725 results about "Feature screening" patented technology

The invention discloses a remote obstacle detection method based on laser radar multi-frame point cloud fusion. A local coordinate system and a world coordinate system are established, an extraction feature point of each laser point is calculated on an annular scanning line of the laser radar according to the original point cloud data under the local coordinate system, and the global pose of the current position relative to the initial position and the de-distortion point cloud in the world coordinate system are obtained through inter-frame feature point matching and map feature point matching; the de-distortion point clouds of the current frame and the previous frame are fused to obtain more compact de-distortion point cloud data, which is unified to the local coordinate system, then projection is performed on two-dimensional grids, and an obstacle is screened according to the height change features of each two-dimensional grid. According to the method in the invention, the problem that the detection rate of the remote barrier caused by sparse laser point clouds is low is solved, the remote barriers can be effectively detected, the error detection rate and the leak detection rateare low, and the system cost can be greatly reduced.

The invention belongs to the technical field of computer-aided diagnosis, and specifically relates to a brain glioma molecular marker nondestructive prediction method and a prediction system based on radiomics. The method comprises the following steps: adopting a three-dimensional magnetic resonance image automatic segmentation method based on a convolution neural network; registering a tumor obtained from segmentation to a standard brain atlas, and acquiring 116 position features of tumor distribution; getting 21 gray features, 15 shape features and 39 texture features through calculation; carrying out three-dimensional wavelet decomposition on the gray features and the texture features to get 480 wavelet features of eight sub-bands; acquiring 671 high-throughput features from the three-dimensional T2-Flair magnetic resonance image of each case; using a feature screening strategy combining p-value screening and a genetic algorithm to get 110 features highly associated with IDH1; and using a support vector machine and an AdaBoost classifier to get a classification of which the IDH1 prediction accuracy is 80%. As a novel method of radiomics, the method provides a nondestructive prediction scheme of important molecular markers for clinical diagnosis of gliomas.

The invention discloses a transfer learning and feature fusion-based ultrasonic thyroid nodule benign and malignant classification method. The method comprises the following steps of firstly preprocessing an ultrasonic image and zooming the ultrasonic image to a uniform size; extracting traditional low-level features of the ultrasonic image; extracting high-level semantic features of the ultrasonic image by using a model obtained in a natural image through deep neural network training through a transfer learning method; fusing the low-level features with the high-level features; carrying out feature screening by utilizing distinction degree of benign and malignant thyroid nodules so as to obtain a final feature vector which is used for training a support vector machine classifier; and carrying out final thyroid nodule benign and malignant classification. According to the method disclosed by the invention, the low-level features and the high-level features are fused, and salient feature screening is carried out, so that the problem that the ability of single features for describing thyroid nodule features on the level of semantic meaning is insufficient is solved, and the classification precision is effectively improved; and through importing the transfer learning, the problems that the medical sample images are few and the deep features can not be obtained by direct training are solved.

The invention relates to a multi-well complex lithology intelligent identification method and system based on logging data. The method comprises the following steps: firstly, determining a target logging data file and carrying out format conversion and normalization preprocessing; performing feature screening and / or feature combination expansion on the logging curve data according to the known lithology of the key coring well in the whole region in the coring well section to obtain the logging curve data sensitive to the lithology; performing labeling calibration on the logging curve data sensitive to the lithology response to form a sample database, and forming a to-be-tested database by using the logging curve data which are not labeled in the whole region; utilizing data of the sample databaseto carry out machine learning training by combining a plurality of machine learning algorithms, and then automatically establishing a plurality of lithology recognition models. An optimal modelis selected through a classification performance evaluation rule, the optimal model is used for carrying out lithology prediction on the data in the database to be tested so as to realize intelligentidentification of multi-well complex lithology in the whole region, and the method is efficient, convenient, applicable in the whole region and very accurate in automatic intelligent prediction result.

