202 results about "Network classification" patented technology

Approaches are generally described for corrupted speech detection in voice-based computer interfaces. First input data including first audio data representing a user utterance may be received. First data representing the first audio data may be generated using a first encoder. First text data representing a transcription of the user utterance may be generated. Second data representing the first text data may be generated using a second encoder different from the first encoder. Third data may be generated by combining the first data and the second data. The third data may be sent to a classifier network trained to predict a relevant corruption state for speech processing inputs. The classifier network may determine that the first input data corresponds to a first corruption state.

The invention discloses a traffic accident detection method based on FFC and GCSA models. The method is innovatively improved based on a YOLOv8 network model. Firstly, a multi-scale feature fusion module FFC is designed in Backbone to replace an original C2f module, and the capability of fusing the multi-scale features of the network can be further improved on the premise of not increasing the network calculation amount, so that the network can fuse the multi-scale features on the level of finer granularity; then, a GCSA attention mechanism is introduced between a C2f module and a Deect module of a Neck layer, and the recognition capability of the model for key features in a complex environment is remarkably enhanced; and finally, establishing a loss function Focal-EIoU Loss for the improved YOLOv8 network structure, so that the network classification detection capability is improved, and the generalization capability of the model is improved. In a specific implementation process, a traffic accident image data set covering multiple scenes is constructed, and specialized data preprocessing is performed; then, end-to-end training and parameter optimization are carried out on the Traffic-YOLO network model obtained after improvement; and finally, integrating the optimized model to a traffic accident detection system for real-time target detection. Compared with the prior art, the method effectively improves the accuracy and robustness of traffic accident detection in a complex scene, and has important practical significance.

The invention discloses an efficient intelligent network attack classification tool based on a federated learning framework, which comprises a central server and a plurality of end-side nodes, local models are deployed in each end-side node, a central model is deployed in the central server, and the models comprise multi-task logistic regression, random forest, XGBoost and long and short-term memory neural network classifiers; the local model collects local network flow data, performs model training and abnormal data detection and classification in combination with a D-S evidence theory optimization algorithm and a four-class complementary classifier, and transmits encrypted abnormal data and related key update parameters to the central server; the central model receives and aggregates the updated parameters, performs attack type classification based on abnormal data in combination with a D-S evidence theory optimization algorithm after updating the global model, and re-allocates a classification result and the updated global model to each end side node; the method has high robustness and high efficiency, the adaptability is enhanced, and the recognition capability of attack scenes is improved.

The invention discloses a cross-domain continuous learning image classification method based on a dynamic knowledge extension network, and relates to the field of image classification, and the method comprises the steps: carrying out the training of a cross-domain continuous learning method based on the dynamic knowledge extension network, obtaining a dynamic knowledge extension network classification model, and carrying out the classification of a target image; the cross-domain continuous learning method based on the dynamic knowledge extension network comprises the following steps: initializing static and dynamic backbone networks, and loading a training task and a cross-domain training data set; according to the knowledge correlation between the new task and the existing task, determining whether to expand the network architecture or utilize the existing architecture for learning, and updating and activating the expert network; utilizing a memory sample to optimize a dynamic backbone, and calculating a clustering prototype; after task learning is completed, evaluating the network; and the trained network parameters are applied to the dynamic knowledge extension network. The problem of disastrous forgetting in cross-domain continuous learning can be effectively solved, and the generalization ability and learning efficiency of the model are improved.

The invention discloses an expandable category guide anomaly detection method and device for multiple categories of targets, and relates to the field of computer vision. The method comprises the following steps: carrying out abnormal region guided adaptive enhancement preprocessing on an industrial image to obtain a target image; judging the category of the target image according to a pre-constructed lightweight neural network classifier; activating at least one anomaly detection model according to the category of the target image; and performing anomaly prediction on the target image by using the activated anomaly detection model, and fusing with the confidence corresponding to the anomaly detection model to realize adaptive anomaly discrimination of the industrial image. According to the method, the sample category is quickly judged through the lightweight classifier, the pre-screening and path guidance of the anomaly detection model are realized, the category guidance weight is generated by using a confidence coefficient mechanism, the subsequent model fusion strategy is endowed with higher adaptability, and the structural clarity, the model selection accuracy and the overall calculation efficiency of the system are effectively improved.

