73 results about "Neural network classification" patented technology

The invention discloses an arrhythmia detection method based on a pulse neural network. The method comprises the following steps: performing incremental modulation pulse coding and pooling operation on a target electrocardiosignal to obtain a corresponding pulse sequence; and inputting the pulse sequence into a trained multi-level pulse neural network classification model to obtain an arrhythmia detection result. Wherein the multi-level pulse neural network classification model comprises a weight sharing layer and a plurality of cascaded classifiers, the weight sharing layer is used for extracting electrocardiosignal feature information from the pulse sequence, and the plurality of classifiers are used for obtaining an arrhythmia detection result based on the electrocardiosignal feature information. According to the method, on the premise that the accuracy is equivalent, the classification result is more refined, and the requirements for hardware storage resources and computing resources are reduced.

The invention belongs to the technical field of artificial intelligence, and particularly relates to a computer network intrusion detection method and system based on artificial intelligence, and the method comprises the steps: constructing a finite state automaton to carry out the state jump compliance verification of a connection flow, intercepting the illegal flow, only sending the compliance flow into a feature engineering module, and extracting the multi-dimensional behavior features; inputting a deep neural network classification model to determine whether the behavior is an intrusion behavior; according to the technical scheme provided by the invention, the data input to the artificial intelligence model can be ensured to have complete compliance on the protocol level, so that the model is prevented from learning false feature association caused by protocol violation, and the antagonistic attack based on protocol state jump can be effectively resisted on the premise of not depending on adversarial training.

The application relates to the technical field of positioning device signal processing and positioning precision, and particularly discloses a non-line-of-sight classification and error compensation method and system based on a double convolution neural network, the method comprising the following steps: acquiring channel impulse response data of an ultra-wideband signal, and pre-processing the channel impulse response data to obtain an amplitude sequence in real number form; constructing a double convolution neural network classification model, the double convolution neural network classification model comprising parallel large-kernel convolution blocks and small-kernel convolution blocks, which are respectively used for extracting macroscopic waveform features and microscopic fluctuation features from the amplitude sequence; and performing classification training after fusing the macroscopic waveform features and the microscopic fluctuation features; and based on the trained classification model, the regression compensation of ranging errors of non-line-of-sight signals is performed, macroscopic waveform contour features and microscopic local fluctuation features can be simultaneously extracted from data, and a complementary feature extraction mechanism can more comprehensively describe the differences between LOS / NLOS signals compared with a single-structure network.

The present application relates to a kind of generalization adversarial sample group generation method based on easily confused class feature injection, belong to the technical field of adversarial attack of computer vision field.It introduces the class disturbance update momentum in the direction of easily confused class feature vector in the process of generating class-oriented generalization disturbance sample group, the class disturbance update momentum makes the adjacent easily confused class of adversarial example faster update, to realize the generalization attack to class.The present application only integrates the features of easily confused class in the same class image, forms disturbance template, realizes class-oriented generalization disturbance sample group;Variable step size iterative attack means is used to explore easily confused class in the process of neural network classification, realize easily confused feature injection, while introducing class momentum parameter in the process of iterative attack, improve attack efficiency and alleviate the over disturbance shortcomings of generalization disturbance.

The application discloses a dynamic convolution deep neural network classification method with global and local attention, and belongs to the technical field of image classification and artificial intelligence, wherein a dynamic convolution kernel module with global and local attention is designed and built to improve the traditional convolution operation by analyzing the convolution layer in the deep convolution neural network. The method comprises the following specific steps: S1, dividing a data set into a training set and a test set; S2, designing and building a dynamic convolution kernel module with global and local attention; and S3, training and testing the built dynamic convolution deep neural network with global and local attention on a natural image classification data set, so as to reduce the parameter quantity and the calculation complexity of the network, thereby better solving the image classification problem.

