531 results about "Temporal information" patented technology

A system and method for multi-scale temporal attention processing in multimodal technology deep learning systems. This system processes time-series, textual, sentiment, and structured tabular data across three hierarchically-organized temporal streams—quarterly, weekly, and intraday levels—with bidirectional cross-temporal information flow. Scale-specific attention mechanisms are optimized for respective temporal granularities, while an adaptive controller dynamically weights each temporal level based on real-time market volatility indicators. A multi-scale fusion processor integrates attention-weighted representations to generate temporally unified representations preserving both short-term market dynamics and long-term trends. This approach enables superior forecasting and risk assessment by leveraging temporal correlations across multiple time scales while automatically adapting to changing market conditions. The system facilitates interpretable AI analysis through attention visualization and enables synthetic scenario generation for model testing.

The invention discloses an insurance claim settlement-oriented multi-modal image video evidence analysis method and system. The method comprises the following steps of: acquiring video / image and multi-source data such as metadata, audio, IMU (Inertial Measurement Unit), GPS (Global Positioning System), OBD (On-Board Diagnostic) and the like; calculating content Hash of the video and the audio according to frames, connecting the content Hash with time information in series to form chained Hash, and adding a verification digital signature and a credible timestamp; realizing cross-modal time sequence alignment based on self-adaptive time anchor-attitude coupling; tampering detection is carried out in combination with PRNU fingerprints, noise field consistency, dual compression, copy-movement and the like; multi-view geometry and monocular depth are fused, IMU scale constraint and micro rendering are introduced, three-dimensional reconstruction and re-projection optimization are completed, and collision dynamics verification is carried out; and constructing an event cause and effect graph, judging responsibility in combination with traffic rules, outputting a confidence coefficient vector and a structured report, and generating a verifiable evidence packet. The scheme has the advantages of high efficiency and traceability in the aspects of space-time restoration and interpretable responsibility judgment.

The invention discloses a short video network public opinion information identification method based on an image processing technology, and relates to the technical field of artificial intelligence and image processing, and the method comprises the following steps: S001, through obtaining image frames, audio tracks and time sequence information of a short video, constructing a multi-modal fusion model, extracting continuous image frames with suspicious identity features, and carrying out the recognition of the short video network public opinion information; generating a forgery risk area distribution map; and S002, performing semantic consistency verification according to the counterfeit risk region distribution map, and extracting space and time anomaly features existing among facial micro-expressions, pronunciation actions and background semantics in the image frame. According to the method, a multi-modal model is constructed by fusing image, audio and time information, fine abnormal features, traceability forgery starting points and propagation paths in a deep forgery video are identified, and an identification strategy and a public opinion response mechanism are dynamically adjusted, so that accurate identification, adaptive processing and closed-loop control of short video public opinion risks are realized; and the identification accuracy and the treatment efficiency are improved.

An artificial intelligence system based on spatial-temporal information pairs is provided by the present disclosure. By integrally deploying paired vision, auditory and olfactory acquisition devices, a device that can collect the data within a 720-degree area is constructed, and the multi-dimensional continuous spatial-temporal information pairs such as positions, morphologies, motion states, sounds and odors from the ambient environment of the acquisition device or a same spatial object in the environment are recorded in real time. These information pairs can not only contain spatial relationships and a clock attribute, but also contain rich label attributes, for example, identifiers of acquisition devices, and a name, a category and a behavior pattern of the spatial object.

The invention discloses a surgical robot dynamic compensation method based on multi-mode real-time 4D digital twinning and a related device. The method comprises the following steps: acquiring preoperative three-dimensional and four-dimensional image data of a patient, synchronously acquiring optical body surface movement data and in-vivo ultrasonic data during an operation, inputting the data into an end-to-end modular network in combination with current state data of a surgical robot for feature extraction and fusion, generating a 4D twinborn space containing spatio-temporal information, and generating a compensation instruction based on the space to control the surgical robot. According to the method, real-time fusion and dynamic modeling of multi-modal data in an operation can be realized, the self-adaption and compensation capability of a surgical robot in a complex surgical environment is improved, and the surgical precision and safety are enhanced.

