33 results about "Shape vector" patented technology

The invention discloses a large language model low-delay reasoning method based on dynamic reasoning graph optimization, and provides a low-delay reasoning method based on dynamic reasoning graph optimization. Constructing a template inference graph which can be rewritten and replayed, and establishing a template library according to input shape vectors; during reasoning, a template is matched with a distance threshold value, and only the attention / feedforward sub-graph is locally recaptured when the distance threshold value exceeds the threshold value; executing a forward execution graph, injecting a key value cache page pointer, numerical value precision and an adapter identifier, and performing playback; dividing the pre-filling and decoding sub-graphs to implement graph-level scheduling; switching the key operator when the key operator operates between a standard / fast kernel and different precisions; page-level backspacing of speculative branches is achieved through a shadow page table and reference counting, and batch and template selection is adjusted in a self-adaptive mode based on online indexes; compared with the prior art, the method has the advantages that the recapture and start overhead is reduced, tail delay and jitter are inhibited, and the hardware utilization rate and the service stability are improved.

The invention discloses a space target three-dimensional reconstruction realization method based on radiation field generation. The method comprises the following steps: acquiring image data of a space target for preprocessing; an improved generated radiation field network is built, and an image is sampled to obtain an image patch; obtaining a group of three-dimensional sampling points in a layered sampling mode, performing multi-frequency position coding, splicing with a shape vector, sending into a shape coding layer to obtain feature representation, and inputting into a color head together with a sight direction code and an appearance variable to calculate the color of the sampling points; for each ray, colors of all sampling points are synthesized into a final pixel color based on a volume rendering equation, and the final pixel color is combined into a patch image to be output; inputting the real patch and the patch image into a discriminator, and calculating confrontation loss and discrimination loss; and acting the weight obtained by training on a generator to obtain rotating videos of the space target at different visual angles. The method can realize three-dimensional reconstruction of a space target, and has the advantages of flexible data driving, cross-view geometric consistency, efficient training and the like.

The invention provides a method and device for reasoning by using an open neural network switching ONNX model based on a fusion operator, electronic equipment and a medium. The method comprises the steps that shape information of an input tensor of the ONNX model is obtained, the shape information is a vector with the dimension being N, each element of the vector indicates each dimension of the input tensor, and N is larger than or equal to 1; generating an input shape vector based on the shape information, the input shape vector being a vector with a dimension of N + 1; the fusion operator is used for operating the input shape vector to obtain a changed shape vector, the fusion operator is fused with at least one ONNX operator, the fusion operator is based on a coefficient matrix, and one dimension of the coefficient matrix is N + 1; and based on the changed shape vector, reasoning is performed by using the ONNX model.

The invention discloses an analysis method for a design image of a power transmission tower. The analysis content comprises information extraction of shape, parameterization, layout and the like of a design structure. Firstly, an original design drawing is converted into a clean topological mask drawing through a semantic segmentation network, and the shape of a design structure is obtained; then, through a vector extraction network, extracting a skeleton center vector and endpoint information from the topological mask map; then, numbers and symbols on the image and in the table are recognized and extracted, corresponding numbers beside the mask image are numbered, the same numbers are found in the table, the actual size corresponding to the numbers is obtained, and parameterization assignment of the actual size of the shape vector is achieved; and finally, constructing a visual angle classification network, analyzing the image layout of each shape, and obtaining the visual angle orientation of the three-dimensional model corresponding to the planar graph. The method lays a foundation for realizing automatic three-dimensional modeling of the drawing, and can be expanded to be used for intelligently verifying, improving and checking the design drawing of the iron tower and the like.

The invention discloses an intelligent virtual accompanying robot scene dialogue rhythm control system, and belongs to the technical field of robot dialogue. Comprising a data acquisition module used for acquiring voice data of historical speaking; the historical vector extraction module is used for storing the low-dimensional rhythm shape vectors in a vector database; the online clustering module is used for generating a plurality of low-dimensional rhythm clusters and response templates; the current vector extraction module is used for extracting a rhythm shape vector; the similarity acquisition module is used for acquiring the similarity; the rhythm response module is used for calling a rhythm response template; and the new cluster generation module is used for generating a new rhythm cluster on line and creating a new template. According to the invention, the voice rhythm of the user is compressed into the low-dimensional shape vector, the exclusive rhythm cluster is formed through online clustering, and zero-delay matching is carried out in the real-time dialogue in a sliding slicing manner, so that the technical effects of instant response to the dialogue rhythm and long-term memory of the personal style are achieved; the problems of rhythm response lag and lack of context long-term memory in the prior art are solved.

