62results about How to "Moderate amount of calculation" patented technology

The invention relates to a method and a system for abnormal target detection and relay tracking under a large-range monitoring scene. The system comprises a target detection module, a target identification and tracking control module and an active tracking module, wherein the target detection module is used for carrying out Gaussian background modeling by utilizing a sub-sampled image to obtain aforeground image, computing the coordinate of a target in a digital map and then sending coordinate information to the target identification and tracking control module; the target identification andtracking control module is used for finishing the target tracking, the track recording and the abnormal behavior detection of the target, and if abnormal behavior occurs, a proper pan-tilt video camera is selected according to the coordinate information of the target, and an alarm signal is sent to the active tracking module; and the active tracking module is used for receiving the alarm information of the target identification and tracking control module, controlling the pan-tilt video camera to carry out preset position steering according to the content of the alarm information and then carrying out the detection and the tracking of the moving target. The invention solves the integration problem of multiple paths of image target information under the large-range monitoring scene and realizes the accurate and robust relay tracking of the same target.

The invention discloses a semi-supervised clustering integrated protocol identification method. The method comprises the following steps: various data packets in a network are acquired; received network data is analyzed, and each field of the data packets is extracted and counted; feature code of network data obtained after the network data is analyzed is matched with various feature codes preset in a data base, if the match is successful, the data packets are corresponding protocols; data not successfully matched is subject to cluster analysis, a plurality of base clustering devices are used to cluster the data packets, and the result is fed back, and a priori label value is modified; and a semi-supervised statistical learning is carried out for the result of the clustering of the network data packets and each known protocol, and a discriminant learner is trained. According to the invention, the terminal protocol identification rate is improved, and the amount of calculation is moderate, so that the efficiency is high; one time of dialog generate less flow, inaccurate identification is not easy; and besides, the method integrates a plurality of identification methods, so as to achieve multi-dimension identification. The invention also discloses a corresponding semi-supervised clustering integrated protocol identification system.

The invention discloses a wide dynamic image synthesis method. First, exposure selection is performed according to single frame images of a wide dynamic scene, and exposure images to be synthesized are obtained, the obtained exposure images to be synthesized are processed through region segmentation, motion vectors corresponding to all region blocks obtained after segmentation are calculated, and frame motion registration is performed with the motion vector with the highest frequency as the motion vector of frame motion registration; MTM classification is performed on irradiation images of the exposure images to be synthesized, motion images are obtained through difference between the irradiation images, and motion of objects in the exposure images to be synthesized is registered; finally, the registered exposure images to be synthesized are fused to obtain a final output wide dynamic image. Through the wide dynamic image synthesis method, the fuzzy frame situations such as smears, ghosts and double images caused by shakes of a camera and motion of the objects can be effectively eliminated, the calculation amount is moderate, and the synthesized image is good in effect.

The invention discloses a rehabilitation robot and a human-machine cooperative interaction force control method thereof. The rehabilitation robot comprises a robot arm with at least one joint. The method for controlling the human-machine interaction force of the rehabilitation robot comprises the following steps: detecting an action moment of a patient's affected side limb acting on the robot arm;Generating a target motion trajectory of the manipulator; Establishing a dynamic model of the manipulator; Calculating a joint moment required for generating the target motion trajectory by using aninverse dynamics principle; Calculating a friction compensation torque of the joint and a dynamic compensation torque of the affected side limb of the patient; The joint torque, the action torque, thedynamics compensation torque, and the friction compensation torque are used as control inputs to the manipulator. The robot can provide the compliant force interactive control between the patient andthe rehabilitation robot arm, provides complete dynamic compensation, improves the equipment motion starting ability and flexibility, and is favorable for improving the experience degree of the patient to the auxiliary rehabilitation training. The invention can provide the compliant force interactive control between the patient and the rehabilitation robot arm, provides complete dynamic compensation, and improves the equipment motion starting ability and flexibility.

The invention discloses an urban road network motor vehicle exhaust real-time remote sensing monitoring base address selecting method is composed of a preparation work layer, an addressing algorithm layer and a scheme decision-making layer. After a monitoring region is determined and relative information is collected, modeling is carried out according to the characteristics of roads, then inherent similarities of an urban road network are searched and classified based on the model, a representative road is selected in each kind, equipment selection is carried out according to road environments, and each scheme is operated individually. Next, each classifying result is assessed, and an optimal scheme is selected. The method is simple in principle and easy to carry out, the calculation amount is modest, and a good effect is obtained.

