167 results about "Search graph" patented technology

An image retrieval system performs both keyword-based and content-based image retrieval. A user interface allows a user to specify queries using a combination of keywords and examples images. Depending on the input query, the image retrieval system finds images with keywords that match the keywords in the query and / or images with similar low-level features, such as color, texture, and shape. The system ranks the images and returns them to the user. The user interface allows the user to identify images that are more relevant to the query, as well as images that are less or not relevant to the query. The user may alternatively elect to refine the search by selecting one example image from the result set and submitting its low-level features in a new query. The image retrieval system monitors the user feedback and uses it to refine any search efforts and to train itself for future search queries. In the described implementation, the image retrieval system seamlessly integrates feature-based relevance feedback and semantic-based relevance feedback.

Techniques for more efficiently identifying specific topological patterns in microdevice design data, such as layout design data. A user provides a topological pattern matching tool with a pattern template. In response, the topological pattern matching tool will analyze the pattern template to create a set of “design rule check” operations that can be performed to identify topological features of the layout design that will include the set of topological features specified for the template. The topological pattern matching tool also specifies properties that should be determined for each set of topological features identified by a design rule check operation. Once the design rule check operations have been created, the tool applies them to the layout design data being analyzed. The results produced by the design rule check operations will be a group of topological features in the layout design that encompass the topological features specified for the template. The results also will include a set of properties for each of the identified topological features. Next, the pattern matching tool creates a search graph based upon the results of the design rule check operations. Once the search graph is constructed, the pattern matching tool traverses the search graph to identify combinations of nodes connected by graph edges representing feature characteristics that match the constraints specified for the pattern template. For each such identified combination of nodes, the tool will output the arrangement of geometric elements corresponding to the nodes as a topological match to the original template.

An efficient person search method utilizing human communications characteristics such as close social network, user-dependent hot list, and high skewness in access popularity. This method makes use of hot lists in the routing process and clusters people of the same community together in the search graph, thereof creating a “small-world” effect to minimize the hop counts routing would have to go through in a Peer-to-Peer environment.

A hypothesis space of a search graph may be determined. The hypothesis space may include n hypothesis-space transcriptions of an utterance, each selected from a search graph that includes t>n transcriptions of the utterance. An evidence space of the search graph may also be determined. The evidence space may include m evidence-space transcriptions of the utterance that are randomly selected from the search graph, where t>m. For each particular hypothesis-space transcription in the hypothesis space, an expected word error rate may be calculated by comparing the particular hypothesis-space transcription to each of the evidence-space transcriptions. Based on the expected word error rates, a lowest expected word error rate may be obtained, and the particular hypothesis-space transcription that is associated with the lowest expected word error rate may be provided.

A system, method and computer program product for efficiently generating on- and off-road travel routes using multi-resolution maps including a map generator (108), map database (106), route generator (100), heuristic generator (102), and search envelope factor generator (104). The map generator (108) converts input raster and vector map data into output multi-resolution raster map data suitable for routing and stores it in the map database (106). The route generator (100) uses the map database (106) to produce a route at a coarse zoom level and then regenerate the route at progressively higher zoom levels. The heuristic generator (102) penalizes high resolution routes that deviate too far from one or more low resolution routes. The search envelope factor generator (104) computes this penalty based on the distance from a given search graph node to the low resolution routes.

The present invention relates to a method and apparatus for adding new vocabulary to interactive translation and dialogue systems. In one embodiment, a method for adding a new word to a vocabulary of an interactive dialogue includes receiving an input signal that includes at least one word not currently in the vocabulary, inserting the word into a dynamic component of a search graph associated with the vocabulary, and compiling the dynamic component independently of a permanent component of the search graph to produce a new sub-grammar, where the permanent component comprises a plurality of words that are permanently part of the search graph.

A method and system for searching a database of graphical data are described. Embodiments of the invention use accelerated image-comparing techniques based on application of the Levenshtein algorithm for matching or searching one-dimensional data strings to recognizing pre-selected targets in graphical contents of 2D images.

The invention discloses a rapid suppression method of interference of linear frequency-modulated radar based on FRFT, wherein the main idea comprises the following steps. A radar echo signal is obtained, and an interference signal is extracted from the radar echo signal,while the delay processed interference signal shown in the description is calculated. FFT of shown in the description is conducted for obtaining the fast fourier transformed signal, and W FRFT order values are obtained after a rough estimate of frequency-modulated slopes of the interference signal and a rough estimate of required orders of the interference signal in FRFT are calculated in order. A two-dimension plane is correspondingly obtained according to the W FRFT order values, a three-dimension searching graph of the interference signal is further obtained, and a precise order p<0> of the three-dimension searching graph of the interference signal is determined, then the interference signal is in suppression process after the radar echo signal with p0-order FRFT is obtained by calculation, a radar echo signal Chi'p<0>(u) with the suppression of the interference signal is obtained. The radar echo signal with inverse FRFT is obtained after an inverse FRFT of Chi'p<0>(u).

