78results about How to "Fully excavated" patented technology

The invention relates to a federated learning-based magnetocardiogram analysis method, device and equipment and a medium, and relates to the technical field of magnetocardiogram analysis, and the method comprises the following steps: aiming at a preset arrhythmia analysis task, configuring a corresponding independent task head on a frozen shared feature backbone network, and caching a corresponding reference logic value; and configuring a corresponding new task head for the new task, and inserting a low-rank adapter module into the frozen shared feature backbone network so as to perform joint optimization on the inserted low-rank adapter module, the new task head and the independent task head, updating parameters of the low-rank adapter module, parameters of the new task head and parameters of the independent task head, according to the method, the limitation of a data island is broken through, data resources of multiple centers are integrated, massive unlabeled data can be fully mined, the model feature extraction capability is improved, the situation that new and old data are used for retraining the model is avoided, and the safety of the federal continuous learning model is improved. And the old category diagnosis capability is ensured.

The invention discloses a combined preprocessing and fusion decoding method based on electroencephalogram signals and functional near-infrared signals, which comprises the following steps of: synchronously acquiring the electroencephalogram signals and the functional near-infrared signals, taking a task trigger event as a unified time reference, and carrying out basic preprocessing of time alignment and modal self-adaption on two modal signals to obtain a functional near-infrared signal; a pre-trained cross-modal noise joint modeling module is utilized to perform joint modeling on cross-modal joint noise features caused by a common noise source, collaborative denoising processing is performed on multi-modal signals based on the cross-modal joint noise features, multi-modal denoising representation is obtained, a cross-modal feature interactive modeling mode based on an attention mechanism is obtained, and the multi-modal noise is obtained. And dynamically modeling the correlation between different modal features, realizing deep fusion of multi-modal complementary information, carrying out decoding processing based on the fused features, and outputting a corresponding brain-computer interface control instruction or task identification result. According to the method, the decoding accuracy and robustness of the brain-computer interface system in a complex task scene are enhanced.

The invention discloses a two-stage multi-mode bearing fault diagnosis method based on a pre-training large model, and belongs to the technical field of bearing fault diagnosis. The method aims at solving the problems that a traditional method is poor in generalization and poor in robustness under multiple working conditions and small sample conditions. The method comprises the following steps: firstly, constructing a learnable multi-modal Tokens which comprises a multi-scale patch Token, a feature Token and a fault Token, and realizing efficient extraction and fusion of multi-modal features; a time-frequency semantic fusion module is introduced, and comprehensive time-frequency features are output through adaptive frequency coding, time coding and multi-modal fusion; and inputting the multi-modal feature sequence into a pre-training BERT model, and adopting a two-stage training strategy, in the first stage, performing self-supervised pre-training by taking mask signal reconstruction as a target, and in the second stage, performing parameter fine tuning by taking fault classification as a target. According to the method, the diagnosis accuracy and the cross-working-condition generalization ability under the small sample condition can be remarkably improved.

The invention discloses a perception method based on multi-semantic space attention and Daubechies wavelet double-branch Mama, and belongs to the field of hyperspectral image processing. The problem that in the existing Mama-based model feature extraction process, the capacity of capturing multi-scale local structures and direction sensing information is insufficient is solved. The method comprises the following steps: inputting a hyperspectral image; projecting the spectral vector to an embedding space through an embedding layer to obtain an embedding feature; inputting the embedded features into an encoder, wherein the encoder comprises an SMSAMama branch, a DWTMama branch and a self-adaptive feature fusion module; the SMSAMama branch is used for extracting spatial features; the DWTAMba branch is used for extracting spectral features; the adaptive feature fusion module performs weighted integration on the spatial features and the spectral features by using randomly initialized fusion weights; and inputting the integrated features into a segmentation head to generate a final perception result. The method is used in agricultural monitoring and urban planning fields.

