33 results about "Multi-scale approaches" patented technology

Aspects relate to systems and methods for improving the operation of computer-implemented neural networks. Some aspects relate to training a neural network using a compressed representation of the inputs either through efficient discretization of the inputs, or choice of compression. This approach allows a multiscale approach where the input discretization is adaptively changed during the learning process, or the loss of the compression is changed during the training. Once a network has been trained, the approach allows for efficient predictions and classifications using compressed inputs. One approach can generate a larger more diverse training dataset based on both simulations from physical models, as well as incorporating domain expertise and other available information. One approach can automatically match the documents to the list, while still allowing a user to input information to update and correct the matching process.

The invention discloses a computer soft hardware system based on a variation multiscale method, belonging to the technical field of high-performance computers. The computer soft hardware system provides a behavior of describing a system bottom layer with a discrete model unit and a calculation mode of describing a stability condition observed by a large quantity of model unit group behaviors with variation restraint aiming at the multiscale structure and the discrete essence of a complex system in the actual world, thereby saving a large amount of calculated amount compared with simple bottom layer discrete simulation and outstandingly enhancing the calculation accuracy compared with single-level equalization simulation. Multilevel calculation hardware with the design directly reflects the calculation mode, adopts the interaction among a large quantity of simple processor calculation model units and acts on the variation restraint of the large quantity of simple processor calculation model units with complex general processor calculation, thereby outstandingly enhancing the resolving speed and the calculating scale of similar problems.

A computer simulation analysis method suitable for describing the mechanical behavior of multiphase composites based on the real microstructure of materials relates to a multidisciplinary field such as computational material science, simulation and high throughput calculation. Through the first-principles calculation under nano scale, the molecular dynamics simulation under micro scale, and the thermodynamic calculation under mesoscopic scale, various physical parameters needed for the finite element simulation under macro scale can be obtained, including the elastic and plastic physical parameters of each phase in the composite at different temperature and different grain sizes. Focused ion beam experiment and image processing are adopted to obtain real material microstructure. Through the parameter coupling and parameter transfer among the calculated results of various scales, combining the microstructure of the material, stress-strain relationship, stress distribution and its evolution law, plastic deformation and other mechanical behaviors of the multiphase composites under complex stress and different temperature can be simulated.

Aspects relate to systems and methods for improving the operation of computer-implemented neural networks. Some aspects relate to training a neural network using a compressed representation of the inputs either through efficient discretization of the inputs, or choice of compression. This approach allows a multiscale approach where the input discretization is adaptively changed during the learning process, or the loss of the compression is changed during the training. Once a network has been trained, the approach allows for efficient predictions and classifications using compressed inputs. One approach can generate a larger more diverse training dataset based on both simulations from physical models, as well as incorporating domain expertise and other available information. One approach can automatically match the documents to the list, while still allowing a user to input information to update and correct the matching process.

A multi-scale method to efficiently determine the fine-scale saturation arising from multi-phase flow in a subsurface reservoir is disclosed. The method includes providing a simulation model that includes a fine-scale grid defining a plurality of fine-scale cells, and a coarse-scale grid defining a plurality of coarse-scale cells that are aggregates of the fine-scale cells. The coarse-scale cells are partitioned into saturation regions responsive to velocity and / or saturation changes from the saturation front. A fine-scale saturation is determined for each region and the saturation regions are assembled to obtain a fine-scale saturation distribution. A visual display can be output responsive to the fine-scale saturation distribution.

