81results about How to "Guaranteed sparsity" patented technology

The invention discloses an image de-noising method based on a sparse self-adapted dictionary, and the method is mainly used for overcoming the defects that over-fitting exists when an existing method is used for training the dictionary, and the self-adaption is insufficient. The realization process comprises the following steps of: (1) obtaining image blocks from images with noises and paralleling the image blocks into vectors to form a training data set; (2) utilizing the training data set to iteratively train the dictionary; in an iteration process, taking the dictionary obtained by iteration as a basic dictionary of the iteration, and after the iteration is finished, obtaining a final dictionary and an encoding coefficient matrix of the training data set on the dictionary; (3) utilizing the dictionary and the encoding coefficient matrix obtained by training to obtain a de-noised data set; and (4) utilizing the de-noised data set to reconstruct a de-noised image. The dictionary trained by the method disclosed by the invention has sparseness and better self-adaptation; the effect of de-noising the image is improved; and the method can be used for de-noising a natural image and a medical CT (Computed Tomography) image.

The invention discloses a maneuvering target radial acceleration and speed estimation method based on an orthogonal match pursuit, belonging to the field of radar signal processing. According to the maneuvering target radial acceleration and speed estimation method, a radial acceleration and a radial speed of a high-speed and large-maneuvering target can be rapidly accurately determined, the tracking performance and the recognition capability of a radar to the high-speed and large-maneuvering target are effectively improved, the response time of a weapon system is increased, and the attacking capacity to the target is enhanced. The maneuvering target radial acceleration and speed estimation method provided by the invention comprises the following steps of: 1, sampling a signal received by a radar, establishing a complete atom library; 2, carrying out orthogonal match pursuit decomposition on the signal; and 3, searching a sparse solution energy diagram peak value, determining a radial acceleration and a radial speed of the signal according to peak value coordinates. The maneuvering target radial acceleration and speed estimation method has the advantages of high instantaneity, accuracy in radial acceleration and speed estimation, easy realization of engineering, stronger engineering application value and popularization prospect, and can be used in the fields of improving high-maneuvering target real-time accurate pursuit and radar target type recognition, and the like.

The invention relates to a traffic accident rate prediction system based on on-line variational Bayesian support vector regression, which comprises a data preprocessing module, an online variational Bayesian support vector regression model building module, an online variational bayesian support vector regression model training module, and an online variational bayesian support vector regression model prediction module. This method effectively solves the problem that the traditional support vector regression model predicts the speed of traffic accident rate is slow, the prediction result is inaccurate, and it is difficult to solve the problem on line and show its practical value.

The invention discloses a modal distance constraint-based multimodal fusion image classification method. The method includes the following steps that: first step, the rs-fMRI (resting-state functionalmagnetic resonance imaging) data and DTI (diffusion tensor imaging) data of a plurality of subjects are obtained; second step, a brain function network feature vector and a brain structure network feature vector are constructed for each subject; third step, feature filtering operation is performed on the feature vectors of two modalities based on the Kendall tau correlation coefficient and an overlap mode; fourth step, the relative distance constraint of the feature vectors of two modalities of the same subject before and after mapping is added on the original basis of the K-support norm, andthe objective function of a multimodal feature selection model is constructed, and the optimal feature vectors of two modalities are screened out; and fifth step, a classifier is trained on the basisof a multi-kernel support vector machine model, and the optimal feature vectors of two modalities of the subjects are inputted into the trained classifier, and the category labels of the subjects arepredicted. The classification accuracy of the method of the invention is high.

The invention discloses an online fault diagnosing method for Fast RVM (relevance vector machine) sewage treatment. The method includes the steps of firstly, removing samples with incomplete attributes in sewage data, normalizing the samples into a [0, 1] interval, and determining a historical data set and an updating test set; secondly, using a relevance vector machine method based on clustering to compress the majority data of the historical data set; thirdly, using a virtual minority upward sampling method to extend the minority data of the historical data set; fourth, building a 'one-to-one' fast relevance vector machine multi-classification training model; fifthly, adding new samples from the updating test set into the model for testing, and updating the historical data set; sixthly, returning to the second step, reprocessing unbalanced historical data, training the model, and repeating the above process until online data testing is finished. By the online fault diagnosing method, the unbalance of the sewage data is lowered effectively, classification accuracy is increased, online updating speed is increased, operation faults can be diagnosed in real time, and the safety operation of a sewage treatment plant is guaranteed.

