45 results about "Deviation matrix" patented technology

A convolutional neural network is implemented on a graphics processing unit. The network is then trained through a series of forward and backward passes, with convolutional kernels and bias matrices modified on each backward pass according to a gradient of an error function. The implementation takes advantage of parallel processing capabilities of pixel shader units on a GPU, and utilizes a set of start-to-finish formulas to program the computations on the pixel shaders. Input and output to the program is done through textures, and a multi-pass summation process is used when sums are needed across pixel shader unit registers.

The invention provides a wind farm predicted wind speed correction method and a wind farm predicted wind speed correction device. The wind farm predicted wind speed correction method comprises steps that a plurality of conventional wind farm wind speed deviation matrixes are established, and wind speed deviations represent differences between the predicted wind speeds of the target wind measuring masts of the conventional wind farms and actual measured wind speeds; based on the plurality of conventional wind farm wind speed deviation matrixes, a wind speed deviation correction model database is established, and comprises a plurality of wind speed deviation correction models corresponding to the conventional wind farms, and the wind speed deviation correction models represent relations between the wind speed deviations of the conventional wind farms and equivalent factors; related parameters of a target wind farm are determined, and according to the related parameters, the matched wind speed deviation correction model is determined in the wind speed deviation correction model database, and then the predicted wind speed of the target wind farm is corrected according to the determined wind speed deviation correction model. Errors generated during the calculation of the wind speeds of the wind farms by using conventional fluid models are effectively remedied.

The invention discloses a multi-target tracking method in the case of loss of measurement data. The multi-target tracking method includes the following steps: initializing probability hypothetical density and potential distribution of a target state set; predicting the probability hypothetical density and potential distribution of the target state set; updating and converting a deviation matrix and a covariance matrix for measurement; introducing a scaling factor, and changing a gain matrix; using an ellipsoid threshold to remove irrelevant measurement data for the weight of a target strength function; performing trimming and merging, and extracting a target state; determining if a target exists or not, returning to the step of predicting the probability hypothetical density and potential distribution of the target state set if so, otherwise ending target tracking. According to the invention, the scaling factor is introduced during updating of a CPHD filter to regulate the filter gain matrix, and the robustness of the filter can be improved; a self-adaption threshold can simplify a measurement set, the observation number of a current observation set is reduced, all the measurement data irrelevant with a detected target is removed, and finally, the computing amount is reduced.

The embodiment of the invention discloses an image stitching method and an image stitching system. The method comprises the following steps: acquiring a plurality of images corresponding to a plurality of scene points through a plurality of sets of image acquisition equipment respectively, wherein the images corresponding to each scene point comprise a reference image and an image to be stitched, and the reference image and the image to be stitched have an overlapped region; carrying out first remapping on the reference image and the image to be stitched respectively; extracting a rough matching characteristic point pair between the remapped reference image and the remapped image to be stitched; acquiring a fine matching characteristic point pair from the rough matching characteristic point pair by utilizing second remapping; estimating a rotary matrix and a deviation matrix between the plurality of sets of image acquisition equipment by utilizing the fine matching characteristic point pair; and stitching the images corresponding to the scene points respectively according to the rotary matrix and the deviation matrix, so as to obtain a stitched image.

Estimating impedance of energy storage devices includes generating input signals at various frequencies with a frequency step factor therebetween. An excitation time record (ETR) is generated to include a summation of the input signals and a deviation matrix of coefficients is generated relative to the excitation time record to determine crosstalk between the input signals. An energy storage device is stimulated with the ETR and simultaneously a response time record (RTR) is captured that is indicative of a response of the energy storage device to the ETR. The deviation matrix is applied to the RTR to determine an in-phase component and a quadrature component of an impedance of the energy storage device at each of the different frequencies with the crosstalk between the input signals substantially removed. This approach enables rapid impedance spectra measurements that can be completed within one period of the lowest frequency or less.

The invention provides a losing lock detection method based on an extended Kalman filter tracking loop in a GPS receiver. According to a characteristic that a loop gain in an extended Kalman filter can dynamically change with change of a measured value, through a gain variable in the loop, whether the loop is in losing lock or not is detected. In the extended Kalman filter, through a state estimation mean square deviation matrix P, a process noise covariance matrix Q, a measurement noise covariance matrix R and a TS filtering equation, an intermediate variable b and a Kalman filtering gain K are calculated, and through subtracting an observation variable matrix estimated value and the intermediate variable b from an observation variable matrix, an residual error is obtained. By multiplying the residual error by the Kalman filtering gain K, a carrier loop gain coefficient matrix G is obtained. An accumulated value of a parameter G (1, 1) in the first row and in the first column of the matrix G is taken as a loop losing locking detection criterion. When the accumulated value of G (1, 1) obtained by accumulating for 100 times of the parameter G (1, 1) exceeds a threshold, the loop losing lock is determined.

