2817 results about "Computation process" patented technology

Exemplary embodiments of methods and apparatuses to dynamically redistribute computational processes in a system that includes a plurality of processing units are described. The power consumption, the performance, and the power / performance value are determined for various computational processes between a plurality of subsystems where each of the subsystems is capable of performing the computational processes. The computational processes are exemplarily graphics rendering process, image processing process, signal processing process, Bayer decoding process, or video decoding process, which can be performed by a central processing unit, a graphics processing units or a digital signal processing unit. In one embodiment, the distribution of computational processes between capable subsystems is based on a power setting, a performance setting, a dynamic setting or a value setting.

In a method of and a system for determining inaccuracy information (600) in positioning a data model (1) of virtual information in an augmented reality system, the data model (1) is projected to the image plane of a display device (4) for mixing with visual impressions of a real environment. The inaccuracy information is calculated from parameters (10-0 to 50-2, 40, 50, 60) in a mapping computing process by means of which the data model (1) is projected to the image plane of the display device (4). It is thus possible to indicate relative precise inaccuracy information in positioning the data model in an augmented reality system.

A system and method automatically generate an on-line document from raw text into an engaging, interactive form for a plurality of viewers. Unstructured articles are read from an information feed. A computation process extracts and tags proper names of people, products, organizations, and places and categorizes them. An image database is used to link these proper names with image files. The image database consists of a series of attribute-value pairs for active searching of names. A URL query string is inputted to the database to extract the location of the image in the database file system. An Extensible Markup Language (XML) file is created from the raw text of the article, the list of proper names in the processed data and the image file references. The XML file is stored in a file system. An Extensible Stylesheet Language (XSL) file provides templates containing computational relationships between the text and images. The XML and XSL style sheets are combined to generate a Hypertext Markup Language (HTML) file containing an on-line story of the unstructured articles in a Java Applet which allows the system to provide a variety of interactive behaviors for a final presentation available by a viewer from a browser.

Methods and systems for performing semiconductor metrology directly on device structures are presented. A measurement model is created based on measured training data collected from at least one device structure. The trained measurement model is used to calculate process parameter values, structure parameter values, or both, directly from measurement data collected from device structures of other wafers. In some examples, measurement data from multiple targets is collected for model building, training, and measurement. In some examples, the use of measurement data associated with multiple targets eliminates, or significantly reduces, the effect of under layers in the measurement result, and enables more accurate measurements. Measurement data collected for model building, training, and measurement may be derived from measurements performed by a combination of multiple, different measurement techniques.

Three-dimensional computer graphics systems and methods and more particularly to structure and method for a three-dimensional graphics processor and having other enhanced graphics processing features. In one embodiment the graphics processor is a Deferred Shading Graphics Processor (DSGP) comprising an AGP interface, a command fetch & decode (2000), a geometry unit (3000), a mode extraction (4000) and polygon memory (5000), a sort unit (6000) and sort memory (7000), a setup unit (8000), a cull unit (9000), a mode injection (10000), a fragment unit (11000), a texture (12000) and texture memory (13000) a phong shading (14000), a pixel unit (15000), a backend unit (1600) coupled to a frame buffer (17000). Other embodiments need not include all of these functional units, and the structures and methods of these units are applicable to other computational processes and systems as well as deferred and non-deferred shading graphical processors.

The invention discloses a three-dimensional reconstruction method based on coding structured light, comprising the following steps: 1) projecting structured light to an object to be measured, and capturing an image modulated by the object to be measured by a camera; 2) matching an optical template, comprising: (2.1) positioning the optical strip boundary, scanning along each row of the image, determining a pixel point with strong gray variation as a candidate marginal point, and searching a local domain; and (2.2) matching the optical strip: adopting a color cluster method to build a color matching proper vector, comparing image color with a projected color, and defining Euclidean distance between the color proper vector and the cluster center to distribute the colors of red, green, blue and white of the candidate optical strip; and 3) using a calibrated system parameter for three-dimensional reconstruction of the object to be measured, determining the relation between a space point coordinate and the image coordinate point thereof by the calibrated conversion matrix parameter; and restoring three-dimensional spatial coordinate from the image coordinate of a feature point. The invention can simplify calculation process and has high matching precision and high reconstruction precision.

