82 results about "CURE data clustering algorithm" patented technology

The trend of analyzing big data in artificial intelligence requires more scalable machine learning algorithms, among which clustering is a fundamental and arguably the most widely applied method. To extend the applications of regular vector-based clustering algorithms, the Discrete Distribution (D2) clustering algorithm has been developed for clustering bags of weighted vectors which are well adopted in many emerging machine learning applications. The high computational complexity of D2-clustering limits its impact in solving massive learning problems. Here we present a parallel D2-clustering algorithm with substantially improved scalability. We develop a hierarchical structure for parallel computing in order to achieve a balance between the individual-node computation and the integration process of the algorithm. The parallel algorithm achieves significant speed-up with minor accuracy loss.

The aspects of the disclosed embodiments include clustering a set of discrete user interface states into groups; presenting the groups on a display of a device; and enabling selection of any state within a presented group, wherein selection of a state returns the user interface to the selected state.

The trend of analyzing big data in artificial intelligence requires more scalable machine learning algorithms, among which clustering is a fundamental and arguably the most widely applied method. To extend the applications of regular vector-based clustering algorithms, the Discrete Distribution (D2) clustering algorithm has been developed for clustering bags of weighted vectors which are well adopted in many emerging machine learning applications. The high computational complexity of D2-clustering limits its impact in solving massive learning problems. Here we present a parallel D2-clustering algorithm with substantially improved scalability. We develop a hierarchical structure for parallel computing in order to achieve a balance between the individual-node computation and the integration process of the algorithm. The parallel algorithm achieves significant speed-up with minor accuracy loss.

A clustering fusion method based on user electrical load data subdivision comprises the following steps of (1) collecting data, (2) modifying the data, (3) converting the data, (4) standardizing the data, (5) constructing a pretreatment clustering algorithm set, (6) building a consensus matrix, (7) running the clustering fusion method, and (8) collecting users. Due to the facts that results obtained in different clustering algorithms or in the same clustering algorithm by using different parameters are automatically combined, and a best clustering result can be automatically judged and generated through clustering fusion, the clustering fusion method based on the user electrical load data subdivision has the advantages of improving self-adaption processing capability of a clustering analysis model, reducing dependency degree to priori knowledge in a user electrical load data clustering analysis process, reducing manual operations, and improving automatic degree of the clustering fusion method based on the user electrical load data subdivision.

The invention relates to the voltage control of the electric power field and especially relates to a reactive voltage partitioning method based on spectral clustering. A topologic matrix with a weight is used to construct a simplified power grid model. According to a spectral clustering definition, a Laplace matrix is acquired. Through an improved K-means clustering algorithm, clustering is performed on different characteristic vectors in a characteristic matrix. During clustering, modularity Q is introduced to be taken as an index of measuring an area partitioning quality. A partitioning scheme with a largest modularity Q value is selected as an initial partitioning scheme. Connectivity verification and reactive verification are performed on each area of the initial partitioning scheme. If the area can not simultaneously satisfy two conditions of area static state reactive balance and an enough reactive reserve margin, under the condition that a value of partitioning modularity Q does not change greatly, node adjusting is performed till that all the verification conditions are satisfied. In the invention, a topology structure of a complex power grid is embodied, calculating complexity is reduced, an integration evaluation system is established based on the modularity, the reactive balance and a reactive reserve index, and integration verification is performed on a partitioning result so as to ensure feasibility of the partitioning scheme.

The invention discloses a static clustering algorithm for a wireless sensor network, which belongs to the technical field of wireless sensor networks. The algorithm is applied to topology control of the wireless sensor network, and is different from the conventional network clustering algorithm. In the algorithm, clusters are partitioned firstly through convergent nodes, and a current turn of cluster head is used for electing a next turn of cluster head. The executing process of the algorithm is periodical, and is partitioned into a cluster partitioning stage, a cluster head selecting stage and a stable data communication stage when a network is established or a large quantity of nodes are added or leave the network, and only the last two stages are performed in the rest of periods. Compared with the conventional hierarchy topology, the algorithm has the advantages that: energy consumption in a network deployment process can be lowered, hierarchical division is more reasonable, and the service life of the network can be effectively prolonged.

