85 results about "Mixture modeling" patented technology

Disclosed is a method and apparatus for building a domain-specific language model for use in language processing applications, e.g., speech recognition. A reference language model is generated based on a relatively small seed corpus containing linguistic units relevant to the domain. An external corpus containing a large number of linguistic units is accessed. Using the reference language model, linguistic units which have a sufficient degree of relevance to the domain are extracted from the external corpus. The reference language model is then updated based on the seed corpus and the extracted linguistic units. The process may be repeated iteratively until the language model is of satisfactory quality. The language building technique may be further enhanced by combining it with mixture modeling or class-based modeling.

A system and method for performing modeling, prediction, optimization, and control, including an enterprise wide framework for constructing modeling, optimization, and control solutions. The framework includes a plurality of base classes that may be used to create primitive software objects. These objects may then be combined to create optimization and / or control solutions. The distributed event-driven component architecture allows much greater flexibility and power in creating, deploying, and modifying modeling, optimization and control solutions. The system also includes various techniques for performing improved modeling, optimization, and control, as well as improved scheduling and control. For example, the system may include a combination of batch and continuous processing frameworks, and a unified hybrid modeling framework which allows encapsulation and composition of different model types, such as first principles models and empirical models. The system further includes an integrated process scheduling solution referred to as process coordinator that seamlessly incorporates the capabilities of advanced control and execution into a real time event triggered optimal scheduling solution.

Automatic selection of training images is enhanced using an instance-weighted mixture modeling framework called ARTEMIS. An optimization algorithm is derived that in addition to mixture parameter estimation learns instance-weights, essentially adapting to the noise associated with each example. The mechanism of hypothetical local mapping is evoked so that data in diverse mathematical forms or modalities can be cohesively treated as the system maintains tractability in optimization. Training examples are selected from top-ranked images of a likelihood-based image ranking. Experiments indicate that ARTEMIS exhibits higher resilience to noise than several baselines for large training data collection. The performance of ARTEMIS-trained image annotation system is comparable to using manually curated datasets.

The invention discloses a suspicious target detection tracking and recognition method based on dual-camera cooperation, and belongs to the technical field of video image processing. The method comprises the steps that a panoramic surveillance camera is utilized for collecting a panoramic image, the improved Gaussian mixture modeling method is adopted for carrying out foreground detection, basic parameters of moving targets are extracted, a Kalman filter is utilized for estimating a movement locus of a specific target, the specific target is recognized according to velocity analysis, the dual-camera cooperation strategy is adopted, a feature tracking camera is controlled to carry out feature tracking on the moving targets, a suspicious target is locked, the face of the suspicious target is detected, face recognition is carried out, face data are compared with a database, and an alarm is given if abnormities exist. According to the suspicious target detection tracking and recognition method, the dual-camera cooperation tracking surveillance strategy is adopted, defects of a single surveillance camera on a specific scene are overcome, and the added face recognition function can assist workers in identifying the specific target to a greater degree; in addition, the tracking algorithm adopted in the method is good in real-time performance, target recognition and judgment standards are simple and reliable, and the operation process is fast and accurate.

A method and apparatus for determining the best fit of a population mixture model to data. In the digital imaging area, the use of histogram data is employed. A plurality of sub-population functions are defined and then optimized to fit the data. An objective function is employed, which is based upon the parameters of the underlying functions. The number of underlying functions is added to the parameter mix, such that no a priori knowledge of the number of sub-populations is required. In an illustrative embodiment, a genetic algorithm is used to evolve the objective function to an optimal fit of the data. Once an optimal fit is found, through comparison with stopping criteria in a fitness function, the data is segmented according to threshold determined based of classification error in the data.

The invention discloses a method and system for tracking targets in a video based on the PTZ. The method comprises the following steps: 100, utilizing a Gaussian mixture modeling method for setting up a background model for a video image obtained by a camera in a preset position scene, and detecting the motion targets in a monitored video image immediately obtained by the camera in the preset position scene; 200, calculating a grey level histogram of the motion targets of the monitored video image, recording template information of the motion targets, and updating parameter information of the motion targets according to the video image immediately obtained by the camera; 300, conducting judgment according to the parameter information and controlling the camera to conduct the corresponding adjustment according to the judged result. The method and system for tracking the targets in the video based on the PTZ can initially and accurately track the motion targets and accurately distinguish the motion targets of the video image.