A decision system for providing customized deployment architectures to users of a cloud computing environment. The decision system can identify one or more parameters for analyzing applications and processes running in a cloud, monitor the applications and processes executed in the cloud, and collect information such as usage of cloud resources, number and type of computing processes instantiated, software programs utilized by the computing processes. The decision system can then generate customized deployment architectures based on the collected information.

The invention discloses a coarse-grained reconfigurable system-oriented convolution neural network loop convolution calculation data reuse system, including a main controller and a connection control module, an input data reuse module, a convolution loop operation processing array, and a data transmission Pathways in four parts. During the convolution cycle operation, the essence is to multiply multiple two-dimensional input data matrices with multiple two-dimensional weight matrices. Generally, these matrices are large in size, and the multiplication takes up most of the time of the entire convolution calculation. The present invention utilizes a coarse-grained reconfigurable array system to complete the convolution calculation process. After receiving the convolution operation request instruction, the register rotation method is used to fully explore the reusability of the input data in the convolution cycle calculation process, which improves the data utilization rate. It also reduces the bandwidth access pressure, and the designed array unit is configurable, and can complete convolution operations with different cyclic convolution scales and step sizes.

A method and an apparatus for easily and efficiently setting an evaluation index suitable for an evaluation purpose when a business or a business portfolio is evaluated. In the method and apparatus, investment distribution is easily and visually confirmed during computation of the evaluation index when an object to be computed is the business portfolio. A table for defining a computation expression model for the evaluation index and a correspondence relationship between the evaluation index and the evaluation purpose is managed by a database to extract and provide an evaluation index model conforming to the evaluation purpose specified by a user. When the business portfolio is evaluated, business investment distributions are represented by a pie chart. The investment distribution information is updated and the evaluation index is computed, by adjusting a distribution surface of the represented pie chart.

Torque estimation techniques in the real-time basis for engine control and diagnostics applications using the measurement of crankshaft speed variation are disclosed. Two different torque estimation approaches are disclosed—“Stochastic Analysis” and “Frequency Analysis.” An estimation model function consisting of three primary variables representing crankshaft dynamics such as crankshaft position, speed, and acceleration is used for each estimation approach. The torque estimation method are independent of the engine inputs (air, fuel, and spark). Both approaches have been analyzed and compared with respect to estimation accuracy and computational requirements, and feasibility for the real-time engine diagnostics and control applications. Results show that both methods permits estimations of the indicated torque based on the crankshaft speed measurement while providing not only accurate but also relatively fast estimations during the computation processes.

The invention provides a deep convolutional neural network computing method applicable to hardware design implementation. The deep convolutional neural network computing method puts forward that a deep convolutional neural network computing structure is readjusted in advance through relevant adjustment parameters and breaks through the constraint that a computing window structure is fixed in a traditional convolutional neural network, so that data, involved in computation, in every computing layer can reach firstly, and computing parallelism in the deep convolutional neural network and streamlining among the computing layers are fully exploited to reduce storage of a great number of intermediate results effectively. The deep convolutional neural network computing method applicable to hardware design implementation has the advantages that a deep convolutional neural network computing structure adjusted by the method is more beneficial to implementation of efficient parallel streamlining in terms of special hardware design, and the problems of resource waste and effective computing delay due to various filling operations during computation are solved effectively, so that system power consumption can be reduced effectively and computing and processing speed can be increased greatly.

A method for monitoring performance of an advanced process control system for at least one process output includes calculating a variance of a prediction error for a processing performance and / or a probability for violating specification limits of the processing performance of the at least one process output. If the variance of the prediction error is calculated, the method also includes calculating a model health index. If the probability for violating specification limits is calculated, the method further includes calculating a process health index.

The present invention provides a neural network processor. The processor comprises at least one computing unit consisting of a host processor and a computing array, wherein the computing array is organized into a row and column two-dimensional matrix by a plurality of processing units, the host processor controls to load neuron data and weight values into the computing array, and the processing unit carries out multiply-accumulate operation on the received neuron data and weight values and passes neuron data and weight values in different directions to a next-level processing unit; and a control unit, for controlling the computing unit to carry out relevant computation of a neural network. By virtue of the computing device provided by the present invention, the computation speed of the neural network can be accelerated and the bandwidth requirements in the computation process can be reduced.