The invention belongs to the field of a radar imaging technology, and discloses a compression perception multilayer ISAR imaging method based on BP optimization, for mainly solving the problems of noise sensitivity, high single-layer compression perception optimization time complexity and difficult sparse control parameters by use of conventional ISAR imaging. The specific realization process is as follows: preprocessing original echo data; determining an orientation resolution step factor sigma; initializing an encoding matrix M1 and a decoding matrix W2; performing training by use of a BP algorithm to optimize an encoding process and a decoding process; reconstructing an ISAR image by use of a compression perception theory; performing reverse Fourier transformation on orientation, and inputting obtained echoes into a next layer, and carrying out the training by use of the same principle; and outputting a high-resolution ISAR image at a final layer. According to the invention, the convergence speed is fast, high-resolution imaging can be realized under the conditions of a low signal-to-noise ratio and a small echo quantity, and the method can be applied to classification and identification of ISAR image objects.

The invention provides a homonymous point matching method of heterogenous remote sensing images taking high-order structural characteristics into consideration, and belongs to the field of plotting science and technology. The method comprises that data is prepared before heterogenous images are matched; a characteristic point cluster is divided and extracted; candidate matching points of characteristic points are determined; homonymous points are matched taking the high-order structural characteristics into consideration; a result is matched for each layer of pyramid images, and a local network adjustment based on an RFM model is carried out by utilizing a fusion selecting weight iteration method; and the matching result is refined layer by layer till an original image layer, and the homonymous points of the heterogenous remote sensing images can be matched automatically and reliably. According to the invention, a rational function model and a hypergraph image matching model are used in an integrated manner, geometrical constraint characteristics and the higher-order structural characteristics are introduced to layer-by-layer pyramid image matching, the sparsity of the hypergraph model is ensured, and information loss caused by hyperedge sampling is reduced to the minimum; and the matching reliability and success rate of the heterogenous remote sensing images are improved, and the workload of manual homonymous point measurement is reduced effectively.

The invention discloses a harmonic noise suppression method based on waveform morphology sparse modeling. The harmonic noise suppression method includes the steps of 1) constructing Chirplet transform according to waveform morphological characteristics of harmonic noise in seismic records acquired via vibroseis slip sweep, and constituting an over-complete dictionary with continuous wavelet transform; 2) quickly implementing Chirplet positive and inverse transforms; 3) determining Chirplet transform parameters based on time-frequency distribution characteristics of correlated data; and 4) according to the starting frequency of a reference scanning signal, determining the filter cutoff frequency of the harmonic noise to achieve fidelity separation of an effective signal from the harmonic noise. The invention solves the problem of harmonic noise interference in the seismic data acquired via vibroseis slip sweep, and achieves the purpose of improving the signal-to-noise ratio of the seismic data. The method of the invention determines the Chirplet transform according to the time-frequency distribution characteristics of the harmonic noise, and ensures the sparseness; the fast implementation of the Chirplet transform guarantees the efficiency of transform, and the correction coefficient guarantees the accuracy of inverse transform reconstruction; and the Chirplet transform parameters are automatically determined according to the data drive, a strong adaptability is gained and single-channel calculation facilitates parallel processing.

The invention discloses an electrocardiogram-signal identity recognition method based on NMF, an evaluation method and a device. The electrocardiogram-signal identity recognition method based on the NMF includes the steps that electrocardiogram signals are received, serve as sample points and are preprocessed; according to the electrocardiogram signal structure, a QRS wave section is subjected tobidirectional particle size scanning, the obtained QRS wave section, a wave section before QRS and a wave section after the QRS are integrated, and the preprocessed electrocardiogram signals are reconstructed; the reconstructed electrocardiogram signals are subjected to non-negative matrix factorization, and a coefficient matrix is obtained; a sub-component of the coefficient matrix is selected with the sub-component analysis method as a final characteristic value matrix of the electrocardiogram signals; electrocardiogram-signal identity recognition is carried out through the characteristic value matrix. Through the characteristic value matrix, a homologous similarity matrix and a heterogenous similarity matrix are calculated through mahalanobis distance, the error recognition rate, the false reject rate and the equal error rate are calculated through the homologous similarity matrix and the heterogenous similarity matrix, and the recognition effect on electrocardiogram-signal identityrecognition is subjected to measurement evaluation.

