674 results about "Model set" patented technology

The invention relates to a method and apparatus for training a multilingual speech model set. The multilingual speech model set generated is suitable for use by a speech recognition system for recognizing spoken utterances for at least two different languages. The invention allows using a single speech recognition unit with a single speech model set to perform speech recognition on utterances from two or more languages. The method and apparatus make use of a group of a group of acoustic sub-word units comprised of a first subgroup of acoustic sub-word units associated to a first language and a second subgroup of acoustic sub-word units associated to a second language where the first subgroup and the second subgroup share at least one common acoustic sub-word unit. The method and apparatus also make use of a plurality of letter to acoustic sub-word unit rules sets, each letter to acoustic sub-word unit rules set being associated to a different language. A set of untrained speech models is trained on the basis of a training set comprising speech tokens and their associated labels in combination with the group of acoustic sub-word units and the plurality of letter to acoustic sub-word unit rules sets. The invention also provides a computer readable storage medium comprising a program element for implementing the method for training a multilingual speech model set.

Disclosed herein are systems, methods, and devices for classifying ophthalmic images according to disease type, state, and stage. The disclosed invention details systems, methods, and devices to perform the aforementioned classification based on weighted-linkage of an ensemble of machine learning models. In some parts, each model is trained on a training data set and tested on a test dataset. In other parts, the models are ranked based on classification performance, and model weights are assigned based on model rank. To classify an ophthalmic image, that image is presented to each model of the ensemble for classification, yielding a probabilistic classification score—of each model. Using the model weights, a weighted-average of the individual model-generated probabilistic scores is computed and used for the classification.

The recently introduced theory of Compressive Sensing (CS) enables a new method for signal recovery from incomplete information (a reduced set of “compressive” linear measurements), based on the assumption that the signal is sparse in some dictionary. Such compressive measurement schemes are desirable in practice for reducing the costs of signal acquisition, storage, and processing. However, the current CS framework considers only a certain task (signal recovery) and only in a certain model setting (sparsity).

We show that compressive measurements are in fact information scalable, allowing one to answer a broad spectrum of questions about a signal when provided only with a reduced set of compressive measurements. These questions range from complete signal recovery at one extreme down to a simple binary detection decision at the other. (Questions in between include, for example, estimation and classification.) We provide techniques such as a “compressive matched filter” for answering several of these questions given the available measurements, often without needing to first reconstruct the signal. In many cases, these techniques can succeed with far fewer measurements than would be required for full signal recovery, and such techniques can also be computationally more efficient. Based on additional mathematical insight, we discuss information scalable algorithms in several model settings, including sparsity (as in CS), but also in parametric or manifold-based settings and in model-free settings for generic statements of detection, classification, and estimation problems.

A state based adaptive PID controller includes a model set component including a plurality of process models, each process model including a plurality of parameters. An error generator generates a model error signal representative of a difference between a model output signal and a process output signal. A model evaluation component computes a model squared error based on the model error signal. A parameter interpolator calculates an adaptive parameter value based on the model squared error. A controller update component updates adaptive controller parameter values based on adaptive parameter values.

A method automatically recognizes speech received through an input. The method accesses one or more speaker-independent speaker models. The method detects whether the received speech input matches a speaker model according to an adaptable predetermined criterion. The method creates a speaker model assigned to a speaker model set when no match occurs based on the input.

A robot simulation device for simulating an operation of a robot having a vision sensor in an off-line mode. The device includes a working-environment model setting section for arranging a sensor model, a robot model and a plurality of irregularly piled workpiece models in a virtual working environment; and an operation simulating section for allowing the sensor model and the robot model to simulate a workpiece detecting operation and a bin picking motion. The operation simulating section includes a workpiece-model image generating section for allowing the sensor model to pick up the workpiece models and generating a virtual image thereof; a workpiece-model position detecting section for identifying an objective workpiece model from the virtual image and detecting a virtual position thereof; and a robot-model operation controlling section for allowing the robot model to pick out the objective workpiece model based on the virtual position.

