86 results about "Synthetic data sets" patented technology

The invention discloses a bearing fault mode diagnosis method and system based on deep learning. The method comprises: a learning rate is adjusted automatically, noises are introduced, and a linear correction unit is introduced as an activation function, and thus a deep belief network is improved based on the activation function; a data set and long- and short-term memory networks are migrated toobtain a synthetic data set, a trained data set is extended based on the synthetic data set, and the improved deep belief network is trained based on training data set training to obtain a bearing fault diagnosis model; and a vibration signal of the bearing is collected and a bearing failure mode is diagnosed based on the bearing vibration signal and the bearing fault diagnosis model. On the basisof combination of a semi-supervised learning and a migration learning algorithm, the diagnostic accuracy is improved while no insufficient data are provided.

Identifying and detecting threats to an enterprise system groups log lines from enterprise data sources and / or from incoming data traffic. The process applies artificial intelligence processing to the statistical outlier in the event of the statistical outliers comprises a sparsely labelled real data set, by receiving the sparsely labelled real data set for identifying malicious data and comprising real labelled feature vectors and generating a synthetic data set comprising a plurality of synthetic feature vectors derived from the real, labelled feature vectors. The process further identifies the sparsely labelled real data set as a local data set and the synthetic data set as a global set. The process further applies a transfer learning framework for mixing the global data set with the local data set for increasing the precision recall area under curve (PR AUC) for reducing false positive indications occurring in analysis of the threats to the enterprise.

The invention provides a dispatching method, apparatus and system for a gas and steam system in iron and steel enterprises. The dispatching method is deployed in a dispatching application server of the gas and steam system in the iron and steel enterprises and includes acquiring historical data of energy production and consumption of the gas and steam system as well as dispatching parameters determined by a user from a synthetic data integration platform server; predicting gas production amount of a gas production device and gas and steam consumption amount of each production user in several cycles in the future by means of the historical data of energy production and consumption of the gas and steam system; determining optimized dispatching policies of each energy user in the gas and steam system through optimization solution of a dispatching model with the optimization object of maximizing electricity generation benefits and optimizing stability of an energy system according to the dispatching parameters, predicted gas production and consumption amount data, and predicted steam consumption amount data.

Embodiments of the present invention provide an improvement to convention machine model training techniques by providing an innovative system, method and computer program product for the generation of synthetic data using an iterative process that incorporates multiple machine learning models and neural network approaches. A collaborative system for receiving data and continuously analyzing the data to determine emerging patterns is provided. Common characteristics of data from the identified emerging patterns are broadened in scope and used to generate a synthetic data set using a generative neural network approach. The resulting synthetic data set is narrowed based on analysis of the synthetic data as compared to the detected emerging patterns, and can then be used to further train one or more machine learning models for further pattern detection.

Techniques for processing data sets and, more particularly, constructing a synthetic data set (test data set) from real data sets (input data sets) in accordance with user feedback. The technique mimics real data sets effectively to generate the corresponding synthetic ones. Multiple real data sets may be used to create a test data set which combines the characteristics of these multiple data sets. Users of the technique have the ability to modify the characteristics of the data sets to create a new data set which has features that a user may desire. For example, a user may change the shape or size of, or distort the different patterns in the data to create a new data set. A user may also choose to inject noise into the system.

The invention discloses a target detection method based on a synthetic data set. The method includes: adding a real environment picture as a background map to the three-dimensional model of the to-be-detected target in 3ds MAX software so as to establish a three-dimensional scene, rendering and generating a required number of synthetic images, and automatically completing the marking of the imagecategory and the marking box so as to complete the construction of a composite data set; using the synthetic data set as a training set to train the target detection network; and carrying out target detection after training is completed. By using the method, the labeled data set of any target can be quickly obtained at low cost, and the problems that the real data set is high in labeling cost andreal data cannot be obtained in a specific scene are solved. Furthermore, the designed target detection network is added with the SOMConv layer, so that the identification capability of the network onreal data can be improved.

