies is improved, and communication obstacles caused by nonstandard use of the terminologies are effectively reduced.

The invention belongs to the technical field of artificial intelligence, and discloses a large model self-iterative optimization method and system based on feature distribution difference feedback, and the method comprises the steps: collecting a human-generated text data set and a model-generated text data set, and carrying out the preprocessing of the data sets; extracting features of multiple dimensions from the data set, and calculating a difference vector; based on the difference vector, using a Softmax function to distribute an optimization weight for each feature dimension to obtain a weight vector; based on the weight vector, using a joint loss function to perform fine tuning on the large model; and performing iterative fine tuning until a preset convergence condition is met. According to the method, the naturalness of the generated text and the human similarity are remarkably improved, automatic and low-cost continuous optimization is realized, a refined and explainable optimization direction is provided, and the method has good universality and expandability. According to the method, an automatic iteration closed loop integrating analysis and optimization is constructed, the problem of a'model cavity 'is solved, and quantifiable alignment of microcosmic language features is achieved.

The invention discloses an automobile production line parameter monitoring and early warning method and system based on dynamic threshold and multi-dimensional comparison. The method comprises the following steps: completing deployment of a sensor device for data acquisition and construction of a data set; constructing a model infrastructure, wherein an input layer synchronously receives a process card parameter vector, an equipment state vector and a process-equipment association feature vector; a time sequence segmentation learning mechanism is introduced into the LSTM layer, and data fragments are divided according to the equipment operation cycle; the model processing layer adopts a dynamic weight adjustment mechanism, and the Attention layer firstly endows features at different moments and in different equipment states with basic weights, and then carries out real-time calibration according to the abnormal contribution degree of each feature in historical early warning data; the output layer calculates and outputs the probability value of the actual parameter meeting the process requirement through a Softmax function; and taking a probability value output by the model as a core, establishing a hierarchical early warning response mechanism and matching a standardized process. And long-term accurate monitoring and efficient early warning of the parameters of the automobile production line are realized.

The invention relates to the technical field of network acceleration, and discloses an edge-end large language model reasoning acceleration method and accelerator, and the method comprises the following steps: reconstructing a calculation process of a decoding stage, and carrying out the deep fusion of a multi-head attention mechanism and the calculation operation of a feedforward network; the weight and key value data are stored in HBM, and the coefficient and the accumulated attention score are stored in DDR; for linear matrix calculation, a unified matrix calculation unit is used for executing multi-precision matrix operation; for nonlinear function calculation, a mathematical transformation and linear fitting method is adopted, a Softmax function is converted into operation with 2 as the bottom through a bottom conversion formula, and truncation and third-order linear fitting are conducted on a Sigmoid function; and constructing a key value screening algorithm based on the accumulated attention score, dynamically adjusting a key value storage position, maintaining a recent key value cache region and an important key value cache region in a limited cache space, and realizing key value efficient cache in long text reasoning.

Mechanisms are provided for optimizing a deep learning (DL) computer model for homomorphic encryption (HE) workload processing. The mechanisms receive an original DL computer model architecture that is to be optimized for HE workload processing, and modifying the original DL computer model architecture by replacing a self-attention layer of the original DL computer model with an HE friendly self-attention layer that comprises a Power SoftMax function that does not have exponent terms, to thereby generate a modified DL computer model architecture. The mechanisms execute a machine learning training of the modified DL computer model architecture, approximate one or more elements of the Power SoftMax function with polynomials to generate a trained HE optimized DL computer model, and output the trained HE optimized DL computer model for execution on HE workloads.