The invention relates to the field of die forging press anomaly detection, in particular to a large die forging press anomaly detection method, which comprises the steps of collecting multi-dimensional time sequence training data of a die forging press, aligning timestamps and processing missing values; dividing the training data into a plurality of segments through a sliding window, and dividing the operation stages of the die forging press into five classes; performing tagging processing on the data fragments, calculating a Pearson's correlation coefficient matrix, and generating a corresponding correlation threshold matrix; training a neural network classification model based on the training data; collecting detection data of the die forging press in real time, and judging the current operation stage; selecting, calculating and comparing a correlation threshold matrix and a real-time correlation coefficient matrix according to the current operation stage; and if the real-time correlation coefficient is lower than the correlation threshold matrix coefficient, abnormal early warning is triggered. According to the method, anomaly detection is carried out on the sensor correlation in different stages, so that the accuracy of anomaly detection of a correlation algorithm is improved, and the false alarm rate and the missing report rate are reduced.

The invention provides a lightweight Internet of Things traffic classification method based on improved MobileNetV2. A lightweight model based on MobileNetV2 is constructed and comprises an input layer, a network introducing an attention mechanism, a classification network and an output layer which are connected in sequence; the input layer performs flow splitting and anonymization processing on input network flow data and constructs a multi-dimensional feature representation matrix comprising a byte level, a data packet level and a data flow level; according to the introduction of the attention mechanism network, an attention mechanism module is added behind a plurality of bottleneck layers so as to combine current layer features with shallow high-resolution features; performing iterative training on the lightweight model by taking the training set as training data and adopting a combined strategy based on cosine annealing and hot restart to obtain a trained model; model parameters are optimized through quantitative compression and structure pruning, the test set serves as test data, the optimized model is evaluated, and a classification result is output. According to the invention, efficient and accurate traffic classification is realized, and the problem of resource limitation of the Internet of Things equipment is solved at the same time.

The invention provides an intelligent early warning method and system for physiological fatigue of a construction equipment operator, and the method comprises the steps: synchronously collecting the limb movement and equipment trajectory data of the operator through an inertial sensor and a construction equipment control terminal, carrying out the parallel prediction of an operation sequence through a two-channel time convolution network TCN, and recognizing a redundant and fatigue movement mode; dynamic structure features of the construction equipment track are extracted in combination with an improved ConvNeXt network, and the abnormal level of the track is judged based on a graph neural network classifier; a soft behavior-reviewer algorithm is introduced, the physiological fatigue index, the trajectory deviation rate and the trajectory control fluctuation degree are used as state input, and optimization operation suggestions are dynamically output; a diagnosis result is transmitted to an operator in real time through a voice and tactile feedback mechanism, and a sustainable closed-loop regulation and control system is formed. According to the invention, the fatigue identification accuracy and the intelligent intervention capability of the operator in the construction process are obviously improved, and the construction safety and the operation efficiency are improved.

The invention discloses a network traffic large model construction method based on heterogeneous federated learning, which comprises the following steps: analyzing, cleaning and formatting network traffic data, and constructing question and answer pairs; performing semantic equivalence reappearing on the public data set by using a large language model to generate an enhanced data set; performing architecture optimization on the generative basic model to replace an original output layer with a customizable network classification head; decoupling the teacher model into a simulator and an adapter; initializing a LoRA module for a simulator and an adapter of the student model; the client adaptively adjusts the LoRA rank according to local resources; the client only finely adjusts parameters of the LoRA module of the adapter and freezes other parameters; the server adopts a weighted stacking algorithm to aggregate the LoRA modules uploaded by the client; the method has good universality and expandability.

The invention discloses an unknown similar special-shaped target radar HRRP classification method based on feature decoupling. A feature decoupling network comprises a CNN network, a mixed attention layer, a common feature extraction branch, a private feature extraction branch and a reconstruction network. The former branch comprises a common feature encoder and a prototype network classifier, and the latter branch comprises private feature encoders of various models. According to the method, a training set is constructed by using HRRP echo samples of various types of targets with multiple models, a prototype network classification framework is adopted, similar training samples are pulled to the same type of prototypes, a deep neural network is promoted to extract common features of the targets, and the generalization ability of similar special-shaped targets is enhanced. And collecting unconcerned features of the prototype network classifier, decoupling common features in the unconcerned features by using orthogonality of the same type of different types of private features, mining target class common features to the maximum extent, and classifying by using the class common features. And compared with a training sample, accurate classification of similar special-shaped targets can be realized.