The invention discloses a general water-cooling central air conditioner energy consumption prediction method based on a long-short-term memory recurrent neural network. The general water-cooling central air conditioner energy consumption prediction method comprises the following steps of 1, acquiring water-cooling central air conditioner data and corresponding environment data of normal operationof a plurality of water-cooling central air conditioner projects provided by a large-stroke energy company; step 2, performing data preprocessing on the obtained water-cooling central air conditionerdata in normal operation and the corresponding environment data, and integrating a plurality of data sets to obtain a comprehensive data set; step 3, training the data set to predict the energy consumption of the air conditioner, and adopting a LSTM-RNN long-term and short-term memory cycle neural network, and using the pre-processed data set and corresponding power consumption as input to LSTM-RNN long-term and short-term memory cycle neural network to be subjected to network training to obtain a final prediction model; and 4, inputting the test data into the prediction model to obtain the energy consumption value of the air conditioner under the current working condition. According to the method, the model training process is simplified, and the prediction accuracy is improved.

Systems and methods to generate artificial intelligence models with synthetic data are disclosed. An example system includes a deep neural network (DNN) generator to generate a first DNN model using first real data. The example system includes a synthetic data generator to generate first synthetic data from the first real data, the first synthetic data to be used by the DNN generator to generate a second DNN model. The example system includes an evaluator to evaluate performance of the first and second DNN models to determine whether to generate second synthetic data. The example system includes a synthetic data aggregator to aggregate third synthetic data and fourth synthetic data from a plurality of sites to form a synthetic data set. The example system includes an artificial intelligence model deployment processor to deploy an artificial intelligence model trained and tested using the synthetic data set.

The invention discloses a convolutional neural network rainfall intensity classification method for rainy day pictures. The method comprises the following steps: (1), synthesizing rainfall pictures through image processing software, and obtaining a synthesized data set; (2) establishing a convolutional neural network, and pre-training the convolutional neural network by using the synthetic data set in the step (1); (3) collecting an actual rainfall picture to obtain a real data set; (4) finely adjusting the pre-trained model by using the real data set in the step (3) to obtain a trained model;and (5) using the trained model in the step (4) for real-time rainfall intensity classification. The classification method provided by the invention has a good effect and a low error rate for classification of the rainfall intensity of the real rainfall picture and the synthetic rainfall picture, and can greatly improve the accuracy of real-time weather information in space.

The invention relates to the technical field of data processing, in particular to a CT image data automatic classification method and device based on CNN and GAN, and the method comprises the following steps: S1, obtaining CT image data to be classified; S2, selecting an image of the nodule to carry out data enhancement processing to obtain a public expansion data set; S3, obtaining a generation network and an identification network for the public expansion data set by using the GAN, and performing training at the same time to obtain a GAN synthesis data set; and S4, classifying the GAN synthetic data set by using a CNN network to obtain a final image data set. According to the method, the problem that most of existing researches about lung adenocarcinoma classification focus on radiomicsfeature modeling and other manual marking features, which are based on manual labeling, thus more burden problems are brought to doctors can be solved; according to the invention, the lightweight CNNmodel is also convenient to arrange in a hospital diagnosis system, daily work of radiologists is facilitated, development of precision medical treatment is promoted, and the invention has a very strong market application prospect.

The application provides a scheduling method, a scheduling server and a scheduling system of a combined heat and power generation system of a fire coal thermal power plant. The scheduling method of the combined heat and power generation system of the fire coal thermal power plant is deployed on the scheduling application server in the scheduling system of the combined heat and power generation system of the fire coal thermal power plant, and includes: obtaining initial data needed by building working condition models of all devices in the combined heat and power generation system from a synthetic data integration platform, and building the working condition models of all the devices in the combined heat and power generation system according to the initial data; judging whether preset scheduling conditions are met or not, and if yes, confirming a scheduling policy of the combined heat and power generation system of the fire coal thermal power plant under a current load according to real-time power and heat load data output by the working condition models, and controllable variables and auxiliary variables, which influence outputting of the working condition models. According to the scheduling method, the scheduling server and the scheduling system of the combined heat and power generation system of the fire coal thermal power plant, scheduling problems are considered by integrating blending coal, a boiler, a turbine generator and a temperature and pressure reducer together, and a scheduling scheme can be globally optimized.

The invention discloses a nuclear radiation dose protection method and system. The method includes: collecting the radiation dose, time and current position information of set time points in a current time period, and saving the radiation dose, time and current position information into a collecting database; acquiring an identification code, outputting all data information in the identification code and the collecting database in the current time period, and emptying all the data information in the collecting database; placing the received nuclear radiation dose into the corresponding personal dose cell library according the received identification code; reading the nuclear radiation dose in the personal dose cell library in the current time period to perform data fusion so as to obtain the comprehensive radiation dose in the current time period, and saving the comprehensive radiation dose into a comprehensive data set; comparing the comprehensive radiation dose in the current time period in the comprehensive data set with the radiation dose of nuclear radiation level representation to obtain the nuclear radiation level of the comprehensive radiation dose; acquiring the protection strategy corresponding to the nuclear radiation level.

