244 results about "Hybrid neural network" patented technology

The invention provides an MRI (Magnetic Resonance Imaging) brain tumor localization and intratumoral segmentation method based on a deep cascaded convolution network, which comprises the steps of building a deep cascaded convolution network segmentation model; performing model training and parameter optimization; and carrying out fast localization and intratumoral segmentation on a multi-modal MRIbrain tumor. According to the MRI brain tumor localization and intratumoral segmentation method provided by the invention based on the deep cascaded convolution network, a deep cascaded hybrid neuralnetwork formed by a full convolution neural network and a classified convolution neural network is constructed, the segmentation process is divided into a complete tumor region localization phase andan intratumoral sub-region localization phase, and hierarchical MRI brain tumor fast and accurate localization and intratumoral sub-region segmentation are realized. Firstly, the complete tumor region is localized from an MRI image by adopting a full convolution network method, and then the complete tumor is further divided into an edema region, a non-enhanced tumor region, an enhanced tumor region and a necrosis region by adopting an image classification method, and accurate localization for the multi-modal MRI brain tumor and fast and accurate segmentation for the intratumoral sub-regions are realized.

The invention belongs to the data mining and machine learning field and relates to a non-invasive load identification algorithm based on a hybrid neural network and ensemble learning. According to the method, experimental data are processed, so that the format of the data conforms to the input formats of models; after the data are processed, a hybrid neural network model is established; the data are input into the model; the model is trained and tested, identification results are obtained; and voting is performed for the results of three different models based on the idea of ensemble learning, so that a final identification result is obtained. With the method adopted, the feature extraction effect and load identification effect of the hybrid neural network are better than the effects of a traditional neural network; an ensemble learning idea-based method is provided, a plurality of feature subsets are selected from a total feature set so as to train a plurality of base classifiers, and the base classifiers are combined, and therefore, variance can be decreased, and the identification effect of the final identification result can be improved, and the problem of adverse influence of the introduction of harmonic features on an identification effect can be solved.

The invention relates to a micro expression recognition method based on a 3D convolution neural network. Based on a constructed 3D convolution neural network (3D-CNN) model, happiness, disgust, depression, surprise as well as five other micro expressions can be recognized effectively. The designed micro expression recognition method is simple and efficient. There is no need to carry out a series of processes such as feature extraction, feature dimension reduction and classification on sample data. The difficulty of preprocessing is reduced greatly. Through receptive field and weight sharing, the number of parameters needing to be trained by the neural network is reduced, and the complexity of the algorithm is reduced greatly. In addition, in the designed micro expression recognition method, through down-sampling operation of a down-sampling layer, the robustness of the network is enhanced, and image distortion to a certain degree can be tolerated.

System and method for optimization of a design associated with a response function, using a hybrid neural net and support vector machine (NN/SVM) analysis to minimize or maximize an objective function, optionally subject to one or more constraints. As a first example, the NN/SVM analysis is applied iteratively to design of an aerodynamic component, such as an airfoil shape, where the objective function measures deviation from a target pressure distribution on the perimeter of the aerodynamic component. As a second example, the NN/SVM analysis is applied to data classification of a sequence of data points in a multidimensional space. The NN/SVM analysis is also applied to data regression.

The invention discloses a hybrid neural network-based gesture recognition method. For a gesture image to be recognized and a gesture image training sample, first a pulse coupling neural network is used to detect to obtain noise points, then a composite denoising algorithm is used to process the noise points, then a cell neural network is used to extract edge points in the gesture image, connected regions are obtained according to the extracted edge points, curvature is used to perform fingertip detection on each connected region to obtain undetermined fingertip points, interference of a face part is eliminated to obtain a gesture region, then the gesture region is partitioned according to gesture shape features, Fourier descriptors which keep phase information are obtained according to contour points of the partitioned gesture region, and first multiple Fourier descriptors are selected as gesture features; and a BP neural network is trained according to gesture features of the gesture image training sample, and the gesture features of the gesture image to be recognized are input to the BP neural network for recognition. The hybrid neural network-based gesture recognition method provided by the invention improves the accuracy rate of gesture recognition through utilization of various neural networks.

