60results about How to "Guaranteed Computational Efficiency" patented technology

The invention discloses a blade aerodynamic robustness evaluation method based on a dual-fidelity collaborative neural network, and relates to the technical field of turbomachinery blade aerodynamic analysis. Acquiring a plurality of groups of uncertainty parameters influencing the aerodynamic characteristics of the blade; inputting the uncertainty parameters into the dual-fidelity collaborative neural network model, extracting feature vectors of the uncertainty parameters through a basic trend prediction module, and predicting low-fidelity aerodynamic performance; weighting the feature vector through a cross-fidelity attention fusion module, fusing the uncertainty parameter and the weighted feature vector, and inputting the fused vector into a residual error correction module to obtain a residual error prediction value; adding the residual prediction value and the prediction value of the low-fidelity aerodynamic performance to obtain a prediction value of the high-fidelity aerodynamic performance; and determining the robustness index of the blade according to the high-fidelity aerodynamic performance predicted values corresponding to the multiple groups of uncertainty parameters. According to the method, the calculation efficiency is ensured while the aerodynamic performance precision of the blade is improved.

The application discloses a kind of based on task partition's shipborne unmanned aerial vehicle navigation beacon group path planning method, belong to offshore navigation beacon inspection technical field.The method first obtains navigation beacon, mother ship, obstacle and unmanned aerial vehicle constraint parameter, is divided into navigation beacon by recursion K-means and is checked by threefold verification of operation radius, load, endurance, and is recursively subdivided if unqualified;Second generation mother ship candidate residence point and correct to safe water area, use A* algorithm to plan the global transfer path of mother ship, model as closed loop traveling salesman problem in cluster inspection, to improve self-organizing mapping neural network is solved unmanned aerial vehicle path in combination with tabu search.The system contains data input, task partition and the like module.The scheme makes mother ship only stay near task cluster, greatly shortens voyage, reduces risk, improves inspection efficiency and engineering feasibility.

The application provides a method and device for simulating a water delivery system of an impulse water turbine and electronic equipment, and the method comprises the following steps: establishing a mathematical model of the water delivery system of the impulse water turbine with variable step grids; dividing the water delivery pipe upstream of the distribution ring and the side pipe of the distribution ring into grids with the variable step grids; determining the boundary conditions of the inlet and outlet of the pipe based on the calculation working condition, and determining the initial flow and pressure values of each grid node according to the constant flow theory; combining the initial flow and pressure values of the grid nodes, iteratively calculating the corresponding flow and pressure values of each grid node at the current time, and determining the unit side parameters at the current time until the termination condition is met. Through the application, certain theoretical basis and technical support are provided for the safe and stable operation of the impulse unit, and the problem that the existing related technologies are difficult to balance the pipe calculation efficiency and the calculation accuracy is solved.

The invention relates to a soil organic matter prediction method based on multi-feature fusion, and belongs to the technical field of soil organic matter prediction. The method comprises the following steps: preprocessing acquired soil data to obtain spectral features, and performing time domain reconstruction of frequency domain signals to obtain time domain data; obtaining an optimal delay time and an optimal embedding dimension based on the time domain data, and performing phase space reconstruction to obtain a phase space trajectory; chaos features are extracted based on the phase space trajectory, and the extracted chaos features, the optimal delay time and the optimal embedding dimension serve as final chaos features; obtaining a vegetation index based on the spectral feature and taking the vegetation index as an index feature; and inputting the spectral features, the final chaotic features and the index features into a constructed double-flow low-rank interaction network model to obtain a soil organic matter prediction result. The objective of the invention is to solve the technical problem of low prediction precision caused by the fact that spectral features extracted in the prior art cannot comprehensively represent complex nonlinear characteristics of soil.

