87 results about "Gaussian mixture distribution" patented technology

In the invention, the detecting system comprises a video camera and a signal processor. It uses an improved Gaussian mixture distribution model to characterize each pixel in image frame; the characterized feature only uses a brightness feature; if there isn't a moving object (vehicle), the video image is relatively static; each pixel obeys a statistic model along with time variation; when getting a new image frame, the Gaussian mixture distribution model is renewed, if the pixel in current image match the Gaussian mixture distribution model, then determining the pixel is a background point; otherwise, determining the pixel is a foreground point.

The invention discloses a software failure positioning method based on machine learning algorithm to solve the technical problem of low positioning efficiency of existing software failure positioning methods. According to the technical scheme, the method comprises the steps of describing failure distribution possibly existing in an actual program based on Gaussian mixture distribution to enable failure distribution in the program to be more definite; removing redundant test samples with a cluster analysis method based on a Gaussian mixture model, and finding a special test set for a specific failure, so that the adverse effect of redundant use cases on positioning precision is reduced; remodifying a support vector machine model to be adapted to an unbalanced data sample, and finding the nonlinear mapping relation between use case coverage information and an execution result by means of the parallel debugging theory, so that machine learning algorithm is free from the local optimal solution problem caused by uneven samples; finally, designing a virtual test suite, placing the virtual test suite in a well trained model for prediction, obtaining a statement equivocation value ranking result, and conducting failure positioning. In this way, software failure positioning efficiency is improved.

The invention discloses a tracking method of a video frequency moving object target self-adapting window, relating to machine vision and pattern recognition technique. A frame image of a video frequency sequence is read into a postpositional memory zone, and staring position and size information of a tracked object is obtained in the frame, then distribution statistical information of a target signature is extracted to build a gauss mixed model as an object template, a mean vector and a covariance matrix in gauss mixed distribution are used to describe the position and the size of the object, then the next frame image of the video frequency sequence is read into the postpositional memory zone. In a new frame video frequency image, a parameter estimation method is used to obtain object gauss mized model parameters in the current frame in iterative computation and find candidate template similar to the object template, and the final model parameters obtained in iteration are used to update the tracking window for realizing self-adaption of the tracked window. According to the tracking method of the video frequency moving object target self-adapting window, tracking reliability is greatly increased, which is widely applied in the fields of robot technology, vision navigation, automatic supervision, traffic control and the like.

The present invention successfully combines neural-net discriminative feature processing with Gaussian-mixture distribution modeling (GMM). By training one or more neural networks to generate subword probability posteriors, then using transformations of these estimates as the base features for a conventionally-trained Gaussian-mixture based system, substantial error rate reductions may be achieved. The present invention effectively has two acoustic models in tandem—first a neural net and then a GMM. By using a variety of combination schemes available for connectionist models, various systems based upon multiple features streams can be constructed with even greater error rate reductions.

The invention discloses a distributed optical fiber vibration sensing system disturbance event identifying and positioning method. The method includes the steps that according to the position and the number of peaks with signals exceeding a certain threshold value, the number of the peaks occurring within the scope ranging from deltal / 2 in front of each point to deltal / 2 behind each point in an optical fiber is obtained, whether disturbance events exist or not is judged, and backgrounds are extracted with an amplitude limiting one-order inertia filtering method when the disturbance events do not exist; space-time terminals of the disturbance events are obtained according to the backgrounds when the disturbance events exist, fitting is carried out within the area range to obtain a background event probability density function, Gaussian mixture distribution parameters including the specific gravity, the expectations and the covariance matrix of Gaussian contents of the event peaks including the backgrounds in the area are solved through iteration of the EM algorithm, the parameters, the parameters of the background event probability density function and the total number of the event peaks in the area are fed into a classifier, the disturbance types are judged, and the expectations of the Gaussian contents represent the space positions of the events.

