6109 results about "Feature extraction" patented technology

The embodiment of the invention provides a health assessment method for train bearings. The health assessment method comprises the steps that a bearing temperature data set collected from a plurality of bearings of a train during the travel of the train and an environment temperature data set around the train during the travel of the train are obtained; performing feature extraction on the bearing temperature data set to generate a first temperature feature set; calculating a difference value between each piece of bearing temperature data in the bearing temperature data set and corresponding environment temperature data in the environment temperature data set to generate a first temperature difference value set, and executing feature extraction on the first temperature difference value set to generate a second temperature feature set; and based on at least one of the first temperature feature set and the second temperature feature set, generating an evaluation result of the health state of each bearing in the plurality of bearings.

The application provides a base station traffic prediction method and related equipment, and relates to the field of communication, wherein the base station traffic prediction method comprises the following steps: determining the adjacent base stations of a to-be-predicted base station according to the geographical distance information between base stations; determining the neighbor base stations in the adjacent base stations, and the historical traffic records of the neighbor base stations and the historical traffic record of the to-be-predicted base station have a causal relationship; performing feature extraction on the historical traffic records of the to-be-predicted base station and the neighbor base stations to obtain feature data; and predicting the predicted traffic of the to-be-predicted base station according to the feature data. In the screening process of the feature data object, the spatial factor is fully considered to avoid the singularity of feature data selection, and the causal relationship between the base station traffics is fully considered, so that the blindness of feature data selection is avoided, the redundant feature interference is reduced, the effectiveness of feature selection is improved, and the base station traffic prediction accuracy and efficiency are improved.

This disclosure provides a cross-channel distributed video encoding and decoding method and system based on multidimensional attention, which can be applied to the field of video encoding and decoding technology. The method includes: acquiring a group of video images to be transmitted, the group of video images including a first key reference frame, a second key reference frame, and at least one intermediate video frame; encoding the first key reference frame and the second key reference frame into first encoded data and second encoded data, respectively; encoding each intermediate video frame into third encoded data based on a frame encoder; extracting multi-scale features from each third encoded data based on a multidimensional attention mechanism; processing the multi-scale features into fourth encoded data based on a multi-channel feature extraction mechanism; processing at least one fourth encoded data into a first bitstream; and converting the first encoded data and the second encoded data into a second bitstream, so as to transmit the video image group to the decoder via the first bitstream and the second bitstream.

Embodiments of the present application disclose an object grabbing method, a computer readable storage medium and an electronic device. The method comprises: obtaining an RGB image collected by an image collection device for a target container in a grabbing work area, the RGB image comprising image content related to a plurality of objects stacked in the target container; calling a feature extraction model with the RGB image as input, and determining a target object to be grabbed and extracting feature information of the target object from the RGB image by the model; if the feature information indicates that a physical label is attached to the outer packaging of the target object, and the hardness information of the outer packaging material is not greater than the hardness information of the physical label, determining a region where the physical label is located as a grabbing region; determining a target end effector capable of performing a grabbing operation on the physical label; and controlling an intelligent grabbing device to perform a grabbing operation on the grabbing region of the target object by the target end effector, thereby realizing intelligent grabbing of the target object. This helps to improve the success rate of grabbing, and the grabbing strategy is more universal and compatible.

The application provides a bearing health state online evaluation method and system based on morphological profile analysis and federal evolution hypergraph, aiming at solving the problems of poor model self-adaptability, difficult cross-device knowledge migration and easy to be submerged early weak fault characteristics of the prior art under dynamic working conditions. The method captures the geometric profile evolution of bearing micro-damage by constructing a morphological multi-scale profile feature extraction engine, topologically preserving morphological decomposition of the vibration signal; adopts a Bayesian Poisson online learning algorithm to realize dynamic threshold adaptive updating and early warning of the health index; introduces an evolutionary hypergraph neural network to model the high-order multi-element fault propagation relationship between the bearing and the adjacent components; finally, through a federal edge collaborative framework, the incremental aggregation and knowledge migration of the cross-device model are realized under the premise of protecting data privacy. The application significantly improves the robustness of bearing fault diagnosis under variable working conditions and the sensitivity of early warning, and provides a lightweight and evolving solution for intelligent operation and maintenance in distributed industrial scenarios.