This invention discloses a dynamic inspection system for weld seams of a robotic arm on a flexible production line, integrating a laser tracker and dual photoelectric frequency combs. Based on numerical simulation, it establishes the discrete optimal intervals of the repetition frequency difference and carrier envelope offset frequency, constructing a precise time-domain measurement benchmark. Furthermore, it integrates fault saliency ranking and parallel convolutional neural networks, extracting signal morphology indicators of interference fringes and rearranging the sampling sequence to eliminate servo lag errors in high-speed tracking by the laser tracker. The inspection system provided by this invention can achieve real-time precise reconstruction of the weld seam trajectory of a robotic arm in complex environments on flexible production lines, effectively solving the problems of nonlinear interference and dynamic asynchrony in cross-scale measurements, significantly improving the inspection efficiency and positioning accuracy of automated welding, and possessing broad application prospects.

The application relates to a variational correlation evaluation method and device, which comprises the following steps: acquiring magnetic resonance image data to obtain to-be-processed data; determining an iteration position sequence of the to-be-processed data and extracting iteration features to obtain a training set and a test set; inputting the training set into a variational correlation evaluation classification model for iteration training and testing, judging whether the trained variational correlation evaluation classification model converges or not, and obtaining the variational correlation evaluation classification model when the model converges, and performing effective feature extraction based on the variational correlation evaluation classification model; determining each effective feature position sequence, converting the effective feature position sequence into a three-dimensional brain structure matrix, covering the three-dimensional brain structure matrix to a preset standard human brain template, and identifying effective features related to each stimulation condition. The application quantifies the contribution of a single voxel in the process of executing a specific cognitive function, and finally identifies and extracts the least amount of features that can best represent the target stimulation condition.

The invention discloses a line loss segmentation monitoring method and system for a power distribution master station and a fusion switch, and particularly relates to the technical field of power distribution network line loss, and the system comprises a line segmentation module, a segmentation interval data collection module, a segmentation interval line loss calculation module, a segmentation interval line loss correction module, and a segmentation interval abnormity processing module. Fusion switches are installed on a line according to a certain distance, measurement data of the fusion switches at the two ends of a line segment are collected through a main station, the basic line loss value and the line loss rate of the line segment are calculated, for a section containing a distributed power supply, the line loss influence coefficient and multi-scene compensation of the distributed power supply to the line segment are introduced, and the line loss value is calculated and corrected in real time. Based on the real-time corrected line loss value, in combination with the preset floating value, the threshold interval is constructed, the line loss abnormity of each line segment on the target to-be-monitored line is judged, the abnormal line segment is positioned, the line segmentation electric quantity metering is realized, the calculation precision under all working conditions is improved, the abnormity management is realized, and the operation and maintenance management complexity is reduced.

The embodiment of the invention provides a method and device for calculating neutronics parameters of a reactor core, and belongs to the technical field of nuclear power. The method comprises the following steps: acquiring vibration frequencies of a plurality of fuel assemblies in a reactor core when the fuel assemblies work under a preset vibration working condition; dividing the working period into M continuous time steps according to the plurality of vibration frequencies; for each time step in the M time steps, the reactor core is divided in the radial direction according to the vibration positions of the multiple fuel assemblies in the time steps, and multiple arbitrary quadrilateral segments are obtained; mapping the plurality of arbitrary quadrilateral segments to a reference coordinate system to obtain a plurality of square segments; carrying out reactor core neutronics calculation on the basis of the plurality of square segments to obtain neutronics parameters corresponding to each square segment; and mapping the neutronics parameter corresponding to each square segment to any quadrilateral segment corresponding to the square segment to obtain a target neutronics parameter corresponding to the time step. According to the embodiment of the invention, the accuracy of the neutronics calculation result of the reactor core can be improved.