Systems and methods using a neural network based portable absorption spectrometer system for real-time automatic evaluation of tissue injury are described. An apparatus includes an electromagnetic signal generator; an optical fiber connected to the electromagnetic signal generator; a fiber optic probe connected to the optical fiber; a broad band spectrometer connected to the fiber optic probe; and a hybrid neural network connected to the broad band spectrometer. The hybrid neural network includes a principle component analyzer of broad band spectral data obtained from said broad band spectrometer.

The invention relates to a method for forecasting short-term wind speed of a wind farm based on hybrid neural network. The method comprises the following steps: S1, determining an input variable and an output variable of a hybrid neutral network forecasting model according to a preset forecasting time interval; and S2, forecasting the wind speed according to the hybrid neutral network forecasting model to obtain corresponding wind speed forecasting value. The invention also relates to a system for forecasting short-term wind speed of the wind farm based on the hybrid neural network. The system comprises a variable determination module for determining the input variable and output variable of the hybrid neutral network forecasting model according to the preset forecasting time interval; and a forecasting module for forecasting the wind speed according to the hybrid neutral network forecasting model to obtain the corresponding wind speed forecasting value. The method and the system provided by the invention have advantages of high computation speed and high reliability, solve the technical problem completely depending on a physical forecasting model and overcome the disadvantage of large forecasting error fluctuation based on a single model.

The invention discloses a construction method of hybrid neural network model for dialogue generation. The construction method of hybrid neural network model for dialogue generation includes the steps: acquiring a data set in a mode of dialogue statement pairs, and constructing a glossary; generating a word embedded table; initializing the convolution neural network with special structure, generating a vocabulary recommending table corresponding to the input statement, determining whether real output is provided, and if so, training the parameters of the convolution neural network in the step; initializing the recurrent neural network with special structure, using the last step to output, generating a vocabulary identity list with word order, determining whether real output is provided, and if so, training the parameters of the recurrent neural network in the step; after the training result satisfies the set index, saving the glossary and the word embedded table, and saving the parameters of the convolution neural network and the recurrent neural network, thus completing construction of the whole model. The construction method of hybrid neural network model for dialogue generation solves the problems that a current neural network dialogue model is slow in the training speed, low in the accuracy and general in statement generation because the glossary is too long.

The invention relates to an abnormal flow detection method and system based on a hybrid neural network, and the method comprises the steps: firstly, collecting network flow data, and carrying out thefeature extraction and data preprocessing through taking network flow as granularity; learning spatial features in the network traffic data through a convolutional neural network; inputting the features containing the spatial information into a bidirectional long-short time memory network to further learn the time sequence features of the bidirectional long-short time memory network; finally, outputting a detection result. Compared with an existing machine learning and deep learning abnormal flow detection method, the method has the advantages that high-dimensional features can be better mined, and the accuracy of an intrusion detection model is improved. The method is reasonable in design, and the accuracy rate, the detection rate and the accuracy rate of the obtained classification modelare all high.

The invention discloses a construction method and device of a project recommendation model based on a hybrid neural network and a project recommendation method. The construction method comprises the following steps: filtering comment information, preprocessing the filtered comment information, and learning context features related to a project in the preprocessed comment information and user features and project features in scoring information by using a convolutional neural network; subsequently, fusing and interacting the project characteristics in the user-project scoring information and the context characteristics in the comment information, integrating the learned user characteristics and the fused project characteristics into a multi-task learning framework, and performing joint training to obtain a project recommendation model based on the hybrid neural network. According to the invention, the two heterogeneous data of the scoring information and the comment information are integrated into one unified model, so that the implicit feature vectors of the user and the project can be learned more accurately, and the purposes of improving the performance of the recommendation system and improving the recommendation effect are achieved.

The invention provides a deep architecture for detecting attributes (such as gender, race and clothing) of a human body in a monitoring video. The method is characterized by, through capability of a hybrid neural network and a part-based method, carrying out decomposition forecasting on an object in an image, and giving robustness; carrying out training based on weighted loss, obtaining attributeprediction scores, re-constructing a network to enable the architecture to have robustness to occlusion, eliminating obstacles, and carrying out classification on occlusion images through a discriminator network; and finally, improving resolution of the images through a super-resolution network, and obtaining attribute identification result of the video human body. The method can improve the resolution of the low-resolution images, process occlusion problems, and can effectively extract attributes even under conditions of poor resolution and strong occlusion, so that identification efficiencyis improved greatly; and the method is suitable for a plurality of application fields.

