199results about How to "Reduce computing time" patented technology

The invention provides a method and a device for avoiding an obstacle of an unmanned aerial vehicle. The method comprises the following steps of: detecting and judging whether the obstacle or an emergent threat exists in the forwarding direction of the unmanned aerial vehicle in real time; if the obstacle or the emergent threat exists, establishing an obstacle region at which the obstacle or the emergent threat is located, and establishing a collision avoidance behavior region according to an obstacle side closest to the unmanned aerial vehicle in the obstacle region; if the unmanned aerial vehicle enters the collision avoidance behavior region, calculating an obstacle avoidance velocity vector according to the closest obstacle side, wherein the obstacle avoidance velocity vector comprises obstacle avoidance velocity and obstacle avoidance direction of the unmanned aerial vehicle; calculating forwarding step of the unmanned aerial vehicle according to the obstacle avoidance velocity vector obtained through calculation; if not any obstacle or emergent threat is detected, calculating a target guide velocity vector according to target position, and calculating forwarding step of the target guide velocity vector; judging whether to achieve the target, if the target is not achieved, detecting and calculating again; or ending.

A digital signature is applied to digital data in real-time. The digital signature serves as a mark of authenticity assuring a recipient that the digital data did in fact originate from an indicated source. The digital signature may be applied to any digital data, including video signals, audio signals, electronic commerce information, data pertaining to land vehicles, marine vessels, aircraft, or any other data that can be transmitted and received in digital form.

The invention discloses a depth learning target detection method based on pre-segmentation and regression, mainly to solve the problems that the existing target detection method is poor in small target detection precision and long in detection time. The method comprises steps: 1) a quadtree segmentation algorithm is used to extract a region of interest in a to-be-detected image; 2) a basic convolution layer and an auxiliary convolution layer are used to carry out feature extraction on the region of interest, and feature graphs of multiple scales are obtained; 3) the position information of a default border is calculated on the feature graphs of multiple scales, a convolution filter is used for detection on the feature graphs of multiple scales, and multiple predicted borders and multiple category scores are obtained; and 4) non-maximum suppression is used for the multiple predicted borders and the multiple category scores, and the final target border position and the category information are obtained. A small target in the image can be quickly and accurately detected, and the method can be used for target real-time detection in unmanned aerial vehicle aerial photographing.

The invention discloses a face identification method and apparatus based on a sequencing neural network model. The method comprises the following steps: performing preprocessing operation on input face images, and correcting angles and expressions of the face images; extracting features of already corrected face images/videos by use of a neural network comprising sequencing operation; and according to feature expression of the face images, calculating similarity between image pairs, and accordingly, obtaining identifies of specific objects in the input face images. According to the invention, a sequencing neural network structure is brought forward for solving the problems of too many parameters and too large calculation cost of a conventional neural network based face identification model in face identification problems, through sequencing expression between different features, network parameters are effectively reduced, and the calculation time is saved; and a training method based on comparison loss and triple loss is brought forward for solving the problem of a quite small number of training data.

The invention belongs to the field of artificial intelligence technology and particularly discloses an intelligent question-answering system for tax consultation service. The system comprises a terminal device where an Android operating system is installed and a computer; an application software program is installed in the terminal, wherein the application software comprises a voice conversion module and a question return module; a service software system is installed on the computer, wherein the service software system comprises a question understanding module and a question retrieval module; when the system operates, the voice conversion module converts voice data output by a user into text data, semantic understanding is performed through the question understanding module, the question retrieval module is used for retrieving answers, and a processing result is transmitted to the terminal user through the question return module. According to the system, voice recognition, text classification, similarity calculation and other technologies are used comprehensively to form a method for performing text similarity matching on an incomplete data set in the area of expertise, profound semantic analysis can be performed on questions raised by taxpayers, and meanwhile ceaseless accurate consultation service is provided for mass users to meet the actual requirement for tax consultation.