The invention discloses an Alzheimer disease prediction method and system based on deep learning and electrocardiosignals. Firstly, electrocardiosignals are collected and preprocessed; then, constructing a deep learning prediction model which comprises a feature extraction module and a classification prediction module; the feature extraction module comprises a plurality of residual blocks, and each residual block comprises a convolution layer, a batch normalization layer, a nonlinear activation layer and a jump connection; the classification prediction module realizes a complete mapping process from a feature space to probability prediction through a multi-layer perceptron structure and a softmax function, and obtains a prediction probability of each electrocardiosignal fragment belonging to each category; and fusing the prediction results of the plurality of electrocardiosignal segments of the same sample by adopting a soft voting mechanism to obtain a final prediction result. The system comprises a signal acquisition module, a signal preprocessing module, a deep learning prediction module and a result output module. According to the method, Alzheimer's disease prediction is realized based on the electrocardiosignals, and the method is simple and convenient in acquisition mode, non-invasive, low in cost and suitable for large-scale screening.

The invention discloses a hardware-based SOFTMAX function calculation method and device. The hardware-based SOFTMAX function calculation method comprises the following steps: receiving a multi-dimensional input vector corresponding to an SOFTMAX function through an input port; uniformly dividing the multi-dimensional input vector into a plurality of data segments according to an element sequence through a channel distributor, and distributing the data segments to corresponding operation channels; performing an exponential operation on the data segment using the operation channel to generate a local exponential vector; aggregating the local exponent vectors with an accumulator to generate a global exponent sum; distributing the global index sum to each operation channel, so that the operation channels generate output sub-vectors according to the local index vectors and the global index sum; and splicing the output sub-vectors according to the element sequence through a recombination unit to generate an SOFTMAX output vector. The efficient SOFTMAX function hardware acceleration calculation under the multi-channel parallel pipeline processing architecture is realized, and the calculation performance and throughput are improved by several times compared with the traditional serial software implementation.

The invention discloses an image small target detection method based on a fusion feature visual model, and relates to the technical field of computer visual detection. And the fusion feature visual detection model performs the following processing on an input image and outputs a detection result: adopting adaptive Daubechies wavelet transform to enhance high-frequency details, combining multi-scale feature fusion with a double attention mechanism to enhance small target feature expression, utilizing regression calculation to predict bounding box coordinates, and realizing category probability distribution prediction based on a Softmax function. A multi-task loss function and an Adam optimizer are adopted for end-to-end training, and the model performance is improved by balancing classification and regression loss. The method effectively solves the problems that small target features are deficient and are susceptible to background interference, significantly improves the detection precision while maintaining the calculation efficiency, and is especially suitable for unmanned aerial vehicle patrol and other practical application scenes.

The application provides a cloud platform computing resource allocation method, system, terminal and storage medium, comprising: collecting idle resource rates of each host in a cloud platform cluster; converting each idle resource rate of each host into a corresponding recommended reference value by using a normalization exponential function; calculating the product of each recommended reference value of each host respectively, and accumulating the product of all hosts as a recommended coefficient; calculating a host idle degree value according to the recommended coefficient and the product of each recommended reference value of the host; and allocating a corresponding host to a virtual machine creation request according to the principle of preferentially allocating the highest idle degree value. The application converts data into relative recommended probability by analyzing the idle resource conditions of all hosts in the cluster through a softmax function, and then obtains the recommended value of the host by using DS evidence theory for information fusion, so as to allocate appropriate hosts for user requests. The application can reasonably allocate resources when planning multiple user creation of virtual machines, thereby improving the efficiency of the system.

This application relates to a method, apparatus, device, and medium for few-sample SAR target recognition. The method includes: dividing SAR samples into support and query sets based on a meta-learning framework; converting the original SAR image into a semantic map and an attribute scattering center topology map; extracting three types of features and calculating class prototypes through three parallel feature branches; obtaining the confidence score of each branch using the Softmax function; assigning teacher and student branches according to the confidence score; minimizing KL divergence to achieve dynamic knowledge transfer; weighted fusion of predicted log odds followed by Softmax output; and backpropagation to optimize the network. This method, through multi-prototype fusion and interactive distillation, mitigates the problems of speckle noise and high intra-class variance in SAR images, maintaining high robustness and generalization ability even in extreme data-scarce scenarios.

