132 results about "Sparse regularization" patented technology

A system and method for specificity-based multimodality three-dimensional optical tomography imaging comprises steps of: optical imaging to obtain a light intensity of body surface optical signal of an imaging target; CT imaging to obtain structure volume data; establishing an equation representing a linear relationship between the distribution of the obtained light intensity of body surface optical signal of the imaging target, the obtained CT discrete mesh data and the distribution of unknown internal self-luminescence light sources; establishing a dynamic sparse regularization target function in every iteration for the equation; and reconstructing a tomography image. The present invention well considers the optical specificity of tissue, in which there is a non-uniform optical characteristic parameter distribution within the same tissue when finite element modeling is used, which is closer to the real situation, so that an accurate imaging effect is achieved.

The invention discloses a method for sparse synthetic aperture radar (SAR) imaging on basis of echo simulation operators and relates to the technology of the synthetic aperture radars. By utilizing SAR echo characters and sparsity of an observed scene, a sparse regularization model based on an SAR echo simulation operator is established. By coalesce of a threshold value iterative algorithm of the echo simulation operator, radar imaging on a target scene in the observed area is realized. Compared with an existing sparse SAR image method based on two-dimensional observation models, the method can improve operating efficiency and lowering operating cost, and compared with a matched filtering imaging method, under a controlled condition, the method is capable of realizing imaging on the condition of being lower than nyquist rate sample, restraining sidelobe and obtaining clearer SAR images.

The invention discloses a polyethylene glycol terephthalate (PET) reconstruction method based on a sparsification and Poisson model. The PET reconstruction method includes: firstly, acquiring projection data, determining image size range and pixel range, and calculating a system probability matrix; obtaining an initial image through a filtered back projection (FBP) traditional algorithm, and using an obtained log-likelihood function as a reconstruction recovery item; using a wavelet transformation and discrete cosine transformation (DCT) mixed base and weighting as sparse regularization constraint, decomposing an objective function by using a split Bregman method to obtain two sub-problems, using a first sub-problem as a sparse regularization problem under a Gaussian model, and using linear Bregman interation for solving; using a second sub-problem as a Poisson denoising problem, and using a close operator method for solving; operating the last one iterative formula, completing an integrated iteration, obtaining a reconstructed image, and serving as an initial value of the next iteration.

The invention relates to a sparse-regularization-based face recognition method capable of realizing multiband face image information fusion and belongs to the technical field of image processing. The method comprises the following steps of: 1, acquiring visible-light, near-infrared, intermediate-infrared, far-infrared and thermal-infrared training images; 2, performing normalization, background removal and illumination preprocessing; 3, extracting visible-light, near-infrared, intermediate-infrared, far-infrared and thermal-infrared sample facial features; 4, selecting the sample facial features based on a sparse regularization method, evaluating coefficient and endowing weight under the classification significance according to the sample facial features, and fusing the sample facial features to obtain feature vectors representing the samples to serve as index vectors; 5, forming a feature set according to the index vector; 6, segmenting a facial part from an image to be tested, performing operation of the step 2 to step 4 on the facial part to be tested to obtain a feature vector of the facial part to be tested; and 7, sequentially calculating the distance between the feature vector of the facial part to be tested and the feature vectors of the feature set, and selecting the sample which corresponds to the feature set with the minimal distance and is the person of the image to be tested. The method has the advantages that change of accessories, shelter and the like is overcome; the recognition precision is high; and the application range is wide.

The invention relates to an image denoising method based on sparse regularization. The image denoising method based on sparse regularization fully utilizes the gradient information of an image and non local self-similarity to construct a sparse regularization denoising model, and utilizes an iteration histogram standardization algorithm to solve the model. The solving process includes: dividing the image into image blocks, and then clustering the image blocks according to the level of structural similarity, and at last using a sparse K-SVD dictionary to train each given image block. Therefore, the structural property of the dictionary is improved and also the over-complete dictionary obtained through training can preferably perform sparse representation of the image blocks. The image denoising method based on sparse regularization has the advantages of effectively denoising the image, being high in the reservation capability for the image texture structure, obtaining a better image visual effect, further reducing the computation complexity, and improving the operation speed.