A method, computer readable storage, and apparatus for implementing a recursive hierarchical segmentation algorithm on a parallel computing platform. The method includes setting a bottom level of recursion that defines where a recursive division of an image into sections stops dividing, and setting an intermediate level of recursion where the recursive division changes from a parallel implementation into a serial implementation. The segmentation algorithm is implemented according to the set levels. The method can also include setting a convergence check level of recursion with which the first level of recursion communicates with when performing a convergence check.

Full-AC load flow constitutes a core computation in power system analysis. The present invention provides a performance gain with a hardware implementation of a sparse-linear solver using a Field Programmable Gate Array (FPGA). The invention also relates to the design, simulation, and hardware verification of a static transmission line model for analog power flow computation. Operational transconductance amplifiers are employed in the model based on a previously proposed DC emulation technique of power flow computation, and provide reconfigurability of transmission line parameters via transconductance gain. The invention also uses Analog Behavioral Models (ABMs) in an efficient strategy for designing analog emulation engines for large-scale power system computation. Results of PSpice simulations of these emulation circuits are compared with industrial grade numerical simulations for validation. The application is also concerned with the development of a generator model using analog circuits for load flow emulation for power system analysis to reduce computation time. The generator model includes reconfigurable parameters using operational transconductance amplifiers (OTAs). The circuit module is used with other reconfigurable circuits, i.e., transmission lines and loads.

The invention relates to an SM2 (streaming multiprocessor 2) digital signature generating algorithm realizing method and device. The method includes calculating an SM2 digital signature generating algorithm by coordination of a CPU (central processing unit) and a GPU (graphics processing unit), pre-calculating elliptical curve dots in the SM2 digital signature generating algorithm in the GPU to obtain a result and storing the result in a storage unit, pre-calculating one-added inverse calculation of a user private key in the CPU to obtain a result and storing the result in the storage unit, calling the results obtained by pre-calculation and stored in the storage unit when the CPU calculates a to-be-signed message sequence according to the SM2 digital signature generating algorithm calculating step. Accordingly, digital signature calculation time for to-be-signed messages is shortened, and calculation efficiency of the SM2 digital signature generating algorithm is improved. Further, the CPU can perform one-added inverse calculation on different private keys and store results in the storage unit, and calculation efficiency is further improved.

The invention provides a face key point tracking system and method applied to a mobile device. The system is composed of an initial face frame detection module, a face key point regression model learning module, a face key point regression calculation module, a working mode determination module and a face key point post-processing module. The face key point regression calculation module includes a detection mode and a tracking mode; an initial face frame is used as an input to carry out regression calculation of a face key point model in the detection mode; and regression calculation of a current frame key point model is carried out by using a last-frame face key point model in the tracking mode. According to the system and method provided by the invention, on the basis of the face key point positioning technology having the detection mode and the tracking mode, initial face frame detection is only carried out at a first frame, the tracking mode is employed during a follow-up positioning process, and the key point positioning of the last frame is used as the input of the current frame. Therefore, the face detection module does not need to be invoked again during the tracking process, so that the calculation time is saved and the system efficiency is improved.

The invention provides a reconfiguration method for a compression sensing signal, which comprises the following steps: A) calculating an inner product (g1=phi Ty) of a signal y subjected to compression sensing treatment and each row of a sensing matrix phi, and taking the inner product as an inner product result of first iteration; B) judging if stopping iteration according to the inner product result gt of the iteration; if not, searching an index value lambda t corresponding to an element with the maximum absolute value from the inner product result gt of the iteration, adding the index value lambda t into an index set, calculating the inner product of the next iteration and starting the next iteration process, and returning to the step B, wherein t is the iteration index value and is the No. lambda t row of the matrix X=phi T phi; and if stopping, forming a matrix phi lambda by rows of the sensing matrix corresponding to different index values in the present index set, and reconfiguring the compression sensing signal according to the formed matrix phi lambda and the signal y, wherein the rows in the matrix phi lambda are arrayed according to the sequence of the index values from low to high, and lambda is the set composed of different index values in the present index set. According to the reconfiguration method provided by the invention, the calculating time and resource are saved.

