731 results about "Feature Dataset" patented technology

Time lapse, characteristic of aging, is a complex process that affects the reliability and security of biometric face recognition systems. Systems and methods use deep learning, in general, and convolutional neural networks (CNN), in particular, for automatic rather than hand-crafted feature extraction for robust face recognition across time lapse. A CNN architecture using the VGG-Face deep (neural network) learning produces highly discriminative and interoperable features that are robust to aging variations even across a mix of biometric datasets. The features extracted show high inter-class and low intra-class variability leading to low generalization errors on aging datasets using ensembles of subspace discriminant classifiers.

A biometrics authentication system utilizes information of the palm of the hand of a body to perform individual authentication. A processing unit obtains an image of the palm of the hand of the same body a plurality of times from an image capture unit, judges the degrees of similarity among the characteristic data sets of the plurality of images of the palm of the hand, and registers a plurality of characteristic data sets with a high degree of similarity in a storage unit. And the shape of the hand in the image is checked from the outlines in the image of the palm of the hand, so it is possible to rapidly judge whether image capture has been successful and extract characteristic data, and registration processing can be executed in a short length of time.

The invention discloses an Internet of things terminal secure access method and system based on edge computing. The method comprises the steps that an edge computing device carries out radio frequencyradiation signal collection on each legal sensing device, thereby obtaining collection results; the edge computing device preprocesses each legal device signal, removes outliers and carries out datanormalization on collected signals of each time; the edge computing device generates features through utilization of the normalized data, extracts feature vectors to generate feature data sets and transmits the feature vector sets to a cloud server; and the cloud server selects a classification algorithm to generate a data model, trains the data model according to the feature data sets and transmits an obtained decision-making model to the edge computing device. According to the method and the system, the data processing and access judgment are carried out at the edge computing side, and the method and the system are applicable to an Internet of things device interconnection scheme with limited resources and have the advantages of low computing complexity and high authentication accuracy.

A locking system for a firearm is disclosed. A lock has a set state and an unset state, and substantial movement of any one or more fire control group components is inhibited with in the set state. A biometric sensor attachable to a grip of the firearm is receptive to an input biometric feature data set corresponding to a physiological feature of a user. A biometric input controller stores biometric feature data sets corresponding to enrolled user identities in a memory, and compares it against the input biometric feature data set to generate a biometric input validation status indicator signal. A proximity sensor attachable to the grip detects possession of the firearm by the user and generates a corresponding grip detection indicator signal. A system controller selectively actuates the lock to the set state and the unset state based upon a received combination and sequence of the biometric input validation status indicator signal and the grip detection indicator signal.

A method of using an addressable array of biopolymers on a substrate includes receiving the addressable array and an associated machine readable identifier carried on an array substrate or array housing. The array is exposed to a sample and read, and the identifier machine read as an identifier signal. Biological function data for one or more of the biopolymers is retrieved from a memory based on the identifier signal. Other methods in which first and updated sets of feature characteristic data may readily be provided to array users, and methods of generating arrays are also provided, as are apparatus and computer program products which can execute a method for generating or using arrays.

A system and method for analyzing and visualizing a local feature of interest includes access of a clinical image dataset comprising clinical image data acquired from a patient, identification of a region of interest (ROI) from the clinical image dataset, and extraction of at least one local feature corresponding to the ROI. The system and method also include definition of a local feature dataset comprising data representing at least one local feature, access of a pre-computed reference dataset comprising image data representing an expected value of the at least one identified derived characteristic of interest, and comparison of the characteristic dataset to the pre-computed reference dataset. Further, the system and method include calculation of at least one deviation metric from the comparison and output of a visualization of the at least one deviation metric.

A fault diagnosis method for power electronic circuits based on optimizing a deep belief network, including steps. (1) Use RT-LAB hardware-in-the-loop simulator to set up fault experiments and collect DC-link output voltage signals in different fault types. (2) Use empirical mode decomposition to extract the intrinsic function components of the output voltage signal and its envelope spectrum and calculate various statistical features to construct the original fault feature data set. (3) Based on the feature selection method of extreme learning machine, remove the redundancy and interference features, as fault sensitive feature data set. (4) Divide the fault sensitive feature set into training samples and test samples, and primitively determine the structure of the deep belief network. (5) Use the crow search algorithm to optimize the deep belief network. (6) Obtain the fault diagnosis result.