The invention discloses a synthetic aperture radar polarimetric calibration method based on reflection asymmetric information, which includes the following steps: getting an uncalibrated data matrix [O]; converting the uncalibrated data matrix into a covariance matrix, and setting initialization phasors U, V, W and Z to 0 and inverted 'e' (partial) to 1; using the initialization phasors U, V, W, Z and inverted 'e' to adjust a variance matrix [C]; getting a polarization azimuth deviation matrix P corresponding to a P band through dual-band registration; calculating U, V, W and Z according to the deviation matrix P; and calculating inverted 'e'. The traditional polarimetric calibration method can only meet the conditions of smooth terrain fluctuation and stable flight attitude, and thus, the problem that a distributed target fails to meet a reflection symmetry hypothesis cannot be solved through polarimetric calibration. The problem in the prior art is solved using the method of the invention.

The invention provides a circuit breaker operating characteristic analytical method based on the virtual prototype technology, which includes steps: firstly, setting up a speed index system, a displacement index system and a force index system for evaluating circuit breaker operating characteristics; secondly, determining eigenvalue of each index in the index systems; thirdly, setting up a circuit breaker model and obtaining simulation value of each eigenvalue by kinematically and dynamically simulating the circuit breaker operation process; and fourthly, calculating a deviation matrix of the circuit breaker operating characteristics according to a simulation value matrix and a reference value matrix. Applying the method to the circuit breaker design stage enables designers to comprehensively consider multiple operating characteristics. Compared with the prior art only considering several operating characteristics including switching-on and switching-off speed and switching-on and switching-off time, the method has the advantages that ten operating characteristics are comprehensively considered to lead the designers to work out more optimized designs, product development risks can be effectively reduced, and foundations can be established for comprehensive product performance improvement.

The invention discloses a method for improving video compression coding efficiency based on deep learning. The method comprises the following specific steps of S1, obtaining a peak signal to noise ratio PSNR1 of original video inter-frame prediction composed of basic motion estimation and motion compensation; S2, importing an SRCNN model to train inter-frame pictures, thereby obtaining a weight matrix and a deviation matrix, modifying parameters of the SRCNN model, and adjusting a network to obtain optimal training parameters; S3, testing test pictures through utilization of the trained model,thereby obtaining a result PSNR2 tested by the SRCNN model, and comparing the PSNR1 with the PSNR2, thereby obtaining the feasibility of applying the SRCNN to inter-frame prediction coding; and S4, applying the SRCNN model to official code HM16.0 provided by the newest coding standard HEVC. According to the method, through application of the deep learning to the field of inter-frame coding, the coding efficiency of an inter-frame motion acute block can be improved.

The invention provides a satellite tracking method, which relates to the field of electromagnetic wave beam tracking, and is used for realizing the rapid tracking and pointing of satellites. The method comprises the steps that the directivity factor function of the antenna is obtained; the Jacobi matrix of the directivity factor function which is used for determining the angle compensation direction of the antenna is obtained; the angle of deviation matrix of the antenna, relative to the satellite can be obtained through scanning, and the angle of deviation matrix is used for providing a function model to obtain the compensation angle; the compensation angle is obtained based on the directivity factor function, the angle of deviation matrix, and the Jacobi matrix, and the compensation angle is used for guiding the antenna to move at the next time; the compensation angle can be dynamically obtained during the aligning process, the antenna is continuously adjusted based on the compensation angle to scan, track and align the satellite. The invention is advantageous in that the satellite tracking errors can be rapidly converged through collecting wave beam signals having quite a few orquite low frequency, and the satellite aligning with high precision can be realized.

The invention provides a process model modification method based on a step matrix and a process tree, belonging to the field of process model modification. A step matrix is proposed based on footprintcomparison, which can directly compare the deviation between the original process model and the event log, and store the deviation in the model deviation matrix, so that the deviation position can beaccurately located according to the model deviation matrix. For deviations occurring at the concurrency structure of Petri nets, all new activities at the specified concurrency structure are collected and added to the original model as concurrency branches for repair. The process model repaired by this method can reflect the actual business process more accurately and reduce the complexity of themodel. Through the analysis of the data of diagnosis and treatment of diseases in a hospital and the comparison with other methods, the validity and correctness of the repair method are illustrated,and the fitness and accuracy of the modified model are higher.

