1377 results about "Feature learning" patented technology

A 1D-CNN-based ((one-dimensional convolutional neural network)-based) distributed optical fiber sensing signal feature learning and classification method is provided, which solves a problem that an existing distributed optical fiber sensing system has poor adaptive ability to a complex and changing environment and consumes time and effort due to adoption of manually extracted distinguishable event features, The method includes steps of: segmenting time sequences of distributed optical fiber sensing acoustic and vibration signals acquired at all spatial points, and building a typical event signal dataset; constructing a 1D-CNN model, conducting iterative update training of the network through typical event signals in a training dataset to obtain optimal network parameters, and learning and extracting 1D-CNN distinguishable features of different types of events through an optimal network to obtain typical event signal feature sets; and after training different types of classifiers through the typical event signal feature sets, screening out an optimal classifier.

The present invention provides a text implication relation recognition method which fuses multi-granularity information, and proposes a modeling method which fuses multi-granularity information fusionand interaction between words and words, words and words, words and sentences. The invention firstly uses convolution neural network and Highway network layer in character vector layer to establish word vector model based on character level, and splices with word vector pre-trained by GloVe; Then the sentence modeling layer uses two-way long-short time memory network to model the word vector of fused word granularity, and then interacts and matches the text pairs through the sentence matching layer to fuse the attention mechanism, finally obtains the category through the integration classification layer; After the model is established, the model is trained and tested to obtain the text implication recognition and classification results of the test samples. This hierarchical structure method which combines the multi-granularity information of words, words and sentences combines the advantages of shallow feature location and deep feature learning in the model, so as to further improve the accuracy of text implication relationship recognition.

The invention provides an image super-resolution reconstruction method based on a fused attention mechanism residual network, and solves the problems of poor image reconstruction quality and non-idealvisual effect in the prior art. The researched image reconstruction method comprises the steps of S1, performing data acquisition and preprocessing, and obtaining a training image data set and a to-be-reconstructed image data set; S2, building a network model, wherein the model structure comprises a feature extraction layer, a feature learning layer and an image reconstruction layer; S3, initializing, training and storing model parameters to obtain an optimal model structure and an optimal parameter set; and S4, performing image super-resolution reconstruction, inputting a to-be-reconstructedimage, and outputting a high-resolution image under a corresponding amplification scale. The image super-resolution network provided by the invention combines global and local residual structures, fuses channel attention and spatial attention mechanisms, pays more attention to high-frequency information of the image, reserves important features in the image to the greatest extent, reduces repeated and redundant features, and greatly improves the details and definition of the reconstructed image.

The invention relates to an adaptive extraction and diagnosis method for degree features of a mechanical fault through a stack-type sparse automatic coding depth neural network, and belongs to the technical field of mechanical equipment state monitoring and reliability evaluation. The method aims at a problem of intelligent diagnosis of the degree of the mechanical fault, and comprises the steps: carrying out the stacking of sparse automatic coding, adding a classification layer, and constructing the stack-type sparse automatic coding depth neural network which integrates the adaptive learning and extraction of the degree features of the fault and fault recognition; employing the advantage that the sparse automatic coding can automatically learn the internal features of data, and adding noise coding to be integrated in the sparse automatic coding for improving the robustness of feature learning; carrying out the layer-by-layer no-supervision adaptive learning and supervision fine tuning of the original input complex data through multilayer sparse automatic coding, completing the automatic extraction and expression of the degree features of the mechanical fault and achieving the intelligent diagnosis of the degree of the fault. The method is used for the diagnosis of the degree of faults of rolling bearings under different work conditions, and obtains a good effect of feature extraction and diagnosis.

The invention discloses a short text clustering method based on deep semantic feature learning. The method includes the steps that dimensionality reduction representation is performed on original features under the restraint of local information preservation through traditional feature dimensionality reduction, binarization is performed on an obtained low-dimension actual value vector, and error back propagation is performed with the binarized vector being supervisory information of a convolutional neural network structure to train a model; non-supervision training is performed on a term vector through an outer large-scale corpus, vectorization representation is performed on all words in text according to the word order, and the vectorized words serve as implicit semantic features of initial input feature learning text of the convolutional neural network structure; after deep semantic feature representation is obtained, a traditional K-means algorithm is adopted for performing clustering on the text. By means of the method, extra natural language processing and other specialized knowledge are not needed, design is easy, deep semantic features can be learnt, besides, the learnt semantic features have unbiasedness, and good clustering performance can be achieved more effectively.