The invention relates to a stereoscopic vision based emergency treatment device and method for running vehicles, wherein the device comprises a binocular-vision image pick-up unit, an on-board bus interface, a central processing unit, a vehicle braking control system and an acousto-optic alarm circuit. The binocular-vision image pick-up unit is used for capturing an image of a road in front of a vehicle; the DSP (digital signal processor) based central processing unit is adopted for carrying out real-time quick calculation on a visual image so as to obtain a three-dimensional road scene, and compares the obtained three-dimensional road scene with a safe driving road model set up by a system so as to judge whether obstacles or dangers exist in the traveling direction of the vehicle; when adanger is found, the vehicle braking control system is started so as to reduce the speed of the vehicle and send an acousto-optic alarm to a driver; meanwhile, the central processing unit is connected with a vehicle sensor by an inter-vehicle bus so as to detect the state of the vehicle, and when the vehicle has mechanical or circuit faults, a braking system is started so as to reduce the speed of the vehicle and send an acousto-optic alarm. The device disclosed by the invention can be arranged on ordinary motor vehicles so as to avoid the occurrence of accidents or reduce the accident loss, thereby improving the driving safety performance.

The invention provides a deep question and answer method, which comprises the following steps of: 1, receiving question data; 2, carrying out a first retrieval in a knowledge base database, wherein the knowledge base database comprises factual information in encyclopedia; and 3, carrying out a third retrieval in a self-defining data set, wherein the third step comprises the steps of: 3.1, classifying the questions by a machine study method on the basis of a sentence pattern model set, and determining the sentence pattern mode of questions; and 3.2, obtaining second type keywords through the questions matched with the sentence pattern mode, and retrieving the self-defining data set through the second type keywords. Corresponding to the method, the invention also provides a medical retrieval system which comprises an input module, a first retrieval module and a third retrieval module, wherein the first retrieval module is used for retrieving the knowledge base database, the knowledge base database comprises the factual information in encyclopedia, and the third retrieval module is used for retrieving the self-defining data set.

A method for combining multiple probability of click models in an online advertising system into a combined predictive model, the method commencing by receiving a feature set slice (e.g. corresponding to demographics or taxonomies or clusters), and using the sliced data for training multiple slice-wise predictive models. The trained slice-wise predictive models are combined by overlaying a weighted distribution model over the trained slice-wise predictive models. The combined predictive model then is used in predicting the probability of a click given a query-advertisement pair in online advertising. The method can flexibly receive slice specifications, and can overlay any one or more of a variety of distribution models, such as a linear combination or a log-linear combination. Using an appropriate weighted distribution model, the combined predictive model reliably yields predictive estimates of occurrence of click events that are at least as good as the best predictive model in the slice-wise predictive model set.

A vortex-exciting vibration test device of flexible pipe fittings model set vertically in pull water pool is prepared as passing pipe fittings model through flow-speed increasing unit; setting said model and said unit vertically in pull water pool; supporting two ends of said model by support mechanism; using vertical test support to connect said unit and support mechanism with trailer and distributing various instruments of measurement-analysis system in said model, said units, said mechanism, said support and said trailer.

An image processing apparatus includes an edge intensity calculator which calculates an edge intensity of a pixel in an image; and a correlation value calculator which calculates a correlation value between the calculated edge intensity and a bubble model set in advance based on characteristics of a bubble image. The apparatus also includes a bubble area detector which detects a bubble area based on the calculated correlation value.

Upon receipt of an end point list, a model setting unit requests the measurement/information collection unit to search a path. The model setting unit receives a path appliance list from the measurement/information collection unit, and generates the configuration of a simulation model of a network. The model setting unit selects an appropriate model for each component of the configuration of the simulation model. Then, the model setting unit sets the parameter value of the model based on the measurement result and the collected information received from the measurement/information collection unit, and automatically generates the simulation model of the network.

The invention relates to a management method in the technical field of the automation of a power system, in particular to a multi-version management method for a model set and difference model-based power grid model. The method comprises the following steps of: 1) expressing a difference model; 2) storing the model; 3) loading the model; 4) browsing the model; 5) merging the model; 6) checking the model; 7) transforming the model; and 8) carrying out the on-line distributing and putting of the model. The method divides a space-dimensional model and an application-dimensional model into corresponding versions through the type of a model space and a model tree, and takes the model set as a base wire to realize the multi-version management of the model time dimension by overlapping the modes which become effective at different times.