The invention belongs to the field of automatic processing, and particularly discloses a robot deburring method based on a 3D laser scanner. The method comprises the following steps that a robot grabsa debugging workpiece to move in the visual field range of the 3D laser scanner according to a preset track to obtain a contour point cloud data Pri of the debugging workpiece; the debugging workpiece moves to the position of a tool for teaching programming to obtain a deburring track point coordinates Pj; a to-be-cleaned workpiece is grabbed to move in the visual field range of the 3D laser scanner to obtain a point cloud data Pwi of the to-be-cleaned workpiece in the same manner; the Pri and the Pwi are converted into a robot tail end coordinate system, and a position deviation matrix T andan attitude deviation matrix R are registered to obtain; the Pj is corrected by utilizing T and R, and the corrected deburring track point coordinates Pj' is obtained; and the robot clamps the to-be-cleaned workpiece to move relative to the tool to achieve deburring treatment according to the Pj'. The method has the advantages that the automation degree is high, the consistency of the parts afterdeburring is good, the deburring efficiency and the percent of pass are improved, and the pollution to the environment caused by the dust is reduced.

The invention provides an information fusion method and device, computer equipment and a computer readable storage medium. The method comprises that according to an absolute position coordinate of a navigation object at present time is determined according to a quick response code position coordinate at present time and a deviation coordinate between the quick response code position coordinate anda practical position coordinate of the navigation object, a first difference is determined according to the absolute position coordinate and a dead reckoning coordinate at present, a second noise measurement value at present time is determined according to Kalman filter gain of last time, a first noise measurement value of last time, a mean square deviation matrix of the dead reckoning coordinateof present time and the first difference, and a fusion coordinate of the navigation object at present time is determined according to the first difference, the dead reckoning coordinate of present time, the mean square deviation matrix of the dead reckoning coordinate of present time and the second noise measurement value. The method can be used to converge a filtering result, greatly improves the precision of the fusion coordinate result, and improves the accuracy in positioning the coordinate of the navigation object.

The invention discloses a method and system for judging the performance of an electric automobile power battery. The method includes obtaining the proportion of voltage of each individual battery in the power battery to the total voltage of the power battery at a measuring moment, and recording as an individual proportion, wherein the measuring moment is any moment; calculating a difference value between an ideal weight and the individual weight to obtain a weight deviation of the individual battery; calculating the weight deviations of all the individual batteries, and building a weight deviation matrix; calculating a mean value and standard deviation of the weight deviation matrix; calculating an overall weight deviation; and according to the mean value of the weight deviation matrix, the standard deviation of the weight deviation matrix and the overall weight deviation, judging the performance of the power battery. The method and system for judging the performance of the electric automobile power battery establishes a system for evaluating the difference of the performance of individual batteries in the same electric automobile and a system for evaluating the performance of power batteries of different electric automobiles.

The invention provides a method for estimating the output deviations of a triaxial magnetometer. The method is applied to spacecrafts. The output deviations of the triaxial magnetometer comprise a deviation vector and a deviation matrix. The method comprises the following steps: determining an attitude matrix from star coordinate system to track coordinate system by using a double-vector method; acquiring the measured value of the magnetometer; establishing a deviation model of the magnetometer which is used for defining the relationship among the measured value of the magnetometer, the deviation vector, the deviation matrix and the attitude matrix; and subjecting the deviation model to iterative operation in virtue of Kalman filtering so as to estimate the deviation vector and the deviation matrix in real time. The method provided by the invention is fast in calculation convergence, high in precision and applicable to spacecrafts with stable attitudes and has reference significance to error analysis, in-orbit correction and subsequent designing of the triaxial magnetometer and other triaxial vector sensors with similar measuring principles.

The invention provides a scheme selection decision-making method and device based on a qualified value adaptive regret theory, and the method comprises the steps: carrying out the matrix standardization of an original evaluation matrix, and obtaining a standardized evaluation matrix; taking intermediate values of all values under each attribute in the standardized evaluation matrix as qualified reference values of the attribute; obtaining a utility value of each attribute under each scheme according to the qualified reference value of each attribute, and forming an attribute interval utility value matrix; calculating a regret value of each scheme under each attribute, and obtaining a perception utility value matrix; obtaining a basic probability distribution matrix; obtaining a phase separation degree deviation matrix according to the standardized evaluation matrix, and obtaining the weight of each attribute; obtaining a weight-based basic probability distribution matrix; and performing evidence synthesis, and sorting all the schemes. Compared with the traditional regret theory, the method is higher in flexibility, and can better describe decision-making habits and risk chasing psychologies of people.