The invention relates to a complex background modeling method based on a variable Gaussian mixture number and belongs to the technical field of photoelectric product application. The complex background modeling method comprises the following steps: assuming that each pixel in a video scene is affected by independent Gaussian noise, establishing a background pixel model; according to the pixel model, calculating the whole video scene model; calculating the probability of a certain pixel point value by using a Gaussian mixture model; based on a current pixel point value, a pixel point mean value and a pixel point variance value, calculating the Mahalanobis distance from a current pixel point to a certain Gaussian distribution; updating background model parameters of Gaussian mixture distribution according to a comparison result between the Mahalanobis distance and a judged threshold so as to model a complex background. In the complex background modeling method, through the Gaussian mixture model and a variable Gaussian mixture number updating strategy, a moving target in a complex scene under dynamic disturbance can be effectively detected; tests on visible light sequence images in a street environment prove that the complex background modeling method has good interference resistance and realizes detection of a moving object in complex scenes such as tree branch waggling, shadow existence.

The invention discloses a method for screening a biomarker related to severe oligoasthenospermia. The method comprises carrying out deep mass spectrometry on multiple sperm protein uncoded amino acids of severe oligoasthenospermia through a NanoHPLC-MS / MS mass spectrometry system and an unlabeled quantitative proteomics method, searching the mass spectrometry data through a nonfinite amino acid protein modification analysis method, carrying out multivariable gaussian mixture distribution clustering analysis to identify the non-coding amino acids in the sperm protein group as much as possible, and comparing the non-coding amino acids of the normal and patient sperm protein groups to obtain protein uncoded amino acid sites related to severe oligoasthenospermia, wherein the protein uncoded amino acid sites are used as molecular markers of severe oligoasthenospermia. The method provides novel diagnostic and therapeutic targets for severe oligoasthenospermia.

The invention relates to a segmentation method of medical images, and specifically relates to a segmentation method of viscera and internal blood vessels thereof in a surgical planning system. The method designs a segmentation method of images of liver parenchyma, portal veins and hepatic veins based on minimal supervised classification of the three-dimensional CT images of the viscera and internal tubular tissues thereof. The method applies methods of statistics and spatial information, introduces high credible points, and obtains segmentation results based on arrival time of fast march of grey level. The method assumes that spatially adjacent points belong to a same organ, and performs segmentation calculation on the images by using a classification algorithm based on the above assumption, wherein the segmentation calculation comprises the steps of: acquiring estimated values of parameters of a Gaussian mixture distribution model; selecting the high credible points; and calculating the mean value of the grey level of the images and calculating the arrival time by using a fast marching algorithm to obtain arrival time images of three tissue classifications. According to the invention, good robustness and good anti-noise interference capability are obtained, accumulated errors are eliminated effectively, classification accuracy is improved, and great application value in clinical treatment is achieved.

The invention discloses an urban trunk road travel time estimation method based on variable weight mixed distribution, and belongs to the technical field of intelligent traffic. The method comprises the steps: collecting a trunk road travel time parameter of a trunk road in a target research region, preprocessing and verifying the distribution mode of the trunk road travel time in all time periods, building a Gaussian mixed distribution model with a fixed weight, and determining an optimal component number K; collecting the flow in all flowing direction and a signal control parameter for an intersection and an road segment, and building a mixed distribution weight coefficient Logistic function with K components; building a variable weight mixed distribution model, estimating an unknown parameter in the variable weight mixed distribution model, and finally carrying out the estimation of trunk road travel time distribution and the estimation of reliability service level. Compared with a conventional single-distribution-function and fixed weight mixed distribution model, the method irons out the defects that the transplantability and the adaptability are poor, and can achieve the more accurate estimation and reliability evaluation of the trunk road travel time distribution.

The invention relates to a security risk assessment method of a wind-power-included electric power system based on Gaussian mixture distribution characteristics. The security risk assessment method comprises the following steps: counting up the distribution of historical data of output power of a wind power plant; establishing a non-parameter probability distribution model of the output power of the wind power plant; establishing a Gaussian mixture distribution model of the output power of the wind power plant; determining the quantity of Gaussian distribution and initializing parameters of each Gaussian distribution; solving the parameters of each Gaussian distribution and determining the Gaussian mixture distribution characteristics; determining cumulative distribution functions of state variable node voltage and branch power flow; and calculating state variable threshold-crossing probability and severity of generated results, and comprehensively estimating safety risks of the electric power system. According to the security risk assessment method provided by the invention, a semi-invariant solving process of node injection power of the system is greatly simplified, so that the efficiency of semi-invariant calculation of the node injection power and the accuracy of the cumulative distribution functions of state variable node voltage and the branch power flow are improved, and thus effective data supports are provided for security risk assessment of the electric power system.

