199results about How to "Avoid heavy computation" patented technology

The invention provides a combined wind power prediction method suitable for a distributed wind power plant. The method comprises the following steps: step 1, acquiring data and pre-processing; step 2, utilizing a training sample set and a prediction sample set which are normalized to build a wind speed prediction model based on a radial basis function neural network and predict the wind speed and variation trend of distribution fans at the next moment; step 3, building a distributed wind power plant area CFD (computational fluid dynamics) model and externally deducing the prediction wind speed of each fan in the plant area according to factors such as the terrain, coarseness and wake current influence of a distributed wind field; step 4, acquiring the power data of an SCADA (supervisory control and data acquisition) system fan of the distributed wind field; and step 5, adopting correlation coefficients. The invention firstly provides a double-layer combined neural network to respectively predict the wind speed and power. Models are respectively built through adopting appropriate efficient neural network types, and improved particle swarm optimization with ideas of 'improvement', 'variation' and 'elimination' is additionally added to optimize the neural network, so that the speed and precision of modeling can be effectively improved, and the decoupling between wind speed and power is realized.

An SYNTAX automatic scoring method based on coronary angiogram image segmentation segments and extracts a coronary arterial tree and central line based on coronary angiogram images of a patient and employs a blood vessel side branch separation, reconnection and K main curve method, and generates a binary image of the coronary arterial tree The blood vessel segment numbers are automatically matched based on the SYNTAX scoring standard by employing a central line direction tracking method, and coronary artery pathological changes and adverse characteristics are identified and scored by selecting a coronary left (right) advantageous type in combination with the width attenuation degree of the cross section of the blood vessel and the curvature change of the center line, and the scores are calculated to give a reasonable score result to allow a doctor to employ optimal operation scheme. The method overcomes the problems of great computation amount, tedious calculating steps, complex scoring standard and long calculating time of artificial calculating SYNTAX scoring in conventional clinics.

The invention discloses a voice recognition method facing a specific crowd. The method comprises the following steps of: first, sampling a voice signal and converting the voice signal to a digital signal from an analogue signal; then, pre-weighting, windowing, en-framing and performing front-end processing of endpoint detection on the digital voice signal; later on, performing feature extraction on the voice signal by adopting discrete wavelet transform; and finally, performing voice recognition on the feature-extracted voice signal by adopting a discrete hidden Markov model after training a sample. In the processes of performing the front-end processing and the feature extraction on the voice signal, spectrum features and pronunciation characteristics of different target crowds are fully taken into consideration and the process of extracting voice information is optimized, so that a processing process and an information extracting process can be simplified; and therefore, recognition precision is ensured, simultaneously calculation amount and information storage capacity in the recognition process are greatly reduced, and the voice recognition on an embedded platform is realized.

The invention provides a navigation route finding method, a robot and a system. The method comprises the steps that a collecting device collects the environmental information of the robot during a training movement process and generates training data according to the environmental information during the training movement process; the training data are used to train a depth neural network through a depth-enhanced learning method, so that the trained depth neural network is adapted to carry out decision analysis based on route finding data; and the robot is instructed to carry out route finding toward a navigation target, wherein the route finding data are acquired according to data collected by the collecting device during a route finding application process and the navigation object. According to the navigation route finding method, the robot and the system, the navigation route finding is accurate.

The invention provides a pollution source prediction method, a pollution source prediction device and electronic equipment, and relates to the technical field of pollution prediction, which comprisesthe following steps of: obtaining mesoscale meteorological prediction data and emission list data of a to-be-predicted destination; Wherein emissions and emission sources corresponding to the emissions are recorded in the emission list data; according to the Industry type, combining the emission sources to obtain reference emission data, inputting the mesoscale meteorological prediction data and the reference emission data into an air quality numerical model to obtain air pollution prediction data, and calculating the air pollution refined prediction contribution rate according to emission list data and the air pollution prediction data. The calculation amount of the air quality numerical model can be reduced, and the calculation speed is increased.