The invention discloses an intelligent computer. The intelligent computer comprises a hybrid neural network application layer, an HnetCP interface layer, a LabGrid middleware layer and a software and hardware resource layer. In the hybrid neural network application layer, a customer realizes a hybrid neural network system and develops a visual hybrid neural network application development environment through an interface of the HnetCP interface layer. The HnetCP interface layer defines various interfaces for operating a hybrid neural network, a top-layer application is an application in the hybrid neural network application layer, and bottom-layer network middleware is the LabGrid middleware layer. The LabGrid middleware layer provides a grid operation environment for the upper-layer application. The software and hardware resource layer is located at the bottom, and software resources comprise kinds of software supporting the upper-layer application. According to the intelligent computer, knowledge stored in the computer and experiential knowledge of people are integrated, and the overall advantages of a computer system are exerted.

The invention belongs to the technical field of medical CT (computed tomogram) image segmentation, and relates to a tooth CT image segmentation method based on deep learning. According to the technical scheme provided by the invention, a traditional Level Set algorithm and a U-net network model are combined, and a Level Set algorithm is used for solving the problem of a training set required by the neural network, so that the neural network can be trained by using unmarked tags, at the same time, the neural network model is used to complete the automatic segmentation of the image, the problemof non-convergence of curve evolution is avoided, and a sufficient and accurate segmentation effect can be obtained under the condition that a medical image training set is insufficient.

The invention discloses a protein-ligand binding site prediction algorithm based on deep learning. For protein to be predicted, the algorithm comprises the steps of: firstly, extracting sequence features and a distance matrix; distributing the sequence features to each residue through adoption of a sliding window method; and inputting the features corresponding to the residues into a residual neural network and a hybrid neural network one by one, and inputting output results of the residual neural network and the hybrid neural network into a Logistic regression classifier to obtain a final result, namely the binding probability corresponding to each residue in the protein. According to the method, a classic bidirectional long-short-term memory network and a residual neural network are fused, the fused network can process heterogeneous protein sequences and structural data at the same time, and complementarity of sequence features and structural features is mined. Compared with an existing method, the protein-ligand binding site prediction algorithm has higher prediction precision, and has good generalization performance for data sets of different ligands.

The invention discloses a digital-analog hybrid neural network chip architecture. The architecture comprises a two-dimensional SRAM module, an analog synaptic circuit, a nerve cell circuit, an AER communication module, and a master control digital unit. The two-dimensional SRAM module is taken as a storage unit of neural network connection relation and a synaptic weight value. The analog synaptic circuit and the nerve cell circuit respectively consist of an MOSFET circuit working in a subthreshold section. The AER communication module serves as the input and output interfaces of a chip, and employs an AER protocol for communication. All control circuits in the architecture are synchronous digital circuits. The architecture is low in power consumption, is high in degree of parallelism, and can achieve a neural network algorithm in a reasonable chip area, wherein the neural network algorithm is more complex in nerve cell functions, is larger in network scale, and is more flexible in connection.

The invention provides a CNN-DNN hybrid neural network based noise reduction method. The method is implemented by the following steps that 1, a CNN-DNN hybrid neural network noise reduction model is established; 2, a training set is established for training the CNN-DNN hybrid neural network noise reduction model established in the step 1; and 3, a speech signal needing to be subjected to noise reduction is input to the trained CNN-DNN hybrid neural network noise reduction model in the step 3, and a clean speech signal spectrum is output. The CNN-DNN hybrid neural network based noise reductionmethod has better automatic identification separation and removal capabilities on transient noise and non-transient noise.

