102 results about "Model composition" patented technology

The invention discloses an evanescent die process. The evanescent die process is characterized by comprising the following steps of: selecting foaming plastic beads, manufacturing a model; enabling the beads to expand to be in a certain size through hot water pre-frothing, steam pre-frothing and vacuum pre-frothing, curing and frothing forming; clustering model composition, coating a model; vibrating and shaping; pouring and displacing; and cooling and cleaning, wherein the full mold casting ramoff is simple, a casting piece can be hoisted out when a sand box is inclined or can be directed hoisted out form the sand box, the casting piece and dried sand are naturally separated, and the separated dried sand is reused after treatment. According to the process, a die is not required to be taken out, and a parting plane and a sand core are not required, and thus the casting piece has no flash, no burrs and no pattern tapers, and the dimension error caused by core combination is reduced; and compared with the conventional sand casting method, for the evanescent die process, the 40%-50% machining time is reduced.

A method includes combining at least two 3-D models, given the positions of the models and constraint areas of the models that should not be changed by the combination. The combining includes generating a weighted graph representation of the models at least in a transition volume and including at least a portion of the constraint areas and finding a minimum cut which separates the weighted graph into two cut graphs representing cut versions of the models.

The invention relates to a sound event recognition method based on optimized parallel model combination. The sound event recognition method includes 1) recording data of a sound event, acquiring a GMM (Gaussian mixture model) according to clean sound event training, and establishing a clean sound event template; 2) acquiring noise data in current environment in indoor actual noisy environment, acquiring a GMM according to the noise data training, and establishing a noise template; 3) processing the noise template and the clean sound event template by the method of optimized parallel model combination, and obtaining a template of a sound event with noise; 4) sampling to obtain sample signals of the sound event with noise, recognizing sound of the sample signals according to the parameters in the template of the sound event with noise. According to the sound event recognition method, a GMM capable of describing background noise feature distribution better is established as one input in a PMC (portable media center) method, a clean GMM of five sound events is established at another input in the PMC. Meanwhile, robustness of a recognition system to noises is guaranteed.

There is provided a voice recognition device and a voice recognition method that enhance the function of noise adaptation processing in voice recognition processing and reduce the capacity of a memory being used. Acoustic models are subjected to clustering processing to calculate the centroid of each cluster and the differential vector between the centroid and each model, model composition between each kind of assumed noise model and the calculated centroid is carried out, and the centroid of each composition model and the differential vector are stored in a memory. In the actual recognition processing, the centroid optimal to the environment estimated by the utterance environmental estimation is extracted from the memory, model restoration is carried out on the extracted centroid by using the differential vector stored in the memory, and noise adaptation processing is executed on the basis of the restored model.

A modelling composition, in which there is no cracking likely to cause the modelling composition to disintegrate either during drying or during drying of the objects made from said composition, contains a compound made up of the components starch, an adhesive agent, common salt, water, calcium chloride, sodium benzoate, potassium hydrogen sulphate, liquid paraffin, propylene glycol and bittern.

The present invention presents two new model types and a new method for evaluating a model used in the control application. These include a compound model, a hybrid model and a directional change coefficient model. The present invention allows the mixing of models with different inputs and outputs and the switching between these models based criteria for measuring optimization accuracy. The present invention allows switching between these models. The compound model is a model type that allows zooming in on the process to model parts of the data space with higher fidelity or resolution without loosing the capability to model the complete data space. The modeler does not loose any functionally over a regular neural network, but instead gains the ability to define the conditions when the model should use network weights best matched to the defined local conditions. The hybrid model is an extended version of a compound model. A hybrid model allows the combining of one or more models into a single model for purposes of interrogation or optimization. Within the hybrid model may reside a compound model itself. The directional change model (DCC) allows better evaluation of the predictive capability of Compound Models. It may also be used with any other model type.

The invention discloses a multi-model control method based on self learning. The multi-model control method comprises the steps that (1) a model base is built, and the model base consists of a group of local models of a non-linear model; (2) a group of controllers are built, and a group of local controllers are designed according to the local model in the model base; (3) the performance evaluation is executed: output errors and differences between system output y and model output yi are observed, and a performance feedback or value function is calculated or sent to an API (application program interface) module on the basis of signals; and (4) a similar policy iteration algorithm is executed: performance feedback signals are observed, error signals between reference output and system output are received, the signals are used as the Markov decision process states, and meanwhile, the states are fed back to become return signals for enhancing the leaning. The multi-model control method has the advantages that the principle is simple, the application range is wide, the reliability is high, the general performance and the convergence of the control can be ensured, and the like.

The invention discloses an eight-section impedance model based body composition analysis method. The body composition analysis method comprises the following steps: according to input currents and measured voltages, acquiring six valid body impedance expressions through the eight-section body impedance model; acquiring the difference value between the resistance values of left and right upper limbs as well as the difference value between the resistance values of left and right lower limbs through a five-section body impedance model; calculating to obtain the body impedance expression of each section; according to at least two groups of different input currents and the body impedance expression of each section, obtaining at least two groups of body impedance values; selecting the optimal group of eight-section impedance values, and determining a fitting model according to the selected optimal group of eight-section impedance values; conducting training on a plurality of known samples in the fitting model to obtain unknown coefficients of the fitting model as well as a body composition predicting formula; according to the body composition predicting formula, analyzing the unknown samples to obtain body composition parameters. Through the adoption of the body composition analysis method, the obtained the body composition is accurate.

