57 results about "Neural network classifier" patented technology

The invention relates to a risk identification method and system. The method comprises the following steps: acquiring communication behaviors, equipment fingerprints and service interaction data in real time through a privacy compliance interface; carrying out parallel preprocessing and safe desensitization on the data; inputting a feature construction engine to extract a communication mode, an equipment behavior and an interactive semantic feature vector in parallel, constructing a dynamic weighted hypergraph communication map, and outputting a social risk feature vector by using a time sequence hypergraph neural network; inputting the four types of features into a privacy perception multi-mode gating attention fusion module to output fusion features; generating a risk score through a deep neural network classifier; and monitoring an abnormal event based on Apache Flink, adaptively adjusting a decision boundary in combination with the dynamic risk entropy, and triggering secondary verification. According to the method, the problems of multi-modal data conflict, privacy disclosure and decision stiffness are solved, and the recognition precision and the real-time performance are improved.

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 relates to the technical field of gesture recognition interaction, and discloses an intelligent interaction system and method based on gesture recognition. The method comprises the following steps: when an environment meets a condition, synchronously acquiring an image sequence of a gesture and depth distance field data; a hand joint point two-dimensional motion track is extracted from an image, surface deformation fluctuation information is separated from depth data, the two-dimensional motion track and the surface deformation fluctuation information are subjected to space-time registration fusion, a continuous motion track curved surface in a three-dimensional space is constructed, and then geometric topology features and dynamic change features of the continuous motion track curved surface are extracted. And inputting the fusion features into a neural network classifier subjected to incremental learning training, outputting corresponding semantic tags and confidence evaluation values, and mapping to generate a control instruction after verification. According to the method, an accurate three-dimensional dynamic model is constructed through deep fusion of multi-source data, and an incremental learning mechanism is utilized to enable the system to have online adaptive capability, so that the recognition precision of complex gestures and the long-term applicability of the system are improved.

Embodiments of the present application provide a quantum neural network classifier training method and device, electronic equipment and medium. The scheme is as follows: obtaining a training data set and a to-be-trained classifier; for each training sample data, classifying the training sample data by using the to-be-trained classifier to obtain a first predicted label; calculating a first loss value of the to-be-trained classifier according to a sample label corresponding to each training sample data and the first predicted label; when the to-be-trained classifier has not converged, adjusting the classifier parameters based on the first loss value, and returning to execute the step of classifying each training sample data by using the to-be-trained classifier to obtain the first predicted label corresponding to the training sample data until the to-be-trained classifier converges at the current time. Through the technical scheme provided by the embodiments of the present application, the optimization of the quantum neural network classifier is realized, and the classification accuracy and attack resistance of the quantum neural network classifier are improved.

A system and method utilizes masked language models in order to provide data-type detection, such as (but not limited to) prediction of columnar headings. Two masked language models are pre-trained on example columnar text. One model predicts missing data at the entity level (e.g., masked entity names that may be made up of whole words), while the other predicts missing data at the character level (e.g., masked individual characters). The table with missing column headings is fed into both models, and the output is contextual word embeddings and contextual character embeddings. These results are merged, and then fed into a neural network classifier to then predict the column names.

The invention discloses a vehicle collector shoe detection method, device and equipment and a storage medium, is applied to vehicle collector shoe detection equipment comprising a collector shoe acquisition unit, a collector shoe image storage and recognition unit and data transmission equipment, and relates to the technical field of automatic detection. Comprising the following steps: generating a pulse signal by using magnetic steel in a collector shoe acquisition unit when a vehicle passes, and sending the pulse signal to an area-array camera in the collector shoe acquisition unit to acquire an image of a collector shoe to obtain a target collector shoe image; transmitting and storing the target collector shoe image to a collector shoe image storage and identification unit by using data transmission equipment; a deep neural network classifier in the collector shoe image storage and recognition unit is used for carrying out component positioning on the target collector shoe image to obtain a component positioning result, geometric dimension, inclination angle and surface defect detection is carried out on the component positioning result in sequence, and an abrasion value detection result, a crack detection result and a notch anomaly detection result are obtained; the efficiency of detecting the collector shoe is improved.