A method and an image-reconstruction apparatus are disclosed for reconstructing image data on the basis of measurement data from an imaging system. In at least one embodiment of the process, initial image data, initially reconstructed from the measurement data, is optimized in an iterative optimization method utilizing a substantially edge-maintaining noise regularization term and an additional sparsity regularization term.

Interference in an audio signal is reduced by estimating a target signal using beam-forming in a direction of the signal source. A set of estimates of interference is determined by using a microphone array filtering matrix to block the target signal in the audio signal. A set of filters is optimized by minimizing an objective function measuring a mismatch between the set of estimates of interference and the estimate of the target signal. The minimizing uses a sparse regularization of coefficients of the set of filters. The set of estimates of interference are filtered using the set of filters after the optimizing. Then, the estimate of interference after the optimizing is subtracted from the target signal.

The invention provides a satellite hyper-spectral image compressed sensing reconstruction method based on image sparse regularization. The method comprises the following steps: Step 1, the three-dimensional cube of known hyper-spectral data is rearranged into a matrix; Step 2, a multi-vector measurement model is constructed with a stochastic convolution transform as a linear observation matrix, and each waveband is independently sampled to generate a measurement vector matrix; Step 3, a hyper-spectral image is decomposed in a sparse transform domain into a spectral association component and a difference component, and an image sparse regularization joint reconstruction model including the association component and the difference component is constructed; and Step 4, an alternating-direction multiplier iteration algorithm for solving the joint reconstruction model is put forward, the association component and the difference component of a transform domain are obtained, and then the association component and the difference component are merged to obtain reestablished hyper-spectral data. The method provided by the invention is high in degree of compression and high in precision during satellite hyper-spectral remote sensing data compression.

The invention discloses a mutual conversion method of visible light and near-infrared human face images. The method comprises the steps of converting a near-infrared human face image or a visible light human face photo into an initial visible light human face photo or an initial near-infrared human face image by use of a method based on sparse learning, and converting the initial visible light human face photo or the initial near-infrared human face image into a high-definition detail photo of the visible light human face photo or the near-infrared human face image by use of a method based on multi-characteristic selection. According to the mutual conversion method, heterogeneous human face images are fitted in a stratified manner by use of a method based on sparse regularization and the visible light human face photo is generated from the near-infrared human face image, and therefore, the detail information of the synthetic photo is increased and the problem of heterogeneous human face recognition is solved.

The invention discloses an optical three-dimensional imaging method based on biological tissue specificity. The problem of the prior art that the accurate and quick optical three-dimensional imaging cannot be performed on a complex organism having an irregular anatomical structure and various scattering characteristic tissues is solved. According to the method, a target function is established onthe basis of an optical transmission mixed mathematical model for the biological tissue specificity and a completely thinning regularizing method, and a mixed optimizing method based on a task orientation is used for solving, thereby realizing the optical three-dimensional imaging for an internal targeted object. The method comprises the following steps: collecting a data; preprocessing the data;acquiring the anatomical structure; acquiring a surface optical data; reestablishing optical characteristic parameters; establishing an optical transmission model; establishing a system equation; establishing the target function; solving the target function; and displaying a result. The optical three-dimensional imaging method has the advantage that the accurate, quick and high-resolution reconstruction for a light source in the complex organism can be realized, and the method can be applied to the field of internal optical three-dimensional imaging.

The invention provides a method and system for face identification. The method comprises the following steps: collecting multi-band training images of visible, near-infrared, intermediate-infrared, far-infrared, and thermal-infrared lights; carrying out evaluation and selection on extracted face image characteristics of all bands based on a sparse regularization method so as to obtain new characteristics of the face images of all the bands and corresponding new characteristic evaluation indexes after dimensionality reduction; carrying out evaluation, selection and fusion on a first characteristic set based on the parse regularization method so as to construct a second characteristic set finally expressing a face and a corresponding second characteristic evaluation index; obtaining a second characteristic set for finally expressing a face of a to-be-tested person from a third characteristic set according to a second characteristic evaluation index; and employing a nearest neighbor classifier to obtaining a classification result. According to the invention, on the basis of fusion of enough face image information, a characteristic set for finally expressing a face is ensured to have the low dimensionality, thereby ensuring the speed of face identification and the low data volume that an individual needs to store and thus improving identification precision.

