44 results about "Deep structure" patented technology

The invention discloses a super-resolution face recognition method based on deep belief networks, and relates to the technical field of face recognition. From a cognitive perspective, it is believed that an intrinsic relation exists between mutually corresponding face images differing in resolutions. But previous studies show that the method for expressing the intrinsic relation by linear approximation is restricted by linear approximation. Therefore, it is believed that the intrinsic relation is nonlinear. In view of outstanding performances of an artificial neural network on the nonlinear classification problem, a neural network algorithm is adopted to capture the nonlinear relation of the mutually corresponding face images differing in resolutions under the condition of posture change. Both theoretical research and neurophysiological research show that a deep structure, such as a system constructed by multiple layers of nonlinear processing units, should be constructed to build an intelligent processing system. According to the face recognition method, deep belief networks are adopted to extract a common nonlinear structure shared by mutually corresponding face images differing in resolutions.

A method is disclosed herein that includes an act of causing a processor to receive a sample, wherein the sample is one of spoken utterance, an online handwriting sample, or a moving image sample. The method also comprises the act of causing the processor to decode the sample based at least in part upon an output of a combination of a deep structure and a context-dependent Hidden Markov Model (HMM), wherein the deep structure is configured to output a posterior probability of a context-dependent unit. The deep structure is a Deep Belief Network consisting of many layers of nonlinear units with connecting weights between layers trained by a pretraining step followed by a fine-tuning step.

Methods and systems for estimating values of process parameters, structural parameters, or both, based on x-ray scatterometry measurements of high aspect ratio semiconductor structures are presented herein. X-ray scatterometry measurements are performed at one or more steps of a fabrication process flow. The measurements are performed quickly and with sufficient accuracy to enable yield improvement of an on-going semiconductor fabrication process flow. Process corrections are determined based on the measured values of parameters of interest and the corrections are communicated to the process tool to change one or more process control parameters of the process tool. In some examples, measurements are performed while the wafer is being processed to control the on-going fabrication process step. In some examples, X-ray scatterometry measurements are performed after a particular process step and process control parameters are updated for processing of future devices.

The invention discloses a method for increasing a seismic imaging resolution ratio and is applied to reflection seismic data processing in seismic exploration . The method includes that by performing frequency-dependent amplitude recovery and frequency dispersion correction on migration stack data volume, the resolution ratio of imaging is increased integrally, and the imaging resolution ratios of middle and deep structures are enabled to be close to the imaging resolution ratio of a shallow structure; amplitude recovery and frequency dispersion correction are performed based on a viscoelastic theory, but a nonuniform Q value field does not need to be provided in advance; viscous absorption of seismic waves is described by introducing an equivalence Q value, and the nonuniform equivalent Q value field needed for calculation is acquired by establishing a single numerical indicator and by utilizing parameter scanning. Compared with methods like unsteady-state deconvolution, the method has the advantages that high-frequency information is restored while frequency dispersion correction is completed; a stable control algorithm is introduced, so that stability of calculation is guaranteed. By the method, breaking of the middle and deep structures and stratum deposition forms can be indicated better; and the method has important application value for oil-gas and mineral resource exploration.

The invention relates to a multi-stage reducing reaming-while-drilling drilling rig and method based on a radio frequency identification technology. The multi-stage reducing reaming-while-drilling drilling rig and method are suitable for deep structure coal seam horizontal well underground multi-stage reaming. The multi-stage reducing reaming-while-drilling drilling rig includes a radio frequencyidentification excitation mechanism, an electro-hydraulic control mechanism and a multi-stage reaming recovery mechanism. The multi-stage reaming recovery mechanism is utilized to complete primary andsecondary expansion and recovery of a reaming knife wing; for reaming activation, an RFID system is adopted to activate the electro-hydraulic control mechanism, needed reaming conditions are changedby changing an electronic tag ball placed down, and remote control of the reaming is achieved; on the premise of no replacement of the drilling rig, reaming drilling in multiple modes such as a horizontal well, a vertical well and an inclined shaft is achieved in a soft formation such as deformed coal, and multi-stage reaming operation in a certain hole section can also be completed according to engineering needs; an excitation and reaming mechanism is reliable in structure, a reaming mode of the reaming-while-drilling drilling rig is reliable, the hole expansion rate is high, large-diameter pore forming can be achieved, the reaming stability is good, and the degree of automation is high.

The invention discloses a transformer fault diagnosis method and a transformer fault diagnosis device. The transformer fault diagnosis method comprises a first gray scale image corresponding to an infrared image of a to-be-diagnosed transformer is acquired, and according to the pixel value of the first gray scale image, whether the to-be-diagnosed transformer is faulted is determined; when faults exist, the characteristic images of the transformer parts of the to-be-diagnosed transformer are acquired, and comprise gray scale images; the first deep structure characteristics of the characteristic images are determined; the first deep structure characteristics are used to reflect the appearance characteristics, the texture characteristics, and the fault position characteristics of the transformer parts; second deep structure characteristics matched with the first deep structure characteristics are determined according to a predetermined corresponding relation, and the part names corresponding to the determined second deep structure characteristics are used as the names of the transformer parts having the overheating faults. According to the invention, the names of the transformer parts having the concrete faults are determined.

