29results about How to "Reduce computing consumption" patented technology

The invention discloses a multimode GNSS / inertia tight combination factor graph optimization method based on variance component estimation, and belongs to the technical field of satellite / inertia combination navigation. The implementation method comprises the following steps: fusing observed quantities of a multimode GNSS and an inertial navigation system on the basis of a factor graph optimization method under the condition of distributing prior weights among multiple GNSS systems; on the basis of a posteriori state estimation result obtained after factor graph optimization, combining a multi-mode GNSS observation equation, calculating a posteriori residual error, and utilizing a variance component estimation method to adaptively adjust weights among the multiple GNSS systems in real time according to the signal quality of satellites of the satellite navigation systems; on the basis of gross error detection and a weight reduction algorithm, weight reduction processing is carried out on the satellite observed quantity containing gross errors before the weight between the systems is adjusted in a self-adaptive mode. According to the invention, the robust variance component estimation is used to determine the weight among multiple GNSS systems, the reliability of satellite observation information of different satellite navigation systems is adjusted in real time, and the precision and robustness of the multimode GNSS / inertial tight integration navigation system are improved.

The invention provides a low-illumination target detection method for a multi-scale receptive field under adaptive enhancement, and the method comprises the steps: 1, introducing a Global module, an RCSOBEL module and an LSKADCNN module to optimize a yolov8 target detection algorithm, and constructing a low-illumination target detection model based on image adaptive enhancement and the multi-scale receptive field; 2, acquiring an image data set under a low-illumination condition, and processing the image data set; training the low-illumination target detection model by using the processed data set to obtain a trained low-illumination target detection model; and 3, inputting a to-be-detected low-illumination image into the trained low-illumination target detection model, and outputting a low-illumination target detection result. The method can achieve a good detection effect in a low-illumination image, and is suitable for target detection of images generated under various conditions of insufficient illumination, such as night monitoring and unmanned driving.

The invention relates to the technical field of power system load prediction, in particular to a photovoltaic output prediction method and device integrating satellite and ground data, equipment and a medium, and the method comprises the steps: carrying out the preprocessing of data, which comprises remote sensing inversion radiation data, ground observation radiation data and satellite remote sensing data; space grid division is carried out based on the spatial continuous remote sensing inversion radiation data, grid division is carried out based on geographic positions and meteorological characteristics, areas with different meteorological characteristics are divided into different grid units, and each grid unit at least comprises one ground observation station; establishing a solar radiation prediction model for each grid unit; according to the trained solar radiation prediction model, calculating a solar radiation prediction value of each grid in combination with a real-time satellite cloud picture and ground observation radiation data; the limitation that a traditional foundation method cannot accurately predict radiation fluctuation caused by cloud layer changes is overcome, and therefore the prediction precision of solar radiation and photovoltaic output is remarkably improved.

The invention relates to the technical field of combustion system optimization, and discloses a gas turbine combustion chamber outlet temperature distribution rapid prediction method and related equipment. The method comprises the following steps: collecting combustion chamber outlet temperature distribution data obtained through a CFD experiment mode; performing main mode extraction on the combustion chamber outlet temperature distribution data based on intrinsic orthogonal decomposition to obtain a POD coefficient matrix and a POD mode matrix; performing statistical fitting on the POD coefficient matrix by using a deep neural network, and establishing a POD coefficient prediction model; constructing an outlet temperature prediction model based on a parameterized reduced-order model based on the POD coefficient prediction model and the POD modal matrix; and utilizing the outlet temperature prediction model to perform rapid simulation prediction on the outlet temperature distribution of the combustion chamber under the combustion parameters to be optimized. According to the method, the calculation cost and time consumption of engine combustion chamber simulation can be effectively reduced, meanwhile, high simulation precision is kept, and powerful support is provided for optimization design of an advanced combustion chamber.