To provide a real-time task distribution method and device for disassembly work by human and robot cooperation.SOLUTION: The present invention obtains an image of a fastening part to be disassembled, then constructs a two stage neural network classifier to detect the quality state of the fastening part to be disassembled, and finally uses a genetic algorithm to solve a disassembly human and robot task distribution plan, thereby realizing the detection and state classification of the fastening part to be disassembled. The real-time human-robot collaborative task distribution can make full use of the advantages of the accuracy and durability of the robotic arm, and at the same time, use the intelligent resources of the operator to greatly reduce the fatigue of the operator. In addition, the flexibility of the disassembly work is improved, the problem of uncertain failure of the product is solved, and the quality, safety and efficiency of the disassembly work are improved by making the best use of the flexible response ability of the human and the accurate repetitive operation ability of the robotic arm.SELECTED DRAWING: Figure 1

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.

Provided are a method for regulating a wind-photovoltaic-storage power station based on electricity, green certificate, and carbon price prediction, a device, a medium, and a product. The method includes: inputting acquired historical price data into a price prediction model, and outputting a predicted price; determining a deviation vector of price data based on the historical price data and the predicted price, and generating an uncertainty set of the predicted price by using a multi-kernel-based one-class support vector machine algorithm; classifying the uncertainty set of the predicted price by using a neural network classifier, to obtain multiple types of price scenarios; solving, based on predicted prices under the multiple types of price scenarios, a joint clearing model by using a Pied Kingfisher Optimization (PKO) algorithm, to obtain an operation strategy for the wind-photovoltaic-storage power station; and regulating the wind-photovoltaic-storage power station based on the operation strategy for the wind-photovoltaic-storage power station.

Implementations of a fully configurable, modularized home robot are described that implement trained neural network classifiers to solve problems of providing home health care and monitoring of the elderly and / or infirm, providing entertainment for the family, providing environmental and safety monitoring coupled with mechanisms to clean the air and remedy indoor climates such as humidity and temperature, and provide a mechanism for caring for pets left at home when the owner is away based upon sensory input.

The present invention pertains to the computer implemented method for identifying a microorganism in a blood comprising: receiving (101) a microscopic image of a blood smear; detecting (102) plurality of microorganism cells in the microscopic image using a deep learning cellular segmentation model; extracting (103) from the segmented microscopic image a patch of size HxW centred around a centroid of the detected microorganism cell for each detected microorganism cell; encoding (104) of each patch by a deep neural network encoding model into a patch representation vector u n of size 1×D; aggregating (105) the patch representation vectors un using a pooling method to generate a sample representation vector u of size 1×D; and indicating (106) a species of microorganism present in the microscopic image by classifying the sample representation vector u by a neural network classifier to class representing specific species. In another aspect the invention relates to the data processing system for carrying out the steps of the method according to the invention.

A method of scanning N×N pixels using a vector-by-matrix multiplication array by (a) associating a filter of M×M pixels adjacent first vertical and horizontal edges, (b) providing values for the pixels associated with different respective rows of the filter to input lines of different respective N input line groups, (c) shifting the filter horizontally by X pixels, (d) providing values for the pixels associated with different respective rows of the horizontally shifted filter to input lines, of different respective N input line groups, which are shifted by X input lines, (e) repeating steps (c) and (d) until a second vertical edge is reached, (f) shifting the filter horizontally to be adjacent the first vertical edge, and shifting the filter vertically by X pixels, (g) repeating steps (b) through (e) for the vertically shifted filter, and (h) repeating steps (f) and (g) until a second horizontal edge is reached.

The invention discloses a medical pathology image adaptive probability calibration classification method and learning based on label noise, and the method comprises the steps: firstly, obtaining a data set, and dividing the data set into a training set and a verification set according to a certain proportion; the training set is used for training a deep neural network classifier, and the verification set is used for calibrating the model. Secondly, probability distribution under a noise label is calculated, and a noise transfer matrix is introduced to adjust prediction distribution of the model; and thirdly, constructing a consistency calibrator by optimizing an optimal temperature scaling factor T and combining a noise transfer matrix, so that the consistency calibrator can be theoretically converged to a calibrator trained on clean data. Finally, the superiority of the method in different data sets and noise environments is verified through experiments, and the improvement of the existing built-in calibration method and post calibration method is realized. According to the method, the noise transfer matrix is introduced, self-adaptive adjustment is carried out on calibration errors in different noise environments, and finally robust probability calibration is achieved.