The invention discloses a pedestrian re-identification method. The method comprises steps of pedestrian characteristic extraction and pedestrian characteristic tolerance, for pedestrian characteristic extraction, image color histograms are extracted through employing sliding windows, main colors are expanded, statistics of appearance frequency of each color mode is carried out at each sliding window, sum of relatively larger frequencies is taken as characteristic output of the color, traversal of an integral image is carried out through the sliding windows, and characteristics are formed after normalization; for more generalization matching, a search source is extracted from characteristics of a full body to a half body; for tolerance calculation, characteristic dimensions with a non-zero search image characteristic vector and to-be-searched-target characteristics are taken as an Euclidean distance, and interference of background factors under a complex scene is further reduced. The method is advantaged in that re-identification under the complex scene is carried out, properties of strong transportability for scene change, stable algorithm and fast speed are realized, a problem of poor quality of monitoring videos and to-be-searched databases can be effectively solved, and strong practicality is realized.

The invention relates to artificial intelligence, and provides a PowerPoint generation method, which comprises the steps of receiving a main body keyword of a PowerPoint input by a user through a client; performing text material search in a text material library by utilizing the main body keyword; splicing and integrating the text materials; carrying out manuscript style analysis processing by utilizing keywords and the sub-topics; determining integral style information of the presentation file; inputting the sub-topics and the topic paragraphs corresponding to the sub-topics into a keyword extraction model for related word extraction; and inputting the paragraph keywords into a picture library for searching, and generating a presentation file corresponding to the keywords. According to the method, material search, picture material search, style recommendation and format typesetting can be performed through simple main body keywords, so that a large amount of information search and integration work time in the early stage is saved, and a corresponding presentation file can be quickly and automatically generated after the main body keywords are input by a client; the problem of lowpresentation file generation efficiency in the prior art is solved.

The invention provides a single-target tracking method based on multiple networks, and the method is a model adopting deep learning. The method comprises the following steps: cutting a first frame image and a current frame image of a video sequence to obtain a template image and a to-be-searched image; inputting the template image and the to-be-searched image into an appearance subnet and a semantic subnet, respectively obtaining low-level appearance features and high-level semantic features of the template image and the to-be-searched image, and performing feature fusion to respectively obtain fusion feature maps of the template image and the to-be-searched image; then, based on the fusion feature map of the template image and the image to be searched, using a similarity discrimination method to obtain a final response map; and finally, obtaining a tracking result according to the information provided by the final response diagram. According to the method, the problems that a traditional single-target tracking method cannot effectively detect a tracking target in a to-be-searched image containing a similarity background and false detection is caused by noise existing in extractedlow-layer appearance features in a feature extraction method based on deep learning are solved.

The invention discloses a two-dimensional code recognition method and system based on connected domain analysis, a device and a medium, and the method comprises the steps: detecting the edge of a two-dimensional code in a to-be-recognized image after receiving the to-be-recognized image, and obtaining the to-be-recognized two-dimensional code; determining an image searching graph in the to-be-identified two-dimensional code by using a connected domain analysis algorithm; determining a coordinate point corresponding to the pixel value mean value in the image searching graph as a positioning point; determining a correction point; and performing rotation correction on the to-be-identified two-dimensional code based on the positioning point and the correction point, analyzing the to-be-identified two-dimensional code after rotation correction, and outputting corresponding information. After the two-dimensional code to be identified is acquired, the image searching graph is firstly determined, and the coordinate point corresponding to the pixel value mean value in the image searching graph is determined as the positioning point, so that the problem of point loss possibly caused by searching the positioning point by point when the shooting quality of the two-dimensional code is poor is avoided. The connected domain analysis algorithm is used to confirm the image searching graph, so that the positioning point determined from the image searching graph is more accurate, and the identification precision of the subsequent two-dimensional code is improved.

In order to provided algorithm for searching graphs and a device for searching graph into which a new concept for solving a disadvantage arisen by using proof numbers or disproof numbers in an AND / OR graph construction is taken, algorithm according to the present invention which searches directed graphs in which OR nodes having at least one choice or more choices desired to reach to solution and AND nodes necessary to reach all choices to the solution are appeared alternately from a starting point to the solution, further comprising: a calculating step for calculating branch numbers of routes except for a specific route reaching at least from the starting point to an optional node at a specific node in the AND nodes or the OR nodes accompanying with said specific route.

