76 results about "Data resampling" patented technology

The voice print system of the present invention is a subword-based, text-dependent automatic speaker verification system that embodies the capability of user-selectable passwords with no constraints on the choice of vocabulary words or the language. Automatic blind speech segmentation allows speech to be segmented into subword units without any linguistic knowledge of the password. Subword modeling is performed using a multiple classifiers. The system also takes advantage of such concepts as multiple classifier fusion and data resampling to successfully boost the performance. Key word / key phrase spotting is used to optimally locate the password phrase. Numerous adaptation techniques increase the flexibility of the base system, and include: channel adaptation, fusion adaptation, model adaptation and threshold adaptation.

The voice print system of the present invention is a subword-based, text-dependent automatic speaker verification system that embodies the capability of user-selectable passwords with no constraints on the choice of vocabulary words or the language. Automatic blind speech segmentation allows speech to be segmented into subword units without any linguistic knowledge of the password. Subword modeling is performed using a multiple classifiers. The system also takes advantage of such concepts as multiple classifier fusion and data resampling to successfully boost the performance. Key word / key phrase spotting is used to optimally locate the password phrase. Numerous adaptation techniques increase the flexibility of the base system, and include: channel adaptation, fusion adaptation, model adaptation and threshold adaptation.

A method of resampling medical imaging data from a first spatial distribution of data points onto a second spatial distribution of data points, including determining a matrix of reverse interpolation coefficients for resampling data from said second spatial distribution onto said first spatial distribution, inverting a matrix based on said reverse interpolation matrix to determine forward resampling coefficients for resampling data from said first spatial distribution to said second spatial distribution, and resampling data from said first spatial distribution onto said second spatial distribution using said forward resampling coefficients.

A Multi-Rate Analog-to-Digital Converter (19) is coupled to a single crystal oscillator (17) as a reference clock and has at least two separate channels arranged to sample and convert input data at two differing clock rates. Each channel derives a clock signal from the reference clock. Associated with each of the channels is a Sigma-Delta converter (10a, 10b) comprising a modulator (12), a filter (14) and a resampler (18). The modulator (12) receives input data and provides a data signal to the filter (14), which itself provides a filtered data signal to the associated data resampler. The data resampler resamples the data and provides a digital output signal. As there is sampling in the digital domain the advantages associated with signal processing, speed and low noise injection are obtained. Similarly as the output of the modulator (12) is in digital form, it can be manipulated and processed readily and with the existing software.

The invention provides a method and a system for estimating ground PM2.5 based on a space-time regression Kriging model. The method comprises the steps of re-sampling ground PM2.5 observation data of a to-be-estimated region to a created mesh, and performing matching, wherein the matching process comprises the steps of averaging the ground PM2.5 observation data monitored in the same day by all PM2.5 stations in a mesh unit corresponding to the to-be-estimated region in the created mesh, and then assigning the averaged data to the corresponding mesh unit; calculating an experimental variance function of a residual error according to the ground PM2.5 observation data of the matched to-be-estimated region, and determining a space-time variance function model according to the experimental variance function of the residual error; performing fitting on the space-time variance function model by adopting a least square method; and estimating a ground PM2.5 concentration value of the to-be-estimated region by adopting the space-time regression Kriging model according to a fitting result of the space-time variance function model. Through the method and the system, the PM2.5 estimation precision can be improved.

A hyperspectral classification method based on the double-branch network provided by the invention firstly ensures that the number of different samples of input data is not constant and equal in eachiteration in the training process, and also ensures that each sample participating in the training is balanced in statistics through the method of data resampling. This not only effectively alleviatesthe sample imbalance problem in network learning, but also keeps the diversity of data. In order to extract the multi-scale features of the data, the invention uses the network structure of two branches to carry out semi-supervised learning through three training strategies, so that not only the training set is expanded, but also the classification accuracy is greatly improved; compared with other classification methods, the method greatly improves the classification precision through the ensemble learning strategy. The hyperspectral classification method based on the dual-branch network notonly is superior to other methods in performance, but also is superior to other methods in training efficiency.

