17 results about "Sparse approximation" patented technology

The invention discloses a non-rigid image registration method based on sparse prior, and the method comprises the following steps: constructing a bounded deformation function of a displacement field between a floating image and a fixed image, and taking the deformation function of the displacement field as a regular term of a variational model; establishing a sparse image registration model by adopting a data item of an L0 norm according to the sparsity characteristic of the gray difference between the registered image and the fixed image; further carrying out sparse approximation by adopting an Lp quasi-norm to obtain an Lp sparse image registration model through approximation; and solving the registration model by using a primal dual iterative weighting algorithm to obtain a registration result. According to the method, the robustness to noise is higher, so that a more effective registration result is obtained.

The invention discloses a mass spectrum data high-throughput alignment and parallel qualitative method based on deep characterization learning, and relates to the field of chromatography-mass spectrometry data processing and calculation mass spectrography. The method comprises the following steps: performing streaming, micro-batch and single traversal analysis on original mass spectrum data files of a plurality of samples to be detected; extracting metadata and main data by taking the fragmented spectrogram as a unit, inputting the metadata and the main data into a pre-trained depth representation learning model to generate a high-dimensional embedded vector, and incrementally writing the metadata and the main data into a column type storage database; constructing a sparse approximate nearest neighbor graph; executing a clustering algorithm on the graph to generate a consensus spectrogram set, and generating a consensus feature vector for each clustering cluster and a single example cluster; determining the nature of the consensus feature vector; and constructing a sample * feature matrix based on the cluster affiliation relationship and the qualitative result. According to the method, the technical problems of I / O bottleneck, memory overflow, combinatorial explosion, computing power waste and the like in large-scale mass spectrum data analysis are effectively solved.

The invention discloses a subarray beam domain high-resolution spatial spectrum estimation method under an under-snapshot condition. The method comprises the following steps: step 1, uniformly dividing N-element towed linear array sonar array element frequency domain data into K subarrays; 2, performing beam forming processing on each piece of sub-array data to obtain a beam domain result of each sub-array; 3, calculating a beam domain steering vector; 4, calculating a beam domain covariance matrix; 5, realizing signal and noise power parameter estimation based on a sparse approximate minimum variance criterion; and 6, carrying out frequency band weighted summation on the estimated signal power to obtain a broadband DOA estimation result. According to the method, sub-array division dimensionality reduction, beam domain coherent processing and sparse parameter iteration estimation are combined for processing, the method has the advantages of low snapshot adaptability and high anti-interference capability, the method can be used for near-field weak signal detection of the towed linear array under multi-target interference, and effectiveness is verified through marine test data.

The invention provides a target orientation estimation algorithm based on covariance matrix imaginary part reconstruction for solving the problems that the underwater environment is complex and weak signals are difficult to detect. The algorithm includes calculating a covariance matrix from array received data; extracting an imaginary part of the covariance matrix, and reconstructing a new covariance matrix according to the characteristic value and the corresponding characteristic vector of the imaginary part; initializing a variable factor and a noise power estimation value; calculating initial values of a spatial orientation spectrum and sparseness thereof; on the basis of a sparse approximate minimum variance principle, iteratively calculating the spatial orientation spectrum until convergence; normalization processing is carried out on the iteration space azimuth spectrum, angles corresponding to the first K maximum values are target azimuth estimation values, and K is the real number of target signals. According to the method, the covariance matrix is reconstructed by extracting the covariance matrix imaginary part, so that the influence of noise on weak signals is weakened; and on the other hand, the estimation performance of the weak signal and the sparseness of the azimuth spectrum are considered by adaptively adjusting the variable factor, and the resolution capability of the weak target is enhanced.

The invention discloses a sparse attention operator acceleration method based on an NPU, and the method comprises the steps: combining a Linformer with an Ascend C programming language, and achieving an Ascend C operator for calculating the Linformer based on the NPU; according to the method, an accelerated calculation mode of combining Linformer with Flash-Attention2 is adopted, so that the whole operator can be subjected to calculation in a blocking manner; according to hardware characteristics of the mercuric chloride NPU, optimization of vector calculation is carried out, the number of instructions used by multiple vector calculations such as vector subtraction, line-by-line summation and line-by-line max calculation for calculating safety softmax is reduced, and time consumption is reduced; aiming at the problem that front and back data cannot be parallel due to more dependence, the calculation efficiency of a sparse approximate attention operator is improved by using an algorithm for constructing parallel calculation by emitting Cube in advance; basic block segmentation is carried out on matrix multiplication, and the carrying time of matrix multiplication is shortened.

The application relates to a real-valued super-resolution direction estimation method of a high-order acoustic field sensor array, which uses a received signal covariance matrix to approximately calculate an estimated value of noise power, avoids the iterative operation process of noise power in the prior art, improves the calculation efficiency, and reduces the floating-point operation amount. By constructing an augmented matrix, the array receiving data matrix and the array manifold matrix with a multi-dimensional structure of an element are made into Hermitian matrices, so that the unitary transformation processing of the related parameters of the high-order acoustic field sensor array is realized. On the basis of obtaining the array receiving data matrix and the array manifold matrix in the real number domain, the array receiving data matrix and the array manifold matrix are applied to a sparse approximate minimum variance method with a variable exponential factor, and the completely real-valued sparse approximate minimum variance direction estimation with a variable exponential factor is realized.

