75 results about "Feature synthesis" patented technology

The invention provides a modeling method of the characteristics of population space-time dynamic moving based on multisource data fusion. The modeling method comprises: A. inputting map data, mobile phone locating data and floating vehicle data into a system and managing data organization according to requirements; B. establishing a spatial analysis model of the characteristics of population moving based on the mobile phone locating data and the floating vehicle data; C. applying the spatial analysis model of the characteristics of the population moving to carry out multisource data fusion of the map data, the mobile phone locating data and the floating vehicle data to obtain integrated information of the characteristics of the population moving; and D. analyzing the characteristics of various population moving according to the integrated information of the characteristics of the population moving and publishing an analyzed result by the geographic information system. The modeling method can acquire data of urban population space-time dynamic distribution and moving characteristics with large data amount, high quality and space-time characteristics, obtain basis of accurate population distribution and population moving characteristics, and provide decision-making supports for urban planning, land use planning, transportation planning and the like.

The invention discloses a method for optimizing tight oil horizontal well crack network parameters through the transformation volume, and belongs to the field of tight oil horizontal well reservoirs. According to the method, according to a tight reservoir, firstly, the geologic feature of the reservoir is evaluated, the specific reservoir condition for volume fracture to form a volume crack network system is explicated, then, the influences of different transformation volumes and crack arrangement modes of horizontal well volume fracture on capacity are studied, and the specific crack network parameters are optimized. The method at least includes the steps that firstly, a basic crack network feature comprehensive evaluation method for tight reservoir volume fracture is established; secondly, a capacity prediction and simulation model method for tight oil volume fracture is established. The tight oil layer horizontal well fracture crack network parameters are optimized through the transformation volume formed in the fracture process, the transformation volume is formed in a tight oil layer, a matrix seeps towards a crack at the shortest distance, the driving pressure of effective flow is greatly reduced, and flow under the extremely-low permeability condition is achieved; the single well capacity can be greatly improved.

A system and method for ink database searching using handwriting feature synthesis is disclosed which allows a digital ink database to be searched using a text-based query. Using a writer-specific handwriting model derived from a handwriting recognition system or suitable training procedure, a text query is converted into feature vectors that are similar to the feature vectors that would have been extracted had the author of the digital ink database written the text query by hand. The feature vectors are then used to search the database. This allows the searching of a digital ink database when the only input mechanism available is text entry, and can allow a person other than the author of the digital ink database to search the digital ink database.

The invention provides a method for identifying an overhead high-voltage wire automatically from an infrared image, and aims to provide a method for extracting an overhead high-voltage wire from an infrared image reliably and accurately in real time. According to the technical scheme, the method comprises the following steps of: reading an infrared digital image which is shot from an outdoor scene into a computer program, and converting the image into a grayscale image; performing image segmentation, geometrical feature extraction, luminance feature extraction, edge contour extraction and piecewise linear feature extraction on the grayscale images by an image extraction program, and dividing the image into a target region and a background region; counting and identifying geometrical features, luminance features and piecewise linear features in the regions automatically by a feature comprehensive analysis and identification module, scanning and screening one by one, and removing suspected targets which do not have the infrared image feature of the high-voltage wire to obtain initial pixels of the image of the overhead high-voltage wire; and outputting and displaying the acquired final pixels of the image of the overhead high-voltage wire by a high-voltage wire pixel marking output module.

The invention discloses a three-dimensional image quality evaluation method based on gradient information guided binocular view fusion. According to the method, a Sobel operator and a LoG operator areutilized to construct a united statistical gradient map which serves as a weight map for binocular view fusion, and a corresponding intermediate reference image and a corresponding intermediate distortion image are obtained; then, image feature information extraction is performed on the intermediate images, wherein image feature information comprises edge information, texture information and contrast information, and depth information is extracted from a disparity map of a reference and distortion three-dimensional image pair; and last, a final image quality objective evaluation score is obtained through measurement of a feature similarity and SVR-based feature integration and quality mapping, and measurement of three-dimensional image quality loss is realized. Experiment results show that an algorithm proposed based on the method has good accuracy and robustness.