The invention discloses a training method of a dialogue generation model and a dialogue generation method and device, and relates to the field of artificial intelligence. The method comprises the following steps: encoding a context sample to obtain a first hidden layer variable, and identifying the first hidden layer variable to obtain a prior hidden layer variable; encoding the reply sample to obtain a second hidden layer variable; encoding the reply similar sample to obtain a third hidden layer variable; according to the Gaussian mixture distribution of the first hidden layer variable, the second hidden layer variable and the third hidden layer variable, identifying to obtain a posterior hidden layer variable; and matching the prior hidden layer variable and the posterior hidden layer variable, and performing adversarial training on the dialogue generation model. According to the method, similar samples in a sample set are adopted, posteriori distribution of a dialogue generation model is fitted through Gaussian mixture distribution of the similar samples, the purpose of fitting more complex semantics is achieved, and complex semantics and high variability of dialogues can be captured.

The invention provides a method for segmenting a moveable outline model image based on edge flow and belongs to the field of image processing. The method comprises the following steps: extracting initial image outline information, namely performing gaussian mixed distribution statistics on the image information, judging a foreground area and a background area by utilizing a corresponding gaussian mixed distribution function of the image information, removing the background area and maintaining the foreground area, so as to obtain the initial image outline information; and evolving the initial image outline information, namely evolving the initial image outline information till an evolving energy function reaches a minimum value by adopting a moveable outline model technique based on the edge flow, so as to obtain final segmented information of the image. The method disclosed by the invention meets the demand on boundary continuity of a segmenting target; the problems of weak image boundaries and unsatisfactory segmenting effect of gray level progressing are solved; and the method disclosed by the invention has the advantages of high segmenting efficiency and high precision, is slightly influenced by noise, so that a segmenting effect with high robustness can be obtained.

The embodiment of the invention provides a training method for an anomaly detection model based on differential privacy. The method comprises the following steps: inputting a first vector of any sample in a training set into an auto-encoding network, outputting a dimension-reduced second vector through an encoder, and outputting a restored third vector through a decoder. Then, constructing an evaluation vector based on the second vector, inputting the evaluation vector into an evaluation network, and obtaining the sub-distribution probability that the sample output by the evaluation network belongs to K sub-Gaussian distributions in the Gaussian mixture distribution; then, according to the evaluation vector and the sub-distribution probability corresponding to each sample in the training set, obtaining a first probability of any sample in Gaussian mixture distribution; and determining the prediction loss which is negatively correlated with the first probability corresponding to each sample and is negatively correlated with the similarity between the first vector and the third vector. Furthermore, noise is added to an original gradient obtained on the basis of prediction loss in a differential privacy mode, and model parameters of the anomaly detection model are adjusted by using a gradient containing the noise.

The invention discloses an indoor positioning method based on particle filtering under the condition of non-Gaussian noises. The indoor positioning method comprises the following steps of modeling movement accelerated speed of an object and measurement noises into random vectors which obey Gaussian mixture distribution in a training stage by using a particle filtering method based on a suboptimum important function; and performing local linearization on a non-linear observation equation in a positioning state so as to obtain a suboptimum important function and a weight coefficient, change a degradation phenomenon in particle filtering and implement optimum estimation on state vectors. The indoor positioning method has the advantages that on one hand, compared with Gaussian noises, modeling of the Gaussian mixture model is close to actual conditions, and errors caused by model approximation can be reduced effectively; and on the other hand, the degradation speed of the weight coefficient in the particle filtering process can be increased through the solved suboptimum important function, the algorithm efficiency and the algorithm precision are improved, and the positioning precision is improved effectively.

