97results about How to "Good clustering result" patented technology

The invention discloses a novel K value optimization method in a point cloud clustering denoising process. The method comprises the steps that (1) a three-dimensional laser scanning instrument obtains space sampling points of the surface of an actual object; (2) the space sampling points are used as K values to optimize a clustering sample for clustering, a K-means clustering method is used for generating different clustering results of point cloud clustering in a clustering number search range, clustering validity indexes are used for evaluating different clustering results, and an obtained best clustering number is used as the optimal K value; (3) the optimal K value is used as a clustering initial value of three-dimensional point cloud clustering denoising, and three-dimensional point clouds are subjected to clustering; and (4) local outlier noise points are identified and removed by carrying out Euclidean-distance-based threshold value judgment in a class of clustering results, and ideal point clouds are obtained. The novel K value optimization method is used, the value is used for carrying out optimization clustering on point clouds with noise, so that the denoising accuracy of ideal point clouds is high, denoising speed is increased, and a later-period reconstructed three-dimensional model is smooth and real.

The invention discloses a significance-based self-adaption threshold segmentation and extraction algorithm of a natural object image. The algorithm comprises following steps of: step 1, carrying out clustering segmentation and significance information extraction on a natural target image so as to obtain an average value significant image; and step 2, according to the average value significant image, carrying out self-adaption threshold segmentation on the natural target image so as to obtain an unprocessed segmentation result binary image and carrying out region filling on the unprocessed segmentation result binary image so as to a complete target binaryzation image. According to the invention, by counting the color distribution information, the number of peak values is effectively calculated and then used as a cluster center number K of a K-means cluster, so effects of artificial factors on a segmentation process are eliminated. Meanwhile, compared with the Mean Shift algorithm, the algorithm provided by the invention is characterized in that the clustering results of the K-means algorithm are better. Therefore, the accuracy of the segmentation result is perfect.

The invention provides a distributed online real-time processing method and system for multi-source and heterogeneous flow-state big data. The method specifically comprises the steps of crawling webpage data of each source by utilizing a distributed crawler duplicate removal algorithm; pre-processing the crawled page, constructing a corresponding tree by utilizing a visual page segmentation algorithm, pruning noise nodes according to a visual rule, classifying multi-layer pages, determining predicates under different types of pages according to different characteristics, and inferring data recording block nodes and data attribute nodes through the rule; Distributing the preprocessed data source by using a distributed message system, providing a data stream, and describing the state of a data node in the data stream to form state information; Selective storage operation is carried out on data streams by utilizing a Hadoop distributed file system based on K- And detecting the processed data by using a means text clustering method, determining a text similar to a predetermined sensitive information text, and screening out the sensitive information.

The invention discloses an effective index FCM and RBF neural network-based substation load characteristic categorization method. The method comprises the following steps that: load constituent ratios of a substation are adopted as characteristic vectors of load characteristic categorization of the substation; clustering analysis is performed on data samples of the load constituent ratios of the substation through using a fuzzy clustering analysis method so as to obtain data categorization results under different numbers of clusters, and an optimal number of clusters is determined through three kinds of clustering effect evaluation indexes, and a fuzzy subordination degree matrix and the clustering center of each category of under the optimal number of clusters are obtained; one group of samples are selected in each clustering category according to a principle of minimum distance, and category numbers corresponding to each group of samples are set, such that a training sample set is formed; a substation load characteristic secondary categorization model is established through adopting an RBF neural network, and the formed training sample set is utilized to train the neural network, and the trained neural network is further utilized to realize the load characteristic categorization of the substation. The effective index FCM and RBF neural network-based substation load characteristic categorization method of the invention has the advantages of simple operation and high accuracy.