The invention provides a video-based non-supervision abnormal event real-time detection method. Specific to the aims of reducing the quantity of inter-frame feature points and lowering the complexity in calculation of the feature points at the same time, the feature points are detected with an interval method, namely, a video is segmented, the feature points are detected on a first frame, and only tracking is required subsequently. The calculation amount of a tracking method is small relatively, so that the calculation complexity is lowered greatly. At the end of one segment of video, the feature points are detected once again. After the feature points of each video segment are obtained, direction, speed and location histograms of motion feature points are extracted and connected in series to serve as features of the video segments. Then, the features are subjected to Gaussian mixture modeling and updated in real time to obtain the probability of abnormal events in order to judge whether any abnormal event occurs or not. Thus, the abnormal events can be detected in real time.

The invention discloses a mill load parameter soft measuring method based on a virtual sample. The mill load parameter soft measuring method comprises the steps of acquiring a multi-dimension time domain sub-signal of mill cylinder vibration and vibration sound sample signal by means of ensemble empirical mode decomposition (EEMD) technology, and performing further processing for obtaining high-dimension spectral data with different time dimensions; then constructing a feasibility-based planning (FBP) model based on the high-dimension spectral data according to an improved selective integrated kernel partial least squares (IGASEN-KPLS) method, and generating a new virtual sample based on priori knowledge and an FBP model; then obtaining a mixed sampling model after mixing the new virtual sample with a true training sample, performing adaptive selection of a multiple-dimension spectral characteristic by means of a mutual information (MI) based characteristic selecting method, constructing a soft measuring model by means of the selected spectral characteristics and performing soft measurement.

The invention provides a method for component content prediction and optimization operation in a wet-process metallurgic extraction process, which adopts the wet-process metallurgic extraction technology of a multi-stage extraction tank to realize real-time prediction on raffinate component content by hybrid modeling of the wet-process metallurgic extraction process, and provide online optimization operation guidance for the extraction process. The method comprises the steps of data acquisition, auxiliary variation selection, standard processing, hybrid model establishment, hybrid model correction, optimization operation guidance determination and the like. The method can greatly improve the leaching rate so as to ensure that the production is maintained in a best loss state, can reduce consumption of raw materials and energy, and prolongs the running period of equipment.

The invention discloses a method for monitoring abnormal behaviors of elderly people living alone in a family environment, and belongs to the technical field of video images. The method includes the first step of carrying out infrared camera shooting by means of an infrared camera to obtain a video picture signal, the second step of carrying out foreground detection in an improved self-adaptation Gaussian mixture modeling method, the third step of recognizing a target block mass to obtain block mass information, the fourth step of carrying out real-time tracking on the target block mass, and the fifth step of carrying out Kalman filtering on the center of the target block mass to obtain a predicted value, comparing the predicted value with an actual value, and if the predicted value is larger than a threshold value, then behaviors are determined to be abnormal, and an emergent message is sent out. According to the method, privacy of a monitored object can be protected in a certain degree through the infrared camera, behavior monitoring is carried out through videos, the purpose of monitoring abnormal behaviors in a non-contact mode is achieved, meanwhile, the tracking algorithm used in the method is good in real-time performance, features of a tracked object are easy to obtain, quickness of the operational process is good, and the judgement standard is simple and reliable.

The invention provides an improved adaptive Gaussian mixture foreground detection method. The method comprises: firstly, performing learning by utilizing a Gaussian mixture model to form an initialized Gaussian mixture background model; secondly, for a new input video sequence, performing sampling at an interval of N frames, obtaining an image frame by utilizing weighted time-domain mean filtering, and performing background model updating by taking the image frame as an input of Gaussian mixture modeling; automatically determining whether background mutation exists in a current frame by Poisson distribution, if the background mutation does not exist, keeping normal sampling interval and learning rate, otherwise, reducing an interval frame number and increasing the learning rate, updating the background model, and extracting a current background frame; and finally, performing difference by utilizing the current frame and the current background frame, obtaining an adaptive threshold with a maximum entropy method, performing weighted mean on the obtained threshold, and performing foreground detection. According to the method, motion interferences of tree leaf shake, water ripples and the like in a video scene are effectively overcome, the calculation amount of frames is reduced through periodic sampling, and the timeliness is improved.