The invention discloses a webpage classification method and system based on a Delaunay triangulation double-graph neural network, and belongs to the technical field of webpage classification. The method comprises the steps that original graphs and node features of webpage classification are acquired; the original graph comprises a plurality of nodes and directed edges among the nodes, the nodes are webpages, the directed edges are hyperlinks among the webpages, and the node features are webpage contents; constructing a shadow graph by adopting a Delaunay triangulation method; respectively carrying out graph convolution operation on the original graph and the shadow graph to obtain local structure information and global relation information of nodes; and the local structure information and the global relation information are fused to generate final node representation, and all nodes are classified by adopting a neural network classifier based on the final node representation. According to the method, the graph expression capability and the information spreading efficiency are effectively enhanced and the webpage classification effect is improved on the premise of not destroying the original graph structure semantics.

The invention discloses a method and a system for intelligently identifying damage of a hydrogen conveying pipeline based on multi-sensor feature fusion. The method comprises the following steps: firstly, acquiring multi-source sensor detection signals including electromagnetic and ultrasonic; performing time domain, frequency domain and time-frequency domain feature extraction on each sensor signal, and constructing an original feature set; then, a restricted Boltzmann machine is adopted to carry out unsupervised deep feature learning and dimension reduction processing on the high-dimensional features, feature-level fusion of multi-sensor data is achieved, and a fusion feature vector with higher characterization capacity is obtained; and finally, inputting the fusion feature vector into a pre-trained BP neural network classifier to realize intelligent identification and classification of different damage types and damage degrees of the hydrogen delivery pipeline. The method has the advantages of being high in detection efficiency and accurate in recognition result, and the accuracy and reliability of hydrogen conveying pipeline damage recognition are effectively improved.

This invention discloses a method for rapid training of neural network models in a small-sample environment, comprising the following steps: S1, acquiring a small-sample dataset containing several categories and preprocessing it to obtain a standardized training sample set, wherein the number of samples in each category of the small-sample dataset satisfies the small-sample distribution characteristics; S2, training an initial neural network classifier based on the training sample set, and constructing prototype point-boundary corridor topology maps for each category in the feature latent space. This invention relates to the field of neural network model training technology. This method for rapid training of neural network models in a small-sample environment, by constructing prototype point-boundary corridor topology maps, can clearly understand the relative positions of each category in the feature latent space and the distribution characteristics of inter-class classification boundaries, thereby accurately identifying boundary gap areas with insufficient sample coverage and prone to misclassification, providing a clear target area for subsequent supplementary sampling operations.

A millimeter-wave radar voice reconstruction and recognition method based on a physical guide network comprises the steps that a millimeter-wave radar is used for transmitting a radio-frequency signal to a to-be-detected target and receiving an echo signal, and meanwhile a reference audio signal is collected; extracting a steady-phase signal Mel spectrum according to the echo signal; generating a simulated radar Mel spectrum by performing an audio signal simulation on the reference audio signal and the common speech data set; synchronizing and standardizing the stable-phase signal Mel spectrum and the simulated radar Mel spectrum, and constructing a voice signal data set; constructing a multi-mode voice reconstruction network model; training a multi-modal voice reconstruction network model according to the voice signal data set; inputting a newly collected real millimeter wave radar signal into the trained multi-mode voice reconstruction network model for voice reconstruction, and outputting a non-contact voice Mel-frequency spectrogram; and inputting the voice Mel spectrogram into the constructed lightweight convolutional neural network classifier, and outputting an identity category label of the speaker.

The invention provides a switched reluctance motor speed regulation system fault diagnosis method based on an adaptive sliding window integration algorithm, and relates to the technical field of motor speed regulation system fault diagnosis. The method comprises the following steps: firstly, collecting current of each phase under different working conditions to construct fault data, extracting features through normalization and harmonic analysis, obtaining fault features through feature selection, and training a plurality of integrated learning classifiers and neural network classifiers; collaborative decision making is sequentially carried out on data to be diagnosed in the graded sliding window, rapid and accurate online identification of single-tube short circuit and winding turn-to-turn short circuit faults is achieved, and the stability and reliability of a speed regulation system are improved.

The application discloses a small sample text classification method fusing data expansion and course learning, first expands data of a labeled sample set by using a strategy, then divides the data set according to a change degree of the expanded sample, then constructs a neural network classifier and trains a classification model, and finally classifies new samples by using the trained model. The learning method is improved and reconstructed, on one hand, data expansion is performed by using multiple strategies, and on the other hand, samples are divided into grades according to change degrees from small to large, and in the learning process, a course learning strategy is used to learn gradually from easy to difficult, and the classification accuracy is significantly improved.