The present invention provides a methodology, apparatus and system for resampling digital data utilizing a Fourier series based interpolation engine 104. A quick means to up-sample or down-sample data is provided without requiring computationally intensive processing. This is accomplished by utilizing low order coefficients of terms of a complete Fourier series expansion for a continuous signal. The summation of the expansion is limited to input samples immediately adjacent in time to the desired output. Generally speaking, the output is normally required to be a constant sampling rate, therefore, the input and output rates are related by an integer ratio. This ratio can be greater or smaller than one, providing up-sampling or down-sampling as appropriate. By employing the present invention, a DSP engine can be constructed that is adjustable to any ratio of sampling rates in a computationally efficient manner with low RMS error while preserving convolution through the resampling process.

The invention provides a low-complexity timing recovery method in a TDS-OFDM system, belonging to the technical field of digital information processing. The method comprises the steps as follows: intermediate frequency signals are received and sampled by fixed frequency; quasi-baseband complex data is gained by carrying out down transformation on the intermediate frequency sampled signals; the quasi-baseband complex data is re-sampled by an interpolation filter on the basis of decimal delay and sampled enabling signals returned by a loop numerically controlled oscillator; PN correlation is carried out on the re-sampled data to gain the correlation value of the local sequence and the receiving sequence. A simplification error detector extracts timing error information based on the gained correlation value; the timing error information, with noise components filtrated by a loop lowpass filter, is corrected by negative timing error and then input into the numerical control oscillator, and further the decimal delay and enabling signal required by the interpolation filter and a sampler are generated, thus forming a closed loop. The method has the advantages of low complexity, easy realization of hardware circuit and strong compatibility.

The invention discloses an existing railway turnout junction center extraction method based on a three-dimensional laser scanning technology. The method comprises the steps of conducting observation station splicing and gross error elimination on original data; resampling the data; filtering the original point cloud data based on the rail surface elevation; performing Hough transformation on the filtered point cloud data, and detecting a straight line; fitting the point cloud data belonging to the same straight line by using a least square method, and calculating a straight line slope k1 and an intercept b1; re-screening the point sets belonging to the same straight line; carrying out second-time least square fitting, accurately calculating the slope k2 and the intercept b2 of the straightlines and finally extracting four straight lines, wherein two straight lines belonging to the same group of rails are parallel, and forming two groups of parallel straight lines; calculating intersection points of the two unparallel straight lines wherein the coordinate average value of the four intersection points is the fork center coordinate. The beneficial effects of the invention are that the method can improve the measurement precision of an existing railway turnout, and avoids the huge potential safety hazards of the online operation of a measurer.

A Multi-Rate Analog-to-Digital Converter (19) is coupled to a single crystal oscillator (17) as a reference clock and has at least two separate channels arranged to sample and convert input data at two differing clock rates. Each channel derives a clock signal from the reference clock. Associated with each of the channels is a Sigma-Delta converter (10a, 10b) comprising a modulator (12), a filter (14) and a resampler (18). The modulator (12) receives input data and provides a data signal to the filter (14), which itself provides a filtered data signal to the associated data resampler. The data resampler resamples the data and provides a digital output signal. As there is sampling in the digital domain the advantages associated with signal processing, speed and low noise injection are obtained. Similarly as the output of the modulator (12) is in digital form, it can be manipulated and processed readily and with existing software.

The present invention provides a methodology, apparatus and system for resampling digital data utilizing a Fourier series based interpolation engine 104. A quick means to up-sample or down-sample data is provided without requiring computationally intensive processing. This is accomplished by utilizing low order coefficients of terms of a complete Fourier series expansion for a continuous signal. The summation of the expansion is limited to input samples immediately adjacent in time to the desired output. Generally speaking, the output is normally required to be a constant sampling rate, therefore, the input and output rates are related by an integer ratio. This ratio can be greater or smaller than one, providing up-sampling or down-sampling as appropriate. By employing the present invention, a DSP engine can be constructed that is adjustable to any ratio of sampling rates in a computationally efficient manner with low RMS error while preserving convolution through the resampling process.

