72results about How to "Avoid heavy calculations" patented technology

The invention discloses a pedestrian detection method. Multiple times of convolution and pooling are performed on an input image through the pedestrian detection method based on a convolutional neuralnetwork; the features of the original image are extracted so as to obtain the corresponding feature graph of the original image; the corresponding feature graph after zooming of the original image isapproximately calculated through the image feature pyramid rules; a candidate window is generated through a region proposal network RPN; a candidate proposal window is further selected and summarizedaccording to the pedestrian size distribution in the candidate window; the corresponding weight of different scales of pedestrian targets on different scales of images is trained by using the training data having the tag; and the classifier network is trained. The summarized candidate window is solved, and the confidence obtained through the classifier and the set threshold are compared and finalpedestrian detection judgment is performed. Heavy calculation amount of obtaining the feature graph through image zooming calculation can be avoided by application of the image feature pyramid, and detection is performed on different feature graphs by using the weighing mode of different weights so that misjudgment and leak detection caused by single feature graph detection can be effectively avoided.

The invention discloses an air mouse and a method and a device for controlling the movement of a mouse pointer. The air mouse comprises at least one kind of inertia device, wherein the inertia device comprises a gravity acceleration sensor; and the method for controlling the movement of the mouse pointer comprises the following steps of: setting a report rate of the air mouse based on a vector sum of acceleration output values of various sensing axes of the gravity acceleration sensor; determining a space coordinate of the air mouse; correspondingly converting the determined space coordinate or variation thereof into a coordinate of the mouse pointer or variation thereof; and outputting the coordinate of the mouse pointer or the variation thereof at the set report rate so as to control the movement of the mouse pointer. By the technical scheme, the report rate can be dynamically adjusted according to actual movement conditions of the air mouse, so that the movement of the mouse pointer is controlled.

The invention relates to the technical field of array signal processing and discloses a near-field sound source positioning method for estimating a direction of arrival (DOA) of a near-field sound source signal and signal source distance by adopting a speed sensor array. The method has the advantages that a high-order accumulated value is not required to be calculated, so that relatively low operation amount is achieved; automatic paring in parameter estimation can be realized; a distance between two array elements does not need to be limited in the 1 / 4 wavelength range; the aperture of the array can be expanded by increasing the distance between two array elements; and therefore, the parameter estimation precision of an algorithm can be improved.

The embodiment of the invention discloses a video thumbnail generation method and device. The method comprises the following steps: carrying out scene boundary detection on a video, and sectioning the video according to different scenes; from all sections of the video, selecting a section with the longest duration as a key section; from all video frames contained in the key section, selecting a video frame with the highest balance as a key video frame; and compressing an image of the key video frame so as to generate a thumbnail to be used as the thumbnail of the video. The video thumbnail generation method and device disclosed by the embodiment of the invention can solve the problems that in the prior art, by analyzing each video frame of the whole video to generate the thumbnail, the calculation amount is large, and the efficiency is low; the invention also solves the problem that in the prior art, a video frame having no reference value for the video content is used as the thumbnail of the video; and a video frame which represents the video content and has a reference value for users can be captured as the thumbnail of the video.

The invention discloses a fuzzy intelligent multiple extreme response surface method (FIMERSM) for calculating blade life. The method comprises the following specific steps of building a finite element model of a structure; selecting and inputting random variables; performing heat-structure coupling analysis on the blade structure, and extracting a certain quantity of sample points by adopting a Latin hypercube sampling technology and calculating corresponding extreme output responses; calculating a membership degree of a sample by adopting a KFCM clustering algorithm, performing normalization processing on data, and finally obtaining a training sample of a fuzzy support vector regression (FSVR) machine; performing parameter optimization on an FSVR model by adopting a genetic algorithm, and building an FSVR mathematical model; based on a basic thought of a multiple extreme response surface method, building an FIMERSM mathematical model by applying the FSVR model; performing coupling failure reliability analysis on a blade by adopting the FIMERSM; and performing validity verification on the FIMERSM. According to the method, a new way is opened up for reliability analysis of a multi-failure mode structure.