The invention discloses a dementia identification method based on brain computer network space cross attention fusion. The dementia identification method comprises the steps that resting-state electroencephalogram signals are acquired, preprocessing and brain network construction are carried out, the frequency band power ratio is calculated, and a training set and a test set are divided; the brain network fuses the spatial information, and spatial features are obtained through an isotropic graph neural network and a local feature enhancement module; the frequency band power ratio is coded by a multi-layer perceptron to obtain frequency spectrum statistical characteristics; a bidirectional cross attention module is input, and complementary fusion features are extracted; and performing mixed pooling and splicing on the complementary fusion features, then sending the fused features to a Chebyshev Kolmogorov-Arnold network classifier, and outputting an Alzheimer's disease / frontotemporal dementia / health control (AD / FTD / HC) classification result and a cognitive scale evaluation score (MMSE). According to the method, the multi-feature complementary information is effectively integrated through joint modeling of the spatial diagram features and the global spectrum features, and the accuracy, stability and generalization ability of AD and FTD classification in a complex brain network are improved.

The invention discloses a deep learning prediction method fused with multi-modal features, which can accurately identify protein hidden binding sites in a ligand-free (apoo) state. The method comprises the following steps of: firstly, constructing a protein graph by taking residues as nodes and taking C alpha distance less than or equal to 14 as edges, wherein node feature sets comprise amino acid one-hot, secondary structures, atomic attributes, protein language model embedding and BLOSUM62 evolutionary information, and edge features comprise distance and angle similarity; then capturing three-dimensional geometric equivariant features by adopting an equivariant graph neural network (EGNN), and modeling a chemical topological relation by using a graph isomorphic network (GINE) with edge features; eGNN and GINE double-branch feature fusion and global dependence integration are realized through gating cross attention and gating multi-head attention; and finally, inputting the fusion features into a Kolmogorov-Arnold network (KAN) classifier, and predicting whether each residue belongs to a hidden binding site or not. The method can adapt to large-scale conformation change without coordinate alignment, AUC and F1 on a standard data set are remarkably superior to those of an existing method, high robustness and generalization are kept for multi-chain protein and complex conformation, and the method can be widely applied to drug target discovery and structure-driven drug design.

The invention relates to a paper archive digital processing method and device, equipment and a medium. The method comprises the following steps: extracting image quality characteristics of a paper archive image, identifying a flaw region based on the image quality characteristics, and classifying flaw types through a pre-trained neural network; according to a preset defect type-image restoration algorithm mapping rule, restoring the defect area by adopting a corresponding algorithm to obtain a finished image; extracting a character sequence from the trimmed image by using an optical character recognition algorithm, encoding the character sequence into a semantic vector, and determining a document theme label in a pre-established archive theme library through vector similarity matching; and a multi-dimensional index structure is generated by combining the original image, the trimmed image and the character sequence, so that efficient retrieval and management of archive contents are realized. According to the method, the accuracy of defect repair and the recognition rate of optical character recognition are improved, and the automation level of digital processing is enhanced.

The invention provides a multi-modal brain network classification method based on feature decoupling and dynamic graph construction, and relates to the technical field of artificial intelligence, and the method comprises the steps: obtaining functional magnetic resonance imaging data and structural magnetic resonance imaging data of a to-be-detected person, and obtaining multi-modal brain network data according to the functional magnetic resonance imaging data and the structural magnetic resonance imaging data, and processing the multi-modal brain network data based on a preset multi-modal brain network classification model to obtain the brain network state of the to-be-tested person. According to the method, by combining functional magnetic resonance imaging (fMRI) and structural magnetic resonance imaging (sMRI) data, the advantages of the two modes can be fully utilized, so that the accuracy and comprehensiveness of brain network state classification are improved, a dynamic graph attention module based on a graph attention network (GAT) can capture the dynamic change characteristics of a brain network, dynamic graph representation is constructed, and the classification accuracy of the brain network state is improved. And the sensitivity of the model to the dynamic connection relationship between brain regions is enhanced.