The invention discloses an aircraft actuator fault detection and diagnosis method based on a depth random forest algorithm. The method includes: firstly, summarizing the fault mode of an aircraft actuator; establishing an RBF neural network, and collecting the input and output data of the aircraft actuator under the normal working condition to serve as training data, and training the parameters inthe neural network model to obtain analysis redundancy of the monitored actuator; analyzing the residual data of the output signals by collecting the output of the actual actuator and the neural network model, and after the feature extraction, inputting the feature data set into a trained depth random forest multi-classifier for fault mode recognition. According to the invention, the complex nonlinear input and output relation of the aircraft actuator can be accurately simulated by the neural network, the fault mode is accurately recognized by a depth random forest strong classifier, and moreover, the method has the advantages of parallel calculation and high running speed, can be integrated into a flight management computer of an aircraft, realizes the online real-time monitoring, and the accuracy and the efficiency of fault diagnosis of the aircraft actuator are improved.

The invention discloses a pathology identification method for a routine scan CT image of the liver based on the random forests. The method comprises that image gray-level texture characteristic is extracted from a pathologic area of the routine scan CT image of the liver and serves as image characteristic vector expression, the random forests is used to select characteristics from the image characteristic vector of the pathologic area of the routine scan CT image of the liver to form a most effective characteristic combination, a most effective characteristic data set is trained and learned, the identification capability of a decision tree of random forests is balanced and optimized, and a final pathology identification model is obtained.

The invention discloses a method for recognizing an early fault of a bearing based on a long and short-term memory recurrent neural network, which comprises the steps of collecting full life vibrationsignals of the bearing, and then extracting common time domain characteristics; constructing waveform entropy characteristics, and verifying the validity of the waveform entropy according to a squaredemodulation method; building a characteristic data set by using the time domain characteristics and the entropy characteristics, and selecting a normal data set and a deep fault data set; taking thenormal data set and the deep fault data set to serve as training samples to train the LSTM (Long and Short-Term Memory) recurrent neural network; and performing time domain characteristic and entropycharacteristic extraction on an online bearing vibration signal, and then inputting the online bearing vibration signal into the trained LSTM recurrent neural network so as to recognize the fault occurrence time. The traditional characteristics and the entropy characteristics of the vibration signals are combined, so that the current state of the bearing is accurately reflected under the condition of ensuring the physical meaning of the vibration characteristics. The adopted recurrent neural network can effectively apply the degraded historical data so as to perform effective recognition on the fault occurrence time of the bearing.

The invention discloses an application scenario identification method for terminal equipment. The method comprises the following steps of compiling and analyzing an application program running on the terminal equipment to acquire the characteristic data of the application program; determining application scenario information corresponding to the characteristic data of the application program from a scenario characteristic dataset according to the characteristic data of the application program, wherein the scenario characteristic dataset comprises corresponding relationships between various kinds of application scenario information and the characteristic data of various kinds of application programs, and the application scenario information corresponding to the characteristic data of the application program is used for representing an application scenario where the terminal equipment is currently used. The embodiment of the invention also provides a power consumption management method. According to the methods, the characteristic data in the application program has higher uniqueness for the description of the corresponding application scenario, so that the application scenario information corresponding to the characteristic data of the application program is more accurate, and the application scenario of the terminal equipment can be more accurately identified.

A user identification based website real-time / non-real-time marketing investment method and system are provided. The method comprises: A, acquiring log information of internet surfing of a user, recorded by a website WEB server within a preset duration, and deleting invalid access data, which influences classification of user behaviors, in the acquired log information of the internet surfing of the user; B, extracting indicators for representing user behavior features according to business flow features, and extracting features from the log information of the internet surfing of the user by the indicators through taking a once browsed session, namely a session, of a single IP user as a unit object; and C, according to a feature data set of the session browsed by the user through the internet surfing, extracted in the step B, obtaining classification rule and model of user identification. Automatic triggering rules of various online marketing modes are written and a server implanted with the automatic triggering rules is used for scanning the data regularly.

