211 results about "Matrix similarity" patented technology

Embodiments are described for a method for processing graph data by executing a Markov Clustering algorithm (MCL) to find clusters of vertices of the graph data, organizing the graph data by column by calculating a probability percentage for each column of a similarity matrix of the graph data to produce column data, generating a probability matrix of states of the column data, performing an expansion of the probability matrix by computing a power of the matrix using a Map-Reduce model executed in a processor-based computing device; and organizing the probability matrix into a set of sub-matrices to find the least amount of data needed for the Map-Reduce model given that two lines of data in the matrix are required to compute a single value for the power of the matrix. One of at least two strategies may be used to computing the power of the matrix (matrix square, M²) based on simplicity of execution or improved memory usage.

The invention discloses a fine-grained vehicle type identification method based on weak supervision positioning and subclass similarity measurement, which comprises the following steps: 1) weak supervision positioning: using a pre-trained VGG-NET network to locate the image object, processing the mask map of convolution layer, and obtaining the boundary frame of the object. 2) constructing a fuzzysimilarity matrix: features of the pictures in the training set after the positioning of the picture are extracted by using B-CNN, and a fuzzy similarity matrix to measure the similarity of each subclass is obtained according to softmax classification results; 3) performing sampling to form a triple set: sampling to form a triple set on that basis of a fuzzy similarity matrix; 4) joint learning of the improved loss function: the improved triplet loss and the weighted softmax loss are used to restrict the distance between samples of the same sub-category and increase the distance between samples of different sub-categories. Compared with the original model, the invention is more accurate in positioning, and the classification accuracy is obviously improved, so that the vehicle target can be positioned well.

The invention discloses a remote sensing image change detection method based on neighbourhood similarity and threshold segmentation and aims to overcome the defect that the traditional method has poor noise immunity and low detection accuracy in terms of the change detection of the target with high noise. The realization process comprises the following steps: (1) using the strength normalization formula to carry out gray level matching on two remote sensing images; (2) using neighbourhood similarity distance measure to construct a similar matrix of the two remote sensing images; (3) combing the similar matrix to construct a difference image of the two remote sensing images; (4) constructing a two-dimension gray level column diagram for the difference image, using the 2D-OTSU method to determine the segmentation threshold value and separating the target area from the background area; and (5) using the fuzzy entropy method to continue classifying the unprocessed edges and noise points. The invention has the advantages of good noise immunity and high detection accuracy for the changing target and can be used for detecting targets with changes of multitemporal remote sensing images.

The invention discloses a music separation method of an MFCC (Mel Frequency Cepstrum Coefficient)-multi-repetition model in combination with an HPSS (High Performance Storage System), and relates to the technical field of signal processing. In consideration of high probability of ignore of a gentle sound source and time-varying change characteristic of music, the sound source type is analyzed through a harmonic/percussive sound separation (HPSS) method to separate out a harmonic source, then MFCC characteristic parameters of the remaining sound sources are extracted, and similar operation is performed on the sound sources to construct a similar matrix so as to establish a multi-repetition structural model of the sound source suitable for tune transformation, so that a mask matrix is obtained, and finally the time domain waveform of a song and background music is obtained through ideal binary mask (IBM) and fourier inversion. According to the method, effective separation can be performed on different types of sound source signals, so the separation precision is improved; meanwhile the method is low in complexity, high in processing speed and higher in stability, and has broad application prospect in the fields such as singer retrieval, song retrieval, melody extraction and voice recognition in a musical instrument background.

The invention discloses an SAR (Synthetic Aperture Radar) image segmentation method based on parallel sparse spectral clustering, relating to the technical field of image processing and mainly solving the problem of limitation of segmentation application of large-scale SAR images in the traditional spectral clustering technology. The SAR image segmentation method comprises the steps of: 1, extracting features of an SAR image to be segmented; 2, configuring an MATLAB (matrix laboratory) parallel computing environment; 3, allocating tasks all to processor nodes and computing partitioned sparse similar matrixes; 4, collecting computing results by a parallel task dispatcher and merging into an integral sparse similar matrix; 5, resolving a Laplacian matrix and carrying out feature decomposition; 6, carrying out K-means clustering on a feature vector matrix subjected to normalization; and 7, outputting a segmentation result of the SAR image. The invention can effectively overcome the bottleneck problem in computation and storage space of the traditional spectral clustering technology, has remarkable segmentation effect on large-scale SAR images, and is suitable for SAR image target detection and target identification.

