928 results about "Deep level" patented technology

A method and system for decreasing power consumption in memory arrays having usage-driven power management provides decreased power consumption in the memory array of a processing system. Per-page usage information is gathered on memory by a memory controller and periodically evaluated by software. The software distinguishes between more frequently accessed pages and less frequently accessed pages by analyzing the gathered usage information and periodically migrates physical memory pages in order to group less frequently accessed pages and more frequently access pages in separately power-managed memory ranks. When used in conjunction with a usage-driven power management mechanism, the ranks containing the less frequently accessed pages can enter deeper power-saving states and/or any power-saving state for longer periods. Operation may be further enhanced by using packed allocation in the memory ranks containing the less-frequently accessed pages and scattered allocation in the memory ranks having more frequently accessed pages.

A magnetic nerve stimulator system is comprised of a core constructed from a material having a high field saturation with a coil winding. A thyrister capacitive discharge circuit pulses the device. A rapidly changing magnetic field is guided by the core, preferably vanadium permendur. For task specific excitation of various nerve groups, specially constructed cores allow for excitation of nerves at deeper levels with higher efficiency than is possible with air-core stimulators. Among the applications possible with this invention are treatment of incontinence, rehabilitation of large muscle groups in the leg and arm, and excitation of abdominal wall muscle groups to aid in weight loss and metabolic rate increase. A C-shape is employed for focussing the stimulation as desired.

The invention discloses a depth image human body joint positioning method based on a convolution nerve network. The method is characterized by comprising a training process and an identification process. The training process comprises the following steps: 1, inputting a training sample; 2, initializing a deep convolution nerve network and its parameters, wherein the parameters comprise a weight and a bias of each layer edge; and 3, by use of a forward algorithm and a backward algorithm, learning the parameters of the constructed convolution nerve network. The identification process comprises the following steps: 4, inputting a test sample; and 5, performing regression on the input test sample by use of the trained convolution nerve network to find positions of human body joints. According to the invention, by use of the deep convolution nerve network and large data, multiple challenges such as shielding, noise and the like can be resisted, and the accuracy is quite high; and at the same time, by means of parallel calculation, the effect of accurately positioning the human body joints in real time can be realized.

The invention discloses a deep learning-based blog text abstract generation method. The method comprises the following steps of: crawling blog data; preprocessing the crawled blog data and selecting blog text data; converting the selected blog text data into vector matrix data according to a Chinese word vector dictionary; constructing a deep learning encoder-decoder model, separately training an encoder and a decoder of the model, and connecting the encoder and the decoder for use after the training is completed; and repeating the steps S01 to S03 to obtain generated data, and generating a predicted abstract from the generated data through the trained model. According to the method, text abstracts of blogs are automatically generated on the basis of a deep learning frame encoder-decoder, and deeper semantic relation of the blogs can be obtained at the same time. The generated text abstracts can visually display the main content of the current blog, so that the text abstracts have wide application prospect.

The invention relates to a low-density ceramic proppant and a preparation method thereof, which belong to the field of exploring petroleum and natural gas. The preparation method of the low-density ceramic proppant comprises the steps of forming an inner core made of bauxite and coal gangue and a shell made of bauxite and trimanganese tetroxide into a ball and sintering to obtain low-density ceramic. The ceramic has higher strength and low density and is applicable to a ceramic proppant for fracturing a deep-level oil-gas well with high closed pressure and low permeability; the sintering temperature is relatively reduced in the range of 1,300-1,380DEG C, so that thus the energy consumption is effectively reduced and the energy can be saved by 20 percent; and meanwhile, the effect with low crushing rate can be achieved.

The invention provides an RGBD image semantic segmentation method. The method comprises the following steps of: S1, acquiring data of a training sample; S2, constructing a configurable depth model and inputting data of the training sample into the depth model to train the depth model; S3, obtaining a color image needing semantic segmentation and a corresponding depth image, analyzing the color image and the depth image by utilizing the trained depth model, and predicting an object to which each pixel in an RGBD image belongs; and S4, forming and outputting a predicted image semantic segmentation image according to a result obtained in S3. According to the method, a deep-level convolutional neural network, a long/short-time memory network and big data are utilized, so that features of color images and depth images can be effectively fused, context information in the images can be effectively mined, and high correctness is provided.