The invention discloses a multiband signal reconstruction method based on a clustering sparse regularization orthogonal matching tracking algorithm, relates to the technical field of information and communication, and aims at solving a problem of restoring an original multiband signal from an unknown-sparseness multi-observation-value vector in a Xampling frame after the conversion through a continuous-finite module and sampling through a modulation bandwidth converter. Because many simulation signals met a multiband signal model in a process of signal processing, the method plays an important role in enabling the compressed sensing theory to be used for simulation signals. The basic idea of the method is to convert an infinite observation value vector problem into a single-observation-value vector problem. The method comprises the following steps: carrying out the column vectorization of observation values; carrying out the extension of an observation matrix through a Kronecker product; estimating a support set of the original signal through employing the above two results and signal sparseness estimation; finally reconstructing the signal, wherein the regularization idea is used in an estimation process of the support set.

The invention relates to a data processing method and device for a customer service staff scoring system. The method includes the following steps that: a plurality of key indexes are automatically identified and analyzed from a large number of preset evaluation indexes based on a sparse regularization model; the weights of the plurality of key indexes relative to customer service scoring can be determined through adopting an analytic hierarchy process; and the comprehensive score of customer service staff is calculated according to the weights of the plurality of key indexes and the actual index values of the key indexes. Compared with the prior art, and according the method of the invention, the coefficient regularization method and the analytic hierarchy process are combined together, and therefore, scoring on the customer service staff can be realized automatically, and the workload of scoring personnel can be decreased, and work efficiency can be improved.

The invention discloses a cone-beam X-ray luminescence computed tomography method based on a simplified ball harmonic wave-diffusion approximation model and solves the problem of fast and accurate imaging for achieving fluorescence image reconstruction based on a diffusion approximation equation or by aid of the traditional X-ray computed tomography technology in the prior art. The cone-beam X-ray luminescence computed tomography method is characterized in that an objective function is constructed on the basis of the simplified ball harmonic wave-diffusion approximation model and a sparse regularization method, and the objective function is solved in a half greedy algorithm to achieve fast and accurate reconstruction of in-vivo fluorescence nano-particle probes. The cone-beam X-ray luminescence computed tomography method specifically includes steps: data collection and preprocessing, living body anatomical structure reconstruction, body surface fluorescence data acquisition, forward light transmission model construction, sparse regularization objective function construction, objective function optimization solution and three-dimensional reconstruction result display. The cone-beam X-ray luminescence computed tomography method has the advantages of being capable of achieving accurate, fast and high-resolution position and concentration distribution reconstruction of the in-vivo fluorescence nano-particle probes, and being capable of being applied to the field of X-ray fluorescence computed tomography.

The invention discloses a weighted graph regularization sparse brain network construction method. According to the method, on the basis of analysis of keeping similar local manifold characteristics ofsimilar data in original data space after projection, constrained modeling is carried out on association between brain connection by using a graph regularization item; with consideration of correlation analysis, the similarity degree of essences in a time sequence signal sequence of a brain area can be measured, sparse modeling is constrained by using measured functional connection strength priorinformation, and a weighted sparse regularization item is established; and then combined modeling is carried out on the whole brain function network and a brain network with the biological significance is constructed. According to the invention, on the basis of integration of correlation analysis, sparsity and graph regularization constraint into the unified modeling frame, a brain function network is constructed effectively by utilizing the similarity and locality of functional magnetic resonance brain image data fully; and the method is also used for neurological disease identification andbiological marker analysis.

The invention discloses a structural damage identification method which combines multi-objective ant lion optimization and sparse regularization. The method comprises the following steps: a finite element model is established according to structural design parameters, modal parameters such as natural frequency and vibration mode of the structure are extracted; according to the principle of model updating, the objective function is established by using the relative frequency variation of the damaged structure and the calculated structure as well as the modal confidence criterion; the objectivefunction is optimized by using weighting strategy and trace sparse regularization; multi-objective ant lion optimization algorithm is used to optimize the objective function until the iterative termination condition is reached; finally, the optimal solution is the result of damage identification. At the same time, the invention optimize a plurality of objective functions, more accurately search for an optimal solution, and introduces a trace sparse regularization and weighting strategy to respectively improve noise robustness and identification accuracy, reduce the influence of noise and damage sensitivity of measurement response on identification result, and has better noise robustness and higher identification accuracy.