The invention discloses a method for detecting small defects on the surface of lightweight steel, and belongs to the technical field of industrial machine vision. The method is based on an improved YOLOv8n architecture and cooperatively works through three core technical means: firstly, a Swin Transform module is introduced into a backbone network to model long-distance spatial dependence and suppress complex background interference; secondly, a high-resolution P2 detection branch is constructed, shallow details and up-sampling semantic features are fused, and microdefect characterization is enhanced through bidirectional refining circulation; finally, P2 exclusive adaptive threshold focus loss ATFL is adopted, threshold update is only limited to P2 detection branch samples, and precise optimization of difficult and tiny defects is achieved in cooperation with a gradient directional return mechanism. According to the scheme, the detection recall rate and the positioning precision of the tiny defects are effectively improved while the model parameter quantity is remarkably reduced, and high-precision and light-weight industrial deployment is realized.

The application belongs to the technical field of target detection, and discloses a wind turbine blade defect detection method based on an improved RT-DETR. The method introduces a SWRepBlock module in the backbone network, thereby effectively solving the problem that a fixed receptive field of a traditional network cannot adaptively extract target features of different scales. In addition, the method replaces an AIFI module in an encoder with a DyT-AIFI module to improve the semantic understanding effect of long-distance feature interaction. In addition, the method introduces a CAA-HSFPN module in the encoder, thereby effectively solving the problem of a semantic gap of a traditional feature pyramid and insufficient distinction of feature importance. The wind turbine blade defect detection method can not only significantly improve the detection capability of small damages on a complex surface texture background of a wind turbine blade, but also effectively identify multiple types of defect features such as cracks, erosion and oil leakage.

An adaptive multi-scale state-space model establishment method and its implementation in a perceptual scanning analysis method are presented. In multi-class pneumonia, especially in scenarios with class imbalance, small sample sizes, and blurred boundaries, it is difficult to simultaneously guarantee spatial structure modeling capability, pathological differentiation capability, and computational efficiency. There is a lack of standardized processing methods that address the detailed issues of spatial structure modeling capability, pathological differentiation capability, and computational efficiency. This invention divides the acquired and processed chest X-ray image into two data streams. One stream is processed through detection and adaptive multi-directional scanning to form a spatial model in the SSM state. The other stream is processed through multi-scale feature extraction to form an edge enhancement module and feature fusion data. The edge enhancement module, feature fusion data, and the SSM-state spatial model are interactively fused to form an interactive fusion model, which is then used to complete the classification output process.

The present application belongs to the technical field of power system frequency stability analysis, and provides a frequency trajectory minimum point prediction method and system based on piecewise linearization analysis, which groups frequency modulation devices according to response speed, respectively aggregates the frequency modulation power in each group in time domain, uniformly represents the frequency modulation power as a function of time and sums up, obtains a piecewise linear model of the total equivalent frequency modulation power of the system, retains the time sequence difference information of the frequency modulation power release of different types of devices, and avoids the information loss caused by the traditional method of mixing all devices into a single speed governor transfer function; by identifying the key moment when the frequency modulation power release rate changes significantly as a candidate breakpoint, and screening through linearization error control, the final breakpoint sequence is obtained, which divides the frequency response process into several linearization periods, and realizes the automatic sparse segmentation effect in the slow change interval, so that higher linearization accuracy can be achieved with fewer segmentation numbers.

This application discloses an analytical method for electrolyte systems based on low-scaling Monte Carlo simulations of electrostatic interactions, relating to the field of molecular simulation technology. The method includes: constructing an initial model of the target system and configuring simulation parameters; initializing local electric displacement auxiliary field variables; using these variables to describe the local electric displacement state of the target system; randomly selecting and updating an object; the updated object being the particle configuration and / or the local electric displacement auxiliary field variables; determining the equivalent local electric field strength at the target spatial location after updating the object; updating the local dielectric constant at the target spatial location based on the equivalent local electric field strength; and after updating the local dielectric constant, determining the various electrostatic interaction energies in the target system and evaluating the total electrostatic energy. This method can accurately describe the complex electrostatic interactions between components caused by local electric fields, high charge density, and the dielectric response properties of the components while ensuring computational efficiency.

