20 results about "Laplacian regularization" patented technology

The invention discloses a semantic-based text content index automatic identification method, which relates to the technical field of information retrieval, and comprises the following steps: initializing sparse projection and LSH signature, performing iterative optimization by using a Lagrange duality form and gradient update, adjusting hash digits, obtaining a fragment index through a k-d tree, and constructing an inverted index. According to the method, compression is performed through Delta coding, an index map is constructed based on Jaccard similarity, compression is performed through WebGraph, CSNMF is used in combination with Z-Laplacian regularization, a low-rank basis matrix and a low-rank coding matrix are generated, a compressed inverted index is reconstructed after iterative optimization, and reconstructed inverted index entries are generated. According to the method, through multi-resolution hash table initialization, joint feature optimization, local adaptive quantization and low-rank index reconstruction, the semantic expression ability and the compression effect of an index structure are improved, the index precision and efficiency are improved, and intelligent identification of index content is achieved.

The invention provides an industrial process operation data filling and repairing method and system, and relates to the technical field of data filling. The global long-term trend of data is captured through cyclic matrix nuclear norm minimization, local short-term fluctuation characteristics of the data are described in combination with time-adaptive Laplacian regularization, and correlation among variables is represented by graph structure regularization of adaptive weight. Balance parameters are dynamically adjusted through a self-adaptive weight mechanism, different time sequence characteristics are adapted, and the interpolation effect is optimized. And finally, solving the optimization model by adopting an alternating direction multiplier method, and generating filled and repaired data. Meanwhile, the global trend, the local fluctuation and the relation between variables of the data are considered, and the accuracy and the reliability of data filling are remarkably improved.

The invention relates to the technical field of semiconductor manufacturing and detection, and discloses a wafer defect detection data processing method and system.The wafer defect detection data processing method comprises the steps that a structured random measurement matrix is constructed based on wafer surface characteristics, and compression measurement data is collected; identifying local manifold structures corresponding to different defect types by using a spectral clustering algorithm; constructing a multi-scale image Laplacian operator, and extracting defect geometric characteristics; performing differential reconstruction processing according to the defect type, and solving an optimization target containing a nuclear norm and a graph Laplacian regularization item; dynamically adjusting a sampling strategy based on reconstruction quality evaluation; according to the method, high-precision defect reconstruction under a low sampling rate is realized, the data processing efficiency is improved, the complex defect reconstruction quality is improved, the method adapts to diversified defect processing, the storage and transmission burden is reduced, and the computing resource allocation is optimized.

The present invention provides a method and system for filling and repairing operating data of an industrial process, and relates to the technical field of data filling. The global long-term trend of the data is captured by minimizing the nuclear norm of the circulant matrix, the local short-term fluctuation characteristics of the data are described in combination with time-adaptive Laplace regularization, and the correlation between variables is characterized by graph structure regularization with adaptive weights. The balance parameters are dynamically adjusted through an adaptive weight mechanism to adapt to different time series characteristics and optimize the interpolation effect. Finally, the alternating direction multiplier method is used to solve the optimization model and generate the filled and repaired data. At the same time, the global trend, local fluctuation and relationship between variables of the data are taken into account, which significantly improves the accuracy and reliability of data filling.

The present application belongs to the field of image denoising, and in particular to a three-dimensional point cloud denoising method. The device and method are based on Laplace regularization subspace non-local low rank learning, and extend the previously proposed low-dimensional flow model of pixel blocks to surface blocks in point clouds. The graph Laplace regularizer of the surface is used to obtain the manifold dimension of the 3D point cloud image. Then, based on the principle that the high-dimensional information of the image point cloud is located in a low-dimensional subspace, the subspace non-local low rank factor is used to study the non-local self-similarity of the subspace, estimate the three-dimensional tensor generated by the non-local similar three-dimensional surface, and finally use the three-dimensional tensor to construct a denoising model to obtain a denoised image with better visual and quantitative indicators. The denoised image point cloud obtained by the method can better preserve the visually significant structural features, and the quantitative indicators are also good.