This invention discloses an ophthalmic disease diagnostic system based on optical coherence tomography (OCT), comprising: an image data acquisition module for acquiring OCT image data; a feature extraction and similarity calculation module for extracting OCT image feature representations and layered structural features of the macular region, calculating macular structural similarity information between different OCT image data, and constructing a structural similarity alignment matrix reflecting the structural similarity among all OCT images; simultaneously constructing an association matrix representing the correspondence between OCT images and disease categories; and a model training module that introduces the structural similarity alignment matrix and association matrix as joint supervision signals into the objective function, adaptively learning to obtain an ophthalmic disease diagnostic model; the ophthalmic disease diagnostic model outputs a classification diagnostic result representing the probability that the input OCT image data belongs to various ophthalmic diseases. This invention achieves automatic identification and classification of ophthalmic diseases, with high diagnostic accuracy and good robustness.

The application discloses a 3D cartoon character motion video generation method oriented to a text sketch guide, takes a text prompt and a sketch as explicit input, takes noise input and a time step as implicit input, and correspondingly obtains a text vector and a motion feature; the noise input, the time step, the text vector and the motion feature are taken as input of a diffusion model to obtain noise of a next time step; after denoising by one step, a motion feature obtained from the sketch is used to calculate a key frame posture loss through a regressor, so that the key frame posture is more suitable for the sketch posture; a natural guide module is responsible for adjusting the natural degree of the key frame and adjacent frame postures to obtain denoised noise input; and after T-step continuous denoising, a final motion sequence is obtained. Through sketch feature extraction and sketch control information fusion branches, the application realizes the guiding function of the sketch on human motion sequence generation, and through the natural guide module, realizes the natural and smooth adjacent frame postures and motion diversity in the human motion sequence.

This invention discloses a method for multi-source data fusion and power prediction of wind turbine units, belonging to the field of power prediction technology. It addresses the technical problems of poor multi-source data fusion and power prediction analysis in existing solutions. By filtering and retaining key features through mutual information, redundant information can be effectively reduced. Dynamic weighting via an attention mechanism makes the fused features more adaptable to real-time operating conditions. Differential weight adjustments for wind speed, terrain, and equipment status ensure high relevance of the fused features even in complex scenarios. Spatial feature extraction captures local spatial correlations, temporal features capture temporal dependencies, and random forests handle nonlinear mappings, solving the problem of insufficient generalization ability of single models. The attention mechanism automatically assigns weights to different time steps and features, avoiding complex mathematical processes such as matrix operations and noise covariance estimation.

A machine learning-based user classification method, comprising: acquiring historical water usage data of a water user to be classified; inputting the historical water usage data into a water usage feature extraction model, to obtain a water usage feature of the water user to be classified; inputting the water usage feature into a pre-trained user pair generation model to obtain a user pair, the user pair comprising the water user to be classified and an already classified water user; and determining the category of the already classified water user to be the category of the water user to be classified.

This invention discloses a lightweight method for detecting foreign objects (FOOs) in power transmission lines, relating to the fields of power line inspection and computer vision. A YOLOv10-based FEO detection model is constructed, comprising a feature extraction part, a feature interaction part, and a detection head part. The C2f module in the feature extraction and feature interaction parts is replaced with an expert dynamic extraction module, and the detection module in the detection head part is replaced with a parameter-sharing detection module. The trained FEO detection model is obtained by collecting and evaluating a calibrated FEO dataset of power transmission lines. FEO keyframes are input into the trained FEO detection model to obtain detection results. The expert dynamic extraction module acquires feature maps from multiple receptive fields, achieving rich representation in FEO detection environments with varying scales. The parameter-sharing detection module, based on this, reduces redundant computation and enhances the semantic representation of small targets through parameter sharing and grouping normalization, effectively improving the efficiency of FEO detection in power transmission lines.