The invention relates to a submerged line tracking method for flood evolution simulation in a complicated river channel landform area. The submerged line tracking method comprises four process of generating a digital river channel multi-resolution grid segmentation model; modeling and computing a simulation forecast model for a drainage area flood evolution state, wherein the process is operated in parallel with the process of generating the digital river channel multi-resolution grid segmentation model; accurately searching the submerging boundary of the complicated river channel landform; and dynamically identifying a drainage area flood evolution submerged area. The submerged line tracking method performs accurate searching on the submerging boundary of the complicated river channel landform through establishing the digital river channel multi-resolution grid segmentation model and the simulation forecast model, thereby realizing dynamic identification on the drainage area flood evolution submerged area, accurately simulating complicated river channels at various landforms, and quickly simulating a flood submerged area. Quick, accurate and visual forecast on time, water depths and flow and even flow velocity in flood evolution to different river segments are realized. A scientific basis is laid for disaster prevention and mitigation operation before flooding.

The invention discloses a method and a system for distinguishing a real face and a picture face. Two cameras are adopted to acquire images. Partial depth information of the acquired images is reestablished via the two cameras so that an objective of distinguishing the picture face and the real face by using the cameras is realized. Meanwhile, a mode (i.e. the two cameras are adopted to acquire the images) based on binocular vision is adopted to realize matching of the face, extracting of characteristic points and reestablishing of three-dimensional coordinates of the characteristic points so that the method and the system for distinguishing the real face and the picture face are high is precision. Besides, a small amount of face characteristic points are used in the calculation process so that an algorithm has a relatively high speed as a whole.

A method for monitoring performance of an advanced process control system for at least one process output includes calculating a variance of a prediction error for a processing performance and / or a probability for violating specification limits of the processing performance of the at least one process output. If the variance of the prediction error is calculated, the method also includes calculating a model health index. If the probability for violating specification limits is calculated, the method further includes calculating a process health index.

A method for control in a process control system, where a graphical user interface of said process control system is arranged with graphic representations of one or more control objects for monitoring and / or control of a process or an equipment. The control system GUI is displayed by a control system client application on a computer or workstation. In addition, one or more applications for monitoring and / or controlling at least one said control object are run in a computing process environment separate from said process control system. No code components execute in the same processing space as that running the GUI for the process control system. Any flaw or error in the extended applications cannot affect the control system program controlling an industrial process in real-time. A control system and a computer program are also disclosed.

The invention discloses a monocular vision range finding method based on geometric relation, and the method comprises the following steps: S1, calculating the distance between a section straight line of a target object and a camera lens in a horizontal plane; S2, calculating the distance between a target object and the optical axis of a monocular lens through the triangle similarity principle. According to the projection model of a camera, the method obtains the relation between a road surface coordinate system and an image coordinate system through geometric derivation, is simple in calculation process, is low in calculation complexity, and is lower in requirements for hardware. Meanwhile, the method can reach higher precision, and the error of the method is less than 0.5m. The method can effectively calculate the distance between the front vehicles or other obstacles under the conditions that the hardware configuration condition is very simple and almost no additional load and electric quantity is added, thereby achieving the effective avoidance. The method and device are good in application prospect, such as the technical field of automobile automatic driving or intelligent robot walking.

A matrix multiplication module and matrix multiplication method are provided that use a variable number of multiplier-accumulator units based on the amount of data elements of the matrices are available or needed for processing at a particular point or stage in the computation process. As more data elements become available or are needed, more multiplier-accumulator units are used to perform the necessary multiplication and addition operations. To multiply an N×M matrix by an M×N matrix, the total (maximum) number of used MAC units is “2*N−1”. The number of MAC units used starts with one (1) and increases by two at each computation stage, that is, at the beginning of reading of data elements for each new row of the first matrix. The sequence of the number of MAC units is {1, 3, 5, . . . , 2*N−1} for computation stages each of which corresponds to reading of data elements for each new row of the left hand matrix, also called the first matrix. For the multiplication of two 8×8 matrices, the performance is 16 floating point operations per clock cycle. For an FPGA running at 100 MHz, the performance is 1.6 Giga floating point operations per second. The performance increases with the increase of the clock frequency and the use of larger matrices when FPGA resources permit. Very large matrices are partitioned into smaller blocks to fit in the FPGA resources. Results from the multiplication of sub-matrices are combined to form the final result of the large matrices.