The invention discloses a signal decomposition-based compression perception processing method. The method comprises the following steps of acquiring a sparse signal; extracting signal points at equal intervals from the sparse signal to form R sub signals, wherein R is equal to N/L, L is set length of each sub signal, and N is the length of the sparse signal; performing compression perception processing on each sub signal by using an observation matrix to obtain the observation vector of each sub signal, and transmitting the observation vector to a receiving terminal. By applying the method provided by the invention, the signal with large data volume can be subject to effective compression perception processing and signal reconstruction, and the method is not constrained by signal data structure.

The invention discloses a super-resolution image reconstruction method, a super-resolution image reconstruction system, a super-resolution image reconstruction device and a storage medium, and the method comprises the steps of inputting an original low-resolution image, preprocessing the image, and carrying out the bicubic interpolation on the input low-resolution image; and then, passing the preprocessed image through a sparse deep convolutional neural network containing pooling manifold constraints, mutually connecting and arranging the image data, learning the missing high-frequency detail information components in the image, and realizing the reconstruction of the high-resolution clear image. The method can effectively reduce the network parameter calculation amount and keep the important detail information of the image at the same time, and by inputting the original low-resolution image into the sparse deep convolutional neural network model trained by a set algorithm, the high-resolution image is realized by the super-resolution reconstruction.

The invention belongs to the field of integrated circuits, and relates to a method and device for reducing an interconnection line model of a great quantity of ports. The method comprises the following steps of: constructing an undirected graph according to connection relationship of resistors and capacitors of an interconnection line circuit of a great quantity of ports, partitioning the undirected graph by utilizing a spectrum partitioning method, and finally carrying out coarse graining on nodes in a same partition set, thus obtaining a reduced circuit. The device comprises an input unit, an output unit, a program storage unit, an external bus, a memory, a storage management unit, an input and output bridging unit, a system bus and a processor; and an AMOR program of the reduction method can be realized through storage of the program storage unit. According to the invention, model reduction is carried out on the interconnection line of a great quantity of the ports without introduction of nonzero components, and the reduced model is ensured to be shorter in simulation time and higher in efficiency, and simultaneously the resistance value and capacitance of the obtained reduced circuit are positive values, thus having physical realizability and ensuring the passiveness of the reduced circuit.

The invention discloses a multi-gaussian kernel self-optimization relevance vector machine based wastewater quality soft-measurement method. The method comprises the following steps: (1) removing abnormal data from wastewater input and output data, performing normalization processing due to different dimensions of input variables and normalizing to be within an interval from 0 to 1; (2) forming a multi-gaussian kernel function relevance vector machine soft-measurement model module; (3) performing multi-gaussian kernel function nuclear parameter self-optimization algorithm; (4) forming an initial parameter optimization module by the genetic optimization algorithm; and (5) realizing modeling of a multi-gaussian kernel function relevance vector machine soft-measurement model. According to the method, nuclear parameters of all dimensions can be determined by a self-optimization method, the initial parameters are optimized by the genetic optimization algorithm, an optimal model is built, the output precision of BOD (Biochemical Oxygen Demand) in wastewater is effectively improved under conditions that the convergence and the sparseness of the model are ensured.

The invention discloses an image classification method based on multi-sample dictionary learning and local constraint coding, and belongs to the technical field of artificial intelligence and the field of image classification. The method comprises the following steps: firstly, generating a virtual training sample and an initialization dictionary for a training sample by utilizing a K-SVD algorithm; learning a dictionary and a coding coefficient of the training sample by using all the training samples and the initialized dictionary; and then learning the coding coefficient of the test sample byusing the learned dictionary, learning the linear classifier coefficient by using the coding coefficient of the training sample, and finally performing classification identification on the coding coefficient of the test sample by using the linear classifier coefficient and outputting a classification result. When the method is used for image classification and recognition, the learned dictionaryand coding coefficient matrix can have better representation capability, and the classification accuracy can be obviously improved.