Provided is an image feature extraction method based on global and local structure fusion, characterized by comprising: 1) constructing a weight adjacent map; 2) determining laplacian matrix of similar matrix, degree matrix and images, 3) determining scatter matrix inside the kind and between the kind; 4) determining projection matrix, 5) identifying. The invention provides a feature extraction method of fusing the global structure information and the local structure information, wherein complex features fused of the global feature and the local feature are extracted, thereby the method has strong resolving power. The method not only has the characteristics of holding the reflection method locally, namely holding the characteristics of manifold structure of data; moreover has the characteristics of linear discrimination analysis method, namely assembling the date of the kind more compact to enlarge the distance between the kinds. The invention is applied in image recognition, thereby increasing identifying performance.

The invention discloses an optimized layout method for pressure monitoring points of an urban water supply pipe network. The method comprises the steps that S1, basic data of the urban water supply pipe network is acquired, EPANETH software is utilized to calculate hydraulic adjustment of the water supply pipe network, and a hydraulic model of the urban water supply pipe network is constructed; S2, a pressure difference matrix, a shortest distance matrix and a water volume influence fuzzy similar matrix are solved according to the hydraulic model of the urban water supply pipe network, and constraint conditions of an optimized layout model of the pressure monitoring points of the urban water supply pipe network are determined; S3, a model objective function is determined, and the optimized layout model of the pressure monitoring points of the urban water supply pipe network is constructed; and S4, a particle swarm algorithm is utilized to solve the optimized layout model of the pressure monitoring points of the urban water supply pipe network, and an optimal pressure monitoring point and monitoring areas of all the pressure monitoring points are determined. The method has the advantages of being simple in principle, easy to realize and high in efficiency and meets the requirements for representativeness, comparability and feasibility of point distribution.

The invention provides a method for identifying faces in videos based on incremental learning of face partitioning visual representations and belongs to the field of pattern recognition. According to the method, an Adaboost algorithm is used for detecting frontal face images in a first frame of the face videos, a Camshift algorithm is used for tracking, all face images are obtained, in the process of reading the face images in the videos, incremental cluttering is carried out on the face images, and a representative image is selected from each kind of face images; the representative images are processed, and a visual dictionary based on the piece visual representations is learnt; the visual dictionary is used for carrying out the representations on the face images; finally, according to similar matrices, the videos composed of the face images are identified. According to the method, an identification rate and robustness of the video faces can be improved under the state that illumination, postures and tracking results are not ideal. The faces in the videos can be detected, tracked and identified effectively, conveniently and automatically.

The invention relates to a method for pushing recommendation based on user historic behavior interaction analysis. The problem that a data platform in the prior art cannot supply an accurate and customized personalized information service to a user can be solved. The method comprises the following steps: presetting behavior and favorite articles of the user and performing weight allocation; collecting the behavior record information of the user in real time, classifying and then storing; establishing a favorite matrix according to the historic behavior with the highest weight of the user and respectively establishing a user factor matrix and an article factor matrix with the user and the data according to the article information contained in the data; performing singular value decomposition, thereby acquiring a similar matrix, comparing the similar matrix with the favorite matrix, selecting the disliked articles with high scores and recommending to the corresponding user. The method provided by the invention has the advantage that the user and the user as well as the data and the data are combined with each other, so as to form a high-precision relation quantitative index. The method provided by the invention is a continuous learning and promoting process.

The invention discloses a sorting method based on non-supervision feature selection. By means of the method, high dimensional data are expressed in similar diagrams, distances between sample points are obtained through the ITML, and a similar matrix of the original high dimensional data is set up; then the SM algorithm is executed on the similar matrix and a diagonal matrix corresponding to the similar matrix to achieve mapping of original sample sets to feather vector space; then through learning of sparse coefficient vectors and MCFS scores, weight coefficients of all attributes in the original sample set are obtained, and the attribute which can best express the original sample information is selected out; finally a support vector machine is used for setting up a sorting model of the selected data to predict fatigue states of a driver. The method selects features of the high dimensional data under the condition of maintaining data aggregate structures before the sorting model is set up, and the negative effect of curse of dimensionality on data sorting is avoided.

