66 results about "Graph regularization" patented technology

The present invention discloses a bearing fault classification diagnosis method based on sparse representation and ensemble learning. The method comprises the steps of acquiring the vibration acceleration signals of a rolling bearing at different working rotating speeds via an acceleration sensor under each working condition as the training samples; selecting m training samples to form m sets of training data, establishing a weak classifier-graph regularization sparse representation model, and carrying out T times iterative operation on the graph regularization sparse representation; obtaining a classification function sequence via a weak classifier, then giving a weight to each classification function, and finally obtaining a strong classification function F by weighting a weak classification function; acquiring the vibration acceleration signal data of the to-be-tested rolling bearing at the rotation work via the acceleration sensor as a test sample; taking the test sample as the input quantity of the strong classification function to introduce in the strong classification function to operate, thereby being able to obtain a fault classification result of the to-be-tested rolling bearing. The method of the present invention enables the accuracy and the validity of the rolling bearing fault diagnosis to be improved.

An improved system and method is provided for determining web page quality using collective inference based on local and global web page information. A classification engine may be provided for classifying a web page using local features of a seed set of web pages and global web graph information about the seed set of web pages. A dual algorithm based on graph regularization formulated as a well-formed optimization solution may be used in an embodiment for applying collective inference for binary classification of the web page using the local web page information and global web graph information of a web page, the local web page information and global web graph information of an authoritative set of web pages, and the local web page information and global web graph information of a non-authoritative set of web pages.

A method of detecting and diagnosing cancers characterized by the presence of at least one nodule / neoplasm from an imaging scan is presented. To detect nodules in an imaging scan, a 3D CNN using a single feed forward pass of a single network is used. After detection, risk stratification is performed using a supervised or an unsupervised deep learning method to assist in characterizing the detected nodule / neoplasm as benign or malignant. The supervised learning method relies on a 3D CNN used with transfer learning and a graph regularized sparse MTL to determine malignancy. The unsupervised learning method uses clustering to generate labels after which label proportions are used with a novel algorithm to classify malignancy. The method assists radiologists in improving detection rates of lung nodules to facilitate early detection and minimizing errors in diagnosis.

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 graph regularization sparse coding-based magnetic resonance super-undersampled K data imaging method which comprises the following steps: (a) performing graph regularization sparse coding expression on a double-layer Bergman iteration frame to obtain an image sparse model; (b) updating a learning dictionary and a sparse coefficient on the inner-layer iteration of double-layer Bergman iteration by introducing an auxiliary variable and an alternate solving technology; (c) performing image updating on the outer-layer iteration of the double-layer Bergman iteration to obtain an imaging result by utilizing a part of super-undersampled K data as constraints. According to the method, a proximity graph is established to code local structural data and dig geometric data constraints thereof by introducing adaptive dictionary learning into graph regularization sparse coding, so that image data can be sparsely expressed better; in addition, an image with more complex local geometric characteristics can be processed, a local image structure can be effectively captured, more image details can be recovered, and an obtained image result is higher in fidelity.

The invention requests to protect a multi-image regularization depth hash multi-modal medical image retrieval method. The method specifically comprises the following steps of: simultaneously extracting features of a multi-modal medical image group through a multi-channel depth model; Correspondingly constructing a plurality of graph regularization matrixes according to the characteristics of the multi-modal medical image group; fusing Multiple graph regularization matrixes, and obtaining Hash codes of the multi-mode medical image set through modal self-adaptive restricted Boltzmann machine learning; solving The distance between a single modal data hash code and a multi-modal medical image group hash code through Hamming distance measurement, carrying out sorting according to an ascending order, and selecting and returning n groups of multi-modal medical images with the minimum distance to a user, so that multi-modal medical image retrieval is realized. According to the method, a doctorcan be helped to quickly find data of other multiple modes through data of a certain mode in multi-mode medical images such as ultrasonic images, dispute end texts and nuclear magnetic resonance images, medical diagnosis of the doctor is facilitated, the workload of the doctor is reduced, and the working efficiency is improved.