Targets are detected in a scene behind a wall by first transmitting a pulse through the wall. Then, a primary impulse response is detected by a sparse regularized least squares inversion applied to received signals corresponding to the reflected pulse. A delay operator that matches the primary impulse response to similar impulse responses in the received signals is also determined. A distortion of the pulse after the pulse passes through the wall but before the pulse is reflected by the target can also be determined. The distortion is used in an iterative process to refine the detection of the target and to suppress ghosting artifacts.

The invention provides a digital blurred image blind restoration method based on gradient screening, and belongs to the technical field of image and video processing. The problems that the estimation of blurring kernels of compound movement is not accurate in an existing method, and a priori movement mode is needed are solved, and meanwhile the problem that the restoration effect on images with more details is poor by a normalization sparse regularization blind restoration method is solved. According to the technical scheme, an input image is subjected to bilateral filtering and impact filtering, gradient screening is conducted to get rid of small-gradient amplitude detail information, normalization sparse regularization blind restoration is conducted, and at last a clear image is output. The digital blurred image blind restoration method based on gradient screening can be applied to the fields of civilian photographing and camera shooting, intelligent video monitoring, intelligent security and protection, optical imaging and remote sensing, military imaging reconnaissance, guided missile imaging and guidance and the like.

The invention relates to a method and a system for inverting the underground magnetic susceptibility by adopting full-tension magnetic gradient data. The method comprises the following steps: acquiring the full-tension magnetic gradient data; building a sparse regularization model according to preset prior constraint conditions and the full-tension magnetic gradient data; performing iterative solution on the sparse regularization model based on an improved conjugate gradient algorithm, and performing inversion to obtain the magnetic susceptibility. The invention further relates to a system. The system comprises a data acquisition module, a model building module and a magnetic susceptibility inversion module. According to the method disclosed by the invention, by the setting of some prior constraint conditions, the sparse regularization model is built, so that bounded constraining of physical parameters can be considered, and the accuracy of inverting the magnetic susceptibility is greatly improved.

The invention discloses a hyperspectral image compressed sensing method based on the heavy weighting laplacian sparse prior. The hyperspectral image compressed sensing method based on the heavy weighting laplacian sparse prior is used for solving the technical problem that an existing hyperspectral image compressed sensing method is low in reconstruction accuracy. According to the technical scheme, a little linear observation of each pixel spectrum is collected randomly as compressed data, a compressed sensing model based on the heavy weighting laplacian sparse prior and a sparse regulated regression model is established, and solving on the established models is conducted. According to the hyperspectral image compressed sensing method based on the heavy weighting laplacian sparse prior, and a little linear observation of each pixel spectrum is collected randomly as compressed data, so that resource consumption in the image collecting process is reduced; the strong sparsity of the hyperspectral image is depicted accurately through the heavy weighting laplacian sparse prior, inhomogeneous constraint on the nonzero element of the traditional laplacian sparse prior is overcome, and the reconstruction accuracy of the hyperspectral image is improved. It is tested that when a sampling rate is 0.15 and the compressed data consist strong noise with 10 db signal-to-noise ratio, and the peak signal-to-noise ratio promotes over 4 db relative to a background technology method.

The invention discloses an optical three-dimensional imaging method based on a voxel and an adaptive optical transmission model. With the method, a problem that an optical three-dimensional imaging reconstruction result can not be obtained until tedious organ segmentation and grid discretization are carried out in the prior art can be solved. According to the invention, a voxel physical model and an adaptive optical transmission mathematic model are combined organically; and a complete sparse regularization objective function is constructed and solved to realize optical three-dimensional imaging of a targeting object in vivo. The method comprises steps of data acquisition and pretreatment, voxel-based physical model construction, adaptive optical transmission mathematic model construction, system equation establishment based on enriched function fusion, objective function establishment based on a sparse regularization strategy and a prior feasible region, and objective function and display result solution. Therefore, the provided method has characteristics of high efficiency and easy usage; and an inaccurate three-dimensional imaging result due to an error caused by uncontrollable factors in organ segmentation and grid discretization can be avoided. The method can be applied to the in-vivo optical three-dimensional imaging field.