The invention discloses an allocation method for wireless resources of an ultra dense network based on dynamic clustering. The allocation method comprises a base station dynamic clustering process and a resource block allocating process, and is characterized in that in the base station dynamic clustering process, dynamic clustering is performed on base stations randomly distributed in the network, a lot of base stations in the network are clustered according to an improved K-mean clustering method, and an effective allocation space is provided for inter-cluster resource block allocation of different modes of users; and in the resource block allocating process, joint processing is performed on single base station resource allocation of center users and inter-cluster CoMP resource allocation of edge users according to a clustering result in the step one, resource blocks with an excellent channel state of the base stations are allocated preferentially in clusters where the users are located according a provided proportional fairness based resource block allocation method, the received interference is reduced at the same time, the proportional fairness among the different modes of users is ensured, and an optimal resource block allocation result is acquired. The method disclosed by the invention can effectively improve the sum rate of the system users and achieves an ultimate objective of overall network resource optimization.

The embodiments of the invention provide a load shape acquisition method and a system. The method comprises the steps: acquiring a power consumption load curve, and determining all possible cluster class numbers; carrying out fuzzy clustering on the power consumption load curve by taking each possible cluster class number as a cluster class number of a fuzzy C-mean clustering algorithm respectively, so as to obtain a clustering result respectively; evaluating all obtained clustering results by using predetermined validity indexes so as to obtain the optimal clustering result, and acquiring all load forms of the power consumption load curve according to the optimal clustering result. According to the method and the system, the applicability of power consumption load curve clustering results and the accuracy of extracted load forms are improved.

A method for clustering large-capacity text includes presenting vector space of text, calculating similarity between each two texts, embedding pattern into dimensional space and using K-means algorithm to cluster texts to be two types, carrying out successive bisect till requirement is satisfied and pattern is not divided any more.

A big-data-oriented distributed density clustering method comprises the following steps that firstly, environment virtualization is performed, and a Hadoop platform is established; secondly, data are pre-processed and loaded, wherein an original data table is extracted from a database, a needed field is intercepted through a sqoop-query command, and the pre-processed data are directly extracted to an Hdfs; thirdly, a distance matrix is calculated; fourthly, a cut-off distance and dot density are calculated; fifthly, the minimum distance between a dot and a higher-density dot is calculated; sixthly, the critical distance of a critical density point and a critical density point are determined; seventhly, dot clustering is performed, so a final clustering result is obtained; eighthly, off-group points are removed. The big-data-oriented distributed density clustering method is fast and effective when a big data set is processed, and has the effect that input parameters have good robustness on the clustering result.

A K-nearest neighbor and multi-class merge based density peak clustering method and image segmentation system are disclosed, which use density calculation mode to describe sample distribution and adopt new evaluation index to obtain clustering center. Iterative allocation strategy is designed to classify the remaining points accurately. A local class merge method is given to prevent class splitting that will contain multiple density peak points. The invention describes the distribution of each data point through the density measurement method, constructs a preference index based on the densityand the distance to evaluate the clustering center, and utilizes the iterative distribution strategy to allocate the remaining points to improve the clustering accuracy. After completing local clustering, multi-class merging strategy is used to complete the local class merging operation. The results of numerical experiments show that it has good applicability in 22 test data sets and real data.

The invention is applicable to the field of data processing, provides a data clustering method, a data clustering system and data processing equipment. The method comprises the following steps: inputting a data set consisting of n objects with a block data feature required to be clustered and an expected class number k; selecting k block data objects from the data set to serve as an initial class center; calculating the distance from each object to the initial class center; distributing each block data object to the center closest to the block data object according to the calculated distance to form k disjointed classes; calculating the center of each class to serve as a new class center; repeatedly executing the step of distributing each block data object to the center closest to the block data object according to the calculated distance to form the k disjointed classes and the step of calculating the center of each class to serve as the new class center until the algorithm is converged; obtaining the division result of the data set. By the data clustering method, the data clustering system, and the data processing equipment, the data with the block feature can be processed directly without compressing the block data, so that the loss of information is avoided, and the obtained clustering result is better than the clustering effect obtained after the block data is compressed.