The application discloses a generalized shared energy storage optimization configuration method and system based on fuzzy chance constrained programming, relates to the technical field of electric power, considers the operation characteristics of various generalized energy storage resources, and respectively models the energy storage, wherein the base station energy storage model needs to consider communication load communication quality and base station power supply reliability, the air conditioner model needs to consider user comfort, and the electric vehicle charging station model needs to consider user driving characteristics, so that the existing large number of idle multi-energy generalized energy storage resources are fully tapped, and the generalized shared energy storage optimization configuration model considering multiple uncertainties is designed. The shared energy storage operator and the user group form a cooperative alliance by transferring the energy storage use right, the generalized shared energy storage optimization configuration model is decomposed into two sub-problems of alliance energy consumption cost minimization and internal payment negotiation based on Nash bargaining theory, and the risk brought by the parameter uncertainty of the user group source and load output and virtual energy storage is quantified based on the fuzzy chance constrained programming theory.

The disclosure provides a fault diagnosis method and device and a prototype acquisition method in a model, and relates to the technical field of fault diagnosis. The method comprises the following steps: acquiring business data; performing first processing on the business data to obtain first feature data and second feature data corresponding to each first index and a plurality of time windows; inputting the first feature data into a first group of network layers of a Gaussian prototype network model to obtain third feature data; inputting the second feature data into a second group of network layers of the Gaussian prototype network model to obtain fourth feature data; splicing the third feature data and the fourth feature data to obtain fifth feature data; inputting the fifth feature data into a third group of network layers of the Gaussian prototype network model to obtain sixth feature data; and performing fault diagnosis according to the sixth feature data and a prototype of the Gaussian prototype network model.

The invention discloses a depression detection method based on EEG channel selection and multi-dimensional feature fusion, which avoids the problem of sensitivity of k-means to the clustering number and initial clustering centers, improves the k-means, determines the number of clustering centers by adopting a Karlinski-Halabass criterion, calculates the PageRank value of an EEG channel based on maximum and minimum similarity, and finally determines the number of the clustering centers by adopting the Karlinski-Halabass criterion and the PageRank value of the EEG channel based on the maximum and minimum similarity. Selecting the first k channels as initial clustering centers, designing a self-adaptive threshold model, and selecting channels with distances smaller than a threshold as key EEG channels; constructing a brain function network and a super-brain function network by using the key EEG channel, and extracting low-dimensional time domain features of the key EEG channel, low-dimensional spatial domain features of the brain function network and high-dimensional spatial domain features of the super-brain function network; and designing a multi-dimensional feature fusion strategy based on standard deviation, and fusing low-dimensional time domain features, low-dimensional spatial domain features and high-dimensional spatial domain features to realize depression detection with high accuracy.

The application provides an edge device-oriented lightweight multi-modal early gastric cancer collaborative auxiliary diagnosis system, which comprises a lightweight multi-modal classification module, a decision-level fusion unit, a first feature extraction network and a second feature extraction network; the decision-level fusion unit is used for evidence theory fusion of first classification probability and second classification probability, and outputs a multi-modal classification result; a lightweight segmentation module is constructed based on an encoding-decoding structure, an encoder of the lightweight segmentation module comprises a plurality of stacked lightweight bottleneck modules, a feature enhancement module is arranged in a skip connection of the encoder, and the feature enhancement module is used for pixel-level lesion segmentation of a white light endoscopy image; a gating cooperation unit is connected with the lightweight multi-modal classification module and the lightweight segmentation module respectively, and is used for generating a gating signal according to the multi-modal classification result; when the multi-modal classification result is positive, the lightweight segmentation module is activated to perform a segmentation task; and when the multi-modal classification result is negative, a subsequent segmentation process is terminated.

The application discloses an operating room active data collection platform based on artificial intelligence, which comprises the following steps: capturing multi-class visual information through multiple cameras and an operating field camera, identifying instruments, anatomical structures and lesion areas through a deep learning model, tracking surgical steps and personnel information and evaluating collaboration efficiency, fusing video, audio and equipment data streams to construct a space-time graph, forming a structured time axis through pattern recognition, constructing a multi-dimensional knowledge graph based on massive data, realizing data deep correlation and semantic understanding, breaking through the barriers of different equipment data formats through a multi-protocol compatible interface and format adaptation technology, solving the information island problem, replacing the traditional identification mode which relies on expert experience, improving the comprehensiveness, accuracy and efficiency of surgical data processing, and providing strong support for medical quality optimization, surgical teaching and medical research.

