80 results about "Edge model" patented technology

A system for diagnosing impairments in computer networking services a plurality of connector gateways for interfacing to a plurality of data sources to provide access by the system to computer network data, business-related data, and extra-enterprise environmental data in the plurality of data sources. The system includes a knowledge manager in communication with each of the plurality of connector gateways for analyzing data from the plurality of data sources to generate diagnostic conclusions for resolving the impairments in the computer networking services. The system also includes an interactor in communication with the knowledge manager for generating format information for displaying at least the diagnostic conclusions to a user. The system can include a graphical user interface for displaying at least the diagnostic conclusions associated with edges of an edge model of the computer networking services of the user.

The invention relates to the technical field of cloud computing and edge computing, and aims to provide an industrial equipment predictive maintenance method based on cloud-edge cooperation. The invention discloses an industrial equipment predictive maintenance method based on cloud-side cooperation, and the method comprises the following steps: S1, heterogeneous sensing equipment collects the equipment state data of industrial equipment, and transmitsg the equipment state data to an edge computing platform; s2, an edge data management module of an edge computing platform obtains feature datarequired by a prediction task of the target equipment according to data uploaded by the heterogeneous sensing equipment; a configuration reloading module of a prediction service orchestrator obtains equipment state prediction model configuration of the cloud computing platform and equipment state prediction model configuration trained by an edge model training module, a model operation module loads a latest equipment state prediction model of target equipment, and extracted feature data is input; whether the target equipment has a fault risk or not is judged according to the output data of themodel operation module, if so, the next step is executed, and if not, the step S1 is returned; and S3, a trigger management module of the edge computing platform notifies a designated person in charge of fault early warning information according to a preset trigger. According to the invention, accurate and efficient industrial equipment predictive maintenance can be realized.

A sub-pixel edge of an ellipse target positioning method mainly relates to a image processing and machine vision such as calculation of accurate parameters and calibration and matching of a pick-up camera of the ellipse target and the like, the method is mainly divided into three main steps: the first step includes noise elimination of images, edge detection by Sobel operators and extraction of edge points of the ellipse target; the second step includes calculation of geometric parameters of the ellipse target; and the third step includes position of the sub-pixel edge, wherein, the position of the sub-pixel edge can be divided into four parts of calculating target grayness and background grayness of the edge model, calculating edge angles, calculating the distance between edge points and real edge points and calculating accurate position of sub-pixel edge points. The sub-pixel edge of positioning method comprehensively utilizes geometric parameters of the ellipse target, distribution characteristics of grayness of the ellipse target and two-dimensional edge models. The method not only effectively improves the accuracy and the robustness of the edge positioning, but also greatly reduces arithmetic quantity so as to enhance the rapidity.

The present invention relates to the technical field of image recognition. A multi-view pointer instrument identification method includes: acquiring an image and uploading the image to a computer; using a SSD algorithm to position an instrument area; using a ResNet34 deep residual neural network to perform classification training on the instrument area, and performing preliminary correction on thesample image according to a classification result; using the SSD algorithm to secondarily position the instrument area of the corrected image; using the network to perform regression training on thesecondarily positioned instrument area to identify the position of a pointer on an instrument dial; using an HED edge detection algorithm to perform dial edge detection on the positioned instrument area; performing random sampling according to an RANSAC algorithm, and calculating the edge model of the instrument dial; correcting the instrument point by a zoom ratio, and calculating the angle between the instrument point and a start point; and looking up a database to obtain the scales of the instrument dial. The method can recognize the pointer instrument captured at different angles.

The invention discloses a captive test (CT) image partitioning method based on adaptive learning. The CT image partitioning method comprises the following steps of: 1) acquiring a CT image; 2) extracting characteristics of the CT image; 3) inputting strokes for representing a lesion region and a non-lesion region on the CT image by a user; 4) constructing a region model of the image by taking the strokes input by the user as a basis according to the extracted characteristics of the CT image, and constructing an edge model of the image by adopting an edge detection method; and 5) combining the region model and the edge model to construct a new model, calculating the new model to obtain a partitioning result. By adopting the method, a difference between the lesion region and the non-lesion region on the CT image can be effectively described; the method adapts to the complexity of the CT image; the problems caused by low signal-to-noise ratio (high noise) of the CT image are solved; the user can quickly and precisely partition the lesion region on the CT image in an interactive manner with high efficiency; and therefore, the production efficiency of a medical department can be greatly improved.