A computer-implemented method and apparatus for generating a quadrilateral grid from an image with an alpha channel. The method includes receiving an image with an alpha channel depicting a shape mapped onto a 3D surface, vectorizing the shape to create a vector path, determining corners by calculating a centroid and identifying the points, correcting a winding order to ensure a clockwise orientation, constructing a quadrilateral grid by setting sub-paths based on detected corners, resampling the sub-paths to match specified rows and columns, performing weighted interpolation across the grid using inverse square distance weighting, and outputting the generated quadrilateral grid for display on a computer device. The apparatus comprises a processor configured to execute these steps, including vectorizing the shape, determining corners, correcting the winding order, constructing and resampling the grid, performing weighted interpolation, and subdividing the 3D mesh for flexible manipulation and extrapolation.

An improved gain-shape vector quantization is achieved by determining a number of bits to be allocated to a gain adjustment-and shape-quantizer for a plurality of combinations of a current bit rate and a first signal property. The bit allocation is derived by using an average of optimal bit allocations for a training data set. A number of bits to the gain adjustment and the shape quantizers for a plurality of combinations of the bit rate and a first signal are pre-calculated, and a table indicating the number of bits to be allocated to the gain adjustment-and the shape-quantizers for a plurality of combinations of the bit rate and a first signal property is created. In this way, the table can be used for achieving an improved bit allocation.

The application discloses a large language model low-delay inference method based on dynamic inference graph optimization, and proposes a low-delay inference method based on dynamic inference graph optimization; a rewritable and replayable template inference graph is constructed, and a template library is established according to an input shape vector; during inference, a template is matched according to a distance threshold value, and only attention / forward subgraphs are locally recaptured when the threshold value is exceeded; a forward execution graph injects a key-value cache page pointer, a numerical precision and an adapter identifier, and is replayed; pre-populated and decoded subgraphs are divided to implement a graph-level scheduling; a key operator is switched between a standard / quick kernel and different precisions during operation; a page-level rollback of a speculative branch is realized through a shadow page table and a reference count, and a batch and a template selection are adaptively adjusted based on online indicators; compared with existing schemes, the application reduces recapture and startup overhead, suppresses tail delay and jitter, and improves hardware utilization and service stability.

The invention discloses a 60-degree phase belt pole-changing double-speed single winding wiring method, which is characterized in that a winding is a three-phase winding, and comprises the following steps: respectively drawing star-shaped vector diagrams of two pole numbers by adopting a 60-degree phase belt method; comparing the star vector diagrams to find out A-X slot numbers of one pole number and A-X slot numbers of the other pole number, wherein the slot numbers of the same phase form a basic unit; repeating the operation to respectively find out basic units in B-Y and C-Z; according to a basic unit rule, sequentially arranging the rest slot numbers in A-X, and finding basic units of the rest slot numbers; and repeating the operation to respectively find out the basic units of the remaining cell numbers in B-Y and C-Z. The pole-changing double-speed single-winding structure has the beneficial effects that the pole-changing double-speed single-winding structure which is high in winding coefficient, few in outgoing line ends, free of subharmonic and circulating current, high in efficiency, low in cost, simple in structure and convenient to control is designed, the main performance of the motor can reach the level of a single-speed motor with corresponding capacity, and the requirement for energy conservation is met.

The present application discloses a rod-shaped object vector graphics generation method and device, which solves the problem that the prior art drawing of rod-shaped object vector graphics is affected by incomplete bottom or top point cloud collection. A rod-shaped object vector graphics generation method, used for high-precision maps of point cloud data, includes: obtaining point cloud data of the target area; responding to GUI to capture point cloud data points, determining the plane coordinates of the first point; the first point is any point in the middle point cloud of the rod-shaped object; determining the first elevation value of any point on the horizontal ground where the rod-shaped object is located, and determining the second point of the rod-shaped object on the ground through the first elevation value and the plane coordinates; determining the second elevation value of the third point at the top of the rod-shaped object according to the length of the rod-shaped object; and determining the rod-shaped object line with the second point and / or the third point. The present application can be compatible with some of the problems of its data and perform specific processing on rod-shaped objects, which can improve the efficiency and accuracy of acquisition production and ensure the current demand for high-precision maps for intelligent driving.