The invention discloses a relation extraction method in combination with clause-level remote supervision and semi-supervised ensemble learning. The method is specifically implemented by the following steps of 1, aligning a relation triple in a knowledge base to a corpus library through remote supervision, and establishing a relation instance set; 2, removing noise data in the relation instance set by using syntactic analysis-based clause identification; 3, extracting morphological features of relation instances, converting the morphological features into distributed representation vectors, and establishing a feature data set; and 4, selecting all positive example data and a small part of negative example data in the feature data set to form a labeled data set, forming an unlabelled data set by the rest of negative example data after label removal, and training a relation classifier by using a semi-supervised ensemble learning algorithm. According to the method, the relation extraction is carried out in combination with the clause identification, the remote supervision and the semi-supervised ensemble learning; and the method has wide application prospects in the fields of automatic question-answering system establishment, massive information processing, knowledge base automatic establishment, search engines, specific text mining and the like.

The invention discloses a text classification technology and decision-making tree based complaint tendency judgment method. The method includes: acquiring user information from a database management system and establishing a user history call information table accordingly, wherein the user information includes user profile information and customer service order information; determining a time window for acquiring the user information; determining a prejudgment period and a feature data set used for prejudgment; determining a group range for prejudgment. With the method, complaint tendency degrees of users can be judged accurately when the users call, reference is provided for designing a strategy library and adopting different pacification and guidance strategies, and customer request information extracted from structural data is fully utilized; meanwhile, unstructured text data in call content in work orders is analyzed systematically, future complaints of the users are precast in advance according to the historical call information, complaint risk of the users can be reduced, and good social image of an electric power company is established.

The invention discloses a mechanical fault diagnosis method for an unsupervised deep learning network. The method comprises the steps of: (1) mounting corresponding sensors near the part such as bearing of a mechanical device to collect mechanical vibration signals; (2) converting the collected vibration signals into a mixed domain fault feature data set, and dividing the data set into a testing and a training sample feature subset; (3) inputting the training sample feature subset into the constructed unsupervised deep learning network (UDLN) model for learning and training, wherein the UDLN model is composed of two improved sparse filtering (L12SF) unsupervised feature extraction layers and one weighted Euclidean distance similarity affine (WE) clustering layer; (4) inputting the test sample into a trained diagnosis model to realize full-range unsupervised feature learning and fault clustering; and (5) calculating the recognition rate of the test sample clustering division according to the membership degree thereof to realize fault identification and diagnosis. According to the mechanical fault diagnosis method for the unsupervised deep learning network, the invention is simple and feasible, and can perform adaptive unsupervised fault diagnosis on various faults of mechanical equipment.

The application relates to a network traffic classification method and system based on deep learning, and electronic equipment. The method comprises the following steps: step a, capturing network traffic sample data; step b, extracting a global feature data set of the network traffic sample data through a deep learning classification algorithm; and step c, constructing a random forest classification model according to the global feature data set, and outputting a network traffic classification result through the random forest classification mode. The random forest classification model is trained by utilizing the extracted global feature, the result shows a stable classification performance, and the ultra-high-dimension traffic data can be processed, and the feature selection is avoided. Compared with the prior art, the high-precision and high-performance of the network traffic classification can be effectively guaranteed; and meanwhile, the classification efficiency can be improved, the training time is shortened, and the computation overhead is reduced.

The invention discloses a deep learning-based transformer patrol image intelligent identification and fault detection system. The system comprises an image storage and calibration module, a deep learning module and a fault detection module located in a storage layer, a modeling layer and an application layer in sequence; the image storage and calibration module obtains massive patrol images from apower grid information system, performs distributed storage, performs calibration on parts in the patrol images and parts in an image of a transformer in a normal / fault state, and forms a part feature set, a normal state feature set and a fault state feature set; a feature data set is transmitted into the deep learning module; the deep learning module constructs and trains a convolutional neuralnetwork to obtain a training result; and the fault detection module performs part identification, fault identification and conclusion pushing in sequence by utilizing the training result. Compared with a manual identification mode, the identification efficiency and accuracy of the transformer patrol images are greatly improved.