The invention provides synthetic data set-based three-dimensional multi-person posture estimation. The main contents include a human body posture hierarchical skeleton, posture inference and a networkframework. A process comprises the steps of firstly giving an RGB image, estimating three-dimensional postures of multiple persons in the image, and coding joint positions relative to reference joints; secondly decomposing a body into trunk, four limbs and head by utilizing a special reading scheme of shielding a robust position graph, establishing vectors, and reading out corresponding limb bodypostures; and finally inferring three-dimensional postures through a reading priority and a two-dimensional joint verification policy. A new annotated multi-person data set is created by using an existing single-person three-dimensional data synthesis method; two-dimensional and three-dimensional joint positions can be predicted without bounding box extraction; even for complex multi-person scenes and shielding situations, person postures can be effectively predicted; and realization of the multi-person posture estimation greatly facilitates estimation of the person postures in actual scenes.

The invention relates to a Down's syndrome screening method based on machine learning at the progestational stage and the pregnant metaphase. The method comprises steps that ns fields of pregnant women's Down's screening result data at the pregnant metaphase are selected as training characteristics; Ns samples are added to a data set A; the samples of the data set A are preprocessed to make the number of samples in a minority class set be balanced with the number of samples in a majority class set to obtain a synthetic data set; samples in the synthetic data set are processed to obtain a prediction model for determining whether a fetus has the Down's syndrome, and the prediction model is utilized to predict a tested sample to obtain the prediction result. The method is advantaged in that the process of artificially dividing the indicator threshold is avoided, human resources are saved, and relatively high accuracy and relatively low false positive rate can be achieved.

Systems and methods to fuse aviation-related data systems for comprehensive aircraft system health monitoring are provided. One example method includes obtaining, by one or more computing devices, fault data indicative of a plurality of fault indications provided by a first plurality of components of an aircraft. The method includes obtaining, by the one or more computing devices, condition indicators describing the respective operational conditions of a second plurality of components of the aircraft. The method includes fusing, by the one or more computing devices, the fault data with the condition indicators to form a comprehensive data set. The method includes identifying, by the one or more computing devices, one or more causes of the plurality of fault conditions based at least in part on the comprehensive data set. One example system includes a data fuser, a cause identifier, and an alert generator.

The invention discloses a coastal wetland land cover information extraction method based on integrated multi-source remote sensing data. The method mainly comprises three steps: 1) data collection andpretreatment: to begin with, collecting cloudless optical and SAR remote sensing data in a research area and carrying out corresponding pretreatment, then, combining the two into an integrated data set, and collecting training data and test data and carrying out discretization processing; 2) constructing a coastal wetland land cover information extraction Bayesian network model through a conditional independence test method; and 3) carrying out land cover classification on the multi-dimension data set by utilizing the established Bayesian network model, and realizing automatic extraction of coastal wetland land cover information. The method can realize automatic extraction of land cover information well through integrated utilization of respective advantages of optical and SAR images, andhas high operability and practicality.

The invention relates to a government affair data sharing oriented localized differential privacy method, and provides a government affair data sharing method based on localized differential privacy,which introduces a data binning thought on the basis of a CMS algorithm, and divides data records into a smaller data domain range through equal-width binning to form a BCS algorithm. Therefore, the problem of large statistical error of the current privacy protection algorithm when the data domain is large and the data volume is small is solved; and then the data binning thought is extended and applied to the GRR algorithm to form a BRR algorithm, and an obvious effect is also achieved. In order to verify the effectiveness, the improved BCS and BRR algorithms and the CMS and GRR algorithms arerespectively compared and analyzed on a synthetic data set and a real data set, and experimental results show that the method provided by the invention effectively reduces statistical errors, improves the effectiveness under different distribution and data domain sizes, and has higher popularization and application values.

The invention discloses a video monitoring system capable of collecting information of a second generation identity card, which comprises display equipment, a computer, a multi-branch video collection card, a comprehensive data concentrator, a camera and a second generation identity card reader, wherein after the camera collects images of clients, the camera sends the images to the multi-branch video collection card, after the second generation identity card reader collects the information of an identity card, the information is transmitted to the comprehensive data concentrator, and the computer collects video data flow output by the multi-branch video collection card, receives data sent by the comprehensive data concentrator, conducts superposition process on two kinds of data, and transmits the two kinds of data to the display equipment, thereby enabling a name and an identity card number of the same person to be superposed on a video monitoring image of the display equipment in word mode, and picture information is displayed in the video monitoring image of the display equipment in image mode. The video monitoring system has the advantages that in important situations, information in the second generation identity cards of the clients is superposed and synthesized in a current video monitoring image simultaneously, thereby being favorable for distinguishing and checking for obtaining evidence afterwards.