The invention provides an unmanned aerial vehicle-based intelligent identification method of electric power facilities. The method includes: carrying out lidar scanning on a to-be-patrolled power lineto obtain three-dimensional point cloud data of the to-be-patrolled power line, and carrying out key feature screening to extract and obtain key feature point information; combining position data inthe three-dimensional point cloud data to determine key point patrolling positions and key point patrolling targets of a current patrolling process on the basis of the key feature point information togenerate flight path planning data of the current patrolling process; collecting image data of the key point patrolling positions and the key point patrolling targets in a process of flying accordingto the flight path planning data; and carrying out content identification processing on the image data through a special image processing identification module to troubleshoot fault conditions existing in the to-be-patrolled power line. According to the method, optimal setting can be carried out for both positions and angles of shooting of the key point targets through refining platform handlingand controlling, and patrolling efficiency and accuracy of image identification are greatly improved.

The invention relates to a vehicle type identification method based on a support vector machine and used for an earth inductor. The vehicle type identification method includes the following steps: vehicle type waveform data which require to be identified are collected by the earth inductor; a plurality of numeralization features are extracted from waveforms, effective data are screened out, and the features are normalized; multilayer feature selection is performed according to the extracted features, and an optimal feature combination is picked out; a vehicle type classification algorithm based on the clustering support vector machine is built, and parameters in a classification function are optimized by adopting a particle swarm optimization algorithm; a binary tree classification mode is built, classifiers on all classification nodes are trained, and a complete classification decision tree is built; and earth induction waveforms of a vehicle type to be identified are input to obtain identification results of the vehicle type. The vehicle type identification method builds a waveform feature extraction and selection mode, simultaneously adopts the classification algorithm based on the support vector machine and the particle swarm optimization algorithm, greatly improves machine learning efficiency, and enables a machine to identify vehicle types rapidly and accurately.

The present invention provides a portable multi-channel depression tendency evaluation system based on emotional stimulation task. In a mood stimulation task, Electroencephalogram (EEG), electrodermatogram (ECG), electrocardiogram (ECG), eye movement (EOM), speech and image information were collected by portable equipment. Supervised learning and unsupervised learning algorithms were used to extract features from multi-channel physiological signals and behavioral data, to screen features, to train and validate models, to integrate the results of multi-channel analysis, to calculate depressionpropensity index (DPI) and to evaluate depression propensity.

The invention discloses a pointer instrument positioning and identifying method based on machine vision. The position of an instrument panel in an instrument cabinet image is accurately positioned andan instrument reading is automatically recognized. The method comprises the steps of (1) acquiring the instrument cabinet image; (2) carrying out segmentation processing on the instrument cabinet image to obtain an instrument panel image; (3) preprocessing the instrument panel image; (4) acquiring all edge contours in the instrument panel image by using a Canny edge detection method; (5) findingall linear contours in the obtained edge contours by using Hough transform; (6) screening out the linear contours where a pointer in the instrument panel is located through length characteristics of the linear contours; (7) calculating an average value theta of an inclination angle between the two linear contours where the pointer is located; and (8) calculating out a pointer reading according toa linear relation between the linear inclination angle and an instrument scale. By the adoption of a positioning and detecting technology, the method has the characteristics of accurate positioning and reading, good anti-noise performance and efficient and rapid image processing.

The invention discloses an underground water grading and zoning evaluation method in an agricultural activity area. The method comprises the following steps of: establishing a set of index system for evaluation, arranging a sampling point in a spatial discretization mode, acquiring an underground water sample, screening the indexes according to the characteristics of primary indexes, calculating the weight of each index and the degree of membership of each sampling point index, comprehensively evaluating the regional underground water pollution risk, and zoning and grading the underground water risk. Aiming at the pollution characteristics of the agricultural activity area in China, the method has high adaptability and popularization, and the underground water risk grade and risk distribution in the agricultural activity area can be scientifically and reasonably evaluated.

The embodiment of the invention discloses an online model training method. The method comprises the steps of obtaining a training sample from streaming data, determining an objective function of the model according to the training sample, historical model parameters and non-convex regular terms, determining current model parameters enabling the objective function to be minimum, and updating the model according to the current model parameters. In the online training process, since the non-convex regular term is adopted to replace the L1 regular term for feature screening, the penalty deviationcan be reduced, effective features can be screened out, the sparsity is guaranteed, and the generalization performance of the model is improved. The invention further provides an information pushing method. The method comprises: obtaining user feature data and content feature data, based on the pushing model obtained by the online training model method, determining the probability that a target user is interested in target information according to the user feature data, the content feature data and the pushing model, and determining whether pushing is conducted or not according to the probability that the target user is interested in. The invention further provides an online model training device and an information pushing device.