The application discloses a kind of robot gas source positioning method, belong to intelligent mobile robot technical field.For the intermittent, broken and strong turbulent disturbance of gas plume in complex unknown environment, the method first constructs a relative time-varying perception model, and converts the historical observation data to the local coordinate system of the robot;By introducing motion uncertainty weight and turbulent perception time decay mechanism, adaptive filtering and dynamic updating of historical observation information are realized;Further design dynamic window mechanism and multi-modal feature fusion strategy to improve the modeling accuracy of local gas concentration field in complex environment;At the same time, combined with the improved dynamic window path planning method driven by environment, the high concentration area and the direction of adverse wind are cooperatively guided.The method can run in unknown environment, improve the accuracy, robustness, real-time and efficiency of source search, and is suitable for industrial leakage detection, disaster rescue and dangerous gas search and other application scenarios.

The application discloses a numerical simulation method for spray combustion of an aero-engine combustion chamber, and belongs to the technical field of simulation of spray combustion of an aero-engine combustion chamber. The numerical simulation method is based on an adaptive matching iterative solution control mechanism of liquid droplet and flow field motion scale to perform outer circulation; starting from mixed fraction definition and component transport, a coupling relationship between fuel evaporation and combustion is established by deducing mixed fraction source terms and component transport equation source terms under different definitions, so that the numerical simulation method has high-fidelity coupling simulation capability of flow-spray-combustion; the RCCE method is introduced to establish a flame surface database, a large number of components in a chemical reaction mechanism are mapped to a limited number of constraint variables for dimension reduction solution, and a constraint equilibrium state is obtained according to the maximum entropy principle, and finally, the concentration of each component is reconstructed by using a constraint potential energy; while ensuring the accuracy of key chemical thermodynamic coupling, the generation efficiency of a complex fuel chemical reaction database is improved.

The invention provides a bridge monitoring method, system and equipment based on deep learning and a medium, and the method comprises the following steps: collecting an original strain sequence and environmental parameters of each monitoring point of a road bridge, calculating a theoretical thermal strain according to the environmental parameters and a material thermal expansion coefficient, and carrying out the alignment subtraction of the theoretical thermal strain and an original data time sequence, so as to obtain a structural strain through correction; constructing a space-time strain matrix in a preset time window according to the monitoring point position and the acquisition time sequence, and uploading the space-time strain matrix to the cloud; inputting the matrix into a deployed convolutional neural network at a cloud end, extracting deep features by adopting multi-layer convolution comprising a time dimension convolution kernel and a space dimension convolution kernel, and then mapping the deep features into settlement state probability vectors through a full connection layer; and comparing the vector with a reference health state vector, and if the deviation exceeds a safety threshold, triggering settlement early warning. By implementing the technical scheme provided by the invention, environmental interference can be effectively eliminated, and the accuracy and real-time performance of bridge settlement monitoring are improved.

The invention discloses an image-based three-dimensional reconstruction method for electromagnetic simulation. Collecting and receiving a multi-view image structure by using a camera, and recovering to obtain sparse point cloud and camera parameters; inputting the data into a computer for processing, and performing initialization processing on each point of the sparse point cloud by using a ball initialization method to obtain respective quadric surface primitives; performing joint optimization on each quadric surface primitive through a simulated annealing algorithm; and extracting a triangular mesh for a symbol distance field formed by all the quadric surface primitives to obtain a three-dimensional reconstruction result, and inputting the three-dimensional reconstruction result into electromagnetic simulation. The method is clear in process and high in automation degree, geometric holes in a reconstruction model can be effectively reduced, surface continuity and geometric consistency are improved, the problem that an existing visual reconstruction method is difficult to meet the requirements for electromagnetic simulation geometric precision and continuity is solved, and the method is worthy of popularization and application. The usability and reliability of a reconstruction model in high-frequency electromagnetic simulation are remarkably improved, and the method has a good engineering application prospect.

