48 results about "Dirichlet process" patented technology

The invention provides a rapid detection method for hot issues of timing sequence massive network news, comprising the following steps of: dividing a network news text sequence into block sequences according to time intervals; clustering a news text of the first block according to a Dirichlet process to form a clustered set; attenuating and filtering a result of clustering the front block to be used as prior distribution of subsequent blocks; clustering the subsequent blocks according to the Dirichlet process; carrying out hot degree sequencing of issues of each cluster according to reporting amount; and taking T clusters with the highest sequencing value as the hot issues; and selecting M characteristics with the highest tf-idf value in each cluster as keywords of hot spots and displaying the hot spots. According to the rapid detection method for the hot issues of the timing sequence massive network news disclosed by the invention, the efficiency of clustering the network news can be greatly improved; and meanwhile, the occupation of an internal memory is not linearly increased along the increasing of data quantity, and the rapid detection method is suitable for large-scale text data analysis.

The invention relates to the field of cognitive radio communication and provides a multi-layer distributed combined spectrum sensing algorithm based on a Dirichlet process so as to realize dynamic spectrum sensing. The sensing data acquired by secondary users in a plurality of hierarchical centers is fused to search the optimized sensing information. The automatic data packet is realized by employing the Dirichlet process, a shared hyper-parameter and a corresponding divergent probability in each packet are estimated by a bayesian model, the hyper-parameter is acquired by employing a standard Viterbi algorithm, and the hyper-parameter is compared with a decision threshold value to acquire a final spectrum decision result so as to determine whether a channel is available. Due to the design, the space diversity information of the compressed sensing data is fully considered, and the uncertainty of an individual secondary user on the compressed sensing data is reduced, so that the normalized mean squared error performance is high, the compressed sensing data information in each hierarchical center can be effectively obtained through the algorithm, high accurate detection probability and low false alarm probability are acquired, and the spectrum sensing performance is improved.

The invention relates to a method for estimating satellite navigation direct signal parameter and multipath signal number and parameter; the technical characteristics of the invention lie in that: Dirichlet process DP (alpha, G0) is adopted as the prior distribution of the multipath signal number M, the distribution of M conforms to DP as shown in formula I, the possible number of the multipath signal is obtained by sampling according to the prior distribution; according to the possible number of the multipath signal, an equation of state with parameters of the direct signal and the multipathsignal as the quantity of state is constructed as x(k+1)=Fx(k)+w(k) and an observation equation of observation vector is constructed as y=aR(tau)+n; according to the possible number of the multipath signal, the equation of state and the observation equation, a particle filtering method is adopted to estimate the parameters of the direct signal and the multipath signal, and simultaneously the number of the multipath can be determined. The invention can estimate number and parameter of the multipath signal under the condition of not knowing the number of the multipath signal, thus being suitablefor the condition that the direct signal is larger than the multipath signal, as well as the condition that the direct signal is smaller than the multipath signal.

According to the invention, an obstacle avoidance radar based on an ultra-wideband cognitive CPPM signal generates a number of wide band CPPM signals which are orthogonal to each other. The signals pass through a transmitting beamforming module to form radar transmitting signals, and the signals are transmitted by a rectangular electron scanning array antenna of a receiving/ transmitting switch module. If another flight target carries out reflection, an echo signal is received by an electronic scanning array antenna to form a received signal, and the signal is collected by a signal acquisition module. A receiver matches a filter module to acquire the distance-doppler information of the target. The echo amplitude, the azimuth and the height information of the target are acquired through a two-dimensional multiple signal classification module. The information enters a Dirichlet process hybrid model clustering module for gather classification, and then different targets are distinguished through a generalized likelihood ratio detection module. The individual target information is sent into a variable point detection module to detect a mutational point in a target movement trajectory. According to the mutational point, the amplitude and the pulse width of the CPPM signals are redesigned, and the movement trajectory of an unmanned aerial vehicle is corrected to avoid collision with other flight targets.

