143results about How to "Efficient and accurate extraction" patented technology

The invention relates to a significant target detection method based on an FCN (fully convolutional network) and a CNN (convolutional neural network). The method comprises the following steps: firstly, extracting the deep semantic information by use of the FCN, wherein a fixed size of an input image is not needed, end-to-end prediction is performed, and the training complexity is reduced. Accuracy optimization is performed by adopting the CNN and extracting local features to obtain a rough detection result of the FCN. In the invention, the semantic information in an image can be extracted accurately and efficiently, and the improvement of significant target detection accuracy in a complex scene is promoted.

An image correction method comprising the steps of: specifying a region including the face of one person in an image of people displayed in a screen of image display means in response to an input from an operator; extracting a correction target region including at least the face of the person in the image, based on information on at least one of hue, saturation, lightness, and a position of an outline of the specified region; setting a degree of correction to be carried out on each position in the correction target region in such a manner that the degree of correction becomes lower from the inside of the correction target region to the outside thereof near a boundary of the correction target region; and correcting at least one of hue, saturation and lightness in the correction target region based on the degree of correction.

The invention discloses a stereoscopic vision comfort level evaluation method based on motion features inside an area of interest. The stereoscopic vision comfort level evaluation method based on the motion features inside the area of interest includes the steps: formulating a strategy of computing motion feature vectors inside a stereoscopic vision interest area, building a stereoscopic vision model which represents a non-liner relationship between stereoscopic video motion features and a stereoscopic vision comfort level through subjective testing, selecting a spatial domain mean value strategy, a minimum value integration strategy and a time domain mid-value strategy based on the motion feature vectors inside the stereoscopic vision interest area and a nonlinearity stereoscopic vision model, and formulating the stereoscopic vision comfort level evaluation method. Uniformity of between evaluation results provided and results of human-eye subjective stereoscopic comfort is good.

The invention relates to a quick blind source separating fetal electrocardioscanner and a detection method. The electrocardioscanner comprises a dual channel acquisition and analog-to-digital conversion module, an FPGA provided with an internal storage, a DSP and an ARM for separating a blind source, and an output device, wherein the dual channel acquisition and analog-to-digital conversion module is electrically connected with a common electrocardiogram lead cable and the FPGA through two data channels; the FPGA is electrically connected with the DSP through two data channels; the DSP is also sequentially electrically connected with the ARM and the output device ; and the DSP and the ARM are all externally connected with a program storage and a data storage. The electrocardioscanner adopts a high-speed acquisition and processing structure, detects fetal electrocardiosignals in aliasing electrocardiosignals in real time, adopts a blind source separation method to solve the problem that time domains and frequency domains of mother electrocardiosignals and the fetal electrocardiosignals are mutually overlapped and difficult to separate, and extracts the fetal electrocardiosignals efficiently and accurately for diagnosis.

The invention relates to a time-frequency-transformation-based blind extraction method of fetal electrocardiography, which is based on the characteristic that the time domain of an original signal is relatively sparse. The method comprises the following steps: extracting a plurality of paths of mother-fetal mixed electrocardiosignals from different parts of mother's abdomen, and preprocessing thecollected mixed signal, wherein the preprocessing comprises the steps of correcting the baseline shift of the signals, filtering off 50Hz power frequency interference, filtering off high-frequency noise interference such as myoelectricity and the like; selecting two paths of the mother-fetus mixed electrocardiosignals according to the maximal signal-to-noise ratio and separately searching mother-fetal electrocardio wave group positions of the two paths of mixed signals with an R-wave positioning technology to obtain time intervals in which the mother-fetus electrocardio in the two paths of mixed signals are relatively sparse, converting the obtained relatively sparse time intervals into a time frequency domain by using a fuzzy function, calculating signal items and cross items in each time frequency distribution, constructing a comparison function by using generalized Rayleigh's quotient, and finally separating fetal electrocardiosignals from the two paths of mother-fetal mixed electrocardiosignals. The method provided by the invention is tested on the basis of actual collected data, and can well separate fetal electrocardiosignals.

The invention relates to implementation of a sea ice parameter extraction method on the basis of a satellite image and a system and belongs to the technical field of computer application. The sea ice parameter extraction method on the basis of the satellite image is characterized in that an algorithm of eliminating warp and weft noise by the satellite image is implemented; image element types are accurately identified and classified; and a land and sea ice boundary and a sea ice and seawater boundary are extracted by a boundary tracing algorithm. On the basis, three parameters, i.e. the sea ice intensity, the sea ice thickness and the sea ice size, in the satellite image are calculated; and a user opens the image at a computer terminal and carries out the operations of preprocessing the image, identifying key elements (sea ice, land, seawater), describing the boundaries, extracting the sea ice parameters and storing a data result. The extraction method is reasonable, efficient and accurate and strong in scientificity; and the system is regular in design, high in reliability and user-friendly.