A cross-modal hash retrieval method based on self-learning belongs to the technical field of computers, and comprises the following steps: 1) learning potential public semantic features of different modals through a co-matrix decomposition technology; 2) learning a unified binary code with discrimination through orthogonal transformation and quantization processes; 3) maintaining and integrating the similarity in the modes and the similarity between the modes into a graph regularization item, and embedding the graph regularization item into a binary code generation process; 4) calculating andoptimizing a target function, and iteratively updating a plurality of matrix variables until a convergence condition is met; and 5) completing learning of a specific modal hash function by adopting aself-learning framework. Aiming at the problem of large quantization error caused by a threshold strategy, the binary coding loss of common representation of different modes is minimized, the similarity between the interiors of the modes and the similarity between the modes are embedded, and a self-learning hash scheme is introduced to learn a hash function with higher discrimination. Coding errors in a binary quantization stage can be effectively reduced, and the quality of hash codes and the performance of cross-modal retrieval are improved.

The invention provides a method of identifying cell types based on single-cell RNA sequencing data. According to the method, a low-rank representation model of a high-dimensional matrix is effectivelycombined with a graph regularization theory; an optimization model is constructed by considering the global structure and local structure characteristics of data, the model is solved by adopting an alternating direction multiplier method (ADMM) to obtain a reliable inter-cell similarity matrix, and then the similarity matrix is clustered by adopting a spectral clustering method, so that single cells are clustered, and the cell types are identified. According to the method, the clustering effect of the single-cell RNA sequencing data can be remarkably improved.

The invention discloses an unsupervised hyperspectral image implicit low-rank projection learning feature extraction method, and aims to provide an unsupervised hyperspectral feature extraction methodcapable of realizing rapidness and high robustness. The method is realized through the following technical scheme: firstly, dividing input hyperspectral image data into a training set and a test setin proportion; designing a robustness weight function, calculating the spectral similarity between every two training set samples, and constructing a spectral constraint matrix and a graph regularization constraint according to the training set; approximately decomposing row representation coefficients of the hidden low-rank representation model; constructing an implicit low-rank projection learning model by combining the spectral constraint matrix and the image regularization constraint; and optimizing and solving the hidden low-rank projection learning model by adopting an alternating iterative multiplier method, obtaining a low-dimensional projection matrix, outputting the categories of all test set samples, taking the low-dimensional features of the training set as the training samplesof the support vector machine, classifying the low-dimensional features of the test set, and evaluating the feature extraction performance according to the quality of a classification result.

The problem that in the prior art, the face recognition accuracy is low is solved. The invention provides an adaptive graph regularization non-negative matrix factorization method for face recognition. The method comprises the following steps: vectorizing a face picture; decomposing the data matrix X into a product of a base matrix and a coefficient matrix; adding a non-negative constraint to eachelement in the basis matrix and the coefficient matrix; constraining a coefficient matrix of the data matrix X; obtaining a data matrix X, obtaining a processed similarity matrix, processing the similarity matrix S by utilizing a diagonal matrix and a Laplace matrix, introducing a weight constraint item Lambda to obtain an adaptive graph, and finally introducing the adaptive graph as a regularization constraint item into a non-negative matrix factorization objective function of the data matrix X to identify the face photo. According to the method, the weight constraint term is introduced, theweight matrix is not kept unchanged any more, a large amount of noise and outliers are removed, local neighborhood structure information is complete, and the recognition accuracy is improved.

