942 results about "Auto encoders" patented technology

Embodiments of the present invention relate to systems and methods for improving the training of machine learning systems to recognize certain objects within a given image by supplementing an existing sparse set of real-world training images with a comparatively dense set of realistic training images. Embodiments may create such a dense set of realistic training images by training a machine learning translator with a convolutional autoencoder to translate a dense set of synthetic images of an object into more realistic training images. Embodiments may also create a dense set of realistic training images by training a generative adversarial network (“GAN”) to create realistic training images from a combination of the existing sparse set of real-world training images and either Gaussian noise, translated images, or synthetic images. The created dense set of realistic training images may then be used to more effectively train a machine learning object recognizer to recognize a target object in a newly presented digital image.

The invention provides an SAR image terrain classification method based on a depth RBF network. The method mainly solves the problem of the prior art that the accuracy of classification is low. The method comprises the steps of (1) extracting the texton features of an SAR image; (2) training the texton features of the SAR image through a first-layer RBF neural network of the depth RBF network to obtain the advanced features of the image; (3) training the advanced features through a second-layer sparse autocoder network SAE of the depth RBF network to obtain more advanced features of the image; (4) training the more advanced features through a third-layer RBF neural network of the depth RBF network to obtain the terrain classification features of the image; (5) comparing the terrain classification features of an image test sample with a test sample label, adjusting the parameters of each layer of the depth RBF network, and obtaining an optimal test classification accuracy. The method is high in classification accuracy and can be used for complicated image classification.

The present invention provides a method for repairing face defect images based on an auto-encoder and generative adversarial networks. Through combination of the auto-encoder and the generative adversarial networks, the method comprises the following steps of: (1) performing face data set defect preprocessing; (2) employing the data set after processing to train the auto-encoder to allow the auto-encoder to reach the optimal state; (3) employing the data set after processing to train the condition generative adversarial networks to allow the condition generative adversarial networks to reach the optimal state; (4) inputting a defect image to be repaired into the trained encoder to generate a face image to be repaired; and (5) inputting the image to be repaired to the condition generative adversarial networks to generate much clearer and more natural restored face image. The method improves the restoring definition of the defect face area and the fidelity of the defect content, avoids the pseudomorphism of the defect area edges to the maximum extent, restricts the generation direction of the defect area and generates much clearer and more natural restoring effect.

The invention discloses a session emotion autoanalysis method based on depth learning and belongs to the natural language processing and data mining field. According to the method, voice and text expression is learned on the basis of a de-noising auto-encoder, further through a depth learning method, depth fusion of two types of expressions is realized to acquire unified high level expression, and emotion analysis is carried out on the basis of high level expression after fusion. Through the method, depth fusion of acoustic and text characteristics is realized, and emotion classification accuracy is improved.

The invention discloses a rolling bearing sound signal fault diagnosis method based on short-time Fourier transform and a sparse laminated automatic encoder. According to the method of the invention, firstly a smart mobile phone is used for acquiring a sound signal of the rolling bearing fault; then short-time Fourier analysis is performed on the sound signal for obtaining a spectrogram matrix; then the modulus value of the matrix is acquired and gray scale normalization processing is performed; then the normalized data are selected and input into a deep studying network for automatically extracting characteristics; and finally the characteristic which is extracted by a neural net is input into a Softmax classifier for identifying the fault mode. The invention provides the rolling bearing sound signal fault diagnosis method based on smart mobile phone sound signal short-time Fourier transform (STFT) and the sparse laminated automatic encoder (SAE). Through testing result analysis, the rolling bearing sound signal fault diagnosis method can accurately determine the fault mode of the rolling bearing.

The invention provides a mobile robot path planning method with combination of a depth automatic encoder and a Q-learning algorithm. The method comprises a depth automatic encoder part, a BP neural network part and a reinforced learning part. The depth automatic encoder part mainly adopts the depth automatic encoder to process images of an environment in which a robot is positioned so that the characteristics of the image data are acquired, and a foundation is laid for subsequent environment cognition. The BP neural network part is mainly for realizing fitting of reward values and the image characteristic data so that combination of the depth automatic encoder and the reinforced learning can be realized. According to the Q-learning algorithm, knowledge is obtained in an action-evaluation environment via interactive learning with the environment, and an action scheme is improved to be suitable for the environment to achieve the desired purpose. The robot interacts with the environment to realize autonomous learning, and finally a feasible path from a start point to a terminal point can be found. System image processing capacity can be enhanced, and environment cognition can be realized via combination of the depth automatic encoder and the BP neural network.