The invention discloses a modulation mode identification system and method based on a convolutional neural network, which solve the problems of complex feature extraction steps and low identificationrate under a low signal-to-noise ratio in the prior art. The simple feature in the identification system is constructed as a simple feature using a co-directional component and a quadrature componentof a baseband signal as signals, and the simple feature is sent to a convolutional neural network module for identification. The identification method comprises the steps of: modulating a transmittedsignal and performing pulse shaping; performing up-conversion on the transmitted signal and then transmitting the transmitted signal through an additive white Gaussian noise channel; performing pre-processing first by a receiving end to obtain the co-directional component r(t) of the analyzed signal; constructing the simple feature, i.e., constructing the co-directional component r(t) and the quadrature component of the analyzed signal into a two-dimensional matrix; performing feature learning and classification by the convolutional neural network; and sending a modulation method to a demodulation end to obtain a demodulated signal. The method is low in feature design complexity, avoids explicit feature extraction, has high classification correctness, and can be applied to communication systems having high recognition performance requirements.

A pose-aware feature learning system includes an object tracker which tracks an object on a subject in a plurality of video frames, a pose estimator which estimates a pose of the subject in a track of the plurality of video frames, an image pair generator which extracts a plurality of image pairs from the track of the plurality of video frames, and labels the plurality of image pairs with the estimated pose and as depicting the same or different object, and a neural network trainer which trains a neural network based on the labeled plurality of image pairs, to predict whether an image pair depicts the same or different object and a pose difference for the image pair.

The invention discloses an online learning cognitive map generation system and method for representing the knowledge learning mastering state of a leaner in a specific field. The system comprises an intelligent terminal and a server. The intelligent terminal is used for online learning interaction. The online learning cognitive map generation system runs in the server comprising a knowledge graph construction module, a learning interaction data acquisition module, a knowledge state calculation and prediction module and a visual output module. The online learning cognitive map generation system and method are suitable for a common online learning platform and can simulate the cognitive structure of the learner and diagnose the learning defect of the leaner to provide personalized learning resources and social network services for the learner, so that learning confusion is reduced, cognitive loads are reduced, and learning pertinency and accuracy are improved.

The invention discloses a data-driven and task-driven image classification method. The data-driven and task-driven classification method comprises the steps that a convolutional neural network structure is designed according to the scale of data sets and image content; a convolutional neural network model is trained through the given classified data sets; feature expression is extracted from training set images through a trained convolution neural network; images to be tested are input into the trained convolutional neural network and are classified. The data-driven and task-driven image classification method is based on nonlinear convolution feature learning, and the model can be adapted to the data sets through a date driving mode, so that the specific data set can be better described; errors of K-nearest neighbors can be directly optimized through a task-driving mode, and therefore a better performance can be obtained with respect to a K-nearest neighbor task; efficient training can be conducted through a GPU in the training stage, and efficient K-nearest neighbor image classification can be achieved just through a CPU in the testing stage; in this way, the data-driven and task-driven image classification method is quite suitable for a large-scale image classification task, a retrieval task and the like.

The invention discloses a brain tumor segmentation network and segmentation method based on a U-Net network. The tail of a contraction path of the segmentation network is connected with a spatial pyramid pooling structure; hole convolution of different scales is introduced into a network jump connection part of the segmentation network; an Add operation and original input are adopted to form a residual block with hole convolution; a receptive field of shallow feature information in the contraction path is expanded; fusing with an expansion path of a corresponding stage is carried out. The segmentation method comprises the following steps: cutting and preprocessing a training data set, then constructing a brain tumor segmentation network DCU-Net based on a U-Net network, then inputting a preprocessed two-dimensional image into a segmentation model for feature learning and optimization, obtaining an optimal parameter model of the segmentation model, and finally inputting a to-be-segmented test data set image into the segmentation model for tumor region segmentation. According to the method, the problems of over-segmentation and under-segmentation in brain tumor segmentation can be effectively solved, and the brain tumor segmentation precision is improved.

The invention discloses a nasopharyngeal-carcinoma (NPC) lesion automatic-segmentation method and nasopharyngeal-carcinoma lesion automatic-segmentation systems based on deep learning. The method comprises: carrying out registration on a PET (Positron Emission Tomography) image and a CT (Computed Tomography) image of nasopharyngeal carcinoma to obtain a PET image and a CT image after registration;and inputting the PET image and the CT image after registration into a convolutional neural network to carry out feature representation and scores map reconstruction to obtain a nasopharyngeal-carcinoma lesion segmentation result graph. The method carries out registration on the PET image and the CT image of the nasopharyngeal carcinoma, obtains a nasopharyngeal-carcinoma lesion by automatic segmentation through the convolutional neural network, and is more objective and accurate as compared with manual segmentation manners of doctors; and the convolutional neural network in deep learning isadopted, consistency is better, feature learning ability is higher, the problems of dimension disasters, easy falling into a local optimum and the like are solved, lesion segmentation can be carried out on multi-modal images of the PET-CT images, and an application range is wider. The method can be widely applied to the field of medical image processing.