The invention discloses a remote-sensing artificial ground object identifying method based on a semantic tree model of an object. The remote-sensing artificial ground object identifying method comprises the steps of: establishing a remote-sensing ground object representative image set; splitting images in the remote-sensing ground object representative image set by adopting a multi-scale method, and obtaining an object tree of each image; modeling for each node of each object tree by adopting an LDA (linear discriminant analysis) method, and computing implied semantic features contained in the tree node objects; obtaining the object tree sets of all the images in the representative set to learn each pair of object trees in a matching way, and extracting the common maximum sub-trees from the object trees; combining all the common maximum sub-trees together by adopting a step-by-step adding method, and forming an object semantic tree of the category of the described ground object; and identifying the artificial ground object according to the object semantic tree and obtaining the area in which the ground object is positioned. The remote-sensing artificial ground object identifying method disclosed by the invention can be used for mostly effectively processing the artificial ground objects in the condition of high-resolution remote-sensing images; the identification result is accurate, the robustness is good, the applicability is high, and manual work is reduced.

The invention relates to the field of damage prediction of a composite laminate, aiming at providing a multi-scale prediction method for progressive failure of a composite laminate under low-velocityimpact. The method comprises the following steps: establishing a low-velocity impact finite element model containing a punching hammer, a composite laminate and a support plate; establishing a multi-scale prediction method of composite materials based on micromechanics; solving stress, strain and damage of the composite laminate under s low-velocity impact load based on a multi-scale method; and calculating the low-velocity impact to further obtain the impact force, displacement, velocity and acceleration. A macroscopic progressive failure behavior is predicted from a microscopic point of view; and compared with a research method based on the macromechanics theory of a composite material, the invention considers the influence of the difference of mechanical properties of microscopic fibersand matrixes on the macromechanics properties of the composite laminate, and more essentially and more accurately predicts the progressive failure behavior of the composite laminate under the low-velocity impact load.

Aspects relate to systems and methods for improving the operation of computer-implemented neural networks. Some aspects relate to training a neural network using a compressed representation of the inputs either through efficient discretization of the inputs, or choice of compression. This approach allows a multiscale approach where the input discretization is adaptively changed during the learning process, or the loss of the compression is changed during the training. Once a network has been trained, the approach allows for efficient predictions and classifications using compressed inputs. One approach can generate a larger more diverse training dataset based on both simulations from physical models, as well as incorporating domain expertise and other available information. One approach can automatically match the documents to the list, while still allowing a user to input information to update and correct the matching process.

The invention puts forward a hot-line work robot bolt identification method based on random Hough transform and a SVM (Support Vector Machine). The hot-line work robot bolt identification method comprises the following steps that using a binocular camera to collect an operation scene binocular image including bolts; according to the image, judging whether haze is in the presence or not, if the haze is in the presence, adopting a multiscale Retinex method with color restoration to carry out enhancement and restoration on the image, and then, extracting image edges; if the haze is not in the presence, further judging whether a highlight phenomenon is in the presence in the image or not, if the highlight phenomenon is in the presence, adopting a homomorphic filtering method to enhance the image, then, extracting the image edges, and if the highlight phenomenon is not in the presence, directly extracting the image edges; adopting a method based on random Hough transform to carry out ellipse fitting on image edge information; and if a plurality of ellipses are fit, adopting a target identification method based on the SVM to carry out classification identification on a plurality of ellipse fitting results. By use of the method, bolts can be quickly and accurately identified under a complex background.

The invention provides a one-dimensional photonic crystal beam structure band gap designing method based on a wavelet finite element model. The wavelet finite element model uses the section B sample wavelet to be combined with a finite element method, a BSWI scaling function is used for replacing the polynomial interpolation of the traditional finite element, and the unit cell technology and periodic boundary condition PBCs are combined to build a real symmetrical feature value problem about the one-dimensional photonic crystal dispersing structure, so the band gap feature of the photonic crystal is obtained by calculating. The wavelet finite element model of the one-dimensional photonic crystal beam structure band gap calculation is capable of absorbing the advantages of the wavelet multi-scale approach feature and the complex solution domain of the finite element method, and obtaining a numerical calculation model with high precision and rapid convergence. The provided wavelet finite element model of the one-dimensional photonic crystal beam structure band gap design has the advantages of high precision and rapid convergence, and is suitable for the one-dimensional photonic crystal beam structure band gap design.