The invention provides an identification method of a woven fabric tissue chart, wherein the identification method relates to image analysis. The identification method comprises the steps of a first step, identifying the type of the woven fabric tissue structure based on a yarn boundary characteristic, namely performing woven fabric image pre-treatment, performing inclination correction and dividing for obtaining each tissue point image, determining the latitude and longitude attributes of the tissue point image by means of the yarn boundary characteristic of the tissue point image, obtaining the number of circulating yarns of the woven fabric and classifying the tissue structure; and a second step, performing woven fabric tissue chart identification based on an improved Gabor characteristic and the tissue structure type, namely extracting the characteristic of a non-twill tissue point image based on the improved Gabor kernel transformation, performing PCA dimension reduction, classifying the non-twill tissue point image by means of a support vector machine, performing classified correction on a woven fabric tissue chart matrix, and outputting a correct woven fabric tissue chart. The identification method of the woven fabric tissue chart can realize a high-robustness identification effect on woven fabric three-elementary tissues which are obtained through weaving yarns with different dimensions and colors and the woven fabric tissue chart with a simple changed tissue structure.

The invention discloses a method for quickly estimating underwater target shell radial vibration low-frequency sound radiation signals based on mode decomposition, and belongs to the technical field of underwater target sound radiation. The method includes the following steps that firstly, based on uncertainty of a researched object, the range of researched frequency bands is determined; secondly, based on a mode decomposition theory and different model structures, a simulation sound pressure equation is determined; thirdly, based on a mode truncation theory, an appropriate lower order axial mode order is calculated by adopting a higher circumferential mode order; fourthly, an appropriate lower order circumferential mode order is calculated by adopting a higher axial mode order; eventually the radiation signals through simulating calculation are acquired through calculation, and then the underwater target sound radiation condition is analyzed. According to the technical scheme, the underwater target sound radiation signals generated by radial vibration are accurately simulated through the few mode orders, and therefore calculation complexity is low and accuracy is high.

The invention relates to a single-viewpoint video depth obtaining method based on scene classification and geometric dimension. The single-viewpoint video depth obtaining method specifically comprises the following steps: (1) judging whether a current frame image belongs to a scene in which a camera is static and an object moves or a camera moving scene; (2) judging whether needing to estimate an initial depth graph of the current frame image; (3) figuring out the initial depth graph of the current frame image; and (4) for the scene in which the camera is static and the object moves, obtaining a motion depth graph of the current frame image and fusing the motion depth graph of the current frame image with the initial depth graph; for the camera moving scene, carrying out global motion compensation, carrying out motion estimation on adjacent frames of images after the global motion compensation by using an optical flow method, judging whether a moving object exists, and determining whether needing to fuse with the initial depth graph. By adopting the single-viewpoint video depth obtaining method provided by the invention, no specific scene is relied on, the calculated amount is appropriate, the generated noise is small, a depth graph better conforming to actual scene distribution is obtained, and a 3D video with a better effect is synthesized.

The invention provides an improved short-wave unit positioning method, and aims to provide a positioning method with the advantages of high-resolution positioning performance, small calculation amount, and high estimation precision. The improved short-wave unit positioning method is implemented by the steps of: acquiring a beam domain covariance matrix through solving an autocorrelation matrix based on beam domain output data; calculating eigenvalues and corresponding eigenvectors of the beam domain covariance matrix through decomposition, judging whether a signal source number exits, and determining signal source subspaces and noise subspaces corresponding to the signal source number, so as to obtain estimated values for automatic pairing of signal sources; then constructing a polynomialf(z) and solving roots of the polynomial f(z); solving a root high resolution spectrum, solving roots (on an approaching unit circle) of which the number is the same as that of dimension numbers of the signal subspaces, combining with array element data, beam number and corresponding physical parameters, performing external radiation signal source direction-of-arrival estimation, acquiring eigenvectors of the signal source number, performing deep learning, and adjusting positioning precision of a short-wave single station automatically.

