318 results about "Posteriori probability" patented technology

A system for the representation, editing, evaluation, and inference of graphical models is disclosed which can be used to construct and evaluate a graphical model or graphical network and to calculate inference values. An efficient method of updating graphical models is demonstrated, and provides the basis for an improved system for manipulation and evaluation of probabilistic models. The graphical network editor is useful in the construction of graphical modes such as Bayesian Networks. The graphical network and network graphical user interface (GUI) are used in conjunction with each other to model a system wherein failure probabilities and the current state of components are taken into account to monitor the health and progress of a system for an engineer or engineering software to evaluate and monitor. The evaluation is useful in the monitoring of assets and other real systems having multiple, dependent, and independently operating components such as a pump, a manufacturing plant, a production line, an assembly line, where asset health and quality control is a concern. The asset components each influencing some overall outcome of a system or situation. Success or failure or probability of success, probability of failure and health of the system can be monitored and manipulated by altering the values of prior probability and posterior probability values. Failure correlation between components can be evaluated wherein failure rates of asset is unknown. Production and quality can be monitored and altered.

The invention discloses a pedestrian recognition method based on combination of depth learning and property learning. According to the invention, a convolution neural network containing five implicit strata is constructed. Network training is performed by an anti-convolution method and a concept of property learning is combined. Preferred features obtained from the convolution neural network are input to property classifiers, so that the posterior probability of the property of a sample is obtained. Then by combining with a property class mapping relation, the posterior probability of the class is obtained, so that the class of the sample can be judged. The method is good in detection recognition performance and intrinsic features of an image can be extracted. Besides, since the property has better semantic expression performance than low-stratum features and due to the insensitivity to light and view angles, the algorithm has a good recognition effect.

Embodiments of a computer-implemented method of calculating estimates of a joint posterior probability of class membership given combinations of attribute values of a pair of attributes are disclosed. The calculating is performed on the basis of data representing a training set of a plurality of instances defined by attribute values for a plurality of attributes together with a class membership outcome. Embodiments of the method comprise calculating first and second estimates of a posterior probability of class membership given attribute values of a first and second attribute of the pair, respectively, and binning the first and second estimates into a respective plurality of first and second probability range bins. Instances of the training set are mapped to combinations of one of each of the first and second pluralities of probability range bins, and on the basis of the mapping, calculating estimates of a joint posterior probability of class membership.

The invention discloses a fault diagnosis method during the industrial process. The method comprises the steps of collecting historical normal data during the industrial process; calculating a detection statistics based on the historical normal data during the industrial process; collecting the to-be-detected data of the industrial process; on the condition that the industrial process is detected to be out of order, extracting a statistic feature based on the relative refactoring contribution method; according to the statistic feature, calculating a conditional probability density function in the fault mode and a conditional probability density function in the normal mode; according to the prior probability and the conditional probability density function, calculating a posterior probability; conducting the fault variable recognition on a current time sample based on the minimum risk Bayesian decision theory; according to a diagnosis result, updating the prior probability for the next time sample and conducting the fault diagnosis and recognition again for the next round. According to the technical scheme of the invention, the major failure variable, the secondary process variable and the normal variable of the current sample are distinguished. Meanwhile, the diagnosis result of the process variable of the previous time sample is applied to the diagnosis of the current sample. Therefore, the pollution effect during the fault diagnosis of the industrial process is eliminated.

The invention discloses an indoor positioning method and system and a positioning platform, and relates to the technical field of positioning of the mobile Internet. The method includes: obtaining prior probability of wireless signal strength detected at reference points according to the wireless signal strength of a positioned terminal and a signal strength probability distribution diagram of the reference points; obtaining posterior probability according to the prior probability and based on a Bayesian decision; and multiplying the posterior probability with a matrix of transition probability of the reference points, and outputting the reference point with the maximal transition probability as a positioning result of the positioned terminal. The method, system and platform build the indoor wireless signal strength probability distribution diagram through a transfer HMM model to realize a universal method for accurate positioning, and by combination of a wireless signal strength acquisition and preprocessing mechanism and a probability distribution matching method based on the transfer HMM model, an indoor accurate positioning service with good experience can be provided to a user.