The invention discloses a TPA-LSTM-based data processing method and device, and relates to the technical field of data processing, and the method comprises the steps: collecting a historical data set and an actual data set of energy-consuming equipment in an industrial park, and carrying out the convolution of the historical data set and the actual data set, and obtaining a convolution data set; performing dimension matching on the historical data set, the actual data set and the convolution data set based on a preset hybrid convolution completion method to obtain a historical data sequence and an actual data sequence; training the historical data sequence and the actual data sequence based on a preset TPA model to obtain fusion data, and optimizing the fusion data based on a preset LSTM model to generate a prediction data sequence; and acquiring an actual data value corresponding to the sampling time point of the prediction data sequence, determining the accuracy of the prediction data sequence based on the actual data value, smoothing the prediction data sequence based on the accuracy, obtaining target data, and completing data processing.

The invention belongs to the field of artificial intelligence networks, and particularly relates to a method for performing visual questions and answers by using an attention mechanism from words to regions. Comprising the following steps: (1) extracting features; and (2) generating candidate answers: firstly positioning related image areas and keywords in questions by adopting a collaborative attention mechanism, then obtaining fine-grained image features and question features, and finally fusing the two features to generate the candidate answers. According to the method, in order to generate candidate answers with higher quality, two question and answer stages are cascaded, a traditional single-stage visual model is expanded into a double-stage model, semantic information contained in the answers is fully mined, and accurate prediction of the final answers is promoted. According to the invention, image areas and keywords related to questions can be extracted and utilized, so that more accurate candidate answers can be generated.

The invention relates to the technical field of dissolved oxygen prediction, and provides a dissolved oxygen prediction method based on data decomposition and related equipment, and the method comprises the steps: obtaining a plurality of water quality indexes of a target water body, and determining a plurality of target water quality indexes from all the water quality indexes; according to all the target water quality indexes, generating a dissolved oxygen data sequence of the target water body, and decomposing the dissolved oxygen data sequence to obtain a trend component and a seasonal component; predicting the trend component by using a trend prediction model to obtain a first dissolved oxygen predicted value, and predicting the seasonal component by using a seasonal prediction model to obtain a second dissolved oxygen predicted value; and fusing the first dissolved oxygen predicted value and the second dissolved oxygen predicted value to obtain a final dissolved oxygen predicted value of the target water body. According to the method, the dissolved oxygen prediction accuracy can be improved.

The application discloses a low-gas mine gas emission abnormal source precision detection and treatment method, which is used for solving the problem of low-gas mine gas abnormal emission. First, the monitoring points and the monitoring point positions are determined, and all gas monitoring data are obtained through a mine monitoring system; data analysis is carried out based on SPC control chart theory, and a gas emission data anomaly criterion is established; the gas emission state is judged through the SPC control chart method, the gas emission abnormal area position is determined, and early warning is carried out; according to the determined gas emission abnormal area, the isotopic analysis method is used to detect the gas emission source and composition; finally, the gas emission abnormal source analysis result is tracked, targeted gas source classification treatment measures are formulated, and then the treatment effect test is carried out. The application can effectively control the low-gas mine gas emission abnormal risk, thereby improving the safety of coal mining.

The application relates to the technical field of wireless sensing, and discloses a human posture estimation and non-contact vital sign monitoring method and system based on CSI. WiFi channel state information (CSI) is collected, and dynamic CSI feature vectors are obtained through preprocessing. Features are mapped to human key point feature space through Transformer cross attention, spatial reasoning is performed through graph convolutional neural network fusion of human skeleton topology constraints, and human posture estimation results are output. Respiratory rate and heart rate are extracted based on multi-band frequency domain analysis, and vital sign monitoring results are output after being optimized in combination with posture information. Without cameras and wearable devices, non-contact human perception that is privacy-safe and wall-penetrating can be realized, and the method has the advantages of high precision, strong generalization and low-cost deployment, and is suitable for smart home, medical monitoring, emergency rescue and other scenes.