The invention discloses a layered and distributed network voltage regulator control system based on an active mechanism. The layered and distributed network voltage regulator control system based on the active mechanism comprises a reactive voltage controller which faces a single branch, a coordination voltage controller which faces a local area and an initiative power distribution network voltage management subsystem which faces a whole power distribution network. The initiative power distribution network voltage control subsystem determines a voltage control indicator according to a distribution condition of power distribution network active power. When a voltage is beyond a certain limit, the reactive voltage controller locally regulates the voltage which is beyond the certain limit. When the regulation fails, the coordination voltage controller is regulated on the basis of overall considering voltage fluctuations and a batch-type energy source grid connecting and grid separating plan, and voltage beyond limit recover based on the active mechanism is achieved through the coordination of the reactive voltage controller and the coordination voltage controller. Meanwhile, the invention further provides a regulating method which corresponds to the layered and distributed network voltage regulator control system based on the active mechanism. The layered and distributed network voltage regulator control system based on the active mechanism and the regulating method which corresponds to the layered and distributed network voltage regulator control system based on the active mechanism achieve active control of the beyond limit voltage of the power distribution network on the premise of meeting power distribution network constraint conditions.

The invention relates to an echo elimination method and system based on a microphone array; the method uses wave beam formation, linearity echo elimination and non-linear echo elimination triple structures to inhibit echo energy in pickup signals, wherein a wave beam forming module can stably inhibit direct sound wave signals in a horn direction without being affected by environment changes; a linearity echo elimination module can inhibit residual echo after wave beam formation, thus eliminating echo reflection portion; a non-linearity echo elimination module uses a reference signal and residual echo signal long time amplitude correlation degree to carry out gain inhibition, thus further inhibiting echo energy. The method and system can greatly improve the echo inhabitation effect while ensuring the pickup signal to be clear and nature, thus having stronger performance and better stability when compared with the prior art.

The invention discloses a multicenter data task processing method, a multicenter data task processing device and a multicenter data task processing system. The processing method comprises the steps: A, collecting data task information; B, generating a data task model graph according to the data task information, cutting the data task model graph, namely grouping all nodes in the data task model graph, wherein each node of the data task model graph is one data task; and C, allocating data tasks of different groups into corresponding data centers. By means of the multicenter data task processing method, the device and the system, through the establishment of a model reflecting the correlations among data tasks, multicenter task allocation is performed based on the model, and thus the task allocation is visual, accurate and balanced. Transfer and redundancy of large amount of data are avoided, data volume transmitted among each center is small, therefore pressure of network load is reduced, overall data processing efficiency is improved, and loads in all the data centers are basically the same.

The invention discloses a nonlinear production plan optimization method applied to oil refineries, which includes the steps: firstly, setting up an oil refining production logic network; secondly, setting up process modules of an atmospheric and vacuum CDU (crude distillation unit) and a FCC (fluid catalytic cracking) unit; thirdly, inputting all constraint condition data and determining a target function in two-stage search; and fourthly, using the mixed searching method to optimally determine production plan decision variables. The genetic algorithm and direct search algorithm are combined on the basis of the simulation modeling technique, global profits of an oil refinery production plan can be effectively increased without destroying the constraint conditions, tremendous computation generated during solving of complex production plan problems can be avoided, and optimized solutions can be quickly obtained. The nonlinear production plan optimization method is simple in principle, convenient in implementation, high in transportability and applicable to different production plan environments.

The invention discloses a walking posture control method of a humanoid biped robot based on a genetic algorithm. A mathematical model of a biped robot is established based on the ZMP control theory and a pose matrix, and the basic gait of a human is simulated. V, height H, arm swing R, body center up and down swing C, body center left and right swing A These six parameters plan the robot's gait, and select optimized T, V, H, R through genetic algorithm The six parameters of , C, and A are brought into the inverse operation of the pose matrix to obtain the values ​​of other joint coordinates of the robot, and solve the method of generating data that stably controls the gait of the robot. It overcomes the problem of large amount of calculation and low efficiency in inverse operation of pose matrix, improves the real-time response ability of the system, and uses the data generated by this method in the humanoid biped robot system to achieve a good control effect.