This invention relates to the field of detection, and in particular to an online method and system for detecting impurities in calcium carbonate ore. Specifically, it includes: acquiring a continuous multispectral image sequence of calcium carbonate ore on a conveyor belt, and performing radiometric correction and spectral normalization preprocessing; identifying dominant spectral clusters of pure calcium carbonate ore by statistically analyzing the pixel spectra of the preprocessed images, and selecting a subset of high-confidence pixel spectra as the initial background training set; training a background dictionary, applying a set of sparse regularization constraints during training, and using the correlation between spectral bands to constrain the atomic energies of the dictionary to continuous or strongly correlated band groups; for each pixel of the detected image, using a sparse approximation algorithm combined with the background dictionary, calculating a sparse reconstruction coefficient vector and a reconstruction residual vector, and obtaining the reconstruction residual and coefficient activation uncertainty measure, respectively; if the reconstruction residual exceeds a first threshold, or the uncertainty measure exceeds a second threshold, it is determined to be an impurity point, and the location and quantity are summarized to generate an online impurity distribution map.

The invention discloses a power customer segmentation and behavior discovery method based on a Mahalanobis distance related Chinese restaurant process, and belongs to the technical field of power customer data analysis, and the method comprises the following steps: S10, data preprocessing: collecting multi-dimensional power consumption data of power customers; s20, the Mahalanobis distance is calculated; s30, sparse approximate optimization is carried out; s40, a Chinese restaurant process related to the Mahalanobis distance; s50, integrating a hierarchical Dirichlet process, and automatically learning optimal clustering configuration in a hierarchical structure; s60, performing abnormal behavior detection, and setting a threshold value to perform abnormal customer marking; and S70, outputting and explaining a result, outputting a customer segmentation result and the feature description of each group, and generating an explainable behavior mode report. According to the method, the potential segmentation structure of the power customer can be automatically found under the condition that the clustering number is not specified in advance, meanwhile, high-dimensional data, noise interference and a complex dependency relationship are effectively processed, and accurate customer behavior pattern recognition is achieved.

The present invention provides an On-grid sparse recovery radar target parameter estimation method and device based on echo reconstruction. The method comprises: reconstructing the target echo signal using an observation matrix, a sparse approximate solution and a residual; oversampling the transmitted signal and the reconstructed echo signal, and calculating the two-dimensional matched filter response; re-searching the target spectrum peak position, and using a second-order polynomial interpolation algorithm to obtain the time delay, Doppler frequency and amplitude parameters. The present invention reconstructs the target echo signal, oversamples the transmitted signal and the reconstructed echo signal, and calculates the two-dimensional matched filter response. After obtaining the target spectrum peak position, the time delay, Doppler frequency and amplitude parameters of the radar wave are finally obtained, which can reduce the computational complexity, improve the theoretical estimation accuracy, and facilitate engineering implementation.

This application discloses a method, apparatus, device, and medium for simulating aircraft flow fields based on CFD block sparse approximate inverse parallel solution, relating to the field of computational fluid dynamics. The method includes: determining the dependencies between lower triangular blocks in the target inverse matrix of the BSR matrix corresponding to the CFD equations of the aircraft flow field to construct a dependency index array; filling the target blocks in the target parallel tasks corresponding to the current column in shared memory, and performing multiplication, accumulation, and outer product compensation operations on the obtained tasks to be executed using tensor kernels and CUDA kernels; updating the approximate inverse results in the current column based on the obtained single-block execution results, determining the new target parallel task corresponding to the next column, and repeating this process until block operations in all columns of the target inverse matrix are completed; determining the block sparse approximate inverse results based on the obtained approximate inverse results of all columns, and then determining the flow field simulation results to quickly solve ultra-large-scale sparse linear equation systems.

The application provides an angle domain hybrid beamforming method suitable for a cell-free millimeter wave MIMO system. Firstly, the method uses angle domain hybrid beamforming instead of full digital beamforming to reduce hardware power consumption; secondly, a user-centered service scheme is adopted in the system, and the transmission power and the power consumption of the backhaul line are reduced through reasonable user association. The method respectively uses variables to represent whether the user is associated with the access point, the beam selection of different access points and the digital beamforming vector, then models the system user association and beamforming design problem as an energy efficiency maximization problem under the user service quality constraint and the power consumption constraint; through sparse approximation, Dinkelbach method, equivalent transformation and first-order Taylor expansion, the initial non-convex problem is converted into a convex problem which can be solved iteratively to obtain the optimal energy efficiency of the system. Compared with the early centralized MIMO system and the full digital beamforming method, the application improves the system energy efficiency while ensuring the user service quality.