A method for analyzing change in a mode of a unidirectional composite material includes applying a physical force of a predetermined pattern onto a unidirectional composite material specimen; sensing a vibration signal generated by the physical force at at least one sensed position thereof; performing modal analysis of a frequency response at a corresponding measurement location, based on the physical force applied to the specimen and the vibration signal measured at the sensed position of the specimen, and calculating at least one mode shape vector of the specimen based on the modal analysis result; compensating the calculated mode shape vector based on a distance between a normal line passing through a center of the specimen and the sensed position, thereby calculating a modified mode shape vector; and calculating a first modal assurance criterion (MAC) of the specimen based on the modified mode shape vector.

The application discloses a multi-level human body grid reconstruction method and system based on millimeter waves. The method comprises the following steps: for a target human body, collecting reflection signals of a millimeter wave radar and performing filtering pretreatment to obtain pretreated millimeter wave radar signals; based on the millimeter wave radar signals, extracting multi-level two-dimensional heat maps, including a heat map generated by a moving body part, a heat map generated by a trunk part and a heat map generated by a static body part; inputting the multi-level two-dimensional heat maps into a trained human body 3D grid estimation model to obtain a prediction value of a human body network, including a posture vector, a shape vector and a translation vector of a human body grid; and inputting the prediction value of the human body grid into a neutral parameterized human body model to estimate skeleton and grid information of the target human body. The application can comprehensively capture features of each part of the target body, especially enhances the recognition ability of static parts, and improves the reconstruction accuracy and integrity of each part of the human body.

The application discloses a smart virtual accompanying robot scene dialogue rhythm control system and belongs to the technical field of robot dialogue. The system comprises the following modules: a data acquisition module for acquiring voice data of historical speeches; a historical vector extraction module for storing low-dimensional rhythm shape vectors in a vector database; an online clustering module for generating multiple low-dimensional rhythm clusters and response templates; a current vector extraction module for extracting rhythm shape vectors; a similarity acquisition module for acquiring similarity; a rhythm response module for calling rhythm response templates; and a new cluster generation module for generating new rhythm clusters online and creating new templates. The application compresses user voice rhythm into low-dimensional shape vectors, forms exclusive rhythm clusters through online clustering, and matches in a sliding slice manner with zero delay in real-time dialogue, so that the technical effect of instant response to dialogue rhythm and long-term memory of personal style is achieved, and the problems of rhythm response lag and lack of context long-term memory in the prior art are solved.

The present invention belongs to the technical field of structural health monitoring, and discloses a time-domain multi-measurement group operational modal identification method, device and terminal. The time-domain signals of a large civil structure to be monitored are measured in a multi-measurement group manner, the collected time-domain signals are analyzed, and a time series prediction recurrence formula is constructed; the integration and fusion of multi-measurement group information and the construction of a joint likelihood function are carried out, and a time-domain non-deterministic modal analysis system is constructed based on the multi-measurement group joint likelihood function and the modal parameter joint prior distribution; time-domain modal non-deterministic judgment and hyperparameter joint inference are carried out to obtain the time-domain modal parameter joint probability distribution, and the prediction and reconstruction of the time-domain free response are carried out. The present invention can naturally realize the effective fusion of any multi-group measurement data in Bayesian time-domain modal analysis, and fundamentally solves the difficulty of unifying the directions of local mode shape vectors of each measurement group during the assembly process.

The present invention relates to an adaptive beam tracking method, comprising the following steps: S1: a base station estimates channel state information based on an uplink detection signal transmitted by a terminal; S2: a shaping vector is obtained based on the channel state information; S3: the shaping vector is corrected to obtain a corrected shaping vector; S4: data symbols of a drone are weighted based on the corrected shaping vector to obtain a transmission signal; S5: the transmission signal is transmitted from the drone's antenna array. The adaptive beam tracking method of the present invention uses the drone's real-time track information to correct the beam, thereby reducing beam direction deviations caused by non-ideal factors and improving beam tracking accuracy; due to the use of a circular antenna array, no matter the target is in any direction on the horizontal plane, there is no need to drive the antenna to mechanically rotate, thereby quickly tracking targets in any direction.

