195 results about "Similarity relation" patented technology

A method and a system for generating sequencing information representing a sequence of items selected in a database. Similarity relation techniques are applied between the items.

The invention discloses an algorithm-based text summary automatic extraction method and relates to the technical field of text extraction. The method comprises the steps of S1, preprocessing text; S2, extracting the features of the text; S3, using an existing similarity calculation method to calculate the similarity of sentences, and performing weighting during the calculation; S4, using the sentences in the text as the nodes, the similarity relations of the sentences as the sides and similarity as the weight of the sides to construct an undirected weighted TextRank network graph; performing iterative computation until convergence is achieved to obtain the nodes containing the weight; S5, selecting core sentences according to the weight of the sentences corresponding to the nodes, the structure of the text and the position information of the sentences, sorting the core sentences, and outputting the sorted sentences as the extraction results. The invention also discloses a summary extraction system. By the algorithm-based text summary automatic extraction method and system, the accuracy of text summary automatic extraction can be increased favorably.

The invention particularly relates to a multi-semantic deep supervision cross-mode hash retrieval method. According to the multi-semantic deep supervision cross-mode hash retrieval method, the methodcomprises the following steps: utilizing a deep neural network, combining supervision semantic information of training data, and learning hash mapping models of multiple modes respectively; sending image query data of a given specific mode into a neural network of a corresponding mode, converting the image mode data into a hash code through hash mapping learned by a deep network, then calculatingthe distance between the hash code and a hash code of another mode in a database, and finally returning data most similar to query. According to the multi-semantic deep supervision cross-mode hash retrieval method, various high-level semantic information is fully utilized, the similarity relation between data modes and label semantic information in the modes are kept, high-quality hash codes can be obtained, hash learning and classification tasks are combined in the same stream for learning, the learning efficiency is improved, and then information retrieval between data of different modes isachieved.

The invention provides a large mechanical equipment structure dynamic similarity test method, comprising the steps as follows: (1) a basic similarity theorem is adopted, all physical quantities are listed, geometrical quantity, density and elastic modulus are determined as basic physical quantities, and the similarity theorem is written into a dimensionless equation expressed by the basic physical quantity; (2) according to concerned basic physical principle, the similarity ratio of all physical quantities adopts similar relation expression, and the similar relation expression is modified according to a structure elastic vibration similarity equation; (3) similar model material is selected, density similarity ratio and elastic modulus similarity ratio are determined, geometrical quantity similarity ratio is determined, and the similar model design is then completed; (4) a structure dynamic test of the similar model is carried out so as to obtain the dynamic characteristic number which is substituted into the similarity relation expression so as to work out the dynamic characteristic data of the original model which is also the natural vibration frequency. The large mechanical equipment structure dynamic similarity test method realizes the conversion of ultra-large structure dynamic natural frequency with small model, effectively reduces the scales of dynamic test objects, simplifies the test instruments, and reduces the test cost.

The invention belongs to the technical field of image processing and geographical information system, relating to an estimation method of latitude and longitude based on sun shadow track in a video; the method comprises the following steps: a natural image sequence or a video frame is firstly obtained, and a track point of the shadow for each image frame is detected; a plurality of vanishing points are then determined, and the horizon line is fitted; mutually vertical vanishing points are fitted, and affine rectification and projection rectification matrixes are calculated; measured and rectified world coordinates are restored; the track of the shadow point in the world coordinates is then fitted, sundial design is referred, and the latitude is estimated by using similarity relation; the time difference is calculated by using the extreme point of a secondary curve so as to efficiently restore the latitude and longitude information of the photographed images. The method has lower complexity and higher precision, which is a method for realizing latitude and longitude estimation by using the natural image sequence or video frame and can obtain the latitude and longitude information of the photographed images according to the position of uncalibrated shadow.

The invention relates to a multi-class small target detection method based on metric learning, and designs a novel deep neural network structure by combining the feature expression capability of deeplearning with the similarity discrimination capability of metric learning according to the recognition characteristics of multi-class small targets. The method is characterized in that a Faster RCNN (Recurrent Convolutional Neural Network) network structure combined with a Feature Pyramid Network (FPN) is adopted to detect multiple types of small targets on the basis of data of a whole image; a graph network module is embedded into the network to carry out transmission calculation on similarity information among all regions in the image; a similarity measurement module based on triple loss isadopted at the rear end of the network to distinguish detail information among samples, feature information of small targets and similarity relations among the targets are fully extracted, and the accuracy of multi-class small target detection is improved.