The invention discloses a box separation method based on k-means clustering. The box separation method comprises the following steps that continuous variables are preprocessed; normalization processing is carried out on the preprocessed data, a k-means clustering algorithm is applied on the data obtained after the normalization processing is carried out to divide the data into a plurality of sections; the equal interval method is adopted for setting the initial center of the k-means clustering algorithm to obtain clustering centers; after the clustering centers are obtained, the midpoint of the adjacent clustering centers is used as a classification division point, each object is added into the closest class, and therefore the data are divided into the multiple sections; each clustering center is calculated again, then the data are divided again until each clustering center does not change any more, and the final clustering result is obtained. According to the box separation method, the technical problem that errors are likely to be caused for a data set with the obvious data density distribution bias according to an existing box separation method is solved, the k-means clustering algorithm does not select the initial center randomly any more, and the data separation result is accurate.

The invention relates to a clustering method, in particular relates to an improved fuzzy C-mean clustering method based on quantum particle swarm optimization, and belongs to the technical field of data mining and artificial intelligence. The improved fuzzy C-mean clustering method comprises the steps of: firstly, based on the conventional fuzzy C-mean clustering algorithm, improving the fuzzy accuracy of the conventional clustering algorithm by using a novel distance standard in place of a Euclidean standard; meanwhile classifying singly and quickly through using an AFCM (Adaptive Fuzzy C-means) algorithm to replace a randomly distributed initial clustering center to reduce the sensitivity of the clustering algorithm on the initial clustering center; and finally, introducing a QPSO (AQPSO (Adaptive-Quantum Particle Swarm Optimization)) parallel optimization concept based on distance improvement in a clustering process, so that the clustering algorithm has relatively strong overall search capability, relatively high convergence precision, and can guarantee the convergence speed and obviously improve the clustering effect.

The invention relates to a brain MR image segmentation method based on superpixel fuzzy clustering. The method comprises the following steps: 1) obtaining an MR image; 2) performing superpixel segmentation on the MR image and thus a plurality of atom regions are obtained; 3) performing secondary refinement segmentation on the atom regions whose grey value variance is big; 4) performing fuzzy clustering on the atom regions, so that category membership degree of each atom region is obtained; 5) defining the atom regions, whose membership degrees are not clear enough, as fuzzy blocks and realizing affiliation category judgment for the fuzzy blocks through a function iterative method; and 6) performing superpixel merging operation on the atom regions and thus image segmentation results are obtained. The method is a combination of a superpixel method and a fuzzy c-mean clustering algorithm; advantages of the superpixel method and the fuzzy c-mean clustering algorithm are effectively utilized to targetedly overcome the defect that the fuzzy c-mean clustering algorithm is sensitive to noise and a bias field during pixel level clustering; and compared with the traditional fuzzy c-mean clustering algorithm, the method is higher in segmentation accuracy and robustness.

The invention discloses a zinc floatation state dividing method based on isomerism textural features. Zinc floatation image textural features are extracted by combining a gray-level co-occurrence matrix algorithm which has a good effect on high-frequency band textural features and a Gauss Markov random field algorithm which has a good modeling effect on low-and-medium-frequency texture images, and the zinc floatation image textural features are subjected to Gauss normalization to serve as a textural feature vector. In an integrated clustering algorithm, partitional clustering with high efficiency is conducted firstly to eliminate the influences of noise points and outliers, then a hierarchical clustering algorithm with high clustering quality and high stability is adopted to combine clustering centers output through partitional clustering, and then a final clustering result is obtained. Experiments prove that the extracted textural feature quantity has high mode separability, and foam in different states can be distinguished with the integrated clustering algorithm; furthermore, the method can be directly realized on a computer and is low in cost, high in efficiency and easy to implement.