The invention provides a vector image generation method and system based on bitmap image adaptive segmentation. According to the method, filtering operation is carried out on a bitmap image through adoption of a bilateral filtering algorithm. Region segmentation processing is carried out on the bitmap image; pixels with similar values are merged into the same class according to a value of each pixel of the image; through repeated iteration, region segmented images of the image are obtained; and regions with excessively small areas are merged into adjacent regions. Region marked images of the segmented images are obtained; region edges are extracted through adoption of an optimized edge model and an extraction algorithm; edge line segments are fitted; and an edge curve of each region is determined. A color average value of a source image corresponding to each class of regions in the region marked images within the region is calculated; the average value is taken as a color of the current region; and a color list corresponding to the region is formed. A corresponding vector image file is generated according to the edge curve data and color list information of each region.

A method for object localization comprises defining an edge model for object localization, and searching an image for an object matching the edge model. Searching comprises defining a region of interest, including a portion of the object, sub-sampling the region of interest, and extracting an edge image comprising edges from the region of interest. Searching further comprises thinning at least one edge of the edge image, determining a distance map between the image and the edge image according to a distance transformation, and matching the edge model to the edge image within a search area defined according to the distance map.

The invention aims to provide a method for carrying out three-dimensional tracking initialization. The method specifically comprises the steps of obtaining the 3D model information about a target object and an initial pose of a corresponding camera device relative to the target object; determining 3D surface edge model information corresponding to the initial pose based on the initial pose and the3D model information; and based on the 3D surface edge model information and the initial pose, carrying out tracking matching on the target object in a video related to the target object, and obtaining a corresponding tracking matching result, the tracking matching result comprising an accurate pose of the camera device relative to the target object. According to the method, the SLAM rapid initialization in a complex scene can be realized, a very good tracking effect can be obtained under various environmental interferences, and the scale of a physical world is provided for an SLAM system.

The invention discloses a space non-cooperative target pose measurement method based on a model. The space non-cooperative target pose measurement method is characterized by comprising the following main steps: obtaining a deep edge and a texture edge of a target model according to a relative pose at the last moment so as to obtain a visible edge model; extracting a Harris characteristic of a current scene image according to a KLT (Karhunen-Loeve Transform) algorithm; projecting the target visible edge model in an image scene, searching an image edge in a normal direction along the edge model and calculating a residual error of a corresponding characteristic point pair; estimating 3D coordinates of the Harris characteristic of a last frame through deep buffer, projecting the 3D coordinates in a current image in a current estimation pose, and calculating a residual error of a corresponding characteristic point pair; determining weights of characteristic point pairs according to an M estimator robust method and distributing the weights of characteristic points according to the average residual error obtained in the two steps; and iteratively calculating a relative pose according to a visual servo method. According to the method provided by the invention, the measurement of a relative pose of a space non-cooperative target can be realized, and the advantage of low calculating amount is achieved.

A compiler that forms an intermediate representation of a program using a flow graph with less than all possible edges used to model asynchronous transfers within the program. The flow graph is formed in multiple phases. In one phase, the flow graph is formed without modeling asynchronous transfers. In later phases, representations of the effects of the asynchronous transfers are selectively added. As part of the later phases, edges modeling a possible asynchronous transfer are added to the flow graph following definitions in protected regions of variables that are live outside the protected region. A modified definition of live-ness of a variable is used to incorporate use of the variable in any region, including the protected region, following an asynchronous transfer. Edges from the protected region are also added to the model if the only use of the defined variable is in a handler.

The invention discloses a DDNN training method and a DDNN-based multi-view target recognition method and system, and belongs to the field of cloud computing. The method comprises the steps of obtaining an information entropy of a distributed deep neural network cloud side model to a sample image; constructing a DDNN target function based on the information entropy of the sample image; and according to the DDNN target function, jointly training an edge side model and a cloud side model of the DDNN. In the knowledge migration method focused on a teacher-student network, under the background of multiple outlets of a DDNN level, the adaptive training method based on sample weighting is provided, scores of samples are obtained from a deep outlet of a DDNN, the samples are weighted through the scores to distinguish simple and complex samples, the weighted samples are used for training a cloud side model and an edge side model at the same time, and the communication traffic is minimum while good classification precision is guaranteed. The cloud side model guides the whole training process of the edge model, and the edge model can learn real labels and cloud side migration knowledge at thesame time.