The invention provides a high slope deformation monitoring method and system. The method comprises the steps of eliminating a carrier observation value gross error, eliminating a residual error by using an unbiased estimation method, calculating single-difference geometrical distance, performing carrier single-difference calculation on each monitoring station relative to a base station, establishing a carrier single-difference expression of all synchronous observation satellites, obtaining a difference between carrier single-difference matrixes of two adjacent epochs, establishing a matrix equation of a base line vector variation and a single-difference receiver clock difference variation between the epochs, and calculating a high slope three-dimensional deformation quantity by using a least square method. The monitoring method greatly improves the monitoring efficiency and the monitoring precision; a GPS receiver is arranged at each of the base station and the monitoring stations, and monitoring data of each monitoring station is transmitted to the base station by a switch via an LXI discovery mechanism based on a TCP / IP communication protocol, so that the accuracy of the monitoring data is not influenced by the environment, misinformation, erroneous transmission and the like are avoided, the environmental adaptability is improved, and the monitoring precision is also improved.

The invention relates to the technical field of image processing, in particular to a remote sensing image registration algorithm. The registration algorithm comprises the following steps of: (1) reading primary remote sensing image data from a storage medium; (2) pre-processing each primary image; (3) performing binarization processing to realize image segmentation; (4) corroding and expanding the binary image; (5) extracting a monotropic edge from the eroded and expanded binary image by the Roberts cross gradient operator; (6) calculating the matching degree of the edge image and finding the optimal matching point; (7) conversing images; and (8) outputting registration information and registration images. The remote sensing image registration algorithm can be applied to multispectral data fusion and remote sensing image mosaicking.

The invention discloses a master-slave minimally invasive surgical robot system and a control method, and relates to the field of robot control. The method includes allowing a user to set a catheter moving position expected value of a master manipulator of the master-slave minimally invasive surgical robot system; allowing a gesture sensor to measure catheter actual movement joint amount, and utilizing a variable universe fuzzy PID control method to adjust the position of the catheter; allowing an imaging device to receive detected image information and imaging; determining whether an expected position is achieved according to images displayed by the imaging device. According to the method, position tracking performance and robustness are enhanced on the basis of conventional PID control, dilation factors are selected based on error grading, the contradiction between fuzzy PID control accuracy and control rule number is solved, once dilation factor calculation during each sampling period is avoided, time is saved, requirements of hardware are reduced, safety of a minimally invasive surgery is guaranteed, and effectiveness and practicality of the master-slave minimally invasive surgical robot system during a minimally invasive surgery are improved.

The invention provides a rehabilitation detection method based on channel state information and BiLSTM-Attention, and the rehabilitation detection method comprises the implementation steps: collectingCSI data of a rehabilitation action in an indoor environment, and extracting amplitude information; performing preprocessing steps of low-pass filtering, normalization, principal component analysis and the like on the CSI data; segmenting the preprocessed signals, detecting starting and ending points of actions, and dividing segmented data segments into a training set and a test set; and inputting the training set into the deep neural network based on BiLSTM-Attention to train the action recognition model to obtain the rehabilitation action recognition model. The collected CSI test set data can be classified by adopting the model, and the purposes of rehabilitation action recognition and rehabilitation degree scoring are achieved. According to the rehabilitation detection method, the deepneural network based on BiLSTM-Attention is adopted for automatic learning and feature selection, and recognition of ten rehabilitation degrees of three different actions is achieved.

A method for on-line parameter identification of doubly-fed machine features that base on multi-innovation identification theory and random gradient identification method and adding variable forgetting factor, that parameter identification method of doubly-fed machine based on variable forgetting factor and multi-innovation random gradient method is use to identify parameters of doubly-fed machine. Considering the nonlinear strong coupling of doubly-fed machine, the stator flux-oriented vector control technology is used to build the vector control system of doubly-fed machine to collect data,and the standard form of motor parameter identification model in dq coordinate system is deduced, and the motor inductance and resistance parameters are calculated iteratively according to the method.The method is applicable to doubly-fed machine in both electric and electric states, and can identify the stator parameters of the machine on-line, improve the accuracy of the control model and improve the control performance.