The invention discloses a ship search-and-rescue method based on ant colony algorithm. The ship search-and-rescue method includes the steps of firstly, establishing an ant colony search graph; secondly, allowing ants to creep from a virtual node O; thirdly, transferring each ant from a node r to a node s according to state transition rules Prskp; fourthly, judging whether optimal solutions searched after all ants of generations creep are trapped in local optimum or not; fifthly, applying algorithm update information pheromone strategy (1) to the ship search and rescue problem when the algorithm is trapped in local optimum; and sixthly, when the logarithm is trapped in local optimum, updating only pheromone density in a local optimal solution path by an update information pheromone strategy (2). Through the ship search-and-rescue method based on ant colony algorithm, such problems that reaching to search and rescue sites accurately is difficult for the existing ships under the influence of factors such as failure, path proficiency of operators and marine environment parameters are solved.

The invention discloses a method of fully automatically classifying and partitioning a branch retinal artery obstruction based on a three-dimensional OCT image. The method comprises the following steps: pretreatment is carried out, a graph searching algorithm is adopted to layer the retina, and each layer of the retina is leveled according to a pigment epithelium layer; an AdaBoost classifier is used for automatically classifying an acute stage and an atrophy stage of the branch retinal artery obstruction; partitioning of the acute stage of the branch retinal artery obstruction is carried out, a Bayesian posterior probability is firstly adopted to carry out initial partitioning on an obstruction area, and then based on a graph searching-graph partitioning algorithm, accurate partitioning is carried out on the obstruction area; and partitioning of the atrophy stage of the branch retinal artery obstruction is carried out, and the obstruction area for the atrophy stage is automatically partitioned through building an inner retina thickness model. The method of the invention can accurately classify and partition the branch retinal artery obstruction area, and can replace manual classifying and partitioning.

The invention discloses a learning target tracking algorithm based on a twin network. The method comprises the following steps: 1, setting a function f to compare the feature map similarity of a template image x and a candidate image y, the maximum response value corresponding to a target position, the x being generally an image with a target as the center in a first frame, and the y being an image searched by taking an (n-1) th frame of image target as the center in an nth frame; substituting the two inputs into a convolutional neural network, learning a network parameter rho by using a convolution processing function, generating two cross-correlation mappings, and performing exhaustive search on x at y to respond to the position of a target corresponding to the maximum value; wherein animproved discriminant correlation filter layer is added between input x and cross-correlation operation in the twin network, 3, performing tracking by measuring the similarity between a template imagex and a search image y, and online fine adjustment is carried out by means of a discriminant correlation filter module during online tracking so as to ensure the tracking accuracy.

The invention provides a method and a device for searching graph data. The method comprises the steps of determining keywords to be searched at present, wherein the keywords include a first identifier capable of uniquely identifying the first top point in a graph, and a second identifier capable of uniquely identifying the side associated with the first top point in the graph; searching the graph data through a first access index according to the keywords to obtain a third identifier which can uniquely identify a second top point associated with the first top point in the graph. According to the method, the operation from some top point of one graph to another associated top point by side traversing involves only one index access, so that the graph data search efficiency can be increased.

The invention relates to an underwater robot cooperative target searching method based on a global information transmission mechanism. The method comprises the steps: preliminarily calculating an activity value of each grid in a to-be-searched task region according to prior probability distribution and obstacle distribution of a target, and enabling the activity values to serve as the prior searchgraph information of the region; in combination with ocean current field distribution in the area, calculating robot navigation time between every two adjacent grids, and determining the connection weight between every two adjacent grids; mutually transmitting active values between the adjacent grids at a certain weight; extracting a high-value sub-region by adopting a Gaussian mixture model, andtransmitting the expected income of the high-value sub-region to each grid to realize global sharing and updating of region search graph information; and enabling each robot to independently maintainand iteratively update the respective region search graph and determine a next behavior until the target search task is completed. The method is simple and feasible, the path is smooth, the efficiency is high, and multi-robot cooperative target search is realized.

The invention provides a target tracking model training method and device. The training method of a target tracking model comprises the follwoiing steps of training a target tracking model based on asample image pair, wherein the sample image pair comprises a template image and a search image; wherein the search image is marked with a classification label and a weight label corresponding to a pixel point of the search image; the method comprises the following steps: carrying out iterative training by utilizing the following steps until the model training requirements of a target tracking model are met: inputting a sample image pair into the target tracking model, and predicting based on the target tracking model to obtain the similarity between a search image and a template image; determining similarity loss of the iteration according to the similarity, the classification label and the weight label; and updating model parameters of the target tracking model according to the similarityloss. The accuracy of target tracking can be improved.