Techniques are described for modeling layered facial reflectance consisting of specular reflectance, single scattering, and shallow and deep subsurface scattering. Parameters of appropriate reflectance models can be estimated for each of these layers, e.g., from just 20 photographs recorded in a few seconds from a single view-point. Spatially-varying specular reflectance and single-scattering parameters can be extracted from polarization-difference images under spherical and point source illumination. Direct-indirect separation can be employed to decompose the remaining multiple scattering observed under cross-polarization into shallow and deep scattering components to model the light transport through multiple layers of skin. Appropriate diffusion models can be matched to the extracted shallow and deep scattering components for different regions on the face. The techniques were validated by comparing renderings of subjects to reference photographs recorded from novel viewpoints and under novel illumination conditions. Related geometry acquisition systems and software products are also described.

The invention discloses a behavior recognition technical method based on deep learning, which solves a problem that the intelligentization of a video monitoring system in the prior art needs to be improved. The method comprises the following steps of constructing a deeper space-time double-flow CNN-GRU neural network model by adopting a mode of combining a double-flow convolutional neural networkand a GRU network; extracting the time domain and space domain features of the video; extracting the long-time serialization characteristics of the spatial-temporal characteristic sequence according to the capability of the GRU network for memorizing information, and carrying out the behavior recognition of a video through employing a softmax classifier; proposing a new related entropy-based lossfunction; and with the help of a method for processing the mass information by referring to a human brain visual nerve attention mechanism, introducing an attention mechanism before a space-time double-flow CNN-GRU neural network model performs space-time feature fusion. The accuracy of the model provided by the technology is 61.5%, and compared with an algorithm based on a double-flow convolutional neural network, the recognition rate is improved to a certain extent.

The invention relates to substrates of semi-insulating silicon carbide used for semiconductor devices and a method for making the same. The substrates have a resistivity above 106 Ohm-cm, and preferably above 108 Ohm-cm, and most preferably above 109 Ohm-cm, and a capacitance below 5 pF/mm2 and preferably below 1 pF/mm2. The electrical properties of the substrates are controlled by a small amount of added deep level impurity, large enough in concentration to dominate the electrical behavior, but small enough to avoid structural defects. The substrates have concentrations of unintentional background impurities, including shallow donors and acceptors, purposely reduced to below 5·1016 cm−3, and preferably to below 1·1016 cm−3, and the concentration of deep level impurity is higher, and preferably at least two times higher, than the difference between the concentrations of shallow acceptors and shallow donors. The deep level impurity comprises one of selected metals from the periodic groups IB, IIB, IIIB, IVB, VB, VIB, VIIB and VIIIB. Vanadium is a preferred deep level element. In addition to controlling the resistivity and capacitance, a further advantage of the invention is an increase in electrical uniformity over the entire crystal and reduction in the density of crystal defects.

A system for generating electrical energy using a naturally occurring temperature difference is disclosed. The system provides electrical energy by thermally conduit a geothermal heat source and cold deep-level water to opposing sides of a thermoelectric element. The thermoelectric element generates electrical energy based on the temperature difference between these two surfaces.

The invention provides an event graph construction method based on social media. The method comprises the steps of firstly, performing multi-source data preprocessing; secondly performing multi-sourceevent information extraction on preprocessed data; thirdly, judging a relation between events through event relation evaluation; and finally, performing entity information fusion, and performing attribute complementation on entities in a heterogeneous graph. According to the method, the events are regarded as the abstract entities; basic elements of the events are extracted based on text data pairs of the social media; the events are associated; an existing structured knowledge base is fused to construct an event graph, so that more comprehensive and more direct event-oriented information retrieval services can be provided, and by converting the research of traditional non-structured text contents into the graph-based research, the exploration of deeper information is facilitated.