The application belongs to the technical field of image processing, and particularly relates to a defect classification method for automobile injection molding parts based on database comparison, which comprises the following steps: extracting a feature vector fused with multi-scale artificial information and deep learning information for normal, defect and to-be-detected areas; realizing efficient preliminary screening of abnormal areas by calculating a local anomaly factor of the to-be-detected area and comparing the local anomaly factor with a dynamic determination threshold generated based on defect data; and splicing the original feature vector and the quantized local anomaly factor into an enhanced feature vector, inputting the enhanced feature vector into a neural network classification model, and performing accurate defect category determination. The two-stage strategy of dynamic threshold preliminary screening and enhanced feature accurate division improves the accuracy and stability of defect detection of automobile injection molding parts.

The invention relates to the technical field of block chain power supply management, and discloses a distributed power supply transaction management method of block chain nodes. According to the method, historical normal power supply transaction data is collected to establish a reference database, and the transaction detection period duration is determined according to the data size of a real-time power supply transaction data set. And monitoring node transaction records in a detection period, calculating a transaction frequency average value as a detection threshold value, and identifying suspected abnormal transactions in combination with the reference database. After the suspected abnormal transaction is confirmed, extracting the total transaction amount and the time distribution characteristic to judge whether the transaction is the abnormal transaction, and counting the occurrence frequency to calculate an initial risk coefficient; and the category of the abnormal transaction, the participation node identifier and the behavior pattern are extracted and converted into feature vectors, and malicious transaction node groups are grouped, analyzed and detected through a neural network classification algorithm. And if the malicious group exists, calculating an adjustment coefficient to correct the initial risk coefficient, and dynamically adjusting parameters of the next detection period.

The invention discloses a buried pipeline leakage signal collaborative classification system based on a twin neural network. The system comprises a data acquisition module, a signal preprocessing module, a twin neural network classification module, an intelligent decision module and a man-machine interaction module. The twin neural network comprises two sub-networks sharing weights, and the two sub-networks respectively process multi-modal fusion signals of the same pipe section at different moments, so that intelligent detection of pipeline state changes is realized. According to the method, the leakage detection precision can be effectively improved, the false alarm rate is reduced, accurate classification of different leakage types is realized, and reliable guarantee is provided for safe operation of buried pipelines. According to the system, the DTS is matched with the DAS technology, and intelligent identification and classification of pipeline leakage signals are realized by utilizing a laying mode that an optical cable is attached to a buried pipeline in an accompanying manner.

The application relates to the technical field of artificial intelligence, and provides a classification prediction method based on table data, which comprises the following steps: inputting a characteristic value, a characteristic type corresponding to the characteristic value and statistical information corresponding to the characteristic value into an embedding layer of a neural network classification model to obtain an embedding representation corresponding to the characteristic value; then determining an embedding representation sequence according to the embedding representation; inputting the embedding representation sequence into an encoding layer of the neural network classification model to obtain a semantic vector sequence; inputting the semantic vector sequence into an output layer of the neural network classification model to obtain a prediction result; determining a loss value according to the prediction result and a labeled label, and training the neural network classification model according to the loss value to obtain a target neural network classification model. Through the above scheme, the neural network model can understand the global statistical information of the characteristics like a tree model, the performance and accuracy of the neural network model in performing a classification prediction task based on table data are improved, and the satisfaction of users in the experience of financial products is improved.

A kind of construction method and device of depression recognition model based on deep network feature fusion, its method includes: data acquisition;Build the video dataset containing video regional level label;Respectively using C3D neural network and channel attention network to the whole video stream is reinforced feature extraction;The reinforced feature is used RPN to generate region proposal and produces region start mark and end mark and region expression class;All video frames are extracted to multiple scale features including local eccentricity features and global features;Weakly supervised classification is carried out to multiple scale features;The operation result of video stream level feature and picture level feature is classified by neural network, and depression recognition model is obtained.The present application effectively utilizes video label, takes into account global and local information, and combines the advantages of high accuracy and reduced label workload, can become a potential depression recognition auxiliary means, and can be widely applied in the field of automatic processing of mental illness.