The invention relates to an object deformation real-time simulation graphic processing technique, particularly a soft tissue deformation simulation method based on coupling of mesh-free Galerkin and mass spring, which comprises the following steps: in the pretreatment process, establishing a linear viscoelasticity biomechanical model for soft tissues; in the deformation computation process, dynamically partitioning a mesh-free region and a mass spring region according to the load carried by the soft tissues, establishing a transitional unit of the connection region between the mesh-free region and mass spring region, and constructing a transitional unit approximation displacement function, thereby implementing self-adapting coupling of a mesh-free Galerkin method and a mass spring method;and in the after-treatment process, outputting the state of the mass or node of each time step in the deformation process onto a screen, carrying out illumination rendering, and finally displaying the real-time deformation process of the soft tissue organ under stressed conditions on the screen, thereby implementing visualization effect of dynamic deformation. By utilizing the advantage of high efficiency in the mass spring method and the advantages of high precision and no need of mesh reconstruction in the mesh-free Galerkin method, the invention overcomes the defect that the Galerkin method is not suitable for solving a large-scale problem, thereby effectively lowering the complexity of computation in the soft tissue deformation simulation and enhancing the operation efficiency.

Disclosed is a system and a method for combining the computational resources of numerous embedded devices to enable any of them to perform complex tasks like speech recognition or natural language understanding. A distinguished master device communicates with a network of embedded devices, and organizes them as the nodes of a neural network. To each node (embedded device) in the neural network, the master device sends the activation function for that node and the connectivity pattern for that node. The master device sends the inputs for the network to the distinguished input nodes of the network. During computation, each node computes the activation function of all of its inputs and sends its activation to all the nodes to which it needs to send output to. The outputs of the neural network are sent to the master device. Thus, the network of embedded devices can perform any computation (like speech recognition, natural language understanding, etc.) which can be mapped onto a neural network model.

A systolic linear-array modular multiplier is provided, which can perform the modular multiplication algorithm of P. L. Montgomery more efficiently. The total execution time for n-bit modular multiplication is 2n+11 cycles. The modular multiplier includes a linear array of processing elements which is constructed based on a pipeline architecture that can reduce the computation procedure by one clock period. Each of the processing elements is simple in structure, which is composed of four full adders and fourteen flip-flops. For n-bit modular multiplication, a total number of 46n+184 gates is required, which is substantially less as compared to the prior art, so that manufacturing cost of the modular multiplier can be significantly reduced. These features make the modular multiplier suitable for use in VLSI implementation of modular exponentiation which is the kernel computation in many public-key cryptosystems, such as the RSA (Rivest-Shamir-Adleman) system. With the 0.8 mu m CMOS technology, a clock signal up to 180 MHz can be used. In average, for n-bit modular multiplication, the encryption speed can reach 116 Kbit / s (kilobits per second), which is substantially twice that achieved by the prior art.

The invention relates to a self-positioning technology in wireless sensor network, especially a distributed method without anchor position. Various network nodes measure the values of distance from each other to get the neighbor information within the scope of their respective perception, and exchange their estimated coordinates many times. The invention adopts the distributed calculation method, firstly estimates the approximate location of nodes, and then each node itself makes iterative calculation of the estimated coordinate location by using the particle spring optimization model. The invention provides a practical method for the wireless sensor network to get geographic information. In addition, the calculation process is implemented on each node without the large cost of communication required by the centralized control, and has a very low complexity of calculation, so that the consumption of positioning calculation is lower, which is suitable for the self-organizing wireless sensor network with nodes measuring distance.