This application discloses an image super-resolution reconstruction method, device, and storage medium, relating to the field of image processing technology. The method includes: performing sparse approximation operations on original image patches and corresponding redundant atom dictionaries based on an initial Mamba model to obtain sparse feature representation data of the original image patches; determining a first image patch reconstructed based on the sparse feature representation data, and constructing a loss function based on the mean square error between the first image patch and the corresponding real high-resolution image patch; performing backpropagation on the initial Mamba model according to the loss function to obtain a target Mamba model; performing super-resolution reconstruction on each original image patch according to the optimal atom kernel of the target Mamba model to generate corresponding target image patches; and stitching the target image patches together to obtain a target image. This improves the reconstruction efficiency of super-resolution reconstruction of low-resolution images.

The application aims at the problem that underwater environment is complex and weak signal is difficult to detect, and proposes a target direction estimation algorithm based on imaginary part reconstruction of covariance matrix. The algorithm includes calculating the covariance matrix according to the array receiving data; extracting the imaginary part of the covariance matrix, and reconstructing a new covariance matrix according to the eigenvalue and the corresponding eigenvector; initializing the variable factor and the noise power estimation value; calculating the spatial direction spectrum and the initial value of its sparsity; based on the sparse approximate minimum variance principle, iteratively calculating the spatial direction spectrum until convergence; normalizing the iterative spatial direction spectrum, and the angles corresponding to the first K maximum values are the target direction estimation values, wherein K is the real number of target signals. The application reconstructs the covariance matrix by extracting the imaginary part of the covariance matrix, which weakens the influence of noise on weak signal; on the other hand, by adaptively adjusting the variable factor, the estimation performance of weak signal and the sparsity of direction spectrum are considered, and the resolution ability to weak target is enhanced.

The application belongs to the technical field of rotating machinery fault diagnosis, and particularly relates to a rolling bearing fault feature extraction method, system and equipment based on a fault period constraint Gaussian process, vibration signals under constant rotating speed are acquired, and a fault feature period is preset; observation values are decomposed into potential impact signals and noise, and the impact signals are endowed with Gaussian process prior; a composite covariance function combined by a period correlation kernel and a local oscillation kernel is constructed to represent cross-period phase repetitiveness and impact oscillation waveform; an induced point is constrained in a single fault period, a variational sparse Gaussian process model is established, and a lower bound of evidence is optimized to synchronously update kernel parameters and induced point phases; impact signal posterior predictive mean is calculated to reconstruct impact waveform, and a fault feature frequency and its multiple frequencies are extracted through envelope demodulation. The application encodes bearing fault physics prior into a covariance structure, combines single-period induced point sparse approximation, realizes effective extraction of weak fault features under strong noise, and efficiently models high sampling rate signals.

The invention relates to a division-based multi-region crossing vector approximate retrieval method and system. The method comprises the following steps: acquiring a data set to be searched, sampling a part of vectors from the data set according to a sampling ratio lambda as routing vectors, randomly dividing unsampled vectors in the data set into m partitions, and taking the routing vectors as shared vectors of the partitions; sparse approximate neighbor graphs are constructed for the m partitions respectively; receiving query information, and performing first-stage search according to a first candidate set queue length to approach to an area near the query information; and second-stage search is carried out in the approaching area according to the length of a second candidate set queue, if a routing vector is expanded in the search process, copies of the routing vector in all partitions are added into the candidate set queue, and cross-partition dynamic search is carried out. Compared with the prior art, the method has the advantages that in tests of various graph algorithms, the query efficiency can be improved by 1.5-2 times almost under all precision requirements.

The present invention discloses a non-rigid image registration method based on sparse prior, comprising the following steps: constructing a bounded deformation function of the displacement field between a floating image and a fixed image, and using the deformation function of the displacement field as a regularization term of a variational model; establishing a sparse image registration model by using the data term of the L0 norm according to the sparsity characteristics of the grayscale difference between the registered image and the fixed image; and further adopting the L p The sparse approximation of the pseudo-norm is approximately L p A sparse image registration model is proposed; a primal-dual iterative weighted algorithm is used to solve the registration model and obtain a registration result. The invention is more robust to noise, thereby obtaining a more effective registration result.

The invention particularly relates to a real-valued super-resolution azimuth estimation method for a high-order sound field sensor array, which approximately calculates an estimated value of noise power by using a received signal covariance matrix, avoids the process of iterative operation of the noise power in the existing method, improves the calculation efficiency and reduces the floating point calculation amount. By constructing an augmented matrix, an array receiving data matrix and an array manifold matrix of which array elements have a multi-dimensional structure become a Hermitian matrix, so that unitary transformation processing of related parameters of a high-order sound field sensor array is realized. On the basis of obtaining an array receiving signal data matrix and an array manifold matrix of a real number field, the array receiving signal data matrix and the array manifold matrix are applied to a sparse approximate minimum variance method of a variable exponential factor, and completely real-valued variable exponential factor sparse approximate minimum variance orientation estimation is realized.