The invention discloses an improved k-means clustering method based on a distributed computing platform, and introduces a distributed computing platform Spark for the problem of slow processing of mass data, the Kruskal's algorithm for the problem of too many iterations and the Tanimoto distance for the problem of giving no consideration to the correlation among features of a vector. First, a minimum spanning tree is constructed for the randomly selected k points by using the Kruskal's algorithm, the corresponding weight sum is obtained, and the process is repeated for n times. Then, according to weight sums obtained within the n times, the maximum weight is selected thereform and that distance values between edges composed of the k points are not much different is ensured. In this way, the relatively uniform distribution of cluster centers can be guaranteed. Finally, a clustering algorithm is performed using k-means algorithm improved by using the Tanimoto distance.

The invention discloses a differential privacy protection-oriented k-means clustering method. The K-means clustering method comprises the following steps: performing data preprocessing; ensuring thatC indicates a clustered centered point set, and C indicates a sum of error square of a given data set and a cluster center C; judging the volume of C; performing cyclic execution until retry is greater than a maximum value retrymatx of given retry times, and then returning to the best central point Cbest; traversing each point of the data set X, classifying the points to the nearest central point;setting added random noises; renewedly calculating the sum of the data points of each cluster and the quantity of the points, and adding the noises and finally updating the quality center of the cluster; and repeatedly carrying out the steps until the sum of error square is converged or iteration times reach the upper limit. According to the differential privacy protection-oriented k-means clustering method disclosed by the invention, the appropriate random noises which are specially distributed are added in an iteration process of a k-means clustering algorithm, so that a clustering result is distorted to a certain extent, the aim of privacy protection is fulfilled, and meanwhile, the availability of data is ensured.

The invention relates to an RBF neural network indoor positioning method based on sample clustering. According to the method, signal packet loss rates under different transmitting power are taken as basic data, a clustering algorithm is adopted for screening out a training sample set with similar feature points, then the sample set is trained through an RBF neural network, and finally the position coordinate of an unknown mobile node is predicted. Due to the relation of communication distances and the packet loss rates, the sample set of the RBF neural network indoor positioning method is rich in information, and relation of signals and the distances can be depicted better; meanwhile, the clustering algorithm is adopted for screening out the position similar feature points and RBF neural network training data, so that data under large-scale and large-range conditions are convenient and easy to collect, the purpose of practicability is achieved really, and meanwhile the algorithm has the advantages of being high in convergence rate, accurate in positioning and the like.

The invention discloses a deep learning clustering method for noise images. The deep learning clustering method comprises the following steps: S1, constructing a deep learning clustering model; S2, adopting an AMsoftmax layer as a clustering device, and generating a clustering result according to the feature vector output by the encoder in the step S1; S3, measuring the similarity between the output of the encoder and the output of the twin network by adopting an L2 norm; S4, adopting KL divergence to measure the distribution difference between the clustering result and the auxiliary target distribution; S5, training a deep learning clustering model; And S6, obtaining a clustering result of the data through the AMsoftmax layer. According to the method, unsupervised clustering can be carried out on image data containing noise, and the problems that most image clustering algorithms do not model noise data and an existing deep clustering algorithm is poor in clustering effect on images with high non-linear characteristics are solved.

The invention discloses a thunderstorm kernel identification and tracing method based on a hybrid clustering algorithm. The method specifically includes the following steps of utilizing deployed thunder and lightning monitoring points to conduct exploration and record cloud-to-ground lightning data, conducting preprocessing on the recorded cloud-to-ground lightning data and dividing the data intolightning data sets of each equal time interval; adopting a GPS clock synchronization technology and an algorithm of time differences of arrival for figuring out spatial positioning coordinates of lightning according to the time differences of arrival of changeable radiation pulses of an electric field generated by the lightning to each station; on the basis of thunder and lightning positioning data figured out by means of the procedures above, utilizing a DBSCAN algorithm and a KMEANS algorithm to figure out the relevance of the thunder and lightning positioning data among a thunderstorm kernel center-of-mass coordinate position, the lightning frequency and a thunderstorm kernel. An experimental result indicates that the method can accurately reflect the change tendency of thunder and lightning on thunderstorm days, and great effects of identifying the thunderstorm kernel and movably tracing a thunderstorm are achieved.