The embodiment of the invention discloses a safety monitoring method and system for building construction, and the method comprises the steps: collecting the original video data of a construction site, carrying out the denoising, illumination correction and frame rate adjustment, and outputting a standardized video data stream; extracting attitude features of the constructors and representing the attitude features as a feature matrix to form an attitude feature matrix set containing spatio-temporal information; the sensitivity and correlation of the feature matrix are analyzed, calibration noise is added after dimension reduction, and feature data conforming to differential privacy are generated; and extracting data advanced representation by utilizing a pre-training model, and completing behavior classification, dangerous area judgment and safety violation detection. And the violation risk is evaluated in combination with the risk level of the construction area, graded early warning is generated, and meanwhile violation information is recorded to form a traceable management mechanism. According to the embodiment of the invention, accurate and efficient safety violation behavior detection is realized, and intelligent technical support is provided for safety management of building construction.

The invention discloses a shield tunnel ground subsidence real-time prediction method considering spatio-temporal information, and the method comprises the steps: constructing an input feature system, including classifying model input into geometric information, multi-ring geological condition information, multi-time step shield parameter information and historical subsidence information; a multi-source information fusion model architecture is designed, geometric information, geological condition information, shield operation parameter information and historical settlement information serve as input of the model, and feature extraction is conducted through multiple encoders. Then feature fusion is carried out, nonlinear mapping is carried out by using a residual network (ResNet), and finally a real-time settlement prediction value is output; according to the scheme, the space-time characteristics of ground subsidence induced by tunneling are deeply excavated, and real-time prediction of subsidence of any position in a disturbance range is realized through deep fusion of multi-source characteristics.

An apparatus is configured to classify indicator lights of surrounding vehicles as either active or inactive. The apparatus may use a Siamese network to determine respective feature vectors from respective images captured by a vehicle at different times. The apparatus may also embed speed and time information in the respective feature vectors based on odometry information associated with the vehicle at a time each respective image was captured, and fuse, using a temporal attention mechanism, features and the speed and time information from the respective feature vectors to produce a fused feature vector. The apparatus may further process the fused feature vector using capsule modules to produce an indicator feature vector, calculate a similarity metric from the indicator feature vector, and process the similarity metric with a classifier to output an indicator classification.

The invention discloses an RGBL tracking method based on target prior autoregression query. The RGBL tracking method comprises the steps that RGBL data sets are collected and aligned, and a training set and a test set are constructed; the method comprises the following steps: on the basis of an RGB tracking network AQATrack of autoregression query, constructing an RGBL tracking model based on target prior autoregression query; by introducing language and visual semantic tokens, learning target features of each mode; designing a language semantic token enhancement module to enhance the target features of the language semantic token, and designing a target feature extraction module and a visual semantic token feature enhancement module to improve the target features of the visual semantic token; and fusing language and visual semantic token features in a decoder, outputting a multi-modal semantic token feature, and using the multi-modal semantic token feature as target prior for query with an initialization value of zero, and capturing spatio-temporal information in an autoregression learning mode. According to the invention, by introducing the target prior, the target features in the spatio-temporal information can be captured more effectively in the initial stage, so that the target positioning and tracking process of the tracker is accelerated.

A method including receiving semantic data from a first user device, the semantic data comprising at least one semantic token; obtaining spatial and temporal information associated with the first user device; associating the semantic data with the spatial and temporal information to generate associated semantic data; grouping the associated semantic data based on spatial and temporal proximity to create a cluster;and fusing data within the cluster to generate a representation of a portion of an environment.

The invention relates to a micro-expression recognition method based on double-flow feature fusion, and belongs to the field of computer vision. The method comprises the steps of obtaining N frames of images of a micro-expression video from a start frame to a vertex frame, and calculating an optical flow field between adjacent frames by adopting an optical flow algorithm so as to effectively extract optical flow features of micro-expressions; the method comprises the following steps: acquiring N frames of images of a micro-expression video from a start frame to a vertex frame, converting the N frames of RGB face images into a CIE Lab color space, calculating a pixel difference between two frames of converted Lab images, and extracting pixel stream features of micro-expressions; and inputting the optical flow features and the pixel flow features into a constructed double-flow three-dimensional convolutional network for feature extraction and fusion, and classifying micro expressions. According to the method, the dynamic and subtle changes of the micro-expression are effectively captured by combining the optical flow and the pixel difference characteristics. By integrating spatial and temporal information, richer feature representations are provided. The improved attention mechanism further focuses on fine facial changes, and the accuracy of micro-expression classification is improved.

