2067 results about "Gradient descent" patented technology

The invention discloses a full convolution neural network (FCN)-based monocular image depth estimation method. The method comprises the steps of acquiring training image data; inputting the training image data into a full convolution neural network (FCN), and sequentially outputting through pooling layers to obtain a characteristic image; subjecting each characteristic image outputted by a last pooling layer sequentially to amplification treatment to obtain a new characteristic image the same with the dimension of a characteristic image outputted by a previous pooling layer, and fusing the twocharacteristic images; sequentially fusing the outputted characteristic image of each pooling layer from back to front so as to obtain a final prediction depth image; training the parameters of the full convolution neural network (FCN) by utilizing a random gradient descent method (SGD) during training; acquiring an RGB image required for depth prediction, and inputting the RGB image into the well trained full convolution neural network (FCN) so as to obtain a corresponding prediction depth image. According to the method, the problem that the resolution of an output image is low in the convolution process can be solved. By adopting the form of the full convolution neural network, a full-connection layer is removed. The number of parameters in the network is effectively reduced.

The invention discloses a semi-supervised image classification method based on a generative adversarial network, which mainly solves the problems that the existing unsupervised learning classificationprecision is low and semi-supervised learning needs a large number of accurate labels, and comprises the following implementation steps of: 1) selecting and downloading a standard image training sample and a test sample; 2) setting relevant parameters of network supervised learning, and establishing a generative adversarial network consisting of a generator network, a discriminator network and anauxiliary classifier in parallel; 3) training the generative adversarial network by using a random gradient descent method; and 4) inputting the test sample to be classified into the trained generative adversarial network model, and outputting the category of the image to be detected. The method improves the image classification precision of unsupervised learning, can obtain a very good image classification effect on a sample set only containing a small amount of accurate annotation samples, and can be used for target classification in an actual scene.

A system and method for projection number and radiation dose reduction in tomographic imaging that creates a three dimensional cross sectional image of an object by the reconstruction of its projections. Images of a superior quality can be obtained with a fewer number projections than seen with conventional methods by reconstruction of projections that have been pre-processed and preferably placed in the Fourier domain with a Fractional Fourier Transform (FrFT) or forward Pseudo-polar Fast Fourier transform (PPFFT). The projections are iteratively refined through formulation of a constrained optimization problem with constraints in object space and Fourier space preferably solved with a gradient descent algorithm incorporating a Bregman iterative regularization or a continuative regularization. By using an exact Fourier-based iterative algorithm as well as physical and mathematical constraints, convergence to a lowest-possible noise state that is also strictly consistent with the measured data can be obtained.

The invention discloses a federation learning method, a system and a readable storage medium. The federated learning method includes the following steps: the data terminal performs federation trainingon the multi-party training samples based on the gradient descent tree GBDT algorithm, to construct a gradient tree model, wherein the data terminal is a plurality of, the gradient tree model comprises a plurality of regression trees, the regression trees comprise a plurality of partition points, and the training sample comprises a plurality of features, the features correspond to the partition points one by one; the data terminal performs joint prediction on a sample to be predicted based on the gradient tree model to determine a prediction value of the sample to be predicted. The inventioncarries out federation training on multi-party training samples through GBDT algorithm, realizes the establishment of gradient tree model, and is suitable for scenes with large data volume and can well meet the needs of realistic production environment through the gradient tree model. Forecast the sample to be forecasted jointly, and realize the forecast of the sample to be forecasted.

The invention discloses an identification method of emotional tendency of network comment texts and a convolutional neutral network model. The method comprises the steps as follows: grabbed network comment texts constitute a data set; word segmentation and text preprocessing are performed; all words subjected to text preprocessing are trained, and word vector representation of all words is obtained; the convolutional neutral network model is constructed and trained on a training set selected from the data set, and network parameters are updated with a back-propagating algorithm; in each step of training, noise is added to word vectors of an input layer for construction of adversarial samples, adversarial training is performed, and network parameters are updated with a random gradient descent algorithm; a classification model is obtained through repeated iteration to identify the emotional tendency of the network review texts. The convolutional neutral network model is used in the method and comprises the input layer, a convolution layer, a pooling layer and a classification layer. The adversarial samples can be classified correctly and the identification accuracy is improved.

The invention discloses an infrared target instance segmentation method based on feature fusion and a dense connection network, and the method comprises the steps: collecting and constructing an infrared image data set required for instance segmentation, and obtaining an original known infrared tag image; Performing image enhancement preprocessing on the infrared image data set; Processing the preprocessed training set to obtain a classification result, a frame regression result and an instance segmentation mask result graph; Performing back propagation in the convolutional neural network by using a random gradient descent method according to the prediction loss function, and updating parameter values of the convolutional neural network; Selecting a fixed number of infrared image data training sets each time and sending the infrared image data training sets to the network for processing, and repeatedly carrying out iterative updating on the convolutional network parameters until the convolutional network training is completed by the maximum number of iterations; And processing the test set image data to obtain average precision and required time of instance segmentation and a finalinstance segmentation result graph.