The invention discloses a radar HRRP deformation target recognition method based on feature correction, and the method comprises the steps: carrying out the preprocessing of the radar one-dimensional HRRP echo data of a to-be-recognized target, and obtaining the HRRP original target data; superposing block sparse signals with different sparseness and Gaussian white noise on the HRRP original target data to obtain HRRP deformation target data; constructing a lightweight feature extraction network, and performing pre-training by using HRRP original target data; a feature correction network comprising a multi-dimensional parallel feature correction module, a feature fusion module and a de-noising post-processing module is constructed, and training is carried out after the feature correction network is cascaded with the feature extraction network; and combining the feature extraction network with the feature correction network to obtain a deformation target identification model so as to realize identification and classification of a deformation target. According to the method, the intrinsic feature distribution of the deformed target can be effectively recovered, and the recognition accuracy and generalization ability of the non-cooperative target in a low signal-to-noise ratio environment are remarkably improved.

The invention discloses a multi-step reasoning information acquisition method, system and device and a storage medium, which are corresponding schemes: performing semantic analysis and dynamic correction on task query by using a large language model to realize multi-step reasoning and adaptive optimization of query semantics; moreover, an incremental multi-path recall system fused by heterogeneous associated signals is designed, and keyword, semantic and structured features are fully fused through a multi-path, multi-view and multi-round progressive recall joint mechanism, so that high-coverage and high-precision candidate information recall is realized; in addition, a fine ranking scheme based on deep thinking (thinking chain) and reasoning expansion is provided, and interpretable logic analysis and accurate ranking are carried out on candidate information. Generally speaking, the information understanding, reasoning depth and retrieval accuracy of the model under the complex incidence relation can be remarkably improved while the calculation efficiency is kept, and universal technical support is provided for a high-reliability knowledge acquisition and scientific intelligent system.

The application discloses a coal mine image segmentation model and method based on a VMamba and multi-expert hybrid network and a construction method thereof. The coal mine image segmentation model is constructed. The image block encoding layer output of an encoder is taken as the input of the first encoding layer of a first VSS network, and the output of the first encoding layer of the first VSS network is taken as the input of the output end feature learning module. The input of the output end feature learning module is taken as the input of the first decoding layer of a decoder, and the output of the first decoding layer of the decoder is taken as the input of the second decoding layer of the decoder. The input of the second decoding layer of the decoder is taken as the input of a segmentation head, and the output of the segmentation head is taken as a segmented image. The application can significantly improve the segmentation precision and calculation efficiency of the coal image.

This invention discloses a method for generating missing modalities in brain MRI based on hypergraphs and attention mechanisms. This method addresses the issue of missing modalities in multimodal brain MRI sequences such as T1, T1ce, T2, and FLAIR in clinical scenarios by constructing a unified multi-input multi-output translation framework. The method introduces hypergraph convolution and region-level self-attention into the generative network, aggregating group-level features from different tumors and healthy tissues and modeling structural relationships between sub-regions, preserving the spatial tissue structure of the tumor. Combined with bidirectional Mamba sequence modeling, it enables the 2D network to efficiently capture long-range dependencies between layers of 3D volumetric data, ensuring voxel-level spatial continuity. Through a teacher-student knowledge distillation mechanism, the student network learns structurally perceptual features without a tumor mask, achieving high-fidelity generation. This method can improve the overall quality of generated images and the detail fidelity of tumor lesions, thereby enhancing the performance of downstream tumor segmentation tasks and showing promising application prospects.

This invention provides a method and system for cross-variable long-term time series prediction that integrates linear and enhanced Transformer methods. The method includes: extracting an input sequence using a sliding window based on historical data and calculating the temporal statistical features of the sequence; generating scaling and bias factors for the window through a shared fully connected network to perform affine normalization on the input and eliminate distribution bias; performing a Fast Fourier Transform on the normalized tensor to extract the main frequency domain components and map them back to the time domain, then fusing positional encoding to form a time-frequency hybrid embedding; constructing an adjacency matrix based on the correlation between variables, using this matrix as a mask in cross-variable attention calculation to suppress interference between weakly correlated variables and enhance variable co-representation; mapping the representation to the prediction length through a time projection layer, and performing an inverse transform using previously generated normalization parameters to restore the original dimensions of the data and obtain the prediction result.