The invention provides a smooth output method of a wind-power combined power generation system. The smooth output method comprises the following steps of separately acquiring a power forecast value of wind power generation and a power forecast value of photovoltaic power generation, adding the power forecast value of wind power generation into the power forecast value of photovoltaic power generation to obtain total power generation power forecast value so as to form a comprehensive data set; fitting the comprehensive data set by a polynomial fitting algorithm to obtain a smooth output formula; calculating a smooth output value according to the smooth output formula; and determining an output mode and a power output value of an energy storage system according to the size relation between the smooth output value and the total power forecast value of wind-power generation and the absolute value of a different value. According to the method, the smooth output formula is obtained by polynomial fitting on the power forecast value of wind-power generation, the whole planning output section can be considered, the smooth output value after optimization is more moderate, and the inertia time of a first-order lowpass filtering method is reduced. Compared with the prior art, the method provided by the embodiment of the invention has the advantages of more optimal smooth output effect.

The invention discloses an object surface reflection attribute extraction method, and the method proposes a Berlin noise-based synthetic data set, generates textures with low-frequency and high-frequency feature distribution, supports the generation of a data set sample, solves a problem of insufficient training data, and carries out the synthesis of sample data based on Berlin noise. A data set similar to a real sample can be expressed, and the generalization of the network is further improved; besides, the invention also provides an auto-encoder network structure based on VGG-19 and UNet network models, the network performs histogram matching on a feature map, feature extraction is performed through a hole convolution strategy, and the network is trained in combination with weak supervision and rendering technologies so as to obtain an auto-encoder network structure based on VGG-19 and UNet network models. The precision and the quality of the extracted object surface reflection attribute are effectively improved. The invention further provides an object surface reflection attribute extraction device and equipment and a readable storage medium, which have the above beneficial effects.

The invention provides a poison-target literature knowledge mining method and system based on network crawling. The poison-target literature knowledge mining method comprises the steps of: acquiring poison and target data information, and processing the poison and target data information, so as to establish a comprehensive data set; developing a web crawler tool; crawling poison and target literature text information by utilizing the web crawler tool based on the comprehensive data set, and processing the poison and target literature text information to establish a literature text database; based on the literature text database, determining a poison-target potential action relationship by utilizing a natural language processing technology to form a poison-target relationship knowledge base; and performing poison-target literature knowledge mining by utilizing the literature text database and the poison-target relationship knowledge base. The poison-target literature knowledge mining method and the system based on network crawling are high in efficiency, good in accuracy and high in intelligent degree.

The invention discloses an unsupervised machine reading understanding method based on large-scale problem self-learning. The method comprises the following steps: firstly, dividing data into four types; and then, carrying out the following steps: S1, training unlabeled general data by using a standard pre-training model to obtain a pre-training language model; s2, training the marked general data by using a pre-training language model to obtain a question generator, and generating a specific task general domain model; s3, generating synthesized intra-domain data from the unlabeled intra-domain data by using a problem generator, filtering by using a specific task general domain model, and training a high-quality synthesized intra-domain data set obtained by filtering to obtain a new pre-training model; s4, mixing the marked intra-domain data through a low-quality synthetic data set obtained by filtering, marking answers, and then training by using a new pre-training model to obtain a final model; and based on the final model, inputting data to obtain a machine reading understanding result.

The embodiment of the invention provides a neural network training method and system, and the method comprises the steps: for the pixel point in the initial reflection laser image, acquiring projectorrays according to the first spatial intersection point and the central point of the speckle projector if it is judged that a first spatial intersection point exists between an infrared camera projection ray corresponding to a pixel point and a target object; based on the first spatial intersection point, the second spatial intersection point, the projector ray, the virtual plane and the referencespeckle pattern, obtaining an intensity value of the pixel point; obtaining a final object speckle pattern according to the intensity values of the pixel points, and obtaining a synthetic data set according to the final object speckle pattern; and training the neural network by using the synthetic data set. According to the embodiment of the invention, under the condition of less training data, the neural network is trained by generating the synthetic data set, so that the training precision of the neural network is improved, and the requirement of large-scale data volume of the deep learningnetwork can be met.