A global K-means clustering method based on a feature weight is disclosed. The implementation process of the method comprises the following steps of: firstly, clustering data into one class, wherein the optimal clustering centre is the centre of mass of all the data; secondly, figuring out the minimal data point of a target function through calculation and taking the data point as the initial clustering centre of a next class; thirdly, performing iterative updating by a K-means method with the feature weight to obtain the optimal clustering centre of clustering into two classes; and finally, performing updating iteration by orderly increasing the number of the clustering centres by the same method until the clustering of K classes predetermined is completed, thus completing the whole process of clustering all the data points into K classes. In the invention, the new global K-means clustering method based on the feature weight is established by combining the global K-means method with the K-means with the feature weight, and the clustering result is very stable; and in contrast with the experimental results of several K-means clustering methods, the effectiveness and the robustness of the clustering method disclosed by the invention are proved.

The invention relates to a method used for establishing semantic scene models for scene images of moving targets by utilizing a computer. The method comprises the following steps: S1, acquiring image data of the moving targets to be processed by utilizing video data input by an image processing device, and detecting and tracking the targets; S2, learning a classifier of pedestrian and vehicles based on a coordinated training method, reducing training marked samples and fully utilizing vehicles characteristics of the targets; and classifying the targets into pedestrian and vehicles according to the classifier obtained by learning; S3, clustering the tracks of all the targets to obtain a track cluster of the targets; and S4, acquiring the distribution region of each track according to the track cluster of the targets, and obtaining the main track in the distribution region of each track as well as the entry point and out point of the track by utilizing a mean shift algorithm, thus the region with certain semantics is obtained, and application is realized.

Provided is a K-means clustering method based on a quotient space theory. The method comprises the steps that firstly, a cluster number K and a data set X are input, then the data set is clustered, and finally a clustering result is output. The method has the good clustering result on class clusters in any shape and noise points, the clustering effect is far better than that of a K-means algorithm, the time performance of the method is far better than that of an MSCA algorithm, the time performance and the clustering effect are combined, and the overall effect is best.

The invention relates to an entropy weight-based global K-means clustering method. The method is realized through the steps of: firstly, clustering data into a class, wherein the optimal clustering center is the centroid of all data; next, determining the minimum data point of an objective function through computation, and taking the data point as an initial clustering center of a next class; then iterating and updating by utilizing an entropy weight-based K-means method to get the optimal clustering centers when the data are clustered into two classes; and sequentially increasing the number of the clustering centers by adopting the same method to iterate and update until the clustering of the set K class is completed, and in this way, completing the whole process of clustering all data points into the K class. According to the method provided by the invention, the global K-means clustering method is combined with K-means with entropy attributes, a new entropy weight value-based global K-means clustering method is constructed, and the clustering results are quite stable; and compared with the experimental results of varieties of K-means clustering methods, the effectiveness and the robustness of the clustering method provided by the invention are proven.

The invention relates to a current transient quantity principal component cluster analysis direction protection method for a power transmission line with a static synchronous series compensator, and belongs to the technical field of relay protection of power systems. The method includes the steps that when a failure occurs in the power transmission line with the static synchronous series compensator, principal component analysis is conducted on line mode current data with failure phases, wherein the line mode current data are acquired at the measurement end of the line in a window at 1ms; whether the failure is the forward failure or the reverse failure is distinguished according to whether the projection q1 of the sample data on a first principal component (PC1) axis. Through principal component cluster analysis direction protection of line mode currents with failure phases, whether the failure is the forward failure or the reverse failure can be reliably distinguished, whether the forward failure is located on the left side of the SSSC or the right side of the SSSC can also be distinguished, and sections of the failure can be reliably recognized, and therefore a basis for highly reliably recognizing the property of the forward failure and the property of the reverse failure is provided.