The invention discloses a method of detection, division, and expression recognition of the natural face based on a layered TDP model. Firstly, an original image is subjected to pre-treatment; an SIFT characteristic and corresponding position information are extracted; the two characteristics are combined by adopting the effective transference Dirichlet process to obtain a characteristic vector with geometric constraint; and the characteristic vector is input to the TDP model to obtain a first layer result-a human face or a non-human face. Then, the human face after division is taken as input of a second layer, and a division result of a subarea is obtained through the same process. Finally, the subarea is taken as the input of a third layer that is an expression recognition layer of the human face, and a result of expression recognition of a human face image is obtained by the same processes of characteristic extraction and combination. According to the invention, the problems that a model is needed to be established for every gesture in conventional multi-gesture expression recognition, and the model recognition rate is low due to factors such as a gesture are solved, and accuracy of expression recognition of a multi-gesture human face image is effectively improved.

The invention provides an interaction behavior detection method in a video monitoring scene, and relates to the technical field of digital image processing. Features of light streams between adjacent frames in a video sequence are extracted, and the video sequence is expressed in a word bag mode; then video files are modeled by means of a layered Dirichlet process model so that each video file can have an atom behavior distribution expression related to the corresponding video file; the dynamic change of an atom behavior is expressed as a multivariate point process, and non-parameter Granger causality analysis is carried out on the multivariate point process; ultimately, a causality directed graph is obtained according to Granger causality, and local and overall interaction behaviors are detected. According to the interaction behavior detection method, local interaction behaviors can be detected and meanwhile overall interaction behaviors can be detected.

The invention provides a short text topic identification method and system, and relates to the technical field of data processing. The method comprises the following steps of S1, obtaining a first corpus set and a second corpus set, wherein the first corpus set is a short text data set to be processed, and the second corpus set is an auxiliary corpus set; S2, obtaining a hidden feature vector based on words on the second corpus set, and constructing a Dirichlet process hybrid model based on the first corpus set; S3, constructing a non-parameter theme model based on the implicit feature vectorand the Dirichlet process hybrid model; S4, performing parameter inference on topic posterior distribution of the non-parameter topic model; S5, inferring and identifying the number of topics in the first corpus set based on the parameters, and obtaining the document-topic distribution and the topic-word distribution in the first corpus set at the same time. According to the method, the Dirichletprocess hybrid model and the implicit feature vector representation of the introduced words are constructed, so that the sparsity problem can be effectively relieved, and the accuracy of short text topic identification is improved.

The invention relates to a cross-camera pedestrian track matching method. The method comprises the following steps of S1, extracting one pedestrian track of a target camera as a target track, and taking all pedestrian tracks of other cameras in this time slot as candidate tracks; S2, using the Chinese chain restaurant process to train and layer the Dirichlet process, extracting the global motion pattern characteristics of all tracks, and obtaining the characteristic weights of the target track and each candidate track in the global motion pattern; and S3, calculating the cosine distance between the characteristic weight of the target track and the characteristic weight of each candidate track as the similarity measurement, and then selecting the candidate track with the smallest cosine distance as the matching track of the target track.

The present invention relates to the field of machine learning, and proposes an ocean trajectory clustering and predicting method. In order to accurately predict future trajectory points, trajectory clustering is required first. According to the trajectory clustering method disclosed by the present invention, similarity measurement is carried out on the trajectory points of complex variability andstrong volatility at sea, and the potential data information is mined; and the method combines the Gaussian mixture model GP with the Dirichlet process DP, and the non-parametric Bayesian framework of the DP is used to determine the number of clusters to improve cluster adaptability. The algorithm uses the process of adding Chinese restaurants based on the DP, and uses the collapsed Gibbs sampling method to solve the model, so that the unsupervised classification from the finite mixed model to the infinite mixed model is implemented, the number of clusters can be automatically obtained, and future trajectory points are predicted for the clustered trajectories by using the Gaussian process regression prediction method. According to the technical scheme of the present invention, the disadvantages of manually specifying the number of clusters and local maximization in parameter estimation are avoided, and the accuracy of prediction is improved under the premise of ensuring adaptive clustering.

The invention relates to a cooperative spectrum sensing method based on multi-user historical sensing data mining, used for judging whether a target channel is occupied. The method comprises the steps of 1, screening historical sensing data in each SU node via a rejection process, thus acquiring effective historical sensing data; 2, performing iterative solution by using a viscous stratified Dirichlet process-Hidden Markov model to acquire an implicit spectrum state of the sensing data and a corresponding Gaussian distribution parameter; and 3, judging the spectrum state of the sensing data. Compared with the prior art, the method provided by the invention has the advantage of accurate and rapid judgment.