The invention relates to a predictive maintenance method based on industrial equipment multi-mode sensor monitoring. The method comprises the following steps: acquiring multi-source synchronous data in an operation process of industrial equipment; screening events meeting a threshold condition from the multi-source synchronous data to obtain a mutation event list; for each vibration pulse segment in the sudden change event list, searching the associated current sudden change event to obtain an associated tag; performing feature extraction on the vibration signal sequence corresponding to the vibration pulse segment to obtain a pulse feature vector; and inputting the pulse feature vector and the relevance label into a pre-trained neural network classifier to obtain pulse event type probability distribution. By adopting the method, the electromagnetic interference signal and the real fault signal can be effectively distinguished, and the accuracy and reliability of prediction maintenance are improved.

A method includes receiving a user utterance. The method also includes providing the user utterance to a first convolutional recurrent neural network (RNN) classifier and a second convolutional RNN classifier to process the user utterance and provide outputs to a first joint layer. The method also includes providing the user utterance to an automated speech recognition (ASR) model to process the user utterance and provide a text transcript to a text classifier. The method also includes combining the outputs from the first convolutional RNN classifier and the second convolutional RNN classifier using the first joint layer. The method also includes combining outputs from the first joint layer and the text classifier using a second joint layer. The method also includes determining an audio class based on a result from the second joint layer, wherein the audio class indicates whether the user utterance includes speech intended for further processing.

The invention discloses a self-adaptive multi-source credit risk model fusion method and device, equipment and a medium, and relates to the field of artificial intelligence, and the method comprises the steps: selecting a sub-score with the optimal distinction degree as a main score, and carrying out the discretization binning processing of the main score; mapping the discretized main score level into an embedded vector to form an embedded matrix; inputting the embedded matrix into a preset neural network gate controller to obtain a gate control probability vector representing the activation probability of each sub-score; performing sparse processing on the gating probability vector to generate a sparse mask; splicing the main score and all the sub-scores into a score matrix, and performing adaptive selection on the score matrix by using a sparse mask to obtain a mask score matrix; and inputting the mask scoring matrix into a pre-trained neural network classifier to obtain a risk probability prediction result. According to the method, the model discrimination performance and the data use efficiency can be considered at the same time, so that the technical bottleneck faced by a traditional static linear fusion mode is broken through.

The invention discloses an event processing flow recommendation method and system based on deep learning, and belongs to the technical field of artificial intelligence and information technology application, and the method comprises the steps: multi-modal feature extraction and fusion: constructing a comprehensive case feature portrait by using the structured information and unstructured text description of a case at the same time; constructing a process recommendation model based on deep learning: constructing a neural network classifier, learning a complex mapping relation from the fused case features to the optimal processing process, and outputting probability distribution of the recommendation process; and closed-loop feedback learning: the system records artificial decision feedback and actual processing effects and is used for continuous iterative optimization of the model, so that the model has self-learning and evolution capabilities. According to the method, instant and automatic recommendation of the case processing flow can be realized, the manual decision-making time is greatly shortened, and the overall case processing throughput is improved.

The invention discloses a method for discriminating true and false tracks of a radar target based on confidence classification, and belongs to the field of radar target recognition. Extracting five-dimensional features of time difference, displacement distance, speed, acceleration and spatial distance in the temporary track set of the given radar target; constructing a deep neural network classifier based on confidence functions to obtain the confidence functions of real, false and uncertain categories to which each temporary track belongs; selecting a plurality of neighbor decision samples according to the feature distance between the temporary tracks, constructing decision evidences based on a corresponding confidence function, and fusing the decision evidences; dividing two subsets according to each temporary track global confidence function obtained by fusion; and outputting a category discrimination result of the temporary track of each radar target. According to the invention, the deep neural network classifier based on a confidence function improves mining learning of an internal relationship between track features and mode categories; the number of decision samples is supplemented through two-step decision, the recognition accuracy of the temporary track of the uncertain category is effectively improved, and rapid track building and stable tracking of a target are achieved.

The invention discloses a damage sensing system and method for a flexible folding and unfolding solar cell panel, and relates to the technical field of state monitoring of flexible photoelectric equipment. In the system, a flexible electronic sensing layer is arranged between a photovoltaic unit and a flexible substrate; the flexible electronic sensing layer obtains original signals of a plurality of monitoring points on the flexible folding and unfolding solar cell panel; the plurality of monitoring points cover a key stress area of the flexible folding and unfolding solar cell panel; the signal acquisition and preprocessing module acquires a plurality of original signals and preprocesses the plurality of original signals to obtain a plurality of preprocessed original signals; the feature recognition and damage judgment algorithm module extracts a feature index group of each preprocessed original signal, and based on the multiple feature index groups, a weighted center positioning algorithm is combined with a convolutional neural network classifier to determine a detection result. The flexible electronic sensing layer is integrated on the folding and unfolding type flexible solar cell panel, so that structural health monitoring and self-diagnosis of the flexible folding and unfolding type flexible solar cell panel are realized.