The invention provides a modal parameter identification method, and relates to the technical field of structural health monitoring, vibration engineering and signal processing. The method comprises the following steps: acquiring an acceleration time domain signal; decomposing the acceleration time domain signal, and performing signal reconstruction on a plurality of multivariate intrinsic mode function components; smoothing each channel of data of the reconstructed signal, and constructing a Toeplitz matrix by using a cross-correlation function sequence; singular value decomposition is carried out on the matrix, and a discrete state space matrix is obtained through calculation by adopting a covariance-driven random subspace method; the discrete state space matrix is decomposed, and the inherent frequency, the damping ratio and the vibration mode vector of the ith-order mode are obtained through calculation; and carrying out statistical analysis on all modal parameters, and verifying to obtain a finally confirmed real modal parameter set. According to the method, multi-channel signals can be comprehensively processed, excitation interference of a complex environment is effectively suppressed, and a robustness method of full-automatic high-precision modal recognition is realized.

The invention discloses an adaptive beam tracking method based on Beidou position correction, which can quickly and accurately track a target in any direction, and the technical scheme comprises the following steps: S1, a base station estimates CSI (Channel State Information) according to an uplink detection signal transmitted by a terminal; s2, a plurality of adjacent antennas with the strongest power are selected from the horizontal plane and the vertical plane respectively, and corresponding estimation channels are extracted; s3, calculating a shaping weight w according to the estimation channel; and S4, correcting w according to the spatial position information of the unmanned aerial vehicle and the task unmanned aerial vehicle, weighting the data symbol s by using the corrected shaping vector w to obtain a transmitting signal x, and transmitting the transmitting signal x from the antenna array.

The invention discloses a welding seam identification method and system based on single-view point cloud feature extraction, and relates to the technical field of welding identification, and the method comprises the steps: obtaining a workpiece three-dimensional cleaning point cloud data set under a current view point; acquiring a sphere neighborhood point cloud set corresponding to each single point cloud, and acquiring single point cloud geometric features corresponding to the single point cloud; performing binary clustering on the feature vectors in the feature vector set by adopting a clustering algorithm; obtaining a viewpoint vector and a shape vector; and identifying non-weld convex points and weld concave points according to the included angle between the viewpoint vector and the shape vector. According to the method provided by the invention, the real welding seam point cloud can be identified without initial welding seam fitting, the real welding seam point is identified by adopting the same concept for a straight welding seam and a curve welding seam, and the universality is high; meanwhile, the technical problem that in the prior art, after the potential welding seam area is recognized, distinguishing of the real point cloud of the welding seam is lacked, and consequently the welding seam recognition precision is low is solved.

The application provides a kind of structural modal parameter intelligent screening method and system based on feature fusion, it is related to engineering structure health monitoring and machine learning cross technical field. Including the following steps: S1, utilize preset time window to carry out block processing to structure response signal;S2, by SSI-Cov, modal parameter identification is carried out to the data block, and the initial modal parameter set containing frequency, damping ratio and mode shape vector is obtained;S3, based on the initial modal parameter set and the structure response signal, construct multi-dimensional feature fusion vector, input the multi-dimensional feature fusion vector into pre-trained classification model, output modal true and false probability, and screen;S4, based on the stable modal parameter obtained by screening, repeat execution for subsequent data block, and output final modal parameter;The application improves the false modal rejection rate, solves the problem that false modal rejection is not thorough;Reduce operation and maintenance cost, improve error identification accuracy.

This invention belongs to the field of image processing and provides a method and system for predicting the short-term posture and behavior of nematodes based on deep learning. The method includes acquiring and preprocessing the original image of the nematode to obtain a preprocessed nematode image; extracting the nematode contour based on the preprocessed nematode image to obtain a nematode contour image; calculating the nematode shape vector based on the nematode contour image to obtain a standardized nematode shape vector sequence; and predicting the nematode posture using a pre-trained nematode posture and shape prediction model based on the standardized nematode shape vector sequence to obtain the short-term posture and behavior prediction result. This invention provides a shape description method for quantitatively describing and characterizing the body shape of nematodes, which simplifies the behavior and body posture of nematodes, ensures the uniqueness of the nematode body shape, and improves the accuracy of model prediction.