The present invention discloses a method and device for detecting infrared dim and small targets based on hypergraphs. The method includes: 1) constructing the image sequence to be detected into a three-dimensional spatio-temporal tensor; 2) establishing a low-rank sparse tensor decomposition model; 3) designing a tensor nuclear norm based on Laplace; 4) designing a hypergraph Laplacian regularization term; 5) establishing a spatio-temporal tensor decomposition model based on the hypergraph regularization term; 6) using an algorithm based on ADMM to solve the model established in step 5) to obtain a target detection result tensor; (7) converting the obtained target detection result tensor into a target detection result sequence as the infrared dim and small target detection result, thereby realizing the detection of infrared dim and small targets. The method for detecting infrared dim and small targets based on hypergraphs in the present invention can effectively improve the detection performance of infrared dim and small targets.

An image foreground-background segmentation method and system based on sparse decomposition and graph Laplacian regularization are disclosed. Firstly, an image is divided into a plurality of non-overlapping image blocks; Then, a foreground-background segmentation model of the image is established according to the image blocks; An image segmentation problem is divided into several sub-problems, which are solved by iteration; Finally, after the iteration, solutions of the problem are obtained; The obtained solutions are respectively matrixed and patched to obtain a foreground image, which is a foreground image of the whole image. The image foreground-background segmentation method uses the linear combination of graph Fourier basis functions to better represent the smooth background region. In addition, the graph Laplacian regularization is used to characterize the connectivity of foreground text and graphics while keeping sharp foreground text and graphics contours. The experimental results show that this method has better foreground-background segmentation effect.

The present invention provides a handwritten handwriting smoothing algorithm based on gradient graph Laplacian regularization, which includes the following steps: Step S1: Read the position information of handwritten handwriting points; Step S2: Reflect the position information of handwritten handwriting points on a matrix; Step S3: Determine whether two pixel points meet the distance requirement; Step S4: Repeat the above steps; Step S5: Concatenate vectors into a matrix; Step S6: Calculate its adjacency matrix according to the gradient graph constructed by the gradient; Step S7: Obtain the Laplacian matrix; Step S8: According to the Laplacian matrix, obtain its GLR expression form with respect to the gradient; Step S9: Define the energy function of the algorithm in this article; Step S10: Minimize the energy to obtain the best smoothing effect of handwritten handwriting; Applying this technical solution can finally achieve a better handwritten handwriting smoothing effect without occupying additional memory.

The invention discloses an X-ray sparse new view synthesis method based on inline prior guidance three-dimensional Gaussian splashing. Compared with an existing method based on three-dimensional Gaussian splashing, the depth constraint inline prior constructed by the method solves the problem of insufficient structure information under a sparse view angle; the constructed mask reconstructs inline prior, and the problems of inherent suboptimum and detail missing of rendered image distribution and over-fitting of three-dimensional gauss to a limited view angle under the sparse view angle condition are solved; on the basis, improved graph Laplacian regularization inline prior is introduced, adaptive weighted constraint is performed on density difference of adjacent Gaussian points, and the problems of space inconsistency and artifacts caused by non-uniform density distribution are solved. Finally, a new solution is provided for X-ray sparse new view synthesis, and the new view synthesis performance and reconstruction quality under the sparse view angle condition are improved.

The invention discloses a multi-scene single-frame large-view-field scattering imaging method based on polarization coding and spatial multiplexing, and the method comprises the steps: building a reflection-type scattering imaging system and a transmission-type scattering imaging system, and respectively measuring the OME ranges of the reflection-type scattering imaging system and the transmission-type scattering imaging system; multiple spatial multiplexing speckles are collected through single exposure of a polarization camera, and a linear multiplexing model is constructed; a signal identification method based on a minimum mean square error and an N-FINDR algorithm are introduced to realize target number identification and polarization specificity speckle initial value extraction; designing a Laplacian regularization term constrained Cozishi non-negative matrix factorization algorithm, and analyzing polarization specific speckles of different OME regions by using the extracted target number and initial value iterative optimization; and reconstructing a speckle result by using a phase recovery algorithm, and hiding the target after recovering the scattering medium. According to the method, the view field range is expanded by 4.5 times in a transmission-type / reflection-type system, the fidelity of a demultiplexing result reaches 32dB, and hidden scene reconstruction without continuous acquisition is realized.