The invention provides a non-linear model prediction control method for a double planetary gear row type hybrid electric vehicle. According to the method, based on the predicted rotation speed of the finished vehicle and the torque requirement, optimization solution is conducted on an objective function during the prediction time interval through a non-linear optimization algorithm, the optimal control sequence of controlled quantities is obtained, and the required torques of a power system engine, a motor, a power generator and a brake system are determined by combining the first controlled quantity of the control sequence with kinetic equations of the double planetary gear row type hybrid electric vehicle in various modes. According to the characteristic that the number of working modes of the double planetary gear row type hybrid electric vehicle is large, control prediction and optimization are achieved through a non-linear model, the combination and disconnection states of all clutches and braking force in a power coupling mechanism can be effectively controlled, optimization of the different working modes and optimal energy distribution among different power components are achieved, and the advantage that the number of the working modes of the double planetary gear row type hybrid electric vehicle is large is brought into full play.

The invention discloses a behavior modeling method of a power amplifier based on a clock recurrent neural network. The problems that an ordinary neural network model is iterated many times, and the long-term memory effect performance is poor are solved. By means of the characteristics that output of a recurrent neural network is not only related to real time input but also related to historical input, the memory effect of the power amplifier is described. On the basis, the weight of an ordinary recurrent neural network hidden layer is divided into multiple modules, each module has its own cycle, the weight of the module is only updated in its own cycle, and the number of weight updates is reduced to accelerate training of the neural network model. Accordingly, the non-linear features and memory effect of the power amplifier can be well described, and the high precision is achieved.

The invention discloses an image identifying method comprising the following steps: scanning an image through scanning windows of different sizes to obtain a plurality of window images; extracting the HOG-LBP (Histograms of Oriented Gradients-Local Binary Pattern) feature vector of each window image; sequentially identifying the HOG-LBP feature vectors through a preset HOG-LBP classifier to obtain a plurality of window images preliminarily judged to be targets; extracting image features of the window images preliminarily judged to be the targets through a preset PHOW (Pyramid Histogram Of Words) dictionary; identifying the image features of the window images through a preset PHOW classifier to obtain a target image window image; clustering to obtain a target image. The invention further provides a image identifying system corresponding to the image identifying method. The window images are filtered by the two classifiers in a cascade way, so that the problem of large calculation amount of simultaneously scanning and calculating various window features on full image and multiscale is solved; the PHOW classifier is independent of the HOG-LBP classifier, so that the misjudgment rate is reduced.

An image processing method for boundary resolution enhancement is disclosed. Firstly, an image is transferred into an image layer. Noise of the image layer is removed by a bilateral filter and crisp edges are retained at the same time. Moreover, the image layer is interpolated by an interpolation filter for resolution enhancement. The image processing method of the present invention can lower the image blur degree substantially, enhance the image resolution and be widely implemented in all sorts of image / video processing hardware devices.

The invention discloses an infrared target tracking method based on the background perception of an activation region, which mainly solves the problems that the instantaneity is poor, a template is easily influenced by interference and a tracking window is short of effective self-adaption adjustment strategies in the existing tracking technique. The infrared target tracking method disclosed by the invention comprises the following realization steps: firstly, marking out a candidate target selection and background perception effective region by establishing the activation region under a particle filtering framework and extracting the apparent characteristics by utilizing a covariance operator; then, distinguishing the current scene state of the target through the background perception of the activation region and extracting the position and dimension observation sets of the target; and finally, formulating a template renewing strategy according to the scene state, avoiding leading the interference into the template and leading and combining the target position observation and apparent characteristics to sieve the candidate targets, thereby improving the tracking accuracy and realizing the self-adaption adjustment of an observation window by simultaneously utilizing the dimension observation of the target. The infrared target tracking method disclosed by the invention has stronginterference resistance and realizes the rapid self-adaption accurate tracking on an infrared target in a strong noise wave interference environment.