The invention discloses a method of improving accuracy of a recommendation system. The method of improving the accuracy of the recommendation system comprises the steps of firstly establishing three data subsets, then marking label-free data in each data subset by respectively applying a scoring model based on Gaussian mixture distribution established by the invention onto the established data subsets to obtain each training set, adding the original label-free data marked in the obtained training sets into labeled data in other subsets, iteratively updating labels of the original label-free data in the other subsets, and finally outputting a final recommendation result. The method enables the recommendation system to be capable of effectively resisting scoring noises, and good robustness is obtained and the recommendation accuracy is improved. Besides, the problem of cold start can be effectively relieved.

The invention discloses a traffic congestion situation analysis system. The traffic congestion situation analysis system includes the following components of: a data acquisition module which is used for acquiring and storing traffic information data in real time; a first analysis module which obtains the traffic information data acquired by the data acquisition module and selects corresponding evaluation indexes according to whether the data are deficient, and calculates real-time evaluation values based on real-time traffic information data, and compares the real-time evaluation values with corresponding evaluation threshold values so as to analyze traffic congestion situations in real time; a second analysis module which obtains all traffic information data which are stored in a set period from the data acquisition module, and establishes a Gauss mixture distribution model of all the traffic information data, and solves estimation values of related parameters of each sub Gauss distribution model, and analyzes traffic flow congestion states corresponding to each sub Gauss distribution model according to the estimation values. According to the traffic congestion situation analysis system of the invention, traffic information data which are acquired in a real-time manner and in a non-real-time manner are analyzed, and therefore, the traffic congestion situation analysis system is advantageous in few analysis variables, low cost and greater practicability.

The invention discloses a random optimization scheduling method considering flexibility for a wind power-containing electric power system. According to the method, the random fluctuation of the net load and the operation characteristics of the source load storage are comprehensively considered, a power system flexibility requirement model is established by utilizing Gaussian mixture distribution,a multi-type flexibility resource supply model is constructed, and then a power system operation flexibility probability balance model is constructed. Risk loss brought to the power system due to insufficient flexibility is measured by utilizing conditional risk values, then system power balance constraints, operation flexibility constraints and equipment operation constraints are considered, a wind power-containing power system random optimization scheduling model considering flexibility is constructed, and system operation economy and flexibility are considered at the same time.

The invention discloses a loan prediction method and device, equipment and a storage medium. The method comprises: acquiring a loan application request of a user; extracting current application feature information in the loan application request; classifying users corresponding to the current application feature information through the Gaussian mixture distribution model; and performing loan passing rate prediction through the preset user loan prediction model according to a classification result and the current application feature information to obtain a prediction result of the loan passing rate. The machine learning-based loan prediction method can predict the default probability of the new customer and quantify the risk of the new customer, thereby helping the loan institution to make a decision about whether the loan is given to the customer or not, and compared with the traditional manual decision, the credit scoring card is higher in accuracy, lower in cost and shorter in consumed time, so that the risk management of the loan institution is benefited.

The invention relates to a rat brain section microscopic image segmentation method based on a markov random field theory. Gaussian mixed distribution is trained through the existing tagged image and used in characteristic field modeling, image characteristics are simulated correctly, great guiding significance for random field modeling is obtained, convergence times of an iterative algorithm are greatly reduced, and the accuracy of a segmentation result is improved. In addition, in order to solve the problem that local neighborhood characteristics of a traditional 8 neighborhood pixel model traced image are too rough, the method introduces a pixel gray value and a distance between pixels into a Potts model, a new potential-energy function is defined, image local information is described correctly, and the accuracy of a segmentation result is improved.

The invention belongs to the field of static random access memory circuit yield analysis in integrated circuit manufacturability design, and particularly relates to a method for obtaining an optimal actual sampling distribution parameter by adopting general Pareto and Gaussian joint distribution as an actual sampling distribution function family and minimizing cross entropy between actual samplingdistribution and ideal sampling distribution. The optimized actual sampling distribution is used for sampling and calculating the SRAM failure rate, so that the number of sampling points can be greatly reduced, and the sampling efficiency is improved. The key point of the invention lies in that general Pareto and Gaussian mixture distribution is adopted as a sampling distribution function family;aiming at the distributed parameter optimization problem, an iteration strategy is provided, sampling is continuously carried out, actual distribution parameters are updated, and the failure rate iscalculated until the failure rate meets the precision requirement. Experimental results show that the method provided by the invention is obviously superior to a method in the prior art.