The invention relates to the technical field of automobile electronic braking, discloses a high-frequency electric signal feature classification method and system oriented to an EMB system, and aims to improve the operation safety and control accuracy of the EMB system. According to the method, high-frequency electric signals generated during operation of the EMB system are collected through a sensor, continuous electric signal flow is formed and then preprocessed in real time, noise is restrained through a digital filter, the signals are smoothed, and purified electric signals are obtained. And then quality evaluation is carried out on the purified electric signal, a signal-to-noise ratio and a stability coefficient are calculated, an evaluation result is generated, and when the result meets a preset standard, a multi-scale feature extraction process is started. In the process, time domain analysis, frequency domain transformation and statistical feature calculation are carried out on the electric signal to obtain a feature vector, the feature vector is input into a pre-trained neural network classifier for pattern recognition, and a classification label is output. And updating the weight of the neural network classifier on line according to the performance index of the classification label.

The application discloses a fault identification system for a radar liquid level meter for volatile substances and belongs to the technical field of radar liquid level meters. The system is used for solving the problems of fault omission and long fault checking process in the detection and analysis of the faults of the radar liquid level meter. The system comprises the following modules: a return signal extraction module, which adopts two-in wavelet transform to perform wavelet decomposition on the return signal of the radar liquid level meter, adopts principal component analysis to perform decorrelation, adopts a Fisher linear classification function to perform feature extraction, utilizes a BP neural network classifier to classify the features, and outputs a classification result; a return signal fitting module, which is used for drawing a fitting curve of the classification result output by the return signal extraction module and the return time of the radar liquid level meter, and obtaining a return graph of the return curves of different classification results; and a return graph analysis module, which is used for judging whether there is crystallization attachment on the antenna according to the fluctuation value on the return graph. The application is used for detecting the organic crystallization on the horn mouth or the antenna of the radar liquid level meter.

The application discloses a fuel cell hybrid railcar layered energy management method based on adaptive working condition sensing; in the identification layer, a sliding window mechanism is used to extract load working condition time domain and frequency domain features, feature data is clustered based on a spectral clustering algorithm driven by a deep auto-encoder, a data set with a category label is obtained, and a deep dynamic learning vector quantization neural network classifier is trained; in the strategy layer, a double-delay deep deterministic policy gradient reinforcement learning algorithm is used to construct a reward function, a lithium battery SOC fluctuation penalty term limit parameter in the reward function is adaptively adjusted according to a real-time load working condition category output by the identification layer, and an optimal power distribution scheme between multiple fuel cell power generation systems and lithium batteries is obtained by training a reinforcement learning intelligent agent; according to the performance degradation degree of different fuel cell stacks, a distributed collaborative control strategy considering performance differences is used to distribute the output power of each stack, so that the coordinated control of the operating states of the multiple fuel cell power generation systems is realized.

The invention provides an online financial fraud user identification method based on a label noise robust graph neural network, and the method comprises the steps: carrying out the data cleaning and feature processing of original data, and obtaining a multi-modal node feature matrix; constructing a transaction association graph topological structure, and screening an initial clean subset and a noisy subset; performing risk assessment on the noisy subset by using the primary node classifier to obtain a label confidence score; combining the screened high-confidence sample and the initial clean subset to obtain a training sample library; traversing the topological structure of the transaction association graph, and training a classifier to obtain an edge-connected classifier; performing connection edge property prediction and differential weighting on the topological structure of the transaction association graph to obtain a weighted adjacency matrix; and constructing and training a graph neural network classifier, and carrying out online financial fraud user identification. According to the method, misleading of label noise in financial historical data to the model is effectively solved, structural disguise of fraudulent molecules is successfully penetrated, and the problem that a traditional model loses efficacy on a disguise diagram is solved.

The presently disclosed subject matter includes a novel computer-implemented method and computer system for the classification of tabular data using a new neural network classifier model (also referred to herein as “Tabular Neural Network Classifier” or TNNC). The disclosed method and system are characterized by improved accuracy and efficiency, as compared to other existing tabular data classification techniques such as Random Forests, XGBoost, etc. The inventor found that the TNNC exhibits in general a better TP to FP ratio in the classification output and a shorter processing time, as compared to existing tabular data classification techniques.