The application discloses a semantic-based automatic identification method for text content index, relates to the technical field of information retrieval, and comprises initialization of sparse projection and LSH signature, iterative optimization by using Lagrange dual form and gradient update, adjustment of hash bit number, obtaining of a sharding index through a k-d tree, construction of an inverted index, compression through Delta coding, construction of an index graph based on Jaccard similarity, compression through WebGraph, generation of a low-rank basis matrix and a low-rank coding matrix by using CSNMF combined with Z-Laplacian regularization, reconstruction of the compressed inverted index after iterative optimization, and generation of a reconstructed inverted index entry. The application improves the semantic expression capability and compression effect of the index structure, improves the precision and efficiency of the index, and realizes intelligent identification of the index content through multi-resolution hash table initialization, joint feature optimization, local adaptive quantization and low-rank index reconstruction.

The invention discloses a therapeutic peptide prediction method based on a nearest neighbor classifier, and belongs to the field of biological information. According to the method, peptide sequence features are extracted from QSP740 and CPP400 data sets by using a UniRep model, a kernel risk sensitive loss function is fused on the basis of a K-nearest neighbor algorithm to construct an objective function, and a multi-Laplacian matrix is constructed in combination with three similarity matrixes of an RBF function, cosine similarity and a Pearson's correlation coefficient to enhance model robustness. And finally, classifying the peptide sequence to be predicted by minimizing the objective function and evaluating the performance. According to the method, features are automatically extracted through deep learning, the model robustness is improved through a kernel risk loss function, and the uncertainty of data is processed by means of multi-view Laplace regularization, so that compared with an existing method, the treatment peptide type can be more accurately predicted, and the classification precision and the anti-noise capability can be improved.

The invention discloses a robust non-negative matrix factorization hyperspectral image demixing method. The method comprises the following steps: step 1, calculating a noise estimation value and a noise estimation matrix of a hyperspectral image; step 2, constructing an improved weighted residual non-negative matrix factorization model based on the noise estimation value and the noise estimation matrix; 3, introducing a second-order adjacency relation to obtain a high-order graph, and carrying out Laplacian regularization to obtain a regularization term; 4, introducing a high-order graph Laplacian regular term to obtain a robust non-negative matrix factorization model based on global structure constraint; and step 5, realizing hyperspectral image unmixing based on the robust non-negative matrix factorization model of the global structure constraint. According to the method, a more robust and more accurate unmixing effect can be obtained, the internal structure of the hyperspectral image can be better protected, and the method is more suitable for scenes with more complex ground features, richer details or higher unmixing precision requirements.

The invention provides a recommendation method and device based on a random walk algorithm and a graph theory, and the method comprises the steps: collecting the credit investigation data of a plurality of users, and the feature data and behavior data of financial products, and constructing a bipartite graph; constructing a scoring matrix based on the bipartite graph, and further constructing a user similarity matrix and a product similarity matrix; performing Laplacian regularization processing on the target user similarity matrix and the target product similarity matrix to obtain a target user similarity matrix and a target product similarity matrix; by setting the initial step length, the number of iterations and the stop rule of the random walk algorithm, the algorithm is executed to extract a user access path. And according to the behavior data and the user access path, calculating the interest probability of the target user on the target financial product, and generating a recommendation list containing a plurality of target financial products. By constructing the bipartite graph and combining the random walk algorithm, the recommendation accuracy is remarkably improved, the complex relationship between the user and the product is accurately captured, the user interest is predicted, and a high-correlation recommendation result is provided. In conclusion, the recommendation accuracy is improved, and the user experience is improved.

This application provides a ToF depth image denoising method based on a Graph Laplacian Regularized Network (GLR). The method includes: acquiring raw data from a ToF sensor; removing noise from the raw data using a GLR to obtain a denoised depth map; and post-processing the denoised depth map to remove regions with confidence scores below a threshold, resulting in a final depth map. This approach utilizes the GLR to achieve better denoising performance and does not require actual data acquisition for training; it can be trained directly using simulated Gaussian noise, exhibiting good generalization ability.