The invention relates to the technical field of wharf AGV scheduling, in particular to an AGV ship loading and unloading task scheduling method, system and device based on DDPG and a medium, and the method comprises the steps: collecting wharf scheduling data in real time; preprocessing the wharf scheduling data; encoding the preprocessed wharf scheduling data into a state vector; inputting the state vector into a trained scheduling decision model, and outputting a scheduling strategy action of each AGV; and performing Softmax function normalization on the task selection probabilities of all AGVs based on the probability of selecting a ship loading task and the probability of selecting a ship unloading task in a scheduling strategy action, generating probability distribution of each AGV for candidate tasks, generating a final task allocation instruction in combination with an epsilon-greedy strategy, and issuing the final task allocation instruction to each AGV for execution. According to the method, the task completion efficiency can be remarkably improved, the energy consumption of the AGVs is reduced, and the high-priority task response capability and the scheduling stability in a multi-AGV conflict scene are enhanced.

The invention provides a power grid fault positioning method, device and equipment based on multi-source data fusion, and relates to the technical field of data processing. The method comprises the following steps: collecting multi-source data in the operation process of power grid equipment in real time, and carrying out space-time alignment processing on the multi-source data; wherein the multi-source data comprises electrical measurement data, meteorological data, SCADA (Supervisory Control And Data Acquisition) data and GIS (Geographic Information System) topological data; inputting the multi-source data subjected to space-time alignment processing into a double-flow deep learning model, respectively extracting spatial features and time features, and fusing the spatial features and the time features to form a high-order fusion feature vector; wherein the double-flow deep learning model comprises a space sub-network and a time sub-network; and inputting the fusion feature vector into a full connection layer, outputting the fault probability of each power grid node through a Softmax function, and superposing node coordinates of which the fault probabilities exceed a threshold value to a GIS map to generate a fault thermodynamic diagram. According to the method, complex time-space correlation information in the multi-source data can be fully mined, and the operation state of the power grid can be analyzed more accurately.

The application provides a traffic target detection and recognition method based on sound-vibration time-frequency characteristics and cross-attention fusion mechanism. First, sound-vibration signals of different traffic targets are collected by a sound-vibration sensor, and normalized least mean square (NLMS) noise filtering preprocessing is performed. Second, the preprocessed sound-vibration signals are subjected to variational mode decomposition (VMD), and scale spectrum segmentation method and summation fuzzy entropy minimum value method are adopted to decompose the sound-vibration signals into multiple intrinsic mode functions (IMFs). Third, Mel spectrograms are extracted from the sound signal IMFs, and wavelet transform time-frequency diagrams are extracted from the vibration signal IMFs, and the results are subjected to CNN convolution pooling, and transformer encoder is further used to extract sound-vibration signal characteristics of different traffic targets. Finally, cross-attention mechanism is used for coding, sound signal characteristics and vibration signal characteristics are fused into new characteristics, and Softmax function and Dropout function are used for normalization and overfitting prevention. The application has the advantages of low algorithm complexity, strong real-time performance and low cost, and solves the traffic target detection problem under extreme climate, weather, light and other scenes.

A computer-implemented method for performing a code generation for calculating a Softmax function of a neural network includes (i) providing a displacement s and a multiplier for the quantized representation of the input tensors of the Softmax function of the neural network, (ii) creating a second lookup table to replace a nested function to calculate the EXP_ON_NEG (MUL_SAT ( )) function from a CMSIS NN library depending on the displacement and the multiplier, wherein element values of an input tensor normalized to a negative value range between 0 and a minimum value are used as arguments, wherein 0 is assigned to a maximum possible element value of the input tensor and the minimum value is assigned to the smallest possible element value of the input tensor, and (iii) implementing an access to the second lookup table in a code generated to calculate the Softmax function, so that it replaces the function call of the EXP_ON_NEG (MUL_SAT ( )) function.