An embodiment of the present specification provides a computer-executed text risk prediction method and apparatus. According to one embodiment of the method, after obtaining a text to be predicted, onthe one hand, keyword features of the text are analyzed, and part-of-speech related features are analyzed for words in the text; On the other hand, mining the important vocabulary in the text, so asto get more related vocabulary and generate extended features. Furthermore, the paper analyzes the keyword features, part-of-speech related features and extended features of the text comprehensively to determine the risk of the text. This embodiment can improve the accuracy of text risk prediction.

The invention discloses a method for classifying imbalance heart beats based on a multi-mode neural network. The method involves an electrocardiosignal preprocessing module, an imbalance data processing module and a feature extracting and classifying module. The preprocessing module conducts denoising and segmenting on an electrocardiosignal; the imbalance data processing module is a core of a system, and sequentially introduces three kinds of methods for processing imbalance in combination with the features of the electrocardiosignal and the features of an algorithm, wherein the methods include boundary sample feature linear synthesis (BLSM), a context feature comprehensive module (CTFM) and a two-phase training (2PT); the feature extracting and classifying module obtains high-order features of all categories of heart beats and realizes final heart beat classification. The method has the advantages that in the prior art, the heart beats cannot be well classified, the corresponding solutions are provided from the aspects of sampling, features and algorithms, and therefore the accuracy of classifying is improved. The method is suitable for solving of the problem of classifying imbalance of time sequence data, images and others, and has generality.

The invention discloses a zero-sample cross-modal retrieval method based on multi-modal feature synthesis. The method comprises the steps: employing two adversarial generative networks, synthesizing feature representations of different modals through class embedding shared by two modal data, then mapping original modal data and synthesized modal data to a common subspace, and carrying out the aligned distribution of the original modal data and the synthesized modal data. Therefore, the relation between different modal data of the same category is established, and knowledge is migrated to unseen categories. The cyclic consistency constraint further reduces the difference between the original semantic feature and the reconstructed semantic feature, and well establishes the association between the original representation and the semantic feature in each modal, so that the common semantic space is more robust, and the accuracy of zero-sample cross-modal retrieval is improved.

The invention discloses a method for distinguishing fog concentration in intelligent image monitoring of a power transmission line, which adopts a multi-feature synthesis method and comprises the following steps of: extracting a plurality of features related to fog in a picture, and synthesizing the plurality of features extracted from each image into a feature parameter table to form a two-dimensional feature map; utilizing a deep learning convolutional neural network to train and model the characteristic parameter graphs of a large number of sample images; and judging whether the unknown image has fog or not by using the trained model, and distinguishing the concentration levels of the fog, namely clear fog, medium fog and dense fog. According to the method, a plurality of deep and irrelevant features of the provider are integrated to serve as image feature parameters, and then a nonlinear classification algorithm model is trained for a large number of sample images by using the convolutional neural network, so that the problem that the accuracy of detecting whether the image is foggy is low is solved.

The invention discloses a computer-assisted ultrasonic diagnosis method for a left atrium/left auricle thrombus. The technical scheme includes that data mining technology, a pattern recognition theory and medical clinical information are combined, a gray-scale video and a real-time three-dimensional dynamic video serve as research objects, all-dimensional information in an image is accurately acquired, potential disease association rules in the image information are mined, and multiclass characteristics are comprehensively analyzed to obtain a detection method for automatically detecting and classifying the left atrium/left auricle thrombus. A thrombus recognition method can avoid missed diagnosis and misdiagnosis caused by subjective reasons such as inadequate experience or visual fatigue of doctors, and patients with suspected left atrium/left auricle thrombi clinically detected by transesophageal echocardiography can be confirmed as early as possible, so that the patients without thrombosis can receive cardioversion treatment of atrial fibrillation as early as possible. The method is simple and convenient to operate and high in practicability, and has important guiding significance for diagnosis and treatment of the left atrium/left auricle thrombus and ventricular fibrillation.