An apparatus and method for use in a cased well having a production tubing assembly installed in the well and terminating with a connection means at a subterranean location. The apparatus and method utilizes a spoolable tubular assembly for installation extending into the well to the connection means wherein the tubular assembly has a plurality of conductors formed connected to electrically operable valve, sensors and tubing latch in the tubular assembly. In the method the assembly is spooled into the well with the tubing latch being operated from outside the well through the tubing conductors to selectively connect and disconnect the assembly from the connector means. Operating the salve and sensors from outside the well when the assembly is installed. The assembly tubing is preferably made of a non-conductive, and preferably composite material that enclosed at least one, and preferably a plurality of electrical conductors.

The invention discloses a network text segmenting method based on the genetic algorithm, used for segmenting short network texts. The method comprises the following steps of: evaluating a Latent Dirichlet allocation (LDA) model corresponding to a corpus by using a Gibbs sampling method, inferring latent topic information using the model, representing texts by using the latent topic information; then transforming a text-segmenting process into a multi-target optimum process by using a parallel genetic algorithm, and calculating the coherency of segmented units, the divergence among the segmented units and fitness functions by using deeper semantic information; and carrying out the genetic iteration of the text segmenting process, and determining whether the segmenting process terminates based on the similarity among multi-iteration results or the upper limit of iterations to obtain the global optimal solution for segmenting the texts. Therefore, the invention improves the accuracy for segmenting the short network texts.

The invention provides a deep learning-based emotion recognition method and system. The system includes a face image acquisition module, a deep learning-based facial expression recognition module and a facial expression early warning module. The facial images of employees are acquired when the employees punch in every time; the emotions of the employees are analyzed through adopting a deep learning algorithm-based facial expression analysis algorithm, and the emotions are compared with historical emotions; and when the emotions are abnormal, the system sends alarm information to relevant personnel. According to the deep learning-based emotion recognition method and system of the present invention, the deep learning algorithm is adopted to perform emotional analysis on the employees, and therefore, deep-level humanistic care can be provided for the employees.

The invention discloses an automatic classification method for electrocardiogram signals. The method is achieved according to the following steps of firstly, obtaining electrocardiogram signals of a human body, conducting filtering on the electrocardiogram signals, and detecting R waves of the electrocardiogram signals where filtering is conducted; secondly, establishing a data set after the R waves are detected, wherein the data set is composed of multiple sets of cardiac beat data, and each set of cardiac beat data has a label; thirdly, establishing a sparse automatic coding deep learning network; fourthly, training the sparse automatic coding deep learning network step by step; fifthly, inputting the to-be-measured cardiac beat data into the sparse automatic coding deep learning network according to the network weight, obtained in the fourth step, of the first hidden layer, the network weight, obtained in the fourth step, of the second hidden layer and the network weight, obtained in the fourth step, of the softmax classifier so as to obtain cardiac data which are output in a classified mode. The sparse automatic coding deep learning network is applied to the classification of the cardiac beat data, and by means of the autonomous leaning capacity and the deep characteristic excavation characteristic of the sparse automatic coding deep learning network, deeper characteristics of signals are extracted, and the cardiac beat data are classified.

The invention provides a method for positioning three-dimensional human body joints in monocular color videos. The method comprises the following steps of: S1, constructing a configurable depth model and importing time sequence information in the depth model; S2, collecting training samples and learning parameters of the depth model by utilizing the training samples; and S3, initializing the depth model by utilizing the parameters learnt in S2, and converting monocular color video data needing three-dimensional human body joint positioning into a plurality of frames of continuous two-dimensional images, inputting the images into the depth model to carry out analysis, and aiming at each frame of two-dimensional image, outputting three-dimensional human body joint coordinates of figures in the image. According to the method, a deep-level convolutional neural network is constructed by utilizing deep learning so as to automatically learn effective spatial-temporal features from a lot of training samples without depending on prior conditions of artificial design and human body joint structural constraints; and through the learnt effective features, the human body joint positions are directly regressed.