The invention discloses an end-to-end image defogging method based on multi-feature fusion. The end-to-end image defogging method comprises the steps of 1, acquiring a sample data set; 2, building an end-to-end image defogging network model based on multi-feature fusion, wherein the model comprises a basic network taking a global feature fusion attention module as a core, a prior feature extraction module supporting back propagation and a prior feature adaptive fusion module; the dark channel priori features and the color attenuation priori features enter a priori feature adaptive fusion module to be fused, and then are fused with deep learning features obtained by the basic network; step 3, constructing a loss function; 4, training an end-to-end image defogging network model based on multi-feature fusion; and step 5, carrying out defogging processing on a to-be-processed image by using the trained model to obtain a defogged image. Experimental results of a synthetic data set and a real data set show that the defogging capability and migration capability of the model in a real scene are improved, the parameter quantity is small, and rapid defogging can be realized.

The invention relates to a defogging method for a foggy image in a real scene based on fog migration and feature aggregation, and belongs to the field of image processing. The invention designs a method for realizing image defogging by migrating fog in a foggy image in a real scene to a clear image to generate a data set and then utilizing a defogging network based on feature aggregation. In a fog migration process, a multi-level feature block identification method is designed to migrate fog in a real scene to a clear image to generate a foggy image training data set, and the image in the data set has a style similar to that of a foggy image in the real foggy scene and distribution characteristics of fog in the foggy image. In addition, extraction features are supplemented in a fine-grained detail information and semantic information aggregation mode so as to realize image defogging. According to the method, a good defogging effect is achieved on the foggy image in a real scene, and the problem that a defogging model trained by a synthetic data set is poor in generalization performance on the real foggy image is greatly solved.

Systems, apparatus, instructions, and methods for medical machine time-series event data generation are disclosed. An example synthetic time series data generation apparatus is to generate a synthetic data set including multi-channel time-series data and associated annotation using a first artificial intelligence network model. The example apparatus is to analyze the synthetic data set with respect to a real data set using a second artificial intelligence network model. When the second artificial intelligence network model classifies the synthetic data set as a first classification, the example apparatus is to adjust the first artificial intelligence network model using feedback from the second artificial intelligence network model. When the second artificial intelligence network model classifies the synthetic data set as a second classification, the example apparatus is to output the synthetic data set.

A data-driven linear filtering method to recover microseismic signals from noisy data / observations based on statistics of background noise and observation, which are directly extracted from recorded data without prior statistical knowledge of the microseismic source signal. The method does not depend on any specific underlying noise statistics and works for any type of noise, e.g., uncorrelated (random / white Gaussian), temporally correlated and spatially correlated noises. The method is suitable for microquake data sets that are recorded in contrastive noise environments. The method is demonstrated with both field and synthetic data sets and shows a robust performance.

Certain embodiments of the present invention provide a system for dynamically displaying clinical information to a user, the system including: a table for displaying the clinical information, the table including entries; an information display subsystem for displaying the clinical information in the table; and an information collection subsystem for collecting a first portion of the clinical information at a first time, for collecting a second portion of the clinical information at a second time, and for providing the first and second portions to the information display subsystem, wherein the information display subsystem displays the first portion of the clinical information at a first display time in the table, and integrates the first and second portions of the clinical information to form an integrated data set, and displays the integrated data set in the table at a second display time such that each of at least a portion of the entries includes information from both of the first and second portions of the clinical information. In an embodiment, the user is capable of interacting with an entry at substantially the first display time.

The invention discloses a natural scene character detection and recognition method based on CRAFT and SCRN-SEED frameworks, and the method comprises the following steps: (1) building an image data set through employing a real data set and a synthetic data set, and dividing the image data set into a training set and a test set; (2) training a CRAFT network by using the image data set; (3) training an irregular text correction network SCRN by using the real data set; (4) combining the SCRN with the SEED network, and training the combined SCRN-SEED network; and (5) the CRAFT network is connected with the SCRN-SEED network, and a complete model is constructed and trained. According to the method, bent and deformed texts or long text instances can be fully detected, each character is accurately positioned, then the detected characters are connected into one text through an affinity mechanism to achieve the purpose of detection, and the method is suitable for bent, deformed or extremely long texts; by correcting irregular text pictures and using semantic information for global information detection, low-quality text instances can be accurately recognized.