The invention discloses an MMVI-CFAR target detection method. Clutter background modes which a target signal possibly undergoes are set to be a model set for overcoming the defect of detection performance at the position where different clutter backgrounds of existing radar maneuvering weak targets are connected. The method comprises the specific steps that 1, uniform clutter background models, clutter edge background models and non-uniform multi-target background models are respectively built; 2, a model set corresponding to clutter background areas is built; 3, reference windows are divided into a left reference window, a right reference window and a full reference window; 4, weight calculation is conducted according to the models, and corresponding constant false alarm processing is conducted according to the value of the weight. The MMVI-CFAR target detection method is applicable to various clutter backgrounds, and the advantages of CA-CFAR, SO-CFAR and GO-CFAR during detection of uniform clutter backgrounds, clutter edge backgrounds and multi-target backgrounds are respectively utilized; detection probability is greatly improved at the positions where different clutter backgrounds are overlapped.

The invention discloses a high-speed train tracking running curve optimization setting method. According to the characteristic of 'movable and dynamic length' of a tracking section of a high-speed train under movable blocking condition, the method establishes a high-speed train echo state network speed prediction model, a movable blocking based tracking running model, a line network and a tracking running curve multi-target setting model adopting innovative evaluation indexes based on line and high-speed train running data acquired in a site; an efficient multi-target particle swarm optimization is adopted to use an algorithm convergence condition as one of model setting constraints, and high-speed train tracking running curve optimization setting is performed based on the real-time data; finally, section operation efficiency and stability are used as the evaluation indexes of the setting method, and a group of optimal running curves are screened out, so that the high-speed train running process is safe and efficient, and meanwhile the high-speed train section operating efficiency and stability under the movable blocking condition are improved.

Techniques for building a model for performing diagnostics. In one embodiment, a set of models is determined based upon a topological relationship created upon receiving an alert or a request for which diagnostics are to be performed. An aggregate model is then generated based upon the set of models and the topological relationship. The aggregate model is then used for performing the diagnostics.

A surface-profile measuring instrument includes a movement-estimating unit (600) that estimates an movement state of a drive mechanism in accordance with a scanning vector command issued by a scanning vector commanding unit (220) to calculate an estimated movement state quantity, and a correction calculating unit (700) that corrects a detection value of a drive sensor in accordance with the estimated movement state quantity calculated by the movement estimating unit (600).

The movement-estimating unit (600) has a nominal model setting unit (311) in which a nominal model as a frequency transfer characteristics from the scanning vector commanding unit to the probe of the scanning probe is set. The correction-calculating unit (700) includes a correction-amount calculating unit (720) that calculates a correction amount for correcting a measurement error generated by a deformation of the drive mechanism in accordance with the movement thereof and a measurement data synthesizing unit (430) that synthesizes the detection value of the drive sensor and the displacement sensor and the calculated correction-amount calculated by the correction-amount calculating unit (720).

The application discloses an image identification method, an image identification device, a metric learning method, a metric learning device, an image source identification method and an image source identification device. The image identification method comprises the following steps: getting a to-be-identified object image; extracting the object characteristics of the to-be-identified object image; selecting a similarity measurement model corresponding to the source category of the to-be-identified object image from a pre-trained measurement model set, and calculating the similarity between the object characteristics and registered image object characteristics as the basis of an output object identification result, wherein the measurement model set contains at least one similarity measurement model, and different similarity measurement models correspond to different source categories of object images. By using the method in image identification, the problem about asymmetric object image identification can be solved effectively, and higher robustness and higher accuracy are achieved for identification of to-be-identified object images of changeable sources.

A system and method for automatically optimizing states of communications and operations in a contact center, using a reinforcement learning module comprising a reinforcement learning server and an optimization server introduced to existing infrastructure of the contact center, that, through use of a model set up a fully observable Markov decision process within a known time period, a resulting hyper-policy is computed through backwards induction to provide an optimal action policy to use in each state of a contact center, thereby ultimately optimizing states of communications and operations for an overall return over the time period considered.