The invention discloses an image automatic annotation method based on digraph unequal probability random search, which comprises the following steps: inputting an image to be annotated and an annotated image set; extracting a plurality of feature vectors of the image to be annotated; selecting an adjacent image set; constructing a digraph model of the image to be annotated; calculating a word similarity matrix Se between tags and a symbiotic relationship matrix Co between tags; fusing the word similarity matrix Se between tags and the symbiotic relationship matrix Co between tags, so as to obtain a tag similarity matrix TT; and carrying out unequal probability random search on each candidate tag in a candidate tag set in the digraph model, so as to calculate the score, and obtaining a plurality of high-score candidate tags to be used as the label results. The invention also discloses an image automatic annotation device based on digraph unequal probability random search. In the invention, the dependency relation between images and similarity relation between tags are utilized fully and reasonably, thus the image automatic annotation can be effectively carried out, and the annotation effect is better.

The invention discloses a method of trajectory clustering based on directional trimmed mean distance (DTMD). The method comprises the following steps of: (1) trajectory extraction: extracting the trajectory from an original dynamic video sequence by using a motion tracking algorithm; (2) trajectory pretreatment: pretreating the extracted trajectory to reduce influences of situations of incomplete trajectory caused by missed tracking, false tracing, sheltering and the like during target tracking or noise point pollution and the like on consequent treatments; (3) similarity degree computation: computing similarity degrees among trajectories by utilizing a DTMD similarity degree formula and constructing a similarity degree matrix; (4) spectrogram clustering: converting the trajectories and similarity relationships thereof into a weighted graph, wherein an apex of the graph stands for the trajectory, edges stand for the similarity degree among corresponding trajectories, computing a characteristic root and a characteristic vector of the similarity degree matrix by utilizing a Laplace equation, and segmenting the graph by utilizing a Fielder value; and (5) clustering result obtaining: converting the segmented result of (4) into trajectory classification, marking the original trajectory and outputting the trajectory clustering result.

The invention discloses a semi-supervised polarized SAR image classification method based on random forest composition. The method mainly solves a problem that a conventional classification method has a defect in representation of a similarity relation between sample points and does not utilize spatial information. The method comprises the following steps of: inputting raw data of a polarized SAR image; extracting relevant features of the data to obtain a data set; constructing an initial random forest model; training two classifiers by using two different sample sets with different attributes to help to train a semi-supervised random forest mode; optimizing the semi-supervised random forest model; constructing a similarity relation graph; constructing a spatial information graph; combining the similarity relation graph and the spatial information graph to obtain a similarity relation matrix between the sample points; and classifying the images and calculating a correct rate. The method constructs an amiable similarity relation graph and spatial information by using the semi-supervised random forest algorithm, improves the classification accuracy of the polarized SAR image, and can be used in civilian and military fields such as geological exploration, disaster relief, target identification and the like.

The invention belongs to the field of data source pattern matching, and relates to a global body-based multi-data-source-pattern matching method. The method includes: converting multiple to-be-matchedpatterns into unified data models, namely pattern bodies; respectively carrying out pattern matching on each pattern body after conversion and a global body according to various related algorithms ofpattern matching, combining calculation results of the multiple matching algorithms to respectively obtain similarity relations between elements in each pattern body and the global body, and using asimilarity degree matrix for representation; and finally, using an aggregation strategy of the similarity relations to aggregate the above-mentioned obtained similarity relations according to transitivity of the similarity relations to obtain a matching result between every two of the multiple data source patterns. According to the method, the problem that one-to-one pattern matching needs to be carried out among multiple data source patterns in or among enterprises can be solved, quality and efficiency of multi-data-source-pattern matching can be significantly improved, and better scalabilityis realized.