The invention discloses a data stream adaptive clustering method for mixed attributes. The method comprises the following steps: 1) data are pretreated and a mesh is initialized, and the partition granularity of the mesh in attributes per dimension and the similarity between mesh objects during an offline clustering process are determined; 2) online maintenance is carried out on the mesh; 3) when a user sends a clustering request, the clustering process is changed to an offline stage from an online stage, the mesh is divided to a dense mesh and a sparse mesh according to the density information of the mesh, as for the dense mesh, an improved DBSCAN algorithm is used for clustering, as for the sparse mesh, a density-distance distribution-based CCFD algorithm is used for clustering, and the finally-obtained clustering result is outputted; and the whole clustering process is completed. Applicability and scalability are good, a related data set can be effectively processed, and a good clustering result is obtained.

The invention discloses a speaker clustering method for simultaneously optimizing deep representation learning and speaker category estimation, and the method comprises the following steps: carrying out the preprocessing of a clustering voice sample, extracting I-vector features, training a convolution self-coding network, and extracting deep representation features; constructing an initial classaccording to the depth representation features to obtain a class number and an initial class label; adding a full connection layer and a Softmax layer to an encoder output layer of the convolutional self-encoding network to form a joint optimization framework, and using the Softmax layer for estimating the category of a speaker; and taking the sum of the reconstruction error of the convolutional self-encoding network and the speaker category estimation cross entropy error of the Softmax layer as a target function, and iteratively updating the joint optimization framework parameters until a convergence condition is met to obtain a voice sample of each speaker. According to the method, the optimized depth representation features and the speaker clustering result can be obtained at the same time, and the speaker clustering effect better than that of a traditional method is obtained.

The invention discloses an REST data service clustering method based on an Internet service domain. The REST data service clustering method comprises the following steps: step 1, performing formalizeddefinition; 2, performing service feature extraction and simplification; 3, constructing a topic clustering model; 4, performing similarity calculation; step 5, performing service clustering; 6, updating the globally optimal object by the data cells according to an operation rule; and 7, enabling each tissue cell in the system to used as an independent execution unit to evolve and run in a parallel structure, so that the system is distributed in parallel. In the system, a series of calculation steps are defined as one calculation step; starting from tissue cells containing an initial data cell object set, in each calculation, it means that one or more evolutionary rules are acted on the current data cell object set, when shutdown constraint conditions of the system are met, the system automatically shuts down, and calculation results are presented in the external environment of the system. According to the method, the characteristics of the service field can be better obtained, and abetter clustering result can be obtained.

The invention provides a high spectral image waveband selection method based on global optimal clustering. The method comprises steps that a ratio of difference between clusters and difference in clusters is taken as a to-be-optimized target function, a bisection method and a dynamic programming method are utilized to acquire the global optimal clustering result, through minimizing a wave band linearity reconstruction error, representative high spectral image waveband selection can be lastly accomplished. The method is advantaged in that high spectral image waveband selection is carried out, selection probability of noise wave bands can be reduced, and higher classification precision can be acquired.

The invention provides a target identification method and device based on laser scanning, and the method comprises the steps: obtaining the data of each point obtained through the laser scanning reflection of a to-be-identified scene through a laser radar, and calculating the three-dimensional coordinate value of each point according to the data of each point, wherein the data comprises a linear distance value between a sampling point scanned by the laser radar and the laser radar; obtaining a two-dimensional coordinate value of each point according to the three-dimensional coordinate value ofeach point, and clustering each point according to the linear distance value of each point and the two-dimensional coordinate value of each point to generate a plurality of categories; calculating each piece of target information according to the two-dimensional coordinate values of each category of points formed by clustering; and determining each target in the to-be-identified scene according to each piece of target information. In order to solve the problem that the horizontal resolution and the vertical resolution of the vehicle-mounted laser radar are different, the density clustering algorithm is adopted for clustering, the clustering radius is self-adaption about the linear distance, the obtained clustering result is good, and therefore the target identification precision is improved.