The invention provides a modeling method and device for voice recognition and equipment. The method comprises the steps of determining N-class labels; constructing a voice recognition model accordingto mandarin voice data training and the N-class labels; acquiring recognition text of voice data of P dialects according to the voice recognition model; determining an error rate according to the recognition text and labeled reference text, determining an acoustic characteristic difference value of each word according to a first error rate of each word and a second error rate of each word in the mandarin for each dialect, and generating new M-class target labels according to M-class labels corresponding to the words with the difference values greater than a preset threshold value; training anacoustic model according to the voice data of the mandarin and P dialects, wherein the output of the acoustic model is the N-class labels and the M-class target labels corresponding to each dialect. Therefore, mixed modeling of mandarin and the dialects is achieved, and one model can support both the mandarin and diversified dialects while the recognition accuracy is ensured.

The invention discloses a complex structure point cloud processing algorithm bases on Poisson reconstruction. Curved surface reconstruction carried out through a point cloud processing technical method based on Poisson reconstruction adopts an indicator function to describe a curved surface; a vector field based on normal vector is constructed through Gaussian filtering; and a Poisson equation is solved through a multi-grid method to obtain a transition portion of the indicator function of directional point cloud to finish the curved surface reconstruction. The curved surface reconstruction technology based on Poisson is mainly formed by five parts: defining an octree, setting function space, estimating the vector field, solving the Poisson equation and extracting contour surface. With the fast development of a 3D laser scanning device, the point cloud obtaining technology also has considerable progress. The point cloud processing algorithm can be widely applied to the fields of reverse engineering, hybrid modeling, visual inspection, medical images and archaeological and cultural relic modeling and the like.

The invention provides a modeling method, device and equipment for voice recognition, wherein the method includes determining N-type labels; training a neural network according to second voice data ofMandarin to generate a recognition model with N-type label output; inputting the second voice data of P-type dialects into the recognition model respectively for processing to acquire an output labelof the second voice data of each frame of the dialects; according to the output label and a real labeled label, determining the error rate of the N-type labels for each dialect in the P-type dialects, and newly generating M-type target labels according to labels with the error rate larger than a preset threshold value; and training an acoustic model according to third voice data of Mandarin and the third voice data of P-type dialects, wherein the output of the acoustic model is the N-type labels and the M-type target labels corresponding to each dialect in the P-type dialects. Therefore, themixed modeling of Mandarin and dialects is realized, while the accuracy of recognition is ensured, the same model can support both Mandarin and multiple dialects.

The invention discloses a gushing operation condition soft sensing modeling method based on hybrid multiple models in a shield tunneling process. Hybrid multiple models include the following four models: 1, a data-driven exploration information and construction data-based soil texture soft sensing model established by using a belief rule base method; 2, a data-driven seepage flow soft sensing model established by using a support vector machine method; 3, a data-driven permeability coefficient model established by using the support vector machine method; and 4, a gushing operation condition soft sensing model established by using a simplification mechanism. The soil texture soft sensing model, the seepage flow soft sensing model and the permeability coefficient model are input of the gushing operation condition soft sensing model. The modeling method disclosed by the invention has the advantages that a hybrid modeling method combining mechanism modeling and data-driven modeling is adopted, and a model modeled by using the modeling method is simple and has high interpretability, high reliability and good extrapolation.

The invention relates to a cloud modeling method of large-scale ALS building points of repetitive buildings based on automatic sensitivity. The method comprises following steps: (1) adopting a deep learning method, precisely cutting ALS point clouds and obtaining four kinds of targets such as buildings, plants, grounds and others; (2) detecting repetitive buildings in local areas as for cloud points of repetitive buildings and adopting a data driving method for rectification and alignment of detected repetitive buildings to construct a roof model for repetitive buildings; taking a mixed model building method including comprehensive data driving and model driving for reminding non-repetitive buildings and constructing a geometric model for building roofs; (3) making qualitative and quantitative evaluations as for precision and efficiency of the modeling model for geometric models of building roofs. The method has following advantages: 1) modeling efficiency and precision is high so that city neighborhoods of many repetitive buildings can be suitably modeled; 2) the method facilitates integration with other methods in order to increase application scope of modeling methods and layered details of models.