The invention discloses a well logging during drilling data real-time processing method. According to the method, real-time data singular point detection and elimination are performed through combination of a thin layer threshold value method and a peak peak / valley valley ratio method, thereby eliminating 'burr' phenomenon caused by high frequency oscillatory interference of well logging during drilling, and improving signal to noise ratio of a real-time well logging during drilling curve; and meshing analysis is performed on well logging during drilling data acquired in real time, meshing of the well logging during drilling real-time data and equal interval processing of sampling intervals are realized through establishment of a well logging during drilling real-time data resampling model, and related comparison demands of different well logging during drilling curves, the well logging during drilling curve and a well logging during drilling curve of internal storage, a real-time well logging during drilling curve and a cable well logging during drilling curve are met.

The invention relates to a data resampling method based on repeated editing nearest neighbor and clustering oversampling. The method comprises the steps: calculating the Euclidean distance between each to-be-sampled book and a nearby sample, selecting the sample with the smallest distance as the nearby sample of the to-be-sampled book, comparing whether the labels of the sample and the nearby sample are the same or not, and deleting the sample if the labels of the sample and the nearby sample are different; dividing the remaining samples into k clusters by using K-means, and filtering out theclusters of which the ratio of the number of majority class samples to the number of minority class samples is less than an imbalance rate threshold c; calculating an Euclidean distance between minority class samples in each cluster, constructing a distance matrix of the cluster, summing all off-diagonal elements in the matrix, and dividing the sum by the number of the off-diagonal elements to obtain an average distance of the cluster; calculating a sparse factor of each cluster; and calculating a resampling weight value of each cluster, and determining the number of generated new samples according to the weight values by using an SMOTE method. According to the method, the problem of class imbalance in the data is solved, so that the classifier can obtain a better classification effect.

The invention discloses a frequency domain optical field digital refocusing algorithm capable of suppressing resampling error, and belongs to the field of optical field image calculation, which solvesthe problem that the existing frequency domain digital refocusing technology causes resampling error in frequency domain data resampling and generates artifacts in the reconstructed image. The specific steps are as follows: At first, that 4D light field is transform into spatial coordinates according to the refocusing depth, Then the center slice in the horizontal direction is taken from the Fourier spectrum, and the refocusing image is obtained by 2D inverse Fourier transform of the central slice. The algorithm avoids the error of discrete data resampling caused by the frequency coordinate transform effectively through the spatial coordinate transformation of the original 4D light field. Compared with the traditional light field digital refocusing method based on the Fourier projection slice theorem, the algorithm of the invention requires fewer operation steps and has lower computational complexity. At the same time, it can achieve the same refocusing effect as the most robust spatial projection integral digital refocusing algorithm.

The invention discloses a normal sampling resampling method based on image reconstruction three-dimensional data, which belongs to the field of three-dimensional data processing and digital image processing, and solves the problem of data resampling in an optical three-dimensional measurement process. The sampling process comprises the following steps of: presetting sampling levels and thresholds, wherein different levels indicate different density; establishing a corresponding relation table between the digital image pixels and three-dimensional data points; adaptively establishing a Hash table, and calculating a normal for each Hash table element, substituting the normal of each data point in the Hash table with corresponding normal of the Hash table, orderly accumulating points through the normal in the region, and summing; judging the sampling level according to the magnitude of the sum value, establishing a corresponding sampling lookup table according to the corresponding relation table between the digital image pixels and the three-dimensional data points, and correspondingly determining the values of the lookup table according to the preset sampling levels; and finally, sampling the three-dimensional data according to the established sampling lookup table. The method has the advantages of being high in efficiency, describing regional characteristics by the least data points without losing the characteristics, accurately dividing sample regions, flexibly setting the sampling levels, being easy to realize and the like.