The invention discloses an early termination method and device for inter-frame coding segmentation, equipment and a storage medium. The method comprises the following steps: calculating a feature distance between to-be-segmented regions based on coding block features; inputting the feature distance into a pre-trained neural network classification model for segmentation threshold prediction, and outputting a target segmentation threshold of the target coding block; based on the feature distance and the target segmentation threshold, judging whether the target coding block meets a segmentation termination condition or not; and if yes, stopping segmentation of the target coding block. According to the embodiment of the invention, features highly related to CU segmentation are selected by using a machine learning algorithm, features having the greatest influence on a segmentation decision are selected by analyzing texture and depth map information of the CU, and early termination of CU segmentation of a texture video and a depth map is realized based on a threshold value extracted from a training model.

The application discloses a power distribution system fault prediction method based on fractal theory prediction and BP neural network classification, collects current data of a line to be predicted in a power distribution system, and performs fractal prediction. The fractal prediction comprises the following steps: S1, establishing a current history sequence, obtaining an accumulated sum sequence and a measurement scale; S2, performing linear regression on the accumulated sum sequence curve in a double logarithmic coordinate system to obtain a fractal dimension; and S3, performing current value prediction of future time. After the fractal prediction, the actual current value is measured, the difference between the actual value and the predicted value is calculated, the BP neural network is inputted, and a fault prediction result is obtained. The power distribution system fault prediction is performed by adopting the method combining fractal theory prediction and BP neural network classification, the method has the characteristics of high accuracy, and is suitable for fault early warning and troubleshooting of a bidirectional power flow and a multi-party participated smart grid.

The application provides a light diffraction neural network classification method and system based on ensemble learning, aiming to improve the object classification accuracy in a scattering medium environment, comprising the following steps: step 1, constructing and generating a simulation dataset suitable for light diffraction neural network training, the dataset containing a target object dataset and a scattering medium dataset; step 2, building multiple differentiated light diffraction neural network forward propagation models, and training each model based on the above simulation dataset; step 3, using a preset ensemble learning strategy to fuse the classification results output by the multiple light diffraction neural networks to improve the overall classification accuracy; step 4, driving the optical system to run, and outputting the classification result of the target object in real time according to the light field intensity data collected by the detector. The application introduces an ensemble learning mechanism to construct an anti-scattering classification strategy, effectively improving the object classification accuracy and robustness under the interference condition of static or dynamic scattering medium.

This invention discloses a method for constructing a pulse data classification model and a pulse data classification method, belonging to the field of classification and recognition technology using spiking neural networks. The construction method includes: training a spiking neural network; the gated parameter neuron layer includes: a synaptic current input terminal for calculating the synaptic current input at time t, and a membrane voltage input terminal for obtaining the membrane voltage v of the input neuron at time t. t The first to fourth logic gates are used to calculate the hidden membrane voltage calculation unit, which is used to calculate the hidden membrane voltage at time t. The pulse emission unit is used to emit the pulse and calculate the output membrane voltage. The output membrane voltage calculation unit is used to calculate the output membrane voltage at time t+1. By introducing the gating structure, the spiking neural network can better fit biological characteristics in terms of parameter distribution, better adapt to backpropagation training, improve the training effect of the spiking neural network and the classification accuracy of the spiking neural network for classification.

This application provides a method and system for adaptive process control in laser cutting, belonging to the field of laser cutting technology. The method involves: acquiring and preprocessing multimodal data of the material to be cut; extracting multidimensional feature vectors characterizing material properties; inputting these vectors into a deep neural network classification model; outputting a preliminary probability distribution of the material type; combining this distribution with electromagnetic and visual features as independent evidence; and using D-S evidence theory for fusion reasoning to determine the final material type. Then, based on a Kalman filter algorithm, spectral distance and geometric contour data are fused to obtain a final thickness estimate. Anomaly detection is performed using the final material type and final thickness estimate, triggering anomaly handling. Based on the final material type and thickness estimate, a preset database is queried to obtain a set of basic process parameters, which are then fine-tuned using a parameter optimization algorithm and configured into the CNC system of the laser cutting machine. This application achieves adaptive control of laser cutting through multimodal fusion recognition.