A system and method for analyzing and visualizing local clinical features includes identification of a first region of interest (ROI) from a medical image dataset acquired from a patient and extraction of a feature dataset representing a feature of interest specific to the ROI. The system also includes identification of a second ROI from the medical image dataset, extraction of a reference dataset comprising reference data representing an expected behavior of the feature of interest, comparison of the feature dataset to the reference dataset, generation of a deviation metric representing a deviation of the feature of interest based on the comparison, and creation of a visual representation of the deviation metric.

The invention provides a machine learning-based domain generation algorithm (DGA) domain name rapid determining method and device, relating to the technical field of network security. The machine learning-based DGA domain name rapid determining method comprises the steps of building a training set comprising multiple DGA domain names and normal domain names; extracting a domain name characteristicof each domain name in the training set; normalizing the domain name characteristics so as to obtain a characteristic data set; and building a domain name classifier model according to the characteristic data set. According to the method and device, richer, more representative domain name characteristics are extracted by researching the domain names; the characteristic data is normalized so as tospeed up training and tests to improve calculation efficiency; and finally, the characteristic data set is trained by using a machine learning algorithm so as to obtain the domain name classifier model, so that determining accuracy is improved, further, generalization capability is improved.

The invention discloses a network intrusion detection method. The network intrusion detection method includes: searching network data to construct a test network data set; performing feature extraction on the test network data set by utilizing a kernel principal component analysis method; constructing a training data set, putting the training data set into a support vector machine classifier for training; obtaining feature datasets, obtaining an optimal feature subset from the feature data set by using a genetic algorithm; utilizing a firefly swarm optimization algorithm to obtain the overalllocal optimal feature subset and the optimal support vector machine parameters from the optimal feature subset, processing the training data set according to the overall local optimal feature subset,and inputting the training data set into a support vector machine classifier for classification modeling to obtain a network intrusion detection model. According to the method, the simplicity and convenience of the algorithm are improved, abnormal data can be more effectively found from samples, the detection accuracy of network intrusion is effectively improved, the missing report rate and the false report rate are reduced, and the overall performance of network intrusion detection is improved.

The embodiment of the invention, which relates to the field of the internet technology, provides a method and apparatus for predicting a re-purchasing probability of a user and electronic equipment. The method for predicting a re-purchasing probability of a user comprises: carrying out learning on a training sample set to obtain a prediction model for a re-purchasing probability of a user; obtaining a feature data set of a to-be-predicted user; and using the feature data set of the to-be-predicted user as an input of the prediction mode and carrying out processing to obtain a re-purchasing probability prediction value of the to-be-predicted user by the prediction model. Therefore, on the basis of multi-feature-dimension data of a user, the re-purchasing probability prediction value of the to-be-predicted user can be obtained automatically by the prediction model, so that the prediction accuracy is improved and the prediction efficiency is enhanced.

The invention discloses a method and a device for locating suggested users. The method comprises the following steps: collecting user transaction data and user video viewing data from a video system; extracting an appointed characteristic data set of training users and an appointed characteristic data set of test users from the collected user transaction data and user video viewing data; training various appointed characteristics in the appointed characteristic data set of the training users according to a training algorithm, so as to obtain weighted values of various appointed characteristics; determining paying probability prediction data of various test users according to the appointed characteristic data set of the test users and the trained weighted values of the appointed characteristics; and locating the suggested users according to the paying probability prediction data of the test users. According to the method and the device, the users with the paying tendency can be relatively accurately picked out from the users in a test set, so that the suggested users can be relatively accurately located.

The technical proposal of the invention comprises a GRNN-based identification method. The method is used for realizing the following steps: an electrocardiogram signal sample data set including a plurality of periodic beats of a plurality of users is acquired, and wavelet transform is adopted to remove noise in the electrocardiogram signal; the denoised signals are segmented to construct the morphological features of ECG signals, and the training set beat feature databases and the test set beat feature databases are constructed respectively; singular value decomposition is used to remove redundant features from ECG signals; dimension reduction of the heart beat feature data set is carried out by linear discriminant analysis; the generalized regression neural network classifier is trained and the identity information of the individual is output according to the principle of multi-beat voting. The method has the advantages that the redundant information in the electrocardiogram signal iseffectively removed, the accuracy rate of the subsequent identification is improved, and the speed of the identification is greatly improved due to the reduction of the heart beat characteristic dimension by the LDA and the use of the GRNN neural network as a classifier.