This invention relates to an energy-saving method and system based on big data in the power industry, belonging to the field of energy-saving optimization data processing technology. The method includes: accessing data from smart meters and external sources; constructing a data model based on spatiotemporal fusion of multi-source data in the cloud; establishing a dynamic benchmark for the operating scenario based on the data model to group nodes; then, based on the deviation of group evaluation indicators, obtaining a multi-dimensional comprehensive ranking and generating a list of energy-saving renovation nodes; and completing energy-saving optimization through bidirectional parsing of the node list, including upward aggregation of nodes to generate scheduling strategies and downward decomposition of load to identify equipment to optimize user energy consumption. This invention constructs a unified data model through real-time acquisition and encrypted transmission of smart meters and multi-source data; utilizes machine learning clustering to optimize the operating benchmark; performs node grouping and feature extraction; calculates node deviations in real time and generates an energy-saving potential list; and achieves node resource scheduling and optimization of high-energy-consuming equipment through bidirectional parsing.

The application provides a complex scene remote sensing image semantic segmentation method and device, performs multi-scale feature extraction on a feature map of an input remote sensing image, performs splicing and then realizes information interaction between channels; performs down-sampling on the feature map, splices pixel features after sampling in a channel dimension, and extracts key semantic information; the features are flattened into a sequence through PatchEmbed, an encoder module based on a Transformer structure is used to strengthen extraction of micro semantic features; an LWGA attention mechanism is used to perform information interaction on features at different depths between an encoding layer and a decoding layer, image features are extracted and integrated from multiple scales in a channel dimension, and the enhanced features are spliced in the channel dimension to obtain final enhanced features. The application can effectively improve semantic segmentation precision of photovoltaic facilities under a complex building background, reduce boundary adhesion, missing segmentation and missegmentation phenomena, and enhance identification capability for small and dense targets.

This application provides a video processing method, apparatus, computer device, and computer-readable storage medium to improve matching accuracy and robustness in various complex scenarios. The method includes: acquiring a video frame to be processed, comprising a left-eye image and a right-eye image, the video frame being included in the video to be processed; performing feature extraction and stitching on the left-eye image to obtain a first target feature, the first target feature being obtained by stitching together the depth features and local features of the left-eye image; performing feature extraction and stitching on the right-eye image to obtain a second target feature, the second target feature being obtained by stitching together the depth features and local features of the right-eye image; generating a cost matrix based on the first target feature and the second target feature; and generating a first disparity map of the video frame to be processed based on the cost matrix.

This application discloses a training method, apparatus, electronic device, and storage medium for a hair recognition model. The method includes: acquiring a first historical hair image for training, and adjusting the image parameters of the first historical hair image to obtain a second historical hair image; training a constructed feature extraction model using the first and second historical hair images, and extracting features from each hair image in the first historical hair image using the trained feature extraction model to obtain a first image feature, and extracting features from each hair image in the second historical hair image to obtain a second image feature; loading a constructed category recognition model, and training the category recognition model using training data to obtain a trained category recognition model; and fine-tuning the feature extraction model and category recognition model using a third historical hair image obtained from the training data to obtain a hair recognition model. This improves the accuracy of hair recognition.

This invention relates to the field of industrial equipment diagnostic technology and discloses a method for fault diagnosis and tracing of industrial equipment based on multi-source heterogeneous parameter fusion and deep learning networks. The method includes: acquiring a multi-dimensional heterogeneous signal sequence representing the operating state of the controlled object; mapping the signal sequence into a two-dimensional spatial image tensor using the Gram angle field algorithm; constructing a dual-stream feature extraction network containing graph convolution branches and cyclic unit branches to extract spatial topological features and temporal evolution features respectively; determining the logical offset matrix based on the spatial flow feature gradient divergence, and resampling the temporal flow features accordingly to achieve spatiotemporal causal phase alignment, and outputting diagnostic analysis results. This invention compensates for the physical system response hysteresis through asynchronous resampling operators in the computational domain, eliminates feature phase contamination caused by conventional rigid alignment, enhances the model's ability to capture weak abnormal signals, and achieves causal consistency in fault root cause localization.

The application discloses a motor imagery electroencephalogram signal denoising method and device, medium and product, relates to the technical field of deep learning and biomedical signal processing, and the method comprises the following steps: acquiring a target electroencephalogram signal containing artifacts; inputting the target electroencephalogram signal containing artifacts into a trained electroencephalogram denoising model to obtain a final denoised electroencephalogram signal; wherein the electroencephalogram denoising model comprises an electroencephalogram denoising branch, an artifact prediction branch and an artifact representation interaction attention fusion reconstruction module; the electroencephalogram denoising branch comprises a multi-scale self-adaptive enhancement module, a frequency domain dynamic enhancement module and a feature extraction module. The application solves the problems of traditional electroencephalogram denoising methods in signal fidelity, spectral fidelity, spatial structure preservation and generalization ability.