The invention discloses a dispersion tensor magnetic resonance image tensor domain non-local mean denoising method, and belongs to the technical field of digital image processing and applied mathematics interdisciplines. The problem that a dispersion tensor magnetic resonance image is easily affected by noise is solved. The method comprises the steps of firstly, sequentially conducting traversal on voxels of the dispersion tensor magnetic resonance image, setting a corresponding search region by using each voxel obtained by traversal as the center, then, conducting tensor matrix similarity comparison between all the voxels inside the search region and the center voxel, finally giving different weights to the voxels inside the search region according to the degree of the tensor matrix similarity, calculating a weighted mean tensor matrix, and obtaining the denoising result of the center voxel. The problem that the dispersion tensor magnetic resonance image is easily affected by the noise is solved.

The invention discloses an improved miRNA-disease relevance prediction method based on collaborative filtering. A miRNA-disease prediction problem can be regarded as a recommendation repair problem. On the basis of a known miRNA-disease-related bipartite network, miRNAs are recommended to use according to known preferences of the miRNAs to related diseases and vice versa. Firstly, an importance matrix SIGd of one disease to another is defined, calculated and measured. When a disease d (i) is perceived to be more important than a disease d (j), the score of SIGd (d(i), d(j)) is higher. Similarly, SIGr is defined and calculated in order to measure the importance of two miRNAs. Secondly, a significant matrix and a similarity matrix are utilized as weight for calculating scores. The similarity matrix is defined to represent similarity between miRNAs or between diseases. The final score of miRNA-disease relevance is the sum of the scores of a miRNA and a disease and score of the miRNA scored by the disease. With the method, higher prediction accuracy is realized.

The invention discloses a method for calculating similarity of short texts by using a deep convolution neural network, and aims to calculate the similarity of short texts by using each word in the short texts and obtain a relatively accurate value of the similarity. According to the technical scheme, the method comprises the following steps: 1) expressing a plurality of short texts as a plurality of matrixes, and sequentially replacing each word in texts by using corresponding word vectors so as to obtain an ordered vector sequence as one matrix; 2) generating a similar matrix of two matrixes representing target short texts, and arranging cosine similarity of word vectors so as to obtain a similarity matrix; 3) paving rows and columns of similar matrixes into same dimensions; and 4) reducing the dimensions of the similar matrixes into one value as the similarity, performing training dimension reduction on the similar matrixes for all similar matrixes of same dimensions by using the deep convolution neural network, and calculating a similarity degree through multi-layer sensation, thereby obtaining the value of the similarity.

The invention discloses a method for clustering low-voltage distribution network transformer districts based on fuzzy clustering. The method comprises the steps that characteristic indexes of the low-voltage distribution network transformer districts are established; characteristic index data to be analyzed are input, and then an original data matrix is established; standard processing is conducted on the original data matrix, so that a fuzzy matrix is obtained, and a fuzzy similar matrix of the fuzzy matrix is established according to the Euclidean distance algorithm; a fuzzy equivalent matrix is established, the fuzzy equivalent matrix is converted into a Lambda-cut matrix equivalent to the fuzzy equivalent matrix, a dynamic clustering diagram is formed, clustering analysis of the low-voltage distribution network transformer districts to be analyzed is achieved, and after the number of categories is determined, a clustering result of the low-voltage distribution network transformer districts is output according to analysis demand; according to the clustering result of the low-voltage distribution network transformer districts, data characteristics of the transformer districts of each category are analyzed, whether the transformer districts of each category are in urgent need for treatment is judged, the transformer districts in urgent need for treatment are screened out, and a follow-up treatment scheme is provided preliminarily. The method for clustering the low-voltage distribution network transformer districts based on fuzzy clustering has the advantages that the recognition speed is high, the classification accuracy is high, and classification effectiveness is high.

The invention provides a symmetric fully-homomorphic encryption method based on a plaintext similarity matrix. The symmetric fully-homomorphic encryption method aims to solve the technical problem of low efficiency of symmetric fully-homomorphic encryption at present, and is implemented by the steps that: a user generates two big prime numbers with the same length according to a requirement, constructs a residue class ring according to the generated big prime numbers, constructing a general linear group according to the residue class ring, calculates a homomorphic calculation public key and a symmetric secret key, encrypting a similarity matrix of plaintext matrixes by using the symmetric secret key, and decrypting ciphertext matrixes by using the symmetric secret key; a cloud server uses the homomorphic calculation public key for conducting homomorphic calculation on the ciphertext matrixes; and the user uses the symmetric secret key for decrypting a homomorphic ciphertext matrix. The symmetric fully-homomorphic encryption method is simple in the secret key selection and encryption process, hides the plaintext matrixes randomly, improves the safety of an encryption algorithm, does not introduce noise in the ciphertext calculation process, can conduct arbitrary calculation on the ciphertext matrixes according to needs, and can be applied to full-course encryption state protection of important data in cloud computing, big data environments and the like.