The application is a medical image segmentation method based on multi-modal self-supervision. First, the multi-modal medical image of the lesion tissue is obtained, including A-mode image and B-mode image, and the image is preprocessed. Then, a cycle-consistent modal contrast domain translation network is constructed, including two generators and two discriminators. The generator is used to convert the image of one mode into the image of another mode, including an encoder, an intermediate shared module and a decoder. The discriminator is used to judge the source of the input. Then, the cycle-consistent modal contrast domain translation network is pre-trained, the training loss is calculated, and the loss function includes multi-modal semantic consistency loss, adversarial loss, cross-domain translation loss and cycle consistency loss. Finally, the A-mode segmentation network and the B-mode segmentation network are constructed, the pre-trained weights are migrated to the two segmentation networks, and the trained two segmentation networks are respectively used for medical image segmentation of the corresponding mode. The contrast cross-domain translation is used as a multi-modal self-supervised pre-training task to learn more comprehensive modal features, promote the network to better learn modal characteristics and common knowledge, and improve the segmentation ability.

This invention discloses a method and system for geological hazard identification based on multimodal semi-supervised learning. The method includes the following steps: acquiring labeled and unlabeled orthophotos and elevation models of geological hazard areas; constructing teacher and student networks containing dual-branch encoders, extracting features from the orthophotos and elevation models respectively, and linearly fusing them, generating ensemble predictions using learnable dynamic weights; in semi-supervised training, calculating the uncertainty of the student network's prediction results in real time, and triggering an exponential moving average update of the teacher network using the student network parameters only when the decrease in uncertainty compared to the historical mean exceeds a preset evolution threshold; optimizing the student network parameters by combining supervised loss, confidence-weighted consistency loss, and student historical consistency loss; inputting the test data into the trained student network, and outputting fine-grained geological hazard segmentation results. This invention can achieve high-precision automatic identification of geological hazards under small sample conditions.

This disclosure provides an entity classification method, apparatus, electronic device, and storage medium. The method includes: acquiring multiple entity relationship networks corresponding to multiple entity nodes; the same entity node corresponding to different entity relationships in different entity relationship networks; performing multi-label prediction on the multiple entity relationship networks based on multiple trained base learners to obtain the multi-label prediction result of each entity node in each entity relationship network; and determining the final multi-label prediction result corresponding to each entity node based on the multi-label prediction result of each entity node in each entity relationship network. This disclosure, from the perspective of multi-view learning, performs multi-label prediction based on multiple entity relationship networks, which can more fully explore the relationships between entities, making the multi-label prediction results for entity nodes more accurate.

This invention belongs to the field of intelligent transportation technology, specifically relating to a traffic situation analysis system based on holographic intersections. It includes edge devices, a data access module, an intersection modeling unit, a data analysis module, a data storage module, and an anomaly backtracking module. The edge devices collect cross-sectional flow, speed, queue length, vehicle trajectory, and traffic signal control data; the data access module receives this information. The intersection modeling unit constructs a detailed 3D digital model including lanes, stop lines, zebra crossings, and other elements. The data analysis module calculates multiple situation indicators such as intersection overflow degree, traffic imbalance index, and congestion degree based on traffic data and the model. The data storage module stores traffic data, the model, and records of anomalies exceeding thresholds. The anomaly backtracking module can trace back to the state before the anomaly occurred based on the anomaly records, supporting the retrieval of monitoring videos, trajectory data, and the situation results at the corresponding time, assisting in the analysis of the causes of the anomalies.

This invention discloses an autonomous unmanned aerial vehicle (UAV) inspection operation monitoring and analysis system, belonging to the field of UAV inspection operation analysis technology. The invention processes collected railway track image data, railway catenary thermal imaging data, and 3D point cloud data along the railway line to obtain visual indices, functional indices, and structural indices. The inspection result generation module compares these indices with thresholds and performs corresponding operations, such as marking the priority of defect maintenance or structural maintenance, and pushing relevant data to maintenance personnel. This fully leverages the value of the data, provides guidance for optimizing subsequent inspection work, and solves the problem of existing technologies lacking efficient analysis and processing mechanisms, thus failing to fully exploit the value of the data.

The embodiment of the application discloses a grid water depth prediction method and device based on an LSH attention mechanism, and relates to the field of hydrological monitoring. The method disclosed by the application first collects B batches of boundary flow time series data and grid water depth time series data of a target basin output by a two-dimensional IFMS model; then, standardizes the collected data, constructs a first feature tensor and transposes the first feature tensor into a second feature tensor; then, a weighted feature tensor is calculated based on an attention algorithm by using a Reformer model; finally, the weighted feature tensor is decoded by using a pre-trained iTransformer to obtain prediction values of grid water depths at future time points. The application can realize cross-time period prediction, improve prediction efficiency and accuracy, and provide a reliable basis for flood control and disaster reduction.