The invention discloses a device and a method for fusing sensor information. The sensor-information fusion method comprises the steps of adopting a self-adaptive weighted-data fusion algorithm to allocate corresponding weight numbers for induction information from a plurality of sensors, adopting a Kalman filtering algorithm to perform optimal estimation to the induction information of the corresponding weight numbers so as to acquire the locally fused induction information, adopting a D-S theory estimation algorithm to perform interval estimation to the uncertain induction information in the locally fused induction information, adopting a multi-Bayesian estimation algorithm to perform individual association probability distribution for the induction information after interval estimation so as to synthesize a joint posteriori probability distribution function, and utilizing the joint distribution function to output a final fusion value of the induction information. The invention has the advantages of improving the reliability, confidence and efficiency of sensor information fusion.

The invention discloses a Polar code efficient adaptive decoding method based on Gaussian structure. Gaussian judgment is introduced based on the Gaussian structure to carry on path splitting and an effective adaptive decoding algorithm is combined. The method calculates the posteriori probability of each information bit by using the Gaussian structure under a Gaussian channel, and uses the posteriori probability as the decoding reliability of each information bit, and applies the decoding reliability of each information bit to a successive cancellation list (SCL) decoding. If the decoding (decoded as 0 or 1) reliability of a current bit is greater than the reliability of correct decoding, decoding is directly performed or if it is not satisfied, the path splitting is maintained. Decoding is performed by obtaining all effective paths of the current bit after the path splitting based on bit reliability in combination with the effective adaptive decoding algorithm. Numerical simulations show that under the same conditions, the method has lower complexity in the case of maintaining the same performance.

The invention provides a recording device clustering method based on Gaussian mean super vectors and spectral clustering. The method comprises the steps that the Melch frequency cepstrum coefficient MFCC characteristic which characterizes the recording device characteristic is extracted from a speech sample; the MFCC characteristics of all speech samples are used as input, and a common background model UBM is trained through an expectation maximization EM algorithm; the MFCC characteristic of each speech sample is used as input, and UBM parameters are updated through a maximum posteriori probability MAP algorithm to acquire the Gaussian mixture model GMM of each speech sample; the mean vector of all Gaussian components of each GMM is spliced in turn to form a Gaussian mean super vector; a spectral clustering algorithm is used to cluster the Gaussian mean super vectors of all speech samples; the number of recording devices is estimated; and the speech samples of the same recording device are merged. According to the invention, the speech samples collected by the same recording device can be found out without knowing the prior knowledge of the type, the number and the like of the recording devices, and the application scope of the method is wide.

The invention discloses a coupled numerical meteorological and hydrological aggregate forecasting reservoir scheduling risk decision method comprising the steps that the numerical meteorological and hydrological aggregate forecasting model of the reservoir basin is established, and the flood process of the basin is forecasted in a rolling way; the uncertainty of hydrological aggregate forecastingis assessed by using the Bayes model averaging method, the Bayes posteriori probability is reckoned and the probability of the flood scene in the future is updated in real time; a flood scene tree isconstructed by using the method based on the probability distance on the basis of the hydrological aggregate forecasting result and branch cutting of the flood scene tree is performed; and a reservoiroptimal scheduling random change constrained programming model is established, the optimal scheduling decision of the reservoir is solved by using the optimization method and the decision risk is assessed. With application of the method, the forecast period of hydrological forecasting can be effectively extended and the forecast precision can be enhanced, and the uncertainty between modes and different data assimilation schemes is comprehensively considered; and the method is suitable for medium-and-short-term real-time reservoir scheduling and can significantly enhance the reliability of thereservoir scheduling decision.

The invention discloses a layered integrated Gaussian process regression soft measurement modeling method used for a complex changeable multi-stage chemical process. The layered integrated Gaussian process regression soft measurement modeling method is an on-line multi-model strategy. A Gaussian mixture model is employed to identify different stages of the process, principal component analysis is carried out on data in each stage, on the basis of the contribution degree of each auxiliary variable in the principal element space, data in each mode is divided into several subspaces, and a corresponding Gaussian process regression soft measurement model is established. When new data comes around, variable selection is carried out by means of subspace PCA, and on the basis of the soft measurement model which is established off line, the prediction output of each model can be obtained. By carrying out mean value fusion on outputs of subspace models, first layer integrated output, i.e., local prediction output in each mode can be obtained, finally new data obtained according to calculation is attached to the posterior probability of each different stage, and local prediction in each mode is fused by means of the posterior probability to obtain second layer integrated output. Key variables can be accurately predicted, and therefore the product quality is improved, and the production cost is reduced.