A system and method are provided to detect, analyze and digitally remove makeup from an image of a face. An autoencoder-based framework is provided to extract attractiveness-aware features to perform an assessment of facial beauty.

The invention discloses a face image aging synthesis method based on a feature separation adversarial network, belongs to the technical field of computer vision, and relates to a face image aging synthesis method. The method comprises the steps: supposing that the age features of a face are distributed in a Manifold, extracting the hidden features of an original image through an Auto-encoder, adding a corresponding age condition, finally generating an adversarial network through the condition, synthesizing a face image of a specific age, and maintaining the identity corresponding to an original face image. The innovativeness of the invention lies in that the method achieves the constraint of the hidden features outputted by an Encoder through the intra-class distance measure, enables the hidden features to be exclusively correlated with the identity, and removes the information correlated with the age, so as to guarantee that the age of the generated image is matched with a target agein a better way after the age is added. The method can be used for the face image aging synthesis, the cross-age face recognition and beatifying software.

The invention discloses a speech emotion recognition method. The method includes the steps that firstly, a speech signal is converted into a spectrogram to serve as initial input; secondly, a deep convolutional neural network is trained to automatically extract emotion features; thirdly, a stack type auto-encoder is trained for each kind of emotions, and all the stack type auto-encoders are fused to automatically construct membership functions of an emotion fuzzy set; fourthly, the features obtained in the second step are subjected to feature optimization by means of the fuzzy optimal theory in the third step; fifthly, emotion classification recognition is conducted by means of a Softmax classifier. The method takes abstract fuzzy properties of speech emotion information into consideration, the extracted emotion features are subjected to selective fuzzy optimization to improve the significance of the features, fuzzy membership functions in the fuzzy theory are automatically constructed by means of the concept of deep neural network layer-by-layer training, and the problem that the proper membership functions in the fuzzy theory are difficult to select and determine is solved.

The invention discloses a convolutional variational auto-encoder neural network-based finger vein identification method and system. The system comprises an image collection module, an image preprocessing module, an image feature extraction module, an image training module and an image identification module. The identification method comprises the steps of obtaining a finger vein image of a to-be-identified user; performing image preprocessing on finger vein image information, and extracting a finger vein region-of-interest (ROI) image; extracting finger vein feature codes in a finger vein ROIthrough the convolutional variational auto-encoder neural network; and inputting the feature codes to a full connection network for performing identification processing, thereby identifying identity information of the to-be-identified user. The finger vein features can be effectively extracted; the noise redundancy is improved; and the identification precision of the finger vein identification system is remarkably improved.

Data indicative of a plurality of observations of an environment are received at a control system. Machine learning using deep reinforcement learning is applied to determine an action based on the observations. The deep reinforcement learning applies a convolutional neural network or a deep auto encoder to the observations and applies a training set to locate one or more regions having a higher reward. The action is applied to the environment. A reward token indicative of alignment between the action and a desired result is received. A policy parameter of the control system is updated based on the reward token. The updated policy parameter is applied to determine a subsequent action responsive to a subsequent observation.

The invention provides an image conversion method based on a variation automatic encoder and the generative adversarial network. The method is mainly characterized by comprising the variation automatic encoder (VAE), weight sharing, generating the generative adversarial network (GAN) and learning, in the process, a non-monitored image is utilized to learn a bidirectional conversion function between two image domains in an image conversion network framework (UNIT), VAE and VAE are comprised, modeling for each image domain is carried out through utilizing the VAE and the VAE, mutual action of an adversarial training target and a weight sharing constraint is carried out, corresponding images are generated in the two image domains, the conversion image is associated with an input image of each domain, and image reconstruction flow and image conversion flow problems can be solved through training network combination. The method is advantaged in that the non-monitoring image is utilized to the image conversion framework, images in the two domains having not any relations are made to accomplish conversion, a corresponding training data set formed by the images is not needed, efficiency and practicality are improved, and the method can be developed to non-monitoring language conversion.

The invention discloses an abnormity detection method for satellite telemetry data through an adversarial network autoencoder, and the method comprises the steps: breaking the limitation of a traditional empirical model, and employing a pure data driving model; on the basis of a variational autoencoder, introducing a confrontation network idea, using a bidirectional LSTM (Long Short Term Memory) (Long-short term memory network) as a discriminator, and judging whether satellite telemetry data is abnormal or not by using errors of reconstructed data and original data; aiming at the redundancy problem of a satellite sensor, the conventional situation is broken through, and a Markov distance is used for measuring a reconstruction error. In combination with periodicity of satellite orbit operation, a dynamic threshold determination method based on a periodic time window is provided. The method has the advantages that pure data driving is adopted, expert experience is not needed, and the method can be suitable for various occasions; By combining the respective advantages of the variational auto-encoder and the generative adversarial network, the proposed network has the characteristics of high training speed and relatively easy convergence; eliminating redundant data influence between satellite telemetry data by adopting a Mahalanobis distance. According to the periodicity of the satellite, the dynamic threshold method based on the periodic time window is provided, and the misjudgment rate is reduced.