This invention discloses a UAV state estimation method based on a two-stage search, relating to the field of low-altitude sensing and control technology. The method includes: constructing a three-dimensional reference coordinate system for a substation area, a set of no-entry flight zones, and a constraint grid table; generating an environmental constraint data packet; configuring and transmitting OTFS sensing frames based on the environmental constraint data packet, acquiring echo sequences, performing time synchronization, frequency offset correction, static clutter suppression, and frequency domain pre-whitening processing to generate a weighted observation sequence and a gridded environmental cost table; performing a coarse search using two-dimensional orthogonal matching pursuit under environmental constraints to obtain a unique candidate state group; performing a continuous domain fine search based on the unique candidate state group to complete position and velocity correction and generate current cycle state data; and generating initial state data for the next cycle based on the current cycle state data. This invention balances search efficiency and estimation accuracy through a two-stage search; and improves the accuracy of UAV state estimation through precise time delay and precise Doppler fitting.

The application discloses an anticancer peptide generation and screening method and system based on a diffusion model and a seed synchronous autoencoder, and comprises the following steps: preprocessing original anticancer peptide and non-anticancer peptide sequences and extracting multi-dimensional physicochemical features; training a diffusion model in a continuous numerical space based on sequence mapping based on a global random seed, adding noise through forward diffusion and learning the potential distribution of anticancer peptide data features through reverse denoising; based on a reverse seed synchronization strategy, training an autoencoder using the same random seed and environment state, and establishing a mapping from continuous random noise to discrete biological sequences; using the diffusion model to sample and generate potential vectors in the latent space and mapping them into candidate anticancer peptide sequences through the decoded synchronously trained decoder; inputting an integrated learning classifier based on grouping features, selecting the optimal feature group and predicting the anticancer activity probability, and outputting the anticancer characteristic peptide chain; the application solves the mapping contradiction between continuous space and discrete sequences and can efficiently screen anticancer peptides with high activity sequences.

This invention relates to a data-driven, multi-agent energy and carbon collaborative decision-making method for low-carbon industrial parks. It constructs a multi-agent dynamic Stackelberg game model, incorporating time-varying grid-side carbon intensity and the carbon offsetting benefits of green electricity into the multi-agent game decision-making process. Energy suppliers act as leaders, and users as followers, aiming to minimize the sum of economic energy consumption costs and carbon costs calculated based on dynamic carbon factors to solve for the optimal energy consumption strategy. A hybrid intelligent solution framework is used to solve the multi-agent dynamic game model. At the game strategy learning level, an improved multi-agent deep reinforcement learning algorithm is employed to approximate the game equilibrium. At the individual optimization level, an improved multi-objective particle swarm optimization algorithm is embedded within the follower agent to solve for the follower's optimal response under the current price strategy. Compared with existing technologies, this invention can coordinate the interests of multiple agents with dynamic carbon objectives, promoting source-load synergy and efficient renewable energy consumption.

The application discloses a kind of long supply and demand mutation water-heat coupling dynamic simulation method and system of heat supply pipe network, belong to central heating technical field, technical scheme is, obtaining heat supply pipe network topological structure parameter, elevation, pipe structure parameter, insulation layer structure parameter and environmental parameter;Matrix is constructed to node association, sets heat source parameter, physical property parameter, mutation working condition parameter and space step;Initial hydraulic calculation is carried out, and the flow of each pipe section of heat supply pipe network and the pressure distribution of each node are obtained;Temperature field in pipe network is calculated when supply and demand load mutation occurs Time advancing, and heat medium physical property parameter and pipe network hydraulic state are updated synchronously in temperature field calculation process, the coupling dynamic simulation of water and heat is realized;On the basis of coupling dynamic simulation, the heat storage or heat release process of heat supply pipe network under supply and demand load mutation working condition is analyzed.The beneficial effects of the application are: provide a kind of long supply and demand mutation water-heat coupling dynamic simulation method and system of heat supply pipe network.