The invention discloses a CAN signal analyzing method and device. The method comprises the steps that a CAN signal of a finished car is obtained; the CAN signal comprises addresses of all data frames; a display screen is controlled to display the addresses of all the data frames; the address of one data frame input by a user is obtained, and data corresponding to the address of the input data frame are extracted; the data are converted into the Boolean quantity, and the display screen is controlled to display the Boolean quantity in the form of an indicator lamp; continuous bits of data of the data frame input by the user are obtained, and the continuous bits of data are converted into decimalism physical quantity data; the physical quantity data are corrected through the factor value and the compensation value, and the corrected physical quantity data are displayed in a coordinate axis graph of the display screen in a curve form. By means of the technical scheme, the time for a CAN analysis engineer to analyze the CAN signal of the finished car can be shortened, and the analyzing efficiency is improved.

The invention discloses a cultural relic line graph drawing method based on vision-dependent curvature estimation, comprising the following steps: estimating the discrete curvature of the vertexes of a triangular mesh model; dividing the model into a flat region and a feature region according to the curvature distribution of the model; sharpening and filtering the vertexes belonging to the feature region, and calculating out new triangular mesh model vertex coordinates; and extracting the feature contour line of the triangular mesh model, and generating lines, thus completing the drawing of a plane line graph of a 3D model of a cultural relic. According to the invention, a 3D model of a cultural relic is locally sharpened through a sharpening algorithm corresponding to a smoothing algorithm based on the mean Epsilon in the field of archaeology so that the texture details of the cultural relic can be drawn. The visual contour line of the surface of the cultural relic model is extracted based on vision-dependent curvature estimation, and then, a cultural relic line graph consistent with the visual perception of archaeologists is drawn automatically. The problem that manual archaeological drawing is slow, low in precision and high in work intensity is solved. The cultural relic line graph drawing method is a simple, efficient and practical computer-aided archaeological line graph drawing method.

The invention discloses a VLSI (Very Large Scale Integration) standard unit placement method based on an electric field energy modeling technology. The method establishes the electric field energy model of a problem, and a global density function and an analytical solution of a Poisson equation are utilized to solve a VLSI standard unit global placement problem. The method has the technical key point that: (1) analogy is carried out on the placement problem and a static electricity system, a unit is taken as an electric charge, original density constraint is converted into zero potential energy constraint, a differential equation is constructed, an explicit expression is solved for the differential equation to more accurately describe the potential energy constraint, then, a penalty function method is adopted to convert the wire length target and the potential energy constraint of the VLSI global placement into an unrestraint nonlinear placement problem, and a proper optimization technology is selected for optimization; and (2) different from a situation that a method for evenly dividing bin is used for obtaining a discrete density function value, the method disclosed by the invention is characterized in that the global density expression of the unit and the overlapping constraint of a whole placement area is calculated so as to more accurately describe the distribution situation of the unit on the placement area.

The invention relates to a Hough transformation linear detection method based on a dynamic threshold range. The method comprises the following steps: performing storage in accordance with a cluster sequence based on adjacent relation, i.e., performing refinement processing on segments with certain line widths in images, outputting a frame line which has only two adjacent pixel points at most in eight neighborhoods of any one pixel point, and then based on the cluster sequence based on the adjacent relation, storing the frame line in a certain segment of arrays; and sampling m adjacent pixels in the certain segment of arrays to construct the threshold range by sampling, determining whether m points are at the same line, and filtering noise at the same time, such that the algorithmic robustness is enhanced. On the basis of constructing the threshold range, linear detection is realized by use of a dynamic sampling mode based on range halving. Each pixel in the pixel arrays is identified, such that the algorithmic detection efficiency and effect are improved.

The invention provides a permanent magnet starter-generator position detection system, and a method. The permanent magnet starter-generator position detection system comprises a current detection circuit, an auxiliary winding, a frequency locking circuit, a phase locking circuit and a control circuit. By adding an auxiliary position sensor and constructing a simulation circuit, the deviation between a phase lock signal of the auxiliary winding and a motor position reference signal under different working modes, rotational speeds and current amplitudes of a starter-generator multiplexed power converter is calibrated. During normal operation, the auxiliary position sensor is removed, and the control circuit compensates the phase locking signal by calibrating the obtained position deviation according to the working mode, the rotational speed and the magnitude of the current of the starter-generator multiplexed power converter, so as to obtain the position information of the permanent magnet starter-generator.