The embodiment of the invention discloses an image material library generation method, an image material recommendation method and related devices, and the method comprises the steps: obtaining a teaching file which comprises a plurality of chapters; for each chapter, determining a chapter keyword from the text content of the chapter; obtaining a plurality of image materials of the chapters according to the chapter keywords; and taking the chapter keywords as labels of image materials to store the image materials so as to generate an image material library of courseware files for making the teaching files. A user does not need to search image materials on the Internet; image materials can be searched in an image material library according to keywords of a target chapter to which a currently-made courseware file belongs and recommended to a user; on one hand, the image material searching time of the user is saved, and on the other hand, the image materials can be recommended according to the chapters to which the courseware files belong, so that the image material screening workload of the user is reduced, and the courseware making efficiency of the user is greatly improved.

A computer-implemented system and method identify Pareto optimal candidate paths between an origin and a destination for which no other candidate path is strictly better on one of a predefined set of criteria and at least as good on all the others. A constraint is defined for each of the criteria, based on user input. A set of Pareto optimal candidate paths is identified, from an origin to a destination, which respect these constraints. The identification may include, in a search graph composed of nodes connected by edges, iteratively advancing each of a set of possible paths from an origin node by exactly one exit node and updating labels of the exit nodes reached. The exit node labels each include a value for each of the criteria. Labels of reached exit nodes that are dominated by another label of that reached node are removed. Pareto optimal candidate path(s) is / are identified.

A system, method and computer program product for efficiently generating on- and off-road travel routes using multi-resolution maps including a map generator (108), map database (106), route generator (100), heuristic generator (102), and search envelope factor generator (104). The map generator (108) converts input raster and vector map data into output multi-resolution raster map data suitable for routing and stores it in the map database (106). The route generator (100) uses the map database (106) to produce a route at a coarse zoom level and then regenerate the route at progressively higher zoom levels. The heuristic generator (102) penalizes high resolution routes that deviate too far from one or more low resolution routes. The search envelope factor generator (104) computes this penalty based on the distance from a given search graph node to the low resolution routes.

The embodiment of the invention provides a picture searching method, device and apparatus, and a computer program product, and relates to the technical field of artificial intelligence. The method comprises the steps of obtaining an OCR recognition result of each picture in a preset picture library in response to a picture search request; traversing the pictures which do not complete the low-dimensional OCR identification processing and do not complete the high-dimensional OCR identification processing, and performing low-dimensional OCR identification processing based on an OCR identificationthreshold on each traversed picture to obtain a low-dimensional OCR identification result of each corresponding picture; determining a target picture matched with the keyword in a preset picture library according to the low-dimensional OCR recognition result or the high-dimensional OCR recognition result of each picture; and determining the target picture as a search result of the picture searchrequest, and displaying the search result. Through the embodiment of the invention, the text information in the pictures can be searched more accurately, refined search is realized, an accurate searchresult is obtained, and the search efficiency can be improved.

The invention discloses a pavement crack rapid identification method based on deep learning, and the method comprises the steps: firstly adjusting the size of a field picture, and then adjusting the exposure of the field picture, wherein original on-site photos are uneven in illumination, and the shielding effect of on-site images can be achieved only by processing uneven exposure; fitting the distribution of grayscale pixel values by using a histogram of pixel intensity values based on grayscale distribution; achieving binary color visualization by using a threshold method based on the mean value of the previous step; if the pixel value is greater than the threshold value, setting the pixel value as a background; enhancing the fracture shape by using a connecting member-based method; using a connection tool to perform denoising; searching all connected objects in the graph, and checking the area of the crack shape; if the area of the fracture shape is less than a threshold, considerting that the object is noise; if the area of the crack shape is greater than the threshold, considerting that the object is a crack; finally, adjusting CNN input, and reconnecting cracks through expansion and erosion; and adjusting CNN hyper-parameters, and determining an optimal CNN frame.

The invention discloses a twin neural network moving target tracking method based on a full-connection attention module, and belongs to the technical field of computer vision tracking. According to the method, after the image features are extracted by using a twin neural network, template features extracted by the template branches are processed by using a full-connection attention module, and the template features are fused with the original template features and then serve as the attention-enhanced template features to be combined with updated template features for performing the same operation; the obtained new template features are fused with search features, so that self-attention and mutual attention of the template features are realized, and the robustness is improved; position information and size offset information of a target in the corresponding search graph are obtained according to a response graph fusing the reinforced template features and the search features; and the input of the update template branch is updated according to a network prediction result of each fixed frame, so that the tracking precision is improved. According to the method, target tracking can still be continuously and stably realized under the conditions of severe deformation, reproduction after transient disappearance or shielding and the like of the target.