The invention discloses a video quality evaluation method and device, a video processing device and a medium. The video quality evaluation method comprises the following steps: constructing a trainingdata set of an evaluation model to be used for a neural network system, wherein the training data set comprises manual annotation data and user click rate data; an attribute feature obtaining step: for each sample in the training data set, obtaining each attribute feature, related to the video attribute data, of each sample through a neural network system; a feature synthesis step: for each sample, performing feature synthesis processing on each attribute feature of the sample to obtain a synthesis feature of the sample; and a training step: performing confrontation training on the syntheticfeatures of each sample by combining the manual annotation task and the user click rate task, and generating an evaluation model for evaluating the video quality. According to the video quality evaluation model provided by the invention, a more accurate and effective evaluation result can be obtained for a new video or an exposed video.

The invention relates to the technical field of speech recognition, in particular to a speech recognition method and device, electronic equipment and a storage medium, which can be applied to various scenes such as cloud technology, artificial intelligence, intelligent traffic and auxiliary driving and are used for efficiently and accurately realizing speech recognition of multiple dialect target languages. The method comprises the following steps: acquiring to-be-recognized voice data of a target language; extracting voice acoustic features corresponding to each frame of voice data in the to-be-recognized voice data; performing deep feature extraction on the voice acoustic features to obtain corresponding dialect embedding features; encoding the voice acoustic features to obtain corresponding acoustic encoding features; and based on the dialect embedding feature and the acoustic coding feature, performing dialect speech recognition on the to-be-recognized speech data to obtain target text information and a target dialect category corresponding to the to-be-recognized speech data. According to the method, the dialect embedding feature and the acoustic coding feature are combined for comprehensive learning, so that speech recognition for recognizing various dialects can be efficiently and accurately realized.

The invention provides a video copy detection method based on time-domain visual attention. The method comprises the following steps of: acquiring visual attention changes between different video frames according to a visual attention mechanism, and acquiring representation of time-domain attention degree; calculating a time-domain attention weight of a video frame in a video clip according to the time-domain attention degree to form a visual attention transfer image of the video clip; and finally, extracting video hash on the generated visual attention transfer image provided with time domain and space domain information. By the method, the time domain information of the video is fully considered, a video frame which makes the video content prominent is weighted, the time domain and space domain information is integrated through the extracted characteristics, and the video copy detection method has high robustness on time domain attacks.

A wireless channel characteristic comprehensive information fusion method includes the steps of characteristic extraction, information entropy calculation, low entropy elimination, key negotiation and staggered fusion. Characteristic extraction is used for obtaining three types of wireless channel characteristic values, namely, the strength, the phase and the Doppler frequency shift of a received signal, from a wireless communication system. Information entropy calculation is used for estimating the information amounts of the three types of wireless channel characteristic values. Low entropy elimination is used for deleting the wireless channel characteristic values lower than a certain threshold value and conducting key negotiation on the wireless channel characteristic values conforming to the conditions. Key negotiation is used for generating symmetric keys through the wireless channel characteristic values. Staggered fusion is used for arraying the keys generated through the different channel characteristic values in a staggered mode and generating a final key. By means of the method, the aim of generating the symmetric keys through the various types of wireless channel characteristics together is achieved, the high key generation capacity, the high key security and the high environment adaption capacity are provided, and the development of the next generation of symmetric key technology is powerfully supported.

The invention discloses a machine learning model construction method and a computer readable storage medium. The method comprises the following steps: acquiring auxiliary data according to a preset keyword; obtaining business data, and determining data items corresponding to the input items and data items corresponding to the output items; performing label marking on the business data of which thevalues of the data items corresponding to the output items are not empty; obtaining a first sample according to the service data; obtaining service data with label marks in the first sample as a second sample; synthesizing feature items through a feature synthesis technology, and merging the feature items into a second sample as input items; performing positive and negative sample equalization processing on the second sample through a synthetic minority class oversampling technology, and taking newly synthesized sample data as a third sample; combining the second sample and the third sample to obtain a fourth sample; and training the fourth sample through a preset machine learning algorithm to obtain a machine learning model. According to the invention, the accuracy of the machine learning model can be improved.