The invention discloses an integrated semi-supervised Fisher's discrimination-based industrial process fault classifying method. In the method, offline modeling is first conducted; non-labeled data is randomly sampled and together with labeled data form a plurality of random training subsets; then semi-supervised Fisher dimensionality reduction is conducted to acquire a plurality of Fisher's discrimination matrixes; sampled data with dimensionality reduction is operated according to a Bayesian statistics method to acquire a series of posterior probability matrixes; the posterior probability matrixes of the labeled data and corresponding labels work as training samples adjacent to a measurement layer fusion algorithm K; during online classification, above semi-supervised Fisher's discrimination classifiers are called to acquire a posterior probability matrix of each online to-be-measured sample; and then the posterior probability matrix is input to a measurement layer fusion K adjacent classifier to acquire a final fault classification result. Compared with other methods, industrial process fault classification effect can be improved, knowledge and operation confidence to the process can be enhanced for operators and automatic implantation of the industrial process can be facilitated.

Businesses typically have large amounts of data about customer transactions and other customer information which is not fully utilized. The present invention provides a means of using this information to make predictions about future customer behavior, for example by estimating the probability that a customer will leave a bank. Using these predictions the business is able to take action in order to improve its performance. Using customer data a Bayesian statistical model is generated and this model used to generate statistical estimators of customer behavior. The statistical model is formed using hidden Markov model techniques by clustering customer data and attributes (e.g. Age, sex, salary) into a finite number of states. The number of states is unobserved and considered random. Bayesian prior probability distributions are specified and combined with the data to produce Bayesian posterior probability distributions. Using these Bayesian posterior probability distributions the statistical estimators are obtained. For example, Monte Carlo sampling techniques are used or alternatively the posterior distributions are calculated numerically or analytically.

The invention relates to a method for tracking multi-target vehicles by adopting the MCMC (Markov Chain Monte Carlo) algorithm, comprising the following steps: modeling the multi-target vehicle tracking process by adopting the MCMC algorithm, establishing various preselection states, traversing the preselection states by using the Metropolis-Hastings sampling based simulated annealing algorithm, and selecting the preselection state with the maximum connection probability as the optimal solution. In the method, a limit is set to the transfer equivalence for the first time, and the parameters in the data association are estimated by the experimental data regression fitting, thus the optimal vehicle motion trajectory with maximum posteriori probability significance is obtained, and the vehicles are tracked. The invention solves the problems of frequent shielding splitting of the multi-target vehicles and has the advantages of high multi-target vehicle tracking precision and good real-time performance.

The application discloses a wakeup word detection method, device and equipment based on artificial intelligence, and a storage medium thereof. The method comprises the following steps of acquiring to-be-identified voice data, and extracting voice characteristics of each voice frame in the to-be-identified voice data; inputting the voice characteristics into a preconstructed deep neural network model, wherein the output voice characteristics are corresponding to posterior probability vectors of syllable identification, and the deep neural network model comprises syllable output units, and the number of the syllable output units is same as that of syllable of a preconstructed pronunciation dictionary; determining a target probability vector from the posterior probability vectors according toa syllable combination sequence, wherein the syllable combination sequence is constructed based on the input wakeup word text; and calculating the credibility according to the target probability vector, and determining that the voice frame comprises the wakeup word text when the credibility is greater than a threshold. Through the scheme provided by the embodiment of the application, the calculation complexity is low, the response speed is fast, an operation of performing special optimization improvement on the fixed wakeup word is avoided, and the wakeup detection efficiency is effectively improved.

The invention provides a method and system for judging the crowd density in an image. The method comprises the following steps: 1. selecting a target region from a video image sample acquired in an image acquisition device by utilizing a block analysis unit, and carrying out block analysis of image blocks in the target region; 2. determining the composite form of two-classifiers by a coding unit;3. selecting a confidence training sample by a training unit, and respectively training each two-classifier; and 4. obtaining the crowd density grade category of the maximum posteriori probability through a decoding unit by means of a channel transmission model. The method can be suitable for obtaining the credible crowd density grade in different scenes and can provide the basis for crowd monitoring and safety guarantee of important regions.