The invention is suitable for the technical field of electric power, and provides a uniform PWM modulation method on the basis of natural commutation. The method comprises the following steps: a per-unit value of a three-phase modulated wave is calculated; a real-time maximum value and a real-time minimum value of voltage of the three-phase modulated wave are calculated according to the per-unit value; the phase of the three-phase modulated wave shifts for 30 degrees; the real-time maximum value and the real-time minimum value of the three-phase modulated wave after phase shifting are calculated; zero-sequence components under three modulation modes of SPWM, SVPWM and DPWMx are calculated according to the real-time maximum value and the real-time minimum value of the voltage of the three-phase modulated wave before phase shifting and the maximum value and the minimum value of the voltage of the three-phase modulated wave after phase shifting; the zero-sequence components under the three modulation modes of SPWM, SVPWM and DPWMx are super-positioned in the three-phase modulated wave before phase shifting so that a uniform modulated wave signal is acquired; and the uniform modulated wave signal and a carrier wave signal are compared by a controller so that a driving signal for controlling a switch tube is acquired.

The embodiment of the invention discloses a key point detection method, device and equipment and a readable medium. The method comprises the steps of obtaining a video frame sequence on which a user image is displayed; Inputting the template frame of the video frame sequence into a key point detection model to obtain position information of a plurality of key points; And inputting the target object comprising the plurality of key points in the template frame and the detection frame into the tracking model, detecting the target object in the detection frame, and obtaining the position information of the plurality of key points in the target object. According to the embodiment of the invention, the efficiency and accuracy of key point detection can be improved.

It is an object of the present invention to accomplish both improvement of a calculation precision and reduction of a calculation time in prediction of a welding distortion of a large welded structure. A method of modeling a welded structure subjected to analysis by generating a mesh and performing thermo-elastic-plastic analysis thereon constructs a global model of the welded structure, and extracts a local model including a welded part from the global model. Next, the method constrains a boundary part of the extracted local model with the remaining portion of the global model, performs thermo-elastic-plastic analysis, and pastes the local model including the analysis result of the thermo-elastic-plastic analysis on the remaining portion of the global model, thereby reconstructing the global model. Thereafter, the method releases the constraint on the boundary part, and performs elastic analysis on the global model, thereby calculating a distortion of the welded structure.

The invention relates to a method for determining the illumination intensity distribution of a light source and a mask pattern of a photo-etching process and belongs to the technical field of manufacturing of an integrated circuit. The method comprises the following steps of: determining the initial shape and initial illumination intensity of the light source, and representing the initial shape and the initial illumination intensity by matrixes; determining an initial pattern of a mask, and representing the initial pattern of the mask by a pixel matrix; adding two compensation items and establishing a target function by using imaging accuracy and focal depth in photo-etching process as a basic optimization target; calculating and normalizing the illumination intensity distribution after the mask pattern is subjected to action of a photo-etching system; repeating the calculation step until a condition of convergence is met; obtaining a light source matrix and calculating an imaging nucleus of the photo-etching system and coefficients of the imaging nucleus; calculating an illumination intensity distribution matrix; calculating a pattern which is etched on a silicon wafer and representing the pattern by the matrix; calculating the gradient of the target function on the matrix; updating the matrix; and repeating the calculation step until the condition of convergence is met. The method for optimizing the light source and the mask of the photo-etching process has the advantages of short time and good optimization effect.

The invention relates to the field of coding of images of field camera shooting equipment, and provides a video coding method allowing a human face to be recognized. The video coding method comprises the steps that signals collected by an image sensor are converted into YUV data; the YUV data are transmitted to a first cache and a second cache respectively, the YUV data fed into the first cache are first data with an original resolution ratio, and the YUV data fed into the second cache are second data with a compressed resolution ratio; a human face detecting algorithm is executed on the second data, and regional information of a macro block meeting human face characteristics is obtained; the first data are coded, and the coding rate of a macro block of a corresponding region in the first data is improved according to the regional information. The invention further provides a video coding device allowing the human face to be recognized. The video coding method and device allowing the human face to be recognized have the advantages that due to the that that the resolution ratio of the second data is low, a higher operating rate is achieved when the human detecting algorithm is executed, and operation consumed time is shortened significantly.