The invention provides a fetus electrocardio blind separation method based on the relative sparsity of time domains of source signals, comprising the following steps of: firstly acquiring mutually overlaid mother-fetus mixed electrocardiosignals of mother and fetus electrocardio in two paths from the body surface of the abdomen of a mother; then carrying out pretreatment on the acquired mother-fetus mixed electrocardiosignals, wherein the pretreatment includes the steps of: correcting the baseline drift of the mother-fetus mixed electrocardiosignals, and filtering the interference of power frequency of 50 Hz, the interference of high-frequency myoelectric signals, and the like; respectively positioning the mother electrocardiosignal and the fetus electrocardiosignal in the pretreated mother-fetus mixed electrocardiosignals; then searching relatively sparse time slices of the mother electrocardio and the fetus electrocardio; and finally separating the mother electrocardiosignal from the fetus electrocardiosignal by utilizing a basic blind source separation algorithm resolved through the general features of a matrix. The invention solves the problem that the time domains and frequency domains of the mother electrocardiosignal and the fetus electrocardiosignal are mutually overlaid and difficult to separate and can efficiently and accurately extract the fetus electrocardiosignal used for medical diagnosis.

An image area extracting method for extracting the desired areas from the image accurately and efficiently. The rough boundary area, in which the boundary area is roughly assigned between the desired area and areas other than the desired area, is divided into a plurality of small areas. The area extracting procedure is executed on each small area and the desired area and areas other than the desired area are detected from each small area. The boundary area mask image is formed based on the detection result and the desired area is extracted based on the boundary area mask image. Thus, the desired area can be extracted from the image accurately and efficiently.

The invention discloses a high-resolution SAR image individual building extraction method, and the method comprises the steps: firstly carrying out domain ontology modeling through combining the imaging features of a high-resolution SAR image individual building and the characteristics of a complex morphological structure, and constructing an individual building body semantic model in a high-resolution SAR image; secondly obtaining image regions with the good homogeneity and clear boundary through employing SAR image segmentation based on an object, wherein the image regions are the basic processing units extracted by building base units; combining with the related rule of the building base units in the body model, wherein the extracted image object characteristics comprise regional features, shape features, geometric features, texture features, and topological features; forming the object expression of body semantic description according to the body semantic rule and the object features, guiding the image objects to be automatically recognized as the building base units, and achieving the large-scale individual building recognition which takes the semantic knowledge as the center. The method can extract a large-scale individual building in the high-resolution SAR image accurately and quickly.

The invention is suitable for the technical field of image processing, and provides a video processing method, a video processing device and terminal equipment. The video processing method comprises the steps: obtaining a to-be-processed video; determining a scene type of each video frame in the to-be-processed video; according to the scene category of each video frame, carrying out target recognition on each video frame in the to-be-processed video through a trained first deep learning model, and obtaining a target recognition result, wherein the target recognition result indicates whether each video frame contains a target object, and the target object corresponds to the scene category; according to the target recognition result, identifying a plurality of target frames of the to-be-processed video, wherein the target frames comprise the target object; and according to the time sequence of each target frame in the to-be-processed video, sequentially combining the plurality of targetframes to obtain the target video.

The invention discloses a named entity identification method oriented to the field of network security. The method comprises two stages of model training and sample selection. The first stage comprises the steps: acquiring an initial character vector which contains semantic information and changes dynamically through ALBERT training of a pre-training language model, sending the initial character vector into a Bi-LSTM + CRF network to be trained, and outputting a label sequence with the maximum input text sequence probability is output. The second stage comprises the steps: based on the model obtained through training in the first stage, selecting network security text data with marking value and training value for manual and machine marking in the mode of combining active learning and self-learning, and iteratively training the model after the network security text data with marking value and training value are added to existing marked text data. According to the method, the accuracy of network security entity identification is remarkably improved, and the problems of lack of marking corpora, high marking cost and the like in the network security field are effectively relieved.