The present invention presents a novel application of a wavelet-based multiscale method in a nanomachining process chemical mechanical planarization (CMP) of wafer fabrication. The invention involves identification of delamination defects of low-k dielectric layers by analyzing the nonstationary acoustic emission (AE) signal collected during copper damascene (Cu-low k) CMP processes. An offline strategy and a moving window-based strategy for online implementation of the wavelet monitoring approach are developed.

The invention relates to an object pose estimation method based on self-supervised learning and template matching, and the method comprises the steps: S1, collecting a color image and a depth image ofa target object, and obtaining a corresponding color candidate image and a corresponding depth candidate image through cutting; s2, segmenting the color candidate image and the depth candidate imageby a trained self-supervised codec with a noise generator to obtain a color segmentation image and a depth segmentation image; s3, matching the color segmentation map and the depth segmentation map with a template library to obtain a matched pose; s4, finely trimming the matching pose to obtain a target object pose, wherein a color sample image and a depth sample image of the target object three-dimensional model are collected through a spherical multi-scale method; training a self-supervised codec by using the color sample graph; extracting features with pose information of the color sample map and the depth sample map under multiple scales, and constructing a template library according to the features. Compared with the prior art, the method has the advantages of being good in robustness, low in cost, free of label information and the like.

The invention discloses a fracture failure behavior prediction method of micro-nano fiber reinforced composite material. The method comprises the following steps of describing mechanical behaviors ofthe composite material through a multi-scale method, and establishing a slippage-enrichment bandwidth model of any fiber embedding length and a non-local slippage model of fibers in any direction; calculating fiber stress distribution and global load displacement response of the crack bridging area; by applying load and boundary conditions and iteratively updating cracks, acquiring a stress-strain relationship and a load displacement curve, making stress-strain information transmission from a fiber reinforced composite material mesoscopic model to macroscopic multi-field coupling and acquiring an analysis result of composite material mesoscopic local characteristics on macroscopic global load displacement response. According to the invention, the fracture failure process of the compositematerial can be accurately simulated while the bridging effect of the fibers is realized.

The present invention presents a novel application of a wavelet-based multiscale method in a nanomachining process chemical mechanical planarization (CMP) of wafer fabrication. The invention involves identification of delamination defects of low-k dielectric layers by analyzing the nonstationary acoustic emission (AE) signal collected during copper damascene (Cu-low k) CMP processes. An offline strategy and a moving window-based strategy for online implementation of the wavelet monitoring approach are developed.

The present invention presents a novel application of a wavelet-based multiscale method in a nanomachining process chemical mechanical planarization (CMP) of wafer fabrication. The invention involves identification of delamination defects of low-k dielectric layers by analyzing the nonstationary acoustic emission (AE) signal collected during copper damascene (Cu-low k) CMP processes. An offline strategy and a moving window-based strategy for online implementation of the wavelet monitoring approach are developed.

A distributed sensing and control network includes a plurality of sensing / control nodes, each of the sensing / control nodes includes a sensor, a local controller, a local memory and a wireless transmitter / receiver. The local memory stores instructions for implementing a multiscale networking protocol for wireless transmissions. The multiscale networking protocol controls at least a first factor at a first timescale and at least a second factor at a second timescale, distinct from the first timescale. The at least a first factor includes an instantiated physical-ratio-K (PRK) interference model for wireless transmissions. The at least a second factor includes at least one of a channel assignment, a node transmission signal power control, a node transmission rate control, a real-time scheduling control, and a local signal map stored in the local memory. Wherein the local signal map, together with instantiated physical-ratio-K (PRK) interference models, defines an interference relationship between a sensing / control node storing the local signal map and each other sensing / control node of the plurality of sensing / control nodes within an exclusion region of the sensing / control node storing the local signal map.