The invention discloses a laser radar positioning and three-dimensional mapping method based on an implicit neural field, and belongs to the technical field of positioning and mapping. The method comprises the following steps: for first frame data in radar point cloud data, converting an input sampling point from a sensor coordinate system to a world coordinate system by using a unit matrix or a constant matrix to initialize a map; the hierarchical potential features are optimized in a self-supervision mode, an octree data structure is used for storing the optimized potential features to generate a potential feature set, and features in the potential feature set are decoded into symbol distance values through a shallow multilayer perceptron (MLP); starting from a second frame of data of the radar point cloud data, obtaining a predicted pose through a matching mode from the frame to the implicit map every time a frame is input; converting a new input point into a world coordinate system by using the obtained predicted pose, and determining whether a new potential feature needs to be added or not according to a Morton value of the new input point; and completing the construction of the NSDF by repeating the operations of predicting the pose and expanding the NSDF.

The invention provides an overcurrent prediction method, medium and system for a liquid-cooling over-charging pile, and belongs to the technical field of electrical variable measurement, and the method comprises the steps: designing a time-sensitive HyperLogLog (TS-HLL) structure through collecting the charging parameters and heat exchange parameters of the liquid-cooling over-charging pile, and coding the parameters; statistical characteristics of charging parameters and heat exchange parameters are extracted by using a time sliding window technology, and a correlation matrix between the charging parameters and the heat exchange parameters is calculated. Based on the correlation matrix, a relation graph of the charging current and other parameters is constructed, a graph neural network is used for learning the relation graph, and a complex nonlinear relation between the charging current and other parameters is extracted. And finally, combining the learned features with historical charging data, inputting the combined features and historical charging data into a recurrent neural network, and predicting the charging current in a period of time in the future, thereby solving the problem of rapid over-current prediction of the liquid cooling over-charging pile, which is difficult to solve in the prior art.

The invention belongs to the technical field of wood defect detection, and particularly relates to a wood surface defect detection method based on improved YOLOv81, electronic equipment and a storage medium. The method comprises the following steps of: improving the Backbone of the traditional YOLOv81 by using Faster Net, and reducing the parameter quantity and the calculation quantity by using the design of partial convolution PConv in the Faster Net; an enhanced weighted bidirectional feature pyramid network EFPN is provided, information exchange between different feature layers is enhanced through an auxiliary branch, and an expansion re-parameter module DRM is designed to fully capture features on different scales. The improved YOLOv81 is more suitable for detecting the multi-scale defects on the surface of the wood, the detection precision is improved, the parameter quantity and the calculated quantity are smaller, and an efficient and accurate solution is provided for detecting the defects on the surface of the wood.

The invention discloses an inspection robot scheduling method and scheduling system. The method comprises the following steps: S1, obtaining task information of each order; s2, generating an initial scheme: calculating a target income of each order, traversing all orders according to the target income, determining a feasible inspection robot trip, calculating time cost, reserving the first a feasible schemes from small to large, and optionally selecting one of the feasible schemes to obtain a target feasible scheme; s3, optimizing a target feasible scheme: presetting four destruction operators and a reconstruction operator, and combining to obtain a plurality of operator combinations; and selecting an operator combination to execute deletion and redistribution operations to obtain a solution, calculating an expected income of the solution, continuously performing iterative optimization until the expected income of the solution is greater than that of the target feasible scheme, and selecting the solution with the maximum expected income as an optimal scheduling scheme. The optimal scheme can be obtained through finite iteration calculation, calculation consumption is small, the response speed is high, and the obtained scheme is more reasonable and reliable.

The invention relates to a mechanism output torque control method based on an electromechanical actuator, and belongs to the technical field of servo control. Building an electromechanical actuator transmission system; the tail end of the transmission mechanism is driven to complete a complete movement process; the actuating distance of a push rod of the electromechanical actuator and the angle of the tail end of the transmission mechanism relative to the horizontal direction in the whole movement process are measured, and relation calibration is carried out; performing relation calibration on the thrust of the electromechanical actuator and the torque of the tail end of the corresponding transmission mechanism; performing relation calibration on the thrust of the electromechanical actuator and the input current data of the electromechanical actuator; semi-closed-loop control is realized by controlling the input current of the electromechanical actuator and the torque at the tail end of the transmission mechanism; according to the invention, a five-order polynomial model fitting mode is adopted, indirect detection of the load torque is realized, and the problem of accurate control of the output torque when no torque feedback exists is solved.