A feature information collecting device includes: a feature image recognizing unit that performs image recognition processing of a feature included in image information in a surrounding area of a vehicle; a construction information obtaining unit that obtains construction information including information of a construction section; a construction information storage unit that stores the obtained construction information; a construction completion determining unit that determines, when the vehicle travels a section of a road corresponding to a construction section according to the already stored construction information, completion of construction indicated by the construction information; and a feature learning unit that causes the feature image recognizing unit to perform image recognition processing of a feature in a construction section according to the construction information when the completion of construction is determined, and that generates, based on an image recognition result thereof and the vehicle position information, learned feature information including position information and attribute information of an image-recognized feature.

The invention discloses a CT image kidney segmentation algorithm based on a residual double-attention deep network. According to the method, the advantage that the residual error unit can repeatedly utilize the features and the excellent feature learning capability of the double attention mechanism are combined; a residual double attention module is designed; and a residual double attention moduleis used as a basic module to construct a U-shaped deep network segmentation model, and a loss function for segmentation is designed at the same time, so that the U-shaped deep network segmentation model can pay more attention to kidney region features, can effectively cope with shape changes of the kidney with cystic lesions, and can maintain robustness for the shape changes of the kidney with cystic lesions. Therefore, the boundary of the kidney area is accurately positioned, and automatic segmentation of the kidney area in the CT image is achieved, and a good segmentation effect is achieved.

The invention discloses a text classification method based on a bidirectional cyclic attention neural network, and belongs to the technical field of learning and natural language processing. The method comprises the following steps: step 1, preprocessing data; Step 2, according to the preprocessed data, generating and training a word vector of each word through a Word2vec method; Step 3, performing text semantic feature extraction on the word vector according to the word vector, fusing an attention mechanism and a bidirectional recurrent neural network, calculating the overall weight of each word, and converting the weight into an output value Y (4) of a model; And step 4, taking the feature vector Y (4) as the input of a softmax classifier according to the feature vector Y (4), and carrying out classification identification. According to the method, the attention mechanism is fused in the text feature learning model, the effect of keywords can be effectively highlighted, the performance of the model is greatly improved, and the text classification accuracy is further improved.

The present disclosure is directed to collaborative feature learning using social media data. For example, a machine learning system may identify social media data that includes user behavioral data, which indicates user interactions with content item. Using the identified social user behavioral data, the machine learning system may determine latent representations from the content items. In some embodiments, the machine learning system may train a machine-learning model based on the latent representations. Further, the machine learning system may extract features of the content item from the trained machine-learning model.

The invention discloses a face unsupervised feature learning method and device based on a generative adversarial network. Through the preprocessing of a collected original face image, the image is converted into a face training image with a set size. The converted face training image is taken as training data to train a target generation network in a constructed depth convolution generative adversarial network. A generated random vector set is inputted into the trained target generation network, and a generative image set corresponding to the random vector set is obtained. The obtained generative image set is inputted into a depth regression network of a constructed depth convolutional neural network, the depth regression network is trained, and a face feature vector of the generative image set is extracted. The invention provides the face unsupervised feature learning method and device based on a generative adversarial network, a mode with the combination of DCGAN and DCNN is used to carry out unsupervised learning, the depth regression network is used to learn a reverse target generation network, the learning effect is good, and the identification accuracy is high.

In a support vector regression approach to forecasting power load in an electrical grid, a feature learning scheme weights each feature in the input data with its correlation with the predicted load, increasing the prediction accuracy. The kernel matrix for the input training data is computed such that features that align better with the target variable are given greater weight. The resulting load forecast may be used to compute commands sent to demand response modules.

The invention relates to a wind turbine generator bearing fault diagnosis method for a multi-channel deep convolutional neural network. The method comprises the steps of: simultaneously acquiring high-frequency vibration acceleration signals of a drive end and a non-drive end of a test bearing in various states by using a vibration acceleration sensor; analyzing the acquired vibration signals by using a time frequency analysis technology to obtain corresponding time frequency spectra; establishing a deep convolutional neural network diagnosis model, and training the diagnosis model by using the time frequency spectra and the states of the bearing as a training sample; evaluating the diagnosis model, and applying the diagnosis model to a bearing to be monitored. The method can realize automatic feature learning, avoids feature engineering, effectively utilizes multi-channel vibration signals, and has good universality and extensibility.