The invention discloses an acceleration method for the similarity propagation data clustering method. The method includes the following steps: firstly, partitioning the similarity matrix of an imported data set and cluster the sub-data set on each sub-matrix using the similarity propagation data clustering method; then, combining the clustering result of the sub-data set in some way, and based on which, clustering the whole data set again using the similarity propagation data clustering method. The invention handles intensive relational data set with a large amount of data and can obtain a basically same result as the similarity propagation data clustering method in a shorter time. Compared with the similarity propagation data clustering method, the invention is provided with an obvious acceleration effect for the clustering of intensive relational data set with a large amount of data.

The invention discloses a fast density clustering double-layer network recommendation method based on graph structure filtering. The method comprises the steps that (1) analog comment data is automatically generated through TextGAN according to historical user comment information to serve as false comments which are accurately annotated with a class mark and extremely similar to real samples; (2)historical real comments and the analog comments marked as the false comments are used as input, a graph-based virtual information filter for studying user access records is designed considering thatthe generated false comments are extremely similar to the real comments, and false users and false comments are detected through continuous iteration of confidence of users, stores and comments; and (3) in order to solve the problem of sparsity of result recommendation data, the recommendation method based on a fast density clustering double-layer network is designed. Through the method, self-adaptive selection of parameters can be realized, a good clustering result can be obtained, therefore, more effective personalized recommendation lists of users can be obtained, and the accuracy of recommendation is improved. An adversarial generative network is utilized to generate false samples extremely similar to the real comment data, and the fast density clustering double-layer network recommendation method based on graph structure filtering is efficient and reliable.

The invention discloses an image block clustering method based on Fourier spectrum characteristics, and the method comprises the steps: firstly carrying out two-dimensional fast Fourier transform of each image block, so as to obtain an amplitude spectrum corresponding to a Fourier domain; secondly setting the direction with the maximum frequency energy of the amplitude spectrum as a first direction D, enabling the number of a main spectrum line direction as the structural complexity C of image blocks, calculating a frequency component, and setting a frequency component mark; finally carrying out the clustering according to the first direction D of each image block, the structural complexity C and the frequency component mark. The method employs the frequency spectrogram to extract the structural complexity, the structural directivity and the frequency component distribution, so as to design a synergetic similarity measurement standard combining the structure, directivity and contrast characteristics and achieve the quick and fine clustering of image block data sets.

The invention relates to a Gauss mixture model tree and an incremental clustering method thereof, and belongs to the field of clustering technologies in computer application technologies. The method includes 1, providing a novel clustering structure, namely a Gauss mixture model tree (clustering tree); 2, on the basis of 1, providing an incremental clustering method based on the Gauss mixture model tree. The incremental clustering method includes four steps of inserting data, updating the clustering tree, deleting the data and determining clustering results. The relation of the four steps includes inserting each new data to an existing Gauss mixture model tree as required, and updating the cluster tree according to inserting results; by means of inserting the new data, detecting whether the data inserted into the clustering tree needs to be deleted or not; if so, deleting the data; after all data is read, determining the clustering results. The method has good effects in accuracy, executing efficiency and stability of clustering.

According to the image deep clustering method and system based on self-supervised comparative learning, comparative learning is utilized to improve the discrimination of embedding, and under the condition of not giving human annotations, the comparative learning can learn the embedding with high cosine similarity and strong discrimination for semantically similar samples by discriminating the samples. On the basis, according to the technical scheme, the subtasks capable of simplifying the learning process are mined, and due to the fact that the intra-class difference of samples of the same class is smaller than that of samples of different classes, it is determined that the subtasks are the most natural division mode according to the classes of the samples. Therefore, compared with a mixedexpert system, highly professional experts are encouraged, each expert is good at processing samples of a specific category, and a good clustering result is naturally obtained. Meanwhile, compared with a hybrid expert system, a single objective function is optimized, clustering degradation can be prevented without processing such as pre-training or regular terms, and the method can be applied tounsupervised clustering tasks of more complex images.