The invention provides an image classification method and a device. The method includes the following steps: extracting classification image characteristics of a to-be-classified image; quantizing each classification image characteristic into a plurality of vision words in a vision dictionary according to a similarity relation of the each classification image characteristic and the vision words in the preset vision dictionary, and determining similarity factors of the each classification image characteristic and the quantized vision words respectively; determining weight of the vision words to set a classified vision word histogram of the to-be-classified image according to the similarity factors of each vision word corresponding to different classification image characteristics in the vision dictionary; inputting the classified vision word histogram into an image classifier which is generated by training of a sample vision word histogram which is preset according to a large amount of sample images, and determining classes of the to-be-classified image according to output results. The image classification method and the device can improve accuracy of image classification and reduce classification errors.

An overheating fault simulating method for GIS (gas insulated switchgear) bus joint includes steps of building a mathematical temperature-rising model according to a physical heating process of the GIS bus joint; performing similarity analysis according to the mathematical temperature-rising model, determining an accurate coupling field similarity relation; simplifying the mathematical temperature-rising model, determining an approximate coupling field similarity relation; determining a simulation test scheme of overheating faults of the GIS bus joint and a physical GIS bus joint similarity model under various contact conditions; implementing an overheating fault simulation test on the basis of the physical similarity model of the GIS bus joint according to the simulation test scheme, and acquiring simulation test data. By simulating the overheating faults of the GIS bus joint with the similarity model of the GIS bus joint, overheating fault mechanisms and reliability features are indirectly researched, the defects that protomodel simulation test is high in cost, equipment manufacturing period is long and reliability during test is poor are overcome, and implementing of temperature monitoring and routing inspection of the GIS bus joint are facilitated.

The invention provides an image retrieval method and device. The method includes acquiring a retrieval key word and performing screening in a database according to the retrieval key word to obtain an image set; establishing a first spectrogram model of the image set according to image characteristics to obtain a similarity relation of each two images in the image set; establishing a semi-supervised learning model according to the similarity relation; subjecting the image set to denoising according to the semi-supervised learning model to obtain a denoising image set; and returning the denoising image set to be used as a retrieval result corresponding to the retrieval key word. According to the image retrieval method and device, by means of the establishing of the spectrogram model of the image set, the semi-supervised learning model is established, the image set is subjected to denoising according to the semi-supervised learning model, the image set which is subjected to denoising is returned to be used as the retrieval result corresponding to the retrieval key word, overall denoising is performed on the retrieved image set, and the accuracy of image retrieval is improved.

The invention provides an image matting method and system. The image matting method comprises the steps of 1, training an image matting model by using an image training set; wherein samples in the image training set comprise an original image, and a tripartite graph, a gold standard mask and a gold standard combined mask which correspond to the original image; wherein the image matting model comprises a depth feature extraction module which is used for learning semantic features and detail information features of an original image; The similarity learning module is used for fusing the semanticfeatures and the detail information features to obtain a similarity relation of the pixel points; The mask propagation module is used for obtaining a mask value of each pixel point through a propagation algorithm according to the similarity relation between the trisection image and the pixel points, and outputting an alpha mask image corresponding to the original image; and 2, inputting the original image to be subjected to image matting and the corresponding tripartite image into the trained image matting model, and outputting an alpha mask image corresponding to the original image to be subjected to image matting.

The invention discloses a sparse representation depth image reconstruction algorithm based on a structure dictionary, belonging to the image processing technology field. Firstly, the sparse representation depth image reconstruction algorithm of the invention considers a corresponding depth image and a corresponding color image as a whole, and, during the solution process, the sparse representation depth image reconstruction algorithm improves the reconstruction effect of the depth image and the color image through constructing a structure dictionary having a logic corresponding relation. During the dictionary construction process, the sparse representation depth image reconstruction algorithm based on the structure dictionary utilizes the logical corresponding relation between the depth image and the color image and the similarity relation of the depth image to improve the efficiency and quality of dictionary training. The depth image reconstruction algorithm combines the related theories like the sparse coding on the basis of collecting and arranging a lot of home and abroad related data, analyzes the correlation between data according to the close correlation between the depth image and the color image, mainly solves the reconstruction problem of the sparse representation depth image based on the structure dictionary, reduces the operation complexity, and improves the reconstruction quality of the depth image and the corresponding color image.