The invention relates to the technical field of intelligent manufacturing, in particular to a plasticizing order matching method, which comprises the following steps of: encoding an order and a production line data event, and generating a form vector and an energy vector through a neural network; constructing a coupling matrix in the hyperbolic space and performing primary matching by using an Isin model; if the output gradient exceeds the threshold, refining is carried out through neuromorphic suppression, and a local entropy gradient is calculated; feeding a refining result and production line feedback into a causal generation graph to judge conflicts, and modulating a solver to generate deviation correction matching if necessary; and finally, the matching vector and the conflict mark are written into the block chain, and each node automatically updates the model through federal learning and drives curvature and threshold value self-adaption through morphological entropy. According to the method, real-time production scheduling with low energy consumption, less line change and sustainable learning is realized.

The present application relates to the technical field of image segmentation, and particularly relates to a quantitative evaluation method of momordica grosvenori based on image segmentation. The method comprises the following steps: performing spectral decomposition on momordica grosvenori data image information by using a spatial-frequency feature component decoupling algorithm to generate decoupled light and shadow features; extracting cross-scale deep semantic information to construct a semantic probability distribution field, performing edge tensor correction and generating a pixel-level fruit body lesion semantic deconstruction set; performing contour topology mapping and spatial domain gravity field analysis to obtain a geometric shape vector represented by a geometric moment operator, a lesion distribution density and a surface texture loss weight value; and performing a preset trait residual energy level discriminant to perform nonlinear dimension reduction mapping and output an evaluation grade. Through deep coupling of signal decoupling, semantic field analysis and geometric topology measurement, the present application establishes a logical mapping link from pixel semantic features to physical energy levels, realizes image signal representation of momordica grosvenori irregular edges in a complex background, and realizes nonlinear coupling between pixels and physical trait residual energy level attributes.

The invention provides an optimal arrangement method of construction scaffold damage identification sensors, which comprises the following steps: in response to an analysis instruction, performing modal analysis on a scaffold numerical model constructed by adopting finite elements in advance to obtain a multi-order modal; constructing a first target information matrix according to the vibration mode vector of each order modal, and determining a modal order selection range in the multi-order modal according to the change rate of the maximum information amount represented by the first target information matrix under different orders; according to a structural strain energy algorithm, determining structural strain energy corresponding to each modal order in the modal order selection range; selecting a plurality of target modal orders in the modal order selection range according to the structural strain energy corresponding to each modal order; based on a plurality of selected target modal orders, determining a target measurement point quantity; and determining a sensor arrangement strategy by using a target algorithm according to the target measurement point quantity and the selected multiple target modal orders. According to the invention, scientificity and high efficiency of sensor arrangement can be improved.

An audio encoder and an audio decoder and methods therein for gain shape vector encoding and decoding. The method performed by an encoder, where the encoding of each vector segment is subjected to a constraint related to a maximum number of bits, BMAX, allowed for encoding a vector segment. The method comprises determining an initial number, Np_init, of segments for a target vector x and determining energy of each Np_init target vector segment. The method further comprises determining an average number of bits per segment, BAVG, based on a vector bit budget and Np_init. The method further comprises determining a final number of segments to be used in the gain shape vector encoding based on energies of the Np_init segments and a difference between BMAX and BAVG. The performing of the method enables an efficient allocation of the bits of the bit budget over the target vector.

This invention discloses a novel electromagnetic characteristic calculation method for electrically large, uncertain metal targets based on AWE technology. First, a model is established using FEKO software and NURBS technology, with the coordinates of all control points of the target's shape defined as shape vectors. Random variables are introduced to change the target's shape. The shape vector is then introduced into the hybrid field integral equation. By expanding the impedance matrix, right-hand side vector, and current into Taylor series form, an expression for the current moment vector is derived. A pseudospectral method is applied to AWE to calculate the product of the matrix derivative and the moment vector, as well as the derivative of the right-hand side vector, to further calculate the current moment vector. The current after each model change is calculated by combining the Taylor series and Padé polynomials. Finally, the radar cross-section is calculated by combining the current and the coordinate information after the model change. This method significantly reduces calculation time and rapidly calculates the scattering characteristics of electrically large metal targets after deformation.