The present invention discloses a deep non - negative matrix factorization method and device for temporal network evolutionary clustering, which relates to the technical field of graph data mining, and includes: using a deep non - negative matrix factorization model to factorize the point mutual information matrix at adjacent times onto the same set of basis matrices to obtain the deep feature matrix at the first time and the deep feature matrix at the second time; using self - representation learning to train the deep feature matrix at the first time to obtain a self - representation matrix; using the self - representation matrix to impose a Laplacian regularization term constraint on the deep feature matrix at the second time; at the same time, using the point mutual information matrix at the second time to impose a Laplacian regularization term constraint on the deep feature matrix at the second time; using the l 2,1 - norm to impose a sparsity constraint on the deep feature matrix at the first time and the deep feature matrix at the second time; constructing a joint deep non - negative matrix factorization model for community detection. The present invention can effectively extract deep features for community detection.

The invention relates to the technical field of industrial equipment state monitoring and signal processing, and discloses a dynamic equipment signal noise self-adaptive suppression method for a complex working condition, and the method comprises the steps: receiving an original one-dimensional time sequence signal of dynamic equipment through calculation equipment, carrying out the multi-dimensional phase space dynamics reconstruction, and generating a phase space trajectory matrix; performing differential calculation and normalization processing on the phase-space state vector to obtain a dynamic tangent vector field; combining the manifold geodesic distance and the dynamic tangent vector field to construct an anisotropic geodesic neighbor graph, and generating an anisotropic weight matrix; and constructing a graph Laplacian regularization optimization model containing a data fidelity item and a manifold smoothing item based on the weight matrix, solving to obtain a denoised phase space trajectory matrix, and inversely mapping the denoised phase space trajectory matrix into a one-dimensional reconstruction signal. According to the method, the geometric distance and the dynamic evolution direction are fused, the false neighbor points are eliminated, and the signal-to-noise ratio and the feature fidelity of the nonlinear signal of the dynamic equipment under the complex working condition are improved.

The invention provides a two-stage vehicle damage segmentation method and system oriented to an abnormal illumination condition, and relates to the technical field of computer vision and image processing. According to the method, a source domain is constructed by using a normal illumination sample, a target domain is constructed by using an abnormal illumination sample, and target domain feature representation is obtained through a backbone network; global and local attention is introduced in the first-stage training, and hierarchical clustering is combined to generate a target domain category pseudo label and a mask pseudo label; in the second stage of training, the cross-illumination features are adaptively aligned by adopting a mask level domain, and graph Laplacian regularization is constructed after convergence so as to strengthen boundary consistency. And finally, weighting multiple losses to form total loss, and realizing stable segmentation of the vehicle damage area under the abnormal illumination condition.

The invention discloses a titanium alloy blade additive manufacturing reversible deformation compensation method and system based on double-grid B splines and graded regularization, and belongs to the field of selective laser melting (SLM) additive manufacturing. The method comprises the following steps: acquiring an original CAD model and a deformed actual measurement model; rigid alignment and three-constraint corresponding point matching are carried out, an observation displacement field is extracted, and a local concave area is intelligently detected; constructing a global sparse + local dense double-grid B-spline displacement field, balancing global torsion smoothness and local recess fitting precision by adopting a hierarchical Laplacian regularization strategy, and solving through least square iteration to obtain a high-precision reverse compensation field; and negative displacement is applied to the original model, the compensation amount constraint is optimized, and a printing model is output for remanufacturing. According to the method, the combined deformation of overall torsion and local depression coexisting in the titanium alloy blade printing process can be efficiently restrained, calculation is efficient, material constitutive parameters are not needed, the method can be directly embedded into an existing additive manufacturing execution system, and closed-loop quality control is achieved.

The invention discloses an underwater hyperspectral clustering method based on bidirectional square attenuation tensor regularization, and relates to the technical field of underwater image processing. Comprising the following steps: reconstructing multispectral data in a feature dimension and a sample dimension; performing singular value decomposition on the tensor of the feature projection matrix and the tensor of the sample similar matrix, and constructing square attenuation tensors of the feature projection matrix and the sample similar matrix based on singular values; introducing a Laplacian diagram into the sample similar matrix, and constructing a hyper-Laplacian regularization constraint; self-adaptive weight is introduced to balance the effect of each view sample; and solving the target function to obtain a comprehensive similarity matrix, and performing spectral clustering on the comprehensive similarity matrix to obtain a clustering result. According to the method, singular values can be contracted adaptively, important spectral features are reserved to the maximum extent, noise and redundant spectral information are strongly suppressed, the separability of different ground feature categories in a feature space is enhanced, and robustness is achieved for interference factors such as illumination changes and sensor noise.