The invention relates to an accurate overprinting method for printing envelope postal codes, which comprises the steps of: capturing an envelope image in real time by a camera; extracting gray level information and red domain information; approximating the gray level information to be a segment; seeking an envelope edge; establishing and updating an envelope edge model according to the continuity of continuously collected images; generating a code red box according to the red domain information; computing whether an average distance delta y between the successfully matched red box and the envelope edge model is within an allowable range of an envelope standard or not; if delta y is within the allowable range of the envelope standard, judging again whether a sequence number of the red box is 1 or not; if the sequence number of the red box is 1, emitting a command of printing a first postal code character in a current position to an envelope printer; if the sequence number of the red box is not 1, computing a distance delta n between the current red box Rn and the previous red box Rn-1; judging that delta n conforms to the envelope standard; and controlling the envelope printer to print corresponding postal code characters in the current position. In the accurate overprinting method, the interference of external factors is effectively eliminated with high control accuracy.

A compiler that forms an intermediate representation of a program using a flow graph with less than all possible edges used to model asynchronous transfers within the program. The flow graph is formed in multiple phases. In one phase, the flow graph is formed without modeling asynchronous transfers. In later phases, representations of the effects of the asynchronous transfers are selectively added. As part of the later phases, edges modeling a possible asynchronous transfer are added to the flow graph following definitions in protected regions of variables that are live outside the protected region. A modified definition of live-ness of a variable is used to incorporate use of the variable in any region, including the protected region, following an asynchronous transfer. Edges from the protected region are also added to the model if the only use of the defined variable is in a handler.

The invention discloses a prospect frame detecting method based on an edge model, which is used for prospect frame determination in the analysis of a security monitoring video frame sequence. The method comprises the following steps of: extracting edge frames only containing edge images from sequence frames by means of a pseudo-sphere edge detection operator; counting the presence probability of edge points of the edge frames in frame sequence statistic time, marking a background attribute and a foreground attribute based on a determination condition of distinguishing according to the attributes of the edge pixel points within the current edge frame; in a foreground edge image, if the number of the edge points connected together in the current frame is less than or equal to 2, determiningthat the points are noise points and removing these points; and if the total number of pixels of the left foreground edge images is less than a noise determination threshold of the sequence frame, determining that the frames are background frames, otherwise, foreground frames. The method of the invention reduces the calculation cost, effectively adapts to the cases of background light change, slow object motion or short stagnation and the like, and creates favourable conditions for subsequent object motion analysis.

The invention provides a power communication network operation mode modeling method and system based on a graph database. The power communication network operation mode modeling method comprises the following steps of acquiring a power communication network data source, extracting equipment information of all power communication network equipment from the power communication network data source; establishing a node model; establishing an edge model; configuring attribute information for the node model and the edge model according to the equipment information; according to the actual operationmode and the routing information of the power communication network, acquiring operation data of the equipment node participating in operation and the operation data of the transmission edge; according to the node model, an edge model, the attribute information and the operation data, constructing a power communication network operation mode diagram model. According to the method, modeling of theoperation mode of the power communication network is achieved, the modeling method is simple and efficient, so that a data foundation is provided for carrying out topology analysis on the operation mode of the power communication network in a follow-up mode, and the performance of topology analysis is improved.

The invention discloses a semantic edge dominated high-resolution remote sensing image segmentation method. The method comprises the steps of data acquisition, network design, model preparation, edgeproduction, morphological post-processing and vectorization. Multi-scale fusion is formed by utilizing the feature learning capability of a convolutional neural network and pooling operation, and ground object edges conforming to visual features are extracted from a remote sensing image. A pre-training edge model or a retraining model based on a local manual sketching result can be adopted to correspond to an unsupervised method and a supervised method in a segmentation task respectively. Relatively accurate and complete ground object boundaries are formed through edge refinement and extensionconnection, and the boundaries are vectorized to form ground object polygons required by segmentation. The ground object polygon generated by the method can be basically matched with the visual boundary of the ground object on the image, and therefore most of over-segmentation and under-segmentation phenomena in a traditional segmentation object are overcome, and effective support is provided forground object fine form determination, type identification and large-scale information extraction.