The invention provides a transonic limit cycle flutter analysis method. The transonic limit cycle flutter analysis method comprises the following steps: through equivalent linearization of a describing function, processing nonlinear characteristics of a transonic aerodynamic force; substituting a frequency-domain aerodynamic coefficient after the equivalent linearization into a flutter equation; obtaining a flutter speed and a flutter frequency by a frequency-domain flutter solving method; and for different limit cycle amplitudes, calculating to obtain different flutter speeds and flutter frequencies to form limit cycle characteristics of a transonic nonlinear flutter. The equivalent linearization is performed on the nonlinear characteristics of the transonic aerodynamic force through the aerodynamic describing function, the flutter equation is solved within a frequency domain, so that the flutter speed and the flutter frequency at a given limit cycle amplitude can be accurately predicted; and the transonic limit cycle flutter analysis method is moderate in computation amount, easy to master and very good in robustness.

An on orbit calibration method based on Weng model for the star-sensitive device includes such steps as creating a posture transform array of star-sensitive device, creating the Weng imaging model, acquiring data, calculating parameters, and calibrating.

The invention is concerned with the harmonic current prediction method for compensating the controlling time-lapse, belongs to the harmonic current compensation technique field. Firstly, it measures the loading current instantaneous value in real time and low passes the filtering, and gets the phase corner of the electric-net voltage by phase-locked loop at the same time; predicts the current volume of the sampling time currently according to the electricity-net current phase corner and the fitting parameter vector; adjusts the fitting parameter vector according to the predicted error; then, according to the adjusted fitting parameter vector, the electricity-net voltage phase corner and the requested compensatory controlling time-lapse, it can predict the loading current volume of the tk+DeltaT time; and predicts the request compensatory harmonic current signal of the tk+DeltaT at last.

A hybrid signal source locating method includes: constructing a fourth-order cumulant matrix according to the received data of a uniform linear array model; estimating the two-dimensional electronic angles of the hybrid signal source one by one; performing eigen-decomposition on the covariance matrix of the received data of the uniform linear array model to estimate the amplitude-phase response of the array model; using a plurality of calibration elements to eliminate the blurring phase in the amplitude-phase response; performing iterative operation on the three processes of the two-dimensional electronic angle estimation, the amplitude-phase response estimation, and the blurring phase elimination until the convergence; obtaining the convergence values of the amplitude-phase response diagonal matrix and the two-dimensional electronic angles; determining the type of the incident signal source; and outputting an incident signal source locating result. The present invention also provides a hybrid signal source positioning system. The implementation of the technical schemes provided by the invention can effectively reduce the computational complexity and shorten the operation time for locating, and meanwhile locates the hybrid signal source of the near field and the far field, improving the locating accuracy and the application range.

The invention discloses a method for embedding and extracting digital watermarks in specific to a three-dimensional point model, comprising the steps of: firstly, reconfiguring the coordinates of a model by adopting a principal component analysis method, then, forming a plurality of projected images inside a spherical surface with a gravity as a sphere by using a visual operator and taking the gravity center as a view point, converting a three-dimensional model into a two-dimensional image, regulating a spectrum coefficient of a low-frequency part by applying a two-dimensional frequency domain analysis method, and mapping the low-frequency part back to the three-dimensional model to realize the embedding and the extracting of the watermarks of the three-dimensional point model. In the invention, the calculated amount is moderate, the watermark capacity is large, and the method is suitable for large-scale point models and can be used for resisting various affine transformation attacks, achieving certain robustness to peak reordering, peak resampling and other various attacks, and effectively protecting the copyright of three-dimensional products.

The invention relates to an Internet of things trusted routing method based on Core-Selecting and a reputation mechanism. The method comprises the following steps that: a source node inquires history interaction record firstly, sends a relay node selection request message to a most trusted node, and broadcasts the relay node selection request message to all neighbor nodes if the most trusted node cannot be found or a selected node does not make feedback; the neighbor nodes assess the confidence level of a request node by a reputation model based on the Core-Selecting after reception of the request message, and provides feedback information indicating whether or not the neighbor nodes wound like to provide relay services; and the source node performs confidence level assessment on the nodes which would like to provide the relay services by the reputation model based on the Core-Selecting after reception of the feedback information, and selects the most trusted relay node finally. Through adoption of the method, the routing security and reliability can be enhanced effectively, and internal collusion attacks can be resisted effectively.