A method for tracking pedestrians in a video sequence, where each image frame of the video sequence corresponds to a time step, includes using marginal space learning to sample a prior probability distribution p(xt|Zt−1) of multi-person identity assignments given a set of feature measurements from all previous image frames, using marginal space learning to estimate an observation likelihood distribution p(zt|xt) of the set of features given a set of multi-person identity assignments sampled from the prior probability distribution, calculating a posterior probability distribution p(xt|Zt) from the observation likelihood distribution p(zt|xt) and the prior probability distribution p(xt|Zt−1), and using marginal space learning to estimate the prior probability distribution p(xt+1|Zt) for a next image frame given the posterior probability distribution p(xt|Zt) and a probability p(xt+1|xt), where the posterior probability distribution of multi-person identity assignments corresponds to a set of pedestrian detection hypotheses for the video sequence.

The invention discloses a multi-language speech recognition method based on language type and speech content collaborative classification. The method comprises the following steps: 1) establishing andtraining a language type and speech content collaborative classification acoustic model; wherein the acoustic model is fused with a language feature vector containing language related information, and model adaptive optimization can be performed on a phoneme classification layer of a specific language by utilizing the language feature vector in a multi-language recognition process; 2) inputting aspeech feature sequence to be recognized into the trained language type and speech content collaborative classification acoustic model, and outputting phoneme posteriori probability distribution corresponding to the feature sequence; the decoder generating a plurality of candidate word sequences and acoustic model scores corresponding to the candidate word sequences in combination with the sequence phoneme posteriori probability distribution of the features; and 3) combining the acoustic model scores and the language model scores of the candidate word sequences to serve as an overall score, and taking the candidate word sequence with the highest overall score as a recognition result of the voice content of the specific language.

The invention discloses a grid model based PDR indoor positioning method and system. The grid model based PDR indoor positioning method comprises the following steps that 1, a grid model of a positioning area is established; the movement trend of a positioning target is obtained; the grid prior probabilities are calculated based on the position information of a target to be positioned at a previous moment and the movement trend at the current moment; the grid prior probabilities are screened according to the semantics and topological characteristics of the grid model to obtain posteriori probabilities of to-be-positioned target position candidate cells, and accordingly positioning is achieved according to the position information, re-calculated based on the grid model, of the target to be positioned at the t moment and the movement trend, re-calculated based on the grid model M, of the target to be positioned at the t moment. The technical problem that accurate positioning and further calculation cannot be achieved in the prior art is solved, and the technical effects of accurate positioning and further calculation are achieved.

The invention discloses a selective hierarchical integration Gaussian process regression soft measurement modeling method based on evolutionary multi-objective optimization. The method comprises the following steps: firstly, constructing a diversity input variable subset by combining Bootstrapping random resampling and partial mutual information; dividing the corresponding original sample subset into different modeling areas by using a Gaussian mixture model algorithm, establishing a corresponding Gaussian process regression base model, carrying out posteriori probability weighted fusion, constructing a first layer of integrated model EGPR, constructing a multi-objective optimization problem from the perspective of evolutionary optimization, and selecting an EGPR model which is good in performance and meets diversity for final integration. The diversity of sample information and input variable information is fully considered, and the diversity and prediction precision of the base modelcan be effectively ensured. Moreover, due to the introduction of the selective integration strategy, the defect that all local models are fused through traditional integrated learning is effectivelyovercome, the complexity of integrated modeling is remarkably reduced, and the model prediction performance is improved.

A method of modelling data, comprising: training an objective function of a linear classifier, based on a set of labeled data, to derive a set of classifier weights; defining a posterior probability distribution on the set of classifier weights of the linear classifier; approximating a marginalized loss function for an autoencoder as a Bregman divergence, based on the posterior probability distribution on the set of classifier weights learned from the linear classifier; and classifying unlabeled data using the autoencoder according to the marginalized loss function.