The invention discloses a neuroblastoma pathological image classification method based on a CMSwinKAN neural network, and the method comprises the steps: firstly building a neuroblastoma pathological image data set, and dividing the data set into a training set and a test set; a CMSwinKAN neural network model is constructed, wherein the CMSwinKAN neural network model comprises a SwinKANsform feature extraction module, a comparison driving multi-scale aggregation module and a kernel activation network classification head; performing feature classification on the slices through an SVM classifier to obtain classification confidence; dynamically weighting the output of the CMSwinKAN neural network model according to the classification confidence, and outputting a final classification result; and taking the training set as input, training the CMSwinKAN model and the SVM classifier, and adjusting and optimizing parameters of the CMSwinKAN neural network model and the SVM classifier.

There is provided a computer-implemented method for verifying the robustness of a neural network classifier with respect to one or more parameterised transformations applied to an input, the classifier comprising one or more convolutional layers, the method comprising: encoding each layer of the classifier as one or more algebraic classifier constraints; encoding each transformation as one or more algebraic transformation constraints; encoding a change in an output classifier label from the classifier as an algebraic output constraint; determining whether a solution exists which satisfies the classifier constraints, transformation constraints and output constraints, and determining the classifier as robust to the local transformations if no such solution exists. A perception system and a computer readable medium are also provided.

The invention provides an intelligent lock panel display abnormity detection method and system based on an AI visual technology, and the method comprises the following steps: 1, classifying the defect types of an intelligent lock panel, including the logo color display error of the intelligent lock panel, the digital color display error, the black spot defect of digital display, and incomplete display; 2, acquiring an intelligent lock panel image to form a data set; step 3, constructing and training a model, wherein the model comprises a YOLOv3 network and 11 ResNet18 network classification models; the YOLOv3 network is composed of a Darknet-53 network and three branch network locks, the Darknet-53 network is a basic network, and the three branch networks are used for carrying out prediction of three scales respectively; after the images pass through the YOLOv3 network, the YOLOv3 model cuts the identified category images and sends the identified category images to the ResNet18 network; and 4, evaluating the model performance through the model evaluation index so as to obtain an accurate detection result. By applying the technical scheme, the defect accuracy of the intelligent lock panel can be improved.

The invention belongs to the technical field of image processing, and discloses a schizophrenia classification method based on connection gradient, which comprises the following steps: acquiring magnetic resonance imaging data, and preprocessing the magnetic resonance imaging data; performing brain region division on the preprocessed magnetic resonance imaging data, and constructing a functional connection network and a morphological similarity network; constructing a function connection gradient and a form similarity gradient by using a connection gradient algorithm based on the function connection network and the form similarity network; performing threshold processing on the function connection network and the morphological similarity network to serve as edge features, taking the bimodal connection gradient as node features, and constructing a graph convolutional network classification model; and training and testing the graph convolutional network classification model. According to the schizophrenia classification method based on the connection gradient, schizophrenia classification identification is realized by using the graph convolutional network classification model, the classification accuracy is improved, and the application value is higher.

The invention belongs to the technical field of ship navigation risk management, and particularly relates to a ship navigation risk control method. Comprising the following steps: acquiring a ship prepared standard data file through a ship navigation service system; determining corresponding elements under the accident type; sorting the comprehensive weight and the influence weight to form a risk accident element graph model; extracting historical navigation accident data according to element types in the risk accident element graph model; establishing a graph form sample database; a multi-layer network classification model is adopted to carry out classification training on the navigation map form samples; and matching ship risk accident features to be analyzed to carry out risk management and control. The method is used for completing risk accident feature classification and identification of different types of ships in the sailing process by analyzing various element features in the historical data of the ship sailing risk accidents, performing classification prediction on the to-be-analyzed ship sailing risk accidents, and determining the sailing accident features so as to take targeted accident prevention measures.