The present application relates to fruit detection technology field, specifically to a kind of fruit maturity real-time detection method and system, comprising the following steps: obtaining channel image signal, extract gray variation trend, judge area enhancement condition, track edge drift direction, analyze time extension characteristics, compare gray jump synchronism, identify light and dark variation trend, screen structure coordination and luminance rising section, output maturity state sequence.In the present application, by tracking the channel luminance enhancement trend in image frame, combined with edge profile drift direction and time extension characteristics, extract the persistence of structural change, measure the degree of area symmetry deviation according to the distribution of gray jump point, combined with the light and dark change direction of gray mean value to construct the reflection change trend, screen the image frame sequence interval with structure stability and luminance rising synchronous appearance, extract the paragraph with time continuity and regional attribute coordination, complete the segmented response and feature induction of fruit surface maturity state change process.

This invention provides a driver state monitoring and intelligent interaction method based on multimodal data, belonging to the field of intelligent driving assistance technology. The method includes: simultaneously acquiring the driver's electroencephalogram (EEG) signals, at least one other physiological signal, and vehicle state data to form a multimodal data stream; preprocessing and extracting features from the multimodal data stream to obtain a multimodal feature vector for state recognition; inputting the multimodal feature vector into a preset multimodal state recognition model to generate a recognition result characterizing the driver's current state; and triggering corresponding intelligent interactive feedback or vehicle control commands based on the recognition result. This invention, through multimodal signal fusion, comprehensively assesses the driver's state from three dimensions: neurophysiology, behavioral performance, and vehicle control, significantly improving the accuracy and reliability of state recognition.

A hybrid forecasting system (100) for tiered cloud pricing using ensemble learning, comprising: a data ingestion module configured to continuously ingest and aggregate heterogeneous data from a variety of sources, including historical cloud usage data, real-time resource consumption metrics, customer subscription profiles, service-level agreement parameters, and external demand indicators; a preprocessing engine functionally coupled with the data ingestion module, the preprocessing engine being configured to perform data cleansing, normalization, transformation, feature extraction, and dimensionality reduction to generate structured and model-compatible datasets;a model training unit that is functionally coupled with the preprocessing engine, wherein the model training unit comprises a variety of heterogeneous predictive models, including at least one statistical model, at least one machine learning model, and at least one deep learning model, each configured to process the structured datasets independently to generate predictive results that meet future cloud resource needs, workload variability, and price sensitivity across multiple service tiers;an ensemble aggregation layer that is functionally coupled to the model training unit, wherein the ensemble aggregation layer is configured to receive and combine the forecast results generated by the multitude of forecasting models using ensemble learning techniques, including weighted averaging, stacking or boosting, with the weights assigned to each forecasting model being dynamically adjusted based on predefined performance evaluation metrics to generate a uniform and optimized forecast;a price optimization engine that is functionally connected to the ensemble aggregation layer, wherein the price optimization engine is configured to determine and dynamically adjust tiered price structures based on the unified forecast, taking into account parameters such as forecasted demand, user segmentation, demand elasticity, infrastructure capacity constraints, and predefined optimization goals such as revenue maximization and resource utilization efficiency;and a feedback adjustment module that is functionally coupled with the price optimization engine and the model training unit, wherein the feedback adjustment module is configured to monitor system performance in real time, user response to price adjustments and resource utilization results, and iteratively updates model parameters and pricing strategies using adaptive learning mechanisms, including reinforcement learning, wherein the system (100) is configured to operate in a multi-tenant cloud environment, supports real-time data processing and decision-making, and enables automated, scalable, and adaptive optimization of tiered cloud pricing.