The invention discloses a reference-free contrast distortion image quality evaluation method based on contrast enhancement. The method comprises the following steps of firstly, calculating a structural similarity index mean value and a variance between a distortion image and a contrast enhancement image as contrast enhancement characteristics; dividing the contrast distortion image into a plurality of superpixels, and calculating the mean value and the variance of the superpixels as superpixel characteristics; calculating the deviation of the contrast distortion image in the 0-degree direction, the 45-degree direction, the 90-degree direction and the 135-degree direction at each pixel point to form a deviation matrix, obtaining the deviation matrix characteristic values, and then obtainingthe deviation characteristics through calculation; combining the features to obtain a final feature vector; and sending into a support vector machine for training and testing to obtain an objective image quality evaluation result. The method fully considers the histogram of the image, the superpixel segmentation, the gray scale local deviation and the contrast distortion, and improves the qualityevaluation precision of the contrast distortion image.

The invention discloses a high-precision attitude fault-tolerant control method for a deep space detector, which comprises the following steps: 1) establishing a mathematical model for a reaction flywheel fault and an installation deviation; 2) based on the attitude error quaternion theory, establishing a attitude control system model of the deep space detector; 3) analyzing the uncertainty of a reaction flywheel; 4) designing a fast terminal sliding mode surface; 5) analyzing the total uncertainty of the system; 6) designing an adaptive fast terminal sliding mode controller. The method reanalyzes and redefines the interference and actuator efficiency matrix of the attitude control system, the installation deviation matrix and the deviation torque, effectively solves the coupling problem under various uncertainties of the actuator, does not require a ground console to transmit commands, can realized autonomously the fault-tolerant control of the attitude; is good in robustness, fast inconvergence rate, and achieves high attitude control precision under faults.

The invention discloses a study algorithm capable of detecting the inorganization hostile attack in a recommendation system so as to better ensure the quality of a recommendation system. The inventionfirstly provides and solves the study problem of formalizing the inorganization attack detection task in the small-scale attack scene into matrix supplementation. A provided matrix supplementation algorithm is used for obtaining a real grading matrix X of the grading matrix M, a system noise matrix Z and a hostile attack deviation matrix Y. A hostile attacker in users is detected out according tothe information of the hostile attack deviation matrix Y.

The invention relates to a self-adaptive straight line detection method based on Hough transform, and belongs to the technical field of image processing detection. The method comprises the steps of detecting straight lines in contour information through Hough transform; dividing data into a first set and a second set according to whether a slope exists or not; forming a first deviation matrix and a second deviation matrix by each set, and summing absolute values of all elements to obtain a total deviation matrix and an all-0 matrix; then comparing elements in the total deviation matrix with a self-adaptive threshold value, and after judgment, delecting rows and columns corresponding to a straight line passing through a small number of edge points appropriately till the elements in the all-0 matrix are all 0; counting the number of remaining straight lines of the first deviation matrix, and determining whether to continue loop iteration according to judgment until the number of the straight lines is equal to a judgment threshold value; then processing the second set likewise. After the method is adopted, the threshold value can be adaptively adjusted, the implementation of subsequent adaptive extraction is ensured, the speed of acquiring the straight lines is accelerated, and finally, the aim of detecting all straight lines in a picture at one time is fulfilled, so that a foundation is laid for ensuring the accuracy of an optical measurement result.

The invention provides a noise weakening method based on cooperative allocation in soft measurement modeling, and realizes a soft measurement modeling instance based on the method. According to the noise weakening method, a credibility vector is constructed by using a correlation coefficient between auxiliary data and a key variable, an offset vector of each data row is constructed by using an Euclidean distance between each data row and sample space center point data, and a real deviation matrix and a deviation direction of each data are constructed by using a difference value between each data and the sample center point data; and finally, the credibility, the offset degree, the deviation and the intensity factors are combined to perform cooperative noise allocation on the original auxiliary data, and different intensity factors are utilized to find the highest point of performance improvement of the model during noise weakening, so that noise data and non-noise data in the collected industrial data are balanced, the problem that noise in soft measurement modeling data is complex is solved, the noise processing difficulty is reduced, the prediction performance of the soft measurement model is improved, and an implemented soft measurement modeling example shows that the method has high adaptability and stability.

The present invention relates to a method (100) for learning at least one pump feature of a pump (32) for a feed module (32) of an SCR-catalyst system (10), including the following steps: (110) obtaining a to-be-learned reference value (nPmpRef) of at least one pump feature (p, T, q), wherein the reference value (nPmpRef) is saved in a reference matrix; and (120) obtaining a to-be-learned deviation value (DeltanPmp), wherein the deviation value is the deviation between the at least one pump feature and the reference value (nPmpRef), and the to-be-learned deviation value is saved in a deviationmatrix.