The invention relates to the technical field of copyright image protection, and provides a decentralized image copyright protection system and method with infringement real-time detection, and the method comprises the following steps of (1), enabling a registered user to upload an image through a client, and enabling an SIFT image feature extraction algorithm in the system to carry out the featuredata collection on the uploaded image; (2) extracting the invariant characteristics, such as visual angle, translation, affine, brightness, rotation, etc., of an existing copyright image by a user through an SIFT algorithm to form a local feature vector set of the image, and taking the local feature vector set as a registered image feature data set; (3) carrying out distributed storage on the feature data sets in the steps (1) and (2) by utilizing an IPFS (Internet Protocol File System), and constructing a network transmission protocol for persistent and distributed storage and file sharing;and (4) performing comparative analysis and calculation, and judging that the image does not accord with the condition of forming a new work. The system and the method provided by the invention have the characteristics of being tamper-proofing, realizing the high safety, permanent preservation and decentralization, being able to trace, being low in cost, etc.

The invention provides a method for mining data of a novel urban electric vehicle charging station system. The method comprises the following steps of establishing a data warehouse, and collecting and recording the data of the charging station system and an electric vehicle user in real time; classifying, pre-processing and discretizing the collected and recorded data, and extracting a required feature data set; evaluating the element membership of the feature set by a fuzzy mathematic and probability statistics theory, and parameterizing; calculating the parameterized set, and obtaining a relation matrix of the user charging demand, the charging area and the charging time; extracting and evaluating the results, and feeding and outputting the evaluation conclusion to a management center of the charging station system and a user mobile terminal, so as to assist to make a decision. The method has the advantages that the benefits of the charging station planning and the movable electric vehicle scheduling are maximized, and the charging requirement of the user is met.

The invention discloses a power electronic circuit fault diagnosis method based on an extremely random forest and a stacked sparse auto-encoding algorithm. The method comprises the following steps of signal acquisition and feature extraction; fault feature dimensionality reduction preprocessing, using an extreme tree algorithm ET to calculate importance scores of all features in an original feature data set, sorting the importance scores in descending order, determining the proportion to be removed, and obtaining a new feature set after removing the proportion; fault feature further extraction, using a stacked sparse auto-encoding SSAE algorithm to cascade multiple sparse self-encoders, and performing layer-by-layer feature extraction to obtain a hidden layer feature of a last sparse self-encoder as a fault sample; performing classification training, inputting fault samples in a training set and a test set into a classifier for training, and obtaining a trained classifier; and pattern recognition, using the trained classifier to classify and recognize faults of a power electronic circuit to be diagnosed, and locating the faults.

The embodiment of the invention discloses an antenna switching method of a user terminal and a user terminal. The method is characterized in that a user terminal is configured with a first antenna and a second antenna; and the first antenna is used for receiving a signal and transmitting a signal in free space and the second antenna is used for receiving a signal and transmitting a signal when the antenna approaches a human body. The user terminal uses the first antenna as the default. The method comprises: pass-by track data of movement of a user terminal to a position at a target distance to a human body are detected; whether the target distance is smaller than or equal to a preset distance is determined; if so, feature data matching the pass-by track data exists in a pre-stored feature data set is determined; and if so, the first antenna is switched to the second antenna. With the method, the sensitivity of signal receiving by the antenna can be improved.

The invention relates to the technical field of wireless communication, in particular to a method and device for building a video traffic prediction model. The method comprises the steps that a historical video traffic data set and at least one characteristic data set which are generated on a preset geographic area in a preset time are acquired; each piece of historical video traffic data includedin the acquired historical video traffic data set is associated with characteristic data in each characteristic data set to obtain a data set for building the video traffic prediction model; characteristic parameter screening is conducted on the data set for building the video traffic prediction model by adopting a preset characteristic selection algorithm combination to determine at least one characteristic parameter related to the video traffic; and model training is conducted by means of the data set for building the video traffic prediction model by adopting a preset model training algorithm to obtain a mapping relation between the at least one characteristic parameter related to the video traffic and the video traffic, and the video traffic prediction accuracy is improved.