The invention provides a face-based video labeling method and a face-based video retrieving method. The labeling method comprises the following steps that: a face image and corresponding face features in a video to be labels are extracted, and the face features and attribute information of the face features are merged to obtain face metadata; all of the face features obtained in a video file to be labeled are subjected to automatic feature clustering, then, feature types are subjected to feature screening and feature recalling, and a feature type set P<2> and a non-classified feature set Q<2> are obtained; for each element in the P<2>, the affiliated feature type recommendation is carried out, and manual confirmation is carried out; for each element in the Q<2>, the non-classified face feature recommendation is carried out, and the manual confirmation is carried out; the feature types and the non-classified features are subjected to name labeling; and the labeled feature type and non-classified face feature information are used for forming a video labeling file of the video file. In the video labeling process, the program automatic recommendation is combined with the manual confirmation, so that the accuracy of the result is ensured, and the efficiency is also improved.

The invention discloses a baseline construction and unknown abnormal behavior detection method oriented to a system log. The method comprises the following steps: I, user operation log quantification, namely, user behavior quantification; II, user behavior feature screening; III, abnormal behavior detection; and IV, abnormal behavior description. The baseline construction and unknown abnormal behavior detection method oriented to the system log disclosed by the invention comprises the following specific steps: firstly, quantifying a user operation log to generate a behavior feature vector; secondly, performing abnormity marking on samples to construct a reference sample set, and evaluating and screening sub-features at the same time to construct a new feature vector; and lastly, performing active clustering analysis on the user behavior feature vector, constructing a behavior baseline, and detecting abnormal behaviors. Abnormal points are eliminated in a detection process, so that an abnormal point effect can be effectively avoided, and typical user behaviors and abnormal behaviors can be detected accurately.

The invention discloses a text classification character screening method based on character distribution information. The method is used for resolving the technical problems that an existing text classification character screening method is poor in accuracy. The technical scheme includes conducting preprocessing for each document of a document set firstly; enabling the whole document collection to be presented as a vector space modal (VSM); constructing a character dictionary; counting document frequency DF (t, Cj), comprising the character t, of each classification Ci; calculating a normalized tf*idf value of each classification Ci, and then calculating the dispersion D Intra and average inter-classification dispersion D Inter Avg of the character in each classification Ci; calculating the weight wi (t) of each character tk in each classification Ci of a text character space; and enabling all the characters to be arranged in a descending order mode according to the weight of all the characters in the whole document set, and preferentially keeping the characters having front orders during character screening. On the basis of a character distribution system, the method enables the character distribution system to be applied to the character screen process, and improves text classification efficiency and accuracy.

The invention provides an automatic sleep stage classification method based on dual character filtering. The stage classification method comprises steps of firstly extracting a two-guidance sleep electroencephalogram signal and an one-guidance horizontal electrooculogram signal; performing filtering for the original electroencephalogram signals and electrooculogram signal; extracting multiple characters from the filtered electroencephalogram signals and the filtered electrooculogram signal; selecting the optimal character subsets by use of a dual character filtering method combining the Fisher score method and the sequential forward selection method. Via the dual characteristic filtering method, character dimension is greatly reduced and redundancy among the characters is reduced. At last, a support vector machine classifier is used for identifying the optimal characters, so automatic stage classification of sleep is finished. According to the invention, objectivity, precision and convenience of automatic sleep stage classification can be well increased; the automatic sleep stage classification method is characterized by high precision, low calculation complexity, simple operation and easy popularization; and considerable social and economic benefit can be gained.