This invention relates to an image classification method, apparatus, and medium for transfer-based black-box adversarial defense. The method includes acquiring an input image to be classified; determining a predefined expansion factor, spatial phase offset, and gap-filling value; calculating signal pixel positions based on the expansion factor and spatial phase offset; mapping the pixel values ​​of each original pixel to a high-dimensional space according to their corresponding signal pixel positions to form an expanded representation; filling gaps in regions other than all signal pixel positions with gap-filling values; and training a target neural network classification model with the obtained expanded representation. During inference on the trained target neural network classification model, the same acquisition, determination, calculation, mapping, and filling operations are performed on the input image, and the obtained expanded representation is input into the target neural network classification model to obtain the classification result. This method achieves a synergistic improvement in clean sample accuracy and black-box transfer robustness without relying on adversarial training.

This invention discloses a method and apparatus for identifying IoT devices in an active detection scenario. The method includes: acquiring first network text data, wherein the first network text data includes network text data based on network devices; extracting network image features from the first network text data to output a first image feature vector, and extracting a first statistical feature vector from the first network text data; fusing and concatenating the first image feature vector and the first statistical feature vector to obtain a network device feature vector; and inputting the network device feature vector into a pre-trained neural network classification model to obtain a probability vector, thereby using the probability vector to identify IoT devices. This invention, by converting HTML text into images, not only preserves the features of the HTML text but also utilizes an advanced image feature extraction model to extract more effective features, significantly improving the accuracy of device identification; furthermore, it avoids such problems by utilizing image features, expanding the scope of device identification.

The invention belongs to the technical field of knowledge graph data classification, and discloses a knowledge graph data multi-level classification method, a knowledge graph data multi-level classification system and a medium. Then establishing a multi-dimensional feature space containing entity dimensions, relation dimensions and attribute dimensions, constructing a feature matrix and carrying out hierarchical processing; then, a deep neural network classification model is adopted for feature extraction and fusion, adaptive optimization of features is achieved through a feature matrix optimization equation set, and the steps of feature importance degree evaluation, feature correlation calculation, feature weight optimization and the like are included. And finally, a final classification model is obtained through model training and optimization, multi-level classification of the knowledge graph data is realized, and the technical problem of low classification precision caused by insufficient feature extraction of the knowledge graph data in the prior art is solved.

The application discloses an anti-interference method of a surface acoustic wave passive wireless sensing system based on artificial intelligence. Firstly, radio frequency signal intensity data of various surface acoustic wave echo signals including noise, normal response and interference under different frequencies are acquired, and different labels are given to the radio frequency signal intensity data. The radio frequency signal intensity data is sequentially preprocessed, standardized, one-hot encoded and an additional dimension is added, so that the data becomes three-dimensional. A one-dimensional CNN neural network classification model of a training attention mechanism is constructed. The model comprises a one-dimensional convolution layer used for extracting time-frequency features of the radio frequency signal intensity data, and the size of the convolution kernel is 1, the padding is 1 and the step is 1, and an attention mechanism module used for weighted average of the time-frequency features extracted by the one-dimensional convolution layer. The one-dimensional CNN neural network model after training is used to output the category of a new surface acoustic wave echo signal. When the anti-interference is realized, the time domain and frequency domain information of the echo signal can be reserved, the loss of information is reduced, and the signal reconstruction quality and reliability are improved.

A deep learning-based method for monitoring anesthesia depth based on high time-frequency resolution bispectral density includes the following steps: acquiring single-channel EEG data, preprocessing it, and dividing it into time segments; extracting features from the EEG data, and calculating a high time-frequency resolution wavelet bispectral matrix by combining variational mode decomposition and synchronous squeeze wavelet transform; constructing and training a deep neural network classification model based on partially convolutional neural networks and gated recurrent units; for test data, inputting the extracted features into the model, and selecting the state with the highest output probability (awake, anesthesia maintenance, or awakening) as the final prediction result for the test data; this invention uses single-channel EEG signals as input, greatly reducing the difficulty of acquiring clinical signals, and the proposed high time-frequency resolution bispectral features combined with a dedicated deep learning network have a strong ability to capture complex nonlinear EEG dynamic features and cross-subject generalization adaptability; therefore, this invention can perform high-precision, multi-classification, and real-time automatic monitoring of anesthesia depth.