The invention relates to an image pre-division method based on a multi-layer multi-motion search result, and belongs to the technical field of video coding. The method comprises the steps that an original image and a prediction search image are acquired, and according to the original image and the prediction search image, motion information of the original image is obtained through multi-layer multi-time motion search; obtaining a video sequence and extracting to obtain an image frame set, coding and extracting features of the image frame set to obtain an image pre-division data set; training and testing according to the image pre-division data set through a decision tree algorithm to obtain an image pre-division decision tree; and obtaining a current image frame, extracting current image motion information through the multi-layer multi-time motion search, and obtaining a current image pre-division state through the image pre-division decision tree according to the current image motion information, thereby assisting an encoder in encoding, accelerating the process of determining the encoding block division form, and improving the encoding efficiency. The calculation consumption of the encoder is reduced, and efficient and accurate image pre-division is realized.

The present invention provides a ligand screening model construction method and device, screening method, equipment and medium, belonging to the field of drug screening, the method includes obtaining a drug ligand training set, the drug ligand training set includes a drug ligand chemical formula and a classification label; based on the ligand chemical formula, drawing a ligand graph network, in which atoms are nodes and chemical bonds are edges connecting nodes; using a random initialization vector to identify the weight vector of each node in the ligand graph network; reconstructing each node of the ligand graph network according to the connection relationship of the ligand graph network to obtain a reconstructed graph network, repeating the reconstruction step to obtain at least two layers of reconstructed graph networks; performing deep learning on the ligand graph network and at least two layers of reconstructed graph networks according to the classification labels to construct a ligand screening model. Through the processing scheme disclosed in the present invention, the properties of compounds in biological and chemical experiments of different scales are accurately predicted.

The application discloses a method for quickly optimizing a trajectory of a robot of an arbitrary shape based on RC-ESDF, which comprises the following steps: generating an occupancy grid map according to the shape of the robot, thereby establishing RC-ESDF on the mechanical system of the robot, wherein the ESDF value of a grid vertex inside the robot is negative and the absolute value is the nearest distance to the surface of the robot, and the ESDF value of a grid vertex outside the robot is 0; obtaining an initial trajectory of the robot; based on the initial trajectory, constructing a target function for trajectory optimization, wherein the target function comprises a collision penalty item for the joint optimization of the position and the yaw angle of the robot; and solving the target function to obtain an optimized trajectory, so that the robot moves according to the optimized trajectory. The application reduces the calculation consumption based on RC-ESDF and is suitable for robots of an arbitrary shape; and based on the joint optimization of the position and the yaw angle, the robot can pass through a narrow area in a complex environment in a suitable pose.

The invention discloses a photoetching simulation method and system for decomposing a superposition pupil shift matrix based on a circular sampling function, and the method comprises the steps: carrying out the sampling of a shape function and a pupil function of an extended light source, and obtaining a corresponding superposition pupil shift matrix; processing the superposed pupil shift matrix, the extended light source and the frequency domain circle sampling function matrix of the pupil plane to obtain a frequency domain cross transmission coefficient matrix and a projection matrix; performing singular value decomposition processing on the projection matrix to obtain a singular value decomposition result; processing the spatial domain circular sampling function matrix based on a singular vector to obtain a convolution kernel matrix; constructing a spatial domain image simulation model based on the singular value decomposition result, the convolution kernel matrix and the initial Hopkinson model; and obtaining a spatial domain image simulation result based on the spatial domain image simulation model. According to the method, direct calculation and storage of the TCC matrix can be reduced by using a pupil shift matrix superposition method and analytic circle sampling function decomposition, and calculation consumption required for obtaining the kernel is remarkably reduced.

The present invention discloses a wavenumber domain three-dimensional ultrasound full-matrix imaging method based on CUDA. The present invention represents the original full-matrix data in the form of a five-dimensional matrix, converts the obtained original full-matrix data into the wavenumber domain through a five-dimensional fast Fourier transform, realizes the wavenumber conversion between the original data coordinate system and the imaging coordinate system by means of Weyl's Identity, obtains the mapping of the image coordinate system through eight linear interpolation, and then obtains the three-dimensional image through a three-dimensional fast inverse Fourier transform. The present invention adopts the VTK module and uses the MIP (maximum intensity projection) ray projection method to obtain the rendered two-dimensional image, and can change the observation angle and distance of the three-dimensional image in real time. The present invention has the real-time high-efficiency, high-quality three-dimensional imaging and two-dimensional rendering capabilities for variable regions of interest, can be flexibly transplanted to computer equipment with a CUDA computing platform, and can cope with the rapid processing of large amounts of three-dimensional full-matrix data.