The invention relates to the technical field of CNN and image classification and recognition, in particular relates to a breast cancer histopathologic grading method based on CNN and image histological feature fusion. The invention provides a method for judging the histopathologic grade of the breast cancer of the molybdenum target image by constructing a feature-fused CNN model, the features withhigh correlation with the histopathologic grade of the breast cancer are selected based on the grayscale features, the texture features and the wavelet features extracted from molybdenum target tumorregion through LASSO logistic regression model for feature selection, and then the high-level semantic features extracted by CNN and the selected image histological features are fused in the newly added full-connection layer of the network, and the fused CNN model is used for recognizing the histopathologic grade of the breast cancer. The breast cancer histopathologic grade of the patient can bedirectly analyzed and judged by the mammography target image scanned by the patient, thereby ensuring the discrimination accuracy and further shortening the discrimination time.

The invention discloses a multi-factor short-time traffic flow prediction method. The method comprises the steps that 1, distances between a specific detector and meteorological stations are calculated, the nearest meteorological station is screened out, and weather data of the meteorological station serves as weather data of the detector; 2, historical traffic flow data and historical weather data of the detector are preprocessed respectively and merged according to the time; 3, a plurality of features are constructed, and feature screening is carried out on the basis of LightGBM; 4, modelingis carried out on time sequence features and periodic features of the traffic flow data through LSTM; 5, the time sequence features and the periodic features of the traffic flow are fused with the features screened out in step 3 through a fully-connected network in a neural network; 6, the model is trained, and the short-time traffic flow is predicted. The method overcomes the defect that existing methods cannot fully use known data, can be used for deeper mining and analysis of the traffic flow data, and is high in prediction accuracy and good in robustness.

The invention discloses a vertical mill operation regulation and control system based on data mining and a method thereof. The method is characterized by using an integration characteristic screening method to carry out mining analysis on work condition data and acquiring a vertical mill health state assessment index; carrying out cluster mining analysis on a work condition state, acquiring a characteristic of each condition cluster, and acquiring each state distribution condition in historical work conditions; defining operation state types in the historical work conditions and acquiring a stable mode condition database; using an ARIMA algorithm to carry out training model on a determined characteristic value in a vertical mill health state characteristic acquisition module and predict a parameter change trend, and using a predicted value to assist state identification; combining the predicted value given by an ARIMA model, determining a vertical mill operation state; when the state is determined to be abnormal, reading a work condition record in the stable work condition mode database, acquiring a recommended regulation and control target value and regulating and controlling a controllable parameter at the moment. In the invention, vertical-mill qualitative and quantitative regulation and control suggestions can be accurately provided and a mill machine can work stably for a long time.

The invention discloses a multi-class entity recognition model training method, an entity recognition method, a server and a terminal. The multi-class entity recognition model training method comprises the steps: carrying out the entity and entity class labeling of corpus information, and obtaining the target annotated corpus information comprising an entity and an entity class label; performing multi-dimensional feature analysis processing on the corpus information in the target annotated corpus information to obtain multi-dimensional information of the target annotated corpus information; performing multi-class entity recognition training on the preset deep learning model based on the multi-dimensional information and entities and entity class tags in the target annotation corpus information to obtain a multi-class entity recognition model, wherein the preset deep learning model comprises a feature input conversion layer, a semantic sequence representation layer, an entity feature screening layer and a class entity output layer. By utilizing the technical scheme provided by the invention, the entities and the entity categories in the corpus information can be quickly and accurately identified, and multi-category entity identification is realized.

The invention discloses a data mining-based slag grinding system health state identification system and method. According to the data mining-based slag grinding system health state identification system and method of the invention, a comprehensive feature screening method is utilized to carry out mining analysis on working condition data so as to obtain key parameters which affect the stability of a vertical mill, and the key parameters are adopted as indicators of vertical mill health state assessment; based on the identified indicators of the vertical mill health state assessment, clustering and mining analysis is performed on working condition states, so that the features of various of working condition clusters are obtained; the state distribution conditions of historical working conditions are obtained, and operating state categories in the historical working conditions are defined; and the ARIMA algorithm is adopted to train a model for determined feature values in a vertical mill health state feature acquisition module, and the change trend of parameters is predicted, and a prediction value is utilized to assist state identification. The data mining-based slag grinding system health state identification system and method of the invention have the advantages of high recognition precision, great generalization ability and high performance, and is suitable for the health state identification and diagnosis of a slag grinding system.