The invention provides a deep convolutional neural network capable of completing an image style recognition task, and provides an image color expression mode migration method. According to the method,content representation features of a to-be-processed content image and an initialized image are extracted through the deep convolutional neural network pre-trained in an object recognition task, calculation of a content loss function is carried out, style representation features of the initialized image and a style image are extracted through the deep convolutional neural network pre-trained in an image style recognition task, calculation of a style loss function is carried out, and finally, a result of a total loss function is obtained; and a gradient descent algorithm is used to start fromthe initialized image to carry out iterative optimization in an image domain according to the total loss function to obtain a corresponding image of a smallest result of the total loss function to usethe same as a final result image. According to the method of the invention, image color expression mode migration can be completed, a case where massive strokes similar to impressionist painting images exist in an output result image can be avoided at the same time, and structure information of original natural images can be retained.

The present invention relates to an example-based concept-orietned data extraction method. In an example labeling phase, the exemplary data string is converted into an exemplary token sequence, in which the target concepts and filler concepts are labeled to be tuples for use as an example, and thus an exemplary concept graph is constructed. In the data extraction phase, the untested data string is converted into an untested token sequence to be processed, and, based on the associated concept recognizers defined by the tuples in the example labeling phase, it is able to detect the concept candidates and establish the composite concepts and aggregate concepts, thereby constructing a hypothetical concept graph. After comparing the exemplary concept graph with the hypothetical concept graph, the optimal hypothetical concept sequence in the hypothetical graph is determined, so as to extract the targeted data from the matched target concepts.

The invention relates to a high-definition aerial remote sensing data automatic road extraction method; the method employs an aerial remote sensing image and DSM data of a corresponding researched area for extracting ground and overground areas, constitutes streets and extracts central lines of roads via a series of morphological conversions, including open and close operation, convex set shell, concave set shell, range conversion, watershed segmentation, and the like. The central lines and border lines of roads obtained by the disclosed method are continuous curves. Comparing to the method of the prior art, the algorithms and theories are complete, the road lines are prices, the calculations are quick, and the algorithms are good in adaptability; indicated after experiments that the disclosed method is very strong in robustness with respect to road extraction.

The invention belongs to the technical field of high voltage electrical equipment partial discharge detection and discloses a signal and noise separation method for a partial discharge signal and an information data processing terminal. A partial discharge waveform signal is collected; the partial discharge signal is extracted, a single pulse signal is acquired, a peak value and the phase of the pulses are recorded, and a phase distribution spectrum is drawn; for the extracted pulses, through multiple band pass filters, the down pulse peak value information of the corresponding filters is obtained, through principal component analysis and dimension reduction, characteristic parameters are obtained; the characteristic parameters are subjected to clustering analysis, the pulses are classified into multiple categories, a single-class pulse phase distribution spectrum is drawn according to the classification category number, the pulse peak value and the pulse phase, and the PD signal and the interference signal are determined. The method is advantaged in that the method can extract various pulse waveforms efficiently and accurately, utilizes the peak information under the multiple bandpass filters as a feature vector, can fully reflect the pulse waveform information and improves precision of separation of the interference signal and the partial discharge signal.

Provided are a screening method and system for a fundus image of glaucoma based on deep learning. The screening method comprises the steps that the fundus image to be identified is acquired; an FCN model is extracted through an optic disk to position and cut the optic disk, and optic disk sub-images are extracted and obtained; the optic disk sub-images are identified through a classification CNN model, and a classification result of the fundus image of glaucoma is obtained. According to the screening method and system for the fundus image of the glaucoma based on deep learning, identificationis only conducted on the cut and extracted optic disk, so that information processing amount is reduced, and the screening method and system can assist a doctor in improving the detection efficiency of glaucoma.

The invention discloses a method for automatically extracting the tree height of a standing forest having high canopy density based on TLS and UAV. By utilization of UAV photographic measurement pointcloud data and ground precise point cloud data, the highest point and the lowest point of an individual tree are extracted; according to an individual tree positioning coordinate and automatic identification highest point information, the judged highest point and a positioning point are projected in a xoy plane; the identification accuracy is judged; and thus, a tree height value is rapidly extracted. The cost is reduced; the efficiency is increased; furthermore, on the premise that the tree height extraction precision is increased, partial manual survey is replaced; a lot of manpower and material resource consumption can be reduced; a test result shows that: the tree height extraction method provided by the invention has relatively high precision for tree height estimation of an artificial metasequoia forest; the tree height extraction value whole is closer to the actual measured value; and furthermore, the method in the invention also makes effective recommendations for forest resource assessment and business planning of small and medium-sized farms.