The present invention discloses improved versatile distribution and versatile mixture distribution models characterizing wind power probability distribution. By selecting an appropriate shape parameter and order, a versatile mixture distribution model is capable of fitting wind power distribution or error distribution in any shape under a precision requirement. A CDF of a distribution function of the model has a closed analytic expression, and an inverse function of the CDF is an implicit function expression, which is applicable to economic dispatch in a wind power integrated system. Compared with fitting performed on actual wind power distribution of an actual wind farm by a Gaussian mixture distribution model, an advantage of a probability distribution model is verified. The method has great promotion value and an excellent application prospect.

The invention provides an intelligent driving automobile obstacle avoidance decision-making method and device. A Gaussian mixture distribution model and a hidden Markov model are models in machine learning and can be automatically executed by a processor and other terminals, and automatic training is performed based on data; compared with rules based on artificial experience and artificial calibration, a confusion matrix and a state transition matrix obtained by the trained Gaussian mixture distribution model and by training of the hidden Markov model are high in accuracy; and therefore, the situation that the accuracy of obstacle avoidance decision making is high on the basis of the confusion matrix and the state transition matrix obtained by the pre-trained Gaussian mixture distributionmodel and by training of the hidden Markov model can be guaranteed, a vehicle can make a correct decision in time, and the driving efficiency and safety are improved.

Through a method in the field of artificial intelligence, a personalized search algorithm with enhanced time information is realized, and on the basis of inputting a user set, two time-aware LSTM structures are designed through an original correlation score obtained by a recurrent neural network to obtain a query intention of a user and a document interest representing the short-term interest of the user. A long-term query intention and a long-term document interest of a user are obtained by utilizing query-based Gaussian mixture distribution taking time as an independent variable; and finally, a neural network model is utilized to consider the original correlation score and the personalized score at the same time to carry out final calculation on the personalized score of the document. Through the above means, the user interests are modeled in the continuous time space, so that more accurate user interest representation is constructed, and the personalized sorting effect is improved.

The invention relates to a g lass furnace temperature forecast method based on a learning machine related to Gaussian mixture distribution, and belongs to the automatic control field, the information technology field, and the advanced manufacturing field. Modeling problems of glass furnace temperature forecast such as complicated glass furnace internal reaction process, complicated asymmetric noises of data, and input variables including time series variables, the glass furnace temperature forecast method based on the learning machine related to robustness in the Gaussian mixture distribution is provided. A kernel function regression model is used as a forecast model structure, and non-zero mean value Gaussian mixture distribution is used as probability density distribution of forecast model residual terms, and the time series variables are parallely arranged, and are used as the input variables of the models, and then a Bayesian inference method is used to acquire approximate posterior probability distribution of model structure parameters, and therefore the structure parameters of the forecast model is acquired. The glass furnace temperature forecast method is effectively used for the forecast of the glass furnace pool bottom temperature, and therefore a glass furnace control and operation optimization effect is improved.

The invention provides a Markov chain model based traffic factor network construction method. According to the method, captured speed and flow data serve as system factors, and a traffic factor network established via higher-order multi-component Markov chain simulates inherent complex space-time relation in the traffic network. Knowledge in the traffic field is applied to modeling, internal connection between traffic data is taken into consideration, and traffic factor values are clustered and analyzed. An EM algorithm based Gaussian mixture distribution model is used to learn historical datareversely, a lot of traffic flow parameter data is clustered and analyzed, and different traffic parameter clusters correspond to different traffic states in the practical traffic system, namely environment influential factor levels. The traffic factor state network is constructed by high order and multiple components on the basis, and thus, more accurate data prediction and correction can be realized.