The invention discloses a Transform-based lightweight early fire detection method. The method comprises the steps of establishing an indoor and outdoor multi-scene fire initial image sample data set; the method comprises the following steps: designing a Transform-based lightweight backbone network, introducing a linear attention enhancement mechanism into a Transform structure, introducing locality into a feed-forward network through sequence and image conversion and depth separable convolution to realize global and local feature processing of an image, and performing down-sampling through an inverse residual block to obtain feature maps with different resolutions; and further enhancing feature extraction through feature enhancement and a multi-scale feature fusion structure, and finally performing detection in the mixed feature map to obtain a flame target detection result. According to the method, the advantages of Transform and the convolutional neural network are combined, while feature extraction is optimized, network parameters and calculation amount are reduced, a lightweight detection model is constructed, high detection precision is ensured, high detection speed is realized, and multi-scene fire early-stage target detection can be well realized.

The invention provides a high-resolution low-sidelobe deconvolution spectrum estimation method based on space-time processing, and belongs to the field of signal processing. The method is suitable forthe high-resolution power spectrum estimation of a received array signal at frequency and azimuth under the condition that the data sample is limited. According to the invention, a deconvolution algorithm is used for the multi-dimensional deconvolution operation of an array signal power spectrum with the limited length and a multi-dimensional window function power spectrum, so the method can completely eliminate the impact caused by a window function under the ideal conditions, and obtains a real power spectrum of a signal. Based on the conventional convolution technology, the method can greatly improve the frequency resolution and angle resolution of a power spectrum estimation result, can greatly inhibit the sidelobe caused by the limited length of data, provides an additional signal tonoise ratio, and provides a new idea for the multi-target detection and recognition and weak signal extraction in the array signal processing. In addition, the method is high in operability, is moderate in calculation burden, is stable in calculation result, and is very wide in application range.

The invention discloses an intelligent English pronunciation self-service learning system. The intelligent English pronunciation self-service learning system comprises a standard data module, a data selecting module, a data collecting module, a data storing module, a data analyzing module, a data exporting module and an intensifying module. The data selecting module carries out multi-level database classification, a user selects the corresponding classification database according to the demand, and the system automatically retrieves the English data of the corresponding stage, field and complexity; the collected voice data is stored in the data storing module, and the data analyzing module carries out voice recognition on the collected voice data, and carries out comparison feedback with standard voice data; the data exporting module carries out pronunciation error correction according to the analysis result, and broadcasts the standard English pronunciation data; and the intensifying module retrieves the relevant words in the database, so that the user can carry out intensive training. With the system, the user can carry out the self-service English pronunciation practice, and meanwhile, error correcting and intensive training can be carried out, so that the pronunciation level is obviously improved.

The invention discloses an iterative phase feeding calculation method for improving beam pointing accuracy of planar phased array antenna. First, the coordinates of each array element are determined by taking the central symmetric point of the planar phased array antenna as the origin, and all elements are numbered in turn according to the central symmetric characteristics. Then, the first half ofthe array elements are taken to calculate the theoretical and actual feeding phases of each array element in turn, and the actual feeding phases are obtained by carrying or rounding quantization of the theoretical feeding phases. The principle of minimum pointing error of cumulative beams and the feeding of the latter half of the array elements are completed according to the centrosymmetric characteristics. The invention improves the beam pointing accuracy of the planar phased array antenna effectively with moderate calculation amount and complexity, meeting the requirements of fast beam scanning and tracking in the phased array antenna system.

The invention relates to a method for predicting turning machining deformation of a thin-wall complex curved surface rotating member, and belongs to the technical field of machining. Firstly, a tool cutting edge angle and cutting line speed are taken as experimental factors, a two-factor multi-level full factor test is performed to obtain residual stress distribution under each kind of combination; then, according to the obtained multiple sets of residual stress distribution, local residual stress values corresponding to different places in a surface stress layer are obtained by a multidimensional linear interpolation method, and the reconstruction of the machining of a non-even residual stress field is completed; then, according to a position relationship of a local coordinate system anda global coordinate system, the conversion from a local stress field to a global stress field is achieved; lastly, loads and boundary conditions are applied, and the rotating member turning machiningdeformation on the thin-wall complex curved surface is calculated. According to the method for predicting the turning machining deformation of the thin-wall complex curved surface rotating member, thenon-uniformity of processing residual stress distribution caused by related parameter changes in the actual processing of the thin-wall complex curved surface rotating member is considered, thereby the calculation amount of the method is moderate, the quality requirement for grids is general, and the efficiency and the accuracy are simultaneously achieved.