The application provides a dynamic contrast-enhanced magnetic resonance image classification method, system, device and medium, the method comprising: acquiring breast dynamic contrast-enhanced magnetic resonance images of different disease ages and preprocessing to generate a breast dynamic enhancement magnetic resonance image dataset; according to the breast dynamic enhancement magnetic resonance image dataset, an image classification prediction model comprising a multi-task classification network and a perception discrimination network connected in turn is constructed; the acquired breast dynamic contrast-enhanced magnetic resonance image to be classified is input into the image classification prediction model for multi-task classification, and an image classification result comprising the position, spatial structure and spatial signal of the tumor is obtained. The method can fully and accurately mine the rich multi-dimensional, multi-scale heterogeneous space, time and semantic correlation representation possessed by the breast contrast-enhanced magnetic resonance image data, and can supplement the missing different spatial dimension information, ensuring the reliability and accuracy of image classification prediction.

The embodiment of the invention provides a photovoltaic output information generation method and device, equipment and a storage medium. The method comprises the steps that meteorological factors are screened based on correlation analysis, a comprehensive meteorological factor curve is generated through dimensionality reduction, and a photovoltaic output curve is synchronously generated. Five types of results of cloudy days, cloudy days, sunny days, sunny-to-cloudy days and multi-cloud-to-sunny days are extracted through three-stage clustering, an error distribution estimation model is constructed based on historical data of each scene, a classification historical data set is used for training a Bayesian neural network, an integrated model is constructed, a path is matched according to input data, and an error distribution estimation model is constructed; and inputting the generated comprehensive curve into an integration and error model, and outputting a photovoltaic daily output curve with a confidence interval as final information. Refining input with high reference value is provided for time sequence production simulation, and the refinement degree and the uncertainty quantification capability of output prediction are improved.

The application discloses a device identification method and device, apparatus and storage medium, and belongs to the technical field of computers. The method comprises: obtaining device feature information of a to-be-identified device; performing feature embedding processing on the device feature information to generate overall feature information of the to-be-identified device; determining cross feature information of the to-be-identified device based on the overall feature information; and determining that the to-be-identified device is an Internet service device in the case that the overall feature information and the cross feature information meet a target condition. In the technical scheme of the application, high-dimensional discrete device features are embedded into low-dimensional dense overall features, the device features are more efficiently represented and the amount of calculation is reduced under the premise of ensuring the completeness of the feature information, then the cross features between the features in the overall features are determined, the feature information is fully mined, and then the type of the to-be-identified device is determined according to the overall features and the cross features, so that the efficiency of identifying the Internet service device is improved, and the identification accuracy is also improved.

The invention relates to a Bluetooth AOA signal outlier elimination method and system based on deep learning, and the method comprises the steps: initializing system configuration, and cooperatively collecting Bluetooth signals through employing a plurality of Bluetooth base stations; acquiring IQ data in a plurality of Bluetooth base stations, and performing data preprocessing; according to the IQ data, performing calculation to obtain a preliminary positioning result of the Bluetooth terminal; constructing an input data set of the deep learning model and designing a network structure of the deep learning model; training a deep learning model according to the input data set; the training precision of the deep learning model is judged, if the training precision meets the requirement, IQ data collected in real time is input into the trained deep learning model, and a real-time positioning result of the Bluetooth terminal is obtained; otherwise, returning to adjust the network structure; and optimizing the real-time positioning result by using a Kalman filtering algorithm to obtain a final positioning result. Compared with the prior art, the method has the advantages of high positioning accuracy and strong robustness.

The invention provides an intelligent tumor auxiliary diagnosis system and method based on deep learning, a medium, a terminal and a program product. The method comprises the steps that pathological section images and multiple single-mode radiology images of tumor tissue are acquired and preprocessed; constructing a pathological encoder for outputting pathological features according to the pathological section image; training according to various single-mode radiology images to obtain a single-mode diagnosis model, and taking the output pathological features as semantic reference in the training process; according to multiple single-mode radiology images, performing fusion and training to obtain a multi-mode diagnosis model, and performing shielding mode input based on a random mode shielding strategy in the training process; and deploying the single-mode diagnosis model and the multi-mode diagnosis model to output a tumor diagnosis result. According to the method, the single-mode image can be independently processed, and the multi-mode image features can be fused, so that complementary information among different modes can be fully mined, and the accuracy of noninvasive tumor diagnosis and the clinical diagnosis efficiency are improved.