The invention discloses a target retrieval method and system. The method comprises the following steps: S1, data access and preprocessing: acquiring multi-source data, performing cleaning, formatting and time-space standardization, and performing target detection and cutting on an image / video to generate a structured target object; s2, feature extraction: extracting deep semantic feature vectors and auxiliary understanding information, which have discriminability and adapt to complex scenes, from the target image; s3, constructing a data index, including constructing a spatio-temporal semantic hypergraph index of multiple types of nodes and hyperedges based on deep features and spatio-temporal information of the targets to express a complex relationship between the targets; and S4, data query: receiving multi-modal query information of a user, performing candidate region screening, feature matching and relation reasoning by utilizing a hypergraph index, and finally outputting a high-confidence target retrieval result. According to the method, the robustness of complex scenes and target changes and multi-dimensional query of depth are improved.

The invention provides a long video understanding method capable of relieving time sequence illusion in a video language large model. The long video understanding method is based on a static bias adaptive frame selection mechanism and a cross-modal feature fusion strategy. According to the static bias mechanism, inter-frame similarity is evaluated through a discriminator, redundant frames are identified, key frames are selected or a complete sequence is reserved, so that calculation overhead is reduced, and spatio-temporal information integrity is kept; a video frame and a text are mapped to a shared semantic space, the single-frame semantic understanding ability is enhanced, then an embedded sequence serves as a soft prompt to be input into a large language model, and a final answer is generated in an autoregression mode. According to the method, the efficiency and accuracy of long video understanding and video question and answer tasks can be remarkably improved; the problem of low training and reasoning efficiency caused by time sequence dependence redundancy and excessive computing resource consumption is effectively relieved; and through a dynamic multi-modal task processing framework and a space-time memory bank compression mechanism, the modeling capability and generalization performance of the model on a long video sequence are further improved.

The invention relates to the technical field of intelligent monitoring of protective forests, and discloses a protective forest monitoring and evaluating system based on intelligent visual identification. According to the system, synchronous perception information of vision, terrain and weather is captured through a multi-source information synchronous acquisition module, and a vegetation growth space field containing canopy form topology and canopy density change process is generated through multi-dimensional feature extraction operation. A degeneration characteristic evolution graph constructed according to the method can present a degeneration plaque contour, a spreading track and an intensity distribution thermodynamic diagram. The map is subjected to deep feature analysis through a trained multi-level degradation identification network to obtain a degradation level and a trend vector, and finally a comprehensive evaluation report with spatio-temporal information is formed. According to the invention, high-precision and automatic monitoring and evaluation of the dynamic degradation process of the protection forest are realized.

The invention discloses a group behavior identification method based on multi-view individual relationship interaction, which comprises the following steps of: firstly, acquiring joint point information of different individuals in a scene by utilizing a skeleton point estimation network and a position information coding technology; then extracting multi-granularity feature representation of the body part through a depth model; then, utilizing an attention mechanism in Transform to respectively construct a structured relation reasoning module in the individual under multiple view angles and a spatio-temporal information interaction module among individuals under multiple view angles, and realizing deep interaction of individual body part information from different view angles; self-adaptive fusion factors are designed, cross-view cross-granularity individual interaction features are integrated, and high-discrimination individual content representation is obtained. And finally, a multi-head loss training strategy is added, the types of the individual and group behaviors are judged from the enhanced individual and group behavior characteristics, and a group behavior recognition task in a complex scene is realized.