The invention relates to a small assemblage robot online automatic welding device comprising a welding gantry, a linear guide rail, a servo motor, a robot, a sensor fixing bracket, a 3D line laser scanning sensor and an upper computer; and the welding gantry is fixedly connected with the robot and the sensor fixing bracket, the sensor fixing bracket is horizontally arranged and is perpendicular tothe moving direction of the welding gantry, and the 3D line laser scanning sensor is mounted on the sensor fixing bracket. The 3D line laser scanning sensor is driven by movement of the welding gantry to scan a welding area to obtain point cloud data, the point cloud data are subjected to model reconstitution to generate a robot execution program, and robot welding is carried out. A welding operation method is also disclosed. The small assemblage robot online automatic welding device and the operation method can realize establishment of small assemblage workpiece models in any shape, and thewelding operation generation efficiency and the robot operation efficiency are improved.

The invention discloses a database abnormality query detecting method based on target condition relevance rules. The method comprises a rule mining method and an abnormality detecting method, wherein the rule mining method comprises an initialization phase, a frequent-item generation phase and a target condition relevance rule generation phase; the abnormality detecting method adopts a rule bank obtained in the rule mining process to carry out abnormality detection; and just when the rule existing in the rule bank meets the rule support conditions, the operation is normal, otherwise, the operation is abnormal. For the invention, the rule mining efficiency, the detection efficiency and precision are high; the detection failure and the omission factor are low; and the universality and the expansibility are high.

The invention provides an online log analysis method and system and electronic terminal equipment thereof. The method comprises the following steps: S1, conducting log preprocessing on each unanalyzedlog to obtain a plurality of unanalyzed log sequences with different log lengths, and classifying the unanalyzed log sequences into corresponding first log groups; S2, acquiring a log character string of each log sequence in the first log group, calculating the similarity of the log character strings, and performing online clustering based on the similarity of the log character strings; and S3, taking the unanalyzed log sequence as a query item, matching the common node with the template in the template spanning tree of the second log group to obtain the template. The method has the advantages that the logs are classified according to the length, the logs are subjected to secondary clustering based on the log character string similarity, finally, the template spanning tree is used for extracting the log template, the method can efficiently and accurately extract the log template from the unstructured logs, and a data analyst can conveniently carry out higher-level analysis and processing on the logs.

An embodiment of the invention provides a vehicle on-line detection and diagnosis system and method. The system comprises a far-end server for detecting vehicle data; an OBD detection device connected with an OBD interface of a vehicle to obtain actual OBD data of the vehicle; a handheld information acquisition device for obtaining vehicle paint thickness; and a detection terminal connected with the OBD detection device, the handheld information acquisition device and the far-end server. The detection terminal is used for comparing the actual OBD data of the vehicle and pre-stored standard operating parameters corresponding to the vehicle model to generate a diagnostic report.

The invention discloses a resume identification method which is characterized by comprising the following steps of a first step, setting all latent keywords in a resume; a second step, selecting a to-be-analyzed resume; a third step, performing preprocessing on the resume according to the preset keywords, and analyzing the keywords which are included in the resume; a fourth step, differentiating a simple information field and a complex information field according to the information of the keywords which are included in the resume; a fifth step, performing secondary analysis processing on the complicated information field, extracting sub-item information; and a sixth step, outputting the simple information field and the complex information field. The resume identification method can realize high-efficiency accurate extraction for resume information and furthermore has high extraction accuracy.

The invention relates to a high resolution seismic wavelet extracting method based on high-order statistics and an ARMA (autoregressive moving average) model, which belongs to the field of seismic signal processing. The high resolution seismic wavelet extracting method provided by the invention is characterized in that: under the precondition of performing ARMA parameter simplified modeling on seismic wavelets, SVD (singular value decomposition) based on autocorrelation function is adopted to determine an order of AR part, an MA order determining method is provided for integrating information content criterion function in a high-order cumulant MA order determining method, and the MA order determining accuracy rate in a seismic wavelet ARMA model is improved; an SV-TLS (singular value decomposition - total least squares estimation) and a cumulant method are respectively adopted to estimate wavelet parameters; and under the precondition of ensuring wavelet precision, the model order is decreased as far as possible to improve the operation efficiency and to finally realize seismic wavelet extraction in high efficiency and high precision. Through the data simulation verification and the practical seismic data processing demonstration, the method provided by the invention is proved to effectively improve the estimated precision and extracting efficiency for the seismic wavelets and to have obvious effect even under short-time seismic data and strong noise pollution.