The invention discloses a driver behavior identification method based on a multi-scale attention convolutional neural network. The method comprises the following steps: (1) shooting an image data setfor driver behavior identification; (2) carrying out data enhancement on the shot driver behavior data set and simultaneously bringing samples obtained by enhancement into training data; (3) constructing a neural network model which comprises three modules, namely a multi-scale convolution module, an attention module and a classification module; (4) training a multi-scale attention convolutional neural network; using a Pytorch open source tool i for building a network model, and using a random gradient descent method for optimizing network parameters; and (5) testing the multi-column convolutional neural network. According to the invention, the multi-scale model and the attention mechanism are introduced into the driver behavior identification task to extract the behavior fine-grained feature representation with the distinction degree, so that the driver behavior identification accuracy can be further improved.

A method and system extracts a matte from images acquired of a scene. A foreground image focused at a foreground in a scene, a background image focused at a background in the scene, and a pinhole image focused on the entire scene are acquired. These three images can be acquired sequentially by a single camera, or simultaneous by three cameras. In the later case, foreground, background and pinhole sequences of images can be acquired. The pinhole image is compared to the foreground image and the background image to extract a matte representing the scene. The comparison classifies pixels in the images as foreground, background, or unknown pixels. An optimizer minimizes an error function in the form of Fourier image equations using a gradient descent method. The error function expresses pixel intensity differences.

This invention relates to the integration of visual information, anatomic constraints, prior shape knowledge, and level set representations for the segmentation of medical images. An embodiment according to the present invention comprises a level set variational framework that uses a bi-directional boundary flow, an intensity-based regional component that maximizes the a posteriori segmentation probability, a physiology-based module that constrains the solution space and a term that accounts for shape-driven consistency. All modules are expressed in an energetic form and the resulting objective function is optimized using a gradient descent method.

A method and apparatus is disclosed for performing blind source separation using convolutive signal decorrelation. For a first embodiment, the method accumulates a length of input signal (mixed signal) that includes a plurality of independent signals from independent signal sources. The invention then divides the length of input signal into a plurality of T-length periods (windows) and performs a discrete Fourier transform (DFT) on the, signal within each T-length period. Thereafter, estimated cross-correlation values are computed using a plurality of the averaged DFT values. A total number of K cross-correlation values are computed, where each of the K values is averaged over N of the T-length periods. Using the cross-correlation values, a gradient descent process computes the coefficients of a finite impulse response (FIR) filter that will effectively separate the source signals within the input signal. A second embodiment of the invention is directed to on-line processing of the input signal—i.e., processing the signal as soon as it arrives with no storage of the signal data. In particular, an on-line gradient algorithm is provided for application to non-stationary signals and having an adaptive step size in the frequency domain based on second derivatives of the cost function. The on-line separation methodology of this embodiment is characterized as multiple adaptive decorrelation.

The invention proposes a dangerous driving behavior real-time detection method based on deep learning, and the method comprises the steps: building a front car image collection system, collecting a training sample, carrying out the manual marking, and building a dangerous driving behavior data set; proposing a spatial pyramid pooling convolution depth confidence normalized classification network (SPP-CDBRNet) model according to the characteristics of the dangerous driving behavior data set based on a deep learning method; carrying out the preprocessing of the built data set, carrying out the training of the SPP-CDBRNet model through a random gradient descent method with a momentum and the data set after preprocessing, and obtaining an SPP-CDBRNet which precisely recognize whether there is a dangerous driving behavior (behaviors of using a cellphone and smoking during driving); carrying out the detection of a front car image through the determined SPP-CDBRNet model, and achieving the real-time detection of the dangerous driving behavior. The method can effectively improve the detection precision of dangerous driving behaviors, is good in instantaneity and mobility, and is good in application prospect.

The invention discloses a fabric defect detection method based on a depth neural network. The method comprises the following steps: (1), an image acquisition system is built to acquire an image; (2), the image is segmented into experimental samples, fabric sample image data are enhanced at the same time, and a fabric image after enhancement serves as a training sample; (3), a depth neural network is designed; (4), parameters are set, the depth neural network is initialized, the training sample is fed to the depth neural network for training, and after network training is completed, the network model is saved; and (5), an inputted new fabric sample is fed to the network model for detection. According to the fabric defect detection method based on the depth neural network provided by the invention, with a convolutional neural network as a core, feature extraction is performed by a convolutional layer, a pooling layer retains effective features and reduces the amount of calculation, and a full connection layer is used for classification. A mini-batch gradient descent method is used for optimization, the generalization ability is enhanced through L2 regularization, defect recognition is carried out through determining the corresponding position of the maximum component outputted by a classifier, effects are shown in Figure 4, Actual presents the actual category of the sample, and Pred presents the predicted category of the sample.