The invention provides a traditional Chinese medicine efficacy-efficacy intelligent semantic alignment and reasoning method and device, and relates to the technical field of natural language processing. The method comprises the steps of obtaining a Chinese herbal medicine efficacy text and a corresponding Chinese herbal medicine efficacy text, performing segmentation to obtain efficacy sub-word units and efficacy sub-word units, and inputting the efficacy sub-word units and the efficacy sub-word units into a fine-tuned BERT model to obtain word vectors; inputting into a double-layer bidirectional long-short-term memory network to obtain efficacy multi-granularity characteristics and efficacy multi-granularity characteristics; and inputting the Chinese herbal medicine efficacy text to a cross attention layer, dynamically distributing matching weights between the efficacy and the efficacy, and classifying the matching weights through a Softmax function to obtain a predicted Chinese herbal medicine efficacy text. Through fusion of the deep learning model, accurate conversion between the efficacy and the efficacy is effectively realized, and a brand new technical support is provided for efficacy analysis in the field of traditional Chinese medicines.

The invention provides a historic building damage prediction method, a model, equipment, a medium and a program. The method comprises the following steps: acquiring multi-source time sequence data of key components of a historic building; preprocessing the multi-source time sequence data; extracting features from the Raman spectrum data by adopting three-layer 3 * 3 convolution, extracting features from the infrared spectrum data by adopting multi-scale convolution, and obtaining spectral features by self-adaptive pooling unified length; converting the structure attribute data and the environment data into low-dimensional features matched with spectral feature dimensions through a two-layer full-connection network; splicing the spectral features and the low-dimensional features, unifying the number of channels through 1 * 1 convolution, screening key features through a gating module, and outputting a global feature vector Y1; inputting the global feature vector Y1 into a three-layer full-connection network, and outputting an original output vector Y2 of the current damage level; and calculating the confidence of each grade of building damage through a Softmax function. The method does not need to contact key components, and secondary damage caused by damage prediction is avoided.

The application relates to a dual-branch hyperspectral image classification method based on a graph convolutional neural network and an attention mechanism, which comprises the following steps: inputting a hyperspectral original image to data for pretreatment; inputting the data after dimension reduction into a CNN attention mechanism branch, and obtaining processed branch features through an attention mechanism module, a data processing module, an attention mechanism module and convolution operation; inputting the data after dimension reduction into a GCN branch, performing simple linear iterative clustering, segmenting into k superpixels, encoding the k superpixels, obtaining superpixel graph nodes, inputting the graph nodes into a GCN module for feature extraction, decoding the extracted superpixel node features into pixel-level branch features; performing feature fusion on the two branches, using a cross-entropy function as a loss function to train the model, and using a softmax function to obtain a class label probability. The application takes limited hyperspectral images as research objects, improves high classification precision, and ensures classification speed.

Techniques for performing secure computing of softmax functions at high speed and with high accuracy are provided. A secure softmax function calculation system that calculates a share ([[softmax (u1)]], . . . , [[softmax (uJ)]]) from a share ([[u1]], . . . , [[uJ]]) includes a subtraction means for calculating a share ([[u1−u1]], [[u2−u1]], . . . , [[uJ−uJ]]), a first secure batch mapping calculation means for calculating, [[exp (u1−u1)]], [[exp (u2−u1)]], . . . , [[exp (uJ−uJ)]], an addition means for calculating a share([[∑ j=1J⁢exp⁡(uj-u1)]],… ,[[∑ j=1J⁢exp⁡(uj-uJ)]],and a second secure batch mapping calculation means for calculating a share ([[softmax (u1)], . . . , [[softmax (uJ)]]).