The invention discloses a method for automatically and accurately measuring the abrasion loss of a cutter based on a gray image probability. The method is characterized in that a method for solving the missing original boundary of a cutting edge based on Bayesian inference is provided for solving the problem of accurate measurement of the abrasion loss of the cutter caused by missing of the original boundary of the cutting edge. A likelihood function of a tool wear boundary and a priori probability model of an original boundary are established by analyzing the change rule of the gray value ofthe cutting edge of the milling cutter, and the missing original boundary is reconstructed by maximizing the posteriori probability of the original boundary of the tool. The area of a worn area is obtained by using a region growing method to realize accurate calculation of the cutter wear extent. According to the invention, automatic and accurate measurement of the cutter abrasion loss is realized, the measurement precision and efficiency are improved, and an effective solution is provided for the marking problem of a large amount of data in cutter state monitoring.

The invention provides a combined Gaussian model radar target steady recognition method based on noise apriority, and the method is mainly used for solving the problem that a statistical model in an existing radar plane target recognition technology is unsteady in noise. The method is implemented through the following steps: preprocessing HRRP (High Resolution Range Profile) data of a radar; performing impression taking on the processed data so as to obtain time domain features of the preprocessed data; determining the average value, loading matrix and noise covariance matrix of each frame ofcombined Gaussian model of HRRP data of radar training targets; accounting the noise variance in the non-signal supporting region of HRRP data in a radar test and calculating the average value and the modified value of the noise covariance matrix of the combined Gaussian model according to the noise variance; calculating the posteriori probability values of time domain features, corresponding to each radar test target, of the HRRP data of the radar test targets; and determining the type attribute of the HRRP data of the radar test targets. The method has the advantage of being steady in noise, and can be used for the steady recognition of radar and plane targets.

The invention discloses a spectrogram abnormal sample point detection method based on a random sampling agree set. The spectrogram abnormal sample point detection method comprises the following steps of: eliminating part of an abnormal sample in advance so as to obtain a correcting sample set through principal component analysis on the base of a maximum posteriori probability random sampling agree set and starting from a given spectroscopic data, carrying out random sampling, establishing multivariated correcting model and evaluating a model property, and selecting an appropriate sample subset as an inner point set through random sampling for many times. The designed spectrogram abnormal sample point detection method based on the random sampling agree set, provided by the invention, has the advantages of being rapid, effective, high in accuracy and wide in application range.

The invention discloses a personalized recommendation method based on a knowledge graph convolution algorithm, and belongs to the field of hybrid recommendation systems. The method comprises the following steps: firstly, constructing a 1-to-d-hop entity receptive field set of each article in a domain knowledge graph, aggregating embedded vectors of d-to-1-hop receptive field entities by using a graph convolution algorithm, and calculating an article embedded vector containing high-order neighbor information; in the aggregation process, on the basis of the graph convolution algorithm, using a knowledge graph for representing a learning model DistMult and representing an attention mechanism of user interest distribution to improve the embedded vector expression ability; by referring to a matrix decomposition model, multiplying a high-order article embedding vector generated by iteration with a user embedding vector, and outputting a predicted interaction probability by using a Sigmoid function; and inferring a combined loss function of a hybrid user article interaction matrix loss function and a knowledge graph representation learning loss function by using maximum posteriori probability estimation. According to the method, the high-order feature representation of an article embedding vector is enriched, and the expression accuracy of the entity and relationship embedding vectoris enhanced.

The invention discloses a method for diagnosing a fault of an oil-immersed transformer on the basis of a rough set and a bayesian network. The method comprises the following steps that (a) the type of the fault is determined, as much as possible input fault characteristic vectors are selected in an original sample set, and an input attribute set is determined; (b) discretization processing is carried out on a fault data set through a data discretization method in the rough set theory, and a discretization decision table is established; (c) establishment of the bayesian network is carried out through Matlab; (d) a conditional probability table is initialized, wherein all the possible conditional probabilities of each node relative to the father node of the node and the quantitative description of the corresponding problem domain are listed in the conditional probability table; (e) parameter learning is carried out, and a deduction engine is established to carry out deduction after the bayesian network is established; (f) a test sample set is input, the posterior probability is solved, and the type of the fault is judged. The method for the oil-immersed transformer on the basis of the rough set and the bayesian network can simplify the scale of a diagnosis network, enhance the anti-interference performance of the network, diagnose various faults of the transformer rapidly, and reduce the outage rate of the transformer greatly.