The invention relates to the technical field of biomedicine, and relates to a biomedicine event trigger word identification method based on characteristic automatic learning. The biomedicine event trigger word identification method comprises the following steps of 1, data pre-processing; 2, construction of an event trigger word dictionary; 3, construction of candidate trigger word examples; 4, characteristic learning by means of a convolutional neural network model; 5, training by means of a neural network model; and 6, classification of event trigger words. The biomedicine event trigger word identification method is advantaged in that 1, complex preprocessing to data is simplified, and tedious steps for carrying out a characteristic design by people are saved; 2, domain knowledge is introduced, and a lot of external resources such as unlabeled linguistic data are effectively utilized; 3, characteristic automatic learning is carried out by means of a convolutional neural network, manual intervention is reduced, sentence level characteristics in a deeper level can be excavated and explored, through the fusion of local characteristics, implicit global characteristics are discovered, and the category of trigger words can be identified; and 4, a better experiment result is obtained in MLEE linguistic data, and the whole performance on event trigger word detection is improved.

The invention provides a web vulnerability scanning method and device based on an infiltration technology. The scanning device comprises a client, a Web browser, a scanning host and a Web server which are sequentially connected and is characterized in that the scanning host comprises a control platform, a scanning mode setting module, a scanning module and a vulnerability library, a scanning modeis set in the scanning mode setting module by a user through the control platform, a scanning depth is extended from a first-class scanning to a second-class scanning or even a deeper scanning in thescanning module, and finally, a target website is scanned by using the vulnerability library and a test report is given. By virtue of the complete crawling-based website security vulnerability scanning method, improvements on a further treatment (strengthening the depth of detection) after vulnerabilities are detected and the flexibility of a detection mode are realized, and a comprehensive and effective solution is provided for system/network vulnerability scanning and processing.

The invention discloses a boiler combustion optimizing control system and a boiler combustion optimizing control method. The method comprises the following steps of: on the basis of the equilibrium distribution and transformation for coal powder in a fire coal unit, accurately measuring parameters such as wind, powder, ash and the like by using a measuring device; analyzing a history behaviour of operation by the deep analysis and digging of data by using the acquired real-time history data of a boiler and taking the work condition optimization as a basic optimizing method; establishing mathematical models among operation parameter, status input parameter and parameters such as boiler efficiency, NOx and the like in the combustion process to obtain a unit operation mode knowledge base; performing the whole plant energy-saving and emission-reducing comprehensive assessment and diagnosis; analyzing the unit operation potential; and providing a knowledge base and a rule for optimizing operation; optimizing boiler combustion parameter configuration aiming at different combustion indexes or an index combination to realize the optimization of multiple optimizing objects and propose an energy-saving and emission-reducing implementation scheme and measurement in a classification mode. The method not only can realize closed-loop optimizing control but also can realize on-line optimizing guide.

The method includes following steps: (1) constructing training set; (2) adding seed web page into queue to be climbed; (3) climbing all URL in queue to be climbed, parsing new climbing web page, and extracting all links; (4) based on class of training set, and then based on class of link to determine priorities of each new link; (5) viewing preferential queue according to sequence, and putting all URL in not void queue with highest priority into queue to be climbed, then jumping to step (3), and holding other queues unchanged; (6) the method ends climbing when all preferential queues are void, or specified climbing cycle index is reached. Using useful information including anchor character and URL, the method analyzes hierarchy of link, and carries out analyzing and climbing topology of web pages in deep layer.

The invention belongs to the field of computer security, and relates to a method for detecting web page Trojan horse based on program execution characteristics, which comprises the following steps: using web crawlers to capture source codes of a web page; then obtaining a recognizable script program through multilevel decoding; carrying out disassembling processing on the script program to obtainassembled source codes while reserving the script program; then, judging whether a large number of filled invalid instructions, calling system level functions and obvious URL links exist in the sourcecodes; and finally detecting whether the Trojan horse exists in the web page through the assembled source codes in a deep level. Because most of the prior web pages with the Trojan horse are embeddedwith ShellCode, to execute the ShellCode in the web pages in a local computer, system vulnerability is needed to realize buffer overflow and enable the program to skip onto the ShellCode code segment. Thus, only by analyzing the condition of executing the ShellCode, and analyzing the source codes according to the execution characteristics, whether the web page to be detected is the web page Trojan horse can be quickly detected.