The invention discloses a mobile robot autonomous exploration method based on a topological graph, and relates to the field of intelligent robots. According to the method, a framework topological structure of an environment is introduced, exploration behaviors are guided by utilizing topology of the environment, and the method comprises incremental framework extraction, framework topological graph updating and exploration strategy finite-state machine. According to the method, the BFMs phenomenon is relieved to a great extent, complex TSP problem solving is avoided, and efficient exploration incremental skeleton extraction is achieved with extremely low calculation consumption; the exploration behavior of the robot is effectively guided, invalid movement is remarkably reduced, information acquisition is maximized, and resource consumption is minimized.

This invention provides a multi-layer memory-enhanced neural network architecture, training method, device, and medium, including a backbone network and an attractor network module. The backbone network extracts features from the input data and generates feature representations for the intermediate layers. The attractor network is connected to the intermediate layers of the backbone network and adopts a Hopfield network structure. Through iterative operations, it enhances the memory of the intermediate layer features of the backbone network, enabling the intermediate layer features to recover to a stable feature pattern even in noisy environments. By embedding the attractor network module into a traditional neural network architecture, this invention effectively solves the problem of insufficient robustness of existing deep neural networks when facing noise or unknown perturbations. The attractor network can memorize the feature patterns of the intermediate layers of the backbone network and correct features affected by noise during the inference stage, causing them to converge to a stable memory pattern. This characteristic makes the model more stable and reliable in complex environments.

The invention discloses a method and device for closing internal and external corners, and the method comprises the steps: calculating a data structure of a plane model on each side at the internal and external corners according to the heights of the plane models on the two sides of the internal and external corners and the angles of the internal and external corners, the data structure comprising the lifting height and the horizontal offset of points on the plane models and the ID of a region to which the data structure belongs; performing oblique modeling on the plane modeling on each side according to the data structure to generate an extension body, wherein the extension body can form internal and external corners of the plane modeling and realize closing of the internal and external corners. According to the oblique modeling mode, the topological structure is not changed according to the modeling result, only the position lengths of the points, the edges and the faces are changed, and a series of follow-up supports such as paving, corner lines and lamp strips related to the modeling result are naturally supported or relatively easy to achieve. Moreover, the judgment of the boundary conditions can increase the complexity of program calculation, and excessive boundary conditions do not need to be considered, so that the calculation consumption is reduced.

The application discloses a trap glycan library construction method and system based on topological graph editing and global optimization, and belongs to the technical field of glycan analysis, which comprises the following steps: analyzing original glycan into a topological structure, calculating the number of each preset type of monosaccharide to obtain a monosaccharide composition vector, applying numerical disturbance to the monosaccharide composition vector to obtain a candidate trap monosaccharide composition vector, and then calculating the difference between the monosaccharide composition vector and the candidate trap monosaccharide composition vector, so as to edit the topological structure to obtain an effective candidate trap topological structure; and performing global distribution on the original glycan group and the effective candidate trap glycan group to screen out a unique effective candidate trap glycan group, and combining all the selected effective candidate trap glycan groups to form a trap glycan library. The application aims to solve the problems of unreasonable trap glycan structure, large quality distribution deviation and difficulty in processing isomers in the prior art, so as to improve the accuracy of glycomics identification error rate estimation and quality control.

This invention relates to the field of weld defect type identification technology. Existing weld defect detection methods fail to achieve intelligent and efficient detection due to the diverse types of weld defects. This invention provides a low-cost self-learning neural network design method for ultrasonic detection of weld defects. It performs high-dimensional spatial domain feature representation on the initial one-dimensional ultrasonic signal, enriches the feature expression of weld defect data, selects the feature domain with better performance, constructs an adaptive scaling network for weld defect detection, and uses a multi-objective, training-free network search and evaluation method to iteratively search and evaluate candidate networks for weld defect detection, obtaining a balanced solution on both network classification accuracy and parameter quantity. Ultimately, it achieves the selection of a weld defect type detection network with superior overall performance without any training.