The invention discloses an end-to-end automatic driving method based on linear time complexity. The end-to-end automatic driving method comprises the following steps: acquiring sensor information of historical and current frames; fusing the information of the multi-modal sensor to obtain a BEV feature map; the BEV features and the current state of the vehicle are spliced, and a differential discarding strategy is executed; and obtaining a multi-modal trajectory through a decoder based on a diffusion strategy, and selecting the trajectory with the highest score as a final output trajectory. The method has the beneficial effects that the adopted encoder and decoder are realized based on a linear attention mechanism, so that the reasoning efficiency can be greatly improved; space and time information can be extracted and fused from continuous multi-frame sensor input, dynamic changes of a scene are accurately captured, and a guarantee is provided for making a more reasonable and safer track decision for a vehicle; the linear time complexity is kept, and meanwhile efficient alignment and information interaction between unequal-length query and features are achieved.

The invention relates to the technical field of network security, and discloses an intelligent control method and system for an Internet protection gateway, and the method comprises the following steps: obtaining network flow, and extracting a multi-mode state feature; and performing causal reasoning in combination with the knowledge graph to generate causal features. After fusing the two features, inputting the two features to three agents, namely a flow analysis agent, a response strategy agent and a resource scheduling agent, for collaborative decision, and generating a security strategy and a resource scheme; according to the scheme, dynamic deployment is carried out on heterogeneous computing resources, flow processing is completed, and data are recorded; and finally, iteratively optimizing the agent model by utilizing the disposal data, and applying the optimized model to the next round of decision. According to the method, the multi-modal state features including the basic features, the application layer semantics and the time sequence information are extracted, the network security knowledge graph is further constructed for causal relationship reasoning, and isolated network events are placed in a wider logic relationship for analysis, so that the depth and accuracy of network threat identification are improved.

The invention discloses a ground wire full life cycle management method, relates to the technical field of power equipment asset management, and aims to solve the problems that manual recording is easy to tamper, cloud diagnosis delay is high and an alarm mode is single and easy to neglect in traditional ground wire management. According to the method, operation fingerprints and spatio-temporal information during grounding wire hooking are recorded through the block chain technology, and it is ensured that data cannot be tampered; a built-in lightweight AI model of the edge computing unit is used for analyzing sensor data in real time to carry out localized fault diagnosis; and triggering a multi-mode sound-light alarm controlled by the PWM signal according to the diagnosis result grade. The system realizes full-life-cycle credible traceability, sub-second fault response and high-recognition-rate alarm of the state of the grounding wire, and is suitable for intelligent operation and maintenance of the grounding wire in the field of a transformer substation and the like.

The invention belongs to the technical field of image processing, and provides a facial paralysis grading method, system and equipment fused with multi-modal data and a medium in order to solve the problem that existing facial paralysis grading is inaccurate. Carrying out joint modeling on the handmade facial paralysis features based on prior knowledge and the depth visual features containing spatio-temporal information; extracting multi-scale features based on static image features and key point features extracted from a single-frame face image of a target individual, gradually transmitting small-scale features representing local asymmetry to medium-scale features and large-scale features, and adaptively performing feature fusion through dynamic weight distribution to obtain static symmetric features; and a bidirectional information interaction channel between the dynamic facial features and the static symmetric features is constructed, so that the generated fusion features simultaneously contain complete pathological information of spatial structure asymmetry and motion abnormality, and diagnosis grading is more accurate.

The invention relates to a multi-modal gait recognition method based on space-time semantic modeling and cross-modal cooperation, and the method comprises the steps: inputting a skeleton energy diagram, a contour energy diagram, a contour diagram and a skeleton diagram into four feature extraction branches with the same structure, and obtaining four modal features and two stage features of each modal; inputting the two stage features of each group of cross-modal combination into a double-stage feature interaction enhancement module to obtain a cross-modal fusion feature; all the modal features and the cross-modal fusion features are mapped and pooled and then input into a classification head, and a multi-modal gait recognition result is obtained. According to the method, key features under different view angles and walking conditions are effectively focused, the integrity of local channel information is reserved, modeling is performed on spatio-temporal information through multiple branches, high-level semantic features are extracted, spatio-temporal changes of gaits under different view angles and walking conditions are accurately captured, features of different modal data are fused, and the gait time-space information is obtained. And the gait recognition performance under multi-view and different walking conditions is improved.