The invention provides a gas pipeline leakage identification method based on a convolution neural network. The method comprise the following steps that after a leakage sound signal and a background sound signal of a typical leakage type are collected, framing processing and short-time Fourier transform are carried out to obtain a time-frequency diagram representing the original leakage sound signal; then a convolution neural network classification model aiming at the leakage sound signal is built, a traditional square convolution kernel is changed into a specific strip-shaped rectangular convolution kernel, so that the line spectrum characteristics in the time-frequency diagram are better extracted; and the time-frequency diagram of the leakage sound and the time-frequency diagram of the background sound are mixed and sent to the built convolution neural network for training, K-fold cross validation is adopted for training, and a network model superparameter is optimized, so that the optimal model superparameter is selected and the robustness and universality of the model are enhanced. Compared with the pipeline leakage identification method in the prior art, the method has the advantages that the identification rate is further improved, and the problem of feature screening which is most difficult to process in the prior art can be effectively solved.

The invention discloses an aero-engine service life prediction method based on label-free, unbalanced and initial value uncertain data, which comprises the following steps of firstly, carrying out feature screening on an engine training data set by utilizing a correlation index and a trend index; then, obtaining a health state label through quantum fuzzy clustering, training a multivariable deep forest classifier, and obtaining an aero-engine health assessment model; meanwhile, training a long-short cycle memory neural network (LSTM) time sequence prediction model by utilizing the engine training data set; and finally, obtaining the maintenance time and the final residual service life (RUL) of the engine at different health stages by using the engine test data set according to the trainedhealth assessment model and the time sequence prediction model. According to the method, the defects of no label, imbalance and initial value uncertainty of observation data are overcome, and a technical reference is provided for maintenance decisions of the aero-engine in different subsequent health stages.

The invention provides an identity authentication method and equipment. The method comprises the steps that an identity authentication request of a user under authentication is received; the identity authentication request is responded and the face image of the user under authentication is automatically acquired; the face features are extracted from the acquired face image of the user under authentication; and existence of face feature templates matched with the face features extracted from the acquired face image of the user under authentication is determined in a preregistered face feature template library, representative feature screening is performed according to the face features extracted from the acquired face image of the user under authentication and the face feature templates if the judgment result is yes, and the face feature template library is updated by using the selected features through screening.

The invention discloses a hypergraph and random forest (HG-RF)-based intrusion detection method, and belongs to the technical field of network intrusion detection. The method comprises the following steps: (1) carrying out data preprocessing; (2) carrying out feature screening to obtain a new feature subset; (3) calculating a Fisher score of each feature in each category, and carrying out descending-order arrangement; and (4) inputting a test sample set to a weighted random forest classifier to obtain final intrusion detection results of test samples. The method is based on a method of featurepreference, firstly carries out dimensional reduction processing on data, then carries out classification, improves intrusion detection speed and an accuracy rate of the classifier, and reduces a detection false-reporting rate.

A detection method based on image fusion comprises the steps of: irradiating a to-be-detected object with a plurality of illuminating modules in different directions in order to obtain a plurality frames of to-be-detected images; performing image fusion treatment to the to-be-detected images; and performing feature screening to an obtained a texture image or a height image. The method not only canreduce influence due to poor image collection quality under low-contrast but also can provide complete surface feature information, which has advantages of fusing the surface feature information in each frame of the to-be-detected images into the texture image or height image and recognizing and screening the surface defect features from the distinguishing characteristics of the images. An apparatus, with combination of the surface defect detection method, can maximumly eliminate influence due to environmental factors during the detection on the to-be-detected object, and can synthesize and obtain the feature image, comprising the surface defect features from a large number of images. The method and apparatus enhance screening effect on the surface defect features and complete the surfacedefect detection operation under the low-contrast condition.

The invention provides a rapid feature matching industrial network DDoS intrusion detection method, relates to the field of industrial network communication and belongs to the aspect of information security. According to the method, a traditional database lookup feature matching model for intrusion detection is replaced through utilization of a deep learning model, thereby finishing designing theintrusion detection system in an industrial network. According to the method, data packet features are extracted; features are screened; a feature matrix is established; the feature matrix is input into the deep learning model for judgment; and the model is trained by a KDD99 data set, and the model has a very high identification rate for a DDoS attack type. According to the method, the defect that the traditional database lookup feature matching speed is low is improved, a system does not need to waste time for waiting for the looking up in a database, a matching rate is greatly improved, thematching precision is high, a false alarm rate is low, the deployment is convenient, the use of system resources is effectively reduced, and a demand of the industrial network for network timelinessis satisfied.