The invention provides a facial emotion recognition method based on a deep sparse convolutional neural network. The method comprises the following steps: to begin with, carrying out emotion image preprocessing; then, carrying out emotion feature extraction; and finally, carrying out emotion feature identification and classification. The facial emotion recognition method based on the deep sparse convolutional neural network carries out optimization on weight of the deep sparse convolutional neural network through a Nesterov accelerated gradient descent algorithm to enable network structure to be optimal, thereby improving generalization of the face emotion recognition algorithm; since the NAGD has a precognition capability, the algorithm can be prevented from being too fast or too slow foreseeingly; and meanwhile, response capability of the algorithm can be enhanced, and better local optimum value can be obtained.

The invention relates to an urban road congestion degree prediction method based on time sequence traffic events. The method comprises the steps of: S1, acquiring historical traffic event data, real-time traffic event data and video monitoring data of an urban road section; S2, identifying traffic congestion forewarning events in the video data through the 3D CNN, and performing data space-time fusion according to the historical traffic events; S3, determining congestion degree classification labels, constructing a time sequence traffic congestion event data dictionary, and screening a training set, a verification set and a test set; S4, establishing an LSTM sequence data classification model, inputting the training set, and iteratively updating model parameters by utilizing a gradient descent method; S5, inputting the verification set into the model with updated parameters, optimizing and adjusting hyper-parameters, and selecting an optimal model; and S6, inputting the test set into the optimal training model, checking the effectiveness of the model, and carrying out road congestion prediction according to real-time traffic monitoring data. According to the method, a sequence dataclassification model is established by using LSTM, and the urban road congestion degree is predicted based on time sequence traffic events.

Methods and systems for finding optimal or near optimal solutions for generic optimization problems by an approach to minimizing functions over high-dimensional domains that mathematically model the optimization problems. Embodiments of the disclosed invention receive a mathematical description of a system, in symbolic form, that includes decision variables of various types, including real-number-valued, integer-valued, and Boolean-valued decision variables, and that may also include a variety of constraints on the values of the decision variables, including inequality and equality constraints. The objective function and constraints are incorporated into a global objective function. The global objective function is transformed into a system of differential equations in terms of continuous variables and parameters, so that polynomial-time methods for solving differential equations can be applied to calculate near-optimal solutions for the global objective function. Embodiments of the present invention also provide for distribution and decomposition of global-gradient-descent- field-based optimization methods, by following multiple trajectories, and local-gradient- descent-field-based optimization methods, by using multiple agents, in order to allow for parallel computation and increased computational efficiency. Various embodiments of the present invention further include approaches for relatively continuous adjustment of solutions to optimization problems in time, to respond to various events, changes in priorities, and changes in forecasts, without needing to continuously recalculate optimization solutions de novo. While many embodiments of the present invention are specifically directed to various classes of optimization problems, other embodiments of the present invention provide a more general approach for constructing complex hierarchical computational processes and for optimally or near optimally controlling general computational processes.

The invention discloses a deep neural network for the fine recognition of vehicle attributes and a training method thereof. The network comprises a depth residual network, a feature migration layer, aplurality of all-connection layers, a plurality of loss calculation units, and a plurality of parameter updating units. The depth residual network is used for carrying out feature extraction on an input image to obtain a feature image. The feature migration layer comprises a plurality of feature migration units and is used for enabling each of all feature migration units to be adapted to specifictasks according to the features shared by all attribute identifying tasks. The plurality of all-connection layers correspond to the branches of all attribute identifying tasks and are connected withthe feature migration layer so as to obtain feature vectors corresponding to all attribute identifying tasks. The plurality of loss calculation units correspond to the branches of all attribute identifying tasks and are respectively connected with the all-connection layers. The plurality of loss calculation units are used for calculating the loss of a loss function by adopting cross entropies as multiple classifiers. The plurality of parameter updating units correspond to the attribute identifying tasks and are connected with the loss calculation units. The parameter updating units are used for returning the loss based on the random gradient descent optimization algorithm, and updating parameters. According to the invention, various fine vehicle attributes can be identified at the same time by adopting only one neural network.