The invention discloses an inter-class inner-class face change dictionary based single-sample face identification method to solve the problem of limitations of the current single-sample face identification algorithm. The method comprises the steps of step1, obtaining expressions of face images in the compression domain; step2, building a face image training sample matrix containing k classes; step3, building an average face matrix and an inter-class face change matrix of a face database; step4, adding low rank and sparse constraints into the inter-class face change matrix; step5, solving an inter-class similarity matrix and an inter-class difference matrix; step6, projecting the average face matrix, the inter-class similarity matrix and the inter-class difference matrix to low-dimensionality space; step7, performing normalization processing on the dimensionality reduced average face matrix, the inter-class similarity matrix and the inter-class difference matrix through a normalization method, and performing iterative solution on the face image training sample matrix based sparse coefficient vectors through a norm optimization algorithm; step8, selecting column vector face labels in the average face matrix, which are corresponding to the sparse coefficient maximum, to serve as the final face identification result.

The invention discloses a high permeability distribution type renewable energy generating cluster dividing method, which comprises the steps that distribution type renewable energy power supply, load,and distribution line form an active power distribution network, and the distribution type renewable energy power supply, and a bus accessed by the load are used as the node, and the dividing of thegenerating cluster can be conducted based on the proper phasor; the characteristic vector is formed by a node power characteristic curve, a node geographic coordinate, and node electrical distance; the hour node power curve at a typical day is used as the characteristic curve of the node power; the geographic plane coordinate of the node is used as the node geographic coordinate; the voltage wattles sensitivity matrix among nodes can be calculated based on network topology and the typical day load curve; fuzzy clustering algorithm can be adopted to construct a similar matrix based on differentcharacteristics of characteristic vectors, and the distribution type renewable energy power supply generating cluster dividing can be conducted. The invention is advantageous in that the regulation and control problem of large scale renewable energy can be simplified, and the group regulation and ground control of the renewable energy can be benefited.

The invention is applicable to the technical field of privacy protection, and provides a spectral clustering method based on differential privacy protection. The method includes the steps of pre-processing sample data; calculating a similarity matrix; based on a k-near-value, simplifying the similarity matrix; adding the random noise satisfying Laplace distribution to the similarity matrix; constructing an adjacent matrix and a degree matrix based on the similarity matrix after random noise perturbation; obtaining the Laplace matrix based on adjacency matrix and degree matrix; obtaining the first m large eigenvalues and corresponding eigenvectors of Laplace matrices; normalizing the eigenvector to form eigenmatrix; using k-means clustering method to cluster the feature matrix to get the label of clustering. A spectral clustering algorithm is used to calculate the sample similarity between the sample data as the weight value between the data points, and then differential privacy algorithm is used to add random noise of Laplace distribution to the weight value to interfere with the weight value to achieve the purpose of privacy preservation. The interfered data can not only achieve privacy preservation but also ensure the effectiveness of clustering.

The invention discloses a personalized object recommending method based on object similarity and a network structure. The personalized object recommending method comprises the following steps of 1, defining that a system has n projects and m users, and according to the project purchasing or browsing history of the users, building an n*m adjacency matrix; 2, according to the project information, building an n*m project approximate matrix; 3, calculating an n*n project resource distribution matrix by the adjacency matrix and the project resource distribution process; 4, according to the project resource distribution matrix and the project approximate matrix, calculating the comprehensive n*n project distribution matrix; 5, according to the initial resource distribution result of the first user and the comprehensive project distribution matrix, calculating the final project resource distribution, and according to the final project resource distribution results, descending the projects; 6, recommending the first S unbrowsed/unpurchased projects of the user to the user. The personalized object recommending method has the advantages that the accuracy of recommending results is improved, and the method can be used for recommending books, movies, music and the like.