The present application relates to the technical field of pedal data processing, and particularly relates to a method and system for replacing pedal data in a playing file. The method comprises: obtaining a playing file of a piece of music; obtaining a score file corresponding to the piece of music; generating an insertion node according to the score file and a pedal insertion rule, and / or according to the playing file, wherein the pedal insertion rule comprises a chord rule; selecting replacement data of a pedal in response to a selection signal of a user, wherein the selection signal comprises a specified style or a specified pedal file; and inserting the replacement data into the insertion node to form a new playing file. The present application can meet the demand of users who want to reference pedal data of others by replacing pedal data, and greatly improves the efficiency of music production, teaching assistance, performance analysis and the like.

The present application relates to the technical field of three-dimensional reconstruction, and in particular to an object three-dimensional reconstruction system based on deep learning, which introduces an adaptive cost aggregation method with visibility perception for cost volume aggregation, acquires the visibility of pixel points in the view through a network, and can improve the reconstruction integrity of the occluded area; a variance-based method is used to predict the disparity range of each pixel, and a spatially-varying depth hypothesis surface is constructed for depth estimation in the next stage, and a residual and channel attention guided fusion depth map optimization module is proposed in the last stage to obtain an optimized depth map; an improved depth map fusion algorithm is used to combine the pixel point and 3D point re-projection error for consistency checking to obtain a dense point cloud. Quantitative and qualitative comparison results of the present application and other methods on the DTU dataset show that the present application can reconstruct a scene with better details, and achieves the purposes of reducing GPU memory consumption and computation time.

The invention relates to the technical field of electric digital data processing, and discloses a task branching control method and system driven by pre-analysis and pre-processing of a large language model, and the method comprises the steps: analyzing a to-be-processed document set, generating semantic data blocks, and distributing an index identifier with a hierarchical relationship for each semantic data block; performing feature extraction on the semantic data blocks by adopting a pre-constructed feature extraction model to obtain a multi-dimensional feature set so as to construct a structured index table, and dynamically retrieving a candidate data block list from the structured index table according to a received user query task; generating a data block loading strategy according to the multi-dimensional feature set associated with each candidate data block in the candidate data block list; and according to a data block loading strategy, performing branching control on processing of the selected quantized large language model on the candidate data blocks to obtain a task processing result. According to the method provided by the invention, the processing capability of the quantized large language model on complex and long document tasks is improved, and the computing power consumption is reduced.

The embodiment of the invention provides a video playing method and device, equipment and a storage medium, and relates to the technical field of computers. The method comprises the steps that a source video is acquired, the source video is displayed in a terminal screen in a first display mode, and the first display mode is one of a horizontal screen mode and a vertical screen mode; cutting the source video to obtain a first video displayed in a second display mode, the second display mode being the other one, different from the first display mode, of the horizontal screen mode and the vertical screen mode; and playing the first video in the terminal screen in the second display mode. According to the method, after the terminal device obtains the source video, the terminal device cuts the source video to adapt to the display mode of the terminal screen, the server does not need to store the source videos corresponding to different display modes respectively, the calculation consumption of the server is reduced, meanwhile, the storage resources, occupied by the source video, of the server are reduced, and the user experience is improved. And releasing the storage space of the server.

The invention discloses an unmanned aerial vehicle road inspection dynamic obstacle elimination algorithm based on a clustering principle, and relates to the field of unmanned aerial vehicles. When an unmanned aerial vehicle carries out road inspection work, environment data of a road to be inspected is extracted through a laser radar and a visible light camera, point cloud data obtained by the laser radar are processed, and the road inspection dynamic obstacle elimination algorithm is obtained; a large amount of redundant information is eliminated, useful point cloud data is extracted, calculation consumption is reduced, dynamic obstacles in an inspection road are eliminated more quickly, fusion operation is performed on two kinds of heterogeneous data, cloud data are converted into pixel points of color texture data, subsequent clustering analysis is facilitated, calculation time cost is reduced through clustering analysis, and the calculation efficiency is improved. The real-time performance is improved, the dynamic point cloud target is identified to facilitate the later elimination work, the point cloud data and the color texture data are spliced and reconstructed to complete the elimination of the dynamic obstacle in the unmanned aerial vehicle routing inspection road, and the unmanned aerial vehicle can conveniently perform the road routing inspection work.