The invention discloses a dynamic weighted hybrid clustering algorithm based circuit breaker fault diagnosis method. The method includes the following steps: (1) capturing the energy changes of mechanical drive during the operation of a circuit breaker by utilizing three-axis vibration and two-way sound signals, decomposing the signals through local mean, and extracting the approximate entropy ofeach PF component as the characteristic quantity of a circuit breaker vibration signal; (2) optimizing the initialized clustering center of fuzzy kernel clustering by utilizing the maximum density peak decision of a density peak clustering algorithm, and considering different influences of different characteristics and different samples on clustering results; (3) performing checking on a clustering number K through a cluster validity index MIA; (4) inputting correctly classified characteristics into a multi-level classifier of a support vector machine to perform training; and (5) finding the optimal parameter of the support vector machine through mesh generation, and inputting test data samples to perform final fault classification prediction so that classification accuracy rates can be obtained;. The method has advantaged of being fast in fault diagnosis speed and high in accuracy rate.

The invention belongs to the network technical field, in particular to an online traffic identification method based on an incremental clustering algorithm. The method includes: an offline recognitionstage and an online recognition stage, wherein in the offline recognition stage,a semi-supervised learning flow algorithm based on an improved K-means algorithm is used to perform preliminary clustering and mapping work on the prepared training data sets, and the data sets which are preliminarily classifiedare obtained; in the online recognition stage,based on the completed clustering and mappingdata sets formed in the offline identification phase, incremental clustering is used to determine the network application type of the newly added data streams online, so as to achieve the purpose oftraffic identification. According to the method,based on machine learning technology, by constructing a suitable recognition model to learn the prepared data, the online traffic can be incrementally clustered in real time, and the preliminary semi-supervised classification can be carried out by combining the prepared training set, which can realize the online recognition of network traffic, and has good real-time performance and high recognition rate.

The invention discloses a microblog topic detection method and system based on an incremental clustering algorithm; the method comprises the steps of S1, acquiring a microblog information set; S2, preprocessing the microblog information set; S3, after preprocessing, extracting characteristic words according to word occurrence frequencies, word distribution in microblog text, and word distribution in a time window; S4, weighting the characteristic words, and vectorizing the characteristic words and their weights; S5, combining topics by means of a similarity judging method based on inter-vector spacing. The microblog topic detection method and system based on the incremental clustering algorithm have good effects in terms of recall rate, accuracy and the like, with operating speed greatly higher than that of the k-means method.

The invention discloses a hot topic detection method based on weighted LDA and improved Single-Pass clustering algorithm. The hot topic detection method of the Pass clustering algorithm comprises thefollowing steps: preprocessing text data, including Chinese word segmentation, stop word removal and feature word weighting; modeling the text data by utilizing a weighted LDA topic model, realizing feature dimension reduction by mining hidden topic information in the text data, and filtering and denoising a vectorized result; subjecting text vectorization result processed by LDA topic model weighted by feature words to improved Single-Pass clustering algorithm to carry out clustering; and calculating a hot value of the topic cluster by utilizing the topic cluster scale and the topic cluster compactness, and identifying the hot topic. The detection method has the advantages of being low in algorithm complexity, low in dependency on text input time sequence and the like.

The invention discloses a pavement crack recognition algorithm method and system based on dual-scale clustering algorithm. The pavement crack recognition algorithm method includes steps of reading a three-dimensional image data matrix to obtain a binary image by a computer; scanning a data matrix corresponding to the binary image to obtain a marked crack area by an eight-neighborhood search algorithm according to the sequence from top to bottom, left to right; clustering cracks by the dual-scale clustering algorithm to obtain the clustered crack area; utilizing the minimum circumscribed ellipse where the clustered crack area is located as pavement cracks. The pavement crack recognition algorithm method is low in complexity, short in operation time and free of manual involvement. Disordered crack data is locally linearly fitted, model-established theory is indicated by a regular and certain mathematical expression, and data processing complexity is lowered; detection of the pavement cracks can be completed by inputting and acquiring pavement crack data only, so that the algorithm is high in detecting efficiency and fast and has a certain research value.