The invention discloses a bridge moving load identification method based on cascaded dictionaries and sparse regularization. The method mainly includes the following steps: 1) arranging a speed measuring device on a bridge to be used for measuring vehicle speed, and arranging a plurality of strain and acceleration sensors on the bridge for actual measurement of bridge response; 2) building a bridge finite element model; 3) adopting cascaded dictionaries formed by a discrete trigonometric function and a discrete rectangular function to expand an unknown moving load; 4) normalizing actual measurement response, and introducing a weighted L1 norm regularization method to build an identification equation; and 5) utilizing a fast iterative shrinkage-thresholding algorithm (FISTA) to solve the equation, and calculating a corresponding time interval signal. The method provided by the invention can accurately identify a bridge moving load, including a slowly-changing integral signal and local impact signal in bridge moving vehicle load signals, has strong noise robustness, and is particularly suitable for field utilization of bridge response to estimate gross vehicle weight.

The invention discloses a method for improving a target real-time identification network structure and suitable for embedded equipment, and relates to the field of computer vision, and the method specifically comprises the steps: firstly, operating a normal target identification network model, and calculating the proportion of weight parameter quantities of convolution modules of all levels; selecting a convolution module with a proportion of 50%-70%, replacing the convolution module with a depth separable convolution module, and completing primary compression; secondly, introducing the influence factor of each channel into a loss function, and carrying out sparse regularization; setting a cutting proportion threshold, selecting an influence factor lower than the proportion threshold, andcutting off the corresponding network channel to complete a channel pruning and compressing process; then, using an image enhancement module to adjust the image, and inputting the image into the neural network added into the SENet module for training; and finally, performing target identification on the image acquired in real time. According to the method, a structure compression method and a channel pruning method are combined together, so that the original network is compressed to a great extent.

The invention discloses a seismic inversion method based on hybrid high order and fractional order Anisotropic Total p-variation (ATpV) sparse regularization, belongs to the field of geophysical inversion and oil and gas reservoir forecasting, and solves the problems caused by existing calculation of a total variation by adopting a first-order difference. The seismic inversion method comprises thefollowing steps: preprocessing a seismic record, extracting horizon position, and then acquiring an initial model of a parameter to be inversed according to well-logging data after the horizon position is extracted; calculating an initial wave impedance logarithm based on the initial model of the parameter to be inversed and the preprocessed seismic record, and then building an inversion model based on the hybrid high order and fractional order ATpV sparse regularization; updating the initial wave impedance logarithm based on the seismic record and a seismic wavelet by combining an alternating direction multiplier method and a forward model to obtain an updated wave impedance logarithm and a parameter for updating a next wave impedance logarithm; and determining whether a not updated value and an updated value meet conditions, if YES, updating the next wave impedance logarithm, and if NO, acquiring the parameter to be inversed. The method is used for seismic inversion.

The invention relates to the technical field of deep learning, particularly to a model compression method and system based on a sparse convolutional neural network, and related equipment, and aims toreduce resource occupation of a model. The compression method comprises the steps of performing sparse regularization training on a model to obtain a to-be-compressed model; calculating the importancescore of each filter according to the parameters of each convolution layer and the BN layer of the to-be-compressed model; setting an importance threshold according to the importance score and a preset pruning rate; cutting off the filters of which the importance scores are lower than the importance threshold and the BN layer parameters corresponding to the filters to obtain a pruned model; and finely adjusting the pruned model to ensure that the precision of the model is not lower than the preset precision. According to the method, on the premise that the model performance is not affected, large-amplitude compression of the model parameter quantity and the calculated quantity is achieved, the problem that resources are occupied by the model can be well solved, and the deep learning modelcan run on the edge computing equipment with limited resources.