The invention provides an azimuth imaging method based on a MIMO rectangular plane array. The method includes a step A of performing backward projection operation on each of four groups of target scattered signals obtained by a receiving antenna when four transmitting antennas are in a time-share transmission condition, obtaining four preliminary radar images and performing modification; a step Bof performing Fourier transform on the modified radar images to a spectral domain, performing truncation and mergence on spectral domain data of the four radar images in the spectral domain and performing Fourier inversion on a comprehensive spectral domain result obtained through mergence and obtaining a final imaging result. The invention effectively solves a problem of the relationship among MIMO array length, array element number and azimuth resolution under the topological structure of the specific sparse rectangular plane MIMO array, and simultaneously gives attention to the operation efficiency. Truncation and mergence are performed on image data in the spectral domain, so that frequency spectrum information of a target is improved and imaging quality is improved. Besides, the computational efficiency is improved and the algorithm real-time performance is ensured by using Fourier transform.

The invention discloses a backlight driving method. The backlight driving method includes the steps that a main control module receives an original synchronous signal and an original backlight drivingsignal of each frame of image data; the main control module conducts N times of frequency doubling on the original synchronous signal to generate a first synchronous signal, a backlight region is divided into N different sub-regions, original backlight driving signals of the N backlight sub-regions are borne in N frames of backlight data, and each frame of backlight data includes effective driving data of one backlight sub-region and ineffective driving data of N-1 backlight sub-regions; the main control module sends the first synchronous signal and the N frames of backlight data N times to abacklight driving module within one original synchronous signal period. In the technical scheme, a backlight driving scheme which is simple in design and easy to use generally among liquid crystal panels with different dimensions is provided, so that synchronization of liquid crystal displaying and backlight driving is realized.

An embodiment of the invention provides a data transmission method, a device, and a base station, and relates to the technology field of communication. The method comprises determining the region where a user device permitting data transmission is in a service cell; when the user device is at the edge of the service cell, performing single stream data transmission with the user device. Via determining the region where the user device permitting data transmission is in the service cell, and performing single stream data transmission with the user device when the user device is at the edge of the service cell, obtaining of SINR of each data stream is prevented, massive calculation via a single-stream or multi-stream self-adaptive switching algorithm upon each user device is prevented, calculated amount is reduced, calculated complex is reduced, and frequent switching between a single-stream transmission mode and a multi-stream transmission mode is prevented.

The invention discloses a scanning and recognizing device and a scanning and recognizing method based on a triaxial accelerometer and relates to the field of image processing. The device comprises an optical sensor, the triaxial accelerometer and a control module, wherein the optical sensor is used for continuously shooting a scanned character to acquire an image frame sequence comprising the scanned character; the triaxial accelerometer is used for measuring a relative displacement among image frames in the image frame sequence; and a control module is used for performing down-sampling on the image frame sequence according to the relative displacement measured by the triaxial accelerometer, splicing acquired sampling frames, performing character recognition on a spliced panoramic image and thus obtaining a character code of the scanned character. In the device and the method, a large amount of calculation among the image frames, particularly large-size image frames, is avoided by measuring the displacement between the adjacent image frames through the triaxial accelerator, performing accurate down-sampling on the image frame sequence according to the displacement and searching from a given window to calculate spliced information of images of the adjacent sampling frames; therefore, load of the control module is reduced greatly, the splicing speed is increased and the high-precision alignment is realized.

The invention discloses a skeleton point behavior recognition system based on a shift graph convolutional neural network. The skeleton point behavior recognition system comprises an image acquisitionmodule, an image processing module, an extraction module and a behavior recognition module, wherein the image acquisition module is used for acquiring a behavior image; the image processing module isused for processing the behavior image acquired by the image acquisition module for image processing; the extraction module is used for extracting skeleton points of the image processed by the image processing module; the behavior identification module is used for identifying the behavior characteristics of the skeleton points extracted by the extraction module. According to the invention, the behavior identification module is designed to identify skeleton point behaviors; the novel graph convolution of the graph convolution calculation amount is reduced; the system is different from traditional image convolution; according to shift graph convolution, a feeling range is not expanded by expanding a convolution kernel, but shift splicing is carried out on graph features through a novel shiftoperation, so that the same or even higher recognition precision is achieved under the condition that the calculation amount is remarkably reduced and the calculation speed is improved, and the situation that the calculation amount of traditional graph convolution is increased along with increase of the convolution kernel is avoided.

