30 results about "Bayes' rule" patented technology

A method for automatically detecting and tracking multiple targets in a multi-camera surveillance zone and system thereof. In each camera view of the system only a simple object detection algorithm is needed. The detection results from multiple cameras are fused into a posterior distribution, named TDP, based on the Bayesian rule. This TDP distribution represents a likelihood of presence of some moving targets on the ground plane. To properly handle the tracking of multiple moving targets with time, a sample-based framework which combines Markov Chain Monte Carlo (MCMC), Sequential Monte Carlo (SMC), and Mean-Shift Clustering, is provided. The detection and tracking accuracy is evaluated by both synthesized videos and real videos. The experimental results show that this method and system can accurately track a varying number of targets.

A probabilistic method for classifying observed CSEM response for a resistive anomaly to classify the response into multiple geologic categories indicative of hydrocarbon production potential. Each category is assigned a prior probability (301). For each category, conditional joint probability distributions for observed CSEM data in the anomaly region are constructed (303) from rock property probability distributions (302) and a quantitative relationship between rock / fluid properties and the CSEM data (304). Then, the joint probability distributions and prior probabilities for each category (305) are combined with observed data (307) and used in Bayes' Rule (306) to update the prior category probabilities (308). Seismic data may be used to supplement CSEM data in the method.

According to some embodiments, system and methods are provided comprising: receiving data; providing a simulation model for the data; generating one or more simulations via a Bayesian module based on the data, wherein the simulation includes one or more nodes in a chain; executing the Bayesian module to determine the acceptability of the nodes in the simulation based on a Bayesian rule, wherein execution of the Bayesian module further comprises: generating a binary decision tree representing the chain in the simulation, wherein the chain includes one or more nodes; prioritizing which nodes in the tree to simulate; generating one or more simulations; executing the simulation model with data associated with the prioritized nodes in the tree in parallel to determine a posterior probability for each prioritized node; and determining whether each prioritized node is accepted or rejected based on the posterior probabilities. Numerous other aspects are provided.

The present invention proposes a 3D-HEVC fast depth coding method based on Bayesian decision theorem, the steps of which are: firstly determine the depth map tree block, and analyze the inter-frame pattern distribution between different quantization parameters; on the one hand, use Bayesian Calculating the posterior probability of the depth map tree block according to the Sri Lankan rule, and judging whether the interaction mode of the depth map tree block is the best. It is determined that the best coding mode of the depth map tree block is the SKIP / Merge mode; on the other hand, it is calculated according to the Bayesian rule The Bayesian cost of the depth map tree block, according to whether the Bayesian cost is less than the tolerance parameter, early determines the termination of the CU pruning of the depth map tree block, determines the best coding mode, or uses the 3D‑HEVC encoder to split the depth map tree block CU, test other coding modes of the depth map tree block to find the best coding mode. The present invention avoids testing other inter-frame modes of depth map coding by introducing the Bayesian rule, not only ensuring the quality of similar virtual viewpoints, but also significantly reducing the complexity of 3D-HEVC depth coding.

The invention discloses a reverse recognition algorithm of a pollution source in an unsteady state flow field based on a Markov chain, which belongs to the technical field of indoor pollutant control, and aims to solve the problem of pollution source recognition in multiregion buildings in an unsteady state flow field. The reverse recognition algorithm comprises the following steps: determining a position value range of a pollution source; calculating a pollutant transfer probability matrix in a building in an unsteady state flow field; operating the transfer probability matrix and a potential pollution source matrix to obtain a responsive matrix; performing inversion by a regularization method to obtain a hourly release rate of a certain pollution source; solving hourly release rates corresponding to different positions of pollution sources in a set; distributing prior probability to pollution source position information; calculating a likelihood function by means of the transfer probability matrix; calculating posterior probability by means of Bayes rule, thereby obtaining the pollution source position and the release rate thereof with the maximum posterior probability. Compared with numerical value calculation, the reverse recognition algorithm has the advantages that the calculating speed is increased, and reversal calculation of the positions and release rates of pollution sources in multiregion buildings in a unsteady state flow field can be realized.

In order to realize effective tracking of multi-target states in complex scenes under the condition of limited resources, the present invention provides a multi-target tracking method based on random set theory. In the track prediction stage, the method first predicts each surviving track in one step according to the Bayesian criterion, and then adapts the birth target track according to the target prior information; in the track update stage, first calculates Corresponding existence weight probability and joint multi-target probability density function, and then calculate the posterior probability parameters of each target label, including existence probability and corresponding probability density function. This method has the advantages of low approximation cost, excellent performance under limited resources, and robustness applicable to any measurement model. In addition, since the algorithm of the invention occupies less system computing resources, it has good engineering application prospects.

A soft measurement method of industrial course under small sample condition includes using support vector computer to structure I / O mapping relation, measurable variables, object control input and object measurable output variable and utilizing them as input variables of soft measurement model, using optimum estimation of estimated variable as output, using genetic algorithm to confirm optimum parameter of support vector computer and applying Bayes rule to speed up convergence velocity of genetic algorithm.