The invention discloses a fashion compatibility prediction method based on low-rank regularization feature enhancement representation. The method comprises the steps that a feature matrix is decomposed into a first objective function composed of main features of multiple visual angles and a sparse error matrix; The features learned in the low-rank subspace are standardized through hypergraph items, and a second objective function of the relation between the fashionable single products is obtained; a Grassmannian manifold is introduced to obtain a third objective function of the maximum distance between dictionary base matrixes under different visual angles; Establishing a relation between the characteristics of the multi-view low-rank subspace and the output matching score, adding a sparseregularization item to the least square loss part, and obtaining a typical Lasso regression, namely a fourth objective function; Obtaining a fifth objective function taking the affinity matrix as a label matrix, establishing a relationship between the affinity matrix and the learned characteristics, and minimizing an error between the affinity matrix and the learned characteristics; And obtaininga total objective function according to the weighting of the first to fifth objective functions, optimizing the total objective function by utilizing an alternating direction multiplier method, introducing a Lagrangian multiplier, and sequentially iteratively updating parameters at each view angle until the value of the objective function is converged to obtain a final prediction score.

The invention relates to a joint constraint random noise suppression method based on sparse regularization, and belongs to the technical field of geophysical exploration. The method specifically comprises the following steps: S1, inputting an original seismic record, and constructing an objective function of curvelet transform-second-order generalized total variation joint constraint denoising according to sparse features of noisy data in a curvelet domain and an image gradient domain; S2, converting the L1-L2 norm regularization model containing the curvelet transform constraint term into a standard basis pursuit noise reduction problem, and inverting a curvelet coefficient with the minimum L1 norm to obtain a seismic record after preliminary denoising; and S3, taking the preliminarily denoised seismic record as an input image, solving a denoising problem of the second-order generalized total variation constraint to realize random noise suppression of the joint constraint, and finallyoutputting seismic data with an enhanced signal-to-noise ratio. According to the method, the denoising effect on random noise in the seismic data is improved, and weak signal energy can be effectively protected, so that high-quality processing of subsequent seismic data and the reliability of a seismic geological interpretation result are guaranteed.

The invention discloses a structural damage identification method based on subspace projection and sparse regularization, comprising the following steps: (1) arranging an acceleration sensor structurally for recording structural response information; (2) arranging a force sensor at the excitation point, performing the experimental modal analysis on the structure under the lossless state and the structure under the damage state separately, and extracting structural modal information; (3) establishing a sensitivity equation; (4) constructing a measurement matrix that satisfies the D-RIP condition; (5) performing subspace projection on the sensitivity equation; (6) introducing l1 norm regularization to establish a structural damage identification equation; (7) solving the recognition equationby using the signal space compression sampling matching tracking algorithm; (8) introducing damage prior information to rationalize the recognition results. According to the structural damage identification method based on subspace projection and sparse regularization, the damage is effectively located and the damage degree is quantified, the computational cost of the solution process is reducedeffectively, and the needs of assessing the structural health status timely and effectively in actual project are satisfied.

The invention provides an SAR (Synthetic Aperture Radar) image resolution-maintaining side-lobe suppression method. According to the technical scheme, optimization processing is performed on an SAR image by using a Newton iteration method to obtain a reconstructed SAR image f^, wherein the optimization objective function is shown in the description. In the formulas shown in the description, y represents SAR image data, and f represents the SAR image to be reconstructed; [lambda] represents a regularization parameter which is generally determined according to an empirical value; k is a very small positive constant, and a is greater than 1 and determined according to an empirical value; and log(1+abs(f)<2> / k) in the formula is used for measuring the sparsity of a vector f to be reconstructed. The method provided by the invention is very simple in step and can be directly used for performing resolution-maintaining side-lobe suppression processing on SAR image domain data.

The invention discloses a composite material structure impact load identification method based on iteration weighting sparsity. The method comprises the steps of acquiring a transfer function matrix between a composite material structure impact excitation position and a response measuring point position; Measuring an impact response signal generated by a to-be-identified impact load applied to thecomposite material structure; Constructing an enhanced sparse regularization model based on an lp norm, solving the sparse regularization model by utilizing an iterative weighting l2 norm method, andidentifying an impact load applied to the composite material structure; The invention further discloses a composite material structure impact load identification device based on iteration weighting sparsity. Compared with a Tikhonov method based on an l2 norm, the method has the advantages that the stability is high, and the amplification of measurement noise in an identification result can be inhibited; Compared with a standard sparse regularization method based on l1 norm, the method has the advantages that the identification precision of the peak force of the impact load can be improved, and the result is more sparse.