The invention discloses a power transmission line large-scale construction breakage-prevention large-scale construction vehicle recognition method based on a BOW image representation model. The method comprises the steps that median filtering is performed on a current image frame obtained by a camera; a Gaussian mixture modeling method is used for background modeling and foreground recognition; an intra-large-region texture-based method is used for eliminating shadows of a moving object; BOW model characteristics of each foreground region are extracted; the extracted characteristics are fed into a multi-classification SVM learned in advance for vehicle type recognition. According to the power transmission line large-scale construction breakage-prevention large-scale construction vehicle recognition method, the defect that other color-based methods can only detect vehicles of a certain specific color can be overcome, types of various large-scale construction vehicles can be detected, and the accuracy is high.

The invention provides a thickener underflow concentration prediction method based on a mixing model, Aiming at the problem that it is difficult to measure the underflow concentration on-line in hydrometallurgy dense washing process, On the basis of in-depth analysis of the characteristics of dense scrubbing process, a hybrid modeling method combining mechanism modeling and three-layer ELM error compensation model based on global distribution optimization algorithm is used to realize the accurate measurement of underflow concentration in dense scrubbing process.

The present invention relates to a modeling method of broadband radio frequency power amplifier. In the method, a memory-free non-linear nominal model is connected with a memory non-linear identification model in parallel so as to realize the modeling of the power amplifier PA. The modeling method makes full use of the prior knowledge of the PA, and adopts the mixed modeling idea which combines the traditional memory-free non-linear nominal model and the memory non-linear identification model, realizes the modeling of the broadband radio frequency PA, and greatly improves the performance of the whole PA model.

The invention discloses a hybrid modeling method for a feeding system of a numerical control machine tool. The hybrid modeling method comprises a dynamic basic model and a neural network deviation model based on big data, wherein the dynamics basic model is obtained through dynamics theory analysis and parameter identification; the neural network deviation model is obtained by analyzing and training an instruction sequence, simulation prediction data of the dynamic basic model and actual response data; and the instruction sequence is input into a system mixing model, and the actual response sequence is predicted to obtain a mixed prediction sequence. Compared with a pure dynamic model, the technical scheme of the invention has the advantages that the simulation of a highly nonlinear process (such as a reverse process) is more accurate, and compared with a pure neural network model, the generalization ability under different processing technologies is stronger. Through a hybrid modelingmode, accurate simulation of a complex dynamic feeding system is realized.

A mechanism for enabling a user to treat 3D scan data model regions as surface bodies so as to eliminate the need to convert model regions to parametric surfaces as a prerequisite to performing part body modeling operations is discussed. During CAD re-modeling raw 3D scan data is imported. The present invention allows a user to use a region directly as an input argument for a part body modeling operation as long as a sheet body (surface body) is applicable as a modeling input argument. Additionally, if any regions used in the body modeling are modified by the user, surfaces which have been generated from fitting regions are also recalculated and the body model is updated automatically. The region modification can happen when the user includes more scan data in the region, excludes scan data from the region, smoothes geometry, or uses other editing functions.

The invention discloses a soil equivalent resistance model modeling method in combination of repeated multipoint actual measured data of a power grid and simulation modeling calculation. The method ischaracterized in that an existing soil equivalent resistance model construction method is improved by use of a mixed modeling method in combination of the actual measured data and the simulation modeling calculation; repeated multipoint measurements are performed on an alternating-current power grid by changing a topological structure of the alternating-current power grid, that is, sequentially switching off any return circuit of alternating current circuits between two transformer substations with direct electrical contact in a modeling range to obtain multiple groups of direct current distribution data of a main transformer neutrality point of each transformer substation in the modeling range; and a soil equivalent resistance model is constructed by simulating calculation and inversionmeans. According to the soil equivalent resistance model modeling method, detailed soil parameters are unnecessary to collect, the calculation accuracy degree is high, the problem that a non-uniform layered soil model is difficult to establish accurately in previous methods is overcome, and the method is suitable for accurate calculation of distribution of direct currents in an alternating currentsystem in the case of complex soil compositions.