The invention discloses a sample clustering and feature recognition method based on integrated non-negative matrix factorization. The method comprises: 1, X = {X1, X2... XP} representing multi-view data composed of P different omics data matrixes of the same cancer; 2, constructing a diagonal matrix Q; 3, introducing graph regularization and sparse constraints into the integrated non-negative matrix factorization framework to obtain target functions O1 and O2; 4, solving the target function O1 to obtain a fusion feature matrix W and a coefficient matrix HI; solving the target function O2 to obtain a feature matrix WI and a fusion sample matrix H; 5, constructing an evaluation vector according to the fusion feature matrix W, and identifying common difference features according to the vector; 6, performing functional explanation on the identified common difference characteristics by using GeneCards; and 7, performing sample clustering analysis according to the fusion sample matrix. According to the method, the complementary and difference information of the multiple omics data can be fully utilized to identify the common difference characteristics, clustering analysis can be carriedout on the sample data provided by the multiple omics data, and a calculation method basis is provided for integrated research of different types of omics data.

The invention discloses a cross-modal data retrieval method and system based on graph regularization and modal independence, and the method comprises the steps of receiving the original data of different modalities, carrying out the feature extraction, building a multi-modal data set which consists of the image text pairs in one-to-one correspondence, wherein the multi-modal data set comprises a training set and a test set; projecting the feature matrixes of different modal data in the training set to a public subspace through optimizing a predefined objective function to obtain an image projection matrix and a text projection matrix; according to the image projection matrix and the text projection matrix, projecting the feature matrixes of different modal data in the test set to a publicsubspace; calculating the similarity between the projected matrix and other projection matrixes in the public subspace; and performing descending sort according to the similarity to obtain the data corresponding to the plurality of first feature projection matrixes, and performing the cross-mode retrieval.

The invention discloses a visual data completion method based on low-rank tensor ring decomposition and factor prior, and aims to solve the problem that a traditional data completion algorithm based on tensor decomposition depends on initial rank selection, so that a recovery result lacks stability and effectiveness, and a layered tensor decomposition model is designed. Tensor ring decomposition and complementation are realized at the same time, and for the first layer, incomplete tensors are expressed as a series of third-order factors through tensor ring decomposition; for the second layer, the transformation tensor nuclear norm is used for representing the low-rank constraint of the factors, and the degree of freedom of each factor is limited in combination with the factor priori of graph regularization; according to the method, the low-rank structure and the prior information of the factor space are utilized at the same time, on one hand, the model has implicit rank adjustment, the robustness of the model to rank selection can be improved, and therefore the burden of searching the optimal initial rank is relieved, and on the other hand, potential information of tensor data is fully utilized, and the complementation performance is further improved.

The invention discloses a drug target interaction relationship prediction method based on collaborative matrix decomposition, which considers drug attribute similarity and target attribute similarity,and combines drug topological structure similarity and target topological structure similarity to improve drug target interaction relationship prediction accuracy. The method comprises the followingsteps of: acquiring drug attribute similarity and target attribute similarity data and drug target interaction relationship data; constructing a drug attribute feature similarity matrix, a target point attribute feature similarity matrix, a drug topological similarity matrix and a target point topological similarity matrix; performing graph regularization-based collaborative matrix decomposition on the drug target incidence matrix, and integrating attribute feature similarity and topological structure similarity information of the drug and the target to obtain a final target function; and performing iterative updating by using a Newton method to obtain the feature representation of the drug and the target, reconstructing a drug-target interaction relationship matrix, and predicting the drug-target interaction relationship.

The invention provides a robust structure non-negative matrix factorization clustering method based on graph regularization, and the method comprises the steps: S10, obtaining m to-be-clustered images, and constructing k nearest neighbor graphs according to the to-be-clustered images; S20, a corresponding data matrix Y is obtained for each nearest neighbor graph, the data matrix Y comprises n datapoints, and a non-negative matrix decomposition method is used for decomposing the data matrix Y to obtain a feature matrix W and a coefficient matrix H; S30, establishing an objective function O ofrobust structure non-negative matrix decomposition based on graph regularization based on l2 and p norms; S40, according to the objective function O, using an iterative weighting method to iterate preset times, and updating the feature matrix W, the coefficient item and the graph regular item; S50 analyzing and clustering the feature matrix W obtained by each nearest neighbor graph by using a k-means clustering algorithm. According to the method, a robust loss function is adopted to measure a reconstruction error therein, mark data is not used in the robust loss function for judgment, and after a semi-supervised method of non-negative matrix factorization is introduced, the efficiency and the accuracy rate can be effectively improved.

