34 results about "Recursive convolution" patented technology

The invention discloses a time domain integral equation method for analyzing plasma electromagnetic scattering properties. The method analyzes electromagnetic scattering properties of a chromatic dispersion plasma model through a time domain integral equation. According to the method, a discrete object model is needed only, a free space green function and a recursive convolution method are used, electromagnetic scattering properties of an inhomogeneous chromatic dispersion plasma target can be analyzed, and compared with a finite difference time domain method and a time domain finite element method both needing total space discretization, the method provided by the invention has very high calculation accuracy and relatively less unknown quantity; meanwhile, since the time domain method is used for solving, electromagnetic scattering properties of each frequency point of broadband can be calculated one time, and compared with a frequency domain method needing a frequency sweeping operation, the method provided by the invention consumes greatly reduced time for solving.

The invention relates to a revolution-modulation serially concatenated coding and decoding system with a star map, which pertains to the field of numerical information coding and decoding and is characterized in that: by artful mapping and analyzing, modulation-analysis is divided into three independent parts: even-odd check coding, recursive convolution coding and a memoryless 8PSK or square M-QAM modulation. Memorability contained in the modulation process is taken as a united internal code and then forms the serially concatenated coding system together with a simple outer code concatenated. Considerable coding gain and excellent error code performance can be obtained at a receiving end by iterative decoding.

The invention discloses a time domain integral equation method for analyzing the electromagnetic scattering characteristic of a hypersonic flight object. The time domain integral equation method comprises that a time domain volume-surface integral equation is established; according to the structural scattering characteristic, the total electric field on the stimulated object is equal to the sum of an incident electric field and a scattered electric field, and the incident electric field is stimulated; the integral equation is discretized, tests in space and time are carried out, and a time domain matrix equation is established; the relationship between the dielectric flux density and the electric field intensity of plasma is established; and the time domain matrix equation is solved, and the electromagnetic scattering characteristic of a mixed object in broadband is analyzed. According to the invention, only the object model needs to be discretized, by means of a free space green function and a recursive convolution method, a mixed object containing heterogeneous plasma medium and metal can be analyzed, the calculating precision is high, and the number of unknowns is relatively small; at the same time, a time domain method is employed for solution, the electromagnetic scattering characteristic of each frequency point in broadband can be calculated for one step, compared with a frequency domain method that frequency sweep operation is needed, the time domain integral equation method is advantaged in that the solving time is reduced.

The invention relates to a video content semantic understanding method based on a recurrent convolutional neural network. The method is used for carrying out content analysis and classification on video data such as network videos and monitoring videos. According to the method, a convolutional neural network is placed in a recurrent neural network to serve as a kernel; Presenting video state concepts, organic combination of target detection and inter-frame association in the video data is realized; By performing recursive convolution operation between video frames, accurate and efficient extraction of video features is realized, video representations with more semantic representations are obtained, and on the basis, tasks of video classification, event detection, scene recognition and thelike are completed by adopting an artificial neural network full-connection classifier. The method provided by the invention overcomes the problems of information loss, poor feature characterization capability, training convergence difficulty and the like in the traditional method, and is an accurate, efficient and advanced method with a wide application prospect.

The invention discloses a method and a system for coding a multi-system network Turbo code. The method comprises the following steps: obtaining a symbol sequence after carrying out error correction coding and modulation on user information; merging each relay node in the maximum ratio to receive and demodulate an antenna receiving sequence, then respectively directly carrying out recursive convolution coding and carrying out recursive convolution coding after carrying out code element interleaving to obtain two groups of check digit sequences, and carrying out orthogonal space-time block coding to obtain a relay node emission matrix; merging each destination node in the maximum ratio to receive and demodulate a user node transmitting sequence received by an antenna, storing and carrying out orthogonal space-time block decoding on a received relay node transmitting sequence, and demodulating to obtain a check digit sequence; and merging and forming the multi-system network Turbo code. The invention can form the virtual MIMO for the single antenna user so as to obtain the transmitting and receiving diversity gain which can be provided by the system, promote the error correcting capacity of the transmission network, improve the network throughput and better solve the coverage problem of the marginal cell.

The present invention relates to a basketball goal detection method and device based on video, the method comprising: constructing a recursive convolutional neural network model based on a convolutional neural network and a recursive neural network; constructing a picture library sample set of a basketball video, said The pictures in the picture library have tags, and the tags include timing information and goal identification information; the recursive convolutional neural network model is trained based on the sample set; images are extracted from the basketball video to be detected, and the training is used The final recursive convolutional neural network model processes the extracted image to obtain an output vector; according to the output vector of the recursive convolutional neural network, it is judged whether there is a goal in the current basketball video. The invention can accurately detect whether a basketball is scored or not based on the video.