The embodiment of the invention provides an automatic driving decision planning method, electronic equipment and a computer storage medium, and the method comprises the steps: obtaining the driving perception information of a to-be-decided object in a continuous behavior space, wherein the driving perception information comprises geometric information, historical driving track information and map information related to the to-be-decided object; according to the driving perception information and the driving target information, obtaining a plurality of planning strategies conforming to Gaussian mixture distribution and strategy evaluation corresponding to each planning strategy; and performing decision planning on the to-be-decided object according to the plurality of planning strategies and the strategy evaluation corresponding to each planning strategy. According to the embodiment of the invention, decision planning can be effectively carried out in a strong interaction scene in automatic driving, and the decision effect is improved.

The invention discloses a collection prediction method and device based on forward propagation, equipment and a storage medium, and the method comprises the steps: obtaining the transaction data of a to-be-predicted user, the transaction data comprising the current application data information and the current post-loan repayment information of the user; preprocessing the current application data information to obtain target independent variable feature information of the current application data information; classifying the users to be predicted corresponding to the current post-loan repayment information by adopting a preset Gaussian mixture distribution model, and marking corresponding target tag information on the users to be predicted according to a classification result; and performing collection prediction on the target independent variable feature information and the target label information through a preset deep neural network back propagation collection model to obtain a target collection prediction result, thereby performing collection prediction through the preset deep neural network back propagation collection model, and improving the prediction precision.

The invention discloses a method for automatically extracting a flood risk area based on an SAR image.The method comprises the steps: calculating the water body probability of two time phase SAR images before and after a flood event through employing a C wave band backscattering coefficient sigma0, and expressing a flood risk area through employing the increment of the water body probability. The method is a rapid, convenient and nearly real-time flood monitoring method, can effectively extract the newly added submerging area (flood risk area) before and after the rainfall event, and comprises the following steps: acquiring two time phase images before and after a flood event, and forming a sigma0 set; generating a frequency histogram of sigma0, assuming that sigma0 accords with Gaussian mixture distribution, and then filtering outliers; performing unsupervised classification on the downsampled sigma0 set, and calculating a prior probability; carrying out iterative calculation on the downsampled sigma0 set by utilizing an expectation maximization (EM) algorithm and the prior probability, calculating a distribution curve parameter, and calculating a water body probability based on sigma0 before downsampling; and extracting a flood risk area by using the water body probability increment of the two time phase images.

The invention discloses a multi-sensor distributed data fusion method based on the Chernoff fusion criterion. According to the method, particle filtering is performed on each sensor to obtain local estimation results firstly, the local estimation results are approximated as Gaussian mixture distribution through adoption of the maximum expectation algorithm, Gaussian mixture parameters are interacted among the multiple sensors, initial fusion results of the multiple sensors are obtained through utilization of a Chernoff fusion method in the case of a first-order approximation model, importance sampling is performed on the results, a multivariate optimization function with a constraint is built, the optimization function is solved through adoption of a particle swarm optimization algorithm, index values of the multiple sensors under the Chernoff fusion criterion are obtained through calculation, Chernoff fusion is finally performed by utilizing particle samples and the optimal index weight of the Chernoff fusion, and then an estimation state of a target is obtained through calculation. The method theoretically, approximately and optimally solves the problem that ideal distributed fusion results for the multiple sensors are difficult to obtain in the case that correlation between the local estimation results exists.

The invention discloses a mouse trajectory tracking method based on background modeling, and relates to the technical field of intelligent video monitoring. The method comprises the following steps: extracting data before a specified frame in an original video, and establishing a background model; creating a blank image as an initialized thermodynamic diagram of a mouse track; performing foreground detection on each frame of image after the specified frame, and extracting target frames of all foreground moving targets in the current frame of image; establishing a Gaussian mixture distributionmodel for the target box; and accumulating the Gaussian mixture distribution model on an initialized thermodynamic diagram to generate a thermodynamic diagram of the moving target, namely the track ofthe mouse. When the mouse track is detected, the mouse track detection method is not limited by kitchen scenes, is easy to implement, high in operation efficiency and convenient to develop and deploy, and can greatly solve the hidden danger problem in the catering industry.