The invention discloses a typhoon intensity prediction method and device fused with a space-time deep network, and a medium, and the method comprises the following steps: 1), collecting typhoon historical observation data, and obtaining typhoon data; 2) data preprocessing; 3) extracting data features; 4) feature fusion; 5) establishing a typhoon intensity prediction model; 6) model training; and 7) performing prediction by using the trained model, outputting a final prediction value, and restoring a prediction result into an actual typhoon intensity value through reverse normalization processing. The influence of the remote sensing satellite image and the characteristics of the tropical cyclone on the tropical cyclone strength estimation accuracy is considered, the typhoon strength prediction integrated with the spatio-temporal information is carried out, the spatio-temporal representation of the typhoon focus area is enhanced through the integration of the spatial characteristics and the time sequence characteristics of the typhoon, and the accuracy of the strength prediction result is improved.

The invention discloses a dynamic response method of a network attack defense strategy based on multi-head attention fusion, and relates to the technical field of network security, and the method comprises the steps: collecting network flow data and equipment state data of IIOT equipment, carrying out the time alignment, obtaining statistical features and frequency domain features through feature extraction, and generating a structured feature matrix; calculating a cross-domain association weight of statistical features and frequency domain features in the structured feature matrix through an attention mechanism, generating a fusion feature vector, carrying out time sequence modeling by using an LSTM model, calculating an abnormal probability score, and generating an abnormal event report; the method comprises the following steps: mapping spatio-temporal information in an abnormal event report into a quantum bit coding scheme in a digital twin environment, generating a Pareto optimal solution set by using a quantum annealing algorithm, screening a final network attack defense strategy, and coding the final network attack defense strategy into a quantum decision instruction. According to the invention, the adaptive security protection capability of the industrial Internet of Things system in a complex attack scene is effectively improved.

The invention relates to the technical field of motion detection alarm, in particular to a multi-agent fusion perception area intrusion abnormity alarm method and system, and the method comprises the following steps: collecting a target coordinate and a timestamp, integrating and sorting tracks, analyzing the direction and time change, judging the stability, recognizing an induction behavior, and screening and verifying the abnormity. According to the method, through integration of the spatial position and time information of the multi-source sensing intelligent agent, the continuous track of the target is reconstructed, the integrity identification of the motion behavior is enhanced, and the track direction change and time interval linkage analysis is carried out, so that the sensing depth of the behavior trend is improved; multi-dimensional recognition of abnormal behaviors is achieved through track stability and regional stay state linkage judgment, continuous conditions are introduced into abnormal verification to filter accidental interference, alarm output is matched with a regional strategy based on a verification result, the accuracy and robustness of intrusion detection are effectively improved, and the practical requirement in a high-dynamic scene is met.

The invention belongs to the technical field of computer vision and robot collaborative perception, and relates to a 3D human body posture estimation method and system based on space time sequence information fusion. The method comprises the steps that a multi-view image sequence is collected, and 2D joint coordinates of all view angles are extracted from the multi-view image sequence through a posture detector; position embedding, global embedding and edge embedding are carried out based on the 2D joint coordinates, and spatial features, including position information, global association information and skeleton edge information, of each joint point are obtained; fusing the features of each visual angle in the spatial features through a cross-visual-angle attention mechanism to obtain global feature representation; performing spatio-temporal feature enhancement and time sequence mixing on the global feature representation to obtain features integrating time sequence information, space information and channel information; and performing 3D posture regression on the features of the integrated time sequence information, the space information and the channel information to obtain a 3D human body posture. According to the invention, accurate and robust multi-view 3D human body posture estimation can be realized.

The invention relates to a photovoltaic power generation data prediction method and device based on multi-source data. The method comprises the following steps: acquiring multi-source data of a photovoltaic power station; preprocessing the multi-source data to obtain preprocessed multi-source data with consistent time and space; performing feature extraction and fusion on the preprocessed multi-source data to generate multi-source data feature representation; inputting the multi-source data feature representation into a space-time prediction network for training, and constructing a photovoltaic power generation prediction model; according to the photovoltaic power generation prediction model, obtaining a feature map fusing photovoltaic power generation spatio-temporal information obtained by combined extraction of a time convolutional network and a graph convolutional neural network; inputting the feature map fused with the photovoltaic power generation spatio-temporal information into a prediction layer; and carrying out feature mapping through two-step convolution operation in sequence in the prediction layer, and outputting a power generation data prediction matrix of the target photovoltaic power station in a future time step by the prediction layer.