A probabilistic digital signal processor using data from multiple instruments is described. In one example, a digital signal processor is integrated into a biomedical device. The processor is configured to: use a dynamic state-space model configured with a physiological model of a body system to provide a prior probability distribution function; receive sensor data input from at least two data sources; and iteratively use a probabilistic updater to integrate the sensor data as a fused data set and generate a posterior probability distribution function using all of: (1) the fused data set; (2) an application of Bayesian probability; and (3) the prior probability distribution function. The processor further generates an output of a biomedical state using the posterior probability function.

The invention discloses a high-spectral image sharpening method based on probability matrix decomposition, and belongs to the field of remote sensing image processing. The method is characterized in that the method comprises the steps: carrying out the preprocessing of two inputted images based on the hypothesis that a pixel spectrum vector of a high-resolution high-spectral image is just formed by the linear superposition of a few of vectors with the hidden spectrum features according to one low-resolution high-spectral image and one high-resolution high-spectral image and the frequency response matrix, decomposition matrix dimensions and algorithm iteration number, corresponding to the high-spectral images, of a multispectral camera, wherein the low-resolution high-spectral image and the high-resolution high-spectral image are taken at the same height in the same target region at the same time; listing mathematical equations of the two processed images and a to-be-solved high-resolution high-spectral image, and building a Bayesian model; calculating the posteriori probability distribution of the decomposition matrix, and obtaining a matrix with the hidden spectrum features in the decomposition matrix, and solving the mean value of linear superposition vectors corresponding to the two images after preprocessing, thereby obtaining the to-be-solved high-resolution high-spectral image. The method greatly reduces the time consumption of calculation while improving the sharpening precision, and is easy to adjust.

The invention discloses a novel optic disc separation method and system. The method comprises the steps that S100, an angular point detection algorithm is utilized to detect points of interest aroundan optic disc in an eye fundus image, and convex hulls surrounding the points of interest are calculated to extract an optic disc region image; S200, according to a feature similarity between pixel points of the optic disc region image, the pixel points are grouped, and super pixels capable of replacing a large quantity of pixel expression image features are obtained; S300, prior probability distribution is calculated based on the convex hulls and the super pixels, color histograms are subjected to statistical analysis inside and outside the convex hulls respectively, and an observation likelihood probability is calculated; S400, the posterior probability that each pixel point f belongs to an optic disc region is calculated according to the Bayesian theorem, and a posterior probability distribution diagram is obtained; and S500, the optic disc is separated from the eye fundus image based on the posterior probability distribution diagram and through standard Hough Transformation circledetection. By the adoption of the separation strategy from rough to fine, precise separation of the optic disc from the eye fundus image is realized under a Bayesian model framework.

Encoding-system includes a check-bit-generation-unit configured to generate N-ary-parity-bits by processing information composed of N-ary-symbols (where N is a power of 2) in modulo-N with a LDPC-matrix composed of binary elements, an encoded-sequence-generation-unit configured to generate an encoded sequence including the information composed of the N-ary-symbols and the N-ary-parity-bits, a modulation unit configured to modulate the encoded sequence in a modulation scheme having N-ary-modulation-symbols to produce a modulated signal, a demodulation-unit configured to demodulate the modulated signal to produce a demodulated signal, a metric-generation-unit configured to generate a metric for each of N-modulation-signal-points from the demodulated signal to obtain a plurality of metrics, and a decoding-unit configured to decode the encoded sequence by obtaining posteriori probabilities of the symbols in accordance with a state transition having an N-state defined by a binary LDPC-matrix corresponding to the LDPC-encoder encoding the encoded sequence, on the basis of the metrics.

The invention discloses an on-line recognizing method for nonlinear dynamic model parameters of a high-speed train. The method comprises the following realization steps that a corresponding system model is built according to a practical high-speed train system, the noise interference possibly suffered by a system is analyzed, the model parameters under the interference are obtained, in addition, the weights of each group of parameters are initialized, the system state is estimated by a particle filter algorithm under the condition of each group of parameters, the current observation output of the system is obtained, the posterior probability of each group of parameters is calculated through a probability density function of observation noise under the condition that the output value is known, the weights of each group of parameters are updated, and along with the system operation, the recognizing result is obtained according to the weights of each group of parameters. The on-line recognizing method has the advantages that the calculation quantity of parameter recognition is reduced, the recognition accuracy is improved, and the system parameters can be recognized in real time.