The invention discloses a fuzzy-clustering-based Aegis system signal sorting method, which can be applied to the signal identification of phased array radar AN/SPY-1 of an Aegis system. The method comprises the following steps of: carrying out standardization treatment on pulse description words of Aegis system signals by using translation standard deviation transformation and translation range transformation; calculating a fuzzy similar matrix among the pulse description words; converting the fuzzy similar matrix into a fuzzy equivalent matrix by using a transitive closure method; and converting the fuzzy equivalent matrix into a lambda-cut matrix equivalent to the fuzzy equivalent matrix, and obtaining a sorted result of the pulse description words of the Aegis system signals through sorting Rlambda of the lambda-cut matrix. According to the fuzzy-clustering-based Aegis system signal sorting method, the limitations of the traditional radar signal sorting method are broken through by adopting a fuzzy mathematical method, and full pulses and intrapulse characteristics are subjected to fusion processing, so that the difficult problem in processing of minimal signals is effectively solved, and the validity of extraction of radar signal characteristics is improved.

The invention relates to a similarity propagation and popularity dimensionality reduction based mixed recommendation method. According to the similarity propagation and popularity dimensionality reduction based mixed recommendation method, sparse data are processed in two phases; firstly, neighbors of the sparse data are expanded due to constant iteration of similar matrixes of users, resources and Tags through a similarity propagation method and accordingly elements for zero are filled; then a score algorithm in a search engine is introduced to calculate the Tag popularity in consideration of the problem that original data is provided with meaningless rubbish Tags, the tags with the popularity smaller than a certain threshold value are deleted to simplify data to perform dimensionality reduction on the matrix; recommendation results are diversified and the sparsity and cold starting problem can be relieved to some extent due to the fact that the recommendation based on contents and the collaborative filtering recommendation are combined. The similarity propagation and popularity dimensionality reduction based mixed recommendation method has the advantages of solving the problem of data sparsity in the individual recommendation process and being high in recommendation result accuracy, high in accuracy and high in reliability.

The invention relates to a random forest technology-based similar file retrieval method. Compared with the prior art, the method has the advantage that the defect of incapability of performing effective retrieval in a specific field is overcome. The method comprises the following steps of organizing a judgment document; constructing a case characteristic tree; performing weight training on the case characteristic tree, performing training for different targets by adopting a random forest method, and calculating a comprehensive weight of case characteristics; obtaining retrieval information, and inputting filtration conditions and query conditions of the retrieval information in an input mode of conditional selection, a text with conditions or the whole judgment document; calculating a case similarity matrix; and outputting a retrieval result, obtaining similar cases from the case similarity matrix, finding n cases most similar to the query conditions or cases with the similarity greater than s, making statistics on the information, and performing visual display. According to the method, the problems of instability and over-fitting caused by a single decision-making tree are effectively avoided by using the random forest-based method.

An image retrieval method based on deep Hash learning optimization comprises the following steps of 1, constructing a multi-layer full connection network firstly, connecting a panh function behind each layer of full connection, and finally conducting the sign operation on the network output; 2, constructing a semantic retention loss function obtained by a classification loss function and a weighted semantic similarity matrix, and a target function with discrete orthogonal constraint obtained by a quantitative loss function and a regular item; 3, optimizing an objective function; 4, dividing the obtained feature data set into a query set Q and a to-be-searched set D, taking a part of data in the to-be-searched set D to form a training data set P, inputting depth characteristics and label information of the training data set P, initializing a weight coefficient and a binary code, performing iterative optimization on the step 3 in sequence to obtain an optimal network weight coefficient,and obtaining a depth hash function by the step 2; and step 4, carrying out the image retrieval and precision testing. The method is relatively higher in precision and relatively higher in retrieval efficiency.

The invention discloses a multi-target tracking method based on an LSTM network and deep reinforcement learning. A target detector is used to detect each frame in a video to be detected, and a detection result qt<j> is output; a number of single-objective trackers based on a deep reinforcement learning technology are constructed, wherein each single-target tracker comprises a convolutional neuralnetwork and a fully connected layer and the convolutional neural network is constructed on the basis of a VGG-16 network; the tracking result pt of each single-target tracker is output; a similarity matrix, which is described in the description, of data association is calculated; a data association module is constructed based on the LSTM network; the similarity matrix is input to acquire a distribution probability vector At; At<ij> is the matching probability between the i-th target and a detection result j; and an acquired target detection result with the maximum matching probability isthe tracking result of the i-th target. The method provided by the invention is not affected by mutual occlusion, similar appearance and continuous quantity change in a multi-target tracking process,and improves the multi-target tracking accuracy and the multi-target tracking precision.