The invention relates to a pyramid slicing method based on dynamic effect model (DEM) data. The pyramid slicing method based on the DEM data comprises the following steps: the first step is determining size of slicing and number of layers of the slicing, the second step is reading DEM data by using an open-source image reading device, the third step is judging whether existing size and number of layers of the slicing are suitable or not and determining the optimal number of layers L and combination number of initial pixel N , the fourth step is determining start-stop positions of a rank and a low of the DEM data needing to be read according to the existing number of layers, the fifth step is reading image pixel values in a partitioning mode according to the start-stop positions of the rank and the low, and carrying out data resampling of the data read, the sixth step is dividing the data according to the size of the slicing, outputting the slicing data, and naming each slicing datum in a standardization mode, and the seventh step is judging whether all layers are divided or not: if all layers are divided, the method is end; and if all layers are not divided, combination number of the initial pixel is modified and returning to the fourth step is needed. Compared with the prior art, the pyramid slicing method based on the DEM data is high in slicing efficiency, good in expandability and the like.

The invention discloses a finite-element quick image reestablishment system and method for an optoacoustic mammary gland imager. The system comprises a data collection module, a data resampling moduleand a quantitative image reestablishment module. The data collection module is used for conducting data collection on the mammary gland of the human body through the optoacoustic mammary gland imager; the data resampling module is used for conducting quick imaging on data collected by the data collection module, observing whether or not a suspected focus exists in an illumination area according to an imaged graph and intercepting data with the suspected focus to obtain valid data of a laser radiation area; the quantitative image reestablishment module is used for conducting quantitative imaging on output data of the data resampling module through a finite element algorithm and reestablishing hemoglobin concentration and oxygen saturation images of the suspected focus and the periphery ofthe suspected focus. According to the system, when the focus is analyzed, the images are cut through the quick image reestablishment system, the illumination area with the diameter of 5 cm is cut, sothat the grid diameter becomes extremely small when the finite-element algorithm is conducted, the imaging precision is guaranteed, and the imaging speed is also guaranteed.

During data resampling, bad samples are ignored or replaced with some combination of the good sample values in the neighborhood being processed. The sample replacement can be performed using a number of approaches, including serial and parallel implementations, such as branch-based implementations, matrix-based implementations, and function table-based implementations, and can use a number of modes, such as nearest neighbor, bilinear and cubic convolution.

The invention discloses a genetic attribute reduction-based palaeontological lineage evolution analysis method. The method comprises the basic steps of establishing a lineage species tree; constructing a concept sample template set of nodes in the lineage species tree by utilizing a genetic attribute reduction algorithm, and by taking the concept sample template set as a decision basis, establishing a concept decision species tree; by utilizing matching degrees between attributes of species and concept sample templates of the species tree, determining the positions of the species containing missing data in the lineage species tree; and through a bootstrap method, performing data re-sampling to obtain likelihood values of the positions of grafted species in the lineage tree, thereby finishing species grafting, and constructing a palaeontological lineage tree. Compared with a method for constructing the lineage tree by a maximum parsimony method and attribute reduction, the problems of concept sample template failure and difficulty in stable construction of the palaeontological lineage tree due to a large amount of the missing data are effectively solved; and the accuracy and stability of species lineage analysis are improved.

The invention provides a data resampling method based on clustering oversampling and an instance hardness threshold. The method comprises the following steps: firstly, performing clustering processingon a data set by utilizing a Kmeans method, and performing filtering processing and sampling weight distribution on clustering; then, adopting an SMOTE algorithm to carry out oversampling on the dataset to generate new data, so that the number of minority class samples in the data set is equal to that of majority class samples, and the data set becomes class balance; and finally, cleaning the data by adopting an instance hardness threshold algorithm to obtain a final balanced data set with fewer noisy points. According to the method, the class imbalance data set can be processed into the balance data set, and the prediction performance of the classifier for minority class samples is improved.