A method is disclosed herein that includes an act of causing a processor to access a deep-structured, layered or hierarchical model, called deep convex network, retained in a computer-readable medium, wherein the deep-structured model comprises a plurality of layers with weights assigned thereto. This layered model can produce the output serving as the scores to combine with transition probabilities between states in a hidden Markov model and language model scores to form a full speech recognizer. The method makes joint use of nonlinear random projections and RBM weights, and it stacks a lower module's output with the raw data to establish its immediately higher module. Batch-based, convex optimization is performed to learn a portion of the deep convex network's weights, rendering it appropriate for parallel computation to accomplish the training. The method can further include the act of jointly substantially optimizing the weights, the transition probabilities, and the language model scores of the deep-structured model using the optimization criterion based on a sequence rather than a set of unrelated frames.

The present invention provides a facial action unit strength estimation-based expression analysis method. The method mainly comprises the content of structured depth condition random field, one-element potential, paired potential, learning and deduction, and comprises the process that the structured depth condition random field has two settings, in the first setting, an input facial image is given, a pre-defined convolutional neural network layer is applied to the (normalization) input image to generate a characteristic graph, and the second setting is to use a data enhancement learning method to utilize the information coming from a plurality of data sets; the one-element node potential is defined in a structural depth condition random field, when the random variables are discrete, the joint distribution can be constructed for the discrete variables, and an objective function of the learned and generated structural depth condition random field is enhanced from the plurality of data sets. The expression analysis method of the present invention utilizes the deep structured learning, can process the high-dimension input characteristics, and enables the image characteristics to be improved substantially, the performances to be improved remarkably and the facial action unit strength estimation accuracy to be improved.

The invention discloses an artificial cultivation method for camphor trees, and belongs to the technical field of afforestation. The artificial cultivation method specifically comprises five steps ofselecting an afforestation land, cutting grass and conducting controlled burning, conducting soil preparation on a furrow zone, transplanting and conducting tending management. By adopting controlledburning and soil preparation on the furrow zone, quick acting nutrients can be greatly increased, the soil fertility can be improved, expelling of harmful animals is facilitated, pests and diseases are reduced, the sanitation status of the afforestation land is improved, the deep structure of soil is improved, water permeability and soil aeration are improved, water and fertilizer retention capacity is improved, and microbial activities are promoted to improve the fertility of the soil, so that the root development of the camphor trees and the rapid growth of overground parts are facilitated,the survival rate of cultivation is 95% or above, the process is simple, the labor intensity is small, and seedlings grow rapidly after transplanting.

The invention relates to a single-blade lateral milling method based on torsional ultrasonic vibration. The single-blade lateral milling method is characterized in that an ultrasound single-blade lateral milling device is driven by a machine tool spindle to rotate, a workpiece is fed on a machine tool workbench, normal contact between a tool bit and a machining face is always maintained during machining, the turning radius of a tool bit and workpiece contact point is smaller than the curvature radius of the contact point on the workpiece, the vibration direction is the tangential direction ofrotation of a single-blade milling cutter, and a single-blade lateral milling mode under torsional ultrasonic vibration is realized. The single-blade lateral milling method has the advantages that themethod is novel, the structure is unique, the process is practical, and reliability is high; the ultrasound single-blade lateral milling device is simplified in structure, downsized and convenient tomount; single-blade milling impact is improved, the method is adapted to milling of narrow and deep structures, frequency variation of a sound and vibration system is overcome, and heat radiation ofthe sound and vibration system is omitted; the problem of milling difficulty of high-hardness high-precision complex undercut hook faces and high-hardness variable-pitch spiral grooves is solved undercommon technological conditions, and the application prospect is abroad.

The invention discloses an electric power defect level identification method based on deep semantic matching, and relates to the technical field of electric power defect level identification. A traditional text classification model has a vector space model based on Boolean values, the defects that deep context semantics cannot be considered, vectors are sparse, interpretability does not exist andthe like are overcome, and the classification precision is not high for defective texts of power equipment. According to the method, a deep neural network is adopted to divide sentences into five layers of structures including an input layer, a representation layer, a matching layer, a sorting layer and an output layer, and a deep structure semantic model is achieved; then, the text is preprocessed on the input layer based on word hashing and a word segmentation model; on the basis of the deep neural network, an input layer, a representation layer and a matching layer are trained in sequence to obtain a low-dimensional representation vector of the defect text; and finally, obtaining the average defect level of the to-be-classified text based on the semantic similarity of the cosin distanceand the TopK sorting model. The level identification rate of the defect text is effectively improved, and high-precision power defect level identification is realized.

An optical cannula configured for implantation in biological tissue or other samples provides minimal invasiveness while accessing image deep structures within the sample. The cannula has a stabilizer portion that is used to affix the cannula to the sample, and an extension portion that protrudes into the sample and is either formed by, or contains, an optical probe that images a target area within the sample to an object image provided to an external optical device. The cannula has an integrated coupler portion that detachably connects to a connector of the external device, providing removable and interchangeable connection to external optical systems.