The invention discloses a recursive convolution-based circuit transient response calculation method. The method comprises the steps of (1) obtaining a decomposition formula of a transfer function H(s), and determining an expression of an excitation function h(t); (2) based on an input voltage Vin(t0) and the decomposition formula of the transfer function H(s), performing Newton-Raphson iteration of a state variable Vout,i,j(t0) and an output voltage Vout(t0) at an initial moment t0; (3) based on a state variable Vout,i,j(tn-1) at a previous moment tn-1 and an input voltage Vin(tn) at a current moment tn, performing Newton-Raphson iteration of the state variable Vout,i,j(t0) and the output voltage Vout(t0) at the current moment tn; and (4) repeatedly executing the step (3) until a simulation ending moment. According to the recursive convolution-based circuit transient response calculation method, in circuit simulation, a method for quickly calculating the output voltage is provided for sub-circuits or elements with given input voltage and transfer function in combination with characteristics of transient circuit simulation by utilizing a basic thought of recursive convolution calculation.

The invention discloses a video behavior identification method and system. The method comprises the following steps: carrying out multi-level feature extraction on a to-be-identified video; performinginitial detection on the ROI of the target object by using a deep full convolutional network; carrying out ROI fine adjustment by utilizing a Markov random field to obtain an ROI set of a final target object; and based on the ROI set of the final target object, carrying out single-person behavior identification and group behavior identification at the same time. According to the method, the consistency of time sequence information in a group is considered, meanwhile, the difference of individual time sequence information is also considered, single-person behavior recognition based on ROI timesequence reasoning is helpful for better extracting discriminative single-person behavior characteristics, and the recognition precision is improved; based on the ROI matching recurrent convolutionalnetwork, the ROI information of a single person in the time domain can be fused and propagated, and the method is an effective method for solving the problem of video behavior recognition.

The invention provides an nport problem self-adaptive fitting and simulation method based on a vector fitting method. The nport problem self-adaptive fitting and simulation method includes the following steps that 1, the vector fitting method is used for self-adaptive fitting of an S parameter, and the optimal fitting result SIJ (s) of the fitting order K and the delay time d is obtained, wherein1< / =i and j< / =n; 2, through Laplace inverse transformation, a transfer function Sij (s) in a frequency domain is converted into an excitation function Sij (t) in a time domain; 3, the convolution of the S parameter is calculated through a recursion method, and thus an algebraic equation of an nport model is obtained. According to the nport problem self-adaptive fitting and simulation method basedon the vector fitting method, the vector fitting method is used for self-adaptive fitting of the S parameter and the recursive convolution is used for quickly calculating the convolution of the S parameter, higher fitting accuracy and better fitting effect are achieved, and an approximate computational format simple in form and very high in accuracy is given through the Taylor.

A method and apparatus for fast digital filtering that requires only filter stages of first and second order. A desired rational filter transfer function is represented as a sum of first and second order intermediate transfer functions. A time dependent input signal is first fed in parallel into a plurality of first and second order intermediate recursive filter stages. Then, the outputs of the intermediate filter stages are summed up to an output filter signal that corresponds to the desired rational filter transfer function. The method and apparatus reduces the amount of calculational effort to the order of O(N), where N denotes the number of sampling points in the time domain, because the digital filtering is based on a discrete recursive convolution in the time domain.

The invention discloses a method for encoding and decoding applicable to multi-user multicast communication and a device thereof. The method comprises the following steps that: an information sequence of each user in n users to be communicated with each other in a small area is saved as an information sequence with length of m bits; n information sequences with length of m are encoded to obtain output information yi<j>=Xi<j> square x(i=1)<j>, and are repeated for m times to obtain information sequences of n users y1<1>y1<2>lambada yn-1<1>lambada y(n-1)<m>; the information sequences are transmitted to the users in the group according to a multicast mode; each user decodes the received data by a recursive convolution decoding mode, the ith user decodes the received data according to a sequence of yi<j>y(n+1)<j>lambada y(n-1)<j>y1<j>lambada y(i-1)<j>, and the ith user can acquire X(i+1)<j>x(i-1)<j>, information of the jth bit of n users can be acquired in the same mode, and information of m*(n-1) of other (n-1) users can be acquired by repeating the steps. The method and the device improve the utilization rate of channel resource and the security of the system, and can be realized simply.