The invention discloses a driver visual dispersion detection method based on a deep fusion neural network, which is used for detecting the sight attention of a driver in real time. The method comprises the following steps: (1) acquiring a driver face image, and tracking facial feature points of a driver by adopting local binary features (LBF); (2) positioning and extracting an image of the eye region according to the positions of the facial feature points of the eye region; (3) estimating the head posture by adopting an N-point perspective (PnP) algorithm according to the positions of the facial feature points, and obtaining head posture parameters of the driver in three directions; And (4) estimating the driver sight direction based on the deep hybrid neural network. According to the invention, the attention of a driver can be effectively detected.

The invention discloses a photovoltaic power prediction method of a deep neural network model fused with an attention mechanism, and belongs to the technical field of renewable energy photovoltaic power. According to the method, firstly, a hybrid neural network based on a long-term and short-term memory neural network and a convolutional neural network is selected as a prediction model according to photovoltaic data characteristics, and an optimal connection mode is considered; and secondly, in order to shorten the calculation time of the model and more accurately extract high-quality featureinformation capable of being used for photovoltaic prediction, an attention mechanism model is added in the aspect of model feature extraction. Through comparison of different prediction models, the advantages of the proposed hybrid deep learning model are proved, and possibility is provided for selection of high-quality features through application of an attention mechanism model. The reasonablehybrid model mode can realize dual pursuits of high prediction precision and low calculation cost.

The invention discloses an electricity stealing detection method and system based on ResNetLSTM; and the method comprises the steps: collecting an electricity consumption data sample of a user and anelectricity consumption type label of the sample, which are collected during the normal operation of a power system, carrying out data preprocessing of electricity consumption data, and enabling a data set to be divided into a training set, a testing set and a verification set; using an automatic encoder to process the labels in the training set as electricity utilization data samples for electricity stealing, and obtaining a new training set; respectively inputting the power consumption data of the original training set into a ResNet model and an LSTM model to carry out an electricity larcenydetection test, selecting a ResNet and LSTM neural network combined structure through a test result, building a hybrid neural network according to the ResNet and LSTM neural network combined structure, and selecting a proper hybrid neural network structure through the test; performing testing by using the selected ResNet-LSTM hybrid neural network structure to select an appropriate neural networkoptimization method, and forming an electricity stealing detection model. The new training set is applied to train the electricity stealing detection model, a complete electricity stealing detectionmethod is constructed, and the electricity stealing detection capability and the detection efficiency are improved.

The invention relates to the technical field of computers, in particular to a social network rumor identification method and system based on a hybrid neural network. The social network rumor identification method and system aims to solve the technical problem that rumors in the social network can not be accurately identified under a situation that rumor forwarding and commenting information is considered. For the purpose, the social network rumor identification method comprises the following steps that: firstly, utilizing three different neural networks to independently obtain a user feature vector, an original text feature vector and a spread information feature vector; then, combining the user feature vector, the original text feature vector and the spread information feature vector to obtain a new feature vector; and finally, utilizing a fourth neural network to carry out rumor identification on the combined feature vectors. On the basis of the above steps, the rumors in the socialnetwork can be quickly and accurately detected. Meanwhile, the system in the invention can execute and realize the above steps.

A computer-implemented method and system for building a neural network is disclosed. The neural network predicts at least one target based upon predictor variables defined in a state space. First, an input data set is retrieved that includes the predictor variables and at least one target associated with the predictor variables for each observation. In the state space, a number of points is inserted in the state space based upon the values of the predictor variables. The number of points is less than the number of observations. A statistical measure is determined that describes a relationship between the observations and the inserted points. Weights and activation functions of the neural network are determined using the statistical measure.

The invention discloses an image recognition method, a device and equipment based on a hybrid neural network model, and a computer readable storage medium. The method comprises the steps of inputtinga to-be-recognized image into a convolutional auto-encoder for preprocessing; extracting image features of the preprocessed to-be-identified image by using a feature extractor constructed based on transfer learning; extracting internal time sequence features of the preprocessed to-be-identified image by using a long short-term memory network model; utilizing a feature fusion door and a feature screening door to fuse and screen the image features and the internal time sequence features to obtain target features of the recognition image; and utilizing a softmax classifier to classify the targetfeatures to obtain a classification result of the to-be-identified image. According to the method, the device, the equipment and the computer readable storage medium provided by the invention, the number of images required for training the neural network model can be greatly reduced, and the accuracy of image recognition is improved.