The invention relates to a method for extracting topographical spatial information based on a generalized self-similarity principle, which comprises: transforming the topographical spatial information into an energy spectral density space by Fourier transformation; eliminating the influence of boundary effect generated by the topographical data boundary part; drawing a double logarithmic scatter graph formed by the energy spectral density value (S) and the area (A) included by the isometric line of the energy spectral density value (S), and detecting the fractal rules of the energy spectral density and the area; determining the number and the interval of generalized self-similarity relations; determining the threshold and a corresponding fractal filter; and transforming filtered energy spectrum information back to a spatial domain by inverse Fourier transformation, and achieving the aims of decomposing anomaly and ambient fields and extracting interested topographical spatial information. The method has the advantages of wide practicality, high extraction precision and the like, and is suitable for topographical data such as geological data, mineral data, geochemical data, geophysical data, remote sensing data and the like, and the operations of topographical information extraction and topographical data mining such as mineral exploration, resource assessment, environmental pollution assessment, natural disaster analysis, marine vortex extraction and the like.

The invention relates to a remote sensing image classification method and a system based on neighbor regular joint sparse representation. The method comprises steps: a to-be-classified remote sensing image is inputted; training samples and testing samples are divided; a data dictionary is built; a regular joint sparse representation model including a neighborhood pixel weight matrix is built, and a joint sparse representation coefficient matrix for each testing sample and the neighborhood pixel weight matrix are optimized in a joint mode; and according to the data dictionary and the optimal joint sparse representation coefficient matrix for the testing sample and the optimal neighborhood pixel weight matrix, the testing sample is classified. While the joint sparse representation coefficient matrix is optimized, the neighborhood pixel weight matrix is also optimized, the neighborhood pixel weight matrix can reflect a similarity relation and a joint sparse representation error relation between neighborhood pixels, the joint sparse representation coefficient can reflect an approximation relation between the testing sample and the data dictionary more accurately, and an accurate and reliable classification result can be acquired.

The invention discloses a universal entropy decoding method and a universal entropy decoding device for a video decoder, belonging to the field of multimedia digital signal processing. The entropy decoding method recombines video decoding calculation into bit decoding, syntactic element decoding and entropy decoding control three parts, in accordance with the similarity relation between entropy decodings in video decoding. The entropy decoding device in the video decoder comprises a bit decoding module, a syntactic element decoding module, a control module, and a memory. In the method, calculation in each step has similar calculating granularity and data local characteristic, which is helpful to distributing calculating resources reasonably, simplifying the internal interfaces of the universal entropy decoding device in the video decoder, and improving the entropy decoding efficiency in video decoding.

Disclosed are rapid detection method and system for television commercials. The method includes: detecting switching of a television video lens and acquiring lens switching image frames and the value of similarity relation between any two image frames within subsequent preset continuous frames so as to obtain a matrix of relation between any two image frames; calculating offset distance between minimum relation values of each image frame and other image frames; calculating average offset distances of the image frames in each group; if the average offset distances of the frames in the certain group are larger than the preset offset distance threshold, confirming that a commercial begins from the group of image frames; if the average offset distances of the frames in the certain group are smaller than or equal to the preset offset distance threshold, confirming that the commercial ends at the image frames in the group. The fast detection method and system are higher in accuracy of television commercials detection and less in massive data calculation, and user experience is improved evidently.

The invention relates to a fault diagnosis rule extraction method, in particular to an imperfection fault diagnosis rule extraction method based on the quantitative characteristic relation. The method solves the problem that according to an existing imperfection fault diagnosis rule extraction method, fault diagnosis rules cannot be extracted from a decision table containing various unknown attribute values, solves the problems that according to the existing imperfection diagnosis rule extraction method, similarity degrees among examples are not considered, and the similarity relation among the examples is not measured, and solves the problem that original information will change due to an existing manual data filling process. The imperfection diagnosis rule extraction method based on the quantitative characteristic relation comprises the steps of (1) preprocessing original imperfection fault diagnosis data, (2) determining the type of an unknown attribute value, (3) analyzing an imperfection fault diagnosis decision table, (4) conducting reduction, and (5) extracting fault diagnosis rules. The imperfection fault diagnosis rule extraction method is applied to the field of fault diagnosis rule extraction.