The invention provides a sampling inertial guidance-based visual IMU direction estimation method, namely, a sampling process of a matching point pair and a removing process of mismatching points in visual direction estimation are guided by utilizing direction estimation information of an IMU (Inertial Measurement Unit). The method comprises three steps of gain adaptive complementary filter-based IMU direction estimation, scale rotational invariance-based feature detection and visual IMU fusion-based direction estimation. According to the method, a gain adaptive complementary filter is adopted and a remarkable mismatching point pair can be removed in an initial iterative process, so that the accuracy of direction estimation is improved; pose estimation information of the IMU serves as an initial value to be introduced in the visual direction estimation, and the mismatching point pair is iteratively removed, so that the direction estimation process is accelerated and the problem of large calculation amount caused by adoption of a random initial value is effectively avoided; and the method is wide in applicability, good in robustness and high in accuracy rate, and can be widely applied to an action capture process of human body rehabilitation training.

The invention relates to a multi-target direct positioning method based on an uncorrected array and a neural network. The method comprises: placing single signal sources at different discrete coordinate points (known positions), establishing a sample library of uncorrected array manifold responses; enabling the uncorrected array to acquire target signal source data and to estimate the array manifold matrix thereof; using the sample library to automatically pair column vectors in the array manifold matrix, and classifying the array manifold vectors corresponding to the same target into the samegroup and merging the same into a high-dimensional data vector, determining the approximate area of each target; training a radial basis neural network by using the data sample corresponding to the approximate area of each target position, and using the high-dimensional data vector corresponding to each target as the input of the neural network, and using the output of the nerve network as the estimated position for the target. The method can avoid the huge computational amount caused by the correction of the antenna array and the grid search, has a good practical application value, and stable, reliable and high performance.

The invention relates to a video data frame rate up conversion method based on a system analysis and design (SAD) and video on demand (VOD) matching rules. The method comprises the following steps of: when two adjacent reference frames in video data are subjected to interpolation processing, performing matching block search processing on the two reference frames by a bidirectional motion estimation method according to an SAD matching rule to determine a standby matching block pair, and judging whether each block has brightness sensitivity; performing matching block search processing again on the blocks with the brightness sensitivity by the bidirectional motion estimation method based on a VOD matching rule to determine the corresponding final matching block pair; for the blocks without brightness sensitivity, directly taking the standby matching block pair as the final matching block pair; and determining a motion vector according to the final matching block pair corresponding to each block, and performing motion compensation. By utilizing the method, the block matching error, which is caused by illumination change, of the adjacent frames in a video image is reduced; the motion vector can be accurately estimated; and therefore, frames to be interpolated with relatively high quality can be obtained.

The invention discloses a bistatic synthetic aperture radar imaging method based on similar single static equivalence. A double radical sign slant distance of bistatic is equivalent to a single radical sign slant distance of a similar single static, an analytical expression of a two dimensional spectrum of a system is obtained, a solving question of a double radical sign of the bistatic is simplified, mature single static imaging methods are fully used, such as an omega-K (wk) algorithm and a range-Doppler algorithm (RD algorithm), a two dimensional STOLT relation of the system is obtained according to the expression of the two dimensional spectrum, and finally, a two dimensional STOLT interpolation is approximated according to the two dimensional non-uniform Fourier transformation, and imaging of the system is achieved. Therefore, complicated degree of an imaging algorithm of the bistatic is largely reduced, and successive imaging process is facilitated.

The present invention provides a high-performance vector font library display method based on a VXWorks system. The method comprises the following steps: 1) initializing FreeType; 2) performing texture initialization; 3) reading a vector font library file stored in a file system into a memory; 4) determining whether character cache data exists, and if yes, displaying the character cache data as characters via texture mapping, and if no, parsing font vector information into lattice data according to the vector font library file, and displaying the well-parsed lattice data as characters via texture mapping; and 5) determining whether a display process ends, if yes, releasing a memory texture space, and if no, continuing executing the step until the display ends. Based on an OpenGL graphic library based on VXWorks and in combination with TrueType, the high-performance vector font library display method based on a VXWorks system is high in drawing efficiency, good in display effect, capable of supporting zooming, rotating and translation and good in anti-aliasing effects.