The invention discloses a passenger destination prediction method and a system. The passenger destination prediction method comprises steps of determining the travel type, performing destination area integration, performing classification on the travel record history of the passenger, calculating the weight of each classification, performing statistic and conclusion on the travel record history of the passenger of the appointed destination to obtain the appointed travel probability of each destination, using a frequency pattern tree FPTree algorithm to obtain the random travel probability between the destinations, and obtaining the passenger travel probability of the destination according to the appointed travel probability, the random travel probability and the weight of each classification. The invention also discloses a prediction system of the passenger, which solves the technical difficulty of the passenger destination prediction on the aspects of time calculation and space calculation, and guarantees the accuracy and practicability of the prediction result.

The invention discloses a dynamic grid optimization-based LOF (Local Outlier Factor) clustering data anomaly point detection method and detection system. The detection method comprises the following steps: 1, obtaining an optimal unit grid vector Mopt according to an initial unit grid vector M0 and a growth vector delta p; 2, dividing data space into a dense region Rd, a sparse region Rs and a transition region Rt according to the optimal unit grid vector Mopt; 3. dividing all grids in the transition region Rt, which is obtained in the step 2, into dense region grids G'd and sparse region grids G's according to grid periphery density ratio vectors F, and adding the sparse region grids G's into the sparse region Rs to form a sparse region Regions of the data space; and 4, applying an LOF algorithm to detect anomaly points of the sparse region Regions of the data space. According to the method, the operation data amount of the LOF algorithm is reduced through the dynamic grids, and the computation time of the LOF algorithm is greatly reduced.

The invention is applied to the technical field of information processing, and provides a method and system for obtaining a recommended video. The method comprises the following steps of: obtaining a video set watched by a user group according to a pre-set screening condition; obtaining video frequent item sets from the video set according to a pre-set video support degree threshold value, and sorting the video frequent item sets according to the video support degree; and, on the basis of a Spark parallel algorithm of Fp-growth, obtaining a recommended video item set from the sorted video frequent item sets. By means of the method and the system disclosed by the invention, the computing complexity of the recommended video can be reduced; and the computing time is reduced.

The invention discloses an EBM variable direction forming dynamic slicing method cooperated with an 840D numerical control system. The EBM variable direction forming dynamic slicing method comprises the steps of establishing a vertex array object VAO, adjusting a projection matrix to the slicing height during rendering of a three-dimensional model and outputting a two-dimensional section binary image through a reverse ray tracing algorithm; acquiring data of a contour line from the binary image through a two-dimensional contour extraction method, outputting coordinates of an end-to-end closedcontour, connecting all adjacent coordinates into straight lines and connecting all the straight lines end to end to form a closed curve; and dividing the contour data into a straight line part and acurve part with the starting point of the small straight lines with the included angle of 140 degrees as the segmentation point when the triangle tolerance is one, removing neighbor points, repeated points and internal points of the same straight lines, conducting segmentation fitting on a slicing data curve through a NURBS curve and then conducting dynamic slicing through the 840D numerical control system. Through the EBM variable direction forming dynamic slicing method, the computing time and conversion errors of STL files are reduced, and meanwhile, the phenomenon of over-accumulation of straight line inflection points is reduced.

The invention discloses a no-reference structural sharpness-based automatic focusing method. The method includes the following steps that: step S1, grayscale processing is performed on an original image; step S2, image preprocessing is performed with a downsampling method; step S3, low-pass smoothing filtering is performed on a downsampled image, so that a reference image is obtained; step S4, overlapped blocking is performed on the downsampled image and the reference image according to a set block size and a moving step size, so that image sub-blocks are obtained; intra-block variances are calculated, a threshold parameter is set, and image sub-blocks of which the intra-block variances are greater than the threshold are selected; step S5, SSIM values are calculated for the selected imagesub-blocks corresponding to the down-sampled image and the reference image sequentially, and the image sharpness evaluation value, namely NRSS value, of the original image is calculated; and step S6,the NRSS value is converted into a voltage signal, and a lens receives the voltage signal and thereafter completes real-time automatic focusing. The method of the invention has strong robustness, saves calculation time, and well satisfies the real-time requirement of micro-circulation monitoring for automatic focusing.