A computational system can include digital circuitry and analog circuitry, for instance a digital processor and a quantum processor. The quantum processor can operate as a sample generator providing samples. Samples can be employed by the digital processing in implementing various machine learning techniques. For example, the computational system can perform unsupervised learning over an input space, for example via a discrete variational auto-encoder, and attempting to maximize the log-likelihood of an observed dataset. Maximizing the log-likelihood of the observed dataset can include generating a hierarchical approximating posterior. Unsupervised learning can include generating samples of a prior distribution using the quantum processor. Generating samples using the quantum processor can include forming chains of qubits and representing discrete variables by chains.

The invention discloses an electroencephalogram signal feature extraction and classification method of combining a denoising auto encoder (DAE) and a convolutional neural network (CNN). The method includes the steps of: collecting electroencephalogram data through an electroencephalogram signal collection instrument; carrying out preprocessing of abnormal-sample removing, mean removing, signal filtering and the like on the collected data; using the auto encoder, into which a noise coefficient is added, and electroencephalogram signals for training; using hidden layers of the denoising auto encoder as feature data output; then converting obtained feature data into an image-like format; utilizing the convolutional neural network for classification; and finally, utilizing a test data set to carry out a performance test on the trained network. Compared with other traditional methods, the method of the invention can obtain a higher classification accuracy rate and higher robustness.

The invention belongs to the technical field of image processing and mode recognition and particularly relates to a license plate recognition method based on a deep convolutional neural network. The method includes: performing license plate detection on vehicle images, performing image segmentation on detected license plates to obtain license plate characters, using the license plate characters as training samples to obtain a training sample block set, inputting the training sample block set into a deep auto-encoder to train the deep auto-encoder, using the trained deep auto-encoder as the convolution kernel of the convolutional neural network, extracting the convolution features of the training sample block set, performing pooling operation on the convolution features of the training sample block set to obtain feature vectors, performing normalization processing on the feature vectors, loading the feature vectors after the normalization processing into an SVM classifier to train the SVM classifier, and testing to-be-recognized vehicles. By the method, license plate recognition accuracy can be increased, and license plate character recognition rate and robustness can be increased when the license plate characters are located in severe environments.

Disclosed is a data analysis and cybersecurity method, which forms a time-based series of behavioral features, and analyzes the series of behavioral features for attack detection, new features derivation, and / or features evaluation. Analyzing the time based series of behavioral features may comprise using a Feed-Forward Neural Networks (FFNN) method, a Convolutional Neural Networks (CNN) method, a Recurrent Neural Networks (RNN) method, a Long Short-Term Memories (LSTMs) method, a principal Component Analysis (PCA) method, a Random Forest pipeline method, and / or an autoencoder method. In one embodiment, the behavioral features of the time-based series of behavioral features comprise human engineered features, and / or machined learned features, wherein the method may be used to learn new features from historic features.

The present invention relates to a zero-sample classification technology in the computer vision field, in particular, a variational automatic encoder-based zero-sample image classification method. Asto the zero-sample image classification method, the distribution of the mappings of semantic features and visual features of categories in a semantic space is fitted, and more efficient semantic associations between the visual features and category semantics are built. According to the variational automatic encoder-based zero-sample image classification method, a variational automatic encoder is adopted to generate embedded semantic features on the basis of the visual features; it is regarded that the variational automatic encoder has a latent variable Z<^>; the latent variable Z<^> is adoptedas an embedded semantic feature; as for a zero-sample image classification task and the visual feature xj of a category-unknown sample, the encoding network of the variational automatic encoder whichis trained on visual categories is utilized to calculate a latent variable Z<^>j which is generated through encoding; the latent variable Z<^>j is adopted as an embedded semantic feature, cosine distances between the latent variable Z<^>j and the semantic feature of each invisible category are calculated, wherein the semantic feature of each invisible category is represented by a symbol describedin the descriptions of the invention; and a category of which the semantic feature is separated from the latent variable Z<^>j by the smallest distance is regarded as the category of the vision sample. The method of the present invention is mainly applied to video classification conditions.