The invention provides optimization processing for a search target task of multiple unmanned aerial vehicles, wherein the optimization processing comprises the steps: enabling the prior probability distribution of a search target of single unmanned aerial vehicles to serve as a prior search graph of a search region, carrying out iterative updating of the prior search graph, and obtaining updated search graphs of the single unmanned aerial vehicles; calculating the communication probability between the unmanned aerial vehicles, determining the communicability between any two unmanned aerial vehicles based on the communication probability, and determining an unmanned aerial vehicle communication network; fusing the search graphs among the unmanned aerial vehicles based on the communication probability among the unmanned aerial vehicles to obtain a fused search graph; updating the fused search graph in combination with the target motion to obtain a fused updated search graph; and enablingthe single unmanned aerial vehicles to update the search graph based on fusion, optimizing the positions of the unmanned aerial vehicles and guiding the unmanned aerial vehicles to fly. The method comprehensively considers and optimizes the communication performance and the search performance in the search task of the multiple unmanned aerial vehicles, and is more suitable for the search of a moving target, especially the target search task in the actual complex marine environment.

The invention provides a method for recognizing different artistic expressions of an object based on a specific data set. The method mainly comprises an ArtData data set, an annotation transfer way and artistic image retrieval, and comprises the steps of firstly, measuring the degree about which knowledge of the object in a medium can be applied to other medium, then comparing the emotion and different features and using styles of medium classes to consider the style and aesthetics so as to improve the aesthetic prediction task on other artistic data set, and finally, constructing a search engine specific to the task in combination with an experiment of learning feature space, wherein the search engine can be used for searching an image according to the content, emotion or medium similarity. The method breaks through the limitation of image retrieval only on daily photography, and improves the generality of image retrieval; and besides daily photography scenes, different artistic expressions can also be recognized and extracted, so that the cognition degrees of computers on arts field and aesthetics are improved, and the view field of machine vision is expanded to other fields.

The invention discloses a webpage search intelligent sorting method and system based on mobile internet data deep mining and a computer storage medium. Pictures needing to be searched by a user are imported into a webpage search engine, meanwhile, a search mode is selected by the user; if the user selects the picture search mode, the comprehensive matching degree conformance coefficients of the search pictures imported by the user are analyzed and calculated, a comprehensive matching degree conformance coefficient difference value of each picture in the webpage search engine is obtained through comparison, and the pictures are sequentially sorted according to the comprehensive matching degree conformance coefficient difference value from small to large; if the user selects a text search mode, the comprehensive matching degree coincidence coefficient of each text in the webpage search engine is analyzed and calculated, and the pictures are sorted according to the sequence of the comprehensive matching degree coincidence coefficients from large to small, so that the function of selecting multiple search modes is realized, the actual search requirement of the user is met, and the intelligent sorting level of the webpage search engine is improved.

One or more embodiments of the invention provide a semantic search method and device based on a case event knowledge graph, and electronic equipment, and the method comprises the steps: firstly carrying out the preprocessing of a law case event question inputted by a user, and then carrying out the information extraction and information fusion of the description of the law case event question; classifying the description based on case event type tags, and outputting a semantic search graph with the case event type tags; performing sub-graph matching on the semantic search graph and a law case event knowledge graph, and outputting law case event answer data. According to the method, the semantic network is established for the user questions, and matching and reasoning are carried out in combination with the case event knowledge graph, so the search intention of the user can be accurately captured, answers meeting the search intention of the user are directly given, and legal questions of the user are practically solved.

The invention discloses an RGB-D feature target tracking method based on a twin network. The method comprises the following steps: constructing a twin network model based on RGB-D features; precessing the template image by a shared network to obtain semantic features of the template image, and inputting the high-level semantic features to a deep convolutional network module to obtain a depth map; performing depth feature extraction on the depth map to obtain depth feature information, and fusing the depth feature information with the semantic features in a cascade mode to obtain fused image features; processing the search image through a shared network to obtain features of the search image, wherein the features of the search image are subjected to convolution and pooling operation to obtain context information of the search image, the fused image features are guided through the context information of the search image, and adaptive features used for target positioning are generated; and carrying out cross-correlation operation on the adaptive features and features obtained by processing the search images through a shared network, and carrying out interpolation calculation on the score graph to obtain a tracking result. The depth map is introduced, high-precision tracking in a complex scene can be achieved, and the effect is good.