The embodiment of the invention provides a method and device for recalling standard questions according to user questions, and the method comprises the steps: carrying out the word segmentation of a current user question, and obtaining a word segmentation set corresponding to the current user question; taking the word segmentation set as the input of a pre-trained first convolutional neural network CNN model, and obtaining a first sentence representation vector of the current user question through the output of the first CNN model; obtaining each second sentence representation vector corresponding to each standard question generated according to a pre-trained second CNN model; determining a recalled standard question sentence according to the first sentence representation vector and each second sentence representation vector; wherein the first CNN model and the second CNN model are used as feature extractors in a deep structure semantic model DSSM model, and are obtained by training the DSSM model. The standard questions can be efficiently recalled according to the user questions, and the recalled standard questions have a large probability to meet the requirements of users.

Provided is a method and system for delivering a diagnostic agent to a site in the brain of a subject for imaging at least a portion of the brain site on a medical imaging system. The method and system includes a catheter device with associated lumens having diagnostic agent ports for delivering the diagnostic agent (e.g., infusate) through the lumens and advancing the diagnostic agent so as to exit out from the lumens to at least a portion of the brain site and while sealing a portion of the brain site thereby preventing the exited diagnostic agent from travelling proximally beyond the sealing location, and at the same time imaging at least a portion of the brain site during at least a portion of the sealing duration so that the brain site can be visualized on a medical imaging system. The diagnostic agent (infusate) is able to highlight borders and internal patterns of the deep structures of the brain thereby enabling direct targeting. Ultimately this leads to reduced complications, enhanced therapy, and the elimination of the need for awake surgery. The method and system provides the capability heretofore not possible to visualize the small, often indistinct regions, which will greatly improve clinical outcomes with therapeutic interventions.

The invention discloses a surface water-based preserving agent for GCr15 steel. The surface water-based preserving agent is composed of, by weight, 20-24% of deionized water, 14-18% of corrosion inhibitors, 11-16% of 10% tannic acid solutions, 2-3% of potassium sorbate, 2-4% of sodium carbonate, 6-9% of triethanolamine, 2-3% of polyisobutene, 3-6% of guar gum, 2-5% of sesbania gum, 10-15% of mixed alcohols, 2-6% of strong phosphoric acid, 3-6% of grape saccharin, 2-3% of wetting agents and 1-2% of defoaming agents. According to the surface water-based preserving agent, the selected materials are superior, the scientific processing method is adopted, and rust protection and film formation act together; the deionized water serving as a dispersion medium is matched with the other raw materials, the environment friendliness is high, and pollution is avoided. In addition, the surface water-based preserving agent is free of methylbenzene, ethylbenzene, dimethylbenzene, chromic acid, sodium nitrite and formaldehyde, low in volatile organic compound (VOC), free of pungent smells and harmless to the human body, can be easily absorbed by deep structures, can effectively protect the surface of the steel against corrosion, cavitation and scale formation, can form a protective film layer, can delay the time of generating rusts on the surface of the steel and has remarkable environmental benefits, economic benefits and social benefits.

The invention discloses a geological rock core space posture restoration device which comprises a base table, a rotary table rotationally arranged on the base table, a middle table body rotationally arranged on the rotary table and a rotary chuck rotationally arranged on the middle table body and used for clamping a rock core. The middle platform body is also provided with a displacement measuringdevice which can move along the rotation axis direction of the rotation chuck and is used for measuring the distance from the rotation axis of the rotation chuck to each point on the outer surface ofthe rock core; the rotation axis of the rotary table is perpendicular to the rotation axis of the middle table body, and the rotation axis of the rotary chuck is perpendicular to the rotation axis ofthe middle table body. The device has the characteristics of accurate restoration of the rock core space attitude, reliable parameter test precision and high automation degree, is suitable for original space attitude restoration of the common non-directional geological rock core, and assists in deep rock mass ground stress test and deep structure analysis.

A variety of deep structured decorative patterns originate with mechanical relief or etching. The present method incorporates a deeply patterned or textured etching or relief into a thin film embossing shim to simulate the look of the deep pattern or texture when used to emboss thin film or material. A transparent mold of the relief surface (such as brushed metal, engine-turned patterns, and textured glass) is formed using UV curable liquid and a transparent substrate. The relief copy in the transparent mold or overlay is mapped onto a photoresist surface or plate by shining or expanding one or more laser lights or laser beams through the transparent mold. The different heights of the relief copy of the transparent mold will cause the light to diffract/refract to form a corresponding patterned etching in the photoresist plate. The resulting photoresist plate is then metalized and electroplated to form a thin film embossing shim. The thin film embossing shim is then used with conventional embossing equipment to form thin film embossings. The embossed thin films can then be metalized and laminated onto substrate to create a product that has shifting patterns that reflect at a variety of viewing angles when (observed under normal lighting conditions) exposed to white light.