Embodiments of the present disclosure provide an indirect memory copy method, a processing unit, a computing device and a system. The processing unit of the embodiments of the present disclosure comprises an operation unit, an addressing unit and an index cache, and the addressing unit and the index cache are located between the main storage area outside the processing unit and the operation unit. The operation unit executes an indirect memory copy instruction, and the indirect memory copy instruction at least has a base address, an index address and a destination address, so as to send the index address to the index cache and send the base address and the destination address to the addressing unit; the index cache loads a corresponding index from the main storage area according to the index address and sends the index to the addressing unit; the addressing unit determines a source address of the main storage area corresponding to source data according to the base address and the index; and the addressing unit loads the source data from the main storage area according to the source address and sends the source data to the destination address of the internal cache of the operation unit. The scheme of the embodiments of the present disclosure saves the computing resources of the operation unit itself.

This invention discloses a method, device, and medium for predicting the tail risk of static stability margin in power systems, belonging to the field of risk prediction technology. The method includes: using a quantile regression model to perform multi-quantile prediction of the output of new energy sources such as wind power and photovoltaics, obtaining the cumulative distribution function of the output of each new energy node; further constructing a discrete probability density function of the new energy output through discretization and differencing to avoid modeling errors caused by the assumption of continuous distribution; based on this, combining the thermal power output configuration and the static stability margin based on converter dynamic parameters, establishing a mapping relationship between the random injection of new energy and the static stability margin of the system, realizing the quantitative prediction of the stability margin probability distribution and its tail risk. This invention can accurately reflect the distribution characteristics of the static stability margin of the receiving-end power system in a probabilistic sense, especially the stability margin variation law under low-probability, high-risk operating conditions.

The application discloses an unmanned aerial vehicle reconnaissance video image target automatic detection method, and belongs to the technical field of unmanned aerial vehicle image processing. The method comprises the following steps: preprocessing and frame extraction are performed on an unmanned aerial vehicle reconnaissance video image, a video frame sequence is constructed, multi-scale feature extraction is performed on the video frame sequence, multi-scale features extracted based on semantic guidance are fused, a time sequence memory model is constructed, the multi-scale features are corrected and enhanced in a time sequence context, a weak and small target detection branch is generated based on multi-dimensional feature perception, the multi-scale features corrected and enhanced are detected in the weak and small target detection branch, and a detection result is output. Through grouping convolution, the method can efficiently fuse, guarantees the calculation efficiency, and enhances the contour perception ability for fuzzy weak and small targets.

This invention discloses a method, apparatus, device, and medium for constructing three-dimensional semantics from multi-view images. The method includes: obtaining multiple three-dimensional Gaussian primitives (3D Gaussian primitives) based on multi-view images of a target scene and camera acquisition parameters for each image; the multiple 3D Gaussian primitives describe the geometric structure and appearance attribute information of the target scene; each 3D Gaussian primitive represents a local region in the target scene; inputting the multiple 3D Gaussian primitives into a trained neural regularization model to obtain the semantic features corresponding to each 3D Gaussian primitive output by the neural regularization model; the neural regularization model is used to learn the mapping relationship from the consistent attribute space of the 3D Gaussian primitives to the semantic feature space, thereby achieving regularization constraints on semantic information. By utilizing the geometric structure and appearance attribute information of the 3D Gaussian primitives to effectively regularize the semantic information at the 3D representation level, semantic noise in the 3D semantic field is suppressed while ensuring computational efficiency.