The invention relates to a T type wiring optical fiber longitudinal differential protection data synchronization method, including a local side data resampling module, an opposite side data sending module, an opposite side data receiving module and an opposite side data resampling module. Devices at three sides in T connection have no master-slave difference, nonsynchronous sampling at equal intervals is performed on the three sides, a frequency tracking function is performed separately at the three sides, and after frequency tracking resampling, local side data are sent to the two opposite sides immediately; at each side, buffering area data are read and received during interruption, channel delay is calculated according to the time scale in a data frame, and opposite side data time scales are mapped to a time shaft of a local side device. Then the corresponding historical interpolation moments of local side data is searched according to data sampling time scales of the two opposite sides, resampling is performed on the received data of the two sides, and the resampled data are filled in resampling serial number buffering areas corresponding to the historical moments. For a singledevice, data of the three sides have synchronous resampling serial numbers, and current differential data are calculated according to the sampling serial number of each side to be used for protectioncalculation.

The invention discloses a high-precision geometric correction method for a spatial image. The method comprises steps of: 1) matching the sampling time of data collected by a IMU / DGPS system with the sampling time of spatial image data, so that the sampling time of the data collected by the IMU / DGPS system is enabled to be matched with the sampling time of the spatial image data; 2) obtaining the element of exterior orientation of a camera through coordinate system conversion; 3) finding the pixel array number matching the ground coordinates of the spatial image data, i.e., the optimal scan array number according to the corresponding relationship between the spatial image data and the corresponding actual ground coordinates; 4) re-sampling the spatial image data according to the optimal scan array number obtained in the step 3) to obtain the corrected spatial image data. The method utilizes the data collected by the IMU / DGPS system to perform geometric correction on the spatial image, and can achieve high-precision image geometric correction for the ground motion measurement platform or air platform.

The invention relates to the technical field of electrocardiogram detection, and particularly discloses an electrocardiographic data resampling method, and an electrocardiogram displaying method and device. The electrocardiographic data resampling method obtains initialization information, the target wave velocity, the displayed point distance and the sampling rate in an automatic / semi-automatic mode, an automatic resampling generation algorithm is adopted to dynamically generate different resampling modes, the electrocardiographic data resampling method can be applied to the displaying device with different parameters, and electrocardiographic waves can displayed on different displaying devices on the premise that the normative relevant provisions such as EC13, YY1079-2008 and JJG1041-2008 are met.

The invention discloses a highway real-time traffic accident risk assessment method based on deep learning, and the method comprises the steps: firstly dividing a highway into road segments through employing the basic information of an ETC portal, highway intercommunication and a toll station, and building the upstream and downstream association relation between the road segments; calculating thetraffic flow, the traffic flow speed and the traffic flow density of each road section respectively, obtaining road information, meteorological information and accident information, converting the road information, the meteorological information and the accident information into one-hot codes, and then performing data fusion, data resampling and standardization on four types of information corresponding to upstream and downstream road sections of an accident occurrence point; distinguishing time sequence features and non-time sequence features according to the acquired data, and constructing and training a deep learning model; and finally, according to the trained deep learning model, carrying out real-time evaluation on the risk level of the traffic accident of each road section of the expressway, and calculating to obtain an accident risk level index. According to the invention, the highway traffic accident risk level can be evaluated timely and accurately.

The invention relates to the technical field of data automatic processing, and particularly relates to an HSD data fire point real-time monitoring and automatic analysis system based on Himawari-8. The HSD data fire point real-time monitoring and automatic analysis system based on Himawari-8 can achieve real-time acquisition, batch correction processing and abnormal fire point discrimination of remote sensing satellite data and comprises the following steps: 1) acquiring a Himawari-8 data set; 2) reading and splicing the data sets according to wave band numbers; 3) resampling the resolution data of different wave bands; 4) calculating and processing the data in the step 3): in a stage of processing and calculating the data, processing technical points include original data radiometric calibration, ephemeris table grid calculation, geometric correction based on GCP control points, linear offset fine adjustment of a corrected reference ephemeris table, cloud amount detection removal andbrightness temperature inversion; and 5) waiting for next task initiation. The HSD data fire point real-time monitoring and automatic analysis system based on the Himawari-8 has the advantages that data processing is fast, correction precision is high, cloud layer interference is effectively reduced, and fire point detection accuracy is improved.