The invention belongs to the technical field of data processing, and discloses an adaptive semi-supervised density clustering method and system. First, density parameters are automatically extracted from data with and without class labels; The initial clustering analysis is performed to obtain the local clustering results; finally, the final global clustering results are obtained by integrating the local clustering results. The algorithm proposed by the present invention does not need to set the number of clusters. In the cluster analysis process, the number of clusters is automatically determined according to the data set density information; Automatically extract multiple sets of density parameters, and then use these density parameters to perform density-based cluster analysis on the target data set, which can obtain excellent cluster analysis results, and has strong adaptability and noise resistance.

The embodiment of the invention provides a lane line clustering method and device, and the method comprises the steps: obtaining a lane line feature image, and enabling pixels in the lane line featureimage to form a feature point set according to a preset rule; sequentially selecting feature point groups from the feature point set according to a preset condition, and performing transverse clustering on the feature point groups to generate clustering groups; respectively calculating a clustering center of each clustering group; and updating the lane lines in the existing lane line set according to a preset lane line reservation condition and a relationship between each clustering center and the existing lane line set, and returning to the step of carrying out transverse clustering on the feature point groups in the feature point set according to a preset condition to generate clustering groups until the feature point set is traversed to obtain a clustering result of the lane line feature image. The clustering method provided by the invention has good clustering speed and robustness, has a good clustering effect on Y-shaped and V-shaped roads, and further improves the accuracy of alane line recognition result.

The invention discloses an image segmentation method and system based on dynamic multi-objective optimization. The method comprises the steps of: constructing a multi-objective function by using a K-Means algorithm and a Fuzzy C-means (FCM) algorithm; defining an environmental change rule by using a background difference method; constructing a self-adaptive inertial dynamic factor; optimizing a timely mutation factor; based on the self-adaptive inertial dynamic factor and the timely mutation factor, dynamically optimizing the multi-objective function by using a multi-objective optimization particle swarm method. The K-Means and the FCM are optimized by dynamically optimizing a particle swarm algorithm, thus a good aggregation result can be acquired; the problem of error segmentation of pixels or edge blur is avoided, thus a good image segmentation effect can be achieved; the image segmentation is high in accuracy rate; the image segmentation method and system based on dynamic multi-objective optimization can provide high-quality result data and technical reference for image recognition.

The invention discloses an improved DPC clustering algorithm and system based on a symmetric neighbor relation, and the algorithm comprises the following steps of S1, obtaining the influence space ofeach sample point; S2, calculating the product of the sample point density and the distance; and S3, performing clustering operation according to the width-first traversal mode. According to the method, parameters can be reasonably selected, the Miller effect is avoided, the clustering effect is improved, and particularly, a very good clustering effect can be obtained on a data set with non-uniform density.

The invention relates to a multi-dimensional graph network node clustering processing method, device and equipment, and the method comprises the steps of converting an original unweighted multi-dimensional graph network into a weighted multi-dimensional graph network according to the attribute similarity and structural similarity of nodes; according to the constructed intra-layer transition probability and cross-layer random walk transition probability, performing intra-layer and cross-layer multi-layer network random walk processing on the weighted multi-dimensional graph network to obtain a sampling sequence of each node of the weighted multi-dimensional graph network; converting the sampling sequence of each node into low-dimensional embedding based on a SkipGram model; performing clustering processing on the low-dimensional embedding of each node by adopting a K-means algorithm to obtain a clustering result of each node of the weighted multi-dimensional graph network; and embedding and projecting each low dimension into a two-dimensional space by adopting a dimension reduction technology, and displaying a clustering result by adopting a graph visualization technology. The clustering effect is obviously improved, and the clustering effect is better.