The invention relates to an algorithm for realizing a perfectly matched layer by truncating plasmas based on a CNDG algorithm, and belongs to the technical field of numerical simulation. The algorithm is technically characterized in that a complex frequency offset stretch coordinate variable is substituted into a two-dimensional corrected Maxwell equation set, and an auxiliary differential equation method is used. In a process of deriving a relationship between dielectric flux density D and an electric field E, a piecewise linear recursive convolution method is applied. By applying a CNDG method, the electric field is solved by solving a three-diagonal equation, so that the calculation efficiency is improved and the time is saved. The algorithm has the advantages of unconditional stability and increased electromagnetic field calculation speed.

The invention discloses a system for realizing sight line estimation and attention analysis based on a recursive convolutional neural network. The system comprises a sight line feature extraction module, a sight line regression module, a sight line drop point mapping module and an attention visualization and analysis module. The invention also relates to a corresponding method which comprises the following steps: simultaneously extracting the apparent characteristics of the eyes and the posture characteristics of the head during implementation, and carrying out spatial domain characteristic fusion; for continuous multi-frame sight line features, performing joint coding on time sequence features of a gazing behavior through a Bi-LSTM network layer, completing time domain feature fusion, and further performing regression to obtain a sight line vector of a middle frame; the invention further provides a sight line drop point resolving method based on the monocular camera, the sight line drop point coordinates can be obtained in real time, and the method is not limited by scenes; data support is provided by a bottom layer module, and abundant real-time sight tracking visualization and relevant sight parameter visualization forms are provided in an attention visualization and analysis module. According to the technical scheme, the accuracy and stability requirements of use scenes can be met, and the application scenes are wide.

The invention provides an nport problem self-adaptive fitting and simulation method based on a vector fitting method. The nport problem self-adaptive fitting and simulation method includes the following steps that 1, the vector fitting method is used for self-adaptive fitting of an S parameter, and the optimal fitting result SIJ (s) of the fitting order K and the delay time d is obtained, wherein1< / =i and j< / =n; 2, through Laplace inverse transformation, a transfer function Sij (s) in a frequency domain is converted into an excitation function Sij (t) in a time domain; 3, the convolution of the S parameter is calculated through a recursion method, and thus an algebraic equation of an nport model is obtained. According to the nport problem self-adaptive fitting and simulation method basedon the vector fitting method, the vector fitting method is used for self-adaptive fitting of the S parameter and the recursive convolution is used for quickly calculating the convolution of the S parameter, higher fitting accuracy and better fitting effect are achieved, and an approximate computational format simple in form and very high in accuracy is given through the Taylor.

Embodiments of the present invention provide a behavior recognition method, device, device, and readable storage medium based on an LRCN network, wherein the method includes: acquiring a video frame sequence to be recognized and a corresponding optical flow graph; The sequence and the corresponding optical flow graph are input into the long-term recursive convolutional network model to obtain the behavior category label of the video frame sequence to be identified, and each adjacent preset number of frames in the video frame sequence to be identified is input In the first convolutional neural network in the long-term recursive convolutional network model, input the optical flow graph corresponding to the preset number of frames into the second convolutional neural network in the long-term recurrent convolutional network model, the convolutional neural network Convolutional layer sharing is performed by data fusion for a preset number of frames and optical flow maps respectively. This scheme introduces sharing between convolutional layers, which reduces the large amount of redundancy in the image information between adjacent frames before performing behavior recognition, thereby helping to reduce the overall computational load of the network.

The invention discloses an image multi-scale feature extraction method based on a cellular neural network, and the method comprises the steps: firstly generating a plurality of pairs of feature maps in a binuclear recursion convolution mode based on a local binary constraint improved cellular neural network under the condition of neuron neighborhood L (r, p) sampling; compressing the state featuremap by using rotation invariant mapping and low-frequency mode combination; generating a single-scale joint mode proportional histogram of the image on the state feature map and the response featuremap according to a joint distribution mode statistical rule; performing softmax optimization on the joint histogram, and adding a standard variance component, so that an optimized single-scale featurevector can be obtained; and finally, connecting the plurality of single-scale vectors in series to obtain a multi-scale feature vector of the image.