The invention discloses a non-coded RNA and disease association prediction method based on sparse subspace learning and belongs to the field of system biology. The method comprises steps of 1, constructing an adjacency matrix associated with non-coded RNA-diseases, and then respectively calculating Gaussian spectrum kernel similarity of the non-coded RNA and Gaussian spectrum kernel similarity ofthe diseases; 2, calculating a graph theory characteristic matrix and a statistic characteristic matrix according to two similarity matrixes and an adjacent matrix, further constructing a target function and solving a mapping matrix G; and 3, solving non-coded RNA-disease association pair relationship score prediction matrixes, and performing sorting to give the final prediction result. The methodis advantaged in that a graph theory, a statistical method and a machine learning method are fused, the information of a negative sample in the non-coded RNA-disease associated data can be effectively utilized, the non-coded RNA with significant correlation to disease occurrence and development can be efficiently, accurately and quickly predicted, and problems of long time consumption and high cost of a biological experiment method are effectively solved.

The invention discloses a robustness-gradable sparse array frequency and DOA (Direction of Arrival) estimation method and device. The method includes the following steps: obtaining a parameter group of frequency, amplitude and phase after correction according to a relaxed co-prime sparse array and a Tsui spectrum corrector; constructing a first remainder array according to the corrected frequency,carrying out frequency reconstruction based on a robustness-oriented remainder system, and averaging the frequency to obtain the estimated value of frequency and the wavelength; constructing a mean vector according to the corrected phase information and amplitude information, obtaining the phase estimation of the signals on the array elements according to the relation between the phase and the mean vector, and obtaining two phase difference values through subtraction; and constructing a second remainder array according to the two phase difference values, carrying out DOA reconstruction basedon the robustness-oriented remainder system to obtain an intermediate parameter, and calculating the estimated value of DOA according to an estimation formula. The device includes a relaxed co-prime sparse array, an ADC sampler, a DSP and a display device. Frequency and DOA estimation is completed by using the relaxed co-prime sparse array.

The invention provides an unsupervised multi-view feature selection method based on graph learning and view weight learning. The method comprises the following steps: firstly, constructing an unsupervised multi-view feature selection model based on graph learning and view weight learning, and solving the model to obtain a feature selection matrix of each view in multi-view data; secondly, sorting2-norm values of matrix row vectors according to features, and selecting to obtain corresponding features according to needs; according to the method, the optimal similarity matrix shared by all viewsis obtained through adaptive learning to describe the manifold structure of the original data, adaptive weight distribution is carried out, complementary information existing among different views can be mined, more comprehensive and more accurate support information is provided for a feature selection process, and thus more valuable features are selected.

The invention provides an image enhancement method based on non-convex total variation regularization, which is used for solving the problem that the existing image enhancement method has little attention to noise, does not consider the sparsity of an image and cannot obtain a high-quality restoration result. According to the method, data fitting items based on exponential transformation and one regular item are adopted to describe noise prior information, and then the sparsity of piecewise constants is punished by utilizing the regular item of a non-convex total variation (simulation) norm. According to the image structure, the index p is selected from (0, 1); and finally, based on a Retinex theory, calculating an illumination part, and removing illumination to obtain a reflection part, i.e., essential characteristics of the image, so as to achieve the purpose of offset field correction. According to the method, the interference of noise on the image is considered, the sparsity of theimage is punished through the regularization term, and meanwhile, logarithm transformation is replaced by exponential transformation, so that the contrast ratio is improved, and the model has higherrobustness.

The invention discloses a cold start project recommendation method based on embedded feature selection. The method comprises the steps of obtaining a user set, a project set and a feature set; generating a user project interaction matrix and a project feature matrix; establishing an optimization model of user project prediction based on the embedded feature selection; carrying out parameter optimization on the optimization model by utilizing a supervised learning method; substituting the optimized parameters into the prediction model, and calculating a prediction rating score of the project; and recommending the project according to the predicted rating score. According to the method provided by the invention, the individual difference of each user in feature selection is considered, and the existing of common features is considered, the individuation of recommended features is ensured, at the same time, the dimension reduction is realized. In addition, the model is optimized through the regularization of the elastic network and the high-efficiency algorithm based on the soft threshold, so that the method provided by the invention has the stronger superiority compared with other recommendation algorithms.