The invention discloses a fusion network drug target relationship prediction method based on network enhancement and graph regularization. The method comprises the following steps of modeling a drug similar network and a protein similar network by using an undirected graph model; performing enhancement processing on the modeled drug similar network and protein similar network by using a network enhancement method based on three-order neighborhood random walk; extracting the enhanced similar network by using a similar matrix decomposition model with graph regularities to respectively obtain a drug network feature representation and a protein network feature representation; and training the prediction model, and inputting the drug network feature representation and the feature representationvector of the protein network into the trained prediction model to obtain a prediction value of the association probability of the drug target pair. According to the method, the global connection relationship between molecules can be better captured, the noise can be effectively suppressed, and robustness is higher when the molecular network data with different scales and different noise degreesare used for prediction.

The invention discloses a multi-modal robust feature learning model based on non-negative matrix factorization, and belongs to the technical field of computers. The model comprises the following stepsof firstly, carrying out normalization and special value preprocessing on a multi-modal data set; secondly, reconstructing modal data in a low-dimensional shared space, simulating a geometric space in a data space by utilizing a graph regularization thought, introducing a noise matrix to remove noise in the data space, and constructing a multi-modal robust feature learning model based on non-negative matrix factorization; thirdly, according to a model optimization result, sequentially updating the mapping matrix of each mode and the shared feature matrix of all modes, updating the noise matrix and updating a modal weight factor; and finally, judging the difference between the current model value and the last model value, and iteratively updating the third step until a model convergence condition is met. An effective model is derived according to the steps to solve the feature learning problem of the multi-modal data containing noise. A large number of experiments prove that the data representation performance obtained by the method is superior to that of the related model at the present stage.

The present invention discloses a bearing fault classification diagnosis method based on sparse representation and ensemble learning. The method comprises the steps of acquiring the vibration acceleration signals of a rolling bearing at different working rotating speeds via an acceleration sensor under each working condition as the training samples; selecting m training samples to form m sets of training data, establishing a weak classifier-graph regularization sparse representation model, and carrying out T times iterative operation on the graph regularization sparse representation; obtaining a classification function sequence via a weak classifier, then giving a weight to each classification function, and finally obtaining a strong classification function F by weighting a weak classification function; acquiring the vibration acceleration signal data of the to-be-tested rolling bearing at the rotation work via the acceleration sensor as a test sample; taking the test sample as the input quantity of the strong classification function to introduce in the strong classification function to operate, thereby being able to obtain a fault classification result of the to-be-tested rolling bearing. The method of the present invention enables the accuracy and the validity of the rolling bearing fault diagnosis to be improved.

The invention relates to a geological disaster multi-modal monitoring data fusion imaging method based on a resistivity method. The method is characterized in that resistivity monitoring data is usedas a complete modal subspace, a spatial and temporal discontinuous data mode including water level, deep displacement and soil humidity is used as an incomplete modal data set, and a geometric structure of the data is coded through a graph regularization factor to ensure the local similarity of each modal data; deep learning and incomplete multi-modal analysis are fused to construct a deep semantic matching model fusing a modal deep neural network and incomplete multi-modal matrix decomposition; the deep semantic matching characteristics of the multi-modal data are obtained by combining deep matrix decomposition, optimization and layer-by-layer modal semantic matching and updating, and the incomplete multi-modal data in the shared space at least comprises classification and clustering analysis. The effectiveness of a dynamic change inversion section of a landslide seepage field, a geologic body structure field and a deep displacement field generated by coupling is improved.