The invention discloses a multi-unmanned aerial vehicle collaborative multi-moving object search search method based on Gaussian distribution prediction. The multi-unmanned aerial vehicle collaborative multi-moving object search method based on Gaussian distribution prediction is characterized by comprising the steps that 1, a task area R is represented by a search probability graph; 2, the Bayes rule is used to update the posterior probability; 3, the objective positions of moving objects are predicted through the application of the Gaussian distribution, and the search probability graph is updated; 4, a distribution model prediction control method is used for constructing the search model of the multi-unmanned vehicle collaborative multi-moving objects search; 5, unmanned aerial vehicles solve a decision information input sequence and obtain heading deflection angles and fly according to the heading deflection angle when the constraint condition is met; 6, a time step is progressively increased, the step 2, the step 3, the step 5 and the step 6 are executed repetitively until the time step exceeds the total time step of search of the unmanned aerial vehicles, and then searching is over. The multi-unmanned aerial vehicle collaborative search method based on the Gaussian distribution prediction can achieve the task of searching multiple moving objects by multiple unmanned aerial vehicles in a collaborative mode, improve the accuracy of searching of the unmanned aerial vehicles, and ensure the accomplished accuracy of the searching task.

The invention disclose a bit error rate (BER) testing system based on the Bayes rule, which consists of a clock unit, an m sequencer, an information output unit, an information receiving unit, a sequence synchronizing and threshold detecting unit, a bit error statistical unit, a BER estimating unit, a bit synchronous clock extracting unit and a local m sequencer. A BER estimating method based on the Bayes rule estimates the range of the BER by utilizing the statistical confidence level principle, simultaneously, posterior probability distribution is adopted to substitute normal distribution so as to be similar to a statistical model of a true BER, the posterior probability distribution of the bit error rate is calculated in real time so as to estimate the BER effectively and accurately according to the actually tested bit error points, and the optimal testing time is selected. The system meets the requirement that the BER of a communication system is estimated by the estimated confidence level in a possibly short time in the practical engineering application of a full optical fiber digital clinometer.

The invention discloses a target tracking method based on progressive unscented Kalman. The method considers the truncation error problem in a measurement linearization process and puts forwards a progressive unscented Kalman filter algorithm to better handle linearization error and numerical calculation error increasing problems existing in the measurement linearization process. Firstly, a homotopy function for a system state is constructed according to a Bayes rule; through carrying out iteration on a measurement updating process, the current observation information is introduced progressively, and further, the posterior state of the system is obtained; and besides, an innovation judgment condition is introduced during the progressive process, the estimated errors of the system during the progressive process are ensured to be bounded, and the stability of the filter is ensured. Compared with the existing target tracking method, the method thoroughly considers linearization error influences, and on the premise of ensuring the calculation complexity, the target tracking precision and the robustness are improved.

The invention discloses a new quantum state transfer method based on Hamiltonian learning, which comprises the steps of preparing the input state by the sender and performing forward unitary evolution on it; the sender sends the input state after forward evolution to the Receiver; the receiver performs reverse unitary evolution of the received evolution quantum state and performs likelihood evaluation; the receiver uses Bayesian rule to update and resample the probability distribution of the measurement output data; the receiver updates and resamples according to As a result, it is judged whether the input state exists and is perfectly transferred to the receiver with probability 1 at a certain moment. The method of the present invention can be evaluated by the likelihood function and the mean square error, the likelihood function is the fidelity of the quantum state transfer, and the mean square error is the Bayesian loss, which is exponential with the measurement times of the interactive quantum likelihood evaluation experiment decline. Therefore, the method of the invention has the characteristics of strong confidentiality, high fidelity and small loss.

The invention discloses an LDA-based Chinese question mapping method, which includes using the LDA topic model to classify the document base, and then using the Softmax regression model to classify the part-of-speech of the question, and according to the difference of the part-of-speech classification, the weight of the content word is higher than that of the function word High, but the weights of different parts of speech in content words are not the same, and then use the syntactic analysis based on dependency grammar to find out the dependency relationship of words in the sentence, and give different weights according to the different components of words in the sentence, so the problem The weight of each word in is obtained by the product of two parts, and finally, according to Bayesian rules, the connection is established through the weighted distribution of words in the question and the distribution of topics and terms in the document. The topic model based on LDA classifies the documents, and at the same time, assigns different weights by referring to the part of speech of the terms in the question sentence and the components in the sentence, so as to improve the role of important terms in classification and improve the mapping of Chinese questions. accuracy.

The invention discloses an image restoration method and system based on multiple regularization of an equation structure. The equation structure is introduced to construct a measured Gaussian likelihood function; based on the constructed measured Gaussian likelihood function, a plurality of sparse transformations are selected, and the sparse transformation coefficient is On the basis, multiple independent probability density functions are given to calculate the conditional prior probability density of the target image; based on the conditional prior probability density of the target image, the mean estimation model of the target image is calculated according to the Bayes rule; Combined with conjugate gradient method, iteratively realizes fast reconstruction of target image, overcomes the invertibility limitation of sparse representation introduced by synthesis method, and allows the use of any number of sparse transformations, and the joint sparse domain can effectively improve the convergence speed of the algorithm.

The invention is applicable to the field of multi-sensor information fusion, and provides a target tracking method and a tracking system for nonlinear Gaussian systems. The steps of the method are as follows: first, according to the marginal distribution and the existence probability of each target at the previous moment, predict the marginal distribution and the existence probability of each target at the current moment, and specify the corresponding marginal distribution and the existence probability for the new target at the current moment . Combined with the position measurement at the current moment, Bayesian rule is used to obtain the updated marginal distribution and existence probability of each target at the current moment, and then the updated marginal distribution related to each target is cut and merged to obtain the marginal distribution of each target at the current moment and its existence probability, and finally cut out the edge distributions whose existence probability is less than the first threshold, and use the remaining edge distributions and their existence probabilities as the recursive input at the next moment, and at the same time, extract the edges whose existence probability is greater than the second threshold distribution as the output at the current moment.