The invention discloses a voice guide walking stick and a deep neutral network optimizing method based on the walking stick, and belongs to the technical field of intelligent guide. The walking stickcomprises a stick body, a handle arranged at the top end of the stick body, and a wheel arranged at the bottom end of the stick body. The guide walking stick further comprises a micro-processor ARM embedded into the stick body, a deep neutral network module, a wide-angle camera, an array radar, a positioning module, a motor and a power module, the deep neutral network module, the wide-angle camera, the array radar, the positioning module, the motor and the power module are connected with the micro-processor ARM, and the motor is connected with the wheel. The voice guide walking stick is basedon a neutral network and the array radar, the image recognition function of the deep neutral network and the obstacle recognition function of the array radar are combined in the walking stick, multiple obstacle avoiding plans can be provided, meanwhile, the voice recognition function of the deep neutral network is utilized by the walking stick, the requirements provided by blind users are recognized and fed back, interaction between the walking stick and the users can be conveniently achieved, and the voice guide walking stick is convenient to use, convenient to operate, safe and reliable.

The invention provides a weighted broadband direction-of-arrival estimation method based on group sparsity. The method comprises the following steps: collecting receiving data of each channel; constructing an observation model under group sparsity; processing the obtained group sparse model by using an L1SVD method; introducing a weighting operation, and constructing a weighting coefficient through a spectrum peak value of the MUSIC algorithm; and obtaining a final broadband DOA estimation result. firstly a decorrelation operation is used, then a regularization parameter lambda is introduced to obtain a weighted objective function, then the weighted objective function is converted into a second-order cone optimization problem, and is solved to obtain a sparse signal S so as to obtain a DOA estimation result. According to the weighted broadband direction-of-arrival estimation method based on group sparsity, the group sparsity characteristic is fully utilized, the number of samples is effectively reduced, and the compression performance and the real-time performance are improved. According to the method, weighting operation is introduced, the sparsity of understanding is ensured, and the resolution and the estimation precision can be effectively improved.

The invention relates to the field of complex product design and manufacture, and particularly to a method of constructing a response surface model on the basis of a genetic algorithm and a system applying the same. The method comprises the following steps: a, carrying out sampling, wherein desired parameters or standards are input, and sampling data are obtained by a method of Latin hypercube sampling; b, constructing a basis function dictionary, wherein the mixed dictionary is constructed; c, seeking an algorithm of sparse representation, wherein after the mixed dictionary in the step b is constructed, coefficients corresponding to all basis functions in the mixed dictionary are obtained by solving by the simulated genetic algorithm according to X and Y in the sampling data in the step a; d, establishing the model; e, obtaining an original simulation model value yt corresponding to an independent variable xt in complex product design; and f, outputting machining parameters to use the same on production and manufacture of a complex mechanical and electrical product. According to the method, simulated genetic thinking is utilized to seek expression, which is low in sparseness and is high enough in precision at the same time, on the mixed dictionary, and thus the more precise and concise response surface model is constructed.

The invention discloses a human body behavior recognition method based on sparse tensors. The human body behavior recognition method comprises the following steps: obtaining a human body behavior silhouette sequence construction sample, and representing each sample with a three-order tensor; using sparse tensor local Fisher discriminant analysis to train a sparse projection matrix group; projecting the training sample and the to-be-tested sample to a tensor subspace by using the obtained sparse projection matrix group; and calculating the similarity between the to-be-tested sample and the training sample, and constructing a nearest neighbor classifier based on the tensor distance to identify the to-be-tested sample in the low-dimensional space. According to the tensor local Fisher discriminant analysis algorithm of the invention, the problems of eigenvalue solving and eigenvector solving are converted into a series of linear regression problems, so that the objective of tensor local Fisher discriminant analysis can be satisfied, and the sparsity of an obtained projection matrix can be ensured.

The invention discloses an equipment defect time prediction method based on capacitive equipment defect data, and relates to the technical field of electrical equipment and information. Feature engineering processing and defect time modeling are carried out by adopting a whole set of method; for data with more abnormal values, missing values and redundant values of a power grid company, firstly missing value filling and data cleaning are performed, new features are constructed through feature decomposition, and feature dimensionality reduction and denoising are performed by using an auto-encoder; and for the processed feature data, feature model construction is performed by using various machine learning methods such as a gradient boosting tree and a deep learning method. The method has the advantages of being easy to implement, high in calculation speed, high in prediction precision, good in prediction robustness and systematized in prediction process.