Techniques for aligning object representations for animation include analyzing a source object representation and a target object representation to identify a category of the source object and a category of the target object. Based on the category or categories of the objects, a feature extraction machine learning models is selected. The source object representation and the target object representation are provided as input to the selected feature extraction machine learning model to generate respective semantic descriptors and shape vectors for the source and target objects. Based on the semantic descriptors and the shape vectors for the source and target objects, an alignment machine learning model generates an aligned target object representation that is aligned with the source object representation and usable for animating the target object.

A face drive generation method based on neural radiance field includes the following steps: S1, extracting feature points of a face region from image frames of a video, calculating expression features and posture features of a corresponding three-dimensional face model according to the feature points, and extracting speech features from a speech part of the video; S2, for each position in a three-dimensional space where the three-dimensional face is located, predicting a mixed shape vector for controlling the three-dimensional face, a posture correction coefficient, and a linear mixed skinning weight through a neural network; S3, based on a deformation technology of the three-dimensional face model, calculating a deformed position of each current position according to the posture features and the speech features obtained in step S1 and the parameters predicted in step S2; S4, predicting a flow field according to each calculated position, aggregating feature information of adjacent time through a multi-frame set manner, and optimizing a face reconstruction loss by using time aggregation, so as to constrain smooth transformation of a dynamic neural radiance field in time. The method improves the accuracy of expression deformation, and makes the generated video more smooth and consistent.

The application provides a trajectory similarity calculation method and system based on road network topology perception and multi-view fusion, comprising a road section encoder, a road section-trajectory joint encoding module, an image encoder and a sequence encoder; the road section encoder is used for encoding road sections to form a road section sequence, the road section-trajectory joint encoding module is used for encoding trajectories according to the road section sequence to generate a road section granularity trajectory vector representation of the trajectory; the image encoder is used for encoding visual images of the trajectory to generate a geometric shape vector representation of the trajectory; the sequence encoder is used for generating a motion state vector of the trajectory; and the three vectors are aggregated to form a final vector representation. The application can effectively deal with problems such as uneven sampling of trajectory data, positioning noise and data skew distribution while ensuring that the calculation speed is optimal, and generates a trajectory vector representation with higher robustness and discrimination, so that a more accurate and efficient trajectory similarity calculation method is realized.

The invention discloses a magnetic suspension fan rotor bending mode test method, which comprises the following steps: obtaining geometric data of a rotor, a main impeller and a tail end heat dissipation impeller, generating a simplified three-dimensional grid structure, and obtaining an initial simulation model; key node coordinates are extracted from the initial simulation model, specified material parameters comprise density and elastic modulus, and a consistent material attribute distribution model is determined; performing modal simulation on the consistent material attribute distribution model, and calculating inherent frequencies and vibration mode vectors of first-order and second-order bending modals to obtain a simulation modal feature data set; according to the frequency distribution in the simulation modal characteristic data set, determining the positions of excitation points and response points tested by a hammering method, and obtaining a point location configuration scheme covering the key part of the rotating shaft; a sensor signal acquisition process is simulated through a point location configuration scheme, a frequency range is estimated, hammering excitation is applied to different rotating shaft states, and an original vibration response data sequence is obtained.

The application discloses a kind of based on single view point cloud feature extraction of weld seam identification method and system, it is related to welding identification technical field, including steps: obtaining the workpiece three-dimensional cleaning point cloud dataset under current view point;Obtain the spherical neighborhood point cloud set corresponding to each single point cloud, obtain the single point cloud geometry feature corresponding to single point cloud;Using clustering algorithm, the feature vectors in feature vector set are classified clustering;Obtain view point vector and shape vector;According to the included angle between view point vector and shape vector, identify non-weld seam convex point and weld seam concave point.The method provided by the application can identify real weld seam point cloud without initial weld seam fitting, uses the same concept for real weld seam point identification for straight-line weld seam and curve weld seam, and has strong versatility;Meanwhile, the technical problem that the present art lacks distinction of real weld seam point cloud after identifying potential weld seam area, resulting in low weld seam identification precision, is solved.