The invention discloses a dynamic wireless charging load prediction control method for electric vehicles, and belongs to the technical field of electric vehicle wireless charging. The method includesutilizing an electric vehicle dynamic wireless charging system to perform prediction control on a receiving end load; taking the collected current and voltage of a load battery as the input signal ofa controller; and judging charging modes and switching the charging modes by estimating the charged state of the load battery, and converting an output signal into a duty cycle control DC / DC conversion module to realize power regulation. The electric vehicle dynamic wireless charging system can realize the reasonable distribution of system resources, so that the electric energy loss during the transmission of electric energy can be saved, magnetic flux leakage can be reduced, the transmission efficiency of the system can be enhanced, situations of overcharging can be decreased, the service life of vehicle batteries can be prolonged, and therefore, the stability of the dynamic charging of electric vehicles can be guaranteed, and safety and reliability can be improved.

The invention provides a real-time finding method for abnormal conditions of the in-orbit satellite thruster ignition period. A temperature relative difference model is built by means of in-orbit satellite historical telemeasuring data, and real-time, automatic and accurate monitoring of the thrust ignition process can be achieved by driving and discriminating real-time satellite telemeasuring data. The method comprises the specific steps of firstly, determining whether a satellite is in the position keeping working condition or not, determining two thrusters adopted by the current position keeping working condition, then, building the temperature relative difference model for building and setting an alarm threshold, finally, utilizing data obtained through the temperature relative difference model of the real-time in-orbit telemeasuring data for real-time comparison with the alarm threshold, and performing real-time alarming when the threshold is exceeded. According to the method, thein-orbit monitoring threshold obtained through the relative difference model can find the thrust in-orbit abnormities of the thrusts in real time, precious time is fought for emergency processing ofabnormities, and reliable guarantees are provided for long-term in-orbit stable operation of satellites.

A limited redundancy discrete wavelet transform method is used in the field of image processing, image fusion and signal processing. The invention firstly designs the filter, which is used respectively in the first class and the subsequent dissociation class, and then takes different sub-sampling rules according to different channels during the process of wavelet dissociation to, and uses the filter corresponding to its positive transformation and super-sampling rules during the process of inverse transformation. The invention, which uses different filter design methods in different class of discrete wavelet transform, and sub-sampling, super-sampling strategies according to different channels, can make the wavelet coefficients to have good invariance and appropriate redundancy. The invention can be applied to image processing, image fusion, signal processing and transformation, which can be used in radar, voice, geographic information processing and other home and military systems.

The invention discloses a real-time remote sensing monitoring base site selection method for city motor vehicle tail gas. The base site selection method comprises a preparation work layer, a site selection algorithm layer and a solution decision layer. After a monitoring area is determined and relevant information is collected, modeling is performed according to characteristics of a road, then an internal similarity of a city road network is searched based on a model and roads are classified, a representative road is selected from each classification, and a corresponding type device is selected according to a road environment, wherein each scheme runs independently. Each classification result is assessed to obtain an optimal scheme.

According to the Chinese medical question classification system based on deep encyclopedia learning, by using a semantic structure of Chinese search encyclopedia in combination with a deep learning method, a method for constructing a feature vector more efficiently and accurately is provided, which comprises: using a semantic association degree efficient convergence method based on the semantic structure of the Chinese search encyclopedia for constructing a network inquiry question feature vector; based on the features of the medical questions, improving a semantic association degree algorithm, solving the defect that the speed is low when feature vectors are constructed, and expanding feature words by extracting Chinese search encyclopedia word links; on the basis of a distributed Chinese word vector space of a CB-CBS language model, achieving efficient dimensionality reduction of network inquiry question feature vectors, avoiding the problem of data sparseness, greatly improving the inquiry classification efficiency; and using the CB-CBS model in combination with Chinese search encyclopedia and deep learning to construct distributed medical question word vectors, constructing a professional medical question corpus, and improving the accuracy of the word association degree and the medical question classification efficiency remarkably.