The invention provides a tag-co-occurred tag clustering method. A tag matrix, a common tag matrix, a tag importance degree matrix and a similarity matrix are defined in order to improve clustering effectiveness; feature vectors of tags are determined by extracting tag co-occurrence information; the method that the distance between one object and another object is calculated by using geometrical distance in the traditional clustering algorithm is changed into the method by using Pearson correlation coefficient in the way that the similarity is calculated by extracting the feature vectors; the tag-co-occurred tag clustering method which is used for clustering the tags by being combined with the K-means clustering algorithm is provided. The clustering method provided by the invention is better than the other clustering methods in effect and has good effectiveness and feasibility.

The invention discloses an SAR image segmentation method based on a manifold distance two-stage clustering algorithm. The method mainly solves the problem that an existing clustering segmentation algorithm is unstable in result and big in calculated amount. Achieving steps includes: (1) setting an ending condition and an operation parameter; (2) inputting an image to be segmented, and conducting pre-segmentation on the image to be segmented; (3) extracting characteristics of an image block obtained by pre-segmentation; (4) taking Euclidean distance to be used as similarity measurement to conduct first stage clustering for the characteristics of the image block; (5) taking a clustering center of a first stage and a point farthest from the center as representative points; (6) calculating a manifold distance between any two representative points; (7) taking the manifold distance as the similarity measurement to conduct clustering of a second stage of the representative points; (8) refreshing a clustering center; and (9) judging whether an end condition is achieved, if the end condition is not achieved, returning the step (7), or outputting a segmentation result image. The method has the advantages of being accurate in segmentation result, stable, short in time, and capable of being used in the technical fields such as image strengthening, pattern recognition and target tracking.

The invention provides a network table semantic recovery method. The method comprises the steps that based on a Probase lexeme database, preliminary semantic recovery is conducted on a network table to be recovered, and a candidate concept set of each column in the network table is obtained; according to combination distances among different tuples in the network table, each initial clustering center in a clustering algorithm is determined, the tuples in the network table are involved into clusters where the initial clustering centers are located, the clustering centers of the clusters are adjusted, and according to the final clustering center of the clusters, a network table after the shrinkage is conducted is obtained; according to the candidate concept set of each column in the network table and the network table after the shrinkage is conducted, column tags and column entities of all the columns of the network table are recovered out. According to the method, by selecting the initial clustering centers and calculating the similarity based on the combination distances, the K-means clustering algorithm can be improved, the scale of the network table is effectively shrunk, the complexity to fulfill a task is reduced, and the accuracy of recovering the column tags and the column entities of the network table is improved.

The invention comprises invention discloses an automatic segmentation method for fuzzy spectral clustering brain tumor images based on super pixel, comprising the following steps of : firstly, performing super pixel segmentation on a FLAIR mode image of magnetic resonance imaging containing brain tumors, and extracting gray histogram features of super pixel blocks as input of an algorithm, calculating a fuzzy similarity matrix of images through the input features; then performing clustering through NJW spectral clustering algorithm to obtain a final segmentation result. According to the automatic segmentation method for fuzzy spectral clustering brain tumor images based on super pixel, fuzzy theory is used to optimize similarity measurement mode of spectral clustering, fuzzy weight parameters are introduced in Gaussian distance measurement method of spectral clustering, and a fuzzy similarity measurement mode based on super pixel features is defined. The invention is an automatic imagesegmentation method, human intervention is not needed, and time complexity of spectral clustering algorithm is greatly reduced and segmentation accuracy can be improved by utilizing fuzzy spectral clustering algorithm based on super pixel.

The invention discloses a physiological data preclinical processing method and system. The physiological data preclinical processing method comprises preprocessing physiological data based on the time series; carrying out association rule analysis according to the calculated abrupt change score and by means of a multidimensional abrupt change detection model and an integrated learning algorithm fused with multiple classifiers, and obtaining a disease associated network according to the result of association rule analysis; selecting a disease associated network characteristic from the disease associated network by means of an improved clustering algorithm, obtaining a disease diagnosis result according to the disease associated network characteristic and historical data, wherein the improved clustering algorithm is based on a nonnegative matrix decomposition theory and a self-learning mechanism, and extracting a corresponding connected subgraph from large graph data of the disease associated network as the disease associated network characteristic through adjusting the subgraph density. The physiological data preclinical processing method and system are advantaged in that the method and system are wide in applicability, and high in efficiency and precision, and is flexible and convenient, and can be widely applied to the field of data processing.