The invention discloses a self-learning emotion interaction method based on multi-modal recognition. The method comprises the following steps: respectively collecting voice, face and gesture signals by a non-contact channel; performing feature extraction on the signals to obtain preliminary features of the signals; inputting the features into a bidirectional LSTM layer to obtain single-mode private information and multi-mode interaction information, and obtaining fusion features according to the information; predicting the emotion of a user based on a classification learning algorithm in combination with the multi-modal fusion features and a historical emotional state curve, and selecting an interaction mode; in the interaction mode, giving an interaction response according to the dialoguememory network; and finally, feeding back and optimizing the emotional state curve and the dialogue memory network according to the interaction effect. According to the method, an operator is allowedto input information through multiple channels of a non-contact man-machine interaction interface, multi-modal fusion features are comprehensively considered, and an interaction task is completed incombination with the historical emotional state and the dialogue memory network.

A method is provided for training and using an object classifier to identify a class object from a captured image. A plurality of still images is obtained from training data and a feature generation technique is applied to the plurality of still images for identifying candidate features from each respective image. A subset of features is selected from the candidate features using a similarity comparison technique. Identifying candidate features and selecting a subset of features is iteratively repeated a predetermined number of times for generating a trained object classifier. An image is captured from an image capture device. Features are classified in the captured image using the trained object classifier. A determination is made whether the image contains a class object based on the trained object classifier associating an identified feature in the image with the class object.

The invention discloses a space function unit-based land area province city and county space planning three-region identification method. The method comprises the steps of performing classification and grading on landforms by adopting refined DEM data; identifying a primary space function unit; extracting the space function unit; and determining a three-region classification rule and automaticallyclassifying and identifying three regions. By integrating element characteristics in the unified space function unit, the accuracy of three-region classification is ensured; and the three-region classification rule is modeled through a decision tree algorithm, and the automation of three-region classification definition is realized, so that the purposes of effectively and finely integrating and representing space and attribute characteristics of geographic elements on space units, accurately classifying the three regions, enabling the minimal space units to be independent mutually, homogeneous internally and stable structurally and improving working efficiency of space planning are achieved.

The invention belongs to the technical field of radar radiation source signal characteristic evaluation in electronic countermeasure, and discloses a radar radiation source signal intra-pulse characteristic comprehensive evaluation method and system. The method comprises the following steps of firstly, carrying out feature extraction on a received radar radiation source signal, and carrying out feature evaluation index measurement and normalization according to an established evaluation system; carrying out improved interval analytic hierarchy process by combining expert priori knowledge and an actual environment, and establishing a nonlinear equation optimization model by using an improved projection pursuit algorithm; and finally, performing final subjective and objective decision fusionby using a projection spectrum gradient algorithm. The radar radiation source signal intra-pulse characteristic evaluation method and system can reasonably and effectively realize various radar radiation source signal intra-pulse characteristic evaluations based on an actual environment, and scientific and effective evaluation is carried out to help to select characteristics capable of highlighting radar radiation source signals so as to facilitate the subsequent radar radiation source signal sorting and identification.

The invention discloses a method and device for obtaining image blurring level classification, electronic equipment and computer readable storage medium. The method for obtaining the image blurring level classification comprises the steps that blurring level graphs and significance graphs of images to be recognized are obtained; and classifications of blurring levels of the images to be recognizedare determined based on the blurring level graphs and the significance graphs. By obtaining the blurring level graphs and the significance graphs, that is, the features of different dimensions comprehensively consider the classifications of the blurring levels of the images to be recognized, the situation that misjudgment is caused, such as images with blurring backgrounds are intended to be badimages, caused by local evolution of image blocks is avoided, the accuracy rating of ensuring the image blurring levels is improved so that managing images better according to the blurring levels is convenient for users, and use experience is improved.