The invention discloses a method and a system for estimating PM2.5 based on an empirical Bayesian skill model, and a storage medium. The method comprises the following steps: resampling ground PM2.5 observation data of a to-be-estimated region into a pre-created grid, and matching the resampled ground PM2.5 observation data; Dividing the matched grid data into a plurality of overlapped subsets with specific sizes; Estimating a semi-variation function through the PM2.5 observation data in the subset; Taking the estimated semi-variation function as a model, and performing unconditional simulation at each input position through the model to generate a new PM2.5 simulation value; Estimating a new semi-variation function through the PM2.5 analog value, and calculating the weight of the semi-variation function according to an empirical Bayesian rule; And repeating the fourth step and the fifth step for several times, weighting the semi-variation function by the weight obtained at the last time, and carrying out PM2.5 prediction on a weighting result at an unknown position. According to the method, the cost can be saved, and the estimation precision of PM2.5 is greatly improved.

The invention relates to an Adaboost ensemble learning power grid fault diagnosis system and method based on data resampling. The system comprises a fault information database, a data preprocessing module and a fault diagnosis module; the data preprocessing module calls data of the fault information database to perform data preprocessing, and the data preprocessing module sends the preprocessed data to the module fault diagnosis module. The fault information database is used for storing fault information packets; the data preprocessing module is used for carrying out vacancy value filling andnormalization operation on the data; the fault diagnosis module comprises a data resampling module used for carrying out balance operation on data, a decision tree base classifier module used for carrying out modeling training and fault prediction on fault data, and an Adaboost integrated classifier module used for carrying out multiple rounds of learning and fault prediction on a decision tree base classifier. According to the invention, the integrity of the fault data is ensured, and the diagnosis of the fault data is more accurate.

The invention relates to a training data resampling method and device, a storage medium and an electronic device. The method comprises: obtaining first original data in a first time period; calculating a first proportion respectively occupied by a plurality of preset classifications in the first original data; according to the size relation of the first proportion, sorting the multiple preset classifications according to a preset rule to obtain a first sorting result; determining a sampling ratio corresponding to each preset classification according to the ranking of each preset classificationand a preset corresponding relationship, the preset corresponding relationship being a corresponding relationship between the ranking and the sampling ratio; and resampling the training data for modeling according to the sampling proportions respectively corresponding to the plurality of preset classifications. Thus, the problem that the classification model is unfriendly to small classes is solved, the classification accuracy of the classification model obtained through training of the training data for different applications is improved, and therefore the user experience is improved.

The invention provides a satellite borne SAR satellite load reduction data rate processing method. The method comprises the steps that 1 SAR echo data are input; 2 Doppler center estimation is carriedout; 3 the downsampling multiple is determined; 4 a downsampling interpolation kernel function is built; 5 on-board data re-sampling is carried out; and 6 signal re-sampling is carried out. Accordingto the invention, the problem of large data rate, which is caused by high azimuth oversampling rate faced by the SAR system design under the requirements of azimuth ambiguity and zebra map design, issolved to a certain extent; the earth observation performance of the SAR system is effectively ensured; and the signal to noise ratio is ensured.

The invention belongs to the technical field of non-destructive detection and discloses a high-precision self-matching boundary dimension measurement method applied to a high-speed condition. The method specifically includes the following steps that: 1, the data of a pipe to be detected are acquired, so that calibration can be performed; 2, periodic analysis is performed on the collected calibration signal of the pipe to be detected; 3, in-phase data statistical analysis is performed on the periodic data of data acquired by two probes; 4, signal acquisition is performed on the pipe to be detected; and 5, calculation is performed, so that the boundary dimensions of the pipe to be detected can be obtained. The step 2 further includes the following steps that: 2.1, signal preprocessing is performed so as to filter out abnormal signals; 2.2 the flag bits of the periods of the acquired data are obtained through calculation; and 2.3, each of the periodic data of the data acquired by the twoprobes is re-sampled. With the high-precision self-matching boundary dimension measurement method applied to the high-speed condition of the invention adopted, in-phase angle comparison is adopted, sothat the influence of mechanical installation errors and acquisition system errors can be eliminated to the greatest extent.