The invention provides a large-scale reservoir model reconstruction method and device based on a Transform-CNN architecture, and relates to the field of three-dimensional geological modeling and deep learning, and the method comprises the steps: S1, constructing an initial MAE network based on a Transform network and a convolutional layer; s2, training the initial MAE network based on progressive generation and partition simulation to obtain a large-scale MAE network; and S3, inputting the reservoir model to be reconstructed into the large-scale MAE network to obtain a reconstructed reservoir model. According to the method, the initial MAE network is constructed based on the Transform network and the convolutional layer, so that the parameter quantity of the neural network is effectively reduced, and the modeling scale is improved. An initial MAE network is trained in a progressive generation and partition simulation mode, a large-scale MAE network is finally obtained by iteratively training different scales of MAE networks, the large-scale MAE network can consider the size of a complete model, the overall space structure of a large-scale reservoir model is effectively restored, and the problem of error propagation in a traditional progressive generation method is solved.

The invention provides a display control method and device in a game, electronic equipment and a storage medium, and the method comprises the steps: responding to a game image drawing instruction, and recognizing a target line-shaped element in a visual range; determining a feature adjustment value of each vertex according to a position relationship between each vertex on the target line-shaped element and a virtual camera; and drawing a game image based on the feature adjustment value of each vertex on the target line-shaped element. According to the invention, the calculation consumption can be effectively reduced.

The invention discloses a three-dimensional geologic structure modeling method and system based on SH-LDM and a medium, and relates to the technical field of deep learning and geologic modeling crossing, and the three-dimensional geologic structure modeling method based on SH-LDM mainly comprises the steps: carrying out the preprocessing of original data, and obtaining a training data set and a test data set; training the multi-branch self-encoding network by using the first-stage loss function, and training the multi-condition diffusion network by using the second-stage loss function; and according to soft data and hard data of a real scene, utilizing the trained multi-branch self-encoding network and the multi-condition diffusion network to obtain a three-dimensional geologic structure generation image. By implementing the SH-LDM-based three-dimensional geologic structure modeling method and system and the medium provided by the invention, high-precision and efficient modeling of a complex non-stationary underground structure can be realized.

This invention discloses a cross-domain authentication method for trusted data spaces based on digital certificates, belonging to the field of trusted data space authentication technology. The method includes S1, constructing a trusted data space cross-domain authentication system; S2, system initialization; S3, trusted data space and user registration; S4, intra-domain two-way authentication between the trusted data space and the user; and S5, cross-domain authentication and key negotiation between the trusted data space and the user. This invention combines a lightweight algorithm to achieve lightweight intra-domain authentication and secure cross-domain communication. Addressing the single point of failure problem in traditional authentication, it designs a blockchain distributed verification mechanism to avoid the risk of system paralysis by a centralized trusted party. During the authentication process, the identity information of the trusted data space and the user is anonymized through pseudo-identity and hash processing. Simultaneously, location privacy and communication security are protected through timestamps and digital signatures. The blockchain stores a permission table, simplifying the cross-domain verification steps for authenticated entities and reducing system computation and communication costs.

This invention discloses a method, apparatus, device, medium, and product for optimizing computing power node scheduling. The method includes quantifying the allocable computing power of each node in a computing power cluster; obtaining a set of nodes that meet the computing power threshold requirements based on the quantified cluster computing power matrix and a preset computing power threshold matrix; and using a greedy binary search algorithm based on the node set and target business requirements to find a subset of nodes corresponding to each type of computing power that meets the target business requirements, thereby optimizing computing power node scheduling. This method can quickly filter the remaining computing power matrix using the computing power quantification method and calculate each node that meets each type of computing power requirement based on the greedy binary search algorithm, reducing the computational cost and algorithm complexity of computing power node scheduling.