Diagnosis of prostate cancer often involves invasive measures, such as biopsies, in which tissue is removed and graded by a pathologist. Non-invasive measures, such as multi-parametric MRI, can be used to examine lesions and grade them using the PI-RADS scoring scale; however, achievable image resolution is limited by motion. A new paradigm, magnetic resonance histopathology (MRH), is disclosed, which aims to evaluate tissue texture at sub-millimeter resolution through a fast clinical acquisition process. The ability of MRH to identify cancerous tissue in the prostate through computer simulation analysis is demonstrated, reproducing the MRH measurement process on a set of high-resolution, histology slides annotated by pathologists (2, 4, 6, 8, 10, 12, 14). A dataset of spectral intensities (20) at sub-millimeter wavelengths is created, and a deep learning model (22) trained to classify the spectral data is based on normal or tumor tissue. A set of spatial frequencies is identified, which is used to optimize diagnostic capability under the constraint of limited acquisition time. In addition to single-region classification, the disclosed method and architecture integrate spatial context and local information, the inclusion of which improves model performance and denoises inference results. In addition to algorithms for estimating the physical length scale of lesions identified by the model, the trained model is demonstrated to be applied to unlabeled high-resolution 3D MRI data.

This invention discloses a quantitative discrimination method and application for the geological types of fracture-vuggy units in fracture-vuggy reservoirs, relating to the field of petroleum geological exploration technology. The quantitative discrimination method of this invention includes the following steps: using a clustering algorithm to perform cluster analysis on the static parameters of fracture-vuggy reservoirs, classifying fracture-vuggy units, analyzing the clustering results, and determining the number and characteristics of categories; using well-understood fracture-vuggy units as a dataset, with the static parameters described in step (1) as input and the geological type corresponding to each fracture-vuggy unit as output, constructing a neural network model; and optimizing the results of cluster analysis and neural network classification. This invention utilizes cluster analysis and neural network technology to achieve accurate classification of the geological types of fracture-vuggy units through quantitative discrimination using five selected key static parameters.

The present application belongs to the technical field of knowledge graph data classification, and discloses a knowledge graph data multi-level classification method, system and medium. The knowledge graph data multi-level classification method proposed by the present application firstly pre-processes the original data to obtain standard data, then establishes a multi-dimensional feature space containing entity dimension, relationship dimension and attribute dimension, constructs a feature matrix and performs hierarchical processing. Then, a deep neural network classification model is used for feature extraction and fusion, and the self-adaptive optimization of features is realized through a feature matrix optimization equation set, including feature importance evaluation, feature correlation calculation and feature weight optimization. Finally, the final classification model is obtained through model training and optimization, realizing multi-level classification of knowledge graph data and solving the technical problem of low classification accuracy caused by insufficient feature extraction of knowledge graph data in the prior art.

The application discloses an image recognition method and device based on contrast knowledge distillation, comprising: obtaining training image samples with category labels; performing supervised learning training on a first neural network classification model based on the category labels by using the training image samples, inputting the training image samples into the trained first neural network classification model to obtain first embedding feature expressions and corresponding first classification soft labels; inputting the training image samples into a second neural network classification model to obtain second embedding feature expressions and corresponding second classification soft labels, calculating a contrast distillation loss based on the first embedding feature expressions and the second embedding feature expressions, calculating a KL divergence loss based on the first classification soft labels and the second classification soft labels, updating parameters of the second neural network classification model in combination with an image recognition task loss, and using the updated second neural network classification model as an image recognition model for image recognition.