The invention discloses a federated learning-based recommendation model training method, a terminal and a storage medium. The method comprises the steps of obtaining a user historical behavior data set recorded by a client application in multiple preset types of application projects; extracting a single characteristic user vector of each client application based on each group of user historical behavior data set; extracting a project feature vector set and a project score set from a user historical behavior data set of the target application; combining each single feature user vector, the project feature vector set and the project score set to obtain a local training sample set; and participating in federated learning based on the local training sample set to obtain a recommendation modelof the target type application project. According to the method, model training is carried out under a federal framework to protect user privacy data, meanwhile, recommendation model training is carried out on the basis of multi-scene data, the recommendation model obtained through training can more accurately locate the preference characteristics of the user, and therefore the recommendation effect of the recommendation model is improved.

The present invention provides a method for detecting image steganography based on deep learning, which comprises: filtering images having steganographic class label or true class label in a training set with a high-pass filter to obtain a training set including steganographic class residual images and true class residual images; training a deep network model on said training set to obtain a trained deep model for steganalysis; filtering the image to be detected with said high-pass filter to obtain a residual image to be detected; detecting said residual image to be detected on said deep model so as to determine whether said residual image to be detected is a steganographic image. The method for detecting image steganography in the present invention can create an automatic blind steganalysis model through feature learning and can identify steganographic images accurately.

The invention discloses a small target detection and recognition method for enhanced feature learning, belongs to the field of image processing, pattern recognition and computer vision, and solves theproblems of low small target detection and recognition task detection precision and low network efficiency in the prior art. The method comprises: sequentially constructing a basic network module, afeature extraction module, a candidate box generation module and a prediction output module to serve as a small target detection and identification network; preprocessing the extracted small target sample image data based on the extracted small target sample image data; inputting the preprocessed small target sample image data into the small target detection and recognition network with initialized parameters for training to obtain a trained small target detection and recognition network; and inputting a to-be-predicted small target image into the trained small target detection and recognitionnetwork, and outputting the prediction box position and category information of the small target end to end through forward propagation. The method is used for small target detection and recognition.

The invention discloses a malicious traffic detection method. The method comprises the following steps: correspondingly establishing malicious and normal data sample libraries by using obtained malicious and normal data traffic samples; executing a data cleaning operation and a preprocessing operation on the data sample libraries in sequence to obtain training data, and constructing a traffic detection model by using the training data and a deep learning algorithm; judging whether to-be-measured data traffic contains malicious data by using the traffic detection model; and if so, sending alarminformation carrying the to-be-measured data traffic belonging to malicious data via a preset oath. Feature learning and training are performed by using the malicious and normal data traffic samplesvia the automatic learning property of the deep learning algorithm, the feature information extraction operation is completed without consuming precious human resources, thereby improving the improving the work efficiency and improving the discrimination of the malicious traffic. Precision. The invention further discloses a malicious traffic detection system and apparatus and a computer readable storage medium, which have the above beneficial effects.

The invention relates to a cell image segmentation method based on automatic feature learning. As a method for learning features of cell images is very good in feature learning capacity, the cell segmentation accuracy can be greatly improved, and meanwhile, a random forest classifier does not need to select the features, so that the method is capable of well solving the confronted problems of feature extraction and selection in a recognition process. The cell image segmentation method based on the automatic feature learning comprises the following steps: 1, preprocessing: preprocessing initial cell images in a training set and a test set; (2) training a feature extractor; (3) performing recognition by utilizing the random forest classifier; and (4) postprocessing.

The invention provides a training method of a neural network model. The method comprises the steps of: acquiring a training data set; setting an initial loss function of the to-be-trained neural network model, and using a regularization term to update the initial loss function; and inputting the training data set to the to-be-trained neural network model after updating of the initial loss functionto carry out feature learning training until convergence occurs. In the embodiment of the invention, correlation among model parameter vectors is removed, and model running efficiency is improved. The invention also provides a training device of the neural network model, a storage medium and terminal equipment.

The present invention provides a vehicle position prediction method based on deep learning. The method comprises the steps of: S1, obtaining an original VLPR data set, performing cleaning and denoising of the original VLPR data set, obtaining a VLPR data set, and performing grouping of the VLPR data set according to a set time quantum; S2, from the VLPR data set, extracting vehicle pass record ofthe same vehicle, generating a vehicle track data set and screening out a track meeting a demand, and performing data conversion of the related feature information of the track; S3, establishing an algorithm model based on deep learning according to the obtained vehicle track features, and achieving analysis and learning of the track features; and S4, after performing feature learning of the vehicle track, employing a full-connection network layer to combine a Softmax classifier to output a next position vector, matching real geographic position information, outputting the real geographic position information, and achieving vehicle position prediction. The method provided by the invention analyzes operation features of a road network and employs the track features in the vehicle driving process so as to obtain high prediction precision.