The invention discloses a propeller design method based on a genetic algorithm. The method comprises the following steps: firstly, presetting initial speed V0; determining value range of initial parameters of a propeller; constructing a genome population; carrying out cavitation checking on the genome population to eliminate / reform genomes not according with requirements; calculating propeller efficiency of genomes; carrying out evolution and elimination on the genome population until a preset evolution times are reached; outputting and selecting highest efficient propeller design parameters,and calculating corresponding speed; and finally, judging, according to the speed, whether optimization is continued. According to the method, an optimization design of the genetic algorithm is employed, limitation that in the prior art, mutual influences in changes among all parameters need to be considered and thus adjustment is continuously carried out for obtaining optimum propeller efficiencyis overcome, and design efficiency of the propeller is improved; and compared with a set of optimal parameters only obtained by using a graph design manner, a set of multi-solution optimization parameter schemes finally obtained by the method more facilitate selection and determination of an optimal propeller design scheme.

The invention discloses an IBBO based complex looped network directional protection MBPS solving method, which is characterized in that the effectiveness of a determined breakpoint is judged by adopting a protection association matrix, and a lot of calculation required by forming all simple directed circuits of the whole network is avoided; an improved biogeography optimization algorithm is proposed, an a transfer model, a transfer operator and a mutation operator of the traditional BBO algorithm are improved; and the breakpoint is solved by using an IBBO method, and efficient solving for a minimum breakpoint optimization model is achieved. The MBPS can be effectively solved for an ultrahigh-voltage power grid containing multiple looped networks according to the invention. Compared with the traditional optimization method, the method disclosed by the invention is high in convergence speed, good in iteration robustness and high in convergence precision, and has excellent optimization performance.

The invention discloses a method and device for optimizing the instantaneous dynamic state of an LLC resonant converter load. Firstly, an actual test of a specific point is carried out on the LLC resonant converter; discrete points related to the LLC switching frequency, the input voltage, the output voltage and the output current are obtained; and fitting is carried out to obtain an optimal objective function of the LLC switching frequency, and then through dynamic load switching judgment, a final output PID result of the voltage loop controller linearly approaches an expected output value ofthe optimal objective function when a load suddenly changes. An effective LLC switching frequency target function is obtained through a simple fitting mode, and the PID output result of the voltage controller linearly approaches the expected value of the LLC switching frequency during dynamic load switching. According to the method, the calculation deviation and a large quantity of calculation caused by LLC small signal model modeling are avoided, and the change speed of the LLC switching frequency is increased, so that the dynamic response speed of the LLC is increased under the condition that the hardware cost of the system is not increased.

The invention discloses a method for detecting an initial magnetic field position of a linear synchronous motor. A motor is controlled by using a three-phase inverter of a servo controller; a system is electrified; under the control of the servo controller, a V-phase upper bridge arm and a W-phase lower bridge arm of the three-phase inverter are conducted, and other bridge arms are cut off; at the moment, a current sequentially flows through the V-phase upper bridge arm, a V-phase winding of the motor, a W-phase winding of the motor and the W-phase lower bridge arm; by using common ends of three-phase windings of the motor as reference points, a positive current flows through the V-phase winding of the motor, and a negative current flows through the W-phase winding of the motor; and after the motor stops rotating, the servo controller acquires a motor position signal which is fed back by an incremental photoelectric coder, and a motor position serves as the initial magnetic field position of the motor. By adoption of the method, the initial magnetic field position of the motor can be simply, conveniently and quickly detected, a large amount of calculation of a microprocessor of the servo controller is avoided, and detection accuracy and the control speed of the servo controller are improved.

The invention discloses a data balancing method based on pseudo-negative samples and a method for improving data classification performance. The method comprises the following steps: step 1, positiveand negative samples are separated to obtain a positive sample set and a negative sample set; 2, calculate a negative sample Pearson correlation coefficient set; 3, initialize a pseudo negative sampleset and a selected sample set; Step 4: using a maximum correlation-least redundancy method to calculate weights and obtains a weight set. 5, selecting the maximum weight, updating the pseudo-negativesample set and the selected sample set; 6, repeat that steps 4 and 5 until a pseudo negative sample set is selected; 7, merging the selected pseudo-negative sample set into the positive sample set, and removing the selected pseudo-negative sample set from the negative sample set. The invention firstly proposes and defines the concept of pseudo-negative sample, and the proposed algorithm can improve the accuracy of data classification and further improve the performance of the classifier, in particular, has obvious advantages in dealing with unbalanced biological information data.