The invention discloses a track modeling method based on a viscous multi-modal dual layered dirichlet process hidden markov model (SMD-HDP-HMM). The method comprises the following steps: expressing each track as a visual document to form a training set comprising a plurality of visual documents; learning the SMD-HDP-HMM model by using the visual documents in the training set. The invention also discloses a track searching method. The track searching method adopts the SMD-HDP-HMM model generated by the track modeling method. The track searching method comprises the following steps: expressing a new input track as a visual document; matching the visual document of the new input track with the SMD-HDP-HMM model, judging that whether the visual document of the new input track is abnormal and/or searching a most similar track in the SMD-HDP-HMM model.

The invention relates to a hybrid state space model-based sensor data blind correction method and belongs to the field of sensor networks. The method includes the following steps that: S1, a sensor data blind correction problem is modeled as an HSSM (hybrid state space model); S2, the a posteriori distribution of nonlinear observation signal parameters in the HSSM is obtained through using a UT-FBalgorithm according to relations between various parameters in the HSSM, and the a posteriori distribution of the other parameters in the HSSM is obtained through using Bayes' theorem and Dirichlet process; S3, the IMCMC (Iterative Markov Chain Monte Carlo) sampling method is used to iteratively collect the samples of the a posterior distribution of the parameters; and S4, some initial samples ofa sample set obtained by means of sampling in the S3 are discarded according the nature of the Markov chain, and the average value of an obtained sample set is solved, so that the estimated values ofcorrected parameter gain and offset are obtained. With the method of the invention adopted, the accuracy of sensor correction is improved under the premise of ensuring that the established model is close to the real application scenario of a sensor network.

The invention discloses a high spectral image denoising method combining Bayes layered learning with space-spectrum combined priori; the method comprises the following steps: carrying out three dimensional slide block segmentation for a high spectral image according to a space spectrum correlation and non-local self-similarity thereof, and using relative distance priori based on fused features tonon-locally select a plurality of block data pieces to serve as synergy block data, wherein the selected block data is most similar to to-be-observed block data; using layered priori to build a Bayeslow-rank decomposition model, and realizing synergy block data learning and expression. The mode uses low-rank decomposition to carve statistics characteristics of synergy block data, and combines Dirichlet process mixing Gaussian distribution to express noise statistics characteristics; the method uses a variation Bayes method to solve the model, thus effectively reducing image noises. An existing method cannot simultaneously inhibit a plurality of noises in the high spectral image; the high spectral image denoising method combining Bayes layered learning with space-spectrum combined priori can solve said problems, is accurate in result, and can provide strong analysis basis for follow up applications.

The invention belongs to the technical field of text segmentation, and particularly relates to a text segmentation method based on a layered Dirichlet model, which comprises the following steps: S1, acquiring a news corpus, preprocessing the news corpus to obtain a word segmentation set of the whole news corpus, and performing word frequency statistics on the word segmentation set; S2, putting a result after word frequency statistics into a layered Dirichlet process model for training, and storing the trained layered Dirichlet process model; And S3, obtaining a topic vector of each word in theto-be-segmented text through the trained layered Dirichlet process model, and realizing text segmentation according to the topic vectors. By using the method and the device, the following effects canbe realized: the method enables text segmentation not to depend on manual setting of the number of topics, the topic vectors are automatically generated through the hierarchical Dirichlet process model, and the text segmentation efficiency is improved.

The invention relates to a radio dynamic spectrum access method for multiple users and multiple channels. The method includes the following steps: S1, on the basis of a Dirichlet process, obtaining sub-carrier loading strategies and data packet arrival rates of all virtual users according to ACK/NACK messages broadcast by a cluster head in a cognitive radio network in a unit time; S2, defining a collision probability of a data packet of a target secondary user; S3, obtaining a probability density function obeyed by a data packet transmission delay of a target secondary user; S4, establishing arelationship between the data packet transmission delay and a queue delay of the data packet of the target secondary user, and obtaining total delay characteristics of multimedia transmission; S5, obtaining channel capacity characteristics of the multimedia transmission according to the frequency of data packet retransmission; and S6, constructing a service quality evaluation system by combiningthe capacity characteristics and the time delay characteristics and making an optimal data packet loading mode for the target secondary user by maximizing the service quality. Compared with the prior art, the method of the invention has the advantages of low cost, high practicability and the like.