The application provides a kind of overhead power line surveying and mapping image key area identification processing method and system, wherein the method comprises: obtaining the original image data collected by camera and the geographic position data of camera shooting position;Match the geographic position data with the preset geographic position database, obtain the natural material prior probability data set corresponding to the camera shooting position;Extract polarization feature and image content feature from the original image data, use double-branch attention network, establish material optical property table based on polarization feature in the first branch processing path, fuse natural material prior probability data set and image content feature in the second branch processing path;Based on material optical property table and fusion result, identify key area from original image data;Conduct confidence assessment on key area, generate key area confidence map.The application improves the identification accuracy of natural material key area in outdoor complex scene.

This invention discloses a fault diagnosis method for wind turbine turntable bearings based on WDCNN-Transformer, belonging to the field of wind power equipment condition monitoring and intelligent fault diagnosis technology. This invention proposes a novel fault diagnosis method by combining dual-branch time-frequency domain parallel feature extraction with the serial structure of WDCNN-Transformer and time-frequency domain feature fusion. This method effectively improves the accuracy and robustness of wind turbine turntable bearing fault diagnosis. It utilizes both time and frequency domain branches simultaneously, automatically extracting features using deep networks to ensure comprehensive mining of signal information from multiple perspectives. The concatenation of WDCNN and Transformer improves the utilization efficiency of temporal and spatial features. Feature fusion fully leverages the complementary advantages of time-frequency information, enhancing the discriminative power of fault features.

The invention discloses a clothes changing pedestrian re-identification method based on graph attention human body part dynamic correlation modeling. The method comprises the following steps: acquiring a video sequence of a pedestrian to be identified, and performing part segmentation through a human body analysis model; a part perception multi-scale feature extraction module is used to extract the features of each part under different receptive fields and suppress the background; then, the combination of each frame and the part is constructed into a graph node, and a heterogeneous space-time part graph containing a space edge, a time edge and a non-local edge is constructed; performing message passing on the three types of edges through a graph attention part association modeling module, and fusing to obtain enhanced node features; then, the enhanced node features are mapped back to original spatial feature dimensions; carrying out adaptive weighted integration on different granularity features; and finally, an identity representation vector is generated, similarity between the identity representation vector and image library features is calculated, and a matching result is output. According to the method, by capturing time sequence dynamic association between human body parts, identity representation generated by the model is more focused on a stable human body structure instead of a variable clothing appearance.

This invention proposes a method for depression detection based on a multimodal speech-language model with instruction fine-tuning. The steps are as follows: First, a multimodal depression instruction dataset is constructed, integrating three types of information—original audio, automatically transcribed text, and emotional descriptions—into structured instruction-response samples. The emotional descriptions are automatically generated using a dedicated large-scale emotional inference model. Next, a low-rank adaptation technique is employed to efficiently fine-tune the parameters of the multimodal speech-language model. Audio features are extracted through an audio encoder and mapped to the text embedding space, then fused with the text and emotional description embeddings in a multimodal manner. The model parameters are optimized using a cross-entropy loss function. Finally, the speech to be detected is input into the fine-tuned model, and the depression recognition result is output. The technical solution provided by this invention enables joint inference based on audio, text, and emotional information, significantly improving the accuracy of depression recognition while offering advantages in parameter efficiency.

This invention discloses a deep learning-based method for ultra-short-term photovoltaic (PV) power prediction. First, historical data is preprocessed to eliminate irrelevant variables and accelerate model training. Then, the advantages of three clustering algorithms are combined to obtain a more reasonable dataset partitioning. Next, particle swarm optimization is used to optimize the parameters of variational mode decomposition. Finally, parallel learning of CNN and GRU networks is employed to identify local and temporal features of the data, enabling the network structure to fully leverage the input data. Finally, a deeper learning process is achieved through CNN neural network concatenation and fusion, resulting in high-precision prediction. The PV power prediction method of this invention demonstrates excellent performance, significantly outperforming other traditional models in predicting PV power under different weather conditions.