The invention provides a node similarity relation detection method based on a combined meta-path in a heterogeneous information network. The method comprises the following steps that: constructing theheterogeneous information network; selecting a reference sample pair; searching a path living example for connecting the source-target nodes of the reference sample pair; carrying out mapping on thepath living example to obtain candidate meta-paths; on the basis of the candidate meta-paths, calculating incidence relation strength among nodes under different path constraints; and calculating information entropy to screen the candidate meta-paths, and finally, obtaining the combined meta-path which describes the incidence relation among nodes. By use of the method, a concept of the combined meta-path is put forward by aiming at the characteristic of rich semantics of the heterogeneous information network to describe the incidence relation among nodes so as to measure a similarity between the incidence relations among nodes, and therefore, the method is suitable for the search task of node pairs with the similarity incidence relation in the heterogeneous information network.

The invention discloses a fuzzy control method of continuous fuzzy interpolation based on relative similarity of a fuzzy set and a system thereof, and relates to the industrial control field. The method comprises the following steps: establishing a continuous fuzzy mapping relation between an input variable and a control variable of a fuzzy controller; performing interpolation computation on a control value based on a similar priority principle according to the relative similarity relation of the fuzzy set in a fuzzy bunch, thus achieving continuous precise control on a controlled object, and comprehensively and perfectly reserving the information from fuzzy reasoning. The method and the system have the advantages of strong anti-interference capability, fast response speed and high control precision, and can meet the actual requirements of the industrial control. The method is applicable to control systems, in which the input variable and the output control variable are continuous variables, especially furnace temperature control systems of large-sized and medium-sized heating furnaces in the industries such as metallurgical industry, chemical industry and the like.

The invention provides a similar traditional Chinese medicinal material excavating method and device; the method comprises the following steps: S2, building a normal compatibility medicament set and a compatibility medicament word frequency set relating to each medicinal material of a given medicament set; S2, calculating similarity between two random medicinal materials in the medicament set according to the normal compatibility set similarity of each medicinal material in S1, thus forming a medicament similarity matrix; S3, taking each medicinal material in the medicament set as a cluster, using a layered cluster mode to build a cluster set with the clusters having similarity bigger a reference threshold according to the medicament similarity matrix obtained in S2, wherein the medicinal material in each cluster is the similar medicinal material. The novel method and device use a big data excavation method to replace an excavation traditional Chinese medicine medicament similarity relation, thus copping with mass traditional Chinese medicine data analysis; the novel method and device are high in excavation precision.

The invention provides a text information processing method and device as well as a terminal. The text information processing method comprises the steps of determining a keyword group corresponding toan obtained search word; obtaining a preset number of keywords corresponding to the search word based on similarity relation between the search word and each keyword in the determined keyword group;and determining text information including one or multiple of the preset number of the keywords based on the preset number of the keywords. Through the text information processing method, the problemthat the search function of a terminal device in related technologies is too mechanical is solved, and the effect of improving the text information search efficiency is further realized.

The invention provides a cross-modal retrieval method based on deep self-supervised sorting hash. The method comprises the following steps of: firstly, learning a label network for reserving a similarity relationship between semantic features and corresponding hash codes; according to the label network, effectively using multi-label information to bridge semantic relevance among different modes; and designing an end-to-end feature learning network for the image and the text, and carrying out feature learning; in one aspect, a semantic correlationship between the label network and an image text network is kept. And on the other hand, the learned features can be perfectly compatible with a specific cross-modal retrieval task. In order to solve the problem that binary partition function coding is very sensitive to partition threshold values, a coding function based on sorting is adopted. The relative order of each dimension is not changed, and the value of the Hash code is not changed, so that the Hash function is not very sensitive to some threshold values, and the robustness of the obtained Hash code is better.

The invention discloses an online cross-modal retrieval method and system based on similarity re-learning, and the method comprises the steps: obtaining an original data sample, dividing the original data sample into a plurality of groups, and constructing a training set; constructing an objective function of Hash code learning, training the objective function by utilizing the training set to obtain a Hash code and a Hash function corresponding to each batch of data, and storing the Hash code and the Hash function into a retrieval library; generating a hash code of the to-be-queried sample according to the sample external expansion mapping; updating the hash code of the original sample data in the retrieval library based on the new sample data in the data stream; and comparing the hash code of the to-be-queried sample with the updated hash code in the retrieval library, sorting according to the Hamming distance from small to large, and returning a retrieval result. According to the method, the Hash representation is generated for the new data on the premise of not retraining the original data, and meanwhile, the retrieval precision is greatly improved by mining the similarity relationship between the new data and the old data and utilizing the label information of the new data.