The invention discloses a sound track spectrum Gaussian mixture model based rapid voice conversion system and method. The method comprises the steps of parameter extraction and synthesis, characteristic parameter time aligning and characteristic parameter training and conversion. By the technologies of fixation of the Gaussian average on Mel frequency spectra, adaptive Gaussian variance adjusting, selecting of sampling points as weight coefficients on logarithm magnitude spectra and the like, the calculation complexity of voice parameter characterization is greatly reduced, and the operating rate is improved greatly.

The invention discloses a method of calculating the fatigue life of a flexible bearing by utilizing finite element modeling. The method includes: taking the flexible bearing in a harmonic reducer as a research object, establishing a three-dimensional model in UG, establishing a grid model of the flexible bearing in ANSYS software, and simulating, through establishing link units, a cage to realize a constraint effect on rolling elements, wherein a ''link unit-entity'' hybrid finite element model of the flexible bearing is established, and a solid cage model is omitted; and obtaining load distribution under working condition of the flexible bearing according to a finite element analysis result, thus calculating equivalent rolling element loads of the flexible bearing, and then calculating the life of the flexible bearing according to an L-P life theory. Compared with equivalent rolling element loads obtained by empirical formulas, the equivalent rolling element loads obtained by the finite element simulation method utilized in the invention are more accurate. At the same time, the finite element model in the invention is established by adopting a hybrid modeling technology, the calculation scale of the finite element model is reduced, and the calculation time is greatly shortened. The method disclosed in the invention is applicable to different types of flexible bearings, and is low in cost and higher in precision.

The invention provides a dynamic stiffness modeling method for a flexible support gear transmission device, belongs to the field of gear system dynamic modeling, and aims to divide a common single-layer vibration isolation gear system into a transmission system, a box body, a vibration isolator and a basic module. The method comprises the steps of establishing a concentrated mass model of the transmission system, and converting to obtain a dynamic stiffness equation; establishing a box body finite element model, and obtaining box body dynamic stiffness parameters through harmonic response analysis; simplifying the vibration isolator into a Timoshenko beam, and obtaining a dynamic stiffness equation through a wave equation; obtaining dynamic stiffness parameters by a foundation through a test; by assembling the dynamic stiffness equations of the subsystems, obtaining a complete dynamic stiffness model of the gear transmission device, and hybrid modeling of theoretical parameters / test parameters can be achieved. The modeling method can greatly improve the modeling efficiency of gear transmission device analysis and the accuracy of calculation results.

The embodiment of the invention discloses a boiler heat storage coefficient measurement method and a boiler heat storage coefficient measurement device. The measurement method comprises the following steps: acquiring boiler measuring point data; and processing the boiler measuring point data based on a hybrid model which is established by a mechanism model and takes the boiler measuring point data as input and takes the compensation result obtained by performing deviation compensation on the output data of the mechanism model as the output by employing the hybrid modeling technology, and calculating to obtain the boiler heat storage coefficient so as to improve the calculation precision of the boiler heat storage coefficient, wherein the mechanism model is a mathematical model established by performing mechanism analysis on the boiler measuring point data and the boiler heat storage coefficient according to the energy conservation law in advance.

The invention discloses a vehicle detection method based on multi-component space position relation GMM (Gaussian Mixture Modeling). The method comprises the following steps: training color features of a license plate and a pair of rear lamps to obtain conversion models specific to color images and gray images, and processing original road traffic video images by the conversion models respectively to obtain gray images highlighting a license plate area and rear lamp areas; performing threshold segmentation and connected component analysis on the grey image of the pair of rear lamps to finish further accurate positioning of the pair of rear lamps; and lastly, building a position relationship probability model for detection results of the pair of rear lamps and the license plate on the same space according to a GMM theory, and further judging whether targets constructed by detected components refer to the same vehicle or different vehicles so as to finish accurate recognition of vehicles. The method has relatively high stability and detection accuracy, and has a relatively good effect particularly for the situations of poor light conditions and component shielding.