The invention discloses a video key frame extraction method fused with self-supervised deep learning, and the method comprises the following steps: carrying out the standardized sampling of an input video according to a fixed interval, extracting a SuperPoint local key point and a Video MAE global semantic feature, and generating a dense descriptor and a semantic vector; constructing multi-dimensional change indexes such as local matching degree and global similarity; constructing a soft distribution matrix and a matching point set based on the fusion features; dynamically judging the key frame through a self-adaptive multi-threshold rule; and outputting the key frame set. The method fuses local and global spatio-temporal information, has robust feature extraction and key frame discrimination capabilities under a weak supervision condition, and can effectively improve the efficiency and precision of video compression, abstract and event detection.

The invention provides a short play subtitle translation system based on artificial intelligence, and relates to the technical field of artificial intelligence, and the system comprises a subtitle recognition module, an erasing module, a translation module and an output module, and can automatically recognize time information, position coordinates and visual style parameters of subtitles in a short play video. And generating a picture sequence without original subtitles in combination with the erasing processing, and translating the recognized original subtitle text into target language subtitles. The system generates target rendering parameters based on explicit style parameters and implicit style embedding vectors, realizes high restoration of fonts, strokes, shadows, gradient, transparency, textures and dynamic special effects, and performs adaptive adjustment according to target subtitle text features. Therefore, the visual consistency and culture adaptability of translated subtitles are improved in a short drama scene with complicated subtitle styles, frequent dynamic changes and obvious cross-culture differences, the audience impression is improved, and the manual post-processing workload is reduced.

The invention relates to a tunneling action generation method and system based on time-space depth fusion multi-task prediction in the technical field of shield engineering data processing, and the method and system integrate local and global time sequence information through a dynamic depth fusion network, achieve the fusion of multiple time-space scales, improve the perception capability of complex working conditions, and improve the efficiency of shield engineering data processing. Meanwhile, future state prediction and control candidates are output, action fusion is carried out on a strategy layer, performance loss caused by prediction-control splitting is reduced, prediction-control integration is achieved, an uncertainty head is introduced, explicit constraint is carried out on a loss function and strategy fusion layer, the risk under stratum sudden change or sensing noise is effectively restrained, and the prediction-control performance is improved. Uncertainty constraint security is realized; on the basis of experience playback, weight self-adaption and noise removal, online self-adaption updating requirements of different stratums and tunneling stages are met, online self-adaption updating can be achieved, and the action generation capacity of tunneling stability control under the complex stratums and noise conditions is remarkably improved.

The invention provides a safety anti-counterfeiting method based on double anti-counterfeiting codes, belongs to the field of anti-counterfeiting technologies, and is used for solving the problems that anti-counterfeiting methods in related technologies are weak in anti-copying, anti-tampering and anti-transfer capabilities and are difficult to cover full life cycle tracing of products. On the basis, generating double codes, associating, binding and storing the double codes and spatio-temporal information, and then performing layered verification; the anti-copying, anti-tampering, anti-transferring and anti-quantum-cracking capabilities of the whole life cycle can be improved, and the verification precision and the scene adaptability are both considered.

The invention discloses a hydroelectric generating set rotor acoustic diagnosis method and device based on a space-time joint feature map and a medium, and belongs to the technical field of hydroelectric machines. Generating an acoustic signal with a complete spatio-temporal information mark; forming a preliminary acoustic feature vector; further analyzing the preliminary acoustic feature vectors, and determining phase differences among the acquisition nodes; generating a space-time synchronization acoustic signal subjected to coupling alignment; further feature extraction is carried out on the space-time synchronous acoustic signals subjected to coupling alignment, and a space-time joint feature map used for describing the operation state of the hydroelectric generating set rotor is constructed; and performing automatic analysis on the space-time joint feature map to generate a final fault diagnosis report. According to the method, weighted focusing is carried out on the high-energy disturbance characteristics in the frequency domain and the time sequence dimension in combination with a self-attention mechanism, and the recognition resolution of complex multiple fault modes is improved.