The present invention relates to a method for determining patient-specific blood vessel information. More specifically, the present invention relates to a method for determining patient-specific cardiovascular information by applying a simplified coronary circulation model thereto. Furthermore, the present invention relates to a method for determining a blood flow rate for branches of a blood vessel having originated from an artery of each patient. According to the present invention, the method for determining cardiovascular information by using a computer system comprises the steps of: receiving image data including a plurality of coronary arteries having originated from the aorta; processing the image data so as to generate a three-dimensional shape model of the plurality of coronary arteries; simulating a blood flow for the generated three-dimensional shape model of the plurality of coronary arteries; and determining a fractional flow reserve (FFR) of the respective coronary arteries with the blood flow simulation result. In the blood flow simulation step for the three-dimensional shape model of the plurality of coronary arteries, a computational fluid dynamics model is applied to the three-dimensional shape model of the coronary arteries, and a centralized parameter model to be combined with the computational fluid dynamics model uses a simplified coronary circulation model including coronary arteries, capillaries of the coronary arteries, and coronary veins.

The invention relates to the technical field of laser point cloud classification, and more specifically, relates to a point cloud top noise elimination method based on LiDAR. A standard space grid isestablished for point cloud based on spatial clustering and spatial autocorrelation. The position of noise is determined according to the spatial distribution characteristic and interdependence of point cloud data. The grid units containing noise are determined according to the growth pattern of grid clustering. European clustering is carried out by using initially identified top noise as a seed point, in order to eliminate noise. The algorithm abandons one-by-one computation, and reduces the computation time. The problem that a good effect cannot be achieved in actual production due to poor general applicability, low efficiency or other reasons is solved. Cloud sheet-shaped, high-density and wide-range noise produced in a special climate environment can be eliminated with high accuracy.

The invention discloses a log-based user behavior data processing method, a medium, a device and a device. The method comprises: 1, collecting log information; 2, determining a plurality of access characteristics, and extracting access behavior values of different users for different access characteristics accord to log information; 3, dividing that access characteristic into N groups, and determine the cluster number of each group; N is an integer greater than or equal to 1; step 4, clustering each grouping according to the number of clusters corresponding to the grouping to obtain clusteringresults. The invention solves the problem of poor high-dimensional clustering effect by constructing high-dimensional data as low-dimensional features composed of a plurality of similar or related features and clustering. And by constructing the high-dimensional data into a plurality of similar or related features of the low-dimensional features after clustering, the number of clusters per sub-clustering process is reduced, saving computational time and saving computational memory.

The invention provides a method for detecting an SAG signal in a DVR (Discharge Voltage Regulator) device. The method comprises the following steps: a, obtaining the number of sampling points according to preset sampling frequency and power frequency cycle, and setting a cyclic buffer region with the length equal to the number of sampling points; b, obtaining sampling values corresponding to the sampling points in the power frequency cycle and storing the sampling values in the cyclic buffer region according to a sampling time sequence to form an analytic window; c, when the sampling time of a next sampling point is reached, storing the sampling value of the point in the window, discarding the sampling value with the earliest sampling time in the window, and updating; d, performing wavelet analysis on each sampling value in the updated window to obtain a modulus maximum value of the point; e, judging whether the modulus maximum value is more than 0 or not; f, if so, determining that the SAG signal exists and the sampling time of the point is the starting moment of the SAG signal, and otherwise, returning to the step c. By implementing the method, the starting moment and amplitude of the SAG signal can be detected in real time and the purpose of reducing SAG fault detection time delay in the DVR device can be achieved.