The invention relates to a Gaussian optics-based focusing type light-field camera parameter calibration method. The invention aims to solve the problem that parameters such as a distance m between thefront end surface of a main lens to a micro lens array, a distance a from the micro lens array to a photosensitive element, and a distance b from the photosensitive element to a virtual image plane cannot be obtained through calibration in the prior art. As for the problem in the prior art, a dimensionless parameter virtual depth v=b / a is introduced, wherein v can be indirectly calculated according to geometrical similarity relations in a fixed focusing type light-field imaging system; a calibration device is built; a plurality of groups of object distances l1 and corresponding v are obtained; and curve fitting is performed on the plurality of groups of (v, l1), a Gaussian optics imaging formula is applied in a combined manner, so that the parameters of a camera can be obtained. Accordingto the method of the invention, calibration can be performed through simple Gaussian optics relationships so as to obtain the key parameters of the focusing type light-field camera; mathematical relationships are simple, and it does not need to define a coordinate system and complex matrix operation; and when the parameters are obtained by means of calibration, the relationship curve of virtual depths and object distances can be obtained. The method can be directly used for three-dimensional depth measurement.

The invention is suitable for the technical field of computers, and provides a similar image retrieval method and device, computer equipment and a storage medium, and the method comprises the following steps: obtaining a to-be-retrieved image; processing the to-be-retrieved image according to a pre-trained unsupervised deep hash model, and determining a hash code of the to-be-retrieved image, wherein the unsupervised deep hash model is generated by iterative optimization training based on a clustering algorithm and a deep hash algorithm in advance, wherein a pseudo tag determined through a clustering algorithm is used as an optimization target in the deep hash algorithm; and according to the hash code of the to-be-retrieved image, determining a similar hash code meeting a preset similarity relationship with the hash code, and determining a corresponding image. According to the method provided by the invention, when the unsupervised deep hash model faces massive images in the training process, the images do not need to be labeled one by one in advance, and the clustering algorithm is directly utilized to carry out false label annotation on the images, so that the method has better applicability.

The invention discloses a feature layer fusion method and device based on graph embedding canonical correlation analysis. The feature layer fusion method comprises the steps that samples in all modesare mapped to a projection matrix of a space of the same classification result, and L21 norm regularization is applied to the projection matrix so that independent complementary features can be selected from multiple single-mode feature spaces at the same time; constructing a data similarity graph matrix to represent the similarity relation of sample points in the single-mode feature space; and learning a corresponding projection matrix for each mode through the regularization target function, and projecting the plurality of mode data to a projection subspace with the maximum discrimination, the maximum correlation and the minimum redundancy to realize multi-mode data fusion. According to the method, multi-modal data fusion is realized, the interference of redundant information in an original feature space is eliminated, the single-modal feature discriminability is improved, the correlation between multi-modal sample sets is enhanced, the recognition performance and stability are improved, the feature fusion effect is good, and the recognition effect is good.

The invention relates to a precision enhancing method for visual media semantic indexing, and belongs to the technical field of visual media processing.The method comprises the steps that semantic indexing is carried out on objects and scenes contained in visual media at first, a confidence coefficient matrix is built, some elements are screened out through threshold value judgment, a weighing non-negative matrix factorization method is applied to reestimate the screened matrix, and overall precision is enhanced; a similarity propagating method is adopted according to an overall precision enhanced result, and a similarity relation between samples is utilized for overall precision enhancing.The precision enhancing method has the advantages that the accuracy of visual media semantic indexing is enhanced by means of various kinds of semantic relations, the method does not depend on a large number of annotation data sets and knowledge bases and has the strong flexibility and adaptability, overall precision enhancing and local precision enhancing are organically combined, and the flexibility and effect of the method are improved; the algorithm is low in calculation complexity, strong in extendibility and suitable for practical industrial application.