The invention provides a maneuvering target tracking method based on variable structure multiple models. The maneuvering target tracking method mainly solves the problems that in the prior art, target tracking accuracy is low and target tracking is not in time. The method includes the implementation steps that firstly, a target motion state detected by radar is sampled, and then a radar measured value is acquired; secondly, a maneuvering target tracking mathematical model is established according to target motion characteristics; thirdly, a maneuvering target motion model is established according to the target tracking mathematical model and is initialized; fourthly, through a model set adaptive strategy, a target motion model set at each moment is acquired; fifthly, according to the radar measured value and the target motion model sets, a variable structure interactive multi-model algorithm is operated, state estimation of a maneuvering target is acquired, and primary tracking of the target is achieved; sixthly, the fourth step and the fifth step are repeatedly executed until tracking is finished. On the premise that high tracking accuracy is guaranteed, the calculation quantity is reduced, and timeliness of maneuvering target tracking is improved.

The invention discloses a method for automatically generating a dynamic wireless application protocol (WAP) website for separation of a page from data. By the method, the dynamic WAP website is generated dynamically on the basis of extensive markup language (XML) identification page contents and data contents. The method comprises the following steps of: firstly, providing concepts of a component unit, a component unit representation model and a component unit data model, and defining a page representation template and a data representation template by using the concepts as basic units; and secondly, implementing a WAP website generator, interpreting a page representation model set which is established according to the page representation template to generate a WAP website target code, and dynamically interpreting a data representation model which is established according to the data representation template in the operating process of the code. In the method, by structuralized and componentized page representation and data representation, a dynamic page can be dynamically generated possibly; static page construction process is separated from dynamic data processing process, so the WAP dynamic website can be simply and efficiently constructed; and fast data exchange channels between the WAP website and other business platforms can be established.

The invention discloses a multi-model human motion tracking method, and relates to the technical field of image processing, which mainly solves the problems that human motion ambiguity and time complexity are high and excellent three-dimensional body posture estimation cannot be obtained by increasing motion models purely in the conventional method. The multi-model human motion tracking method comprises the following steps of: (1) inputting a human motion video image to acquire a human silhouette and an edge and a skeleton line thereof; (2) detecting the positions of human articulation points; (3) training the motion models by a ridge regression method; (4) initializing a model set M1; (5) performing an interactive multi-model algorithm to acquire a human motion posture; and (6) activating the motion models which meet the activating condition, and terminating the motion models which meet the termination condition. The multi-model human motion tracking method has the advantages of low time complexity and good tracking effect, has high cost effectiveness, and can be applied in fields of sports training, animation production and the like.

The invention discloses a voice emotion recognition and application system for conversations of a call center. The system comprises a voice processing module, a voice keyword detection module, an emotion model set module, an emotion classification module and a business application module, wherein the voice processing module is used for extracting voice information and preprocessing the voice information; the voice keyword detection module is used for recognizing voice data of a voice feature analysis submodule as emotion keywords and topic keywords, and obtaining emotion data information and reflected problem data information; the emotion model set module is used for dynamically capturing and tracking emotion states of callers; the emotion classification module is used for judging emotioncategories of voices in to-be-detected conversations; and the service application module is used for providing response assistance for the customer service personnel and providing management assistance for the management personnel. The voice emotion recognition and application system for conversations of call centers is capable of enabling the customer service personnel to accurately understand the emotions of customers, providing effective response plans, and accurately evaluating the customer service personnel.

The embodiment of the invention discloses a figure replacing method in video live. A problem of how to replace the figure in the live can be solved. The method disclosed by the embodiment of the invention comprises the following steps: acquiring the current video frame image of a live target video in time; detecting whether a human face exists in the video frame image; if the human face exists in the video frame image, recognizing body silhouette corresponding to the human face at the position of the human face in the video frame image; determining a body coordinate position of the body silhouette in the video frame image; selecting a figure model matched with the body silhouette from a preset figure model set as a target model; replacing the body silhouette to the body coordinate position by use of the target model so as to synthesize a new video frame image; and outputting the new video frame image to the network to perform the live streaming. The embodiment of the invention further provides a figure replacing device in the video live and a recording and playing system.