The invention discloses a method for realizing voice reduction through rapid voice waveform repairing. The method comprises the steps of filter processing, smooth processing, audio energy analysis, tread processing and outputting. By the method, an audio curve can be effectively reduced through the smooth processing, and only burrs and jitter in the audio curve can be removed, so that a large amount of computation brought by Fourier transform is avoided, and the audio curve is reduced by repairing discontinuity and compensating the jitter.

The invention provides a three-dimensional reconstruction system and reconstruction method based on PAL cameras. The reconstruction system comprises two industrial cameras respectively provided with a PAL lens, the two PAL lenses are arranged in a perpendicular manner, and the two industrial cameras are connected with a computer through USB data wires. The three-dimensional reconstruction method comprises the following steps of: arranging the two industrial cameras respectively provided with one PAL lens in the perpendicular manner and connecting the two industrial cameras with the computer; b, using the industrial cameras to collect image data, and transmitting the image data to the computer; c, carrying out processing and calculation on the image data, and obtaining optical characteristic relations of the cameras; d, correcting image distortion based on the optical characteristic relations of the cameras; e, utilizing a dynamic programming algorithm to match two images obtained by shooting; f, converting points of a panoramic image in a cylindrical coordinate system into points in a three-dimensional coordinate system; and g, drawing a three-dimensional scene to be reconstructed. The three-dimensional reconstruction system and reconstruction method have the advantages that the structure is simple, the operation speed is high, and the system and the method are applicable to the reconstruction of a large-range scene.

The invention discloses a low peak-to-average ratio multi-antenna pilot optimization and low-complexity transmission method, which comprises the following steps of: firstly, searching for an optimal constant modulus zero auto-correlation sequence according to a pilot length, the size of discrete Fourier transform and a sub-carrier mapping format; secondly, performing the discrete Fourier transform, sub-carrier mapping, inverse discrete Fourier transform and cyclic prefix addition on the sequence, and storing the processed sequence into a read-only memory of a field programmable array; thirdly, multiplying signals of different transmission antennae by using different phases according to the sub-carrier mapping format; and finally, judging whether a current signal block transmits a pilot signal or not according to the number of a current radio access unit and the number of a signal block in a timeslot. By the method, the requirements of a multiple-input and multiple-output wireless channel measurement system on the low peak-to-average ratio of the transmitted signals and low implementation complexity are met.

The embodiment of the invention provides a server recommending method and device, and relates to the technical field of the Internet. The method comprises the steps of receiving a server recommending request, wherein the server recommending request includes a user IP address; searching a digital IP address matched with the user IP address and a corresponding server identifier in a preset digital IP address database by a dichotomy, wherein the preset digital IP address database comprises the digital IP addresses which are arranged according to a number sequence, and the corresponding server identifiers; and sending the server identifier corresponding to the digital IP address matched with the user IP address. According to the server recommending method and device provided by the embodiment of the invention, the digital IP address matched with the user IP address and the corresponding server identifier are searched in the preset IP address database by the dichotomy, so that a user selects the server corresponding to the server identifier, and the problem that the calculation amount is huge due to traversal of the IP database is avoided.

The invention provides the elastic wave forward modeling of simulated micro earthquake. The method comprises the steps that the rock property is analyzed according to the actual data of geological survey; the mode of rock fracture is determined; a moment tensor matrix expressing a micro earthquake source is generated; the geological velocity field model of an explored area is established; under a fluctuation theory, an elastic wave fluctuation equation which is suitable for describing the propagation of a micro earthquake wave field is established; under the fluctuation theory, a micro earthquake elastic wave fluctuation equation is solved to realize micro earthquake wave field propagation simulation; simulation is completed; and multi-component recording of micro earthquake ground and in a well and longitudinal and transverse wave energy radiation patterns are output. According to the elastic wave forward modeling of simulated micro earthquake, a simulation result which is more accurate than a ray theory can be quickly acquired; the amount of information is abundant; and better technical means and tools are provided for subsequent characteristic law analysis, location method research and earthquake source parameter inversion.