The invention discloses a monitor image anomaly detection method and device, and belongs to the field of video monitoring. The method comprises the steps: foreground pixel points in a target monitor image are determined based on a mixture Gaussian background model; the amount of the foreground pixel points in the target monitor image can be counted to obtain first pixel amount; second pixel amount can be determined based on an edge detection algorithm, and the second pixel amount is the amount of pixel points of which edge detection values in the target monitor image are greater than a first preset threshold value and positions in the target monitor image do not belong to a plurality of specific positions contained by an edge model; weighting and combining are performed on the first pixel amount and the second pixel amount, and a weighted statistical value is obtained; and when the value is greater than a second preset threshold value, the target monitor image is determined as an abnormal image. According to the invention, a way of combing foreground information and edge information is adopted to perform anomaly detection on a monitor image, and the accuracy and stability of the anomaly detection can be improved.

The invention discloses a social network model construction module of a company image improvement system. The social network model construction module comprises the following five sub-modules: construction of a complex social network user model, construction of an inter-user relationship module, construction of a multi-source heterogeneous complex social network topological graph, identification of key nodes, discovering and dividing of communities, wherein the construction of the complex social network user model comprises user data extraction and user attribute feature definition; the construction of the inter-user relationship module comprises user relationship extraction and potential relationship prediction; and the identification of the key nodes comprises user node importance indexes and event propagation node importance indexes. According to the invention, related data on social media is collected efficiently; the complex social network user model is constructed on the basis ofacquired data; meanwhile, a specific relationship among users is modeled; a one-way edge model among the users is constructed; a complex social network topological structure model is comprehensivelyobtained; and the complex social network topological structure model is taken as an object.

The invention provides an augmented reality unmarked tracking registration method and device based on an edge model, and relates to real-time performance and robustness of tracking registration in theaugmented reality unmarked field. At present, the defects that the surface of a three-dimensional object lacks enough textural features, pose jitter or disturbance is caused by few feature points, and the algorithm search space is too large are overcome. Aiming at these problems, according to the method, a linear parallel multi-mode LINE-MOD template matching method keeping rotation and scale unchanged is adopted to quickly identify a target object, and a reference view close to a current view angle is obtained to complete estimation of the pose of a camera. And the pose is used as an initialization pose based on an edge model tracking registration method, so that the initialization pose is prevented from being given manually. Therefore, the method can achieve the quick tracking and registration, and is good in real-time performance and robustness.

The invention belongs to the machine vision and related image processing techniques, relating to the improvement of an image edge extracting method. The steps of the invention: making smooth filtering processing on the original image; searching a boundary point Ps with steady-state grey scale maximum value and a boundary point Pe with steady-stage grey scale minimum value of a step edge; calculating initial values a0ú¼b0 and c0 of edge model parameters; calculating the optimum values of aú¼b and c and determining the step edge. The initial value selection of the model parameters is simple, and the step edge extraction has high speed, strong robustness and high accuracy.

The invention discloses a federated learning poisoning detection method based on neuron distribution characteristics, which comprises the following steps: (1) obtaining a plurality of edge models trained and uploaded at clients, and according to the similarity of adjacent edge models uploaded for several times corresponding to each client, screening a plurality of edge models meeting screening requirements from the plurality of edge models uploaded each time to serve as candidate poisoning models; (2) screening at least one model from the candidate poisoning models as a poisoning model according to the distribution state of the model parameters, and removing the poisoning model, (3) inverting the poisoning model according to the sample data and the label to obtain poisoning patch data, optimizing the aggregation model parameters of the server according to the poisoning patch data, and obtaining an optimized aggregation model; and issuing the optimized aggregation model to the client bythe server to serve as an edge model of the client for edge training of the next round. The federated learning poisoning detection method can rapidly detect the poisoning model.

a pattern evaluation system which receives image data of a pattern to be evaluated to evaluate the pattern includes an edge model producing part which produces a pattern edge model and an edge point coordinate detecting part which carries out an image matching processing to an image of the pattern with the pattern edge model to detect coordinates of an edge point of the pattern.