The invention discloses a graph reasoning-based judgment document event joint extraction method. According to the method, a lexical graph inference mechanism and a trigger word centralization decoding architecture are fused, the lexical graph inference mechanism strengthens the relevance between trigger words and argument roles and solves the problem of semantic segmentation of entities and events, and the trigger word centralization decoding architecture avoids multi-stage cascade errors and improves the complex logic chain processing capacity. The method comprises the following steps: constructing a joint event extraction model: performing corpus labeling and BIO labeling on a data set, and constructing an event type-role label set; processing the text subjected to BIO labeling into numeric vector representation capable of being input into a model, and constructing a lexical element pair label matrix; the hidden lexical elements output by the text embedding layer are classified through a classification layer, a softmax function is adopted to predict relation marks between lexical element pairs, label distribution corresponding to the lexical elements is generated, and a basis is provided for subsequent event extraction; and establishing an event center graph according to lexical tag distribution, performing graph structure decoding to obtain an event prediction result, and training a model.

This invention discloses a method for accurate audio warning identification based on hybrid features. This method analyzes collected audio speech to determine the issuance time of the warning signal, thereby accurately evaluating the timeliness of the audio warning. First, a double noise reduction method using logmmse-spectral subtraction is employed to filter out noise information in the recorded speech. Next, endpoint detection based on short-time energy is used to mark the effective speech segments in the test speech. Then, MFCC features and waveform polynomial features are extracted from each frame of the effective speech segment. Subsequently, the two features are used as inputs to two channels of a convolutional neural network, and the outputs of the two channels are summed to obtain the hybrid features. Finally, the hybrid features are used as input to a softmax function, and the speech segment containing the target speech (audio warning signal) is determined by the maximum probability value. The starting position of this speech segment is the issuance time of the warning signal.

The application discloses a remote sensing image change detection and change description method based on multi-task learning, obtains shared features of double-time remote sensing images through a SegformerB1 network based on weight sharing, introduces a deformable attention mechanism to focus on key areas, introduces a dynamic convolution mechanism to perform multi-scale convolution operation and feature splicing on the shared features, obtains general features capable of comprehensively explaining change information of the remote sensing images, performs multi-task learning on a change detection task and a change description task by using the general features, and outputs remote sensing image change positioning and change description results through a Softmax function. The method constructs general features of the change detection and the change description, improves expression capability and robustness of the features, performs multi-task learning on the change detection task and the change description task, improves collaborative efficiency between the tasks, and improves accuracy and generalization of the remote sensing image change detection and change description model.

The invention discloses a large language model softmax function hardware acceleration circuit and method, and belongs to the field of neural network hardware acceleration of super-large scale integrated circuits. An input sequence is divided into a plurality of data blocks which are processed in parallel, and three-stage pipeline division is adopted, so that average single calculation delay is shortened to G clock cycles, the calculation parallelism is improved, the calculation speed of a softmax function is improved, and the reasoning delay is reduced; and a sparse mask strategy of sparse threshold comparison is introduced, the sparsity of data is fully utilized, the problems of high calculation complexity and high calculation delay of the softmax function and the data access bottleneck of the softmax function are solved, the calculation cost is remarkably reduced, and the calculation efficiency is improved. In addition, a softmax function hardware circuit adaptive to software optimization is constructed, calculation delay and memory access pressure are reduced through the hardware circuit, and data processing efficiency is improved.

The invention discloses an optical fiber distributed voiceprint recognition method based on multi-scale decomposition and hybrid recombination, and belongs to the technical field of distributed optical fiber sensing and artificial intelligence, and the method comprises the following steps: constructing a DAS system, and employing the DAS system to demodulate and restore the phase and intensity signal of backward Rayleigh scattering light to obtain a voiceprint signal; converting the voiceprint signal into basic voiceprint feature representation by using a multi-scale decomposition and hybrid recombination network; performing multi-scale decomposition on the basic voiceprint feature representation, and extracting multi-scale long-term trend features and short-term detail information; performing feature extraction and fusion on the multi-scale short-term detail information to obtain multi-scale short-term detail features; carrying out adaptive weighting on the multi-scale short-term detail features, and carrying out hybrid recombination on the multi-scale short-term detail features and the long-term trend features to obtain hybrid recombination features; and outputting an identification result through a plurality of layers of linear mapping and a Softmax function on the hybrid recombined features. By adopting the method, the accuracy and the stability of voiceprint recognition are improved.