The invention relates to an event trigger word extraction method based on auto-encoder fusion document information, which comprises the following steps: generating a training set by using unlabeled free text corpora, and training a GRU model to construct an auto-encoder; pre-processing and labeling the training corpus, and extracting to-be-identified words; obtaining a document vector of the to-be-identified word in a document where the to-be-identified word is located by the auto-encoder, and taking the document vector as a global feature of the to-be-identified word; expressing word vectorsand entity types of the to-be-identified words in a distributed manner to serve as local features of the to-be-identified words; carrying out vector splicing on the global feature and the local feature to obtain context features of the to-be-identified word; inputting the context features to the Bi-GRU model to perform multi-classification to identify whether the to-be-identified word is an eventtrigger word and the corresponding event type of the to-be-identified word.

Methods and systems for embedding a network in a latent space include generating a representation of an input network graph in the latent space using an autoencoder model and generating a representation of a set of noise samples in the latent space using a generator model. A discriminator model discriminates between the representation of the input network graph and the representation of the set of noise samples. The autoencoder model, the generator model, and the discriminator model are jointly trained by minimizing a joint loss function that includes parameters for each model. A final representation of the input network graph is generated using the trained autoencoder model.

The present invention relates to an image fusion method, especially to an image fusion method based on depth learning. The method comprises: employing a convolution layer to construct basic units based on an automatic encoder; stacking up a plurality of basic units for training to obtain a depth stack neural network, and employing an end-to-end mode to regulate the stack network; employing the stack network to decompose input images, obtaining high-frequency and low-frequency feature mapping pictures of each input image, and employing local variance maximum and region matching degree to merge the high-frequency and low-frequency feature mapping pictures; and putting a high-frequency fusion feature mapping picture and a low-frequency fusion feature mapping picture back to the last layer of the network, and obtaining a final fusion image. The image fusion method based on depth learning can perform adaptive decomposition and reconstruction of images, one high-frequency feature mapping picture and one low-frequency mapping picture are only needed when fusion, the number of the types of filters do not need artificial definition, the number of the layers of decomposition and the number of filtering directions of the images do not need selection, and the dependence of the fusion algorithm on the prior knowledge can be greatly improved.

The invention relates to the technical field of hydroelectric generating set fault diagnosis, in particular to a hydroelectric generating set fault diagnosis method and system based on a DdAE (Difference Differential Algebraic Equations) deep learning model. The method and the system are established on the basis of the analysis of the original vibration data of the hydroelectric generating set, adeep learning characteristic extraction method based on a multilayer neural network model is adopted, a complex manual processing and feature extraction process is not required, an ASFA (Aquatic Sciences and Fisheries Abstracts) method based on random search is adopted to carry out the structural parameter adjustment and optimization of the DdAE to achieve a purpose of strategy optimization. A deep denoising automatic encoder model is used for realizing the distributed expression of original data, and reconstruction data subjected to feature extraction is input into a Softmax regression modelto judge the work state and the fault type of the hydroelectric generating set. The analysis of a network experiment result indicates that the method can be effectively applied to the hydroelectric generating set fault diagnosis.

Autoencoder-based, semi-supervised approaches are used for anomaly detection. Defects on semiconductor wafers can be discovered using these approaches. The model can include a variational autoencoder, such as a one that includes ladder networks. Defect-free or clean images can be used to train the model that is later used to discover defects or other anomalies.

The embodiment of the invention provides a training sample data expansion method and device based on a variational auto-encoder, and relates to the technical field of big data. The method comprises the steps of obtaining an original sample; inputting the original sample into the encoder of the variational autoencoder, wherein the encoder of the variational autoencoder comprises two neural networks, the two neural networks output Mu and sigma respectively, and Mu and sigma are both functions of the original sample; according to the square of the Mu and sigma, namely sigma 2, generating a randomnumber of corresponding Gaussian distribution; randomly sampling the standard normal distribution to obtain a sampling value epsilon, and determining a sampling variable Z according to the sampling value epsilon and the random number of the Gaussian distribution; and inputting the sampling variable Z into a decoder of the variational autoencoder, decoding the sampling variable Z by the decoder ofthe variational autoencoder, and then outputting similar samples of the original samples, and taking the similar samples as extended samples. Therefore, the technical scheme provided by the embodiment of the invention can solve the problems that the time and labor are wasted and the efficiency is low when sample data is manually expanded in the prior art.