The invention discloses a control method of a three-phase grid-connected inverter based on a modified proportional resonant regulator. The method comprises the following steps of: firstly, computing according to given active and reactive powers of the inverter and a voltage of a power grid to obtain required d-axis and q-axis currents under a synchronous coordinate system, and performing coordinate transformation by combining with phase information which is measured by a phase locking loop and a measured three-phase voltage signal of the power grid to obtain a current reference signal under a stationary alpha-beta coordinate system; secondly, performing coordinate transformation on an output current signal of the inverter to the alpha-beta coordinate system, controlling and tracking the current reference signal by the modified proportional resonant regulator with a harmonic compensator; and finally, returning to an abc coordinate system by coordinate transformation to generate a pulse-width modulation (PWM) pulse-control three-phase full-bridge inverter output, so as to make a distributed generation system grid-connected for power generation. By the method, a large number of complicated computation processes are eliminated, and a control structure is simplified. The control method of the three-phase grid-connected inverter based on the modified proportional resonant regulator is an accurate and high-efficiency novel inverter control method and has a bright development future.

A method and system for site path evaluation using web session clustering is provided. The method identifies properties of a plurality of web site traversal paths. Data is acquired from a plurality of sessions corresponding to at least a portion of the plurality of web page traversal paths. Portions of the web site traversal paths are grouped into a unified web page category. The plurality of sessions is clustered into a plurality of web session clusters according to a similarity measure. One of the plurality of web session clusters most closely exhibiting at least one predefined metric is selected for analysis of the propertied of a web page traversal path contained therein. A system includes a plurality of web pages, a monitoring program and a computational process configured to receive data and identify properties of the plurality of traversal paths.

The invention provides a method and device for route programming in a dynamic unknown environment. In the moving process of a robot along a preset optimal route, a collision-free optimal route is obtained through calculation by using a linear programming gradient method and through environment information acquired by a preset rolling window, and then partial routes are obtained through calculation carried out on the obtained collision-free optimal route and acquired perceptual information in the rolling window. In the calculation process, the rolling window is used for reducing calculation amount and improving efficiency, and the linear programming gradient method is applied to ensure global convergence and to prevent from being caught in the problems of local minimum and oscillation.

The invention discloses an on-line calibration system for an external parameter of a monocular camera. The system comprises a straight line detection module, a lane extraction module, a fast-perspective-removal (FPR) module, an optimal value iterative module, and an external parameter output module. The straight line detection module obtains an effective straight line position of an inputted infrared image; the lane extraction module carries out effective straight line searching and determination at a possibly occurring lane position; the FPR module carried out perspective transformation removing on a lane based on an imaging principle; the optimal value iterative module carries out loop iteration to enable a fitness function to be optimized, thereby obtaining an angle with the smallest error relative to a practical value; and the external parameter output module is used for outputting the angle obtained by optimal iteration. No calibration object is needed; the influence on the calibration precision by any external factor can be reduced; dependence on the straight line information of the lane is only needed during the calculation process; the flexibility is high; and on-line calibration on the camera can be realized. Moreover, no control by an external device and no strict experiment condition are needed during the calculation process. The system and method are simple and convenient to realize.

The invention discloses a method for identifying and locating fruit to be picked based on depth target detection, comprising the following steps: image acquisition, image annotation. data Set Preparation. Feature extraction includes using depth convolution neural network CNN to extract the fruit features in the image first; VGG16 is used as a convolution neural network for feature extraction. Model training, For all the training samples, the feature maps on each convolution layer are extracted firstly, and the prediction boxes are generated for the extracted feature maps, the categories of thetargets in the prediction boxes are calculated, the distance between the prediction boxes and the real annotation boxes is calculated, and the joint category loss and the prediction box offset loss are used as the target loss function of the training during the training process. A training is completed based on the complete one-time calculation process described above, and the training ends whenthe number of training times reaches a predetermined threshold or the loss is less than the predetermined threshold.

The invention discloses a neural network online learning system based on a memristor. A pulse coding mode of a K-bit input vector is improved; the coding pulse corresponding to each bit is expanded into 2m pulses; In this way, the total number of needed coded pulses is K * 2m, each bit of weighted summation calculation is actually carried out for 2m times, finally, summation averaging operation iscarried out at the output end, the influence of accidental factors and noise on the calculation result in the calculation process is reduced through the mode, and therefore the calculation precisionis improved. The memristor array is simultaneously used for forward weighted summation calculation and weight small storage in the neural network; Different from offline learning, The weight in the memristor array needs to be updated once every time a signal is input through online learning, the variable quantity of the weight is mapped into the number of pulses, then the pulses are applied for one-time weight write-in operation, the neural network training speed can be increased, the hardware cost can be reduced, and the power consumption of neural network training is reduced.