The invention discloses a lane line detection method based on deep learning. The method comprises the following steps: S01, obtaining a lane detection data set, wherein the lane detection data set comprises a plurality of images containing lanes; S02, building a training model, and setting a loss function and constraint parameters, wherein the training model comprises a convolution layer, a pooling layer, a residual block and a full connection layer which are connected in sequence, the loss function is used for limiting the smoothness and rigidity of the lane line; S03, training the training model by adopting the lane detection data set, and iterating for multiple times to obtain a convergent detection model; and S04, placing the detection model in a vehicle-mounted camera to obtain a lanedetection result. According to the lane line detection method based on deep learning provided by the invention, the lane line can be rapidly detected while the accuracy is ensured.

The invention discloses a block data checking method and device. The method includes the steps that a data node used for storing new data is selected from a block containing multiple data nodes; for the data node, the proof test value of the new data is determined, and according to the proof test value of the new data, the updating proof test value of the data node is calculated; according to the updating proof test value of the data node and original proof test values of other data nodes contained in the block, the proof test value of the block is obtained, wherein the proof test value of the block is used for verifying whether data stored in the block is consistent with data stored in other blocks in a block chain. By means of the new data, the updating proof test value of the data node used for storing new data is determined, the problem that in the prior art, all the data in the block are used for calculating the proof test value when the new data occurs in the block and thus much time is consumed is solved, the calculation time of the proof test value is effectively reduced, and the verifying efficiency of the data in the block is thus improved.

The invention provides a simulation method for the flow field and convection/radiation coupling heat transfer of a multi-nozzle rocket. Firstly, a three-dimensional model of the multi-nozzle rocket isestablished. Secondly, the multi-block structured mesh method is used to mesh the three-dimensional model. The mathematical model of plume field and convection heat transfer of multi-nozzle rocket isestablished again. S multicomponent equation is discretized and integrated by AUSM + scheme. The inviscid flux at the boundary of the model is obtained by differentiation. Then the improved radiationmodel DOM is opened to obtain the radiation absorption coefficient of the model: the integral of radiation heat transfer by DOM model. Differential fundamental equation discretization; The radiationintensity values of each point are obtained; Finally, use RNG k-Epsilon turbulence model, output Mach number, temperature, pressure flow field and convection/coupled heat flux cloud map; The method ofthe invention can not only improve the calculation accuracy but also reduce the calculation cost, and the calculation result can provide guiding significance for the thermal protection of the bottomof the rocket.

The invention provides a space manipulator path planning method for reducing base attitude disturbance. The method includes the following steps that firstly, a kinematical equation is built for a space manipulator; then, parameterization is conducted on a joint trajectory through a sine polynomial function, and a fitness function is defined according to the motion constraint of the manipulator and attitude control accuracy of a base; finally, the optimal trajectory of joint movement of the space manipulator is searched for through a simulated annealing-particle swarm algorithm so that the minimum disturbance can be generated by the base attitude to the movement of the manipulator. The space manipulator path planning method mainly aims to achieve the purpose that path planning is conducted on the joint trajectory of the seven-freedom-degree space manipulator through an intelligent optimization algorithm.

The invention provides a prediction method for an insulator equivalent salt deposit density ESDD and non-soluble deposit density (NSDD) by a least squares support vector machine and genetic algorithm. The method includes: step 1. firstly designing an insulator leakage current on-line monitoring system for statistical analysis of leakage current time domain feature change; step 2. determining the input and output phasors of a least squares support vector machine prediction model and performing normalization processing on sample data; step 3. determining the radial basis kernel function of the least squares support vector machine and the model performance evaluation index; and step 4. optimizing the least squares support vector machine prediction model by genetic algorithm.

A computer readable storage medium stores instructions to receive a source data set and a target data set. Instructions to identify differences between the target data set and the source data set are also stored. These instructions include dividing the target data set into a set of target data blocks. Among the target data blocks at least one duplicate block in which an unbroken copy is fully duplicated within the source data set is identified. At least one modified block among the target data blocks in which an unbroken copy is not fully duplicated within the source data set is also identified. Differences between the modified block and the source data set are then determined.