The invention discloses an L2,1 partial mark learning-based age estimation method. Firstly, a feature matrix and a label matrix of a face data set are obtained, and then an objective function of the method is constructed. In order to enable the label distribution of the sample to be as sparse as possible and to be simpler to solve, L2,1 norm is embedded into an objective function; in order to enable label distribution of adjacent samples to be similar as much as possible, a manifold hypothesis thought is adopted, and a graph regularization item is embedded into an objective function. Then, an optimization problem is solved by utilizing an alternating iteration method to obtain a discrimination coefficient A(t); and finally, the label distribution of a given face test sample is estimated by using the discrimination coefficient A(t), and the age of the face test sample is judged according to a probability maximization principle. On one hand, potential useful information of a feature space is fully utilized, so that label distribution of adjacent samples is close as much as possible, and the accuracy and robustness of the method are effectively improved; on the other hand, disambiguation can be carried out on the candidate mark set, and the label of the sample can be accurately estimated.

The invention discloses a collaborative filtering recommendation method based on a parallel auto-encoderparallel self-encoding machine, which comprises the following steps: 1, constructing a sparse auto-encoderself-encoding machine model to complete an objective function expressed by potential characteristics of a user, and learning high-level abstract characteristics based on the user to obtain areconstruction matrix of a user scoring matrix; 2, constructing a graph regularization automatic coding machine model to complete an objective function of commodity potential feature representation,and learning high-level abstract features based on commodities to obtain a the reconstruction matrix of a commodity scoring matrix; and 3, performing matrix multiplication on the reconstruction matrixbased on the user scoring matrix and the reconstruction matrix based on the commodity scoring matrix to obtain a prediction matrix in which the user is interested in the commodity, and recommending the user according to a result. According to the method, tThe autoencoders of different structures can be utilized in parallel, different feature information of the user and the commodity can be learned at the same time, more accurate high-level abstract features of the user and the commodity are extracted, prediction is conducted through the extracted abstract features, and the purpose of conducting more accurate recommendation for the user is achieved.

The invention provides a missing multi-modal representation learning algorithm for Alzheimer's disease diagnosis. The invention provides a brand-new missing multi-modal data completion model. The missing multi-modal representation learning algorithm for Alzheimer's disease diagnosis comprises the following steps: firstly, mapping original complete modal data to a hidden space by using an automaticencoder network; then supplementing a kernel matrix of an incomplete view in the kernel space by utilizing potential representation learned from the complete mode, meanwhile, constraining the correlation between samples and between modes through graph regularization and an HSIC algorithm, and finally mapping the kernel matrix into a feature matrix through correlation analysis of a kernel method to complete a subsequent diagnosis task. The Alzheimer's disease can be rapidly and accurately diagnosed under the condition that certain modal data are lost, and the algorithm has good performance andobvious advantages, can be applied to scenes such as assisting in diagnosis of doctors and community screening, and has a good development prospect along with development of a population aging process.

The invention provides a medical image classification method based on a self-weighted grading biological characteristic. According to the method, through adding a graph regularization item in a ridgeregression process, neighborhood relation information between brain regions of an MCI patient is sufficiently utilized, and an obtained regression coefficient more accurately reflect an information transmission condition of AD and NC to the MCI. Then a self-weighted grading method is used for performing fusion on the regression coefficient, thereby ensuring a fact that a sample with a relatively high regression coefficient value in a fusion process acquires a relatively high weight and realizing higher distinguishability and robustness of the calculated self-weighted grading biological characteristic. Finally, a support vector machine classifier is used for performing classification. The medical image classification method remarkably improves classification accuracy of pMCI patients and sMCI patients.

The invention relates to the technical field of data processing, in particular to a robust local and global regularized non-negative matrix factorization clustering method, which comprises the following steps of: acquiring an image clustering sample; constructing a nearest adjacency graph on the local scattering of the sample and introducing smooth regularization; using transformation to represent a global geometric structure of the space, and taking the global geometric structure as an additional principal component graph regularization item to be incorporated into an NMF algorithm; graph regularization term constraint is applied to the original NMF model through joint modeling, and the basis matrix is constrained by using LP smoothness constraint; in error measurement, correlation entropy is used to replace Euclidean norm, so that a robust local and global regularized non-negative matrix factorization objective function is obtained; iteration is carried out for preset times by using an iterative weighting method according to the target function, the variables U and V are updated, and robust local and global regularized non-negative matrix factorization is completed; and carrying out clustering analysis on the coefficient matrix by adopting a K-means clustering algorithm.