The invention provides a Kawasaki disease classification and prediction method based on medical data modeling. The method comprises the following steps: S1, data sample selection: extracting a valid sample which can be modeled from a sample data set; S2, feature sieving: sieving 19 features conforming to field medical auxiliary diagnosis application from a feature set for constructing sample data for modeling; S3, Kawasaki disease classifying model construction and evaluation: fitting an Xtrain data set on a training set by using a random forest classifying method, recording optimal modeling parameters and weights of all selected features, and classifying and predicting a test set sample according to the classifying model. In the Kawasaki disease classification and prediction method, Kawasaki disease relevant data are analyzed and modelled systemically, and an evaluation method for model prediction is provided, so that the model can be used for performing effective auxiliary diagnosis on a Kawasaki disease of a patient based on Kawasaki disease data, effective prevention and intervention and treatment are performed at the early stage of attacking, and a basis is laid for a best treatment effect.

The invention discloses a Structure sparse multi-kernel learning-based multi-modal data feature screening and classification method. The method comprises the following steps of: extracting data features and normalizing the features; constructing a kernel matrix for each data feature by utilizing a kernel function; grouping all the data features and the corresponding kernel functions according to the data features; training a Structure sparse multi-kernel classification model by utilizing training data with class labels, and optimizing parameters of the model; and classifying test data by using the trained Structure sparse multi-kernel classification model. According to the method, the data feature selection and data modal fusion are modeled in the uniform structure sparse multi-modal classification model, and the structure sparse feature selection and classifier learning expressed on the basis of an optimal kernel are carried out at the same time, so that a multi-modal data feature screening, fusion and classification method is provided.

The invention belongs to the technical field of medical image processing and computer vision, and particularly relates to an MRI image liver fibrosis automatic grading method based on imaging omics analysis. The invention particularly relates to an imaging omics analysis method for grading magnetic resonance dynamic enhanced imaging liver fibrosis of a Gd-EOB-DTPA contrast agent. The method comprises the following steps: establishing an initial data set by taking a Gd-EOB-DTPA dynamic enhanced image of a hepatitis B patient as source data; registering the initial data; carrying out automatic liver segmentation by utilizing transfer learning to obtain an ROI; performing image omics analysis: for the ROI, carrying out image omics feature extraction and feature screening on a plurality of DCEsequences to obtain a feature subset of important features, and carrying out training of a plurality of classifiers to select an optimal classifier; and finally, predicting and evaluating the classification performance in an external test set by using the classifier. According to the method, the accuracy and reliability of automatic liver fibrosis grading based on DCE images can be effectively improved, and meanwhile, the time and energy of clinicians are saved to a great extent.

The invention discloses an industrial control system intrusion detection method based on integrated learning. The method comprises the following steps: acquiring field data of an industrial control system; carrying out message analysis on communication data to obtain a structured sample, selecting an appropriate feature set through feature screening and extraction, inputting the feature set into an integrated learning model consisting of a plurality of machine learning algorithms, and finally judging whether the communication data of a specific industrial control system is normal or abnormal through the learning algorithm. Effective information of communication data of the industrial control system is fully mined by utilizing an intelligent learning algorithm, the intrusion detection accuracy is effectively improved through an integrated learning model fusion method, and the missing report rate is reduced.

The invention discloses a multimode honeypot system and a data analysis method thereof. The system comprises a plurality of single-node honeypot systems, and the single-node honeypot systems are connected with each other to form a network topological structure. The method comprises the steps: obtaining the original data of each single-node honeypot system; carrying out the feature screening of the original data through a PCA (principal component analysis) method, and obtaining the intermediate data; carrying out the clustering analysis and supervised learning of the intermediate data, and obtaining a plurality of clusters, wherein each cluster comprises a plurality of examples; carrying out the association rule extraction of an hostile attack command in each cluster; taking the extracted association rules as the attack features, and storing the attack features in a feature mark set.