The invention provides a design method and system of a diffraction grating. The method comprises the steps of S1 obtaining conditions of diffraction efficiency enhancement and default of a zero diffraction order and a high diffraction order of diffraction light of the diffraction grating; S2 obtaining the groove depth of the diffraction grating according to a first condition of zero diffraction order default; S3 obtaining the grid ridge width of the diffraction grating and the distance between grid ridges according to the condition of high diffraction order enhancement; and S4 copying the structure obtained in the step S3, and carrying out symmetrical arrangement, so that the even diffraction order of the high diffraction order is lack of order. According to the method, the widths and thepositions of the corresponding grid level and the grid groove are selected according to the diffraction order needing to be enhanced, the whole grating structure does not need to be set as an optimization condition to be optimized, and a large amount of calculation is avoided.

The invention discloses a rapid simulation method and device for equivalent signal level TOA measurement, and the method comprises the steps of 1, carrying out the simulation through a signal level TOA algorithm module, and obtaining a mathematical relation between signal receiving power and TOA; and step 2, calculating the signal receiving power according to the signal transmitting power of the target, and calculating to obtain a TOA estimated value by utilizing a mathematical relation between the signal receiving power and the TOA. According to the invention, the mathematical relation between signal receiving power and TOA is obtained through simulation of the algorithm module; during system simulation, a TOA estimation value is directly obtained through the mathematical relation, and asignal transmitting and receiving simulation processing flow is not carried out any more, so that a large amount of calculation is avoided, the problem that the multi-point positioning system-level simulation speed is low is effectively solved, and the rapid verification of system indexes is facilitated under the condition that the overall performance is not lost.

The invention belongs to the field of image processing and target detection, and discloses a target classification and attitude detection method based on a deep convolutional neural network. The core content of the method includes: 1. a method for making a sample set; 2. a method for generating candidate windows; 3. a method for feature extraction; 4. a method for network training; After the image to be detected gets the deep convolution feature through the convolutional neural network, the candidate window with the attitude angle is mapped to the feature layer to obtain the directional regional feature vector, and the final detection result is obtained by classifying and predicting the feature vector . This method can extract more pure target features from samples, improve the accuracy of classification, and realize the function of target attitude angle detection.

The invention discloses a high-permeability active power distribution network electric energy quality management method based on FCS-MPC control. The method comprises the steps of building a high-permeability active power distribution network as a power quality management test system; building an adjustable reactor mathematical model; combining the finite state model prediction control FCS-MPC with an adjustable reactor model, constructing a step length function, and predicting the output of an adjustable reactor; and setting a self-adaptive weight selection method of the FCS-MPC so that the FCS-MPC can judge according to the severity of the power quality disturbance to output an optimal switching state. According to the method provided by the invention, the adjustable reactor is used forcomprehensively treating the composite power quality disturbance of the high-permeability active power distribution network, meanwhile, after the adjustable reactor is combined with the FCS-MPC, different disturbance types can be controlled and treated, and a new control strategy for the adjustable reactor is provided; therefore, self-adaptive weight factor determination can be carried out according to the amplitudes of the voltage and the current and the harmonic distortion rate, and the multi-objective optimization is completed.

The invention discloses a welding robot energy optimization method based on welding gun rotation angles. The method comprises the steps that 1, an arc welding robot power measurement platform is established; 2, a to-be-optimized trajectory of a working path of an arc welding robot is determined; 3, the welding gun rotation angles at a welding start point and a welding ending point are changed, andthe value range of the welding gun rotation angles is determined under the condition of meeting the requirement for reachability of the arc welding robot; 4, a power consumption measurement experiment is performed according to the value range of the welding gun rotation angles; and 5, in combination with experimental data, average power consumption is compared to obtain welding starting point andending point welding gun rotation angles with the optimal energy. According to the method, a welding starting point and ending point welding gun rotation angle combination with the minimum power consumption is obtained through a small number of tests, the power consumption of the arc welding robot in the welding operation process is reduced, and the method has great significance in energy conservation in the actual welding production process.