The invention discloses a zero-sample image classification method and device based on double auto-encoders, and relates to the technical field of image classification, visual and semantic features are projected to a public space to learn potential semantics, and a consistent weight matrix is constructed based on graph knowledge to enable double projections to keep a consistent data structure. An epsilon-traction technology is introduced, a visible class classifier based on label relaxation is designed, the discrimination of potential language meaning and the generalization ability of a model are enhanced, and the method comprises the following steps: acquiring a sample image; constructing a visual feature vector, then establishing visual and semantic feature spaces and constructing a consistency weight matrix, constructing a regularization self-encoder based on double graph embedding, introducing an epsilon-traction technology, and establishing a visible class potential semantic classifier based on label relaxation, training a double discrimination graph regularization self-encoding model to obtain a zero sample classification model, and obtaining class labels of unseen class test samples in a public space by using a distance calculation formula.

The invention discloses a visual noise reduction processing method based on regular graph Laplace transform. According to the method, a graph regularization sparse coding method is introduced. The feature vector of the Laplacian graph capable of representing the overall structure information of the image is further utilized and participates in the sparse coding process of the image block. A new sparse coding model based on the feature vector of the Laplacian graph is constructed. Therefore noise reduction of the image is achieved.

The invention discloses a multilayer network clustering method based on semi-supervision, relates to the technical field of artificial intelligence and complex networks, and not only takes obtained consensus prior information as a preprocessing means to enable a low-dimensional representation matrix H (i) of each layer obtained through symmetric non-negative matrix factorization to be more excellent. Moreover, the obtained consensus prior information is coded into a consensus subspace graph regularization item, and the consensus low-dimensional subspace H is optimized during the overall non-negative matrix factorization, so the method can make full use of the complementary topological structure information of each network layer, and can also make full use of the obtained consensus prior information; and the method is especially suitable for a multi-layer network with a large amount of noise and a sparse structure. The method is applied to social networks, protein networks and other multi-layer networks, cluster structures of various types of multi-layer networks are successfully recognized, and the method has great significance in understanding some social interaction behaviors among people, recognizing crowds with specific social attributes and improving social cooperation efficiency.

The invention discloses a fatigue driving monitoring method based on a graph regularization extreme learning machine. The fatigue driving monitoring method comprises the steps that electro-oculogram signals and face image information of a driver in the driving process are collected through electro-oculogram signal collecting equipment and a camera; respectively carrying out corresponding preprocessing on the electro-oculogram signal and the face image information; extracting eye movement features of the preprocessed facial image information, classifying three eye movement signals of the preprocessed eye movement signals, and extracting related features; when the eye closing duration in the eye movement characteristics exceeds a preset threshold value, determining that the driver is in fatigue driving and giving an alarm; when the eye closing duration in the eye movement features does not exceed a preset threshold value, the average value of the same features extracted from the electro-oculogram signals and the face image information is calculated; and inputting the average value and the extracted different features into a trained graph regularization extreme learning machine classification model, and discriminating the fatigue state through a GELM classifier. The monitoring method is objective and accurate.

The invention discloses a brain development data analysis method, system and device and a storage medium, and the method comprises the steps: constructing a graph regularization sparse depth self-encoding model, wherein a hidden layer in the graph regularization sparse depth self-encoding model is formed by stacking N sparse self-encoders with graph Laplacian regularization, a graph Laplacian regularization term is added into a loss function of the graph regularization sparse depth self-encoding model; carrying out training and fine tuning on the graph regular sparse depth self-encoding model; and performing brain development data analysis by using the trained graph regular sparse depth self-encoding model. The method, system and device and the storage medium can effectively solve the overfitting problem occurring when high-latitude small sample brain development data is processed.