A device for the autonomous detection of cyber threats, consisting of: a housing that encloses a multitude of interconnected hardware components; a network interface unit configured to receive and send data packets from one or more communication networks; a data acquisition unit that is operationally connected to the network interface unit and configured to capture packet-level data, metadata, and system event logs; a preprocessing processor configured to analyze captured data, decodecode protocols, reconstruct communication flows, and generate structured data representations; a feature extraction processor that is operationally coupled with the preprocessing processor and is configured to calculate statistical, temporal and entropy-based features from the structured data representations; a storage unit consisting of volatile memory for real-time processing and non-volatile memory for storing historical data and learned patterns; an inference processor that is operationally coupled with the feature extraction processor and the storage unit, wherein the inference processor is configured to execute a variety of trained models to identify anomalous behavior based on deviations from stored patterns; a classification unit that is operationally coupled with the inference processor and configured to assign detected anomalies to one or more threat categories based on calculated confidence values; a response control unit configured to generate and transmit remedial actions, including blocking network traffic, isolating network segments, and terminating suspicious processes; and a control processor configured to coordinate the data flow between the network interface unit, the data acquisition unit, the preprocessing processor, the feature extraction processor, the inference processor, the classification unit, the response control unit, and the storage unit, with the device operating autonomously to detect and respond to cyber threats in real time.

This invention belongs to the field of protocol conversion communication and avionics data communication technology, specifically a smart prefetching system and method based on network-to-1553B stream characteristics. The system includes a stream feature extraction module, a feature analysis module, a prefetching decision module, a prefetching execution module, and a prefetching feedback module. The stream feature extraction module extracts multi-dimensional feature information from the network-to-1553B data stream; the feature analysis module establishes a data stream behavior prediction model; the prefetching decision module uses a multi-strategy fusion mechanism to calculate the optimal prefetching strategy; the prefetching execution module completes data prefetching and protocol conversion; and the prefetching feedback module monitors the prefetching effect and optimizes the model. This invention targets the protocol characteristics of the 1553B bus, improving data transmission efficiency through intelligent prefetching, and is suitable for network-to-1553B data conversion scenarios in avionics systems.

The application provides a spatial perception enhancement and pathological image classification method for processing class imbalance, aiming to enhance the model's perception of spatial information and process class imbalance. The specific steps are as follows: in the preprocessing stage, the image is segmented, the background is filtered out, and the image block is cut, and the coordinates are recorded; in the feature extraction and fusion stage, the pathological features are extracted through a pre-trained convolutional model, then the coordinates are normalized to form position features, and the pathological features and position features are fused to form comprehensive features with spatial information. In the reasoning stage, a multi-scale feature enhancement network is introduced, which can capture local features through convolution operations and model spatial relationships through attention mechanisms, thus achieving more comprehensive processing of local and spatial information. Finally, the parallel classifier is used to solve the problems of class imbalance and sample shortage, and the final pathological classification result is obtained. This method can better assist pathologists in improving the diagnosis accuracy and reducing subjective differences in the diagnosis of chondroma.

The application discloses a kind of hearing-impaired person intelligent terminal adaptive feedback method, intelligent terminal and system, it is related to smart home and barrier-free interaction technical field.The present application is aimed at the defects of single feedback mode, lack of intelligent scheduling and no adaptive capacity, constructs an end-to-end intelligent feedback scheme.First, the user situation characteristics are collected in real time by multidimensional sensor, and the situation characteristic vector is constructed;At the same time, the importance score of visitor is calculated by extracting the multi-modal features of visitor.Secondly, the situation characteristics and visitor characteristics are input into the deep Q network, and the optimal multi-modal feedback device combination and feedback intensity level are intelligently decided.Then, the graded response strategy is executed, and the reminder intensity is dynamically adjusted based on the user response.Finally, the decision model is continuously optimized through online learning mechanism.The present application realizes the leap from passive fixed reminder to active intelligent perception feedback, significantly improves the use experience and efficiency of hearing-impaired person intelligent terminal.

Systems and methods for evaluating progression of an anatomical object over a plurality of timepoints are provided. Longitudinal medical images of an anatomical object of a patient acquired over a plurality of timepoints are received. For each respective timepoint of the plurality of timepoints, features are extracted from the longitudinal medical images acquired at the respective timepoint using a machine learning based feature extractor network and the anatomical object in the longitudinal medical images acquired at the respective timepoint is analyzed based on the extracted features using a machine learning based prediction model. Progression of the anatomical object over the plurality of timepoints is evaluated based on results of the analyses using a machine learning based progression model. The evaluation of the progression of the anatomical object over the plurality of timepoints is output.