The invention discloses an internet data analysis system which comprises a data preprocessing module and a data analysis module, wherein the data preprocessing module extracts main content from webpage information of the internet, a text corresponding to each webpage is obtained through filtration, the obtained texts are firstly segmented by a segmentation device to obtain a plurality of segmentation words, and segmentation words highlighting characteristics of the texts are reserved through dimensionality reduction of characteristic values; and the data analysis module selects one or more categories of algorithms from a classification algorithm, a clustering algorithm, an association rule algorithm and special rule matching algorithm according to analysis requirements, each category of algorithm adopts one or more algorithms for processing the segmentation words which are subjected to dimensionality reduction and correspond to the webpages output by the data preprocessing module, and the analysis result is stored. With the adoption of the internet data analysis system, the defect of inaccurate data analysis result caused by a single data mining algorithm is overcome, or the time cost due to the need of secondary system development when other algorithms are used on the basis of one algorithm is saved, and efficiency and accuracy of data analysis are improved.

The present invention discloses a medical MR image segmentation method based on Hough transform and a geometric active contour and belongs to the technical field of image processing. According to the method, firstly, the prior shape knowledge that inner and outer contours of a left ventricle myocardium on a short axis image are of a round-like shaped is utilized, Hough transform is adopted to estimate an initial contour of a left ventricle and the reason of adopting Hough transform is that Hough transform has strong robustness, is not quite sensitive for the incompleteness of data or noise and can recognize partially deformed or partially shielded objects; the reason of using a K-means clustering algorithm is that as a square error based clustering method, the K-means clustering algorithm is simple and has a high clustering speed; and by using the method provided by the present invention, the inner and outer contours of the left ventricle can be segmented effectively, relative positions of evolving curves of the inner and outer contours of the left ventricle can be controlled and a function of shape constraint can be realized.

A model selection method is provided for choosing the number of clusters, or more generally the parameters of a clustering algorithm. The algorithm is based on comparing the similarity between pairs of clustering runs on sub-samples or other perturbations of the data. High pairwise similarities show that the clustering represents a stable pattern in the data. The method is applicable to any clustering algorithm, and can also detect lack of structure. We show results on artificial and real data using a hierarchical clustering algorithm.

The invention discloses a discovery method of the core drug of a traditional Chinese medicine prescription. The discovery method consists of an improved clustering algorithm and a weighting TF-IDF (Term Frequency-Inverse Document Frequency) algorithm, wherein the clustering algorithm comprises three parts including prescription data pretreatment, the selection of a clustering distance function and a clustering mining algorithm; the prescription data pretreatment is used for processing prescription data into a model suitable for the clustering algorithm; the selection of the clustering distance function is used for selecting a reasonable clustering distance function; the clustering mining algorithm is used for clustering similar prescriptions into one cluster; and the weighting TF-IDF algorithm is used for calculating the weight of the drug. A weight calculation formula is combined with a clustering result, a medicine sequence importance degree and the TF-IDF algorithm, and the algorithm is high in accuracy.

The invention belongs to the technical field of image processing, in particular to an image division method based on biogeography optimization capable of being used for image enhancement, mode identification, target tracking and the like. The method comprises the following steps that parameters are initialized; pictures to be divided are input, and clustering centers are initialized; a fuzzy matrix is calculated; the clustering centers are calculated again; the fitness value of each clustering center is extracted; the immigration rate and the emigration rate of each emigration center are extracted; the emigration centers are updated according to mutation operators; and division results are output. Because the biogeography migration strategy is adopted for optimizing the clustering centers, the calculation quantity is reduced, the selected clustering centers after the optimization have the overall situation characteristics, the problem of initialization sensitivity of the traditional clustering algorithm is solved, and the stability and the clustering performance of the clustering algorithm are improved. Because the immigration and emigration strategies are introduced for optimizing the clustering centers, the data center distribution characteristics can be more accurately reflected, in addition, the corresponding clustering center updating rules are designed, and the calculation quantity is reduced.