The invention discloses a target tracking method based on difficult positive sample generation. According to the method, for each video in training data, a variation auto-encoder is utilized to learna corresponding flow pattern, namely a positive sample generation network, codes are slightly adjusted according to an input image obtained after encoding, and a large quantity of positive samples aregenerated; the positive samples are input into a difficult positive sample conversion network, an intelligent body is trained to learn to shelter a target object through one background image block, the intelligent body performs bounding box adjustment continuously, so that the samples are difficult to recognize, the purpose of difficult positive sample generation is achieved, and sheltered difficult positive samples are output; and based on the generated difficult positive samples, a twin network is trained and used for matching between a target image block and candidate image blocks, and positioning of a target in a current frame is completed till processing of the whole video is completed. According to the target tracking method based on difficult positive sample generation, the flow pattern distribution of the target is learnt directly from the data, and a large quantity of diversified positive samples can be obtained.

The invention discloses an abnormal behavior detection method based on a space-time automatic encoder, and belongs to the technical field of image processing and mode recognition. The method comprisesthe steps of firstly obtaining a video image of a to-be-detected area; i; inputting the optical flow information and the RGB information of the video into a space-time auto-encoder respectively; p; passing through a 3D convolution layer in an automatic encoder; a; a pooling layer, an LSTM layer and a deconvolution layer, DERIVING RECONSTRUCTION INFORMATION, T, the method comprises the steps thatinput information is compared with reconstructed information, abnormal scores are obtained according to an abnormal scoring formula, t, then the two abnormal scores are fused to obtain a comprehensiveabnormal score, t, the comprehensive abnormal score is compared with a threshold value, whether abnormal behaviors exist or not and the occurrence time of abnormal conditions are judged, and the higher the abnormal score is, t, the higher the probability of occurrence of the abnormal conditions is. According to the invention, t, through the space-time automatic encoder fusing multi-modal input information, abnormal behaviors in crowds are detected in public areas such as banks, and early warning is sent to security personnel, so that emergencies such as major abnormal events are reduced.

The invention discloses a stacked noise reduction self-coding motor fault diagnosis method based on vibration and current signals, and the method comprises the following steps: 1, obtaining the time domain signals of the vibration and current of the motor during different faults, carrying out the preprocessing, and taking the processed signals as network input; 2, determining network parameters; 3, carrying out the layer by layer training, taking a hiding layer of an AE (Auto encoder) at an upper level as the input layer of an AE at a lower level, thereby obtaining a final feature code which is used for training a Softmax network; 4, carrying out the fine tuning of the whole network, judging whether the expected precision is met or not: finishing the training of the network if the expectedprecision is met, or else adjusting the network parameters, and repeatedly carrying out the step 3; 5, finishing the network construction. According to the invention, the multilayer SDAE network is constructed, and the vibration frequency domain signal and the current time domain signal are combined as the input. The SDAE network and a classifier are sequentially trained, and the supervised finetuning of the whole network is carried, thereby achieving the precise diagnosis of the fault of the motor.

The invention discloses an increment compensation dynamic adaptive enhancement-based fault diagnosis method. The method comprises the following steps of 1, collecting vibration signals in various states for a vibration sensor at a driving end of a motor; 2, preprocessing fault data of a bearing device; 3, adding a training sample into a random noise to serve as an input of a denoising autocoder to perform unsupervised greedy layer-by-layer pre-training; 4, when new device state data exists, performing new fault mode extraction by using an existing trained DAE model, performing similar mode comparison by utilizing a mode similarity algorithm, performing increment combination on new fault modes by adopting an increment active fusion algorithm, and calculating dynamic weighting by utilizing a weight dynamic compensation algorithm; 5, training an SVM classifier by taking labeled fault data and unlabeled fault data subjected to dynamic deep learning training weighting as input vectors; 6, performing global fine adjustment on related parameters in the whole model by utilizing a BP algorithm; and 7, performing classification diagnosis of fault types.

The invention relates to the technical field of multi-modal data retrieval, in particular to a cross-modal retrieval method and device, computer equipment and a storage medium. The method comprises the steps of obtaining first modal to-be-matched data, wherein the first modal to-be-matched data comprises image data and text data; When the first modal to-be-matched data is image data, carrying outfeature vector extraction by using a deep residual network ResNet model, and when the first modal to-be-matched data is text data, carrying out feature vector extraction by using a variational auto-encoder model; Mapping the feature vector to a common representation space by using a preset mapping function; And calculating the similarity between the first modal to-be-matched data and the second modal matching data in the common representation space, and outputting the corresponding second modal matching data according to the similarity to complete cross-modal retrieval. Characteristics of dataare extracted more fully, and retrieval accuracy is improved.