The invention discloses a method for classifying imperfect sorghum grains based on hyperspectral decision-level data fusion, which comprises the following steps of: taking sorghum as a research object, collecting hyperspectral spectrum information of the sorghum, respectively extracting statistical texture and frequency domain texture of a grayscale image, simultaneously extracting spectral information, analyzing overlapping and difference of characteristic peaks, and determining a characteristic wave band; respectively establishing a random forest and a long and short time sequence network classification model based on various kinds of preprocessed spectrum information and image texture information; on the basis of the baseline model, model classification performance is compared, and a cross-domain feature level fusion model is established; performing feature screening and data dimension reduction on the spectral information and the texture information to establish an intermediate data fusion model; and finally, realizing decision level fusion through Bayesian consensus and a weighting mechanism. Through the data fusion method under different levels, the data dimension is reduced, the model performance and robustness are improved, and an efficient solution is provided for development and practical application of sorghum nondestructive testing.

The invention discloses a washing machine motor detection method, and relates to the field of electric digital data processing, and the method comprises the following steps: S1, collecting signals of a plurality of sensors when a motor operates; s2, performing feature extraction on the signal of each sensor by using a multi-head self-attention feature extraction module fused with a Transform block to obtain a feature vector reflecting the running state of the motor; s3, constructing a residual network classification model, inputting the extracted feature vector into the model, and outputting a prediction type of a motor fault; s4, based on a sample set training model composed of normal operation data and fault data of the washing machine motor, a cross entropy loss function quantification model is adopted to predict the difference between a fault type and an actual fault type; and S5, minimizing a loss function by adjusting model parameters until the model prediction precision reaches a preset standard, processing a real-time operation signal of the motor by using the trained model, and carrying out fault detection. The optimal sensor configuration and high-precision fault detection of the motor power system of the washing machine are realized.

The invention discloses a superheat degree identification method based on self-supervised pre-training and feature fusion, and the method comprises the steps: the first stage is self-supervised pre-training based on a priori guidance learnable mask, and the second stage is supervised fine tuning based on feature fusion and a KAN network. The first stage comprises video slicing and space-time embedding, learnable masks guided by priori knowledge, and asymmetric encoder-decoder training; and the second stage comprises double-flow feature extraction, cross attention-based feature fusion, KAN network classification and supervised fine tuning. The invention relates to the technical field of industrial process intelligent perception and computer vision, and solves the problems of strong dependence on labeled data, weak generalization ability under small samples, incomplete feature expression and insufficient interpretability in the prior art.

The invention belongs to the field of auxiliary diagnosis of autism spectrum disorders, and particularly relates to a time sequence-based autism spectrum disorder recognition method, which comprises the following steps of: constructing and training an autism recognition model, and inputting to-be-recognized data into the trained autism recognition model to obtain a recognition result; the autism identification model comprises a graph attention space network feature extractor, a bidirectional long and short term network time feature extractor and a multi-layer perceptron network classifier; according to the method, potential graph structure information and time sequence dependence among all the regions of the brain region template can be comprehensively utilized, and the method has the advantages in the aspect of capturing important features in time sequence data and complex relations among nodes.

The invention relates to the technical field of data transmission, in particular to a transmission efficiency improvement system based on big data, which comprises a parameter acquisition module for periodically acquiring the transmission rate of each network, bandwidth occupation information in different time periods and transmission delay information, the efficiency evaluation module evaluates the efficiency level of each network bandwidth according to the transmission rate, the bandwidth occupation information and the transmission delay information; the classification module monitors time periods in a period for classification and classifies each network; the transmission optimization module determines network adjustment modes in different time periods based on a time period classification result and a network classification result, the dynamic adjustment module dynamically adjusts the data sending rate through a back pressure mechanism in the transmission process, and transmission fault tolerance and rapid recovery are achieved based on the asynchronous check point technology. According to the method, the network communication delay is reduced, the transmission efficiency is improved, and overload of a receiver is avoided; business pause is reduced, and the fault recovery speed is improved.

The application provides a multi-modal brain network classification method based on feature decoupling and dynamic graph construction, and relates to the technical field of artificial intelligence, and the method comprises the following steps: acquiring functional magnetic resonance imaging (fMRI) data and structural magnetic resonance imaging (sMRI) data of a to-be-tested person, obtaining multi-modal brain network data according to the fMRI data and the sMRI data, processing the multi-modal brain network data based on a preset multi-modal brain network classification model, and obtaining the brain network state of the to-be-tested person. The application can fully utilize the advantages of the two modalities by combining the fMRI data and the sMRI data, thereby improving the accuracy and comprehensiveness of the brain network state classification, and the dynamic graph attention module based on a graph attention network (GAT) can capture the dynamic change characteristics of the brain network, construct a dynamic graph representation, and enhance the sensitivity of the model to the dynamic connection relationship between brain regions.