The application provides an air traffic control primary radar signal processing automatic test method and system, relates to the radar test field, and solves the technical problems of fragmentation of the prior art process, single evaluation dimension and poor traceability. The method specifically comprises: obtaining radar test data and structured test configuration; recursively analyzing the test configuration to generate an instruction set; loading data and executing signal replay according to the instructions to drive the software; synchronously collecting video data and processing results of each stage to form a multi-dimensional feature sequence; correlating and analyzing the two and extracting features to generate a holographic feature vector sequence; progressively comparing the sequence with a baseline determination model in layers to obtain a test determination result; integrating test item identifiers, video data, multi-dimensional feature sequences, holographic feature vector sequences and test determination results to record and generate a final test log report. An end-to-end automatic, multi-dimensional deep analysis and full-link evidence tracing radar automatic test method is realized.

This invention discloses a parametric sketch-driven method for the deconstruction and reassembly assembly design of metal tubular furniture, belonging to the field of digital furniture design and intelligent manufacturing technology. The method includes: decoupling the furniture structure into multiple sheet-like substructures and establishing a parametric template library; extracting primitive data and determining connection types by identifying the geometric and topological features of user-defined 2D sketches; automatically matching and driving the templates to generate 3D substructures, completing the process design of connection details; calculating assembly positioning benchmarks and contact areas based on the spatial relationships between sketches, generating connection features to achieve automatic assembly; and finally, performing dimensional verification and closed-loop updates on the assembly. The entire process requires no manual intervention and can output a production-ready digital model and processing data within minutes, significantly improving the efficiency, accuracy, and manufacturability of customized metal tubular furniture design.

The disclosure provides a multi-dimensional time series anomaly detection method based on decoupled representation learning. For the multi-dimensional time series collected by multiple sensors in an industrial system, the multi-dimensional time series is input into a multi-dimensional time series anomaly detection model based on decoupled representation learning, and a judgment result for an abnormal device is output. The implementation mode of the multi-dimensional time series anomaly detection model based on decoupled representation learning is that, for input data, first, a heterogeneous hybrid expert network is used for multi-scale feature extraction to adaptively capture global and local patterns; then, a decoupled double-decoder architecture is designed to separate normal representation learning and abnormal discrimination tasks; finally, a special training method is designed to dynamically generate pseudo abnormal samples adapted to the training progress during the training process, and an abnormal sensitivity enhancement strategy based on data disturbance is introduced to gradually introduce a data disturbance mechanism to dynamically enhance abnormal sensitivity. The problem that single-scale modeling cannot comprehensively describe time series features is solved.

The application discloses a kind of action strategy generation method, system, equipment and medium of humanoid robot, wherein, according to the end-to-end big model, the multi-source data set is extracted and fused with different characteristics, to obtain several different types of single modal features and target cross-modal features;According to the degree of freedom general coefficient and the joint coordination weight coefficient, all the single modal features and the target cross-modal features are calibrated, to obtain calibration features;According to the domain knowledge graph, the calibration features are enhanced with feature knowledge, to obtain knowledge enhanced features;According to the knowledge enhanced features, strategy generation processing is carried out, to obtain the target action strategy of the humanoid robot in the current work stage.The method can effectively improve the generation quality of humanoid robot action strategy.The application relates to the field of artificial intelligence technology.

The application discloses a kind of power transmission line fault diagnosis methods based on fault data and text label multimodal fusion.The application is based on the adaptive window signal interception method of wavelet transform, detects traveling wave head mutation point by continuous wavelet transform, and accurately captures high-frequency transient characteristics by dynamically adjusting the interception window according to signal attenuation rate, effectively suppresses noise interference;The normalized signal is input into the Transformer architecture model together with the associated text data, the signal visual features are extracted by Vision Transformer, the text semantic features are extracted by BERT, and the cross-modal attention mechanism is used to realize feature deep fusion, and finally the fault type is output.The application ensures that the multi-modal fusion model maintains high generalization performance in complex environments across regions and seasons, accurately diagnoses the fault type, and ultimately supports second-level fault early warning, minute-level accurate positioning and natural disaster diagnosis decision, significantly reduces manual intervention and improves power grid operation efficiency.