The invention relates to a power balancing-based non-uniform clustering method for a cluster wireless sensor network, relating to wireless communication technologies. The method comprises the steps of dividing a wireless sensor network monitoring region with uniformly distributed nodes into a non-uniform a ring, determining the positions of cluster head nodes according to an equivalent distance method, ensuring a reliability link index deltap of end-to-end communication between the cluster head nodes to cluster nodes, and setting up an energy consumption solving equation. An optimal ring radius vector is obtained by optimizing parameters of the energy consumption solving equation of the cluster head nodes with a steepest gradient descent algorithm based on the idea of 'power-balancing'. The cluster size is finally obtained by adjusting the positions and the transmission power of the cluster head nodes in accordance with the optimal ring radius vector, so as to balance energy consumption among the cluster head nodes and prolong the network life cycle. The method remarkably prolongs the network life cycle, improves the efficiency of the routing protocol and the media control protocol, and provides a support basis for time synchronization, data integration and target positioning technologies.

The invention discloses a convolution neural network based face detection method and apparatus. The method comprises the steps of: zooming all pictures in a training set into grayscale images of a predetermined size, and assigning label information to each pixel point of each picture, wherein the label information is used to show whether the pixel point that corresponds to the information is a face; establishing a convolution neural network, wherein layers of the convolution neural network are sequentially an input layer, a plurality of convolution layers, a plurality of full connectivity layers and an output layer; training the convolution neural network by using a gradient descent method and a back propagation algorithm; inputting a photograph to be detected into the well trained convolution neural network, so as to obtain an output characteristic value of a final layer; comparing the output characteristic value of a final layer to a predetermined threshold, so as to determine whether each pixel point of the photograph is a face region; and using a method of minimum closure, and detecting a face position according to pixel points that are determined as the face region.

The invention relates to a visual inertial navigation SLAM method based on ground plane hypothesis. According to the method, feature points are extracted from an image to perform IMU pre-integration,a camera projection model is established, and camera internal parameter calibration and external parameter calibration between an IMU and a camera are performed; a system is initialized, a visually observed point cloud and a camera pose are aligned to the IMU pre-integration, and a ground equation and the camera pose are restored; the ground is initialized to obtain a ground equation, the ground equation under the current camera pose is determined and back projected to an image coordinate system, and a more accurate ground region is acquired; and based on state estimation, all sensor observation models are derived, camera observation, IMU observation and ground feature observation are fused to do state estimation, a graph optimization model is used to do state estimation, and a sparse graph optimization and gradient descent method is used to realize overall optimization. Compared with previous algorithms, the precision of the method is greatly improved, estimation of the camera pose can be limited globally, and therefore accuracy is greatly improved.

The invention discloses a speech classification method based on a deep neural network, and aims at solving different speech classification problems through a unified algorithm model. The method of theinvention includes the steps of S1, converting a speech into a corresponding spectrogram; segmenting the complete spectrogram along the frequency domain into blocks to obtain a local frequency domaininformation set; S2, taking the complete and local frequency domain information as inputs of a model respectively, and based on the different inputs, the convolutional neural network being capable ofextracting local and global features; S3, using an attention mechanism to fuse global and local feature expressions to form a final feature expression; S4, using tagged data to train the network by gradient descent and back propagation algorithms; and S5, using the trained parameters for an untagged speech and taking the classification of highest probability that the model outputs as a predictionresult. The method of the invention realizes a unified algorithm model for different speech classification problems, and improves the accuracy on multiple speech classification problems.

An image editing application (or a blur classification module thereof) may automatically estimate a coherent defocus blur map from a single input image. The application may represent the blur spectrum as a differentiable function of radius r, and the optimal radius may be estimated by optimizing the likelihood function through a gradient descent algorithm. The application may generate the spectrum function over r through polynomial-based fitting. After fitting, the application may generate look-up tables to store values for the spectrum and for its first and second order derivatives, respectively. The use of these tables in the likelihood optimization process may significantly reduce the computational costs of a given blur estimation exercise. The application may minimize an energy function that includes a data term, a smoothness term, and a smoothness parameter that is adaptive to local image content. The output blur map may be used for image object depth estimation.

The invention discloses a deep learning-based text keyword extraction method. The method comprises the following steps of: firstly training a recurrent neural network model, wherein the used training data comprise a large amount of texts and keywords thereof, and the training target is maximizing text-based condition probability of the keywords; converting each text and the keyword thereof into word vectors, inputting the word vectors into the recurrent neural network model and updating network parameters by using a random gradient descent method; and after the model training is finished, converting a section of text, the keyword of which is to be extracted, into a word vector, inputting the word vector into the trained recurrent neural network model so as to generate the keyword of the section of text. According to the method disclosed by the invention, the extraction of text keywords is realized by learning an end-to-end model through data driving; and compared with the traditional statistics and linguistics-based method, the method disclosed by the invention is stronger in adaptability, and can be used for obtaining different models according to different training data so as to extract keywords according to the requirements of specific fields.