The invention provides an image rendering model training method and device and an image rendering method and device, and the model training method comprises the steps: inputting a multi-angle target scene graph into a volume rendering model, and obtaining a volume rendering image outputted by the volume rendering model; based on the volume rendering image and the multi-angle target scene graph, training the initialized neural radiation field with implicit scene expression ability to obtain an image rendering model, wherein the volume rendering model is obtained based on multi-angle sample scene graph training; enabling the volume rendering model to firstly carry out projection reconstruction on the multi-angle target scene graph to obtain an explicit density distribution matrix used for representing 3D scene density of a target scene, and after sampling points in the projection direction are determined based on the density distribution matrix, generating a volume rendering image based on voxel features including density and color values in the sampling points. According to the method and device, the sampling points can be quickly and directly determined based on the explicit density distribution matrix, and the training and reasoning efficiency of the image rendering model is improved.

The invention discloses an ellipsoidal fruit dimension rapid detection method based on characteristic vector orientation. A fruit edge image is obtained through performing such operation as threshold segmentation, filtering, edge extraction and the like on an obtained fruit image. A rectangular coordinate system is established for the edge image; a covariance matrix of edge coordinates is solved; a characteristic value of the covariance matrix and a unit characteristic vector are solved accordingly; and by use of the unit characteristic vector, a fruit is oriented to enable the longitudinal or transverse diameter direction of the fruit to be parallel to the horizontal axis of the rectangular coordinate system, and then dimension detection is finished by calculating upper, lower, left and right extreme points of the boundary of the fruit. According to the invention, the longitudinal diameter and the transverse diameter of the fruit are rapidly oriented through solving the characteristic vector for boundary coordinate information of the fruit image, enormous operation brought by rotating the fruit image multiple times by use of an MER method is avoided, and the detection speed is improved while the detection precision is guaranteed. Therefore, the method provided by the invention is applied to the real-time detection need of the dimension of the fruit in an ellipsoidal fruit commercial processing process.

The invention relates to the technical field of artificial intelligence, in particular to an image retrieval method and device, equipment and a storage medium. The image retrieval method comprises the steps of obtaining a to-be-recognized image; inputting the to-be-recognized image into a pre-trained image clustering model, and predicting a target class cluster identifier corresponding to the to-be-recognized image, wherein the image clustering model divides a pre-created image library into a plurality of pre-divided class clusters; obtaining a pre-divided class cluster corresponding to the target class cluster identifier from the plurality of pre-divided class clusters as a target retrieval class cluster, wherein the target retrieval cluster comprises at least one to-be-matched image; and performing image retrieval on the to-be-recognized image based on the target retrieval cluster. According to the method, the clustering precision can be effectively ensured, and the problems of large calculation amount and time consumption caused by traditional image retrieval can be effectively avoided.

The invention discloses a graph-based convolutional network training method, device and system. The method comprises the following steps: establishing a first storage space for each layer, except thelowest layer and the highest layer, of a convolution model; based on the training data and the graph of each batch, determining each center node of the training data and each neighbor node of the center node; for each center node, obtaining a representation vector of each neighbor node from a first storage space corresponding to each neighbor node identifier of the previous layer of center node; according to the characterization vector of the central node transmitted from the previous layer and the obtained characterization vector of each neighbor node, determining a representation vector of the central node in the current layer, and when the current layer is not the lowest layer or the highest layer, transmitting the determined representation vector to the next layer adjacent to the current layer and updating the representation vector corresponding to the central node identifier in the first storage space of the current layer until the representation vector of the center node at the highest layer is obtained. According to the invention, the training calculation amount and training time can be effectively reduced.