The invention provides a phase fraction low moment-based covariance difference propagation algorithm, and mainly solves the problem of parameter estimation of a far-near field mixed signal source of an array radar in an alpha noise environment. The method comprises the following steps: firstly, establishing a model for receiving signals by a uniform and symmetrical array antenna, constructing a covariance matrix based on a phase fraction low-order moment to suppress alpha noise, and performing angle estimation on a far-field source through a MUSIC algorithm; then separating a near-field covariance matrix by combining a covariance difference thought and utilizing structural differences of a far-field covariance matrix, a noise covariance matrix and the near-field covariance matrix; and finally, introducing a propagation operator to solve a noise subspace corresponding to the covariance difference matrix at the moment, and further estimating a near-field source angle parameter and a distance parameter through spectrum peak search. According to the method, the alpha noise is suppressed while certain calculation complexity is reduced, a subjective standard does not need to be used for distinguishing far-near field source types, and the parameter estimation precision is high.

The invention discloses a license plate locating method and device. The method includes: carrying out pre-detection to determine a candidate license plate region; selecting proposal sampling status, which corresponds to proposal distribution, in the candidate license plate region; calculating license plate feature probability of the proposal sampling status according to a color feature, a shape feature and a character feature of a sampling region corresponding to the proposal sampling status; according to the obtained license plate feature probability, determining whether the proposal samplingstatus is added into a sampling status set; returning to repeated execution of the step, of selecting proposal sampling status, which corresponds to the proposal distribution, in the candidate license plate region, to acquire a sampling status set; and determining that a sampling region corresponding to sampling status of highest license plate feature probability in the sampling status set is a license plate location region. According to the method, the license plate region can be presented by using the maximum posterior probability of the image appearance features, design of a proposal probability distribution function is coordinated, a Markov Chain Monte Carlo (MCMC) method is adopted, a calculation process is enabled to be fast and effective, and the purpose of precise locating is achieved.

The invention discloses a method and a device for predicting a thin reservoir in lacustrine facies beach bar sand bodies. The method comprises the steps of: respectively dividing strata sequence levels of different single wells according to historical data and logging data in a target region; comparing the sand bodies of different single wells at the same strata sequence level to obtain a sand body comparison result; preprocessing and standardizing logging curves of the single wells, and screening an inversion sample curve from the logging curves according to lithologic data of the logging curve; performing band-pass filtering on the seismic data according to the sand body comparison result to obtain seismic reflection data of the thin reservoir; carrying out waveform phase-controlled random inversion according to the inversion sample curve, a sound wave curve and a density curve in the seismic data and the thin reservoir seismic reflection data, and acquiring a thin reservoir inversion result; and determining the position of the thin reservoir according to the thin reservoir inversion result. By adopting the method, the prediction precision of the thin reservoir in the lacustrinefacies beach bar sand bodies can be improved.

The invention discloses a cross-domain knowledge assistance system based on a neural network and deep learning. The cross-domain knowledge assistance system comprises a plurality of devices (1) distributed in different knowledge fields, a deep learning coordination module (2) and a global data storage module (9), each device (1) comprises a data cleaning module (3), a data acquisition module (4) and a data reading module (5); the device (1) further comprises a single-machine storage module (7), and the single-machine storage module (7) is in data communication connection with the data cleaning module (3), the data acquisition module (4) and the data reading module (5). According to the federated machine learning-oriented full-process service migration method and system, before data record training is carried out, data cleaning is carried out on the data records, so that abnormal parts of the data records are eliminated, the accuracy of the data records is ensured, and the accuracy of a data model is ensured.

The invention discloses an infrared image noise suppression method for enhancing details. Different noise reduction methods are used for images in different areas by using noise classification; main noise in the selected area is effectively eliminated, detail information of the infrared image and edge information of the image are effectively reserved through a Sobel operator, the two images are fused through a Laplace pyramid fusion algorithm, and an image with a large number of image details and lower noise compared with a traditional method is obtained. The method is simple in thought, low in complexity, moderate in calculated amount and good in effect.