The invention discloses a method for classifying brain slices by combining an improved capsule network and a residual network. The method comprises the following steps: acquiring a magnetic resonance image of a to-be-measured target; inputting the magnetic resonance image into a trained residual capsule network to obtain a reconstructed image; and obtaining a classification result of the brain slices by using the reconstructed image. Wherein the residual capsule network comprises an encoder and a decoder, the encoder comprises a residual network and a capsule network, the decoder comprises a deconvolution layer, and the residual network is used for extracting feature maps of different depths from an input image and transmitting the feature maps to the capsule network; the capsule network carries out vectorized weighted merging on the feature maps of the different depths, and the decoder reconstructs the image by using the longest vector in the connected capsules of the capsule network. According to the method, the definition of the reconstructed image is improved, the image reconstruction time is shortened, and the method is particularly suitable for processing small, unbalanced and large DE-MRI data sets.

The present invention discloses an underwater target tracking method based on a twin prototype hybrid network. The method first obtains original underwater image data, defines the target and support set images, and performs background separation preprocessing. Secondly, an adaptive cutoff filter and a twin network are designed, a prototype network is established and it is integrated with the twin network. Then, two different prototype expressions under the same support set are introduced, and after mutual convolution, the anchor points near the target are judged and screened through an LSTM network. Finally, a regional candidate network classifier for classification and regression is constructed. Based on the anchor points near the target, the classification channel and the regression channel are output, and multi-dimensional aggregation is performed to obtain multi-category classification results and regression results. The item with the highest probability in the classification result is selected, and its corresponding regression result is output as the target position coordinate to realize category recognition and positioning. The present invention can effectively realize underwater target tracking in the presence of background interference.

The invention provides a dual confidence calibration method and system for a neural network classifier, equipment and a medium, and effectively solves the problems that an existing single-stage calibration method is difficult to give consideration to excessive confidence, under confidence, class imbalance sensitivity and the like. A mixed loss function fusing bifocus loss and difference between multi-class confidence and accuracy is introduced in a training stage, so that a neural network classifier is promoted to generate well-calibrated prediction distribution; a class-by-class multi-partition temperature scaling model optimized based on a coupling simulated annealing method is adopted in the reasoning stage, and calibration requirements of different classes and different confidence intervals are more accurately met compared with a traditional temperature scaling technology of a single temperature coefficient; in the training process, classifiers of different rounds are stored, multi-model calibration results are averaged class by class in the test stage, the calibration stability is effectively improved, prediction confidence errors are reduced, and the method is particularly suitable for the safety key fields such as medical diagnosis and automatic driving which have extremely high requirements for prediction reliability.

The embodiment of the application relates to the technical field of digital medical treatment, and specifically provides a text classification model training method, a classification method, a terminal and a storage medium. The method comprises the following steps: inputting a training sample into a text coding network to obtain a first text vector corresponding to the training sample; inputting the first text vector into a context coding network to obtain a second text vector corresponding to the first text vector; inputting the second text vector into a first classification network and a second classification network respectively to obtain a first category probability distribution of the training sample classified by the first classification network and a second category probability distribution of the training sample classified by the second classification network, constructing a first loss function according to the first category probability distribution and constructing a second loss function according to the second category probability distribution, and determining a fusion loss function according to the first loss function and the second loss function; and iteratively updating a text classification model based on the training sample and the fusion loss function to obtain a target text classification model.

The present application provides a method, apparatus, computer device and storage medium for detecting objects thrown from height, the method including: setting up a video surveillance area for capturing a target video image in real time; detecting all moving objects within the target video image using a background modeling algorithm; tracking the detected moving objects using Kalman filtering to obtain trajectory characteristics and related parameters of the moving objects; identifying whether the moving objects are objects thrown from height through rule-based identification of objects thrown from height; performing classification using a Long Short-Term Memory (LSTM) network classification model, including inputting parameters of the moving objects as feature data into the LSTM, to obtain a classification result and determine whether a false positive occurs. The method has strong anti-interference ability and can effectively filter out moving objects other than objects thrown from height.