The invention discloses an image motion deblurring method, which includes a first step of establishing infinite student-t distribution mixed models based on the Dirichlet process as an image gradient distribution model and a point spread function model, and obtaining the number of infinite student-t distribution models automatically based on an observation image; and a second step of taking the image gradient distribution model and the point spread function model respectively as an image a priori model and a point spread function priori model, conducting motion deblurring processing on the image using a maximum posteriori estimation method, and estimating a model parameter using variational Bayesian inference. The adaptive selection of the number of models is realized, the degree of fitting of the image gradient distribution is improved, and the technical effect of accurate image motion deblurring is achieved.

The invention discloses a driving behavior modeling and analysis method based on online car-hailing data, and the method specifically comprises the steps: extracting online car-hailing trajectory data and order origin-destination data, and carrying out the preprocessing, and obtaining a plurality of groups of driving feature sequence data containing speed and acceleration variables; and carrying out driving behavior modeling by adopting an improved nonparametric Bayesian learning model based on a layered Dirichlet process-hidden semi-Markov, and identifying the driving behavior state of the vehicle. Constructing a state space vector to represent the driving style type of the driver, and clustering the driver by adopting an improved K-means clustering algorithm; according to driving behavior state recognition and driving style classification results, driving behavior states of drivers of different styles and types in different traffic operation states can be analyzed. According to the invention, the vehicle driving behavior state is identified and the driver types are divided, so that the analysis of different types of driver driving behaviors in a specific scene is realized.

The invention provides a talent recommendation method and apparatus. The method comprises the steps of obtaining text data from a preset database, wherein the text data at least comprises one of articles, papers and webpage texts; according to release time of the text data, classifying the text data; by adopting a layered Dirichlet process, performing topic extraction processing on the text data corresponding to each type; according to a processing result, obtaining hot topics of a current time period; and recommending authors of the text data corresponding to the hot topics as talents. According to the method and the apparatus, the topic in each time period and the current hot topics can be flexibly and accurately obtained, so that the problem of high extracted topic redundancy or topic omission caused by manual setting of a topic number is avoided, and the accuracy and flexibility of recommending the talents according to the hot topics are improved.

The invention relates to a method for extracting bilingual parallel segments of comparable corpus based on a Dirichlet process, belonging to the technical field of machine learning translation and natural language processing. At first, that subject distribution of the bilingual comparable corpus pair is obtain through the bilingual subject model, Bilingual comparable corpus is segmented randomly by Poisson distribution, and then a theme threshold is set, and the parallel fragment set of comparable corpus is initially screened by the threshold, and finally the matching probability of each parallel fragment is obtained by Dirichlet process, and the final accurate parallel fragment pair is further obtained by Gibbs sampling. Under the same comparable corpus environment, the extraction methodbased on the Dirichlet process of the invention has better effect of obtaining parallel fragment pairs.

An embodiment of the invention provides a method for processing data. The method includes: acquiring a training sample, initiating a Dirichlet process and a Gaussian process, acquiring a maximum Gaussian process parameter corresponding to a maximum posterior likelihood value of Gaussian process parameters and a region partition corresponding to the maximum Gaussian process parameter according to the training sample, and fitting mapping relations of unit structure geometry parameters and corresponding electromagnetic response curve cubic spline coefficients; wherein the training sample comprises the unit structure geometry parameters and the corresponding electromagnetic response curve cubic spline coefficients. During specific application, as long as the unit structure geometry parameters are known, highly close electromagnetic response curve cubic spline coefficients can be estimated, and accordingly corresponding electromagnetic response curves can be known.

The embodiment of the invention provides an artwork recommendation method and device, a readable medium and electronic equipment, and the method comprises the steps: firstly, the learning of the picture style characteristics of artworks is performed, and then the clustering is performed, thereby enabling the artworks with similar styles to be classified, enabling the dimensionality of the artworksto be reduced, and achieving the purpose of solving the problem of the sparsity of data; Then, the characteristic division Dirichlet process model can screen the preferences of the user groups to obtain style labels which can represent the group interests of the user groups, and then the group interests of the user groups are divided in a finer-grained manner according to the style labels; And finally, performing personalized recommendation according to the obtained style label distribution of each potential group and the probability distribution of the user belonging to each potential group,and generating a recommendation list. According to the technical scheme provided by the embodiment of the invention, the prediction accuracy and recall rate can be effectively improved.