The invention provides a recursive convolution method for calculating coupling response and stability of a rotorcraft body. The recursive convolution method comprises the following steps: establishing a rotor body coupling dynamics analysis model; deriving a differential equation based on a rotor body coupling dynamics analysis model, linearizing a coefficient matrix of the differential equation in a 0 response manner, forming a linear equation, and reducing the order of the linear equation into a first-order standard equation form; obtaining a homogeneous solution and a forced response steady-state solution based on a first-order standard equation; calculating the state transition matrix of the time point of one circle based on the period integral of the state transition matrix; calculating the characteristic value of the state transition matrix, and judging the coupling stability of the rotor and the aircraft body based on the characteristic value; if the characteristic value of the state transition matrix is smaller than 1 and the rotor and body coupling response meets the convergence requirement, calculating a periodic value of a right-end excitation item, and a first-cycle convolution integral response; and on the basis of the response of the first cycle convolution integral, the state transition array and the periodicity of the excitation item, using a recursive convolution integral calculation method to calculate the response of the second cycle convolution integral.

A model for a circuit cell used in timing and signal integrity analysis in an integrated circuit design is automatically generated. A behavioral model, such as a gate current model is used in which the current in the circuit cell is determined as a function of the input voltage and the output voltage of the circuit cell as well as the history of at least one of the current, voltage, and charge values of the circuit cell. For example, the current in the circuit cell may be a function of the history of the current, which may be calculated incrementally using recursive convolution at each time step when using the model.

The invention relates to a dynamic community discovery method based on a recurrent convolutional neural network and an auto-encoder. The method comprises the following steps: firstly, constructing a network spatial feature learning model based on a convolutional neural network, and learning spatial topological features of the network to obtain a network spatial feature vector; secondly, fusing a network spatial feature learning model based on a convolutional neural network, constructing a network spatial-temporal feature learning model based on a recurrent neural network, the convolutional neural network and an auto-encoder by taking a network spatial feature vector as an input of the model, and learning spatial-temporal features of the network to obtain a network spatial-temporal featurevector; and finally, community discovery is performed on the basis of the network space-time feature vector so as to detect the dynamic community structure of the social network. The method can be applied to analyzing the social network, autonomously learning and extracting the spatial and temporal features of the social network, and can further improve the modularity of a community structure, thereby revealing the topological structure and the like of a real network, and further effectively predicting network user behaviors, information propagation and the like.

The invention relates to the technical field of electromagnetic compatibility, and in particular relates to a transient response analysis method for a transmission line system terminated with a frequency-variable load. The method comprises the following steps: firstly, measuring or calculating the admittance of a frequency-variable load port at a sampling frequency point to obtain corresponding sampling admittance, and carrying out equivalence on the admittance of the port by adopting a rational function approximation mode; secondly, solving a pole and a residue required by a rational approximation function by adopting a matrix pencil method, and substituting the pole and the residue into a piecewise linear recursive convolution technology to realize piecewise linear recursive convolution expression of voltage at a connection point of the transmission line and the frequency-variable load; and finally, realizing transient response analysis of the transmission line system with the frequency-variable load in combination with a transmission line equation. According to the method, the matrix pencil method (MPM) and the piecewise linear recursive convolution (PLRC) technology are combined, transient response analysis of the transmission line system of the terminating frequency-variable load is achieved, meanwhile, the method has the advantages of being rapid in calculation and high in applicability, and the method can be applied to analysis of the electromagnetic compatibility problem in the fields of electric power, control, communication and the like.

The invention discloses a method and a system for recognizing human behavior in video. An optical flow frame generation network and a long-term recursive convolutional neural network are cascaded to form a time-flow deep convolutional neural network, and a long-term recursive convolutional neural network constitutes a space. The streaming deep convolutional neural network uses a multi-dimensional weighted fusion model to fuse the prediction results of the dual-stream network to obtain the prediction of the human behavior type of the video data. The invention has high detection accuracy, wide application occasions and good generalization ability.

The invention relates to the field of artificial intelligence, and discloses a music data identification method and device based on multi-voice, equipment and a storage medium, which are used for improving the accuracy of identifying single-voice music data. The music data identification method comprises the steps of acquiring multi-voice music data, adopting a convolutional neural network to convert the multi-voice music data into a music sequence, wherein the music sequence comprises a pitch sequence and a rhythm sequence; inputting the music sequence into a multi-voice music identification model for recursive convolution to generate a plurality of sample music score sequence groups; according to the plurality of sample music score sequence groups, a pre-trained music language model and a conditional probability model, generating a plurality of matching probability groups, the plurality of sample music score sequence groups being in one-to-one correspondence with the plurality of matching probability groups; and determining a plurality of pieces of target single-voice music data in the plurality of sample music score sequence groups based on the plurality of matching probability groups. In addition, the invention also relates to a block chain technology, and the multi-voice music data can be stored in a block chain.