The invention discloses a blind deblurring image restoration method, and belongs to the field of image processing. According to the invention, an lq / l2 norm is utilized to serve as a regularization prior term, a multiscale method is adopted to gradually solve in an iterative mode from a coarse scale to a fine scale, high-frequency information of an estimated image is updated by using an l1 norm fidelity term when the model is solved according to a splitting method, an analytical solution is provided by adopting a closed threshold formula during restoration of a clear image, and the algorithm speed is improved. Meanwhile, a linear progressive increase weight parameter is proposed when a blurring kernel is updated, and the blurring kernel is gradually estimated from coarse to fine accordingto the multiscale method, so that the convergence of the blurring kernel is further improved, and the quality of the deblurred image is improved.

The invention relates to the technical field of image processing, in particular to a single-scale motion blurred image frame restoration method for a cab environment, which comprises the following steps: carrying out iterative training on a restoration model to obtain a trained restoration model; and inputting the motion blurred image frame and the optimal blurred kernel into a trained recovery model, recovering the motion blurred image by the recovery model, and outputting the recovered image. Compared with a multi-scale method, the method has the advantages that the overall structure of theblurred image can be better recovered, the running time can be obviously shortened, and the image frame recovery efficiency is improved.

Disclosed is a method for measuring movement and / or deformation of a study part subjected to an external stress, the method implementing a color digital image acquisition device and image correlation unit. The method includes: acquisition (250) of a plurality of color digital images of the surface of the study part by way of the color digital image acquisition device during the application of an external stress on the study part, and processing (260) of the color digital images acquired in the step of acquisition of a plurality of color digital images so as to highlight contrasting details on variable scales.

The invention discloses a method for accurately calculating the actual width of a single pixel in a crack image. The method specifically comprises the following steps: preprocessing a shot picture byadopting a multi-scale Retinex method; correcting the image by using perspective transformation; according to the actual length of a standard rectangular mark block and the pixel length occupied in the corrected image, obtaining the actual length a represented by each pixel, wherein the product of the calculated crack width pixel value and a is the actual crack width value. According to the method, the standard rectangle is obtained by preprocessing the picture and correcting the image, and the actual crack width value is calculated by indirectly establishing the proportional relation betweenthe image coordinate system and the actual coordinate system, so that the technical problem that the distance between the camera and the bridge crack is difficult to measure accurately is solved.

The invention discloses a reconstructed low-frequency data multi-scale full-waveform inversion method based on down-sampling. The method comprises the following steps: acquiring earth surface observation data; reconstructing missing low-frequency part data of the earth surface observation data through a sliding down-sampling method; performing data processing on the obtained reconstructed data; and constructing a background speed model by using the processed reconstruction data through a multi-scale method, taking the background speed model as an initial speed model of traditional full-waveform inversion, and performing iteration to obtain an inversion result. The method comprises the following steps: selecting a sampling point with the maximum cross-correlation value with real observationdata by using a sliding downsampling method to interpolate and reconstruct low-frequency data; according to the invention, the reconstructed low-frequency data is divided into two scales of 0-8Hz andfull-band time domain full waveform inversion on the full band to obtain a high-precision speed model. The modle is used as an initial model of traditional full-waveform inversion, and the cycle skipping phenomenon of full-waveform inversion is relieved.

The invention discloses a day-ahead planed multi-scale imminent generation scheduling closed-loop control method and relates to a generation scheduling closed-loop control method. At present, following up on plans strictly is required on a unit by generation scheduling closed-loop control, and regulating capacity of the unit is limited from a certain extent, and a mass of thermal power generating units are required to input and participate regulation to guarantee area whole control preferences. The generation scheduling closed-loop control method includes the following steps: 1), arranging a master control area module in a interconnection power network system; 2), connecting the master control area module with an automatic generation control module; 3), designedly following the tracks of a model unit, a plan band width model unit and a plan deviation priority model unit by the automatic generation control module. With the technical scheme, the unit is provided with different plan control models according to self needs, multi-scale approaching of a day-ahead plan of the unit is achieved. On the premise that whole control performances of a control area are guaranteed, smooth construction of the day-ahead plan of a generating unit is taken into account, and win-win of a unit plan, unit performances and area indicators is achieved.