A continuous fiber 3D printing process detection method based on artificial intelligence image recognition selects a process monitoring technology based on computer vision and pattern recognition to monitor the continuous fiber reinforced composite material printing process, acquires the continuous fiber printing image through the camera, combines the neural network classification model to realize the image recognition classification function, and further realizes the automatic detection of the printing defects in the printing process; the present application can realize the automatic monitoring of defects in the continuous fiber 3D printing process, and can lay a foundation for subsequent real-time printing control or repair, and further realize the intelligent controllable printing quality, improve the qualified rate of continuous fiber 3D printing products, reduce material waste and shorten the manufacturing time.

The application discloses an unknown threat identification method based on adaptive snapshot integration, which is applied to instant fine classification of network encrypted traffic and can identify unknown threats caused by concept drift. The method comprises an encrypted traffic preprocessing and feature extraction module, an encrypted traffic time sequence feature neural network classification module, a new category traffic discrimination module, a snapshot integration module and a transfer learning module. The method can cope with increasing unknown abnormal traffic, give unknown abnormal traffic clustering results for experts to mark, effectively solve the concept drift and category imbalance problem, and thus guarantee the reliability of abnormal traffic detection.

The invention relates to the field of article classification and recognition, in particular to a medicine bottle classification method based on visual recognition, which comprises the following steps: firstly, acquiring an image of a medicine bottle to be classified through a visual acquisition device, and carrying out denoising, size normalization, brightness enhancement and the like on the image; the processed images are input into a pre-trained neural network classification model to obtain the medicine bottle category confidence coefficient, a category result meeting a preset threshold value is output, and the result can also be sent to a control module to control the sorting device. The technical effects that the medicine bottles can be efficiently and accurately classified and recognized, related devices can be controlled to execute corresponding operations, and the model can support new medicine bottle category training and updating through transfer learning are achieved.

The application provides a neural network classification model training method, the model comprising a feature extractor for extracting sample features and a classifier for classifying samples according to sample features, the method comprising one or more rounds of training of an initial model trained based on source domain samples and labels in the following manner: S1, obtaining a training set comprising source domain sample pairs and target domain sample pairs, any sample pair comprising two samples, each sample being sensor data collected from a person, the two samples corresponding to the same person and the same behavior category of the person when collected; S2, training the current model one or more times using the training set, updating the parameters of the model based on a predetermined loss function each time, the loss function being configured to: penalize classification bias of source domain samples, penalize edge distribution difference and conditional distribution difference between sample features of source domain and target domain samples, and penalize time series distribution difference corresponding to the two samples in each sample pair.

The application relates to the field of article classification and recognition, in particular to a medicine bottle classification method based on visual recognition. First, a vision acquisition device is used to acquire a medicine bottle image to be classified, the image is subjected to denoising, size normalization, brightness enhancement and the like, the processed image is input into a pre-trained neural network classification model to obtain a medicine bottle category confidence, and a category result meeting a preset threshold is output; the result can also be sent to a control module to control a sorting device. The application has the technical effect that medicine bottles can be efficiently and accurately classified and recognized, relevant devices can be controlled to perform corresponding operations, and the model can support the training and updating of new medicine bottle categories through transfer learning.

The present application relates to the field of intelligent terminal perception, and discloses a human body recognition method and device based on radar sensing and RGB camera fusion. The radar sensor analyzes the target micro-Doppler characteristic fingerprint vector and compares it with the pre-stored interference characteristic library, generates a wake-up signal after filtering the known interference source. The processor responds to the wake-up, calculates the predicted coordinates of the target at the exposure moment in combination with the system time delay, maps the predicted coordinates to the light measurement area parameters before starting the camera, converts the target speed into the upper limit parameters of the exposure time and the focus driving value and issues them. The camera collects images according to the parameters, and the processor classifies using a neural network: if it is not a human body, the interference suppression logic is triggered and the current characteristics are written into the interference library, and if it is a human body, the business response is executed. Through radar bottom filtering and camera parameter feedforward configuration, the present application solves the problems of high radar false alarm rate and poor cold start imaging quality of the camera, reduces the system power consumption and improves the recognition accuracy and response speed.