The invention relates to the technical field of remote sensing image processing, in particular to a PM2.5 inversion method based on MODIS and machine learning model fusion. An MODIS image and PM2.5 monitoring data are acquired; the PM2.5 data are interpolated into PM2.5 interpolation image; the MODIS image is subjected to cloud detection; a training set and a test set are constructed; performanceindicators are calculated according to the training set and the test set; a histogram of the performance indicators is made; all corresponding models in a histogram interval with the highest frequencyin the histogram are selected and taken as the optimal model combination; the optimal model combination is used for the full MODIS image for inversion of model fusion. In the method, the relation between the remote sensing image and actually measured PM2.5 is directly constructed on the basis of data of the remote sensing image through a machine learning algorithm and model fusion, so that the inversion result with higher accuracy is realized. Error transferring is avoided and the inversion accuracy is high.

The invention discloses a 3D vertebra CT image active contour segmentation method fusing weighted random forest, and relates to the field of medical image processing. For the problem of sensitivity ofa vertebra CT image segmentation method to an initial contour, a method for automatically locating a vertebra and segmenting a vertebra CT image is proposed. The method comprises the steps of firstly, proposing a weighted random regression and classification forest algorithm to determine a vertebra center; secondly, putting an initial contour ball of active contour segmentation in the vertebra center, and segmenting out the vertebra in the image by adopting a 3D active contour segmentation method in combination with an energy function; and finally, performing combination output on trained models to obtain a complete vertebra CT image segmentation model. A spinal CT segmentation model proposed in the method can automatically locate the vertebra center and can perform automatic 3D segmentation on the vertebra, so that the spinal CT image segmentation steps and processes are simplified.

The invention relates to a modeling method for combustion optimization of a porous medium combustor, and provides a modeling method considering both model prediction accuracy and generalization capacity aiming at the bottleneck problem in the combustion optimization of the porous medium combustor. Before modeling, modeling data is selected according to uniform distribution and number equalization in topological structure, and proper preprocessing is performed, so that prediction capacity and generalization capacity of a model are guaranteed; combustion characteristic models of the porous medium combustor are established by applying a support vector machine and a radial machine neural network aiming at different fuels respectively; the support vector machine and the radial machine neural network model are integrated by applying a weighted average method to form a combustion optimization characteristic model of the porous medium combustor, wherein a weighting coefficient is obtained through a particle swarm optimization algorithm; and finally, the combustion optimization models of different fuels are combined to form an integral model. By using the method, the combustion optimization characteristic model of the porous medium combustor, which has higher accuracy and generalization capacity, can be established.

A modeling composition consisting of a wax-based and oil-based binder, a filler, and a coloring agent, wherein the filler is substantially a light filler comprised of hollow microbeads. The binder is comprised of solid wax, pasty wax, and a liquid component of a wax base, an oil base, or a wax and oil base. The binder is present in the modeling composition in an amount of 45 to 90% by weight. The modeling composition can be used as play dough for children or as a therapeutic composition in the medical field for training and rehabilitation.

The invention provides a two-dimensional animation production method. The method comprises the following steps of: drawing a plurality of limb element models of each limb element; selecting suitable limb element models and combining the models into a role; selecting limb presentation of each limb element according to posture requirements of the role, and placing the body presentation of each limb element at a corresponding time point; and generating body presentation of each limb element between any two adjacent role postures through ordered changes to complete continuous changes of role postures, and performing animation demonstration through a playhead. The invention further discloses the limb element model and the role. According to the method, a small number of limb element models and internal states of the limb element models are designed, the limb element models and the internal states of the element models are selected to form the role postures at the time points, and continuity processing is performed on the role postures, so that animation production is implemented; and the design is simple, the workload is greatly reduced, and requirements for designers and computer configurations are not high, thereby facilitating popularization.

The invention discloses a general topic-embedding-model joint-training method. The method comprises the following steps: step 1, preprocessing an input original document corpus to obtain target text;step 2, constructing a vocabulary table for the target text; step 3, initializing network structures, initializing parameter matrices of models, and constructing a negative sampling table; and step 4,carrying out joint modeling on the topic embedding models, and training the models in a manner of multiple iterations, wherein each iteration process is divided into the following three steps: step 1, using an expectation-maximization (EM) algorithm to train a topic model part; step 2, using a stochastic-gradient-descent algorithm to train an embedding model part; and step 3, using a complete-gradient-descent algorithm to train a regularization term part. The invention can provide a general manner for jointly training the topic model and the embedding model, and problems such as the problemsthat existing model combination manners are too dependent on unique models, generality is insufficient, and two models are difficult to improve at the same time are solved.