The invention discloses a movie and television entity identification method based on IDCNN-crf and a knowledge graph. The method comprises the following steps: A, collecting movie and television datainformation; B, collecting a large amount of user search film and television data converted into texts through voice, and performing data analysis to obtain training data for model training; C, training the entity recognition model; D, collecting prediction data needing to be predicted, and inputting the prediction data into the entity identification model for prediction after data preprocessing;and E, carrying out verification processing on the model prediction result and outputting the result. The method provided by the invention can solve the entity identification problems of few annotation data, short text and spoken movie and television text data.

The invention relates to a screening method of dipeptide functional monomers having excellent separating capacity on saccharide. A dipeptide compound obtained on the basis of screening of standard deviation parameter D and R values is constructed, the dipeptide compound obtained through screening is further coupled with a group containing double bonds, polymerizable dipeptide functional monomers are obtained and are polymerized on surfaces of various matrixes, and a series of polymer modified chromatographic stationary phase materials are obtained. The materials have excellent separating capacity on oligomeric galactose, oligomeric fructose and sialic acid derivatives; meanwhile, the materials have excellent selective enrichment performance on the glycopeptide and separating capacity on a carbohydrate chain structure, have very high glycopeptide recovery rate and repeatability and have the wide application prospect in the fields of posttranslational modification proteomics research and the like.

The invention discloses a fetal electrocardiogram signal extracting method based on wavelet threshold denoising. The method comprises the following steps of: carrying out stationary wavelet transform processing on a maternal abdomen signal, and decomposing to obtain each layer of wavelet coefficient; independently processing each layer of wavelet detail coefficient, removing fetal electrocardiogram wavelet coefficients, reconstructing remained maternal electrocardiogram wavelet coefficients, and obtaining maternal electrocardiogram in the abdomen signal; removing the reconstructed maternal electrocardiogram from the abdomen signal to obtain a fetal electrocardiogram signal; and denoising the extracted fetal electrocardiogram by utilizing a wavelet correlation denoising algorithm to obtain a clear fetal electrocardiogram signal. Due to the method, the accurate extraction of the fetal electrocardiogram signal can be realized. The method is based on a single channel, the computation of wavelet transform modulus maxima is avoided, and better instantaneity is realized.

The invention relates to the technical field of information security, in particular to a method for automatically extracting worm features based on behavior trace analysis. The method comprises the following steps: modeling for the access habit of a network by learning the network traffic; detecting based on a CUSUM (cumulative sum) algorithm to determine that the flow which goes against a habit model in the network is a suspicious worm flow; defining and classifying behavior traces of network worms; performing correlation analysis on the behavior traces of the network worms by using Petri network in the suspicious worm flow; finally determining and extracting the feature codes of the network worms in the traces by applying an evaluation function. Experiments prove that the method can effectively and accurately extract the feature codes of the network worms. The following conclusion is obtained by combining theoretical analysis and experimental data: although identities of the worms cannot be accurately distinguished by the behavior traces of the network worms, the determination of the positions of the feature codes of the worms can be facilitated, so that the feature codes of the worms are effectively extracted.

The invention relates to a text classification method based on a tfidf algorithm and related word weight correction. The text classification method comprising the steps of S1, extracting category keywords; S2, forming a sliding text window, setting a word weight and modifying the position thereof in the sliding text window; S3, calculating the word frequency of words according to a word frequencystatistical correction function; S4, performing weighting calculation according to a TFIDF algorithm to realize vectorization of the words in a text; and S5, classifying the text by a SVM classifier.In the process of text classification, the weight of the category keywords is increased, so that the result of the text vectorization better reflects the text information. The method of the inventionintroduces the text sliding window and takes full account of the position information of the words in the text. The category keywords come from partially training data and users, the category keywordsare extracted by using the tfidf algorithm, the characteristics of the keywords can be extracted efficiently and accurately, meanwhile, the case of a few category keywords of an actual application scene is balanced, and the category keywords are extracted comprehensively and accurately.

The invention discloses a trans-scale geographic information linkage updating technical method based on spatial automatic matching. The method comprises large-scale data analysis, variable information extraction, data editing, quality check and update result storage; the large-scale data analysis comprises integrity analysis, currency analysis and update state mark analysis; the variable information extraction comprises update state marking, trans-scale model matching and artificial contrast analysis; the data editing comprises graph variation, coordinated processing and supplementary acquisition; and the quality check comprises comprehensive choice of check factors, correct geometrical expression normalization, filling accuracy and integrity of check attributes, mutual spatial relationship rationality between check factors and accuracy of check factor update state marks.